Subversion Repositories Kolibri OS

Rev

Rev 5354 | Rev 6935 | Go to most recent revision | Blame | Compare with Previous | Last modification | View Log | Download | RSS feed

  1. /*
  2.  * Copyright © 2012 Intel Corporation
  3.  *
  4.  * Permission is hereby granted, free of charge, to any person obtaining a
  5.  * copy of this software and associated documentation files (the "Software"),
  6.  * to deal in the Software without restriction, including without limitation
  7.  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  8.  * and/or sell copies of the Software, and to permit persons to whom the
  9.  * Software is furnished to do so, subject to the following conditions:
  10.  *
  11.  * The above copyright notice and this permission notice (including the next
  12.  * paragraph) shall be included in all copies or substantial portions of the
  13.  * Software.
  14.  *
  15.  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16.  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17.  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18.  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19.  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20.  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21.  * IN THE SOFTWARE.
  22.  *
  23.  * Authors:
  24.  *    Eugeni Dodonov <eugeni.dodonov@intel.com>
  25.  *
  26.  */
  27.  
  28. //#include <linux/cpufreq.h>
  29. #include "i915_drv.h"
  30. #include "intel_drv.h"
  31. //#include "../../../platform/x86/intel_ips.h"
  32. #include <linux/module.h>
  33.  
  34.  
  35. #define FORCEWAKE_ACK_TIMEOUT_MS 2
  36.  
  37. void getrawmonotonic(struct timespec *ts);
  38.  
  39. /**
  40.  * RC6 is a special power stage which allows the GPU to enter an very
  41.  * low-voltage mode when idle, using down to 0V while at this stage.  This
  42.  * stage is entered automatically when the GPU is idle when RC6 support is
  43.  * enabled, and as soon as new workload arises GPU wakes up automatically as well.
  44.  *
  45.  * There are different RC6 modes available in Intel GPU, which differentiate
  46.  * among each other with the latency required to enter and leave RC6 and
  47.  * voltage consumed by the GPU in different states.
  48.  *
  49.  * The combination of the following flags define which states GPU is allowed
  50.  * to enter, while RC6 is the normal RC6 state, RC6p is the deep RC6, and
  51.  * RC6pp is deepest RC6. Their support by hardware varies according to the
  52.  * GPU, BIOS, chipset and platform. RC6 is usually the safest one and the one
  53.  * which brings the most power savings; deeper states save more power, but
  54.  * require higher latency to switch to and wake up.
  55.  */
  56. #define INTEL_RC6_ENABLE                        (1<<0)
  57. #define INTEL_RC6p_ENABLE                       (1<<1)
  58. #define INTEL_RC6pp_ENABLE                      (1<<2)
  59.  
  60. static void bxt_init_clock_gating(struct drm_device *dev)
  61. {
  62.         struct drm_i915_private *dev_priv = dev->dev_private;
  63.  
  64.         /* WaDisableSDEUnitClockGating:bxt */
  65.         I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
  66.                    GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
  67.  
  68.         /*
  69.          * FIXME:
  70.          * GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ applies on 3x6 GT SKUs only.
  71.          */
  72.         I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
  73.                    GEN8_HDCUNIT_CLOCK_GATE_DISABLE_HDCREQ);
  74. }
  75.  
  76. static void i915_pineview_get_mem_freq(struct drm_device *dev)
  77. {
  78.         struct drm_i915_private *dev_priv = dev->dev_private;
  79.         u32 tmp;
  80.  
  81.         tmp = I915_READ(CLKCFG);
  82.  
  83.         switch (tmp & CLKCFG_FSB_MASK) {
  84.         case CLKCFG_FSB_533:
  85.                 dev_priv->fsb_freq = 533; /* 133*4 */
  86.                 break;
  87.         case CLKCFG_FSB_800:
  88.                 dev_priv->fsb_freq = 800; /* 200*4 */
  89.                 break;
  90.         case CLKCFG_FSB_667:
  91.                 dev_priv->fsb_freq =  667; /* 167*4 */
  92.                 break;
  93.         case CLKCFG_FSB_400:
  94.                 dev_priv->fsb_freq = 400; /* 100*4 */
  95.                 break;
  96.         }
  97.  
  98.         switch (tmp & CLKCFG_MEM_MASK) {
  99.         case CLKCFG_MEM_533:
  100.                 dev_priv->mem_freq = 533;
  101.                 break;
  102.         case CLKCFG_MEM_667:
  103.                 dev_priv->mem_freq = 667;
  104.                 break;
  105.         case CLKCFG_MEM_800:
  106.                 dev_priv->mem_freq = 800;
  107.                 break;
  108.         }
  109.  
  110.         /* detect pineview DDR3 setting */
  111.         tmp = I915_READ(CSHRDDR3CTL);
  112.         dev_priv->is_ddr3 = (tmp & CSHRDDR3CTL_DDR3) ? 1 : 0;
  113. }
  114.  
  115. static void i915_ironlake_get_mem_freq(struct drm_device *dev)
  116. {
  117.         struct drm_i915_private *dev_priv = dev->dev_private;
  118.         u16 ddrpll, csipll;
  119.  
  120.         ddrpll = I915_READ16(DDRMPLL1);
  121.         csipll = I915_READ16(CSIPLL0);
  122.  
  123.         switch (ddrpll & 0xff) {
  124.         case 0xc:
  125.                 dev_priv->mem_freq = 800;
  126.                 break;
  127.         case 0x10:
  128.                 dev_priv->mem_freq = 1066;
  129.                 break;
  130.         case 0x14:
  131.                 dev_priv->mem_freq = 1333;
  132.                 break;
  133.         case 0x18:
  134.                 dev_priv->mem_freq = 1600;
  135.                 break;
  136.         default:
  137.                 DRM_DEBUG_DRIVER("unknown memory frequency 0x%02x\n",
  138.                                  ddrpll & 0xff);
  139.                 dev_priv->mem_freq = 0;
  140.                 break;
  141.         }
  142.  
  143.         dev_priv->ips.r_t = dev_priv->mem_freq;
  144.  
  145.         switch (csipll & 0x3ff) {
  146.         case 0x00c:
  147.                 dev_priv->fsb_freq = 3200;
  148.                 break;
  149.         case 0x00e:
  150.                 dev_priv->fsb_freq = 3733;
  151.                 break;
  152.         case 0x010:
  153.                 dev_priv->fsb_freq = 4266;
  154.                 break;
  155.         case 0x012:
  156.                 dev_priv->fsb_freq = 4800;
  157.                 break;
  158.         case 0x014:
  159.                 dev_priv->fsb_freq = 5333;
  160.                 break;
  161.         case 0x016:
  162.                 dev_priv->fsb_freq = 5866;
  163.                 break;
  164.         case 0x018:
  165.                 dev_priv->fsb_freq = 6400;
  166.                 break;
  167.         default:
  168.                 DRM_DEBUG_DRIVER("unknown fsb frequency 0x%04x\n",
  169.                                  csipll & 0x3ff);
  170.                 dev_priv->fsb_freq = 0;
  171.                 break;
  172.         }
  173.  
  174.         if (dev_priv->fsb_freq == 3200) {
  175.                 dev_priv->ips.c_m = 0;
  176.         } else if (dev_priv->fsb_freq > 3200 && dev_priv->fsb_freq <= 4800) {
  177.                 dev_priv->ips.c_m = 1;
  178.         } else {
  179.                 dev_priv->ips.c_m = 2;
  180.         }
  181. }
  182.  
  183. static const struct cxsr_latency cxsr_latency_table[] = {
  184.         {1, 0, 800, 400, 3382, 33382, 3983, 33983},    /* DDR2-400 SC */
  185.         {1, 0, 800, 667, 3354, 33354, 3807, 33807},    /* DDR2-667 SC */
  186.         {1, 0, 800, 800, 3347, 33347, 3763, 33763},    /* DDR2-800 SC */
  187.         {1, 1, 800, 667, 6420, 36420, 6873, 36873},    /* DDR3-667 SC */
  188.         {1, 1, 800, 800, 5902, 35902, 6318, 36318},    /* DDR3-800 SC */
  189.  
  190.         {1, 0, 667, 400, 3400, 33400, 4021, 34021},    /* DDR2-400 SC */
  191.         {1, 0, 667, 667, 3372, 33372, 3845, 33845},    /* DDR2-667 SC */
  192.         {1, 0, 667, 800, 3386, 33386, 3822, 33822},    /* DDR2-800 SC */
  193.         {1, 1, 667, 667, 6438, 36438, 6911, 36911},    /* DDR3-667 SC */
  194.         {1, 1, 667, 800, 5941, 35941, 6377, 36377},    /* DDR3-800 SC */
  195.  
  196.         {1, 0, 400, 400, 3472, 33472, 4173, 34173},    /* DDR2-400 SC */
  197.         {1, 0, 400, 667, 3443, 33443, 3996, 33996},    /* DDR2-667 SC */
  198.         {1, 0, 400, 800, 3430, 33430, 3946, 33946},    /* DDR2-800 SC */
  199.         {1, 1, 400, 667, 6509, 36509, 7062, 37062},    /* DDR3-667 SC */
  200.         {1, 1, 400, 800, 5985, 35985, 6501, 36501},    /* DDR3-800 SC */
  201.  
  202.         {0, 0, 800, 400, 3438, 33438, 4065, 34065},    /* DDR2-400 SC */
  203.         {0, 0, 800, 667, 3410, 33410, 3889, 33889},    /* DDR2-667 SC */
  204.         {0, 0, 800, 800, 3403, 33403, 3845, 33845},    /* DDR2-800 SC */
  205.         {0, 1, 800, 667, 6476, 36476, 6955, 36955},    /* DDR3-667 SC */
  206.         {0, 1, 800, 800, 5958, 35958, 6400, 36400},    /* DDR3-800 SC */
  207.  
  208.         {0, 0, 667, 400, 3456, 33456, 4103, 34106},    /* DDR2-400 SC */
  209.         {0, 0, 667, 667, 3428, 33428, 3927, 33927},    /* DDR2-667 SC */
  210.         {0, 0, 667, 800, 3443, 33443, 3905, 33905},    /* DDR2-800 SC */
  211.         {0, 1, 667, 667, 6494, 36494, 6993, 36993},    /* DDR3-667 SC */
  212.         {0, 1, 667, 800, 5998, 35998, 6460, 36460},    /* DDR3-800 SC */
  213.  
  214.         {0, 0, 400, 400, 3528, 33528, 4255, 34255},    /* DDR2-400 SC */
  215.         {0, 0, 400, 667, 3500, 33500, 4079, 34079},    /* DDR2-667 SC */
  216.         {0, 0, 400, 800, 3487, 33487, 4029, 34029},    /* DDR2-800 SC */
  217.         {0, 1, 400, 667, 6566, 36566, 7145, 37145},    /* DDR3-667 SC */
  218.         {0, 1, 400, 800, 6042, 36042, 6584, 36584},    /* DDR3-800 SC */
  219. };
  220.  
  221. static const struct cxsr_latency *intel_get_cxsr_latency(int is_desktop,
  222.                                                          int is_ddr3,
  223.                                                          int fsb,
  224.                                                          int mem)
  225. {
  226.         const struct cxsr_latency *latency;
  227.         int i;
  228.  
  229.         if (fsb == 0 || mem == 0)
  230.                 return NULL;
  231.  
  232.         for (i = 0; i < ARRAY_SIZE(cxsr_latency_table); i++) {
  233.                 latency = &cxsr_latency_table[i];
  234.                 if (is_desktop == latency->is_desktop &&
  235.                     is_ddr3 == latency->is_ddr3 &&
  236.                     fsb == latency->fsb_freq && mem == latency->mem_freq)
  237.                         return latency;
  238.         }
  239.  
  240.         DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
  241.  
  242.         return NULL;
  243. }
  244.  
  245. static void chv_set_memory_dvfs(struct drm_i915_private *dev_priv, bool enable)
  246. {
  247.         u32 val;
  248.  
  249.         mutex_lock(&dev_priv->rps.hw_lock);
  250.  
  251.         val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
  252.         if (enable)
  253.                 val &= ~FORCE_DDR_HIGH_FREQ;
  254.         else
  255.                 val |= FORCE_DDR_HIGH_FREQ;
  256.         val &= ~FORCE_DDR_LOW_FREQ;
  257.         val |= FORCE_DDR_FREQ_REQ_ACK;
  258.         vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val);
  259.  
  260.         if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) &
  261.                       FORCE_DDR_FREQ_REQ_ACK) == 0, 3))
  262.                 DRM_ERROR("timed out waiting for Punit DDR DVFS request\n");
  263.  
  264.         mutex_unlock(&dev_priv->rps.hw_lock);
  265. }
  266.  
  267. static void chv_set_memory_pm5(struct drm_i915_private *dev_priv, bool enable)
  268. {
  269.         u32 val;
  270.  
  271.         mutex_lock(&dev_priv->rps.hw_lock);
  272.  
  273.         val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ);
  274.         if (enable)
  275.                 val |= DSP_MAXFIFO_PM5_ENABLE;
  276.         else
  277.                 val &= ~DSP_MAXFIFO_PM5_ENABLE;
  278.         vlv_punit_write(dev_priv, PUNIT_REG_DSPFREQ, val);
  279.  
  280.         mutex_unlock(&dev_priv->rps.hw_lock);
  281. }
  282.  
  283. #define FW_WM(value, plane) \
  284.         (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK)
  285.  
  286. void intel_set_memory_cxsr(struct drm_i915_private *dev_priv, bool enable)
  287. {
  288.         struct drm_device *dev = dev_priv->dev;
  289.         u32 val;
  290.  
  291.         if (IS_VALLEYVIEW(dev)) {
  292.                 I915_WRITE(FW_BLC_SELF_VLV, enable ? FW_CSPWRDWNEN : 0);
  293.                 POSTING_READ(FW_BLC_SELF_VLV);
  294.                 dev_priv->wm.vlv.cxsr = enable;
  295.         } else if (IS_G4X(dev) || IS_CRESTLINE(dev)) {
  296.                 I915_WRITE(FW_BLC_SELF, enable ? FW_BLC_SELF_EN : 0);
  297.                 POSTING_READ(FW_BLC_SELF);
  298.         } else if (IS_PINEVIEW(dev)) {
  299.                 val = I915_READ(DSPFW3) & ~PINEVIEW_SELF_REFRESH_EN;
  300.                 val |= enable ? PINEVIEW_SELF_REFRESH_EN : 0;
  301.                 I915_WRITE(DSPFW3, val);
  302.                 POSTING_READ(DSPFW3);
  303.         } else if (IS_I945G(dev) || IS_I945GM(dev)) {
  304.                 val = enable ? _MASKED_BIT_ENABLE(FW_BLC_SELF_EN) :
  305.                                _MASKED_BIT_DISABLE(FW_BLC_SELF_EN);
  306.                 I915_WRITE(FW_BLC_SELF, val);
  307.                 POSTING_READ(FW_BLC_SELF);
  308.         } else if (IS_I915GM(dev)) {
  309.                 val = enable ? _MASKED_BIT_ENABLE(INSTPM_SELF_EN) :
  310.                                _MASKED_BIT_DISABLE(INSTPM_SELF_EN);
  311.                 I915_WRITE(INSTPM, val);
  312.                 POSTING_READ(INSTPM);
  313.         } else {
  314.                 return;
  315.         }
  316.  
  317.         DRM_DEBUG_KMS("memory self-refresh is %s\n",
  318.                       enable ? "enabled" : "disabled");
  319. }
  320.  
  321.  
  322. /*
  323.  * Latency for FIFO fetches is dependent on several factors:
  324.  *   - memory configuration (speed, channels)
  325.  *   - chipset
  326.  *   - current MCH state
  327.  * It can be fairly high in some situations, so here we assume a fairly
  328.  * pessimal value.  It's a tradeoff between extra memory fetches (if we
  329.  * set this value too high, the FIFO will fetch frequently to stay full)
  330.  * and power consumption (set it too low to save power and we might see
  331.  * FIFO underruns and display "flicker").
  332.  *
  333.  * A value of 5us seems to be a good balance; safe for very low end
  334.  * platforms but not overly aggressive on lower latency configs.
  335.  */
  336. static const int pessimal_latency_ns = 5000;
  337.  
  338. #define VLV_FIFO_START(dsparb, dsparb2, lo_shift, hi_shift) \
  339.         ((((dsparb) >> (lo_shift)) & 0xff) | ((((dsparb2) >> (hi_shift)) & 0x1) << 8))
  340.  
  341. static int vlv_get_fifo_size(struct drm_device *dev,
  342.                               enum pipe pipe, int plane)
  343. {
  344.         struct drm_i915_private *dev_priv = dev->dev_private;
  345.         int sprite0_start, sprite1_start, size;
  346.  
  347.         switch (pipe) {
  348.                 uint32_t dsparb, dsparb2, dsparb3;
  349.         case PIPE_A:
  350.                 dsparb = I915_READ(DSPARB);
  351.                 dsparb2 = I915_READ(DSPARB2);
  352.                 sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 0, 0);
  353.                 sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 8, 4);
  354.                 break;
  355.         case PIPE_B:
  356.                 dsparb = I915_READ(DSPARB);
  357.                 dsparb2 = I915_READ(DSPARB2);
  358.                 sprite0_start = VLV_FIFO_START(dsparb, dsparb2, 16, 8);
  359.                 sprite1_start = VLV_FIFO_START(dsparb, dsparb2, 24, 12);
  360.                 break;
  361.         case PIPE_C:
  362.                 dsparb2 = I915_READ(DSPARB2);
  363.                 dsparb3 = I915_READ(DSPARB3);
  364.                 sprite0_start = VLV_FIFO_START(dsparb3, dsparb2, 0, 16);
  365.                 sprite1_start = VLV_FIFO_START(dsparb3, dsparb2, 8, 20);
  366.                 break;
  367.         default:
  368.                 return 0;
  369.         }
  370.  
  371.         switch (plane) {
  372.         case 0:
  373.                 size = sprite0_start;
  374.                 break;
  375.         case 1:
  376.                 size = sprite1_start - sprite0_start;
  377.                 break;
  378.         case 2:
  379.                 size = 512 - 1 - sprite1_start;
  380.                 break;
  381.         default:
  382.                 return 0;
  383.         }
  384.  
  385.         DRM_DEBUG_KMS("Pipe %c %s %c FIFO size: %d\n",
  386.                       pipe_name(pipe), plane == 0 ? "primary" : "sprite",
  387.                       plane == 0 ? plane_name(pipe) : sprite_name(pipe, plane - 1),
  388.                       size);
  389.  
  390.         return size;
  391. }
  392.  
  393. static int i9xx_get_fifo_size(struct drm_device *dev, int plane)
  394. {
  395.         struct drm_i915_private *dev_priv = dev->dev_private;
  396.         uint32_t dsparb = I915_READ(DSPARB);
  397.         int size;
  398.  
  399.         size = dsparb & 0x7f;
  400.         if (plane)
  401.                 size = ((dsparb >> DSPARB_CSTART_SHIFT) & 0x7f) - size;
  402.  
  403.         DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
  404.                       plane ? "B" : "A", size);
  405.  
  406.         return size;
  407. }
  408.  
  409. static int i830_get_fifo_size(struct drm_device *dev, int plane)
  410. {
  411.         struct drm_i915_private *dev_priv = dev->dev_private;
  412.         uint32_t dsparb = I915_READ(DSPARB);
  413.         int size;
  414.  
  415.         size = dsparb & 0x1ff;
  416.         if (plane)
  417.                 size = ((dsparb >> DSPARB_BEND_SHIFT) & 0x1ff) - size;
  418.         size >>= 1; /* Convert to cachelines */
  419.  
  420.         DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
  421.                       plane ? "B" : "A", size);
  422.  
  423.         return size;
  424. }
  425.  
  426. static int i845_get_fifo_size(struct drm_device *dev, int plane)
  427. {
  428.         struct drm_i915_private *dev_priv = dev->dev_private;
  429.         uint32_t dsparb = I915_READ(DSPARB);
  430.         int size;
  431.  
  432.         size = dsparb & 0x7f;
  433.         size >>= 2; /* Convert to cachelines */
  434.  
  435.         DRM_DEBUG_KMS("FIFO size - (0x%08x) %s: %d\n", dsparb,
  436.                       plane ? "B" : "A",
  437.                       size);
  438.  
  439.         return size;
  440. }
  441.  
  442. /* Pineview has different values for various configs */
  443. static const struct intel_watermark_params pineview_display_wm = {
  444.         .fifo_size = PINEVIEW_DISPLAY_FIFO,
  445.         .max_wm = PINEVIEW_MAX_WM,
  446.         .default_wm = PINEVIEW_DFT_WM,
  447.         .guard_size = PINEVIEW_GUARD_WM,
  448.         .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
  449. };
  450. static const struct intel_watermark_params pineview_display_hplloff_wm = {
  451.         .fifo_size = PINEVIEW_DISPLAY_FIFO,
  452.         .max_wm = PINEVIEW_MAX_WM,
  453.         .default_wm = PINEVIEW_DFT_HPLLOFF_WM,
  454.         .guard_size = PINEVIEW_GUARD_WM,
  455.         .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
  456. };
  457. static const struct intel_watermark_params pineview_cursor_wm = {
  458.         .fifo_size = PINEVIEW_CURSOR_FIFO,
  459.         .max_wm = PINEVIEW_CURSOR_MAX_WM,
  460.         .default_wm = PINEVIEW_CURSOR_DFT_WM,
  461.         .guard_size = PINEVIEW_CURSOR_GUARD_WM,
  462.         .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
  463. };
  464. static const struct intel_watermark_params pineview_cursor_hplloff_wm = {
  465.         .fifo_size = PINEVIEW_CURSOR_FIFO,
  466.         .max_wm = PINEVIEW_CURSOR_MAX_WM,
  467.         .default_wm = PINEVIEW_CURSOR_DFT_WM,
  468.         .guard_size = PINEVIEW_CURSOR_GUARD_WM,
  469.         .cacheline_size = PINEVIEW_FIFO_LINE_SIZE,
  470. };
  471. static const struct intel_watermark_params g4x_wm_info = {
  472.         .fifo_size = G4X_FIFO_SIZE,
  473.         .max_wm = G4X_MAX_WM,
  474.         .default_wm = G4X_MAX_WM,
  475.         .guard_size = 2,
  476.         .cacheline_size = G4X_FIFO_LINE_SIZE,
  477. };
  478. static const struct intel_watermark_params g4x_cursor_wm_info = {
  479.         .fifo_size = I965_CURSOR_FIFO,
  480.         .max_wm = I965_CURSOR_MAX_WM,
  481.         .default_wm = I965_CURSOR_DFT_WM,
  482.         .guard_size = 2,
  483.         .cacheline_size = G4X_FIFO_LINE_SIZE,
  484. };
  485. static const struct intel_watermark_params valleyview_wm_info = {
  486.         .fifo_size = VALLEYVIEW_FIFO_SIZE,
  487.         .max_wm = VALLEYVIEW_MAX_WM,
  488.         .default_wm = VALLEYVIEW_MAX_WM,
  489.         .guard_size = 2,
  490.         .cacheline_size = G4X_FIFO_LINE_SIZE,
  491. };
  492. static const struct intel_watermark_params valleyview_cursor_wm_info = {
  493.         .fifo_size = I965_CURSOR_FIFO,
  494.         .max_wm = VALLEYVIEW_CURSOR_MAX_WM,
  495.         .default_wm = I965_CURSOR_DFT_WM,
  496.         .guard_size = 2,
  497.         .cacheline_size = G4X_FIFO_LINE_SIZE,
  498. };
  499. static const struct intel_watermark_params i965_cursor_wm_info = {
  500.         .fifo_size = I965_CURSOR_FIFO,
  501.         .max_wm = I965_CURSOR_MAX_WM,
  502.         .default_wm = I965_CURSOR_DFT_WM,
  503.         .guard_size = 2,
  504.         .cacheline_size = I915_FIFO_LINE_SIZE,
  505. };
  506. static const struct intel_watermark_params i945_wm_info = {
  507.         .fifo_size = I945_FIFO_SIZE,
  508.         .max_wm = I915_MAX_WM,
  509.         .default_wm = 1,
  510.         .guard_size = 2,
  511.         .cacheline_size = I915_FIFO_LINE_SIZE,
  512. };
  513. static const struct intel_watermark_params i915_wm_info = {
  514.         .fifo_size = I915_FIFO_SIZE,
  515.         .max_wm = I915_MAX_WM,
  516.         .default_wm = 1,
  517.         .guard_size = 2,
  518.         .cacheline_size = I915_FIFO_LINE_SIZE,
  519. };
  520. static const struct intel_watermark_params i830_a_wm_info = {
  521.         .fifo_size = I855GM_FIFO_SIZE,
  522.         .max_wm = I915_MAX_WM,
  523.         .default_wm = 1,
  524.         .guard_size = 2,
  525.         .cacheline_size = I830_FIFO_LINE_SIZE,
  526. };
  527. static const struct intel_watermark_params i830_bc_wm_info = {
  528.         .fifo_size = I855GM_FIFO_SIZE,
  529.         .max_wm = I915_MAX_WM/2,
  530.         .default_wm = 1,
  531.         .guard_size = 2,
  532.         .cacheline_size = I830_FIFO_LINE_SIZE,
  533. };
  534. static const struct intel_watermark_params i845_wm_info = {
  535.         .fifo_size = I830_FIFO_SIZE,
  536.         .max_wm = I915_MAX_WM,
  537.         .default_wm = 1,
  538.         .guard_size = 2,
  539.         .cacheline_size = I830_FIFO_LINE_SIZE,
  540. };
  541.  
  542. /**
  543.  * intel_calculate_wm - calculate watermark level
  544.  * @clock_in_khz: pixel clock
  545.  * @wm: chip FIFO params
  546.  * @pixel_size: display pixel size
  547.  * @latency_ns: memory latency for the platform
  548.  *
  549.  * Calculate the watermark level (the level at which the display plane will
  550.  * start fetching from memory again).  Each chip has a different display
  551.  * FIFO size and allocation, so the caller needs to figure that out and pass
  552.  * in the correct intel_watermark_params structure.
  553.  *
  554.  * As the pixel clock runs, the FIFO will be drained at a rate that depends
  555.  * on the pixel size.  When it reaches the watermark level, it'll start
  556.  * fetching FIFO line sized based chunks from memory until the FIFO fills
  557.  * past the watermark point.  If the FIFO drains completely, a FIFO underrun
  558.  * will occur, and a display engine hang could result.
  559.  */
  560. static unsigned long intel_calculate_wm(unsigned long clock_in_khz,
  561.                                         const struct intel_watermark_params *wm,
  562.                                         int fifo_size,
  563.                                         int pixel_size,
  564.                                         unsigned long latency_ns)
  565. {
  566.         long entries_required, wm_size;
  567.  
  568.         /*
  569.          * Note: we need to make sure we don't overflow for various clock &
  570.          * latency values.
  571.          * clocks go from a few thousand to several hundred thousand.
  572.          * latency is usually a few thousand
  573.          */
  574.         entries_required = ((clock_in_khz / 1000) * pixel_size * latency_ns) /
  575.                 1000;
  576.         entries_required = DIV_ROUND_UP(entries_required, wm->cacheline_size);
  577.  
  578.         DRM_DEBUG_KMS("FIFO entries required for mode: %ld\n", entries_required);
  579.  
  580.         wm_size = fifo_size - (entries_required + wm->guard_size);
  581.  
  582.         DRM_DEBUG_KMS("FIFO watermark level: %ld\n", wm_size);
  583.  
  584.         /* Don't promote wm_size to unsigned... */
  585.         if (wm_size > (long)wm->max_wm)
  586.                 wm_size = wm->max_wm;
  587.         if (wm_size <= 0)
  588.                 wm_size = wm->default_wm;
  589.  
  590.         /*
  591.          * Bspec seems to indicate that the value shouldn't be lower than
  592.          * 'burst size + 1'. Certainly 830 is quite unhappy with low values.
  593.          * Lets go for 8 which is the burst size since certain platforms
  594.          * already use a hardcoded 8 (which is what the spec says should be
  595.          * done).
  596.          */
  597.         if (wm_size <= 8)
  598.                 wm_size = 8;
  599.  
  600.         return wm_size;
  601. }
  602.  
  603. static struct drm_crtc *single_enabled_crtc(struct drm_device *dev)
  604. {
  605.         struct drm_crtc *crtc, *enabled = NULL;
  606.  
  607.         for_each_crtc(dev, crtc) {
  608.                 if (intel_crtc_active(crtc)) {
  609.                         if (enabled)
  610.                                 return NULL;
  611.                         enabled = crtc;
  612.                 }
  613.         }
  614.  
  615.         return enabled;
  616. }
  617.  
  618. static void pineview_update_wm(struct drm_crtc *unused_crtc)
  619. {
  620.         struct drm_device *dev = unused_crtc->dev;
  621.         struct drm_i915_private *dev_priv = dev->dev_private;
  622.         struct drm_crtc *crtc;
  623.         const struct cxsr_latency *latency;
  624.         u32 reg;
  625.         unsigned long wm;
  626.  
  627.         latency = intel_get_cxsr_latency(IS_PINEVIEW_G(dev), dev_priv->is_ddr3,
  628.                                          dev_priv->fsb_freq, dev_priv->mem_freq);
  629.         if (!latency) {
  630.                 DRM_DEBUG_KMS("Unknown FSB/MEM found, disable CxSR\n");
  631.                 intel_set_memory_cxsr(dev_priv, false);
  632.                 return;
  633.         }
  634.  
  635.         crtc = single_enabled_crtc(dev);
  636.         if (crtc) {
  637.                 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
  638.                 int pixel_size = crtc->primary->state->fb->bits_per_pixel / 8;
  639.                 int clock = adjusted_mode->crtc_clock;
  640.  
  641.                 /* Display SR */
  642.                 wm = intel_calculate_wm(clock, &pineview_display_wm,
  643.                                         pineview_display_wm.fifo_size,
  644.                                         pixel_size, latency->display_sr);
  645.                 reg = I915_READ(DSPFW1);
  646.                 reg &= ~DSPFW_SR_MASK;
  647.                 reg |= FW_WM(wm, SR);
  648.                 I915_WRITE(DSPFW1, reg);
  649.                 DRM_DEBUG_KMS("DSPFW1 register is %x\n", reg);
  650.  
  651.                 /* cursor SR */
  652.                 wm = intel_calculate_wm(clock, &pineview_cursor_wm,
  653.                                         pineview_display_wm.fifo_size,
  654.                                         pixel_size, latency->cursor_sr);
  655.                 reg = I915_READ(DSPFW3);
  656.                 reg &= ~DSPFW_CURSOR_SR_MASK;
  657.                 reg |= FW_WM(wm, CURSOR_SR);
  658.                 I915_WRITE(DSPFW3, reg);
  659.  
  660.                 /* Display HPLL off SR */
  661.                 wm = intel_calculate_wm(clock, &pineview_display_hplloff_wm,
  662.                                         pineview_display_hplloff_wm.fifo_size,
  663.                                         pixel_size, latency->display_hpll_disable);
  664.                 reg = I915_READ(DSPFW3);
  665.                 reg &= ~DSPFW_HPLL_SR_MASK;
  666.                 reg |= FW_WM(wm, HPLL_SR);
  667.                 I915_WRITE(DSPFW3, reg);
  668.  
  669.                 /* cursor HPLL off SR */
  670.                 wm = intel_calculate_wm(clock, &pineview_cursor_hplloff_wm,
  671.                                         pineview_display_hplloff_wm.fifo_size,
  672.                                         pixel_size, latency->cursor_hpll_disable);
  673.                 reg = I915_READ(DSPFW3);
  674.                 reg &= ~DSPFW_HPLL_CURSOR_MASK;
  675.                 reg |= FW_WM(wm, HPLL_CURSOR);
  676.                 I915_WRITE(DSPFW3, reg);
  677.                 DRM_DEBUG_KMS("DSPFW3 register is %x\n", reg);
  678.  
  679.                 intel_set_memory_cxsr(dev_priv, true);
  680.         } else {
  681.                 intel_set_memory_cxsr(dev_priv, false);
  682.         }
  683. }
  684.  
  685. static bool g4x_compute_wm0(struct drm_device *dev,
  686.                             int plane,
  687.                             const struct intel_watermark_params *display,
  688.                             int display_latency_ns,
  689.                             const struct intel_watermark_params *cursor,
  690.                             int cursor_latency_ns,
  691.                             int *plane_wm,
  692.                             int *cursor_wm)
  693. {
  694.         struct drm_crtc *crtc;
  695.         const struct drm_display_mode *adjusted_mode;
  696.         int htotal, hdisplay, clock, pixel_size;
  697.         int line_time_us, line_count;
  698.         int entries, tlb_miss;
  699.  
  700.         crtc = intel_get_crtc_for_plane(dev, plane);
  701.         if (!intel_crtc_active(crtc)) {
  702.                 *cursor_wm = cursor->guard_size;
  703.                 *plane_wm = display->guard_size;
  704.                 return false;
  705.         }
  706.  
  707.         adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
  708.         clock = adjusted_mode->crtc_clock;
  709.         htotal = adjusted_mode->crtc_htotal;
  710.         hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
  711.         pixel_size = crtc->primary->state->fb->bits_per_pixel / 8;
  712.  
  713.         /* Use the small buffer method to calculate plane watermark */
  714.         entries = ((clock * pixel_size / 1000) * display_latency_ns) / 1000;
  715.         tlb_miss = display->fifo_size*display->cacheline_size - hdisplay * 8;
  716.         if (tlb_miss > 0)
  717.                 entries += tlb_miss;
  718.         entries = DIV_ROUND_UP(entries, display->cacheline_size);
  719.         *plane_wm = entries + display->guard_size;
  720.         if (*plane_wm > (int)display->max_wm)
  721.                 *plane_wm = display->max_wm;
  722.  
  723.         /* Use the large buffer method to calculate cursor watermark */
  724.         line_time_us = max(htotal * 1000 / clock, 1);
  725.         line_count = (cursor_latency_ns / line_time_us + 1000) / 1000;
  726.         entries = line_count * crtc->cursor->state->crtc_w * pixel_size;
  727.         tlb_miss = cursor->fifo_size*cursor->cacheline_size - hdisplay * 8;
  728.         if (tlb_miss > 0)
  729.                 entries += tlb_miss;
  730.         entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
  731.         *cursor_wm = entries + cursor->guard_size;
  732.         if (*cursor_wm > (int)cursor->max_wm)
  733.                 *cursor_wm = (int)cursor->max_wm;
  734.  
  735.         return true;
  736. }
  737.  
  738. /*
  739.  * Check the wm result.
  740.  *
  741.  * If any calculated watermark values is larger than the maximum value that
  742.  * can be programmed into the associated watermark register, that watermark
  743.  * must be disabled.
  744.  */
  745. static bool g4x_check_srwm(struct drm_device *dev,
  746.                            int display_wm, int cursor_wm,
  747.                            const struct intel_watermark_params *display,
  748.                            const struct intel_watermark_params *cursor)
  749. {
  750.         DRM_DEBUG_KMS("SR watermark: display plane %d, cursor %d\n",
  751.                       display_wm, cursor_wm);
  752.  
  753.         if (display_wm > display->max_wm) {
  754.                 DRM_DEBUG_KMS("display watermark is too large(%d/%ld), disabling\n",
  755.                               display_wm, display->max_wm);
  756.                 return false;
  757.         }
  758.  
  759.         if (cursor_wm > cursor->max_wm) {
  760.                 DRM_DEBUG_KMS("cursor watermark is too large(%d/%ld), disabling\n",
  761.                               cursor_wm, cursor->max_wm);
  762.                 return false;
  763.         }
  764.  
  765.         if (!(display_wm || cursor_wm)) {
  766.                 DRM_DEBUG_KMS("SR latency is 0, disabling\n");
  767.                 return false;
  768.         }
  769.  
  770.         return true;
  771. }
  772.  
  773. static bool g4x_compute_srwm(struct drm_device *dev,
  774.                              int plane,
  775.                              int latency_ns,
  776.                              const struct intel_watermark_params *display,
  777.                              const struct intel_watermark_params *cursor,
  778.                              int *display_wm, int *cursor_wm)
  779. {
  780.         struct drm_crtc *crtc;
  781.         const struct drm_display_mode *adjusted_mode;
  782.         int hdisplay, htotal, pixel_size, clock;
  783.         unsigned long line_time_us;
  784.         int line_count, line_size;
  785.         int small, large;
  786.         int entries;
  787.  
  788.         if (!latency_ns) {
  789.                 *display_wm = *cursor_wm = 0;
  790.                 return false;
  791.         }
  792.  
  793.         crtc = intel_get_crtc_for_plane(dev, plane);
  794.         adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
  795.         clock = adjusted_mode->crtc_clock;
  796.         htotal = adjusted_mode->crtc_htotal;
  797.         hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
  798.         pixel_size = crtc->primary->state->fb->bits_per_pixel / 8;
  799.  
  800.         line_time_us = max(htotal * 1000 / clock, 1);
  801.         line_count = (latency_ns / line_time_us + 1000) / 1000;
  802.         line_size = hdisplay * pixel_size;
  803.  
  804.         /* Use the minimum of the small and large buffer method for primary */
  805.         small = ((clock * pixel_size / 1000) * latency_ns) / 1000;
  806.         large = line_count * line_size;
  807.  
  808.         entries = DIV_ROUND_UP(min(small, large), display->cacheline_size);
  809.         *display_wm = entries + display->guard_size;
  810.  
  811.         /* calculate the self-refresh watermark for display cursor */
  812.         entries = line_count * pixel_size * crtc->cursor->state->crtc_w;
  813.         entries = DIV_ROUND_UP(entries, cursor->cacheline_size);
  814.         *cursor_wm = entries + cursor->guard_size;
  815.  
  816.         return g4x_check_srwm(dev,
  817.                               *display_wm, *cursor_wm,
  818.                               display, cursor);
  819. }
  820.  
  821. #define FW_WM_VLV(value, plane) \
  822.         (((value) << DSPFW_ ## plane ## _SHIFT) & DSPFW_ ## plane ## _MASK_VLV)
  823.  
  824. static void vlv_write_wm_values(struct intel_crtc *crtc,
  825.                                 const struct vlv_wm_values *wm)
  826. {
  827.         struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
  828.         enum pipe pipe = crtc->pipe;
  829.  
  830.         I915_WRITE(VLV_DDL(pipe),
  831.                    (wm->ddl[pipe].cursor << DDL_CURSOR_SHIFT) |
  832.                    (wm->ddl[pipe].sprite[1] << DDL_SPRITE_SHIFT(1)) |
  833.                    (wm->ddl[pipe].sprite[0] << DDL_SPRITE_SHIFT(0)) |
  834.                    (wm->ddl[pipe].primary << DDL_PLANE_SHIFT));
  835.  
  836.         I915_WRITE(DSPFW1,
  837.                    FW_WM(wm->sr.plane, SR) |
  838.                    FW_WM(wm->pipe[PIPE_B].cursor, CURSORB) |
  839.                    FW_WM_VLV(wm->pipe[PIPE_B].primary, PLANEB) |
  840.                    FW_WM_VLV(wm->pipe[PIPE_A].primary, PLANEA));
  841.         I915_WRITE(DSPFW2,
  842.                    FW_WM_VLV(wm->pipe[PIPE_A].sprite[1], SPRITEB) |
  843.                    FW_WM(wm->pipe[PIPE_A].cursor, CURSORA) |
  844.                    FW_WM_VLV(wm->pipe[PIPE_A].sprite[0], SPRITEA));
  845.         I915_WRITE(DSPFW3,
  846.                    FW_WM(wm->sr.cursor, CURSOR_SR));
  847.  
  848.         if (IS_CHERRYVIEW(dev_priv)) {
  849.                 I915_WRITE(DSPFW7_CHV,
  850.                            FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) |
  851.                            FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC));
  852.                 I915_WRITE(DSPFW8_CHV,
  853.                            FW_WM_VLV(wm->pipe[PIPE_C].sprite[1], SPRITEF) |
  854.                            FW_WM_VLV(wm->pipe[PIPE_C].sprite[0], SPRITEE));
  855.                 I915_WRITE(DSPFW9_CHV,
  856.                            FW_WM_VLV(wm->pipe[PIPE_C].primary, PLANEC) |
  857.                            FW_WM(wm->pipe[PIPE_C].cursor, CURSORC));
  858.                 I915_WRITE(DSPHOWM,
  859.                            FW_WM(wm->sr.plane >> 9, SR_HI) |
  860.                            FW_WM(wm->pipe[PIPE_C].sprite[1] >> 8, SPRITEF_HI) |
  861.                            FW_WM(wm->pipe[PIPE_C].sprite[0] >> 8, SPRITEE_HI) |
  862.                            FW_WM(wm->pipe[PIPE_C].primary >> 8, PLANEC_HI) |
  863.                            FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) |
  864.                            FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) |
  865.                            FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) |
  866.                            FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) |
  867.                            FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) |
  868.                            FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI));
  869.         } else {
  870.                 I915_WRITE(DSPFW7,
  871.                            FW_WM_VLV(wm->pipe[PIPE_B].sprite[1], SPRITED) |
  872.                            FW_WM_VLV(wm->pipe[PIPE_B].sprite[0], SPRITEC));
  873.                 I915_WRITE(DSPHOWM,
  874.                            FW_WM(wm->sr.plane >> 9, SR_HI) |
  875.                            FW_WM(wm->pipe[PIPE_B].sprite[1] >> 8, SPRITED_HI) |
  876.                            FW_WM(wm->pipe[PIPE_B].sprite[0] >> 8, SPRITEC_HI) |
  877.                            FW_WM(wm->pipe[PIPE_B].primary >> 8, PLANEB_HI) |
  878.                            FW_WM(wm->pipe[PIPE_A].sprite[1] >> 8, SPRITEB_HI) |
  879.                            FW_WM(wm->pipe[PIPE_A].sprite[0] >> 8, SPRITEA_HI) |
  880.                            FW_WM(wm->pipe[PIPE_A].primary >> 8, PLANEA_HI));
  881.         }
  882.  
  883.         /* zero (unused) WM1 watermarks */
  884.         I915_WRITE(DSPFW4, 0);
  885.         I915_WRITE(DSPFW5, 0);
  886.         I915_WRITE(DSPFW6, 0);
  887.         I915_WRITE(DSPHOWM1, 0);
  888.  
  889.         POSTING_READ(DSPFW1);
  890. }
  891.  
  892. #undef FW_WM_VLV
  893.  
  894. enum vlv_wm_level {
  895.         VLV_WM_LEVEL_PM2,
  896.         VLV_WM_LEVEL_PM5,
  897.         VLV_WM_LEVEL_DDR_DVFS,
  898. };
  899.  
  900. /* latency must be in 0.1us units. */
  901. static unsigned int vlv_wm_method2(unsigned int pixel_rate,
  902.                                    unsigned int pipe_htotal,
  903.                                    unsigned int horiz_pixels,
  904.                                    unsigned int bytes_per_pixel,
  905.                                    unsigned int latency)
  906. {
  907.         unsigned int ret;
  908.  
  909.         ret = (latency * pixel_rate) / (pipe_htotal * 10000);
  910.         ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
  911.         ret = DIV_ROUND_UP(ret, 64);
  912.  
  913.         return ret;
  914. }
  915.  
  916. static void vlv_setup_wm_latency(struct drm_device *dev)
  917. {
  918.         struct drm_i915_private *dev_priv = dev->dev_private;
  919.  
  920.         /* all latencies in usec */
  921.         dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM2] = 3;
  922.  
  923.         dev_priv->wm.max_level = VLV_WM_LEVEL_PM2;
  924.  
  925.         if (IS_CHERRYVIEW(dev_priv)) {
  926.                 dev_priv->wm.pri_latency[VLV_WM_LEVEL_PM5] = 12;
  927.                 dev_priv->wm.pri_latency[VLV_WM_LEVEL_DDR_DVFS] = 33;
  928.  
  929.                 dev_priv->wm.max_level = VLV_WM_LEVEL_DDR_DVFS;
  930.         }
  931. }
  932.  
  933. static uint16_t vlv_compute_wm_level(struct intel_plane *plane,
  934.                                      struct intel_crtc *crtc,
  935.                                      const struct intel_plane_state *state,
  936.                                      int level)
  937. {
  938.         struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
  939.         int clock, htotal, pixel_size, width, wm;
  940.  
  941.         if (dev_priv->wm.pri_latency[level] == 0)
  942.                 return USHRT_MAX;
  943.  
  944.         if (!state->visible)
  945.                 return 0;
  946.  
  947.         pixel_size = drm_format_plane_cpp(state->base.fb->pixel_format, 0);
  948.         clock = crtc->config->base.adjusted_mode.crtc_clock;
  949.         htotal = crtc->config->base.adjusted_mode.crtc_htotal;
  950.         width = crtc->config->pipe_src_w;
  951.         if (WARN_ON(htotal == 0))
  952.                 htotal = 1;
  953.  
  954.         if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
  955.                 /*
  956.                  * FIXME the formula gives values that are
  957.                  * too big for the cursor FIFO, and hence we
  958.                  * would never be able to use cursors. For
  959.                  * now just hardcode the watermark.
  960.                  */
  961.                 wm = 63;
  962.         } else {
  963.                 wm = vlv_wm_method2(clock, htotal, width, pixel_size,
  964.                                     dev_priv->wm.pri_latency[level] * 10);
  965.         }
  966.  
  967.         return min_t(int, wm, USHRT_MAX);
  968. }
  969.  
  970. static void vlv_compute_fifo(struct intel_crtc *crtc)
  971. {
  972.         struct drm_device *dev = crtc->base.dev;
  973.         struct vlv_wm_state *wm_state = &crtc->wm_state;
  974.         struct intel_plane *plane;
  975.         unsigned int total_rate = 0;
  976.         const int fifo_size = 512 - 1;
  977.         int fifo_extra, fifo_left = fifo_size;
  978.  
  979.         for_each_intel_plane_on_crtc(dev, crtc, plane) {
  980.                 struct intel_plane_state *state =
  981.                         to_intel_plane_state(plane->base.state);
  982.  
  983.                 if (plane->base.type == DRM_PLANE_TYPE_CURSOR)
  984.                         continue;
  985.  
  986.                 if (state->visible) {
  987.                         wm_state->num_active_planes++;
  988.                         total_rate += drm_format_plane_cpp(state->base.fb->pixel_format, 0);
  989.                 }
  990.         }
  991.  
  992.         for_each_intel_plane_on_crtc(dev, crtc, plane) {
  993.                 struct intel_plane_state *state =
  994.                         to_intel_plane_state(plane->base.state);
  995.                 unsigned int rate;
  996.  
  997.                 if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
  998.                         plane->wm.fifo_size = 63;
  999.                         continue;
  1000.                 }
  1001.  
  1002.                 if (!state->visible) {
  1003.                         plane->wm.fifo_size = 0;
  1004.                         continue;
  1005.                 }
  1006.  
  1007.                 rate = drm_format_plane_cpp(state->base.fb->pixel_format, 0);
  1008.                 plane->wm.fifo_size = fifo_size * rate / total_rate;
  1009.                 fifo_left -= plane->wm.fifo_size;
  1010.         }
  1011.  
  1012.         fifo_extra = DIV_ROUND_UP(fifo_left, wm_state->num_active_planes ?: 1);
  1013.  
  1014.         /* spread the remainder evenly */
  1015.         for_each_intel_plane_on_crtc(dev, crtc, plane) {
  1016.                 int plane_extra;
  1017.  
  1018.                 if (fifo_left == 0)
  1019.                         break;
  1020.  
  1021.                 if (plane->base.type == DRM_PLANE_TYPE_CURSOR)
  1022.                         continue;
  1023.  
  1024.                 /* give it all to the first plane if none are active */
  1025.                 if (plane->wm.fifo_size == 0 &&
  1026.                     wm_state->num_active_planes)
  1027.                         continue;
  1028.  
  1029.                 plane_extra = min(fifo_extra, fifo_left);
  1030.                 plane->wm.fifo_size += plane_extra;
  1031.                 fifo_left -= plane_extra;
  1032.         }
  1033.  
  1034.         WARN_ON(fifo_left != 0);
  1035. }
  1036.  
  1037. static void vlv_invert_wms(struct intel_crtc *crtc)
  1038. {
  1039.         struct vlv_wm_state *wm_state = &crtc->wm_state;
  1040.         int level;
  1041.  
  1042.         for (level = 0; level < wm_state->num_levels; level++) {
  1043.                 struct drm_device *dev = crtc->base.dev;
  1044.                 const int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1;
  1045.                 struct intel_plane *plane;
  1046.  
  1047.                 wm_state->sr[level].plane = sr_fifo_size - wm_state->sr[level].plane;
  1048.                 wm_state->sr[level].cursor = 63 - wm_state->sr[level].cursor;
  1049.  
  1050.                 for_each_intel_plane_on_crtc(dev, crtc, plane) {
  1051.                         switch (plane->base.type) {
  1052.                                 int sprite;
  1053.                         case DRM_PLANE_TYPE_CURSOR:
  1054.                                 wm_state->wm[level].cursor = plane->wm.fifo_size -
  1055.                                         wm_state->wm[level].cursor;
  1056.                                 break;
  1057.                         case DRM_PLANE_TYPE_PRIMARY:
  1058.                                 wm_state->wm[level].primary = plane->wm.fifo_size -
  1059.                                         wm_state->wm[level].primary;
  1060.                                 break;
  1061.                         case DRM_PLANE_TYPE_OVERLAY:
  1062.                                 sprite = plane->plane;
  1063.                                 wm_state->wm[level].sprite[sprite] = plane->wm.fifo_size -
  1064.                                         wm_state->wm[level].sprite[sprite];
  1065.                                 break;
  1066.                         }
  1067.                 }
  1068.         }
  1069. }
  1070.  
  1071. static void vlv_compute_wm(struct intel_crtc *crtc)
  1072. {
  1073.         struct drm_device *dev = crtc->base.dev;
  1074.         struct vlv_wm_state *wm_state = &crtc->wm_state;
  1075.         struct intel_plane *plane;
  1076.         int sr_fifo_size = INTEL_INFO(dev)->num_pipes * 512 - 1;
  1077.         int level;
  1078.  
  1079.         memset(wm_state, 0, sizeof(*wm_state));
  1080.  
  1081.         wm_state->cxsr = crtc->pipe != PIPE_C && crtc->wm.cxsr_allowed;
  1082.         wm_state->num_levels = to_i915(dev)->wm.max_level + 1;
  1083.  
  1084.         wm_state->num_active_planes = 0;
  1085.  
  1086.         vlv_compute_fifo(crtc);
  1087.  
  1088.         if (wm_state->num_active_planes != 1)
  1089.                 wm_state->cxsr = false;
  1090.  
  1091.         if (wm_state->cxsr) {
  1092.                 for (level = 0; level < wm_state->num_levels; level++) {
  1093.                         wm_state->sr[level].plane = sr_fifo_size;
  1094.                         wm_state->sr[level].cursor = 63;
  1095.                 }
  1096.         }
  1097.  
  1098.         for_each_intel_plane_on_crtc(dev, crtc, plane) {
  1099.                 struct intel_plane_state *state =
  1100.                         to_intel_plane_state(plane->base.state);
  1101.  
  1102.                 if (!state->visible)
  1103.                         continue;
  1104.  
  1105.                 /* normal watermarks */
  1106.                 for (level = 0; level < wm_state->num_levels; level++) {
  1107.                         int wm = vlv_compute_wm_level(plane, crtc, state, level);
  1108.                         int max_wm = plane->base.type == DRM_PLANE_TYPE_CURSOR ? 63 : 511;
  1109.  
  1110.                         /* hack */
  1111.                         if (WARN_ON(level == 0 && wm > max_wm))
  1112.                                 wm = max_wm;
  1113.  
  1114.                         if (wm > plane->wm.fifo_size)
  1115.                                 break;
  1116.  
  1117.                         switch (plane->base.type) {
  1118.                                 int sprite;
  1119.                         case DRM_PLANE_TYPE_CURSOR:
  1120.                                 wm_state->wm[level].cursor = wm;
  1121.                                 break;
  1122.                         case DRM_PLANE_TYPE_PRIMARY:
  1123.                                 wm_state->wm[level].primary = wm;
  1124.                                 break;
  1125.                         case DRM_PLANE_TYPE_OVERLAY:
  1126.                                 sprite = plane->plane;
  1127.                                 wm_state->wm[level].sprite[sprite] = wm;
  1128.                                 break;
  1129.                         }
  1130.                 }
  1131.  
  1132.                 wm_state->num_levels = level;
  1133.  
  1134.                 if (!wm_state->cxsr)
  1135.                         continue;
  1136.  
  1137.                 /* maxfifo watermarks */
  1138.                 switch (plane->base.type) {
  1139.                         int sprite, level;
  1140.                 case DRM_PLANE_TYPE_CURSOR:
  1141.                         for (level = 0; level < wm_state->num_levels; level++)
  1142.                                 wm_state->sr[level].cursor =
  1143.                                         wm_state->wm[level].cursor;
  1144.                         break;
  1145.                 case DRM_PLANE_TYPE_PRIMARY:
  1146.                         for (level = 0; level < wm_state->num_levels; level++)
  1147.                                 wm_state->sr[level].plane =
  1148.                                         min(wm_state->sr[level].plane,
  1149.                                             wm_state->wm[level].primary);
  1150.                         break;
  1151.                 case DRM_PLANE_TYPE_OVERLAY:
  1152.                         sprite = plane->plane;
  1153.                         for (level = 0; level < wm_state->num_levels; level++)
  1154.                                 wm_state->sr[level].plane =
  1155.                                         min(wm_state->sr[level].plane,
  1156.                                             wm_state->wm[level].sprite[sprite]);
  1157.                         break;
  1158.                 }
  1159.         }
  1160.  
  1161.         /* clear any (partially) filled invalid levels */
  1162.         for (level = wm_state->num_levels; level < to_i915(dev)->wm.max_level + 1; level++) {
  1163.                 memset(&wm_state->wm[level], 0, sizeof(wm_state->wm[level]));
  1164.                 memset(&wm_state->sr[level], 0, sizeof(wm_state->sr[level]));
  1165.         }
  1166.  
  1167.         vlv_invert_wms(crtc);
  1168. }
  1169.  
  1170. #define VLV_FIFO(plane, value) \
  1171.         (((value) << DSPARB_ ## plane ## _SHIFT_VLV) & DSPARB_ ## plane ## _MASK_VLV)
  1172.  
  1173. static void vlv_pipe_set_fifo_size(struct intel_crtc *crtc)
  1174. {
  1175.         struct drm_device *dev = crtc->base.dev;
  1176.         struct drm_i915_private *dev_priv = to_i915(dev);
  1177.         struct intel_plane *plane;
  1178.         int sprite0_start = 0, sprite1_start = 0, fifo_size = 0;
  1179.  
  1180.         for_each_intel_plane_on_crtc(dev, crtc, plane) {
  1181.                 if (plane->base.type == DRM_PLANE_TYPE_CURSOR) {
  1182.                         WARN_ON(plane->wm.fifo_size != 63);
  1183.                         continue;
  1184.                 }
  1185.  
  1186.                 if (plane->base.type == DRM_PLANE_TYPE_PRIMARY)
  1187.                         sprite0_start = plane->wm.fifo_size;
  1188.                 else if (plane->plane == 0)
  1189.                         sprite1_start = sprite0_start + plane->wm.fifo_size;
  1190.                 else
  1191.                         fifo_size = sprite1_start + plane->wm.fifo_size;
  1192.         }
  1193.  
  1194.         WARN_ON(fifo_size != 512 - 1);
  1195.  
  1196.         DRM_DEBUG_KMS("Pipe %c FIFO split %d / %d / %d\n",
  1197.                       pipe_name(crtc->pipe), sprite0_start,
  1198.                       sprite1_start, fifo_size);
  1199.  
  1200.         switch (crtc->pipe) {
  1201.                 uint32_t dsparb, dsparb2, dsparb3;
  1202.         case PIPE_A:
  1203.                 dsparb = I915_READ(DSPARB);
  1204.                 dsparb2 = I915_READ(DSPARB2);
  1205.  
  1206.                 dsparb &= ~(VLV_FIFO(SPRITEA, 0xff) |
  1207.                             VLV_FIFO(SPRITEB, 0xff));
  1208.                 dsparb |= (VLV_FIFO(SPRITEA, sprite0_start) |
  1209.                            VLV_FIFO(SPRITEB, sprite1_start));
  1210.  
  1211.                 dsparb2 &= ~(VLV_FIFO(SPRITEA_HI, 0x1) |
  1212.                              VLV_FIFO(SPRITEB_HI, 0x1));
  1213.                 dsparb2 |= (VLV_FIFO(SPRITEA_HI, sprite0_start >> 8) |
  1214.                            VLV_FIFO(SPRITEB_HI, sprite1_start >> 8));
  1215.  
  1216.                 I915_WRITE(DSPARB, dsparb);
  1217.                 I915_WRITE(DSPARB2, dsparb2);
  1218.                 break;
  1219.         case PIPE_B:
  1220.                 dsparb = I915_READ(DSPARB);
  1221.                 dsparb2 = I915_READ(DSPARB2);
  1222.  
  1223.                 dsparb &= ~(VLV_FIFO(SPRITEC, 0xff) |
  1224.                             VLV_FIFO(SPRITED, 0xff));
  1225.                 dsparb |= (VLV_FIFO(SPRITEC, sprite0_start) |
  1226.                            VLV_FIFO(SPRITED, sprite1_start));
  1227.  
  1228.                 dsparb2 &= ~(VLV_FIFO(SPRITEC_HI, 0xff) |
  1229.                              VLV_FIFO(SPRITED_HI, 0xff));
  1230.                 dsparb2 |= (VLV_FIFO(SPRITEC_HI, sprite0_start >> 8) |
  1231.                            VLV_FIFO(SPRITED_HI, sprite1_start >> 8));
  1232.  
  1233.                 I915_WRITE(DSPARB, dsparb);
  1234.                 I915_WRITE(DSPARB2, dsparb2);
  1235.                 break;
  1236.         case PIPE_C:
  1237.                 dsparb3 = I915_READ(DSPARB3);
  1238.                 dsparb2 = I915_READ(DSPARB2);
  1239.  
  1240.                 dsparb3 &= ~(VLV_FIFO(SPRITEE, 0xff) |
  1241.                              VLV_FIFO(SPRITEF, 0xff));
  1242.                 dsparb3 |= (VLV_FIFO(SPRITEE, sprite0_start) |
  1243.                             VLV_FIFO(SPRITEF, sprite1_start));
  1244.  
  1245.                 dsparb2 &= ~(VLV_FIFO(SPRITEE_HI, 0xff) |
  1246.                              VLV_FIFO(SPRITEF_HI, 0xff));
  1247.                 dsparb2 |= (VLV_FIFO(SPRITEE_HI, sprite0_start >> 8) |
  1248.                            VLV_FIFO(SPRITEF_HI, sprite1_start >> 8));
  1249.  
  1250.                 I915_WRITE(DSPARB3, dsparb3);
  1251.                 I915_WRITE(DSPARB2, dsparb2);
  1252.                 break;
  1253.         default:
  1254.                 break;
  1255.         }
  1256. }
  1257.  
  1258. #undef VLV_FIFO
  1259.  
  1260. static void vlv_merge_wm(struct drm_device *dev,
  1261.                          struct vlv_wm_values *wm)
  1262. {
  1263.         struct intel_crtc *crtc;
  1264.         int num_active_crtcs = 0;
  1265.  
  1266.         wm->level = to_i915(dev)->wm.max_level;
  1267.         wm->cxsr = true;
  1268.  
  1269.         for_each_intel_crtc(dev, crtc) {
  1270.                 const struct vlv_wm_state *wm_state = &crtc->wm_state;
  1271.  
  1272.                 if (!crtc->active)
  1273.                         continue;
  1274.  
  1275.                 if (!wm_state->cxsr)
  1276.                         wm->cxsr = false;
  1277.  
  1278.                 num_active_crtcs++;
  1279.                 wm->level = min_t(int, wm->level, wm_state->num_levels - 1);
  1280.         }
  1281.  
  1282.         if (num_active_crtcs != 1)
  1283.                 wm->cxsr = false;
  1284.  
  1285.         if (num_active_crtcs > 1)
  1286.                 wm->level = VLV_WM_LEVEL_PM2;
  1287.  
  1288.         for_each_intel_crtc(dev, crtc) {
  1289.                 struct vlv_wm_state *wm_state = &crtc->wm_state;
  1290.                 enum pipe pipe = crtc->pipe;
  1291.  
  1292.                 if (!crtc->active)
  1293.                         continue;
  1294.  
  1295.                 wm->pipe[pipe] = wm_state->wm[wm->level];
  1296.                 if (wm->cxsr)
  1297.                         wm->sr = wm_state->sr[wm->level];
  1298.  
  1299.                 wm->ddl[pipe].primary = DDL_PRECISION_HIGH | 2;
  1300.                 wm->ddl[pipe].sprite[0] = DDL_PRECISION_HIGH | 2;
  1301.                 wm->ddl[pipe].sprite[1] = DDL_PRECISION_HIGH | 2;
  1302.                 wm->ddl[pipe].cursor = DDL_PRECISION_HIGH | 2;
  1303.         }
  1304. }
  1305.  
  1306. static void vlv_update_wm(struct drm_crtc *crtc)
  1307. {
  1308.         struct drm_device *dev = crtc->dev;
  1309.         struct drm_i915_private *dev_priv = dev->dev_private;
  1310.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  1311.         enum pipe pipe = intel_crtc->pipe;
  1312.         struct vlv_wm_values wm = {};
  1313.  
  1314.         vlv_compute_wm(intel_crtc);
  1315.         vlv_merge_wm(dev, &wm);
  1316.  
  1317.         if (memcmp(&dev_priv->wm.vlv, &wm, sizeof(wm)) == 0) {
  1318.                 /* FIXME should be part of crtc atomic commit */
  1319.                 vlv_pipe_set_fifo_size(intel_crtc);
  1320.                 return;
  1321.         }
  1322.  
  1323.         if (wm.level < VLV_WM_LEVEL_DDR_DVFS &&
  1324.             dev_priv->wm.vlv.level >= VLV_WM_LEVEL_DDR_DVFS)
  1325.                 chv_set_memory_dvfs(dev_priv, false);
  1326.  
  1327.         if (wm.level < VLV_WM_LEVEL_PM5 &&
  1328.             dev_priv->wm.vlv.level >= VLV_WM_LEVEL_PM5)
  1329.                 chv_set_memory_pm5(dev_priv, false);
  1330.  
  1331.         if (!wm.cxsr && dev_priv->wm.vlv.cxsr)
  1332.                 intel_set_memory_cxsr(dev_priv, false);
  1333.  
  1334.         /* FIXME should be part of crtc atomic commit */
  1335.         vlv_pipe_set_fifo_size(intel_crtc);
  1336.  
  1337.         vlv_write_wm_values(intel_crtc, &wm);
  1338.  
  1339.         DRM_DEBUG_KMS("Setting FIFO watermarks - %c: plane=%d, cursor=%d, "
  1340.                       "sprite0=%d, sprite1=%d, SR: plane=%d, cursor=%d level=%d cxsr=%d\n",
  1341.                       pipe_name(pipe), wm.pipe[pipe].primary, wm.pipe[pipe].cursor,
  1342.                       wm.pipe[pipe].sprite[0], wm.pipe[pipe].sprite[1],
  1343.                       wm.sr.plane, wm.sr.cursor, wm.level, wm.cxsr);
  1344.  
  1345.         if (wm.cxsr && !dev_priv->wm.vlv.cxsr)
  1346.                 intel_set_memory_cxsr(dev_priv, true);
  1347.  
  1348.         if (wm.level >= VLV_WM_LEVEL_PM5 &&
  1349.             dev_priv->wm.vlv.level < VLV_WM_LEVEL_PM5)
  1350.                 chv_set_memory_pm5(dev_priv, true);
  1351.  
  1352.         if (wm.level >= VLV_WM_LEVEL_DDR_DVFS &&
  1353.             dev_priv->wm.vlv.level < VLV_WM_LEVEL_DDR_DVFS)
  1354.                 chv_set_memory_dvfs(dev_priv, true);
  1355.  
  1356.         dev_priv->wm.vlv = wm;
  1357. }
  1358.  
  1359. #define single_plane_enabled(mask) is_power_of_2(mask)
  1360.  
  1361. static void g4x_update_wm(struct drm_crtc *crtc)
  1362. {
  1363.         struct drm_device *dev = crtc->dev;
  1364.         static const int sr_latency_ns = 12000;
  1365.         struct drm_i915_private *dev_priv = dev->dev_private;
  1366.         int planea_wm, planeb_wm, cursora_wm, cursorb_wm;
  1367.         int plane_sr, cursor_sr;
  1368.         unsigned int enabled = 0;
  1369.         bool cxsr_enabled;
  1370.  
  1371.         if (g4x_compute_wm0(dev, PIPE_A,
  1372.                             &g4x_wm_info, pessimal_latency_ns,
  1373.                             &g4x_cursor_wm_info, pessimal_latency_ns,
  1374.                             &planea_wm, &cursora_wm))
  1375.                 enabled |= 1 << PIPE_A;
  1376.  
  1377.         if (g4x_compute_wm0(dev, PIPE_B,
  1378.                             &g4x_wm_info, pessimal_latency_ns,
  1379.                             &g4x_cursor_wm_info, pessimal_latency_ns,
  1380.                             &planeb_wm, &cursorb_wm))
  1381.                 enabled |= 1 << PIPE_B;
  1382.  
  1383.         if (single_plane_enabled(enabled) &&
  1384.             g4x_compute_srwm(dev, ffs(enabled) - 1,
  1385.                              sr_latency_ns,
  1386.                              &g4x_wm_info,
  1387.                              &g4x_cursor_wm_info,
  1388.                              &plane_sr, &cursor_sr)) {
  1389.                 cxsr_enabled = true;
  1390.         } else {
  1391.                 cxsr_enabled = false;
  1392.                 intel_set_memory_cxsr(dev_priv, false);
  1393.                 plane_sr = cursor_sr = 0;
  1394.         }
  1395.  
  1396.         DRM_DEBUG_KMS("Setting FIFO watermarks - A: plane=%d, cursor=%d, "
  1397.                       "B: plane=%d, cursor=%d, SR: plane=%d, cursor=%d\n",
  1398.                       planea_wm, cursora_wm,
  1399.                       planeb_wm, cursorb_wm,
  1400.                       plane_sr, cursor_sr);
  1401.  
  1402.         I915_WRITE(DSPFW1,
  1403.                    FW_WM(plane_sr, SR) |
  1404.                    FW_WM(cursorb_wm, CURSORB) |
  1405.                    FW_WM(planeb_wm, PLANEB) |
  1406.                    FW_WM(planea_wm, PLANEA));
  1407.         I915_WRITE(DSPFW2,
  1408.                    (I915_READ(DSPFW2) & ~DSPFW_CURSORA_MASK) |
  1409.                    FW_WM(cursora_wm, CURSORA));
  1410.         /* HPLL off in SR has some issues on G4x... disable it */
  1411.         I915_WRITE(DSPFW3,
  1412.                    (I915_READ(DSPFW3) & ~(DSPFW_HPLL_SR_EN | DSPFW_CURSOR_SR_MASK)) |
  1413.                    FW_WM(cursor_sr, CURSOR_SR));
  1414.  
  1415.         if (cxsr_enabled)
  1416.                 intel_set_memory_cxsr(dev_priv, true);
  1417. }
  1418.  
  1419. static void i965_update_wm(struct drm_crtc *unused_crtc)
  1420. {
  1421.         struct drm_device *dev = unused_crtc->dev;
  1422.         struct drm_i915_private *dev_priv = dev->dev_private;
  1423.         struct drm_crtc *crtc;
  1424.         int srwm = 1;
  1425.         int cursor_sr = 16;
  1426.         bool cxsr_enabled;
  1427.  
  1428.         /* Calc sr entries for one plane configs */
  1429.         crtc = single_enabled_crtc(dev);
  1430.         if (crtc) {
  1431.                 /* self-refresh has much higher latency */
  1432.                 static const int sr_latency_ns = 12000;
  1433.                 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
  1434.                 int clock = adjusted_mode->crtc_clock;
  1435.                 int htotal = adjusted_mode->crtc_htotal;
  1436.                 int hdisplay = to_intel_crtc(crtc)->config->pipe_src_w;
  1437.                 int pixel_size = crtc->primary->state->fb->bits_per_pixel / 8;
  1438.                 unsigned long line_time_us;
  1439.                 int entries;
  1440.  
  1441.                 line_time_us = max(htotal * 1000 / clock, 1);
  1442.  
  1443.                 /* Use ns/us then divide to preserve precision */
  1444.                 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
  1445.                         pixel_size * hdisplay;
  1446.                 entries = DIV_ROUND_UP(entries, I915_FIFO_LINE_SIZE);
  1447.                 srwm = I965_FIFO_SIZE - entries;
  1448.                 if (srwm < 0)
  1449.                         srwm = 1;
  1450.                 srwm &= 0x1ff;
  1451.                 DRM_DEBUG_KMS("self-refresh entries: %d, wm: %d\n",
  1452.                               entries, srwm);
  1453.  
  1454.                 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
  1455.                         pixel_size * crtc->cursor->state->crtc_w;
  1456.                 entries = DIV_ROUND_UP(entries,
  1457.                                           i965_cursor_wm_info.cacheline_size);
  1458.                 cursor_sr = i965_cursor_wm_info.fifo_size -
  1459.                         (entries + i965_cursor_wm_info.guard_size);
  1460.  
  1461.                 if (cursor_sr > i965_cursor_wm_info.max_wm)
  1462.                         cursor_sr = i965_cursor_wm_info.max_wm;
  1463.  
  1464.                 DRM_DEBUG_KMS("self-refresh watermark: display plane %d "
  1465.                               "cursor %d\n", srwm, cursor_sr);
  1466.  
  1467.                 cxsr_enabled = true;
  1468.         } else {
  1469.                 cxsr_enabled = false;
  1470.                 /* Turn off self refresh if both pipes are enabled */
  1471.                 intel_set_memory_cxsr(dev_priv, false);
  1472.         }
  1473.  
  1474.         DRM_DEBUG_KMS("Setting FIFO watermarks - A: 8, B: 8, C: 8, SR %d\n",
  1475.                       srwm);
  1476.  
  1477.         /* 965 has limitations... */
  1478.         I915_WRITE(DSPFW1, FW_WM(srwm, SR) |
  1479.                    FW_WM(8, CURSORB) |
  1480.                    FW_WM(8, PLANEB) |
  1481.                    FW_WM(8, PLANEA));
  1482.         I915_WRITE(DSPFW2, FW_WM(8, CURSORA) |
  1483.                    FW_WM(8, PLANEC_OLD));
  1484.         /* update cursor SR watermark */
  1485.         I915_WRITE(DSPFW3, FW_WM(cursor_sr, CURSOR_SR));
  1486.  
  1487.         if (cxsr_enabled)
  1488.                 intel_set_memory_cxsr(dev_priv, true);
  1489. }
  1490.  
  1491. #undef FW_WM
  1492.  
  1493. static void i9xx_update_wm(struct drm_crtc *unused_crtc)
  1494. {
  1495.         struct drm_device *dev = unused_crtc->dev;
  1496.         struct drm_i915_private *dev_priv = dev->dev_private;
  1497.         const struct intel_watermark_params *wm_info;
  1498.         uint32_t fwater_lo;
  1499.         uint32_t fwater_hi;
  1500.         int cwm, srwm = 1;
  1501.         int fifo_size;
  1502.         int planea_wm, planeb_wm;
  1503.         struct drm_crtc *crtc, *enabled = NULL;
  1504.  
  1505.         if (IS_I945GM(dev))
  1506.                 wm_info = &i945_wm_info;
  1507.         else if (!IS_GEN2(dev))
  1508.                 wm_info = &i915_wm_info;
  1509.         else
  1510.                 wm_info = &i830_a_wm_info;
  1511.  
  1512.         fifo_size = dev_priv->display.get_fifo_size(dev, 0);
  1513.         crtc = intel_get_crtc_for_plane(dev, 0);
  1514.         if (intel_crtc_active(crtc)) {
  1515.                 const struct drm_display_mode *adjusted_mode;
  1516.                 int cpp = crtc->primary->state->fb->bits_per_pixel / 8;
  1517.                 if (IS_GEN2(dev))
  1518.                         cpp = 4;
  1519.  
  1520.                 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
  1521.                 planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
  1522.                                                wm_info, fifo_size, cpp,
  1523.                                                pessimal_latency_ns);
  1524.                 enabled = crtc;
  1525.         } else {
  1526.                 planea_wm = fifo_size - wm_info->guard_size;
  1527.                 if (planea_wm > (long)wm_info->max_wm)
  1528.                         planea_wm = wm_info->max_wm;
  1529.         }
  1530.  
  1531.         if (IS_GEN2(dev))
  1532.                 wm_info = &i830_bc_wm_info;
  1533.  
  1534.         fifo_size = dev_priv->display.get_fifo_size(dev, 1);
  1535.         crtc = intel_get_crtc_for_plane(dev, 1);
  1536.         if (intel_crtc_active(crtc)) {
  1537.                 const struct drm_display_mode *adjusted_mode;
  1538.                 int cpp = crtc->primary->state->fb->bits_per_pixel / 8;
  1539.                 if (IS_GEN2(dev))
  1540.                         cpp = 4;
  1541.  
  1542.                 adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
  1543.                 planeb_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
  1544.                                                wm_info, fifo_size, cpp,
  1545.                                                pessimal_latency_ns);
  1546.                 if (enabled == NULL)
  1547.                         enabled = crtc;
  1548.                 else
  1549.                         enabled = NULL;
  1550.         } else {
  1551.                 planeb_wm = fifo_size - wm_info->guard_size;
  1552.                 if (planeb_wm > (long)wm_info->max_wm)
  1553.                         planeb_wm = wm_info->max_wm;
  1554.         }
  1555.  
  1556.         DRM_DEBUG_KMS("FIFO watermarks - A: %d, B: %d\n", planea_wm, planeb_wm);
  1557.  
  1558.         if (IS_I915GM(dev) && enabled) {
  1559.                 struct drm_i915_gem_object *obj;
  1560.  
  1561.                 obj = intel_fb_obj(enabled->primary->state->fb);
  1562.  
  1563.                 /* self-refresh seems busted with untiled */
  1564.                 if (obj->tiling_mode == I915_TILING_NONE)
  1565.                         enabled = NULL;
  1566.         }
  1567.  
  1568.         /*
  1569.          * Overlay gets an aggressive default since video jitter is bad.
  1570.          */
  1571.         cwm = 2;
  1572.  
  1573.         /* Play safe and disable self-refresh before adjusting watermarks. */
  1574.         intel_set_memory_cxsr(dev_priv, false);
  1575.  
  1576.         /* Calc sr entries for one plane configs */
  1577.         if (HAS_FW_BLC(dev) && enabled) {
  1578.                 /* self-refresh has much higher latency */
  1579.                 static const int sr_latency_ns = 6000;
  1580.                 const struct drm_display_mode *adjusted_mode = &to_intel_crtc(enabled)->config->base.adjusted_mode;
  1581.                 int clock = adjusted_mode->crtc_clock;
  1582.                 int htotal = adjusted_mode->crtc_htotal;
  1583.                 int hdisplay = to_intel_crtc(enabled)->config->pipe_src_w;
  1584.                 int pixel_size = enabled->primary->state->fb->bits_per_pixel / 8;
  1585.                 unsigned long line_time_us;
  1586.                 int entries;
  1587.  
  1588.                 line_time_us = max(htotal * 1000 / clock, 1);
  1589.  
  1590.                 /* Use ns/us then divide to preserve precision */
  1591.                 entries = (((sr_latency_ns / line_time_us) + 1000) / 1000) *
  1592.                         pixel_size * hdisplay;
  1593.                 entries = DIV_ROUND_UP(entries, wm_info->cacheline_size);
  1594.                 DRM_DEBUG_KMS("self-refresh entries: %d\n", entries);
  1595.                 srwm = wm_info->fifo_size - entries;
  1596.                 if (srwm < 0)
  1597.                         srwm = 1;
  1598.  
  1599.                 if (IS_I945G(dev) || IS_I945GM(dev))
  1600.                         I915_WRITE(FW_BLC_SELF,
  1601.                                    FW_BLC_SELF_FIFO_MASK | (srwm & 0xff));
  1602.                 else if (IS_I915GM(dev))
  1603.                         I915_WRITE(FW_BLC_SELF, srwm & 0x3f);
  1604.         }
  1605.  
  1606.         DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d, B: %d, C: %d, SR %d\n",
  1607.                       planea_wm, planeb_wm, cwm, srwm);
  1608.  
  1609.         fwater_lo = ((planeb_wm & 0x3f) << 16) | (planea_wm & 0x3f);
  1610.         fwater_hi = (cwm & 0x1f);
  1611.  
  1612.         /* Set request length to 8 cachelines per fetch */
  1613.         fwater_lo = fwater_lo | (1 << 24) | (1 << 8);
  1614.         fwater_hi = fwater_hi | (1 << 8);
  1615.  
  1616.         I915_WRITE(FW_BLC, fwater_lo);
  1617.         I915_WRITE(FW_BLC2, fwater_hi);
  1618.  
  1619.         if (enabled)
  1620.                 intel_set_memory_cxsr(dev_priv, true);
  1621. }
  1622.  
  1623. static void i845_update_wm(struct drm_crtc *unused_crtc)
  1624. {
  1625.         struct drm_device *dev = unused_crtc->dev;
  1626.         struct drm_i915_private *dev_priv = dev->dev_private;
  1627.         struct drm_crtc *crtc;
  1628.         const struct drm_display_mode *adjusted_mode;
  1629.         uint32_t fwater_lo;
  1630.         int planea_wm;
  1631.  
  1632.         crtc = single_enabled_crtc(dev);
  1633.         if (crtc == NULL)
  1634.                 return;
  1635.  
  1636.         adjusted_mode = &to_intel_crtc(crtc)->config->base.adjusted_mode;
  1637.         planea_wm = intel_calculate_wm(adjusted_mode->crtc_clock,
  1638.                                        &i845_wm_info,
  1639.                                        dev_priv->display.get_fifo_size(dev, 0),
  1640.                                        4, pessimal_latency_ns);
  1641.         fwater_lo = I915_READ(FW_BLC) & ~0xfff;
  1642.         fwater_lo |= (3<<8) | planea_wm;
  1643.  
  1644.         DRM_DEBUG_KMS("Setting FIFO watermarks - A: %d\n", planea_wm);
  1645.  
  1646.         I915_WRITE(FW_BLC, fwater_lo);
  1647. }
  1648.  
  1649. uint32_t ilk_pipe_pixel_rate(const struct intel_crtc_state *pipe_config)
  1650. {
  1651.         uint32_t pixel_rate;
  1652.  
  1653.         pixel_rate = pipe_config->base.adjusted_mode.crtc_clock;
  1654.  
  1655.         /* We only use IF-ID interlacing. If we ever use PF-ID we'll need to
  1656.          * adjust the pixel_rate here. */
  1657.  
  1658.         if (pipe_config->pch_pfit.enabled) {
  1659.                 uint64_t pipe_w, pipe_h, pfit_w, pfit_h;
  1660.                 uint32_t pfit_size = pipe_config->pch_pfit.size;
  1661.  
  1662.                 pipe_w = pipe_config->pipe_src_w;
  1663.                 pipe_h = pipe_config->pipe_src_h;
  1664.  
  1665.                 pfit_w = (pfit_size >> 16) & 0xFFFF;
  1666.                 pfit_h = pfit_size & 0xFFFF;
  1667.                 if (pipe_w < pfit_w)
  1668.                         pipe_w = pfit_w;
  1669.                 if (pipe_h < pfit_h)
  1670.                         pipe_h = pfit_h;
  1671.  
  1672.                 pixel_rate = div_u64((uint64_t) pixel_rate * pipe_w * pipe_h,
  1673.                                      pfit_w * pfit_h);
  1674.         }
  1675.  
  1676.         return pixel_rate;
  1677. }
  1678.  
  1679. /* latency must be in 0.1us units. */
  1680. static uint32_t ilk_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
  1681.                                uint32_t latency)
  1682. {
  1683.         uint64_t ret;
  1684.  
  1685.         if (WARN(latency == 0, "Latency value missing\n"))
  1686.                 return UINT_MAX;
  1687.  
  1688.         ret = (uint64_t) pixel_rate * bytes_per_pixel * latency;
  1689.         ret = DIV_ROUND_UP_ULL(ret, 64 * 10000) + 2;
  1690.  
  1691.         return ret;
  1692. }
  1693.  
  1694. /* latency must be in 0.1us units. */
  1695. static uint32_t ilk_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
  1696.                                uint32_t horiz_pixels, uint8_t bytes_per_pixel,
  1697.                                uint32_t latency)
  1698. {
  1699.         uint32_t ret;
  1700.  
  1701.         if (WARN(latency == 0, "Latency value missing\n"))
  1702.                 return UINT_MAX;
  1703.  
  1704.         ret = (latency * pixel_rate) / (pipe_htotal * 10000);
  1705.         ret = (ret + 1) * horiz_pixels * bytes_per_pixel;
  1706.         ret = DIV_ROUND_UP(ret, 64) + 2;
  1707.         return ret;
  1708. }
  1709.  
  1710. static uint32_t ilk_wm_fbc(uint32_t pri_val, uint32_t horiz_pixels,
  1711.                            uint8_t bytes_per_pixel)
  1712. {
  1713.         return DIV_ROUND_UP(pri_val * 64, horiz_pixels * bytes_per_pixel) + 2;
  1714. }
  1715.  
  1716. struct skl_pipe_wm_parameters {
  1717.         bool active;
  1718.         uint32_t pipe_htotal;
  1719.         uint32_t pixel_rate; /* in KHz */
  1720.         struct intel_plane_wm_parameters plane[I915_MAX_PLANES];
  1721. };
  1722.  
  1723. struct ilk_wm_maximums {
  1724.         uint16_t pri;
  1725.         uint16_t spr;
  1726.         uint16_t cur;
  1727.         uint16_t fbc;
  1728. };
  1729.  
  1730. /* used in computing the new watermarks state */
  1731. struct intel_wm_config {
  1732.         unsigned int num_pipes_active;
  1733.         bool sprites_enabled;
  1734.         bool sprites_scaled;
  1735. };
  1736.  
  1737. /*
  1738.  * For both WM_PIPE and WM_LP.
  1739.  * mem_value must be in 0.1us units.
  1740.  */
  1741. static uint32_t ilk_compute_pri_wm(const struct intel_crtc_state *cstate,
  1742.                                    const struct intel_plane_state *pstate,
  1743.                                    uint32_t mem_value,
  1744.                                    bool is_lp)
  1745. {
  1746.         int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0;
  1747.         uint32_t method1, method2;
  1748.  
  1749.         if (!cstate->base.active || !pstate->visible)
  1750.                 return 0;
  1751.  
  1752.         method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), bpp, mem_value);
  1753.  
  1754.         if (!is_lp)
  1755.                 return method1;
  1756.  
  1757.         method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
  1758.                                  cstate->base.adjusted_mode.crtc_htotal,
  1759.                                  drm_rect_width(&pstate->dst),
  1760.                                  bpp,
  1761.                                  mem_value);
  1762.  
  1763.         return min(method1, method2);
  1764. }
  1765.  
  1766. /*
  1767.  * For both WM_PIPE and WM_LP.
  1768.  * mem_value must be in 0.1us units.
  1769.  */
  1770. static uint32_t ilk_compute_spr_wm(const struct intel_crtc_state *cstate,
  1771.                                    const struct intel_plane_state *pstate,
  1772.                                    uint32_t mem_value)
  1773. {
  1774.         int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0;
  1775.         uint32_t method1, method2;
  1776.  
  1777.         if (!cstate->base.active || !pstate->visible)
  1778.                 return 0;
  1779.  
  1780.         method1 = ilk_wm_method1(ilk_pipe_pixel_rate(cstate), bpp, mem_value);
  1781.         method2 = ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
  1782.                                  cstate->base.adjusted_mode.crtc_htotal,
  1783.                                  drm_rect_width(&pstate->dst),
  1784.                                  bpp,
  1785.                                  mem_value);
  1786.         return min(method1, method2);
  1787. }
  1788.  
  1789. /*
  1790.  * For both WM_PIPE and WM_LP.
  1791.  * mem_value must be in 0.1us units.
  1792.  */
  1793. static uint32_t ilk_compute_cur_wm(const struct intel_crtc_state *cstate,
  1794.                                    const struct intel_plane_state *pstate,
  1795.                                    uint32_t mem_value)
  1796. {
  1797.         int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0;
  1798.  
  1799.         if (!cstate->base.active || !pstate->visible)
  1800.                 return 0;
  1801.  
  1802.         return ilk_wm_method2(ilk_pipe_pixel_rate(cstate),
  1803.                               cstate->base.adjusted_mode.crtc_htotal,
  1804.                               drm_rect_width(&pstate->dst),
  1805.                               bpp,
  1806.                               mem_value);
  1807. }
  1808.  
  1809. /* Only for WM_LP. */
  1810. static uint32_t ilk_compute_fbc_wm(const struct intel_crtc_state *cstate,
  1811.                                    const struct intel_plane_state *pstate,
  1812.                                    uint32_t pri_val)
  1813. {
  1814.         int bpp = pstate->base.fb ? pstate->base.fb->bits_per_pixel / 8 : 0;
  1815.  
  1816.         if (!cstate->base.active || !pstate->visible)
  1817.                 return 0;
  1818.  
  1819.         return ilk_wm_fbc(pri_val, drm_rect_width(&pstate->dst), bpp);
  1820. }
  1821.  
  1822. static unsigned int ilk_display_fifo_size(const struct drm_device *dev)
  1823. {
  1824.         if (INTEL_INFO(dev)->gen >= 8)
  1825.                 return 3072;
  1826.         else if (INTEL_INFO(dev)->gen >= 7)
  1827.                 return 768;
  1828.         else
  1829.                 return 512;
  1830. }
  1831.  
  1832. static unsigned int ilk_plane_wm_reg_max(const struct drm_device *dev,
  1833.                                          int level, bool is_sprite)
  1834. {
  1835.         if (INTEL_INFO(dev)->gen >= 8)
  1836.                 /* BDW primary/sprite plane watermarks */
  1837.                 return level == 0 ? 255 : 2047;
  1838.         else if (INTEL_INFO(dev)->gen >= 7)
  1839.                 /* IVB/HSW primary/sprite plane watermarks */
  1840.                 return level == 0 ? 127 : 1023;
  1841.         else if (!is_sprite)
  1842.                 /* ILK/SNB primary plane watermarks */
  1843.                 return level == 0 ? 127 : 511;
  1844.         else
  1845.                 /* ILK/SNB sprite plane watermarks */
  1846.                 return level == 0 ? 63 : 255;
  1847. }
  1848.  
  1849. static unsigned int ilk_cursor_wm_reg_max(const struct drm_device *dev,
  1850.                                           int level)
  1851. {
  1852.         if (INTEL_INFO(dev)->gen >= 7)
  1853.                 return level == 0 ? 63 : 255;
  1854.         else
  1855.                 return level == 0 ? 31 : 63;
  1856. }
  1857.  
  1858. static unsigned int ilk_fbc_wm_reg_max(const struct drm_device *dev)
  1859. {
  1860.         if (INTEL_INFO(dev)->gen >= 8)
  1861.                 return 31;
  1862.         else
  1863.                 return 15;
  1864. }
  1865.  
  1866. /* Calculate the maximum primary/sprite plane watermark */
  1867. static unsigned int ilk_plane_wm_max(const struct drm_device *dev,
  1868.                                      int level,
  1869.                                      const struct intel_wm_config *config,
  1870.                                      enum intel_ddb_partitioning ddb_partitioning,
  1871.                                      bool is_sprite)
  1872. {
  1873.         unsigned int fifo_size = ilk_display_fifo_size(dev);
  1874.  
  1875.         /* if sprites aren't enabled, sprites get nothing */
  1876.         if (is_sprite && !config->sprites_enabled)
  1877.                 return 0;
  1878.  
  1879.         /* HSW allows LP1+ watermarks even with multiple pipes */
  1880.         if (level == 0 || config->num_pipes_active > 1) {
  1881.                 fifo_size /= INTEL_INFO(dev)->num_pipes;
  1882.  
  1883.                 /*
  1884.                  * For some reason the non self refresh
  1885.                  * FIFO size is only half of the self
  1886.                  * refresh FIFO size on ILK/SNB.
  1887.                  */
  1888.                 if (INTEL_INFO(dev)->gen <= 6)
  1889.                         fifo_size /= 2;
  1890.         }
  1891.  
  1892.         if (config->sprites_enabled) {
  1893.                 /* level 0 is always calculated with 1:1 split */
  1894.                 if (level > 0 && ddb_partitioning == INTEL_DDB_PART_5_6) {
  1895.                         if (is_sprite)
  1896.                                 fifo_size *= 5;
  1897.                         fifo_size /= 6;
  1898.                 } else {
  1899.                         fifo_size /= 2;
  1900.                 }
  1901.         }
  1902.  
  1903.         /* clamp to max that the registers can hold */
  1904.         return min(fifo_size, ilk_plane_wm_reg_max(dev, level, is_sprite));
  1905. }
  1906.  
  1907. /* Calculate the maximum cursor plane watermark */
  1908. static unsigned int ilk_cursor_wm_max(const struct drm_device *dev,
  1909.                                       int level,
  1910.                                       const struct intel_wm_config *config)
  1911. {
  1912.         /* HSW LP1+ watermarks w/ multiple pipes */
  1913.         if (level > 0 && config->num_pipes_active > 1)
  1914.                 return 64;
  1915.  
  1916.         /* otherwise just report max that registers can hold */
  1917.         return ilk_cursor_wm_reg_max(dev, level);
  1918. }
  1919.  
  1920. static void ilk_compute_wm_maximums(const struct drm_device *dev,
  1921.                                     int level,
  1922.                                     const struct intel_wm_config *config,
  1923.                                     enum intel_ddb_partitioning ddb_partitioning,
  1924.                                     struct ilk_wm_maximums *max)
  1925. {
  1926.         max->pri = ilk_plane_wm_max(dev, level, config, ddb_partitioning, false);
  1927.         max->spr = ilk_plane_wm_max(dev, level, config, ddb_partitioning, true);
  1928.         max->cur = ilk_cursor_wm_max(dev, level, config);
  1929.         max->fbc = ilk_fbc_wm_reg_max(dev);
  1930. }
  1931.  
  1932. static void ilk_compute_wm_reg_maximums(struct drm_device *dev,
  1933.                                         int level,
  1934.                                         struct ilk_wm_maximums *max)
  1935. {
  1936.         max->pri = ilk_plane_wm_reg_max(dev, level, false);
  1937.         max->spr = ilk_plane_wm_reg_max(dev, level, true);
  1938.         max->cur = ilk_cursor_wm_reg_max(dev, level);
  1939.         max->fbc = ilk_fbc_wm_reg_max(dev);
  1940. }
  1941.  
  1942. static bool ilk_validate_wm_level(int level,
  1943.                                   const struct ilk_wm_maximums *max,
  1944.                                   struct intel_wm_level *result)
  1945. {
  1946.         bool ret;
  1947.  
  1948.         /* already determined to be invalid? */
  1949.         if (!result->enable)
  1950.                 return false;
  1951.  
  1952.         result->enable = result->pri_val <= max->pri &&
  1953.                          result->spr_val <= max->spr &&
  1954.                          result->cur_val <= max->cur;
  1955.  
  1956.         ret = result->enable;
  1957.  
  1958.         /*
  1959.          * HACK until we can pre-compute everything,
  1960.          * and thus fail gracefully if LP0 watermarks
  1961.          * are exceeded...
  1962.          */
  1963.         if (level == 0 && !result->enable) {
  1964.                 if (result->pri_val > max->pri)
  1965.                         DRM_DEBUG_KMS("Primary WM%d too large %u (max %u)\n",
  1966.                                       level, result->pri_val, max->pri);
  1967.                 if (result->spr_val > max->spr)
  1968.                         DRM_DEBUG_KMS("Sprite WM%d too large %u (max %u)\n",
  1969.                                       level, result->spr_val, max->spr);
  1970.                 if (result->cur_val > max->cur)
  1971.                         DRM_DEBUG_KMS("Cursor WM%d too large %u (max %u)\n",
  1972.                                       level, result->cur_val, max->cur);
  1973.  
  1974.                 result->pri_val = min_t(uint32_t, result->pri_val, max->pri);
  1975.                 result->spr_val = min_t(uint32_t, result->spr_val, max->spr);
  1976.                 result->cur_val = min_t(uint32_t, result->cur_val, max->cur);
  1977.                 result->enable = true;
  1978.         }
  1979.  
  1980.         return ret;
  1981. }
  1982.  
  1983. static void ilk_compute_wm_level(const struct drm_i915_private *dev_priv,
  1984.                                  const struct intel_crtc *intel_crtc,
  1985.                                  int level,
  1986.                                  struct intel_crtc_state *cstate,
  1987.                                  struct intel_wm_level *result)
  1988. {
  1989.         struct intel_plane *intel_plane;
  1990.         uint16_t pri_latency = dev_priv->wm.pri_latency[level];
  1991.         uint16_t spr_latency = dev_priv->wm.spr_latency[level];
  1992.         uint16_t cur_latency = dev_priv->wm.cur_latency[level];
  1993.  
  1994.         /* WM1+ latency values stored in 0.5us units */
  1995.         if (level > 0) {
  1996.                 pri_latency *= 5;
  1997.                 spr_latency *= 5;
  1998.                 cur_latency *= 5;
  1999.         }
  2000.  
  2001.         for_each_intel_plane_on_crtc(dev_priv->dev, intel_crtc, intel_plane) {
  2002.                 struct intel_plane_state *pstate =
  2003.                         to_intel_plane_state(intel_plane->base.state);
  2004.  
  2005.                 switch (intel_plane->base.type) {
  2006.                 case DRM_PLANE_TYPE_PRIMARY:
  2007.                         result->pri_val = ilk_compute_pri_wm(cstate, pstate,
  2008.                                                              pri_latency,
  2009.                                                              level);
  2010.                         result->fbc_val = ilk_compute_fbc_wm(cstate, pstate,
  2011.                                                              result->pri_val);
  2012.                         break;
  2013.                 case DRM_PLANE_TYPE_OVERLAY:
  2014.                         result->spr_val = ilk_compute_spr_wm(cstate, pstate,
  2015.                                                              spr_latency);
  2016.                         break;
  2017.                 case DRM_PLANE_TYPE_CURSOR:
  2018.                         result->cur_val = ilk_compute_cur_wm(cstate, pstate,
  2019.                                                              cur_latency);
  2020.                         break;
  2021.                 }
  2022.         }
  2023.  
  2024.         result->enable = true;
  2025. }
  2026.  
  2027. static uint32_t
  2028. hsw_compute_linetime_wm(struct drm_device *dev, struct drm_crtc *crtc)
  2029. {
  2030.         struct drm_i915_private *dev_priv = dev->dev_private;
  2031.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  2032.         const struct drm_display_mode *adjusted_mode = &intel_crtc->config->base.adjusted_mode;
  2033.         u32 linetime, ips_linetime;
  2034.  
  2035.         if (!intel_crtc->active)
  2036.                 return 0;
  2037.  
  2038.         /* The WM are computed with base on how long it takes to fill a single
  2039.          * row at the given clock rate, multiplied by 8.
  2040.          * */
  2041.         linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8,
  2042.                                      adjusted_mode->crtc_clock);
  2043.         ips_linetime = DIV_ROUND_CLOSEST(adjusted_mode->crtc_htotal * 1000 * 8,
  2044.                                          dev_priv->cdclk_freq);
  2045.  
  2046.         return PIPE_WM_LINETIME_IPS_LINETIME(ips_linetime) |
  2047.                PIPE_WM_LINETIME_TIME(linetime);
  2048. }
  2049.  
  2050. static void intel_read_wm_latency(struct drm_device *dev, uint16_t wm[8])
  2051. {
  2052.         struct drm_i915_private *dev_priv = dev->dev_private;
  2053.  
  2054.         if (IS_GEN9(dev)) {
  2055.                 uint32_t val;
  2056.                 int ret, i;
  2057.                 int level, max_level = ilk_wm_max_level(dev);
  2058.  
  2059.                 /* read the first set of memory latencies[0:3] */
  2060.                 val = 0; /* data0 to be programmed to 0 for first set */
  2061.                 mutex_lock(&dev_priv->rps.hw_lock);
  2062.                 ret = sandybridge_pcode_read(dev_priv,
  2063.                                              GEN9_PCODE_READ_MEM_LATENCY,
  2064.                                              &val);
  2065.                 mutex_unlock(&dev_priv->rps.hw_lock);
  2066.  
  2067.                 if (ret) {
  2068.                         DRM_ERROR("SKL Mailbox read error = %d\n", ret);
  2069.                         return;
  2070.                 }
  2071.  
  2072.                 wm[0] = val & GEN9_MEM_LATENCY_LEVEL_MASK;
  2073.                 wm[1] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) &
  2074.                                 GEN9_MEM_LATENCY_LEVEL_MASK;
  2075.                 wm[2] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) &
  2076.                                 GEN9_MEM_LATENCY_LEVEL_MASK;
  2077.                 wm[3] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) &
  2078.                                 GEN9_MEM_LATENCY_LEVEL_MASK;
  2079.  
  2080.                 /* read the second set of memory latencies[4:7] */
  2081.                 val = 1; /* data0 to be programmed to 1 for second set */
  2082.                 mutex_lock(&dev_priv->rps.hw_lock);
  2083.                 ret = sandybridge_pcode_read(dev_priv,
  2084.                                              GEN9_PCODE_READ_MEM_LATENCY,
  2085.                                              &val);
  2086.                 mutex_unlock(&dev_priv->rps.hw_lock);
  2087.                 if (ret) {
  2088.                         DRM_ERROR("SKL Mailbox read error = %d\n", ret);
  2089.                         return;
  2090.                 }
  2091.  
  2092.                 wm[4] = val & GEN9_MEM_LATENCY_LEVEL_MASK;
  2093.                 wm[5] = (val >> GEN9_MEM_LATENCY_LEVEL_1_5_SHIFT) &
  2094.                                 GEN9_MEM_LATENCY_LEVEL_MASK;
  2095.                 wm[6] = (val >> GEN9_MEM_LATENCY_LEVEL_2_6_SHIFT) &
  2096.                                 GEN9_MEM_LATENCY_LEVEL_MASK;
  2097.                 wm[7] = (val >> GEN9_MEM_LATENCY_LEVEL_3_7_SHIFT) &
  2098.                                 GEN9_MEM_LATENCY_LEVEL_MASK;
  2099.  
  2100.                 /*
  2101.                  * WaWmMemoryReadLatency:skl
  2102.                  *
  2103.                  * punit doesn't take into account the read latency so we need
  2104.                  * to add 2us to the various latency levels we retrieve from
  2105.                  * the punit.
  2106.                  *   - W0 is a bit special in that it's the only level that
  2107.                  *   can't be disabled if we want to have display working, so
  2108.                  *   we always add 2us there.
  2109.                  *   - For levels >=1, punit returns 0us latency when they are
  2110.                  *   disabled, so we respect that and don't add 2us then
  2111.                  *
  2112.                  * Additionally, if a level n (n > 1) has a 0us latency, all
  2113.                  * levels m (m >= n) need to be disabled. We make sure to
  2114.                  * sanitize the values out of the punit to satisfy this
  2115.                  * requirement.
  2116.                  */
  2117.                 wm[0] += 2;
  2118.                 for (level = 1; level <= max_level; level++)
  2119.                         if (wm[level] != 0)
  2120.                                 wm[level] += 2;
  2121.                         else {
  2122.                                 for (i = level + 1; i <= max_level; i++)
  2123.                                         wm[i] = 0;
  2124.  
  2125.                                 break;
  2126.                         }
  2127.         } else if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
  2128.                 uint64_t sskpd = I915_READ64(MCH_SSKPD);
  2129.  
  2130.                 wm[0] = (sskpd >> 56) & 0xFF;
  2131.                 if (wm[0] == 0)
  2132.                         wm[0] = sskpd & 0xF;
  2133.                 wm[1] = (sskpd >> 4) & 0xFF;
  2134.                 wm[2] = (sskpd >> 12) & 0xFF;
  2135.                 wm[3] = (sskpd >> 20) & 0x1FF;
  2136.                 wm[4] = (sskpd >> 32) & 0x1FF;
  2137.         } else if (INTEL_INFO(dev)->gen >= 6) {
  2138.                 uint32_t sskpd = I915_READ(MCH_SSKPD);
  2139.  
  2140.                 wm[0] = (sskpd >> SSKPD_WM0_SHIFT) & SSKPD_WM_MASK;
  2141.                 wm[1] = (sskpd >> SSKPD_WM1_SHIFT) & SSKPD_WM_MASK;
  2142.                 wm[2] = (sskpd >> SSKPD_WM2_SHIFT) & SSKPD_WM_MASK;
  2143.                 wm[3] = (sskpd >> SSKPD_WM3_SHIFT) & SSKPD_WM_MASK;
  2144.         } else if (INTEL_INFO(dev)->gen >= 5) {
  2145.                 uint32_t mltr = I915_READ(MLTR_ILK);
  2146.  
  2147.                 /* ILK primary LP0 latency is 700 ns */
  2148.                 wm[0] = 7;
  2149.                 wm[1] = (mltr >> MLTR_WM1_SHIFT) & ILK_SRLT_MASK;
  2150.                 wm[2] = (mltr >> MLTR_WM2_SHIFT) & ILK_SRLT_MASK;
  2151.         }
  2152. }
  2153.  
  2154. static void intel_fixup_spr_wm_latency(struct drm_device *dev, uint16_t wm[5])
  2155. {
  2156.         /* ILK sprite LP0 latency is 1300 ns */
  2157.         if (INTEL_INFO(dev)->gen == 5)
  2158.                 wm[0] = 13;
  2159. }
  2160.  
  2161. static void intel_fixup_cur_wm_latency(struct drm_device *dev, uint16_t wm[5])
  2162. {
  2163.         /* ILK cursor LP0 latency is 1300 ns */
  2164.         if (INTEL_INFO(dev)->gen == 5)
  2165.                 wm[0] = 13;
  2166.  
  2167.         /* WaDoubleCursorLP3Latency:ivb */
  2168.         if (IS_IVYBRIDGE(dev))
  2169.                 wm[3] *= 2;
  2170. }
  2171.  
  2172. int ilk_wm_max_level(const struct drm_device *dev)
  2173. {
  2174.         /* how many WM levels are we expecting */
  2175.         if (INTEL_INFO(dev)->gen >= 9)
  2176.                 return 7;
  2177.         else if (IS_HASWELL(dev) || IS_BROADWELL(dev))
  2178.                 return 4;
  2179.         else if (INTEL_INFO(dev)->gen >= 6)
  2180.                 return 3;
  2181.         else
  2182.                 return 2;
  2183. }
  2184.  
  2185. static void intel_print_wm_latency(struct drm_device *dev,
  2186.                                    const char *name,
  2187.                                    const uint16_t wm[8])
  2188. {
  2189.         int level, max_level = ilk_wm_max_level(dev);
  2190.  
  2191.         for (level = 0; level <= max_level; level++) {
  2192.                 unsigned int latency = wm[level];
  2193.  
  2194.                 if (latency == 0) {
  2195.                         DRM_ERROR("%s WM%d latency not provided\n",
  2196.                                   name, level);
  2197.                         continue;
  2198.                 }
  2199.  
  2200.                 /*
  2201.                  * - latencies are in us on gen9.
  2202.                  * - before then, WM1+ latency values are in 0.5us units
  2203.                  */
  2204.                 if (IS_GEN9(dev))
  2205.                         latency *= 10;
  2206.                 else if (level > 0)
  2207.                         latency *= 5;
  2208.  
  2209.                 DRM_DEBUG_KMS("%s WM%d latency %u (%u.%u usec)\n",
  2210.                               name, level, wm[level],
  2211.                               latency / 10, latency % 10);
  2212.         }
  2213. }
  2214.  
  2215. static bool ilk_increase_wm_latency(struct drm_i915_private *dev_priv,
  2216.                                     uint16_t wm[5], uint16_t min)
  2217. {
  2218.         int level, max_level = ilk_wm_max_level(dev_priv->dev);
  2219.  
  2220.         if (wm[0] >= min)
  2221.                 return false;
  2222.  
  2223.         wm[0] = max(wm[0], min);
  2224.         for (level = 1; level <= max_level; level++)
  2225.                 wm[level] = max_t(uint16_t, wm[level], DIV_ROUND_UP(min, 5));
  2226.  
  2227.         return true;
  2228. }
  2229.  
  2230. static void snb_wm_latency_quirk(struct drm_device *dev)
  2231. {
  2232.         struct drm_i915_private *dev_priv = dev->dev_private;
  2233.         bool changed;
  2234.  
  2235.         /*
  2236.          * The BIOS provided WM memory latency values are often
  2237.          * inadequate for high resolution displays. Adjust them.
  2238.          */
  2239.         changed = ilk_increase_wm_latency(dev_priv, dev_priv->wm.pri_latency, 12) |
  2240.                 ilk_increase_wm_latency(dev_priv, dev_priv->wm.spr_latency, 12) |
  2241.                 ilk_increase_wm_latency(dev_priv, dev_priv->wm.cur_latency, 12);
  2242.  
  2243.         if (!changed)
  2244.                 return;
  2245.  
  2246.         DRM_DEBUG_KMS("WM latency values increased to avoid potential underruns\n");
  2247.         intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
  2248.         intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
  2249.         intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
  2250. }
  2251.  
  2252. static void ilk_setup_wm_latency(struct drm_device *dev)
  2253. {
  2254.         struct drm_i915_private *dev_priv = dev->dev_private;
  2255.  
  2256.         intel_read_wm_latency(dev, dev_priv->wm.pri_latency);
  2257.  
  2258.         memcpy(dev_priv->wm.spr_latency, dev_priv->wm.pri_latency,
  2259.                sizeof(dev_priv->wm.pri_latency));
  2260.         memcpy(dev_priv->wm.cur_latency, dev_priv->wm.pri_latency,
  2261.                sizeof(dev_priv->wm.pri_latency));
  2262.  
  2263.         intel_fixup_spr_wm_latency(dev, dev_priv->wm.spr_latency);
  2264.         intel_fixup_cur_wm_latency(dev, dev_priv->wm.cur_latency);
  2265.  
  2266.         intel_print_wm_latency(dev, "Primary", dev_priv->wm.pri_latency);
  2267.         intel_print_wm_latency(dev, "Sprite", dev_priv->wm.spr_latency);
  2268.         intel_print_wm_latency(dev, "Cursor", dev_priv->wm.cur_latency);
  2269.  
  2270.         if (IS_GEN6(dev))
  2271.                 snb_wm_latency_quirk(dev);
  2272. }
  2273.  
  2274. static void skl_setup_wm_latency(struct drm_device *dev)
  2275. {
  2276.         struct drm_i915_private *dev_priv = dev->dev_private;
  2277.  
  2278.         intel_read_wm_latency(dev, dev_priv->wm.skl_latency);
  2279.         intel_print_wm_latency(dev, "Gen9 Plane", dev_priv->wm.skl_latency);
  2280. }
  2281.  
  2282. static void ilk_compute_wm_config(struct drm_device *dev,
  2283.                                   struct intel_wm_config *config)
  2284. {
  2285.         struct intel_crtc *intel_crtc;
  2286.  
  2287.         /* Compute the currently _active_ config */
  2288.         for_each_intel_crtc(dev, intel_crtc) {
  2289.                 const struct intel_pipe_wm *wm = &intel_crtc->wm.active;
  2290.  
  2291.                 if (!wm->pipe_enabled)
  2292.                         continue;
  2293.  
  2294.                 config->sprites_enabled |= wm->sprites_enabled;
  2295.                 config->sprites_scaled |= wm->sprites_scaled;
  2296.                 config->num_pipes_active++;
  2297.         }
  2298. }
  2299.  
  2300. /* Compute new watermarks for the pipe */
  2301. static bool intel_compute_pipe_wm(struct intel_crtc_state *cstate,
  2302.                                   struct intel_pipe_wm *pipe_wm)
  2303. {
  2304.         struct drm_crtc *crtc = cstate->base.crtc;
  2305.         struct drm_device *dev = crtc->dev;
  2306.         const struct drm_i915_private *dev_priv = dev->dev_private;
  2307.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  2308.         struct intel_plane *intel_plane;
  2309.         struct intel_plane_state *sprstate = NULL;
  2310.         int level, max_level = ilk_wm_max_level(dev);
  2311.         /* LP0 watermark maximums depend on this pipe alone */
  2312.         struct intel_wm_config config = {
  2313.                 .num_pipes_active = 1,
  2314.         };
  2315.         struct ilk_wm_maximums max;
  2316.  
  2317.         for_each_intel_plane_on_crtc(dev, intel_crtc, intel_plane) {
  2318.                 if (intel_plane->base.type == DRM_PLANE_TYPE_OVERLAY) {
  2319.                         sprstate = to_intel_plane_state(intel_plane->base.state);
  2320.                         break;
  2321.                 }
  2322.         }
  2323.  
  2324.         config.sprites_enabled = sprstate->visible;
  2325.         config.sprites_scaled = sprstate->visible &&
  2326.                 (drm_rect_width(&sprstate->dst) != drm_rect_width(&sprstate->src) >> 16 ||
  2327.                 drm_rect_height(&sprstate->dst) != drm_rect_height(&sprstate->src) >> 16);
  2328.  
  2329.         pipe_wm->pipe_enabled = cstate->base.active;
  2330.         pipe_wm->sprites_enabled = sprstate->visible;
  2331.         pipe_wm->sprites_scaled = config.sprites_scaled;
  2332.  
  2333.         /* ILK/SNB: LP2+ watermarks only w/o sprites */
  2334.         if (INTEL_INFO(dev)->gen <= 6 && sprstate->visible)
  2335.                 max_level = 1;
  2336.  
  2337.         /* ILK/SNB/IVB: LP1+ watermarks only w/o scaling */
  2338.         if (config.sprites_scaled)
  2339.                 max_level = 0;
  2340.  
  2341.         ilk_compute_wm_level(dev_priv, intel_crtc, 0, cstate, &pipe_wm->wm[0]);
  2342.  
  2343.         if (IS_HASWELL(dev) || IS_BROADWELL(dev))
  2344.                 pipe_wm->linetime = hsw_compute_linetime_wm(dev, crtc);
  2345.  
  2346.         /* LP0 watermarks always use 1/2 DDB partitioning */
  2347.         ilk_compute_wm_maximums(dev, 0, &config, INTEL_DDB_PART_1_2, &max);
  2348.  
  2349.         /* At least LP0 must be valid */
  2350.         if (!ilk_validate_wm_level(0, &max, &pipe_wm->wm[0]))
  2351.                 return false;
  2352.  
  2353.         ilk_compute_wm_reg_maximums(dev, 1, &max);
  2354.  
  2355.         for (level = 1; level <= max_level; level++) {
  2356.                 struct intel_wm_level wm = {};
  2357.  
  2358.                 ilk_compute_wm_level(dev_priv, intel_crtc, level, cstate, &wm);
  2359.  
  2360.                 /*
  2361.                  * Disable any watermark level that exceeds the
  2362.                  * register maximums since such watermarks are
  2363.                  * always invalid.
  2364.                  */
  2365.                 if (!ilk_validate_wm_level(level, &max, &wm))
  2366.                         break;
  2367.  
  2368.                 pipe_wm->wm[level] = wm;
  2369.         }
  2370.  
  2371.         return true;
  2372. }
  2373.  
  2374. /*
  2375.  * Merge the watermarks from all active pipes for a specific level.
  2376.  */
  2377. static void ilk_merge_wm_level(struct drm_device *dev,
  2378.                                int level,
  2379.                                struct intel_wm_level *ret_wm)
  2380. {
  2381.         const struct intel_crtc *intel_crtc;
  2382.  
  2383.         ret_wm->enable = true;
  2384.  
  2385.         for_each_intel_crtc(dev, intel_crtc) {
  2386.                 const struct intel_pipe_wm *active = &intel_crtc->wm.active;
  2387.                 const struct intel_wm_level *wm = &active->wm[level];
  2388.  
  2389.                 if (!active->pipe_enabled)
  2390.                         continue;
  2391.  
  2392.                 /*
  2393.                  * The watermark values may have been used in the past,
  2394.                  * so we must maintain them in the registers for some
  2395.                  * time even if the level is now disabled.
  2396.                  */
  2397.                 if (!wm->enable)
  2398.                         ret_wm->enable = false;
  2399.  
  2400.                 ret_wm->pri_val = max(ret_wm->pri_val, wm->pri_val);
  2401.                 ret_wm->spr_val = max(ret_wm->spr_val, wm->spr_val);
  2402.                 ret_wm->cur_val = max(ret_wm->cur_val, wm->cur_val);
  2403.                 ret_wm->fbc_val = max(ret_wm->fbc_val, wm->fbc_val);
  2404.         }
  2405. }
  2406.  
  2407. /*
  2408.  * Merge all low power watermarks for all active pipes.
  2409.  */
  2410. static void ilk_wm_merge(struct drm_device *dev,
  2411.                          const struct intel_wm_config *config,
  2412.                          const struct ilk_wm_maximums *max,
  2413.                          struct intel_pipe_wm *merged)
  2414. {
  2415.         struct drm_i915_private *dev_priv = dev->dev_private;
  2416.         int level, max_level = ilk_wm_max_level(dev);
  2417.         int last_enabled_level = max_level;
  2418.  
  2419.         /* ILK/SNB/IVB: LP1+ watermarks only w/ single pipe */
  2420.         if ((INTEL_INFO(dev)->gen <= 6 || IS_IVYBRIDGE(dev)) &&
  2421.             config->num_pipes_active > 1)
  2422.                 return;
  2423.  
  2424.         /* ILK: FBC WM must be disabled always */
  2425.         merged->fbc_wm_enabled = INTEL_INFO(dev)->gen >= 6;
  2426.  
  2427.         /* merge each WM1+ level */
  2428.         for (level = 1; level <= max_level; level++) {
  2429.                 struct intel_wm_level *wm = &merged->wm[level];
  2430.  
  2431.                 ilk_merge_wm_level(dev, level, wm);
  2432.  
  2433.                 if (level > last_enabled_level)
  2434.                         wm->enable = false;
  2435.                 else if (!ilk_validate_wm_level(level, max, wm))
  2436.                         /* make sure all following levels get disabled */
  2437.                         last_enabled_level = level - 1;
  2438.  
  2439.                 /*
  2440.                  * The spec says it is preferred to disable
  2441.                  * FBC WMs instead of disabling a WM level.
  2442.                  */
  2443.                 if (wm->fbc_val > max->fbc) {
  2444.                         if (wm->enable)
  2445.                                 merged->fbc_wm_enabled = false;
  2446.                         wm->fbc_val = 0;
  2447.                 }
  2448.         }
  2449.  
  2450.         /* ILK: LP2+ must be disabled when FBC WM is disabled but FBC enabled */
  2451.         /*
  2452.          * FIXME this is racy. FBC might get enabled later.
  2453.          * What we should check here is whether FBC can be
  2454.          * enabled sometime later.
  2455.          */
  2456.         if (IS_GEN5(dev) && !merged->fbc_wm_enabled &&
  2457.             intel_fbc_enabled(dev_priv)) {
  2458.                 for (level = 2; level <= max_level; level++) {
  2459.                         struct intel_wm_level *wm = &merged->wm[level];
  2460.  
  2461.                         wm->enable = false;
  2462.                 }
  2463.         }
  2464. }
  2465.  
  2466. static int ilk_wm_lp_to_level(int wm_lp, const struct intel_pipe_wm *pipe_wm)
  2467. {
  2468.         /* LP1,LP2,LP3 levels are either 1,2,3 or 1,3,4 */
  2469.         return wm_lp + (wm_lp >= 2 && pipe_wm->wm[4].enable);
  2470. }
  2471.  
  2472. /* The value we need to program into the WM_LPx latency field */
  2473. static unsigned int ilk_wm_lp_latency(struct drm_device *dev, int level)
  2474. {
  2475.         struct drm_i915_private *dev_priv = dev->dev_private;
  2476.  
  2477.         if (IS_HASWELL(dev) || IS_BROADWELL(dev))
  2478.                 return 2 * level;
  2479.         else
  2480.                 return dev_priv->wm.pri_latency[level];
  2481. }
  2482.  
  2483. static void ilk_compute_wm_results(struct drm_device *dev,
  2484.                                    const struct intel_pipe_wm *merged,
  2485.                                    enum intel_ddb_partitioning partitioning,
  2486.                                    struct ilk_wm_values *results)
  2487. {
  2488.         struct intel_crtc *intel_crtc;
  2489.         int level, wm_lp;
  2490.  
  2491.         results->enable_fbc_wm = merged->fbc_wm_enabled;
  2492.         results->partitioning = partitioning;
  2493.  
  2494.         /* LP1+ register values */
  2495.         for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
  2496.                 const struct intel_wm_level *r;
  2497.  
  2498.                 level = ilk_wm_lp_to_level(wm_lp, merged);
  2499.  
  2500.                 r = &merged->wm[level];
  2501.  
  2502.                 /*
  2503.                  * Maintain the watermark values even if the level is
  2504.                  * disabled. Doing otherwise could cause underruns.
  2505.                  */
  2506.                 results->wm_lp[wm_lp - 1] =
  2507.                         (ilk_wm_lp_latency(dev, level) << WM1_LP_LATENCY_SHIFT) |
  2508.                         (r->pri_val << WM1_LP_SR_SHIFT) |
  2509.                         r->cur_val;
  2510.  
  2511.                 if (r->enable)
  2512.                         results->wm_lp[wm_lp - 1] |= WM1_LP_SR_EN;
  2513.  
  2514.                 if (INTEL_INFO(dev)->gen >= 8)
  2515.                         results->wm_lp[wm_lp - 1] |=
  2516.                                 r->fbc_val << WM1_LP_FBC_SHIFT_BDW;
  2517.                 else
  2518.                         results->wm_lp[wm_lp - 1] |=
  2519.                                 r->fbc_val << WM1_LP_FBC_SHIFT;
  2520.  
  2521.                 /*
  2522.                  * Always set WM1S_LP_EN when spr_val != 0, even if the
  2523.                  * level is disabled. Doing otherwise could cause underruns.
  2524.                  */
  2525.                 if (INTEL_INFO(dev)->gen <= 6 && r->spr_val) {
  2526.                         WARN_ON(wm_lp != 1);
  2527.                         results->wm_lp_spr[wm_lp - 1] = WM1S_LP_EN | r->spr_val;
  2528.                 } else
  2529.                         results->wm_lp_spr[wm_lp - 1] = r->spr_val;
  2530.         }
  2531.  
  2532.         /* LP0 register values */
  2533.         for_each_intel_crtc(dev, intel_crtc) {
  2534.                 enum pipe pipe = intel_crtc->pipe;
  2535.                 const struct intel_wm_level *r =
  2536.                         &intel_crtc->wm.active.wm[0];
  2537.  
  2538.                 if (WARN_ON(!r->enable))
  2539.                         continue;
  2540.  
  2541.                 results->wm_linetime[pipe] = intel_crtc->wm.active.linetime;
  2542.  
  2543.                 results->wm_pipe[pipe] =
  2544.                         (r->pri_val << WM0_PIPE_PLANE_SHIFT) |
  2545.                         (r->spr_val << WM0_PIPE_SPRITE_SHIFT) |
  2546.                         r->cur_val;
  2547.         }
  2548. }
  2549.  
  2550. /* Find the result with the highest level enabled. Check for enable_fbc_wm in
  2551.  * case both are at the same level. Prefer r1 in case they're the same. */
  2552. static struct intel_pipe_wm *ilk_find_best_result(struct drm_device *dev,
  2553.                                                   struct intel_pipe_wm *r1,
  2554.                                                   struct intel_pipe_wm *r2)
  2555. {
  2556.         int level, max_level = ilk_wm_max_level(dev);
  2557.         int level1 = 0, level2 = 0;
  2558.  
  2559.         for (level = 1; level <= max_level; level++) {
  2560.                 if (r1->wm[level].enable)
  2561.                         level1 = level;
  2562.                 if (r2->wm[level].enable)
  2563.                         level2 = level;
  2564.         }
  2565.  
  2566.         if (level1 == level2) {
  2567.                 if (r2->fbc_wm_enabled && !r1->fbc_wm_enabled)
  2568.                         return r2;
  2569.                 else
  2570.                         return r1;
  2571.         } else if (level1 > level2) {
  2572.                 return r1;
  2573.         } else {
  2574.                 return r2;
  2575.         }
  2576. }
  2577.  
  2578. /* dirty bits used to track which watermarks need changes */
  2579. #define WM_DIRTY_PIPE(pipe) (1 << (pipe))
  2580. #define WM_DIRTY_LINETIME(pipe) (1 << (8 + (pipe)))
  2581. #define WM_DIRTY_LP(wm_lp) (1 << (15 + (wm_lp)))
  2582. #define WM_DIRTY_LP_ALL (WM_DIRTY_LP(1) | WM_DIRTY_LP(2) | WM_DIRTY_LP(3))
  2583. #define WM_DIRTY_FBC (1 << 24)
  2584. #define WM_DIRTY_DDB (1 << 25)
  2585.  
  2586. static unsigned int ilk_compute_wm_dirty(struct drm_i915_private *dev_priv,
  2587.                                          const struct ilk_wm_values *old,
  2588.                                          const struct ilk_wm_values *new)
  2589. {
  2590.         unsigned int dirty = 0;
  2591.         enum pipe pipe;
  2592.         int wm_lp;
  2593.  
  2594.         for_each_pipe(dev_priv, pipe) {
  2595.                 if (old->wm_linetime[pipe] != new->wm_linetime[pipe]) {
  2596.                         dirty |= WM_DIRTY_LINETIME(pipe);
  2597.                         /* Must disable LP1+ watermarks too */
  2598.                         dirty |= WM_DIRTY_LP_ALL;
  2599.                 }
  2600.  
  2601.                 if (old->wm_pipe[pipe] != new->wm_pipe[pipe]) {
  2602.                         dirty |= WM_DIRTY_PIPE(pipe);
  2603.                         /* Must disable LP1+ watermarks too */
  2604.                         dirty |= WM_DIRTY_LP_ALL;
  2605.                 }
  2606.         }
  2607.  
  2608.         if (old->enable_fbc_wm != new->enable_fbc_wm) {
  2609.                 dirty |= WM_DIRTY_FBC;
  2610.                 /* Must disable LP1+ watermarks too */
  2611.                 dirty |= WM_DIRTY_LP_ALL;
  2612.         }
  2613.  
  2614.         if (old->partitioning != new->partitioning) {
  2615.                 dirty |= WM_DIRTY_DDB;
  2616.                 /* Must disable LP1+ watermarks too */
  2617.                 dirty |= WM_DIRTY_LP_ALL;
  2618.         }
  2619.  
  2620.         /* LP1+ watermarks already deemed dirty, no need to continue */
  2621.         if (dirty & WM_DIRTY_LP_ALL)
  2622.                 return dirty;
  2623.  
  2624.         /* Find the lowest numbered LP1+ watermark in need of an update... */
  2625.         for (wm_lp = 1; wm_lp <= 3; wm_lp++) {
  2626.                 if (old->wm_lp[wm_lp - 1] != new->wm_lp[wm_lp - 1] ||
  2627.                     old->wm_lp_spr[wm_lp - 1] != new->wm_lp_spr[wm_lp - 1])
  2628.                         break;
  2629.         }
  2630.  
  2631.         /* ...and mark it and all higher numbered LP1+ watermarks as dirty */
  2632.         for (; wm_lp <= 3; wm_lp++)
  2633.                 dirty |= WM_DIRTY_LP(wm_lp);
  2634.  
  2635.         return dirty;
  2636. }
  2637.  
  2638. static bool _ilk_disable_lp_wm(struct drm_i915_private *dev_priv,
  2639.                                unsigned int dirty)
  2640. {
  2641.         struct ilk_wm_values *previous = &dev_priv->wm.hw;
  2642.         bool changed = false;
  2643.  
  2644.         if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] & WM1_LP_SR_EN) {
  2645.                 previous->wm_lp[2] &= ~WM1_LP_SR_EN;
  2646.                 I915_WRITE(WM3_LP_ILK, previous->wm_lp[2]);
  2647.                 changed = true;
  2648.         }
  2649.         if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] & WM1_LP_SR_EN) {
  2650.                 previous->wm_lp[1] &= ~WM1_LP_SR_EN;
  2651.                 I915_WRITE(WM2_LP_ILK, previous->wm_lp[1]);
  2652.                 changed = true;
  2653.         }
  2654.         if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] & WM1_LP_SR_EN) {
  2655.                 previous->wm_lp[0] &= ~WM1_LP_SR_EN;
  2656.                 I915_WRITE(WM1_LP_ILK, previous->wm_lp[0]);
  2657.                 changed = true;
  2658.         }
  2659.  
  2660.         /*
  2661.          * Don't touch WM1S_LP_EN here.
  2662.          * Doing so could cause underruns.
  2663.          */
  2664.  
  2665.         return changed;
  2666. }
  2667.  
  2668. /*
  2669.  * The spec says we shouldn't write when we don't need, because every write
  2670.  * causes WMs to be re-evaluated, expending some power.
  2671.  */
  2672. static void ilk_write_wm_values(struct drm_i915_private *dev_priv,
  2673.                                 struct ilk_wm_values *results)
  2674. {
  2675.         struct drm_device *dev = dev_priv->dev;
  2676.         struct ilk_wm_values *previous = &dev_priv->wm.hw;
  2677.         unsigned int dirty;
  2678.         uint32_t val;
  2679.  
  2680.         dirty = ilk_compute_wm_dirty(dev_priv, previous, results);
  2681.         if (!dirty)
  2682.                 return;
  2683.  
  2684.         _ilk_disable_lp_wm(dev_priv, dirty);
  2685.  
  2686.         if (dirty & WM_DIRTY_PIPE(PIPE_A))
  2687.                 I915_WRITE(WM0_PIPEA_ILK, results->wm_pipe[0]);
  2688.         if (dirty & WM_DIRTY_PIPE(PIPE_B))
  2689.                 I915_WRITE(WM0_PIPEB_ILK, results->wm_pipe[1]);
  2690.         if (dirty & WM_DIRTY_PIPE(PIPE_C))
  2691.                 I915_WRITE(WM0_PIPEC_IVB, results->wm_pipe[2]);
  2692.  
  2693.         if (dirty & WM_DIRTY_LINETIME(PIPE_A))
  2694.                 I915_WRITE(PIPE_WM_LINETIME(PIPE_A), results->wm_linetime[0]);
  2695.         if (dirty & WM_DIRTY_LINETIME(PIPE_B))
  2696.                 I915_WRITE(PIPE_WM_LINETIME(PIPE_B), results->wm_linetime[1]);
  2697.         if (dirty & WM_DIRTY_LINETIME(PIPE_C))
  2698.                 I915_WRITE(PIPE_WM_LINETIME(PIPE_C), results->wm_linetime[2]);
  2699.  
  2700.         if (dirty & WM_DIRTY_DDB) {
  2701.                 if (IS_HASWELL(dev) || IS_BROADWELL(dev)) {
  2702.                         val = I915_READ(WM_MISC);
  2703.                         if (results->partitioning == INTEL_DDB_PART_1_2)
  2704.                                 val &= ~WM_MISC_DATA_PARTITION_5_6;
  2705.                         else
  2706.                                 val |= WM_MISC_DATA_PARTITION_5_6;
  2707.                         I915_WRITE(WM_MISC, val);
  2708.                 } else {
  2709.                         val = I915_READ(DISP_ARB_CTL2);
  2710.                         if (results->partitioning == INTEL_DDB_PART_1_2)
  2711.                                 val &= ~DISP_DATA_PARTITION_5_6;
  2712.                         else
  2713.                                 val |= DISP_DATA_PARTITION_5_6;
  2714.                         I915_WRITE(DISP_ARB_CTL2, val);
  2715.                 }
  2716.         }
  2717.  
  2718.         if (dirty & WM_DIRTY_FBC) {
  2719.                 val = I915_READ(DISP_ARB_CTL);
  2720.                 if (results->enable_fbc_wm)
  2721.                         val &= ~DISP_FBC_WM_DIS;
  2722.                 else
  2723.                         val |= DISP_FBC_WM_DIS;
  2724.                 I915_WRITE(DISP_ARB_CTL, val);
  2725.         }
  2726.  
  2727.         if (dirty & WM_DIRTY_LP(1) &&
  2728.             previous->wm_lp_spr[0] != results->wm_lp_spr[0])
  2729.                 I915_WRITE(WM1S_LP_ILK, results->wm_lp_spr[0]);
  2730.  
  2731.         if (INTEL_INFO(dev)->gen >= 7) {
  2732.                 if (dirty & WM_DIRTY_LP(2) && previous->wm_lp_spr[1] != results->wm_lp_spr[1])
  2733.                         I915_WRITE(WM2S_LP_IVB, results->wm_lp_spr[1]);
  2734.                 if (dirty & WM_DIRTY_LP(3) && previous->wm_lp_spr[2] != results->wm_lp_spr[2])
  2735.                         I915_WRITE(WM3S_LP_IVB, results->wm_lp_spr[2]);
  2736.         }
  2737.  
  2738.         if (dirty & WM_DIRTY_LP(1) && previous->wm_lp[0] != results->wm_lp[0])
  2739.                 I915_WRITE(WM1_LP_ILK, results->wm_lp[0]);
  2740.         if (dirty & WM_DIRTY_LP(2) && previous->wm_lp[1] != results->wm_lp[1])
  2741.                 I915_WRITE(WM2_LP_ILK, results->wm_lp[1]);
  2742.         if (dirty & WM_DIRTY_LP(3) && previous->wm_lp[2] != results->wm_lp[2])
  2743.                 I915_WRITE(WM3_LP_ILK, results->wm_lp[2]);
  2744.  
  2745.         dev_priv->wm.hw = *results;
  2746. }
  2747.  
  2748. static bool ilk_disable_lp_wm(struct drm_device *dev)
  2749. {
  2750.         struct drm_i915_private *dev_priv = dev->dev_private;
  2751.  
  2752.         return _ilk_disable_lp_wm(dev_priv, WM_DIRTY_LP_ALL);
  2753. }
  2754.  
  2755. /*
  2756.  * On gen9, we need to allocate Display Data Buffer (DDB) portions to the
  2757.  * different active planes.
  2758.  */
  2759.  
  2760. #define SKL_DDB_SIZE            896     /* in blocks */
  2761. #define BXT_DDB_SIZE            512
  2762.  
  2763. static void
  2764. skl_ddb_get_pipe_allocation_limits(struct drm_device *dev,
  2765.                                    struct drm_crtc *for_crtc,
  2766.                                    const struct intel_wm_config *config,
  2767.                                    const struct skl_pipe_wm_parameters *params,
  2768.                                    struct skl_ddb_entry *alloc /* out */)
  2769. {
  2770.         struct drm_crtc *crtc;
  2771.         unsigned int pipe_size, ddb_size;
  2772.         int nth_active_pipe;
  2773.  
  2774.         if (!params->active) {
  2775.                 alloc->start = 0;
  2776.                 alloc->end = 0;
  2777.                 return;
  2778.         }
  2779.  
  2780.         if (IS_BROXTON(dev))
  2781.                 ddb_size = BXT_DDB_SIZE;
  2782.         else
  2783.                 ddb_size = SKL_DDB_SIZE;
  2784.  
  2785.         ddb_size -= 4; /* 4 blocks for bypass path allocation */
  2786.  
  2787.         nth_active_pipe = 0;
  2788.         for_each_crtc(dev, crtc) {
  2789.                 if (!to_intel_crtc(crtc)->active)
  2790.                         continue;
  2791.  
  2792.                 if (crtc == for_crtc)
  2793.                         break;
  2794.  
  2795.                 nth_active_pipe++;
  2796.         }
  2797.  
  2798.         pipe_size = ddb_size / config->num_pipes_active;
  2799.         alloc->start = nth_active_pipe * ddb_size / config->num_pipes_active;
  2800.         alloc->end = alloc->start + pipe_size;
  2801. }
  2802.  
  2803. static unsigned int skl_cursor_allocation(const struct intel_wm_config *config)
  2804. {
  2805.         if (config->num_pipes_active == 1)
  2806.                 return 32;
  2807.  
  2808.         return 8;
  2809. }
  2810.  
  2811. static void skl_ddb_entry_init_from_hw(struct skl_ddb_entry *entry, u32 reg)
  2812. {
  2813.         entry->start = reg & 0x3ff;
  2814.         entry->end = (reg >> 16) & 0x3ff;
  2815.         if (entry->end)
  2816.                 entry->end += 1;
  2817. }
  2818.  
  2819. void skl_ddb_get_hw_state(struct drm_i915_private *dev_priv,
  2820.                           struct skl_ddb_allocation *ddb /* out */)
  2821. {
  2822.         enum pipe pipe;
  2823.         int plane;
  2824.         u32 val;
  2825.  
  2826.         memset(ddb, 0, sizeof(*ddb));
  2827.  
  2828.         for_each_pipe(dev_priv, pipe) {
  2829.                 if (!intel_display_power_is_enabled(dev_priv, POWER_DOMAIN_PIPE(pipe)))
  2830.                         continue;
  2831.  
  2832.                 for_each_plane(dev_priv, pipe, plane) {
  2833.                         val = I915_READ(PLANE_BUF_CFG(pipe, plane));
  2834.                         skl_ddb_entry_init_from_hw(&ddb->plane[pipe][plane],
  2835.                                                    val);
  2836.                 }
  2837.  
  2838.                 val = I915_READ(CUR_BUF_CFG(pipe));
  2839.                 skl_ddb_entry_init_from_hw(&ddb->plane[pipe][PLANE_CURSOR],
  2840.                                            val);
  2841.         }
  2842. }
  2843.  
  2844. static unsigned int
  2845. skl_plane_relative_data_rate(const struct intel_plane_wm_parameters *p, int y)
  2846. {
  2847.  
  2848.         /* for planar format */
  2849.         if (p->y_bytes_per_pixel) {
  2850.                 if (y)  /* y-plane data rate */
  2851.                         return p->horiz_pixels * p->vert_pixels * p->y_bytes_per_pixel;
  2852.                 else    /* uv-plane data rate */
  2853.                         return (p->horiz_pixels/2) * (p->vert_pixels/2) * p->bytes_per_pixel;
  2854.         }
  2855.  
  2856.         /* for packed formats */
  2857.         return p->horiz_pixels * p->vert_pixels * p->bytes_per_pixel;
  2858. }
  2859.  
  2860. /*
  2861.  * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching
  2862.  * a 8192x4096@32bpp framebuffer:
  2863.  *   3 * 4096 * 8192  * 4 < 2^32
  2864.  */
  2865. static unsigned int
  2866. skl_get_total_relative_data_rate(struct intel_crtc *intel_crtc,
  2867.                                  const struct skl_pipe_wm_parameters *params)
  2868. {
  2869.         unsigned int total_data_rate = 0;
  2870.         int plane;
  2871.  
  2872.         for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) {
  2873.                 const struct intel_plane_wm_parameters *p;
  2874.  
  2875.                 p = &params->plane[plane];
  2876.                 if (!p->enabled)
  2877.                         continue;
  2878.  
  2879.                 total_data_rate += skl_plane_relative_data_rate(p, 0); /* packed/uv */
  2880.                 if (p->y_bytes_per_pixel) {
  2881.                         total_data_rate += skl_plane_relative_data_rate(p, 1); /* y-plane */
  2882.                 }
  2883.         }
  2884.  
  2885.         return total_data_rate;
  2886. }
  2887.  
  2888. static void
  2889. skl_allocate_pipe_ddb(struct drm_crtc *crtc,
  2890.                       const struct intel_wm_config *config,
  2891.                       const struct skl_pipe_wm_parameters *params,
  2892.                       struct skl_ddb_allocation *ddb /* out */)
  2893. {
  2894.         struct drm_device *dev = crtc->dev;
  2895.         struct drm_i915_private *dev_priv = dev->dev_private;
  2896.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  2897.         enum pipe pipe = intel_crtc->pipe;
  2898.         struct skl_ddb_entry *alloc = &ddb->pipe[pipe];
  2899.         uint16_t alloc_size, start, cursor_blocks;
  2900.         uint16_t minimum[I915_MAX_PLANES];
  2901.         uint16_t y_minimum[I915_MAX_PLANES];
  2902.         unsigned int total_data_rate;
  2903.         int plane;
  2904.  
  2905.         skl_ddb_get_pipe_allocation_limits(dev, crtc, config, params, alloc);
  2906.         alloc_size = skl_ddb_entry_size(alloc);
  2907.         if (alloc_size == 0) {
  2908.                 memset(ddb->plane[pipe], 0, sizeof(ddb->plane[pipe]));
  2909.                 memset(&ddb->plane[pipe][PLANE_CURSOR], 0,
  2910.                        sizeof(ddb->plane[pipe][PLANE_CURSOR]));
  2911.                 return;
  2912.         }
  2913.  
  2914.         cursor_blocks = skl_cursor_allocation(config);
  2915.         ddb->plane[pipe][PLANE_CURSOR].start = alloc->end - cursor_blocks;
  2916.         ddb->plane[pipe][PLANE_CURSOR].end = alloc->end;
  2917.  
  2918.         alloc_size -= cursor_blocks;
  2919.         alloc->end -= cursor_blocks;
  2920.  
  2921.         /* 1. Allocate the mininum required blocks for each active plane */
  2922.         for_each_plane(dev_priv, pipe, plane) {
  2923.                 const struct intel_plane_wm_parameters *p;
  2924.  
  2925.                 p = &params->plane[plane];
  2926.                 if (!p->enabled)
  2927.                         continue;
  2928.  
  2929.                 minimum[plane] = 8;
  2930.                 alloc_size -= minimum[plane];
  2931.                 y_minimum[plane] = p->y_bytes_per_pixel ? 8 : 0;
  2932.                 alloc_size -= y_minimum[plane];
  2933.         }
  2934.  
  2935.         /*
  2936.          * 2. Distribute the remaining space in proportion to the amount of
  2937.          * data each plane needs to fetch from memory.
  2938.          *
  2939.          * FIXME: we may not allocate every single block here.
  2940.          */
  2941.         total_data_rate = skl_get_total_relative_data_rate(intel_crtc, params);
  2942.  
  2943.         start = alloc->start;
  2944.         for (plane = 0; plane < intel_num_planes(intel_crtc); plane++) {
  2945.                 const struct intel_plane_wm_parameters *p;
  2946.                 unsigned int data_rate, y_data_rate;
  2947.                 uint16_t plane_blocks, y_plane_blocks = 0;
  2948.  
  2949.                 p = &params->plane[plane];
  2950.                 if (!p->enabled)
  2951.                         continue;
  2952.  
  2953.                 data_rate = skl_plane_relative_data_rate(p, 0);
  2954.  
  2955.                 /*
  2956.                  * allocation for (packed formats) or (uv-plane part of planar format):
  2957.                  * promote the expression to 64 bits to avoid overflowing, the
  2958.                  * result is < available as data_rate / total_data_rate < 1
  2959.                  */
  2960.                 plane_blocks = minimum[plane];
  2961.                 plane_blocks += div_u64((uint64_t)alloc_size * data_rate,
  2962.                                         total_data_rate);
  2963.  
  2964.                 ddb->plane[pipe][plane].start = start;
  2965.                 ddb->plane[pipe][plane].end = start + plane_blocks;
  2966.  
  2967.                 start += plane_blocks;
  2968.  
  2969.                 /*
  2970.                  * allocation for y_plane part of planar format:
  2971.                  */
  2972.                 if (p->y_bytes_per_pixel) {
  2973.                         y_data_rate = skl_plane_relative_data_rate(p, 1);
  2974.                         y_plane_blocks = y_minimum[plane];
  2975.                         y_plane_blocks += div_u64((uint64_t)alloc_size * y_data_rate,
  2976.                                                 total_data_rate);
  2977.  
  2978.                         ddb->y_plane[pipe][plane].start = start;
  2979.                         ddb->y_plane[pipe][plane].end = start + y_plane_blocks;
  2980.  
  2981.                         start += y_plane_blocks;
  2982.                 }
  2983.  
  2984.         }
  2985.  
  2986. }
  2987.  
  2988. static uint32_t skl_pipe_pixel_rate(const struct intel_crtc_state *config)
  2989. {
  2990.         /* TODO: Take into account the scalers once we support them */
  2991.         return config->base.adjusted_mode.crtc_clock;
  2992. }
  2993.  
  2994. /*
  2995.  * The max latency should be 257 (max the punit can code is 255 and we add 2us
  2996.  * for the read latency) and bytes_per_pixel should always be <= 8, so that
  2997.  * should allow pixel_rate up to ~2 GHz which seems sufficient since max
  2998.  * 2xcdclk is 1350 MHz and the pixel rate should never exceed that.
  2999. */
  3000. static uint32_t skl_wm_method1(uint32_t pixel_rate, uint8_t bytes_per_pixel,
  3001.                                uint32_t latency)
  3002. {
  3003.         uint32_t wm_intermediate_val, ret;
  3004.  
  3005.         if (latency == 0)
  3006.                 return UINT_MAX;
  3007.  
  3008.         wm_intermediate_val = latency * pixel_rate * bytes_per_pixel / 512;
  3009.         ret = DIV_ROUND_UP(wm_intermediate_val, 1000);
  3010.  
  3011.         return ret;
  3012. }
  3013.  
  3014. static uint32_t skl_wm_method2(uint32_t pixel_rate, uint32_t pipe_htotal,
  3015.                                uint32_t horiz_pixels, uint8_t bytes_per_pixel,
  3016.                                uint64_t tiling, uint32_t latency)
  3017. {
  3018.         uint32_t ret;
  3019.         uint32_t plane_bytes_per_line, plane_blocks_per_line;
  3020.         uint32_t wm_intermediate_val;
  3021.  
  3022.         if (latency == 0)
  3023.                 return UINT_MAX;
  3024.  
  3025.         plane_bytes_per_line = horiz_pixels * bytes_per_pixel;
  3026.  
  3027.         if (tiling == I915_FORMAT_MOD_Y_TILED ||
  3028.             tiling == I915_FORMAT_MOD_Yf_TILED) {
  3029.                 plane_bytes_per_line *= 4;
  3030.                 plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
  3031.                 plane_blocks_per_line /= 4;
  3032.         } else {
  3033.                 plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
  3034.         }
  3035.  
  3036.         wm_intermediate_val = latency * pixel_rate;
  3037.         ret = DIV_ROUND_UP(wm_intermediate_val, pipe_htotal * 1000) *
  3038.                                 plane_blocks_per_line;
  3039.  
  3040.         return ret;
  3041. }
  3042.  
  3043. static bool skl_ddb_allocation_changed(const struct skl_ddb_allocation *new_ddb,
  3044.                                        const struct intel_crtc *intel_crtc)
  3045. {
  3046.         struct drm_device *dev = intel_crtc->base.dev;
  3047.         struct drm_i915_private *dev_priv = dev->dev_private;
  3048.         const struct skl_ddb_allocation *cur_ddb = &dev_priv->wm.skl_hw.ddb;
  3049.         enum pipe pipe = intel_crtc->pipe;
  3050.  
  3051.         if (memcmp(new_ddb->plane[pipe], cur_ddb->plane[pipe],
  3052.                    sizeof(new_ddb->plane[pipe])))
  3053.                 return true;
  3054.  
  3055.         if (memcmp(&new_ddb->plane[pipe][PLANE_CURSOR], &cur_ddb->plane[pipe][PLANE_CURSOR],
  3056.                     sizeof(new_ddb->plane[pipe][PLANE_CURSOR])))
  3057.                 return true;
  3058.  
  3059.         return false;
  3060. }
  3061.  
  3062. static void skl_compute_wm_global_parameters(struct drm_device *dev,
  3063.                                              struct intel_wm_config *config)
  3064. {
  3065.         struct drm_crtc *crtc;
  3066.         struct drm_plane *plane;
  3067.  
  3068.         list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
  3069.                 config->num_pipes_active += to_intel_crtc(crtc)->active;
  3070.  
  3071.         /* FIXME: I don't think we need those two global parameters on SKL */
  3072.         list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
  3073.                 struct intel_plane *intel_plane = to_intel_plane(plane);
  3074.  
  3075.                 config->sprites_enabled |= intel_plane->wm.enabled;
  3076.                 config->sprites_scaled |= intel_plane->wm.scaled;
  3077.         }
  3078. }
  3079.  
  3080. static void skl_compute_wm_pipe_parameters(struct drm_crtc *crtc,
  3081.                                            struct skl_pipe_wm_parameters *p)
  3082. {
  3083.         struct drm_device *dev = crtc->dev;
  3084.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  3085.         enum pipe pipe = intel_crtc->pipe;
  3086.         struct drm_plane *plane;
  3087.         struct drm_framebuffer *fb;
  3088.         int i = 1; /* Index for sprite planes start */
  3089.  
  3090.         p->active = intel_crtc->active;
  3091.         if (p->active) {
  3092.                 p->pipe_htotal = intel_crtc->config->base.adjusted_mode.crtc_htotal;
  3093.                 p->pixel_rate = skl_pipe_pixel_rate(intel_crtc->config);
  3094.  
  3095.                 fb = crtc->primary->state->fb;
  3096.                 /* For planar: Bpp is for uv plane, y_Bpp is for y plane */
  3097.                 if (fb) {
  3098.                         p->plane[0].enabled = true;
  3099.                         p->plane[0].bytes_per_pixel = fb->pixel_format == DRM_FORMAT_NV12 ?
  3100.                                 drm_format_plane_cpp(fb->pixel_format, 1) :
  3101.                                 drm_format_plane_cpp(fb->pixel_format, 0);
  3102.                         p->plane[0].y_bytes_per_pixel = fb->pixel_format == DRM_FORMAT_NV12 ?
  3103.                                 drm_format_plane_cpp(fb->pixel_format, 0) : 0;
  3104.                         p->plane[0].tiling = fb->modifier[0];
  3105.                 } else {
  3106.                         p->plane[0].enabled = false;
  3107.                         p->plane[0].bytes_per_pixel = 0;
  3108.                         p->plane[0].y_bytes_per_pixel = 0;
  3109.                         p->plane[0].tiling = DRM_FORMAT_MOD_NONE;
  3110.                 }
  3111.                 p->plane[0].horiz_pixels = intel_crtc->config->pipe_src_w;
  3112.                 p->plane[0].vert_pixels = intel_crtc->config->pipe_src_h;
  3113.                 p->plane[0].rotation = crtc->primary->state->rotation;
  3114.  
  3115.                 fb = crtc->cursor->state->fb;
  3116.                 p->plane[PLANE_CURSOR].y_bytes_per_pixel = 0;
  3117.                 if (fb) {
  3118.                         p->plane[PLANE_CURSOR].enabled = true;
  3119.                         p->plane[PLANE_CURSOR].bytes_per_pixel = fb->bits_per_pixel / 8;
  3120.                         p->plane[PLANE_CURSOR].horiz_pixels = crtc->cursor->state->crtc_w;
  3121.                         p->plane[PLANE_CURSOR].vert_pixels = crtc->cursor->state->crtc_h;
  3122.                 } else {
  3123.                         p->plane[PLANE_CURSOR].enabled = false;
  3124.                         p->plane[PLANE_CURSOR].bytes_per_pixel = 0;
  3125.                         p->plane[PLANE_CURSOR].horiz_pixels = 64;
  3126.                         p->plane[PLANE_CURSOR].vert_pixels = 64;
  3127.                 }
  3128.         }
  3129.  
  3130.         list_for_each_entry(plane, &dev->mode_config.plane_list, head) {
  3131.                 struct intel_plane *intel_plane = to_intel_plane(plane);
  3132.  
  3133.                 if (intel_plane->pipe == pipe &&
  3134.                         plane->type == DRM_PLANE_TYPE_OVERLAY)
  3135.                         p->plane[i++] = intel_plane->wm;
  3136.         }
  3137. }
  3138.  
  3139. static bool skl_compute_plane_wm(const struct drm_i915_private *dev_priv,
  3140.                                  struct skl_pipe_wm_parameters *p,
  3141.                                  struct intel_plane_wm_parameters *p_params,
  3142.                                  uint16_t ddb_allocation,
  3143.                                  int level,
  3144.                                  uint16_t *out_blocks, /* out */
  3145.                                  uint8_t *out_lines /* out */)
  3146. {
  3147.         uint32_t latency = dev_priv->wm.skl_latency[level];
  3148.         uint32_t method1, method2;
  3149.         uint32_t plane_bytes_per_line, plane_blocks_per_line;
  3150.         uint32_t res_blocks, res_lines;
  3151.         uint32_t selected_result;
  3152.         uint8_t bytes_per_pixel;
  3153.  
  3154.         if (latency == 0 || !p->active || !p_params->enabled)
  3155.                 return false;
  3156.  
  3157.         bytes_per_pixel = p_params->y_bytes_per_pixel ?
  3158.                 p_params->y_bytes_per_pixel :
  3159.                 p_params->bytes_per_pixel;
  3160.         method1 = skl_wm_method1(p->pixel_rate,
  3161.                                  bytes_per_pixel,
  3162.                                  latency);
  3163.         method2 = skl_wm_method2(p->pixel_rate,
  3164.                                  p->pipe_htotal,
  3165.                                  p_params->horiz_pixels,
  3166.                                  bytes_per_pixel,
  3167.                                  p_params->tiling,
  3168.                                  latency);
  3169.  
  3170.         plane_bytes_per_line = p_params->horiz_pixels * bytes_per_pixel;
  3171.         plane_blocks_per_line = DIV_ROUND_UP(plane_bytes_per_line, 512);
  3172.  
  3173.         if (p_params->tiling == I915_FORMAT_MOD_Y_TILED ||
  3174.             p_params->tiling == I915_FORMAT_MOD_Yf_TILED) {
  3175.                 uint32_t min_scanlines = 4;
  3176.                 uint32_t y_tile_minimum;
  3177.                 if (intel_rotation_90_or_270(p_params->rotation)) {
  3178.                         switch (p_params->bytes_per_pixel) {
  3179.                         case 1:
  3180.                                 min_scanlines = 16;
  3181.                                 break;
  3182.                         case 2:
  3183.                                 min_scanlines = 8;
  3184.                                 break;
  3185.                         case 8:
  3186.                                 WARN(1, "Unsupported pixel depth for rotation");
  3187.                         }
  3188.                 }
  3189.                 y_tile_minimum = plane_blocks_per_line * min_scanlines;
  3190.                 selected_result = max(method2, y_tile_minimum);
  3191.         } else {
  3192.                 if ((ddb_allocation / plane_blocks_per_line) >= 1)
  3193.                         selected_result = min(method1, method2);
  3194.                 else
  3195.                         selected_result = method1;
  3196.         }
  3197.  
  3198.         res_blocks = selected_result + 1;
  3199.         res_lines = DIV_ROUND_UP(selected_result, plane_blocks_per_line);
  3200.  
  3201.         if (level >= 1 && level <= 7) {
  3202.                 if (p_params->tiling == I915_FORMAT_MOD_Y_TILED ||
  3203.                     p_params->tiling == I915_FORMAT_MOD_Yf_TILED)
  3204.                         res_lines += 4;
  3205.                 else
  3206.                         res_blocks++;
  3207.         }
  3208.  
  3209.         if (res_blocks >= ddb_allocation || res_lines > 31)
  3210.                 return false;
  3211.  
  3212.         *out_blocks = res_blocks;
  3213.         *out_lines = res_lines;
  3214.  
  3215.         return true;
  3216. }
  3217.  
  3218. static void skl_compute_wm_level(const struct drm_i915_private *dev_priv,
  3219.                                  struct skl_ddb_allocation *ddb,
  3220.                                  struct skl_pipe_wm_parameters *p,
  3221.                                  enum pipe pipe,
  3222.                                  int level,
  3223.                                  int num_planes,
  3224.                                  struct skl_wm_level *result)
  3225. {
  3226.         uint16_t ddb_blocks;
  3227.         int i;
  3228.  
  3229.         for (i = 0; i < num_planes; i++) {
  3230.                 ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][i]);
  3231.  
  3232.                 result->plane_en[i] = skl_compute_plane_wm(dev_priv,
  3233.                                                 p, &p->plane[i],
  3234.                                                 ddb_blocks,
  3235.                                                 level,
  3236.                                                 &result->plane_res_b[i],
  3237.                                                 &result->plane_res_l[i]);
  3238.         }
  3239.  
  3240.         ddb_blocks = skl_ddb_entry_size(&ddb->plane[pipe][PLANE_CURSOR]);
  3241.         result->plane_en[PLANE_CURSOR] = skl_compute_plane_wm(dev_priv, p,
  3242.                                                  &p->plane[PLANE_CURSOR],
  3243.                                                  ddb_blocks, level,
  3244.                                                  &result->plane_res_b[PLANE_CURSOR],
  3245.                                                  &result->plane_res_l[PLANE_CURSOR]);
  3246. }
  3247.  
  3248. static uint32_t
  3249. skl_compute_linetime_wm(struct drm_crtc *crtc, struct skl_pipe_wm_parameters *p)
  3250. {
  3251.         if (!to_intel_crtc(crtc)->active)
  3252.                 return 0;
  3253.  
  3254.         if (WARN_ON(p->pixel_rate == 0))
  3255.                 return 0;
  3256.  
  3257.         return DIV_ROUND_UP(8 * p->pipe_htotal * 1000, p->pixel_rate);
  3258. }
  3259.  
  3260. static void skl_compute_transition_wm(struct drm_crtc *crtc,
  3261.                                       struct skl_pipe_wm_parameters *params,
  3262.                                       struct skl_wm_level *trans_wm /* out */)
  3263. {
  3264.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  3265.         int i;
  3266.  
  3267.         if (!params->active)
  3268.                 return;
  3269.  
  3270.         /* Until we know more, just disable transition WMs */
  3271.         for (i = 0; i < intel_num_planes(intel_crtc); i++)
  3272.                 trans_wm->plane_en[i] = false;
  3273.         trans_wm->plane_en[PLANE_CURSOR] = false;
  3274. }
  3275.  
  3276. static void skl_compute_pipe_wm(struct drm_crtc *crtc,
  3277.                                 struct skl_ddb_allocation *ddb,
  3278.                                 struct skl_pipe_wm_parameters *params,
  3279.                                 struct skl_pipe_wm *pipe_wm)
  3280. {
  3281.         struct drm_device *dev = crtc->dev;
  3282.         const struct drm_i915_private *dev_priv = dev->dev_private;
  3283.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  3284.         int level, max_level = ilk_wm_max_level(dev);
  3285.  
  3286.         for (level = 0; level <= max_level; level++) {
  3287.                 skl_compute_wm_level(dev_priv, ddb, params, intel_crtc->pipe,
  3288.                                      level, intel_num_planes(intel_crtc),
  3289.                                      &pipe_wm->wm[level]);
  3290.         }
  3291.         pipe_wm->linetime = skl_compute_linetime_wm(crtc, params);
  3292.  
  3293.         skl_compute_transition_wm(crtc, params, &pipe_wm->trans_wm);
  3294. }
  3295.  
  3296. static void skl_compute_wm_results(struct drm_device *dev,
  3297.                                    struct skl_pipe_wm_parameters *p,
  3298.                                    struct skl_pipe_wm *p_wm,
  3299.                                    struct skl_wm_values *r,
  3300.                                    struct intel_crtc *intel_crtc)
  3301. {
  3302.         int level, max_level = ilk_wm_max_level(dev);
  3303.         enum pipe pipe = intel_crtc->pipe;
  3304.         uint32_t temp;
  3305.         int i;
  3306.  
  3307.         for (level = 0; level <= max_level; level++) {
  3308.                 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
  3309.                         temp = 0;
  3310.  
  3311.                         temp |= p_wm->wm[level].plane_res_l[i] <<
  3312.                                         PLANE_WM_LINES_SHIFT;
  3313.                         temp |= p_wm->wm[level].plane_res_b[i];
  3314.                         if (p_wm->wm[level].plane_en[i])
  3315.                                 temp |= PLANE_WM_EN;
  3316.  
  3317.                         r->plane[pipe][i][level] = temp;
  3318.                 }
  3319.  
  3320.                 temp = 0;
  3321.  
  3322.                 temp |= p_wm->wm[level].plane_res_l[PLANE_CURSOR] << PLANE_WM_LINES_SHIFT;
  3323.                 temp |= p_wm->wm[level].plane_res_b[PLANE_CURSOR];
  3324.  
  3325.                 if (p_wm->wm[level].plane_en[PLANE_CURSOR])
  3326.                         temp |= PLANE_WM_EN;
  3327.  
  3328.                 r->plane[pipe][PLANE_CURSOR][level] = temp;
  3329.  
  3330.         }
  3331.  
  3332.         /* transition WMs */
  3333.         for (i = 0; i < intel_num_planes(intel_crtc); i++) {
  3334.                 temp = 0;
  3335.                 temp |= p_wm->trans_wm.plane_res_l[i] << PLANE_WM_LINES_SHIFT;
  3336.                 temp |= p_wm->trans_wm.plane_res_b[i];
  3337.                 if (p_wm->trans_wm.plane_en[i])
  3338.                         temp |= PLANE_WM_EN;
  3339.  
  3340.                 r->plane_trans[pipe][i] = temp;
  3341.         }
  3342.  
  3343.         temp = 0;
  3344.         temp |= p_wm->trans_wm.plane_res_l[PLANE_CURSOR] << PLANE_WM_LINES_SHIFT;
  3345.         temp |= p_wm->trans_wm.plane_res_b[PLANE_CURSOR];
  3346.         if (p_wm->trans_wm.plane_en[PLANE_CURSOR])
  3347.                 temp |= PLANE_WM_EN;
  3348.  
  3349.         r->plane_trans[pipe][PLANE_CURSOR] = temp;
  3350.  
  3351.         r->wm_linetime[pipe] = p_wm->linetime;
  3352. }
  3353.  
  3354. static void skl_ddb_entry_write(struct drm_i915_private *dev_priv, uint32_t reg,
  3355.                                 const struct skl_ddb_entry *entry)
  3356. {
  3357.         if (entry->end)
  3358.                 I915_WRITE(reg, (entry->end - 1) << 16 | entry->start);
  3359.         else
  3360.                 I915_WRITE(reg, 0);
  3361. }
  3362.  
  3363. static void skl_write_wm_values(struct drm_i915_private *dev_priv,
  3364.                                 const struct skl_wm_values *new)
  3365. {
  3366.         struct drm_device *dev = dev_priv->dev;
  3367.         struct intel_crtc *crtc;
  3368.  
  3369.         list_for_each_entry(crtc, &dev->mode_config.crtc_list, base.head) {
  3370.                 int i, level, max_level = ilk_wm_max_level(dev);
  3371.                 enum pipe pipe = crtc->pipe;
  3372.  
  3373.                 if (!new->dirty[pipe])
  3374.                         continue;
  3375.  
  3376.                 I915_WRITE(PIPE_WM_LINETIME(pipe), new->wm_linetime[pipe]);
  3377.  
  3378.                 for (level = 0; level <= max_level; level++) {
  3379.                         for (i = 0; i < intel_num_planes(crtc); i++)
  3380.                                 I915_WRITE(PLANE_WM(pipe, i, level),
  3381.                                            new->plane[pipe][i][level]);
  3382.                         I915_WRITE(CUR_WM(pipe, level),
  3383.                                    new->plane[pipe][PLANE_CURSOR][level]);
  3384.                 }
  3385.                 for (i = 0; i < intel_num_planes(crtc); i++)
  3386.                         I915_WRITE(PLANE_WM_TRANS(pipe, i),
  3387.                                    new->plane_trans[pipe][i]);
  3388.                 I915_WRITE(CUR_WM_TRANS(pipe),
  3389.                            new->plane_trans[pipe][PLANE_CURSOR]);
  3390.  
  3391.                 for (i = 0; i < intel_num_planes(crtc); i++) {
  3392.                         skl_ddb_entry_write(dev_priv,
  3393.                                             PLANE_BUF_CFG(pipe, i),
  3394.                                             &new->ddb.plane[pipe][i]);
  3395.                         skl_ddb_entry_write(dev_priv,
  3396.                                             PLANE_NV12_BUF_CFG(pipe, i),
  3397.                                             &new->ddb.y_plane[pipe][i]);
  3398.                 }
  3399.  
  3400.                 skl_ddb_entry_write(dev_priv, CUR_BUF_CFG(pipe),
  3401.                                     &new->ddb.plane[pipe][PLANE_CURSOR]);
  3402.         }
  3403. }
  3404.  
  3405. /*
  3406.  * When setting up a new DDB allocation arrangement, we need to correctly
  3407.  * sequence the times at which the new allocations for the pipes are taken into
  3408.  * account or we'll have pipes fetching from space previously allocated to
  3409.  * another pipe.
  3410.  *
  3411.  * Roughly the sequence looks like:
  3412.  *  1. re-allocate the pipe(s) with the allocation being reduced and not
  3413.  *     overlapping with a previous light-up pipe (another way to put it is:
  3414.  *     pipes with their new allocation strickly included into their old ones).
  3415.  *  2. re-allocate the other pipes that get their allocation reduced
  3416.  *  3. allocate the pipes having their allocation increased
  3417.  *
  3418.  * Steps 1. and 2. are here to take care of the following case:
  3419.  * - Initially DDB looks like this:
  3420.  *     |   B    |   C    |
  3421.  * - enable pipe A.
  3422.  * - pipe B has a reduced DDB allocation that overlaps with the old pipe C
  3423.  *   allocation
  3424.  *     |  A  |  B  |  C  |
  3425.  *
  3426.  * We need to sequence the re-allocation: C, B, A (and not B, C, A).
  3427.  */
  3428.  
  3429. static void
  3430. skl_wm_flush_pipe(struct drm_i915_private *dev_priv, enum pipe pipe, int pass)
  3431. {
  3432.         int plane;
  3433.  
  3434.         DRM_DEBUG_KMS("flush pipe %c (pass %d)\n", pipe_name(pipe), pass);
  3435.  
  3436.         for_each_plane(dev_priv, pipe, plane) {
  3437.                 I915_WRITE(PLANE_SURF(pipe, plane),
  3438.                            I915_READ(PLANE_SURF(pipe, plane)));
  3439.         }
  3440.         I915_WRITE(CURBASE(pipe), I915_READ(CURBASE(pipe)));
  3441. }
  3442.  
  3443. static bool
  3444. skl_ddb_allocation_included(const struct skl_ddb_allocation *old,
  3445.                             const struct skl_ddb_allocation *new,
  3446.                             enum pipe pipe)
  3447. {
  3448.         uint16_t old_size, new_size;
  3449.  
  3450.         old_size = skl_ddb_entry_size(&old->pipe[pipe]);
  3451.         new_size = skl_ddb_entry_size(&new->pipe[pipe]);
  3452.  
  3453.         return old_size != new_size &&
  3454.                new->pipe[pipe].start >= old->pipe[pipe].start &&
  3455.                new->pipe[pipe].end <= old->pipe[pipe].end;
  3456. }
  3457.  
  3458. static void skl_flush_wm_values(struct drm_i915_private *dev_priv,
  3459.                                 struct skl_wm_values *new_values)
  3460. {
  3461.         struct drm_device *dev = dev_priv->dev;
  3462.         struct skl_ddb_allocation *cur_ddb, *new_ddb;
  3463.         bool reallocated[I915_MAX_PIPES] = {};
  3464.         struct intel_crtc *crtc;
  3465.         enum pipe pipe;
  3466.  
  3467.         new_ddb = &new_values->ddb;
  3468.         cur_ddb = &dev_priv->wm.skl_hw.ddb;
  3469.  
  3470.         /*
  3471.          * First pass: flush the pipes with the new allocation contained into
  3472.          * the old space.
  3473.          *
  3474.          * We'll wait for the vblank on those pipes to ensure we can safely
  3475.          * re-allocate the freed space without this pipe fetching from it.
  3476.          */
  3477.         for_each_intel_crtc(dev, crtc) {
  3478.                 if (!crtc->active)
  3479.                         continue;
  3480.  
  3481.                 pipe = crtc->pipe;
  3482.  
  3483.                 if (!skl_ddb_allocation_included(cur_ddb, new_ddb, pipe))
  3484.                         continue;
  3485.  
  3486.                 skl_wm_flush_pipe(dev_priv, pipe, 1);
  3487.                 intel_wait_for_vblank(dev, pipe);
  3488.  
  3489.                 reallocated[pipe] = true;
  3490.         }
  3491.  
  3492.  
  3493.         /*
  3494.          * Second pass: flush the pipes that are having their allocation
  3495.          * reduced, but overlapping with a previous allocation.
  3496.          *
  3497.          * Here as well we need to wait for the vblank to make sure the freed
  3498.          * space is not used anymore.
  3499.          */
  3500.         for_each_intel_crtc(dev, crtc) {
  3501.                 if (!crtc->active)
  3502.                         continue;
  3503.  
  3504.                 pipe = crtc->pipe;
  3505.  
  3506.                 if (reallocated[pipe])
  3507.                         continue;
  3508.  
  3509.                 if (skl_ddb_entry_size(&new_ddb->pipe[pipe]) <
  3510.                     skl_ddb_entry_size(&cur_ddb->pipe[pipe])) {
  3511.                         skl_wm_flush_pipe(dev_priv, pipe, 2);
  3512.                         intel_wait_for_vblank(dev, pipe);
  3513.                         reallocated[pipe] = true;
  3514.                 }
  3515.         }
  3516.  
  3517.         /*
  3518.          * Third pass: flush the pipes that got more space allocated.
  3519.          *
  3520.          * We don't need to actively wait for the update here, next vblank
  3521.          * will just get more DDB space with the correct WM values.
  3522.          */
  3523.         for_each_intel_crtc(dev, crtc) {
  3524.                 if (!crtc->active)
  3525.                         continue;
  3526.  
  3527.                 pipe = crtc->pipe;
  3528.  
  3529.                 /*
  3530.                  * At this point, only the pipes more space than before are
  3531.                  * left to re-allocate.
  3532.                  */
  3533.                 if (reallocated[pipe])
  3534.                         continue;
  3535.  
  3536.                 skl_wm_flush_pipe(dev_priv, pipe, 3);
  3537.         }
  3538. }
  3539.  
  3540. static bool skl_update_pipe_wm(struct drm_crtc *crtc,
  3541.                                struct skl_pipe_wm_parameters *params,
  3542.                                struct intel_wm_config *config,
  3543.                                struct skl_ddb_allocation *ddb, /* out */
  3544.                                struct skl_pipe_wm *pipe_wm /* out */)
  3545. {
  3546.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  3547.  
  3548.         skl_compute_wm_pipe_parameters(crtc, params);
  3549.         skl_allocate_pipe_ddb(crtc, config, params, ddb);
  3550.         skl_compute_pipe_wm(crtc, ddb, params, pipe_wm);
  3551.  
  3552.         if (!memcmp(&intel_crtc->wm.skl_active, pipe_wm, sizeof(*pipe_wm)))
  3553.                 return false;
  3554.  
  3555.         intel_crtc->wm.skl_active = *pipe_wm;
  3556.  
  3557.         return true;
  3558. }
  3559.  
  3560. static void skl_update_other_pipe_wm(struct drm_device *dev,
  3561.                                      struct drm_crtc *crtc,
  3562.                                      struct intel_wm_config *config,
  3563.                                      struct skl_wm_values *r)
  3564. {
  3565.         struct intel_crtc *intel_crtc;
  3566.         struct intel_crtc *this_crtc = to_intel_crtc(crtc);
  3567.  
  3568.         /*
  3569.          * If the WM update hasn't changed the allocation for this_crtc (the
  3570.          * crtc we are currently computing the new WM values for), other
  3571.          * enabled crtcs will keep the same allocation and we don't need to
  3572.          * recompute anything for them.
  3573.          */
  3574.         if (!skl_ddb_allocation_changed(&r->ddb, this_crtc))
  3575.                 return;
  3576.  
  3577.         /*
  3578.          * Otherwise, because of this_crtc being freshly enabled/disabled, the
  3579.          * other active pipes need new DDB allocation and WM values.
  3580.          */
  3581.         list_for_each_entry(intel_crtc, &dev->mode_config.crtc_list,
  3582.                                 base.head) {
  3583.                 struct skl_pipe_wm_parameters params = {};
  3584.                 struct skl_pipe_wm pipe_wm = {};
  3585.                 bool wm_changed;
  3586.  
  3587.                 if (this_crtc->pipe == intel_crtc->pipe)
  3588.                         continue;
  3589.  
  3590.                 if (!intel_crtc->active)
  3591.                         continue;
  3592.  
  3593.                 wm_changed = skl_update_pipe_wm(&intel_crtc->base,
  3594.                                                 &params, config,
  3595.                                                 &r->ddb, &pipe_wm);
  3596.  
  3597.                 /*
  3598.                  * If we end up re-computing the other pipe WM values, it's
  3599.                  * because it was really needed, so we expect the WM values to
  3600.                  * be different.
  3601.                  */
  3602.                 WARN_ON(!wm_changed);
  3603.  
  3604.                 skl_compute_wm_results(dev, &params, &pipe_wm, r, intel_crtc);
  3605.                 r->dirty[intel_crtc->pipe] = true;
  3606.         }
  3607. }
  3608.  
  3609. static void skl_clear_wm(struct skl_wm_values *watermarks, enum pipe pipe)
  3610. {
  3611.         watermarks->wm_linetime[pipe] = 0;
  3612.         memset(watermarks->plane[pipe], 0,
  3613.                sizeof(uint32_t) * 8 * I915_MAX_PLANES);
  3614.         memset(watermarks->plane_trans[pipe],
  3615.                0, sizeof(uint32_t) * I915_MAX_PLANES);
  3616.         watermarks->plane_trans[pipe][PLANE_CURSOR] = 0;
  3617.  
  3618.         /* Clear ddb entries for pipe */
  3619.         memset(&watermarks->ddb.pipe[pipe], 0, sizeof(struct skl_ddb_entry));
  3620.         memset(&watermarks->ddb.plane[pipe], 0,
  3621.                sizeof(struct skl_ddb_entry) * I915_MAX_PLANES);
  3622.         memset(&watermarks->ddb.y_plane[pipe], 0,
  3623.                sizeof(struct skl_ddb_entry) * I915_MAX_PLANES);
  3624.         memset(&watermarks->ddb.plane[pipe][PLANE_CURSOR], 0,
  3625.                sizeof(struct skl_ddb_entry));
  3626.  
  3627. }
  3628.  
  3629. static void skl_update_wm(struct drm_crtc *crtc)
  3630. {
  3631.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  3632.         struct drm_device *dev = crtc->dev;
  3633.         struct drm_i915_private *dev_priv = dev->dev_private;
  3634.         struct skl_pipe_wm_parameters params = {};
  3635.         struct skl_wm_values *results = &dev_priv->wm.skl_results;
  3636.         struct skl_pipe_wm pipe_wm = {};
  3637.         struct intel_wm_config config = {};
  3638.  
  3639.  
  3640.         /* Clear all dirty flags */
  3641.         memset(results->dirty, 0, sizeof(bool) * I915_MAX_PIPES);
  3642.  
  3643.         skl_clear_wm(results, intel_crtc->pipe);
  3644.  
  3645.         skl_compute_wm_global_parameters(dev, &config);
  3646.  
  3647.         if (!skl_update_pipe_wm(crtc, &params, &config,
  3648.                                 &results->ddb, &pipe_wm))
  3649.                 return;
  3650.  
  3651.         skl_compute_wm_results(dev, &params, &pipe_wm, results, intel_crtc);
  3652.         results->dirty[intel_crtc->pipe] = true;
  3653.  
  3654.         skl_update_other_pipe_wm(dev, crtc, &config, results);
  3655.         skl_write_wm_values(dev_priv, results);
  3656.         skl_flush_wm_values(dev_priv, results);
  3657.  
  3658.         /* store the new configuration */
  3659.         dev_priv->wm.skl_hw = *results;
  3660. }
  3661.  
  3662. static void
  3663. skl_update_sprite_wm(struct drm_plane *plane, struct drm_crtc *crtc,
  3664.                      uint32_t sprite_width, uint32_t sprite_height,
  3665.                      int pixel_size, bool enabled, bool scaled)
  3666. {
  3667.         struct intel_plane *intel_plane = to_intel_plane(plane);
  3668.         struct drm_framebuffer *fb = plane->state->fb;
  3669.  
  3670.         intel_plane->wm.enabled = enabled;
  3671.         intel_plane->wm.scaled = scaled;
  3672.         intel_plane->wm.horiz_pixels = sprite_width;
  3673.         intel_plane->wm.vert_pixels = sprite_height;
  3674.         intel_plane->wm.tiling = DRM_FORMAT_MOD_NONE;
  3675.  
  3676.         /* For planar: Bpp is for UV plane, y_Bpp is for Y plane */
  3677.         intel_plane->wm.bytes_per_pixel =
  3678.                 (fb && fb->pixel_format == DRM_FORMAT_NV12) ?
  3679.                 drm_format_plane_cpp(plane->state->fb->pixel_format, 1) : pixel_size;
  3680.         intel_plane->wm.y_bytes_per_pixel =
  3681.                 (fb && fb->pixel_format == DRM_FORMAT_NV12) ?
  3682.                 drm_format_plane_cpp(plane->state->fb->pixel_format, 0) : 0;
  3683.  
  3684.         /*
  3685.          * Framebuffer can be NULL on plane disable, but it does not
  3686.          * matter for watermarks if we assume no tiling in that case.
  3687.          */
  3688.         if (fb)
  3689.                 intel_plane->wm.tiling = fb->modifier[0];
  3690.         intel_plane->wm.rotation = plane->state->rotation;
  3691.  
  3692.         skl_update_wm(crtc);
  3693. }
  3694.  
  3695. static void ilk_update_wm(struct drm_crtc *crtc)
  3696. {
  3697.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  3698.         struct intel_crtc_state *cstate = to_intel_crtc_state(crtc->state);
  3699.         struct drm_device *dev = crtc->dev;
  3700.         struct drm_i915_private *dev_priv = dev->dev_private;
  3701.         struct ilk_wm_maximums max;
  3702.         struct ilk_wm_values results = {};
  3703.         enum intel_ddb_partitioning partitioning;
  3704.         struct intel_pipe_wm pipe_wm = {};
  3705.         struct intel_pipe_wm lp_wm_1_2 = {}, lp_wm_5_6 = {}, *best_lp_wm;
  3706.         struct intel_wm_config config = {};
  3707.  
  3708.         WARN_ON(cstate->base.active != intel_crtc->active);
  3709.  
  3710.         intel_compute_pipe_wm(cstate, &pipe_wm);
  3711.  
  3712.         if (!memcmp(&intel_crtc->wm.active, &pipe_wm, sizeof(pipe_wm)))
  3713.                 return;
  3714.  
  3715.         intel_crtc->wm.active = pipe_wm;
  3716.  
  3717.         ilk_compute_wm_config(dev, &config);
  3718.  
  3719.         ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_1_2, &max);
  3720.         ilk_wm_merge(dev, &config, &max, &lp_wm_1_2);
  3721.  
  3722.         /* 5/6 split only in single pipe config on IVB+ */
  3723.         if (INTEL_INFO(dev)->gen >= 7 &&
  3724.             config.num_pipes_active == 1 && config.sprites_enabled) {
  3725.                 ilk_compute_wm_maximums(dev, 1, &config, INTEL_DDB_PART_5_6, &max);
  3726.                 ilk_wm_merge(dev, &config, &max, &lp_wm_5_6);
  3727.  
  3728.                 best_lp_wm = ilk_find_best_result(dev, &lp_wm_1_2, &lp_wm_5_6);
  3729.         } else {
  3730.                 best_lp_wm = &lp_wm_1_2;
  3731.         }
  3732.  
  3733.         partitioning = (best_lp_wm == &lp_wm_1_2) ?
  3734.                        INTEL_DDB_PART_1_2 : INTEL_DDB_PART_5_6;
  3735.  
  3736.         ilk_compute_wm_results(dev, best_lp_wm, partitioning, &results);
  3737.  
  3738.         ilk_write_wm_values(dev_priv, &results);
  3739. }
  3740.  
  3741. static void
  3742. ilk_update_sprite_wm(struct drm_plane *plane,
  3743.                      struct drm_crtc *crtc,
  3744.                      uint32_t sprite_width, uint32_t sprite_height,
  3745.                      int pixel_size, bool enabled, bool scaled)
  3746. {
  3747.         struct drm_device *dev = plane->dev;
  3748.         struct intel_plane *intel_plane = to_intel_plane(plane);
  3749.  
  3750.         /*
  3751.          * IVB workaround: must disable low power watermarks for at least
  3752.          * one frame before enabling scaling.  LP watermarks can be re-enabled
  3753.          * when scaling is disabled.
  3754.          *
  3755.          * WaCxSRDisabledForSpriteScaling:ivb
  3756.          */
  3757.         if (IS_IVYBRIDGE(dev) && scaled && ilk_disable_lp_wm(dev))
  3758.                 intel_wait_for_vblank(dev, intel_plane->pipe);
  3759.  
  3760.         ilk_update_wm(crtc);
  3761. }
  3762.  
  3763. static void skl_pipe_wm_active_state(uint32_t val,
  3764.                                      struct skl_pipe_wm *active,
  3765.                                      bool is_transwm,
  3766.                                      bool is_cursor,
  3767.                                      int i,
  3768.                                      int level)
  3769. {
  3770.         bool is_enabled = (val & PLANE_WM_EN) != 0;
  3771.  
  3772.         if (!is_transwm) {
  3773.                 if (!is_cursor) {
  3774.                         active->wm[level].plane_en[i] = is_enabled;
  3775.                         active->wm[level].plane_res_b[i] =
  3776.                                         val & PLANE_WM_BLOCKS_MASK;
  3777.                         active->wm[level].plane_res_l[i] =
  3778.                                         (val >> PLANE_WM_LINES_SHIFT) &
  3779.                                                 PLANE_WM_LINES_MASK;
  3780.                 } else {
  3781.                         active->wm[level].plane_en[PLANE_CURSOR] = is_enabled;
  3782.                         active->wm[level].plane_res_b[PLANE_CURSOR] =
  3783.                                         val & PLANE_WM_BLOCKS_MASK;
  3784.                         active->wm[level].plane_res_l[PLANE_CURSOR] =
  3785.                                         (val >> PLANE_WM_LINES_SHIFT) &
  3786.                                                 PLANE_WM_LINES_MASK;
  3787.                 }
  3788.         } else {
  3789.                 if (!is_cursor) {
  3790.                         active->trans_wm.plane_en[i] = is_enabled;
  3791.                         active->trans_wm.plane_res_b[i] =
  3792.                                         val & PLANE_WM_BLOCKS_MASK;
  3793.                         active->trans_wm.plane_res_l[i] =
  3794.                                         (val >> PLANE_WM_LINES_SHIFT) &
  3795.                                                 PLANE_WM_LINES_MASK;
  3796.                 } else {
  3797.                         active->trans_wm.plane_en[PLANE_CURSOR] = is_enabled;
  3798.                         active->trans_wm.plane_res_b[PLANE_CURSOR] =
  3799.                                         val & PLANE_WM_BLOCKS_MASK;
  3800.                         active->trans_wm.plane_res_l[PLANE_CURSOR] =
  3801.                                         (val >> PLANE_WM_LINES_SHIFT) &
  3802.                                                 PLANE_WM_LINES_MASK;
  3803.                 }
  3804.         }
  3805. }
  3806.  
  3807. static void skl_pipe_wm_get_hw_state(struct drm_crtc *crtc)
  3808. {
  3809.         struct drm_device *dev = crtc->dev;
  3810.         struct drm_i915_private *dev_priv = dev->dev_private;
  3811.         struct skl_wm_values *hw = &dev_priv->wm.skl_hw;
  3812.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  3813.         struct skl_pipe_wm *active = &intel_crtc->wm.skl_active;
  3814.         enum pipe pipe = intel_crtc->pipe;
  3815.         int level, i, max_level;
  3816.         uint32_t temp;
  3817.  
  3818.         max_level = ilk_wm_max_level(dev);
  3819.  
  3820.         hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
  3821.  
  3822.         for (level = 0; level <= max_level; level++) {
  3823.                 for (i = 0; i < intel_num_planes(intel_crtc); i++)
  3824.                         hw->plane[pipe][i][level] =
  3825.                                         I915_READ(PLANE_WM(pipe, i, level));
  3826.                 hw->plane[pipe][PLANE_CURSOR][level] = I915_READ(CUR_WM(pipe, level));
  3827.         }
  3828.  
  3829.         for (i = 0; i < intel_num_planes(intel_crtc); i++)
  3830.                 hw->plane_trans[pipe][i] = I915_READ(PLANE_WM_TRANS(pipe, i));
  3831.         hw->plane_trans[pipe][PLANE_CURSOR] = I915_READ(CUR_WM_TRANS(pipe));
  3832.  
  3833.         if (!intel_crtc->active)
  3834.                 return;
  3835.  
  3836.         hw->dirty[pipe] = true;
  3837.  
  3838.         active->linetime = hw->wm_linetime[pipe];
  3839.  
  3840.         for (level = 0; level <= max_level; level++) {
  3841.                 for (i = 0; i < intel_num_planes(intel_crtc); i++) {
  3842.                         temp = hw->plane[pipe][i][level];
  3843.                         skl_pipe_wm_active_state(temp, active, false,
  3844.                                                 false, i, level);
  3845.                 }
  3846.                 temp = hw->plane[pipe][PLANE_CURSOR][level];
  3847.                 skl_pipe_wm_active_state(temp, active, false, true, i, level);
  3848.         }
  3849.  
  3850.         for (i = 0; i < intel_num_planes(intel_crtc); i++) {
  3851.                 temp = hw->plane_trans[pipe][i];
  3852.                 skl_pipe_wm_active_state(temp, active, true, false, i, 0);
  3853.         }
  3854.  
  3855.         temp = hw->plane_trans[pipe][PLANE_CURSOR];
  3856.         skl_pipe_wm_active_state(temp, active, true, true, i, 0);
  3857. }
  3858.  
  3859. void skl_wm_get_hw_state(struct drm_device *dev)
  3860. {
  3861.         struct drm_i915_private *dev_priv = dev->dev_private;
  3862.         struct skl_ddb_allocation *ddb = &dev_priv->wm.skl_hw.ddb;
  3863.         struct drm_crtc *crtc;
  3864.  
  3865.         skl_ddb_get_hw_state(dev_priv, ddb);
  3866.         list_for_each_entry(crtc, &dev->mode_config.crtc_list, head)
  3867.                 skl_pipe_wm_get_hw_state(crtc);
  3868. }
  3869.  
  3870. static void ilk_pipe_wm_get_hw_state(struct drm_crtc *crtc)
  3871. {
  3872.         struct drm_device *dev = crtc->dev;
  3873.         struct drm_i915_private *dev_priv = dev->dev_private;
  3874.         struct ilk_wm_values *hw = &dev_priv->wm.hw;
  3875.         struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
  3876.         struct intel_pipe_wm *active = &intel_crtc->wm.active;
  3877.         enum pipe pipe = intel_crtc->pipe;
  3878.         static const unsigned int wm0_pipe_reg[] = {
  3879.                 [PIPE_A] = WM0_PIPEA_ILK,
  3880.                 [PIPE_B] = WM0_PIPEB_ILK,
  3881.                 [PIPE_C] = WM0_PIPEC_IVB,
  3882.         };
  3883.  
  3884.         hw->wm_pipe[pipe] = I915_READ(wm0_pipe_reg[pipe]);
  3885.         if (IS_HASWELL(dev) || IS_BROADWELL(dev))
  3886.                 hw->wm_linetime[pipe] = I915_READ(PIPE_WM_LINETIME(pipe));
  3887.  
  3888.         active->pipe_enabled = intel_crtc->active;
  3889.  
  3890.         if (active->pipe_enabled) {
  3891.                 u32 tmp = hw->wm_pipe[pipe];
  3892.  
  3893.                 /*
  3894.                  * For active pipes LP0 watermark is marked as
  3895.                  * enabled, and LP1+ watermaks as disabled since
  3896.                  * we can't really reverse compute them in case
  3897.                  * multiple pipes are active.
  3898.                  */
  3899.                 active->wm[0].enable = true;
  3900.                 active->wm[0].pri_val = (tmp & WM0_PIPE_PLANE_MASK) >> WM0_PIPE_PLANE_SHIFT;
  3901.                 active->wm[0].spr_val = (tmp & WM0_PIPE_SPRITE_MASK) >> WM0_PIPE_SPRITE_SHIFT;
  3902.                 active->wm[0].cur_val = tmp & WM0_PIPE_CURSOR_MASK;
  3903.                 active->linetime = hw->wm_linetime[pipe];
  3904.         } else {
  3905.                 int level, max_level = ilk_wm_max_level(dev);
  3906.  
  3907.                 /*
  3908.                  * For inactive pipes, all watermark levels
  3909.                  * should be marked as enabled but zeroed,
  3910.                  * which is what we'd compute them to.
  3911.                  */
  3912.                 for (level = 0; level <= max_level; level++)
  3913.                         active->wm[level].enable = true;
  3914.         }
  3915. }
  3916.  
  3917. #define _FW_WM(value, plane) \
  3918.         (((value) & DSPFW_ ## plane ## _MASK) >> DSPFW_ ## plane ## _SHIFT)
  3919. #define _FW_WM_VLV(value, plane) \
  3920.         (((value) & DSPFW_ ## plane ## _MASK_VLV) >> DSPFW_ ## plane ## _SHIFT)
  3921.  
  3922. static void vlv_read_wm_values(struct drm_i915_private *dev_priv,
  3923.                                struct vlv_wm_values *wm)
  3924. {
  3925.         enum pipe pipe;
  3926.         uint32_t tmp;
  3927.  
  3928.         for_each_pipe(dev_priv, pipe) {
  3929.                 tmp = I915_READ(VLV_DDL(pipe));
  3930.  
  3931.                 wm->ddl[pipe].primary =
  3932.                         (tmp >> DDL_PLANE_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
  3933.                 wm->ddl[pipe].cursor =
  3934.                         (tmp >> DDL_CURSOR_SHIFT) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
  3935.                 wm->ddl[pipe].sprite[0] =
  3936.                         (tmp >> DDL_SPRITE_SHIFT(0)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
  3937.                 wm->ddl[pipe].sprite[1] =
  3938.                         (tmp >> DDL_SPRITE_SHIFT(1)) & (DDL_PRECISION_HIGH | DRAIN_LATENCY_MASK);
  3939.         }
  3940.  
  3941.         tmp = I915_READ(DSPFW1);
  3942.         wm->sr.plane = _FW_WM(tmp, SR);
  3943.         wm->pipe[PIPE_B].cursor = _FW_WM(tmp, CURSORB);
  3944.         wm->pipe[PIPE_B].primary = _FW_WM_VLV(tmp, PLANEB);
  3945.         wm->pipe[PIPE_A].primary = _FW_WM_VLV(tmp, PLANEA);
  3946.  
  3947.         tmp = I915_READ(DSPFW2);
  3948.         wm->pipe[PIPE_A].sprite[1] = _FW_WM_VLV(tmp, SPRITEB);
  3949.         wm->pipe[PIPE_A].cursor = _FW_WM(tmp, CURSORA);
  3950.         wm->pipe[PIPE_A].sprite[0] = _FW_WM_VLV(tmp, SPRITEA);
  3951.  
  3952.         tmp = I915_READ(DSPFW3);
  3953.         wm->sr.cursor = _FW_WM(tmp, CURSOR_SR);
  3954.  
  3955.         if (IS_CHERRYVIEW(dev_priv)) {
  3956.                 tmp = I915_READ(DSPFW7_CHV);
  3957.                 wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED);
  3958.                 wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC);
  3959.  
  3960.                 tmp = I915_READ(DSPFW8_CHV);
  3961.                 wm->pipe[PIPE_C].sprite[1] = _FW_WM_VLV(tmp, SPRITEF);
  3962.                 wm->pipe[PIPE_C].sprite[0] = _FW_WM_VLV(tmp, SPRITEE);
  3963.  
  3964.                 tmp = I915_READ(DSPFW9_CHV);
  3965.                 wm->pipe[PIPE_C].primary = _FW_WM_VLV(tmp, PLANEC);
  3966.                 wm->pipe[PIPE_C].cursor = _FW_WM(tmp, CURSORC);
  3967.  
  3968.                 tmp = I915_READ(DSPHOWM);
  3969.                 wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9;
  3970.                 wm->pipe[PIPE_C].sprite[1] |= _FW_WM(tmp, SPRITEF_HI) << 8;
  3971.                 wm->pipe[PIPE_C].sprite[0] |= _FW_WM(tmp, SPRITEE_HI) << 8;
  3972.                 wm->pipe[PIPE_C].primary |= _FW_WM(tmp, PLANEC_HI) << 8;
  3973.                 wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8;
  3974.                 wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8;
  3975.                 wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8;
  3976.                 wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8;
  3977.                 wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8;
  3978.                 wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8;
  3979.         } else {
  3980.                 tmp = I915_READ(DSPFW7);
  3981.                 wm->pipe[PIPE_B].sprite[1] = _FW_WM_VLV(tmp, SPRITED);
  3982.                 wm->pipe[PIPE_B].sprite[0] = _FW_WM_VLV(tmp, SPRITEC);
  3983.  
  3984.                 tmp = I915_READ(DSPHOWM);
  3985.                 wm->sr.plane |= _FW_WM(tmp, SR_HI) << 9;
  3986.                 wm->pipe[PIPE_B].sprite[1] |= _FW_WM(tmp, SPRITED_HI) << 8;
  3987.                 wm->pipe[PIPE_B].sprite[0] |= _FW_WM(tmp, SPRITEC_HI) << 8;
  3988.                 wm->pipe[PIPE_B].primary |= _FW_WM(tmp, PLANEB_HI) << 8;
  3989.                 wm->pipe[PIPE_A].sprite[1] |= _FW_WM(tmp, SPRITEB_HI) << 8;
  3990.                 wm->pipe[PIPE_A].sprite[0] |= _FW_WM(tmp, SPRITEA_HI) << 8;
  3991.                 wm->pipe[PIPE_A].primary |= _FW_WM(tmp, PLANEA_HI) << 8;
  3992.         }
  3993. }
  3994.  
  3995. #undef _FW_WM
  3996. #undef _FW_WM_VLV
  3997.  
  3998. void vlv_wm_get_hw_state(struct drm_device *dev)
  3999. {
  4000.         struct drm_i915_private *dev_priv = to_i915(dev);
  4001.         struct vlv_wm_values *wm = &dev_priv->wm.vlv;
  4002.         struct intel_plane *plane;
  4003.         enum pipe pipe;
  4004.         u32 val;
  4005.  
  4006.         vlv_read_wm_values(dev_priv, wm);
  4007.  
  4008.         for_each_intel_plane(dev, plane) {
  4009.                 switch (plane->base.type) {
  4010.                         int sprite;
  4011.                 case DRM_PLANE_TYPE_CURSOR:
  4012.                         plane->wm.fifo_size = 63;
  4013.                         break;
  4014.                 case DRM_PLANE_TYPE_PRIMARY:
  4015.                         plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, 0);
  4016.                         break;
  4017.                 case DRM_PLANE_TYPE_OVERLAY:
  4018.                         sprite = plane->plane;
  4019.                         plane->wm.fifo_size = vlv_get_fifo_size(dev, plane->pipe, sprite + 1);
  4020.                         break;
  4021.                 }
  4022.         }
  4023.  
  4024.         wm->cxsr = I915_READ(FW_BLC_SELF_VLV) & FW_CSPWRDWNEN;
  4025.         wm->level = VLV_WM_LEVEL_PM2;
  4026.  
  4027.         if (IS_CHERRYVIEW(dev_priv)) {
  4028.                 mutex_lock(&dev_priv->rps.hw_lock);
  4029.  
  4030.                 val = vlv_punit_read(dev_priv, PUNIT_REG_DSPFREQ);
  4031.                 if (val & DSP_MAXFIFO_PM5_ENABLE)
  4032.                         wm->level = VLV_WM_LEVEL_PM5;
  4033.  
  4034.                 /*
  4035.                  * If DDR DVFS is disabled in the BIOS, Punit
  4036.                  * will never ack the request. So if that happens
  4037.                  * assume we don't have to enable/disable DDR DVFS
  4038.                  * dynamically. To test that just set the REQ_ACK
  4039.                  * bit to poke the Punit, but don't change the
  4040.                  * HIGH/LOW bits so that we don't actually change
  4041.                  * the current state.
  4042.                  */
  4043.                 val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
  4044.                 val |= FORCE_DDR_FREQ_REQ_ACK;
  4045.                 vlv_punit_write(dev_priv, PUNIT_REG_DDR_SETUP2, val);
  4046.  
  4047.                 if (wait_for((vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2) &
  4048.                               FORCE_DDR_FREQ_REQ_ACK) == 0, 3)) {
  4049.                         DRM_DEBUG_KMS("Punit not acking DDR DVFS request, "
  4050.                                       "assuming DDR DVFS is disabled\n");
  4051.                         dev_priv->wm.max_level = VLV_WM_LEVEL_PM5;
  4052.                 } else {
  4053.                         val = vlv_punit_read(dev_priv, PUNIT_REG_DDR_SETUP2);
  4054.                         if ((val & FORCE_DDR_HIGH_FREQ) == 0)
  4055.                                 wm->level = VLV_WM_LEVEL_DDR_DVFS;
  4056.                 }
  4057.  
  4058.                 mutex_unlock(&dev_priv->rps.hw_lock);
  4059.         }
  4060.  
  4061.         for_each_pipe(dev_priv, pipe)
  4062.                 DRM_DEBUG_KMS("Initial watermarks: pipe %c, plane=%d, cursor=%d, sprite0=%d, sprite1=%d\n",
  4063.                               pipe_name(pipe), wm->pipe[pipe].primary, wm->pipe[pipe].cursor,
  4064.                               wm->pipe[pipe].sprite[0], wm->pipe[pipe].sprite[1]);
  4065.  
  4066.         DRM_DEBUG_KMS("Initial watermarks: SR plane=%d, SR cursor=%d level=%d cxsr=%d\n",
  4067.                       wm->sr.plane, wm->sr.cursor, wm->level, wm->cxsr);
  4068. }
  4069.  
  4070. void ilk_wm_get_hw_state(struct drm_device *dev)
  4071. {
  4072.         struct drm_i915_private *dev_priv = dev->dev_private;
  4073.         struct ilk_wm_values *hw = &dev_priv->wm.hw;
  4074.         struct drm_crtc *crtc;
  4075.  
  4076.         for_each_crtc(dev, crtc)
  4077.                 ilk_pipe_wm_get_hw_state(crtc);
  4078.  
  4079.         hw->wm_lp[0] = I915_READ(WM1_LP_ILK);
  4080.         hw->wm_lp[1] = I915_READ(WM2_LP_ILK);
  4081.         hw->wm_lp[2] = I915_READ(WM3_LP_ILK);
  4082.  
  4083.         hw->wm_lp_spr[0] = I915_READ(WM1S_LP_ILK);
  4084.         if (INTEL_INFO(dev)->gen >= 7) {
  4085.                 hw->wm_lp_spr[1] = I915_READ(WM2S_LP_IVB);
  4086.                 hw->wm_lp_spr[2] = I915_READ(WM3S_LP_IVB);
  4087.         }
  4088.  
  4089.         if (IS_HASWELL(dev) || IS_BROADWELL(dev))
  4090.                 hw->partitioning = (I915_READ(WM_MISC) & WM_MISC_DATA_PARTITION_5_6) ?
  4091.                         INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
  4092.         else if (IS_IVYBRIDGE(dev))
  4093.                 hw->partitioning = (I915_READ(DISP_ARB_CTL2) & DISP_DATA_PARTITION_5_6) ?
  4094.                         INTEL_DDB_PART_5_6 : INTEL_DDB_PART_1_2;
  4095.  
  4096.         hw->enable_fbc_wm =
  4097.                 !(I915_READ(DISP_ARB_CTL) & DISP_FBC_WM_DIS);
  4098. }
  4099.  
  4100. /**
  4101.  * intel_update_watermarks - update FIFO watermark values based on current modes
  4102.  *
  4103.  * Calculate watermark values for the various WM regs based on current mode
  4104.  * and plane configuration.
  4105.  *
  4106.  * There are several cases to deal with here:
  4107.  *   - normal (i.e. non-self-refresh)
  4108.  *   - self-refresh (SR) mode
  4109.  *   - lines are large relative to FIFO size (buffer can hold up to 2)
  4110.  *   - lines are small relative to FIFO size (buffer can hold more than 2
  4111.  *     lines), so need to account for TLB latency
  4112.  *
  4113.  *   The normal calculation is:
  4114.  *     watermark = dotclock * bytes per pixel * latency
  4115.  *   where latency is platform & configuration dependent (we assume pessimal
  4116.  *   values here).
  4117.  *
  4118.  *   The SR calculation is:
  4119.  *     watermark = (trunc(latency/line time)+1) * surface width *
  4120.  *       bytes per pixel
  4121.  *   where
  4122.  *     line time = htotal / dotclock
  4123.  *     surface width = hdisplay for normal plane and 64 for cursor
  4124.  *   and latency is assumed to be high, as above.
  4125.  *
  4126.  * The final value programmed to the register should always be rounded up,
  4127.  * and include an extra 2 entries to account for clock crossings.
  4128.  *
  4129.  * We don't use the sprite, so we can ignore that.  And on Crestline we have
  4130.  * to set the non-SR watermarks to 8.
  4131.  */
  4132. void intel_update_watermarks(struct drm_crtc *crtc)
  4133. {
  4134.         struct drm_i915_private *dev_priv = crtc->dev->dev_private;
  4135.  
  4136.         if (dev_priv->display.update_wm)
  4137.                 dev_priv->display.update_wm(crtc);
  4138. }
  4139.  
  4140. void intel_update_sprite_watermarks(struct drm_plane *plane,
  4141.                                     struct drm_crtc *crtc,
  4142.                                     uint32_t sprite_width,
  4143.                                     uint32_t sprite_height,
  4144.                                     int pixel_size,
  4145.                                     bool enabled, bool scaled)
  4146. {
  4147.         struct drm_i915_private *dev_priv = plane->dev->dev_private;
  4148.  
  4149.         if (dev_priv->display.update_sprite_wm)
  4150.                 dev_priv->display.update_sprite_wm(plane, crtc,
  4151.                                                    sprite_width, sprite_height,
  4152.                                                    pixel_size, enabled, scaled);
  4153. }
  4154.  
  4155. /**
  4156.  * Lock protecting IPS related data structures
  4157.  */
  4158. DEFINE_SPINLOCK(mchdev_lock);
  4159.  
  4160. /* Global for IPS driver to get at the current i915 device. Protected by
  4161.  * mchdev_lock. */
  4162. static struct drm_i915_private *i915_mch_dev;
  4163.  
  4164. bool ironlake_set_drps(struct drm_device *dev, u8 val)
  4165. {
  4166.         struct drm_i915_private *dev_priv = dev->dev_private;
  4167.         u16 rgvswctl;
  4168.  
  4169.         assert_spin_locked(&mchdev_lock);
  4170.  
  4171.         rgvswctl = I915_READ16(MEMSWCTL);
  4172.         if (rgvswctl & MEMCTL_CMD_STS) {
  4173.                 DRM_DEBUG("gpu busy, RCS change rejected\n");
  4174.                 return false; /* still busy with another command */
  4175.         }
  4176.  
  4177.         rgvswctl = (MEMCTL_CMD_CHFREQ << MEMCTL_CMD_SHIFT) |
  4178.                 (val << MEMCTL_FREQ_SHIFT) | MEMCTL_SFCAVM;
  4179.         I915_WRITE16(MEMSWCTL, rgvswctl);
  4180.         POSTING_READ16(MEMSWCTL);
  4181.  
  4182.         rgvswctl |= MEMCTL_CMD_STS;
  4183.         I915_WRITE16(MEMSWCTL, rgvswctl);
  4184.  
  4185.         return true;
  4186. }
  4187.  
  4188. static void ironlake_enable_drps(struct drm_device *dev)
  4189. {
  4190.         struct drm_i915_private *dev_priv = dev->dev_private;
  4191.         u32 rgvmodectl = I915_READ(MEMMODECTL);
  4192.         u8 fmax, fmin, fstart, vstart;
  4193.  
  4194.         spin_lock_irq(&mchdev_lock);
  4195.  
  4196.         /* Enable temp reporting */
  4197.         I915_WRITE16(PMMISC, I915_READ(PMMISC) | MCPPCE_EN);
  4198.         I915_WRITE16(TSC1, I915_READ(TSC1) | TSE);
  4199.  
  4200.         /* 100ms RC evaluation intervals */
  4201.         I915_WRITE(RCUPEI, 100000);
  4202.         I915_WRITE(RCDNEI, 100000);
  4203.  
  4204.         /* Set max/min thresholds to 90ms and 80ms respectively */
  4205.         I915_WRITE(RCBMAXAVG, 90000);
  4206.         I915_WRITE(RCBMINAVG, 80000);
  4207.  
  4208.         I915_WRITE(MEMIHYST, 1);
  4209.  
  4210.         /* Set up min, max, and cur for interrupt handling */
  4211.         fmax = (rgvmodectl & MEMMODE_FMAX_MASK) >> MEMMODE_FMAX_SHIFT;
  4212.         fmin = (rgvmodectl & MEMMODE_FMIN_MASK);
  4213.         fstart = (rgvmodectl & MEMMODE_FSTART_MASK) >>
  4214.                 MEMMODE_FSTART_SHIFT;
  4215.  
  4216.         vstart = (I915_READ(PXVFREQ(fstart)) & PXVFREQ_PX_MASK) >>
  4217.                 PXVFREQ_PX_SHIFT;
  4218.  
  4219.         dev_priv->ips.fmax = fmax; /* IPS callback will increase this */
  4220.         dev_priv->ips.fstart = fstart;
  4221.  
  4222.         dev_priv->ips.max_delay = fstart;
  4223.         dev_priv->ips.min_delay = fmin;
  4224.         dev_priv->ips.cur_delay = fstart;
  4225.  
  4226.         DRM_DEBUG_DRIVER("fmax: %d, fmin: %d, fstart: %d\n",
  4227.                          fmax, fmin, fstart);
  4228.  
  4229.         I915_WRITE(MEMINTREN, MEMINT_CX_SUPR_EN | MEMINT_EVAL_CHG_EN);
  4230.  
  4231.         /*
  4232.          * Interrupts will be enabled in ironlake_irq_postinstall
  4233.          */
  4234.  
  4235.         I915_WRITE(VIDSTART, vstart);
  4236.         POSTING_READ(VIDSTART);
  4237.  
  4238.         rgvmodectl |= MEMMODE_SWMODE_EN;
  4239.         I915_WRITE(MEMMODECTL, rgvmodectl);
  4240.  
  4241.         if (wait_for_atomic((I915_READ(MEMSWCTL) & MEMCTL_CMD_STS) == 0, 10))
  4242.                 DRM_ERROR("stuck trying to change perf mode\n");
  4243.         mdelay(1);
  4244.  
  4245.         ironlake_set_drps(dev, fstart);
  4246.  
  4247.         dev_priv->ips.last_count1 = I915_READ(DMIEC) +
  4248.                 I915_READ(DDREC) + I915_READ(CSIEC);
  4249.         dev_priv->ips.last_time1 = jiffies_to_msecs(jiffies);
  4250.         dev_priv->ips.last_count2 = I915_READ(GFXEC);
  4251.         dev_priv->ips.last_time2 = ktime_get_raw_ns();
  4252.  
  4253.         spin_unlock_irq(&mchdev_lock);
  4254. }
  4255.  
  4256. static void ironlake_disable_drps(struct drm_device *dev)
  4257. {
  4258.         struct drm_i915_private *dev_priv = dev->dev_private;
  4259.         u16 rgvswctl;
  4260.  
  4261.         spin_lock_irq(&mchdev_lock);
  4262.  
  4263.         rgvswctl = I915_READ16(MEMSWCTL);
  4264.  
  4265.         /* Ack interrupts, disable EFC interrupt */
  4266.         I915_WRITE(MEMINTREN, I915_READ(MEMINTREN) & ~MEMINT_EVAL_CHG_EN);
  4267.         I915_WRITE(MEMINTRSTS, MEMINT_EVAL_CHG);
  4268.         I915_WRITE(DEIER, I915_READ(DEIER) & ~DE_PCU_EVENT);
  4269.         I915_WRITE(DEIIR, DE_PCU_EVENT);
  4270.         I915_WRITE(DEIMR, I915_READ(DEIMR) | DE_PCU_EVENT);
  4271.  
  4272.         /* Go back to the starting frequency */
  4273.         ironlake_set_drps(dev, dev_priv->ips.fstart);
  4274.         mdelay(1);
  4275.         rgvswctl |= MEMCTL_CMD_STS;
  4276.         I915_WRITE(MEMSWCTL, rgvswctl);
  4277.         mdelay(1);
  4278.  
  4279.         spin_unlock_irq(&mchdev_lock);
  4280. }
  4281.  
  4282. /* There's a funny hw issue where the hw returns all 0 when reading from
  4283.  * GEN6_RP_INTERRUPT_LIMITS. Hence we always need to compute the desired value
  4284.  * ourselves, instead of doing a rmw cycle (which might result in us clearing
  4285.  * all limits and the gpu stuck at whatever frequency it is at atm).
  4286.  */
  4287. static u32 intel_rps_limits(struct drm_i915_private *dev_priv, u8 val)
  4288. {
  4289.         u32 limits;
  4290.  
  4291.         /* Only set the down limit when we've reached the lowest level to avoid
  4292.          * getting more interrupts, otherwise leave this clear. This prevents a
  4293.          * race in the hw when coming out of rc6: There's a tiny window where
  4294.          * the hw runs at the minimal clock before selecting the desired
  4295.          * frequency, if the down threshold expires in that window we will not
  4296.          * receive a down interrupt. */
  4297.         if (IS_GEN9(dev_priv->dev)) {
  4298.                 limits = (dev_priv->rps.max_freq_softlimit) << 23;
  4299.                 if (val <= dev_priv->rps.min_freq_softlimit)
  4300.                         limits |= (dev_priv->rps.min_freq_softlimit) << 14;
  4301.         } else {
  4302.                 limits = dev_priv->rps.max_freq_softlimit << 24;
  4303.                 if (val <= dev_priv->rps.min_freq_softlimit)
  4304.                         limits |= dev_priv->rps.min_freq_softlimit << 16;
  4305.         }
  4306.  
  4307.         return limits;
  4308. }
  4309.  
  4310. static void gen6_set_rps_thresholds(struct drm_i915_private *dev_priv, u8 val)
  4311. {
  4312.         int new_power;
  4313.         u32 threshold_up = 0, threshold_down = 0; /* in % */
  4314.         u32 ei_up = 0, ei_down = 0;
  4315.  
  4316.         new_power = dev_priv->rps.power;
  4317.         switch (dev_priv->rps.power) {
  4318.         case LOW_POWER:
  4319.                 if (val > dev_priv->rps.efficient_freq + 1 && val > dev_priv->rps.cur_freq)
  4320.                         new_power = BETWEEN;
  4321.                 break;
  4322.  
  4323.         case BETWEEN:
  4324.                 if (val <= dev_priv->rps.efficient_freq && val < dev_priv->rps.cur_freq)
  4325.                         new_power = LOW_POWER;
  4326.                 else if (val >= dev_priv->rps.rp0_freq && val > dev_priv->rps.cur_freq)
  4327.                         new_power = HIGH_POWER;
  4328.                 break;
  4329.  
  4330.         case HIGH_POWER:
  4331.                 if (val < (dev_priv->rps.rp1_freq + dev_priv->rps.rp0_freq) >> 1 && val < dev_priv->rps.cur_freq)
  4332.                         new_power = BETWEEN;
  4333.                 break;
  4334.         }
  4335.         /* Max/min bins are special */
  4336.         if (val <= dev_priv->rps.min_freq_softlimit)
  4337.                 new_power = LOW_POWER;
  4338.         if (val >= dev_priv->rps.max_freq_softlimit)
  4339.                 new_power = HIGH_POWER;
  4340.         if (new_power == dev_priv->rps.power)
  4341.                 return;
  4342.  
  4343.         /* Note the units here are not exactly 1us, but 1280ns. */
  4344.         switch (new_power) {
  4345.         case LOW_POWER:
  4346.                 /* Upclock if more than 95% busy over 16ms */
  4347.                 ei_up = 16000;
  4348.                 threshold_up = 95;
  4349.  
  4350.                 /* Downclock if less than 85% busy over 32ms */
  4351.                 ei_down = 32000;
  4352.                 threshold_down = 85;
  4353.                 break;
  4354.  
  4355.         case BETWEEN:
  4356.                 /* Upclock if more than 90% busy over 13ms */
  4357.                 ei_up = 13000;
  4358.                 threshold_up = 90;
  4359.  
  4360.                 /* Downclock if less than 75% busy over 32ms */
  4361.                 ei_down = 32000;
  4362.                 threshold_down = 75;
  4363.                 break;
  4364.  
  4365.         case HIGH_POWER:
  4366.                 /* Upclock if more than 85% busy over 10ms */
  4367.                 ei_up = 10000;
  4368.                 threshold_up = 85;
  4369.  
  4370.                 /* Downclock if less than 60% busy over 32ms */
  4371.                 ei_down = 32000;
  4372.                 threshold_down = 60;
  4373.                 break;
  4374.         }
  4375.  
  4376.         I915_WRITE(GEN6_RP_UP_EI,
  4377.                 GT_INTERVAL_FROM_US(dev_priv, ei_up));
  4378.         I915_WRITE(GEN6_RP_UP_THRESHOLD,
  4379.                 GT_INTERVAL_FROM_US(dev_priv, (ei_up * threshold_up / 100)));
  4380.  
  4381.         I915_WRITE(GEN6_RP_DOWN_EI,
  4382.                 GT_INTERVAL_FROM_US(dev_priv, ei_down));
  4383.         I915_WRITE(GEN6_RP_DOWN_THRESHOLD,
  4384.                 GT_INTERVAL_FROM_US(dev_priv, (ei_down * threshold_down / 100)));
  4385.  
  4386.          I915_WRITE(GEN6_RP_CONTROL,
  4387.                     GEN6_RP_MEDIA_TURBO |
  4388.                     GEN6_RP_MEDIA_HW_NORMAL_MODE |
  4389.                     GEN6_RP_MEDIA_IS_GFX |
  4390.                     GEN6_RP_ENABLE |
  4391.                     GEN6_RP_UP_BUSY_AVG |
  4392.                     GEN6_RP_DOWN_IDLE_AVG);
  4393.  
  4394.         dev_priv->rps.power = new_power;
  4395.         dev_priv->rps.up_threshold = threshold_up;
  4396.         dev_priv->rps.down_threshold = threshold_down;
  4397.         dev_priv->rps.last_adj = 0;
  4398. }
  4399.  
  4400. static u32 gen6_rps_pm_mask(struct drm_i915_private *dev_priv, u8 val)
  4401. {
  4402.         u32 mask = 0;
  4403.  
  4404.         if (val > dev_priv->rps.min_freq_softlimit)
  4405.                 mask |= GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_DOWN_THRESHOLD | GEN6_PM_RP_DOWN_TIMEOUT;
  4406.         if (val < dev_priv->rps.max_freq_softlimit)
  4407.                 mask |= GEN6_PM_RP_UP_EI_EXPIRED | GEN6_PM_RP_UP_THRESHOLD;
  4408.  
  4409.         mask &= dev_priv->pm_rps_events;
  4410.  
  4411.         return gen6_sanitize_rps_pm_mask(dev_priv, ~mask);
  4412. }
  4413.  
  4414. /* gen6_set_rps is called to update the frequency request, but should also be
  4415.  * called when the range (min_delay and max_delay) is modified so that we can
  4416.  * update the GEN6_RP_INTERRUPT_LIMITS register accordingly. */
  4417. static void gen6_set_rps(struct drm_device *dev, u8 val)
  4418. {
  4419.         struct drm_i915_private *dev_priv = dev->dev_private;
  4420.  
  4421.         /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
  4422.         if (IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0))
  4423.                 return;
  4424.  
  4425.         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
  4426.         WARN_ON(val > dev_priv->rps.max_freq);
  4427.         WARN_ON(val < dev_priv->rps.min_freq);
  4428.  
  4429.         /* min/max delay may still have been modified so be sure to
  4430.          * write the limits value.
  4431.          */
  4432.         if (val != dev_priv->rps.cur_freq) {
  4433.                 gen6_set_rps_thresholds(dev_priv, val);
  4434.  
  4435.                 if (IS_GEN9(dev))
  4436.                         I915_WRITE(GEN6_RPNSWREQ,
  4437.                                    GEN9_FREQUENCY(val));
  4438.                 else if (IS_HASWELL(dev) || IS_BROADWELL(dev))
  4439.                         I915_WRITE(GEN6_RPNSWREQ,
  4440.                                    HSW_FREQUENCY(val));
  4441.                 else
  4442.                         I915_WRITE(GEN6_RPNSWREQ,
  4443.                                    GEN6_FREQUENCY(val) |
  4444.                                    GEN6_OFFSET(0) |
  4445.                                    GEN6_AGGRESSIVE_TURBO);
  4446.         }
  4447.  
  4448.         /* Make sure we continue to get interrupts
  4449.          * until we hit the minimum or maximum frequencies.
  4450.          */
  4451.         I915_WRITE(GEN6_RP_INTERRUPT_LIMITS, intel_rps_limits(dev_priv, val));
  4452.         I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
  4453.  
  4454.         POSTING_READ(GEN6_RPNSWREQ);
  4455.  
  4456.         dev_priv->rps.cur_freq = val;
  4457.         trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val));
  4458. }
  4459.  
  4460. static void valleyview_set_rps(struct drm_device *dev, u8 val)
  4461. {
  4462.         struct drm_i915_private *dev_priv = dev->dev_private;
  4463.  
  4464.         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
  4465.         WARN_ON(val > dev_priv->rps.max_freq);
  4466.         WARN_ON(val < dev_priv->rps.min_freq);
  4467.  
  4468.         if (WARN_ONCE(IS_CHERRYVIEW(dev) && (val & 1),
  4469.                       "Odd GPU freq value\n"))
  4470.                 val &= ~1;
  4471.  
  4472.         I915_WRITE(GEN6_PMINTRMSK, gen6_rps_pm_mask(dev_priv, val));
  4473.  
  4474.         if (val != dev_priv->rps.cur_freq) {
  4475.                 vlv_punit_write(dev_priv, PUNIT_REG_GPU_FREQ_REQ, val);
  4476.                 if (!IS_CHERRYVIEW(dev_priv))
  4477.                         gen6_set_rps_thresholds(dev_priv, val);
  4478.         }
  4479.  
  4480.         dev_priv->rps.cur_freq = val;
  4481.         trace_intel_gpu_freq_change(intel_gpu_freq(dev_priv, val));
  4482. }
  4483.  
  4484. /* vlv_set_rps_idle: Set the frequency to idle, if Gfx clocks are down
  4485.  *
  4486.  * * If Gfx is Idle, then
  4487.  * 1. Forcewake Media well.
  4488.  * 2. Request idle freq.
  4489.  * 3. Release Forcewake of Media well.
  4490. */
  4491. static void vlv_set_rps_idle(struct drm_i915_private *dev_priv)
  4492. {
  4493.         u32 val = dev_priv->rps.idle_freq;
  4494.  
  4495.         if (dev_priv->rps.cur_freq <= val)
  4496.                 return;
  4497.  
  4498.         /* Wake up the media well, as that takes a lot less
  4499.          * power than the Render well. */
  4500.         intel_uncore_forcewake_get(dev_priv, FORCEWAKE_MEDIA);
  4501.         valleyview_set_rps(dev_priv->dev, val);
  4502.         intel_uncore_forcewake_put(dev_priv, FORCEWAKE_MEDIA);
  4503. }
  4504.  
  4505. void gen6_rps_busy(struct drm_i915_private *dev_priv)
  4506. {
  4507.         mutex_lock(&dev_priv->rps.hw_lock);
  4508.         if (dev_priv->rps.enabled) {
  4509.                 if (dev_priv->pm_rps_events & (GEN6_PM_RP_DOWN_EI_EXPIRED | GEN6_PM_RP_UP_EI_EXPIRED))
  4510.                         gen6_rps_reset_ei(dev_priv);
  4511.                 I915_WRITE(GEN6_PMINTRMSK,
  4512.                            gen6_rps_pm_mask(dev_priv, dev_priv->rps.cur_freq));
  4513.         }
  4514.         mutex_unlock(&dev_priv->rps.hw_lock);
  4515. }
  4516.  
  4517. void gen6_rps_idle(struct drm_i915_private *dev_priv)
  4518. {
  4519.         struct drm_device *dev = dev_priv->dev;
  4520.  
  4521.         mutex_lock(&dev_priv->rps.hw_lock);
  4522.         if (dev_priv->rps.enabled) {
  4523.                 if (IS_VALLEYVIEW(dev))
  4524.                         vlv_set_rps_idle(dev_priv);
  4525.                 else
  4526.                         gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq);
  4527.                 dev_priv->rps.last_adj = 0;
  4528.                 I915_WRITE(GEN6_PMINTRMSK, 0xffffffff);
  4529.         }
  4530.         mutex_unlock(&dev_priv->rps.hw_lock);
  4531.  
  4532.         spin_lock(&dev_priv->rps.client_lock);
  4533.         while (!list_empty(&dev_priv->rps.clients))
  4534.                 list_del_init(dev_priv->rps.clients.next);
  4535.         spin_unlock(&dev_priv->rps.client_lock);
  4536. }
  4537.  
  4538. void gen6_rps_boost(struct drm_i915_private *dev_priv,
  4539.                     struct intel_rps_client *rps,
  4540.                     unsigned long submitted)
  4541. {
  4542.         /* This is intentionally racy! We peek at the state here, then
  4543.          * validate inside the RPS worker.
  4544.          */
  4545.         if (!(dev_priv->mm.busy &&
  4546.               dev_priv->rps.enabled &&
  4547.               dev_priv->rps.cur_freq < dev_priv->rps.max_freq_softlimit))
  4548.                 return;
  4549.  
  4550.         /* Force a RPS boost (and don't count it against the client) if
  4551.          * the GPU is severely congested.
  4552.          */
  4553.         if (rps && time_after(jiffies, submitted + DRM_I915_THROTTLE_JIFFIES))
  4554.                 rps = NULL;
  4555.  
  4556.         spin_lock(&dev_priv->rps.client_lock);
  4557.         if (rps == NULL || list_empty(&rps->link)) {
  4558.                 spin_lock_irq(&dev_priv->irq_lock);
  4559.                 if (dev_priv->rps.interrupts_enabled) {
  4560.                         dev_priv->rps.client_boost = true;
  4561.                         queue_work(dev_priv->wq, &dev_priv->rps.work);
  4562.                 }
  4563.                 spin_unlock_irq(&dev_priv->irq_lock);
  4564.  
  4565.                 if (rps != NULL) {
  4566.                         list_add(&rps->link, &dev_priv->rps.clients);
  4567.                         rps->boosts++;
  4568.                 } else
  4569.                         dev_priv->rps.boosts++;
  4570.         }
  4571.         spin_unlock(&dev_priv->rps.client_lock);
  4572. }
  4573.  
  4574. void intel_set_rps(struct drm_device *dev, u8 val)
  4575. {
  4576.         if (IS_VALLEYVIEW(dev))
  4577.                 valleyview_set_rps(dev, val);
  4578.         else
  4579.                 gen6_set_rps(dev, val);
  4580. }
  4581.  
  4582. static void gen9_disable_rps(struct drm_device *dev)
  4583. {
  4584.         struct drm_i915_private *dev_priv = dev->dev_private;
  4585.  
  4586.         I915_WRITE(GEN6_RC_CONTROL, 0);
  4587.         I915_WRITE(GEN9_PG_ENABLE, 0);
  4588. }
  4589.  
  4590. static void gen6_disable_rps(struct drm_device *dev)
  4591. {
  4592.         struct drm_i915_private *dev_priv = dev->dev_private;
  4593.  
  4594.         I915_WRITE(GEN6_RC_CONTROL, 0);
  4595.         I915_WRITE(GEN6_RPNSWREQ, 1 << 31);
  4596. }
  4597.  
  4598. static void cherryview_disable_rps(struct drm_device *dev)
  4599. {
  4600.         struct drm_i915_private *dev_priv = dev->dev_private;
  4601.  
  4602.         I915_WRITE(GEN6_RC_CONTROL, 0);
  4603. }
  4604.  
  4605. static void valleyview_disable_rps(struct drm_device *dev)
  4606. {
  4607.         struct drm_i915_private *dev_priv = dev->dev_private;
  4608.  
  4609.         /* we're doing forcewake before Disabling RC6,
  4610.          * This what the BIOS expects when going into suspend */
  4611.         intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
  4612.  
  4613.         I915_WRITE(GEN6_RC_CONTROL, 0);
  4614.  
  4615.         intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
  4616. }
  4617.  
  4618. static void intel_print_rc6_info(struct drm_device *dev, u32 mode)
  4619. {
  4620.         if (IS_VALLEYVIEW(dev)) {
  4621.                 if (mode & (GEN7_RC_CTL_TO_MODE | GEN6_RC_CTL_EI_MODE(1)))
  4622.                         mode = GEN6_RC_CTL_RC6_ENABLE;
  4623.                 else
  4624.                         mode = 0;
  4625.         }
  4626.         if (HAS_RC6p(dev))
  4627.                 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s RC6p %s RC6pp %s\n",
  4628.                               (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off",
  4629.                               (mode & GEN6_RC_CTL_RC6p_ENABLE) ? "on" : "off",
  4630.                               (mode & GEN6_RC_CTL_RC6pp_ENABLE) ? "on" : "off");
  4631.  
  4632.         else
  4633.                 DRM_DEBUG_KMS("Enabling RC6 states: RC6 %s\n",
  4634.                               (mode & GEN6_RC_CTL_RC6_ENABLE) ? "on" : "off");
  4635. }
  4636.  
  4637. static int sanitize_rc6_option(const struct drm_device *dev, int enable_rc6)
  4638. {
  4639.         /* No RC6 before Ironlake and code is gone for ilk. */
  4640.         if (INTEL_INFO(dev)->gen < 6)
  4641.                 return 0;
  4642.  
  4643.         /* Respect the kernel parameter if it is set */
  4644.         if (enable_rc6 >= 0) {
  4645.                 int mask;
  4646.  
  4647.                 if (HAS_RC6p(dev))
  4648.                         mask = INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE |
  4649.                                INTEL_RC6pp_ENABLE;
  4650.                 else
  4651.                         mask = INTEL_RC6_ENABLE;
  4652.  
  4653.                 if ((enable_rc6 & mask) != enable_rc6)
  4654.                         DRM_DEBUG_KMS("Adjusting RC6 mask to %d (requested %d, valid %d)\n",
  4655.                                       enable_rc6 & mask, enable_rc6, mask);
  4656.  
  4657.                 return enable_rc6 & mask;
  4658.         }
  4659.  
  4660.         if (IS_IVYBRIDGE(dev))
  4661.                 return (INTEL_RC6_ENABLE | INTEL_RC6p_ENABLE);
  4662.  
  4663.         return INTEL_RC6_ENABLE;
  4664. }
  4665.  
  4666. int intel_enable_rc6(const struct drm_device *dev)
  4667. {
  4668.         return i915.enable_rc6;
  4669. }
  4670.  
  4671. static void gen6_init_rps_frequencies(struct drm_device *dev)
  4672. {
  4673.         struct drm_i915_private *dev_priv = dev->dev_private;
  4674.         uint32_t rp_state_cap;
  4675.         u32 ddcc_status = 0;
  4676.         int ret;
  4677.  
  4678.         /* All of these values are in units of 50MHz */
  4679.         dev_priv->rps.cur_freq          = 0;
  4680.         /* static values from HW: RP0 > RP1 > RPn (min_freq) */
  4681.         if (IS_BROXTON(dev)) {
  4682.                 rp_state_cap = I915_READ(BXT_RP_STATE_CAP);
  4683.                 dev_priv->rps.rp0_freq = (rp_state_cap >> 16) & 0xff;
  4684.                 dev_priv->rps.rp1_freq = (rp_state_cap >>  8) & 0xff;
  4685.                 dev_priv->rps.min_freq = (rp_state_cap >>  0) & 0xff;
  4686.         } else {
  4687.                 rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
  4688.                 dev_priv->rps.rp0_freq = (rp_state_cap >>  0) & 0xff;
  4689.                 dev_priv->rps.rp1_freq = (rp_state_cap >>  8) & 0xff;
  4690.                 dev_priv->rps.min_freq = (rp_state_cap >> 16) & 0xff;
  4691.         }
  4692.  
  4693.         /* hw_max = RP0 until we check for overclocking */
  4694.         dev_priv->rps.max_freq          = dev_priv->rps.rp0_freq;
  4695.  
  4696.         dev_priv->rps.efficient_freq = dev_priv->rps.rp1_freq;
  4697.         if (IS_HASWELL(dev) || IS_BROADWELL(dev) || IS_SKYLAKE(dev)) {
  4698.                 ret = sandybridge_pcode_read(dev_priv,
  4699.                                         HSW_PCODE_DYNAMIC_DUTY_CYCLE_CONTROL,
  4700.                                         &ddcc_status);
  4701.                 if (0 == ret)
  4702.                         dev_priv->rps.efficient_freq =
  4703.                                 clamp_t(u8,
  4704.                                         ((ddcc_status >> 8) & 0xff),
  4705.                                         dev_priv->rps.min_freq,
  4706.                                         dev_priv->rps.max_freq);
  4707.         }
  4708.  
  4709.         if (IS_SKYLAKE(dev)) {
  4710.                 /* Store the frequency values in 16.66 MHZ units, which is
  4711.                    the natural hardware unit for SKL */
  4712.                 dev_priv->rps.rp0_freq *= GEN9_FREQ_SCALER;
  4713.                 dev_priv->rps.rp1_freq *= GEN9_FREQ_SCALER;
  4714.                 dev_priv->rps.min_freq *= GEN9_FREQ_SCALER;
  4715.                 dev_priv->rps.max_freq *= GEN9_FREQ_SCALER;
  4716.                 dev_priv->rps.efficient_freq *= GEN9_FREQ_SCALER;
  4717.         }
  4718.  
  4719.         dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
  4720.  
  4721.         /* Preserve min/max settings in case of re-init */
  4722.         if (dev_priv->rps.max_freq_softlimit == 0)
  4723.                 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
  4724.  
  4725.         if (dev_priv->rps.min_freq_softlimit == 0) {
  4726.                 if (IS_HASWELL(dev) || IS_BROADWELL(dev))
  4727.                         dev_priv->rps.min_freq_softlimit =
  4728.                                 max_t(int, dev_priv->rps.efficient_freq,
  4729.                                       intel_freq_opcode(dev_priv, 450));
  4730.                 else
  4731.                         dev_priv->rps.min_freq_softlimit =
  4732.                                 dev_priv->rps.min_freq;
  4733.         }
  4734. }
  4735.  
  4736. /* See the Gen9_GT_PM_Programming_Guide doc for the below */
  4737. static void gen9_enable_rps(struct drm_device *dev)
  4738. {
  4739.         struct drm_i915_private *dev_priv = dev->dev_private;
  4740.  
  4741.         intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
  4742.  
  4743.         gen6_init_rps_frequencies(dev);
  4744.  
  4745.         /* WaGsvDisableTurbo: Workaround to disable turbo on BXT A* */
  4746.         if (IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0)) {
  4747.                 intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
  4748.                 return;
  4749.         }
  4750.  
  4751.         /* Program defaults and thresholds for RPS*/
  4752.         I915_WRITE(GEN6_RC_VIDEO_FREQ,
  4753.                 GEN9_FREQUENCY(dev_priv->rps.rp1_freq));
  4754.  
  4755.         /* 1 second timeout*/
  4756.         I915_WRITE(GEN6_RP_DOWN_TIMEOUT,
  4757.                 GT_INTERVAL_FROM_US(dev_priv, 1000000));
  4758.  
  4759.         I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 0xa);
  4760.  
  4761.         /* Leaning on the below call to gen6_set_rps to program/setup the
  4762.          * Up/Down EI & threshold registers, as well as the RP_CONTROL,
  4763.          * RP_INTERRUPT_LIMITS & RPNSWREQ registers */
  4764.         dev_priv->rps.power = HIGH_POWER; /* force a reset */
  4765.         gen6_set_rps(dev_priv->dev, dev_priv->rps.min_freq_softlimit);
  4766.  
  4767.         intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
  4768. }
  4769.  
  4770. static void gen9_enable_rc6(struct drm_device *dev)
  4771. {
  4772.         struct drm_i915_private *dev_priv = dev->dev_private;
  4773.         struct intel_engine_cs *ring;
  4774.         uint32_t rc6_mask = 0;
  4775.         int unused;
  4776.  
  4777.         /* 1a: Software RC state - RC0 */
  4778.         I915_WRITE(GEN6_RC_STATE, 0);
  4779.  
  4780.         /* 1b: Get forcewake during program sequence. Although the driver
  4781.          * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
  4782.         intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
  4783.  
  4784.         /* 2a: Disable RC states. */
  4785.         I915_WRITE(GEN6_RC_CONTROL, 0);
  4786.  
  4787.         /* 2b: Program RC6 thresholds.*/
  4788.  
  4789.         /* WaRsDoubleRc6WrlWithCoarsePowerGating: Doubling WRL only when CPG is enabled */
  4790.         if (IS_SKYLAKE(dev))
  4791.                 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 108 << 16);
  4792.         else
  4793.                 I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 54 << 16);
  4794.         I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
  4795.         I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
  4796.         for_each_ring(ring, dev_priv, unused)
  4797.                 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
  4798.  
  4799.         if (HAS_GUC_UCODE(dev))
  4800.                 I915_WRITE(GUC_MAX_IDLE_COUNT, 0xA);
  4801.  
  4802.         I915_WRITE(GEN6_RC_SLEEP, 0);
  4803.  
  4804.         /* 2c: Program Coarse Power Gating Policies. */
  4805.         I915_WRITE(GEN9_MEDIA_PG_IDLE_HYSTERESIS, 25);
  4806.         I915_WRITE(GEN9_RENDER_PG_IDLE_HYSTERESIS, 25);
  4807.  
  4808.         /* 3a: Enable RC6 */
  4809.         if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
  4810.                 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
  4811.         DRM_INFO("RC6 %s\n", (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ?
  4812.                         "on" : "off");
  4813.         /* WaRsUseTimeoutMode */
  4814.         if ((IS_SKYLAKE(dev) && INTEL_REVID(dev) <= SKL_REVID_D0) ||
  4815.             (IS_BROXTON(dev) && INTEL_REVID(dev) <= BXT_REVID_A0)) {
  4816.                 I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us */
  4817.                 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
  4818.                            GEN7_RC_CTL_TO_MODE |
  4819.                            rc6_mask);
  4820.         } else {
  4821.                 I915_WRITE(GEN6_RC6_THRESHOLD, 37500); /* 37.5/125ms per EI */
  4822.                 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
  4823.                            GEN6_RC_CTL_EI_MODE(1) |
  4824.                            rc6_mask);
  4825.         }
  4826.  
  4827.         /*
  4828.          * 3b: Enable Coarse Power Gating only when RC6 is enabled.
  4829.          * WaRsDisableCoarsePowerGating:skl,bxt - Render/Media PG need to be disabled with RC6.
  4830.          */
  4831.         if ((IS_BROXTON(dev) && (INTEL_REVID(dev) < BXT_REVID_B0)) ||
  4832.             ((IS_SKL_GT3(dev) || IS_SKL_GT4(dev)) && (INTEL_REVID(dev) <= SKL_REVID_F0)))
  4833.                 I915_WRITE(GEN9_PG_ENABLE, 0);
  4834.         else
  4835.                 I915_WRITE(GEN9_PG_ENABLE, (rc6_mask & GEN6_RC_CTL_RC6_ENABLE) ?
  4836.                                 (GEN9_RENDER_PG_ENABLE | GEN9_MEDIA_PG_ENABLE) : 0);
  4837.  
  4838.         intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
  4839.  
  4840. }
  4841.  
  4842. static void gen8_enable_rps(struct drm_device *dev)
  4843. {
  4844.         struct drm_i915_private *dev_priv = dev->dev_private;
  4845.         struct intel_engine_cs *ring;
  4846.         uint32_t rc6_mask = 0;
  4847.         int unused;
  4848.  
  4849.         /* 1a: Software RC state - RC0 */
  4850.         I915_WRITE(GEN6_RC_STATE, 0);
  4851.  
  4852.         /* 1c & 1d: Get forcewake during program sequence. Although the driver
  4853.          * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
  4854.         intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
  4855.  
  4856.         /* 2a: Disable RC states. */
  4857.         I915_WRITE(GEN6_RC_CONTROL, 0);
  4858.  
  4859.         /* Initialize rps frequencies */
  4860.         gen6_init_rps_frequencies(dev);
  4861.  
  4862.         /* 2b: Program RC6 thresholds.*/
  4863.         I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
  4864.         I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
  4865.         I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
  4866.         for_each_ring(ring, dev_priv, unused)
  4867.                 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
  4868.         I915_WRITE(GEN6_RC_SLEEP, 0);
  4869.         if (IS_BROADWELL(dev))
  4870.                 I915_WRITE(GEN6_RC6_THRESHOLD, 625); /* 800us/1.28 for TO */
  4871.         else
  4872.                 I915_WRITE(GEN6_RC6_THRESHOLD, 50000); /* 50/125ms per EI */
  4873.  
  4874.         /* 3: Enable RC6 */
  4875.         if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
  4876.                 rc6_mask = GEN6_RC_CTL_RC6_ENABLE;
  4877.         intel_print_rc6_info(dev, rc6_mask);
  4878.         if (IS_BROADWELL(dev))
  4879.                 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
  4880.                                 GEN7_RC_CTL_TO_MODE |
  4881.                                 rc6_mask);
  4882.         else
  4883.                 I915_WRITE(GEN6_RC_CONTROL, GEN6_RC_CTL_HW_ENABLE |
  4884.                                 GEN6_RC_CTL_EI_MODE(1) |
  4885.                                 rc6_mask);
  4886.  
  4887.         /* 4 Program defaults and thresholds for RPS*/
  4888.         I915_WRITE(GEN6_RPNSWREQ,
  4889.                    HSW_FREQUENCY(dev_priv->rps.rp1_freq));
  4890.         I915_WRITE(GEN6_RC_VIDEO_FREQ,
  4891.                    HSW_FREQUENCY(dev_priv->rps.rp1_freq));
  4892.         /* NB: Docs say 1s, and 1000000 - which aren't equivalent */
  4893.         I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 100000000 / 128); /* 1 second timeout */
  4894.  
  4895.         /* Docs recommend 900MHz, and 300 MHz respectively */
  4896.         I915_WRITE(GEN6_RP_INTERRUPT_LIMITS,
  4897.                    dev_priv->rps.max_freq_softlimit << 24 |
  4898.                    dev_priv->rps.min_freq_softlimit << 16);
  4899.  
  4900.         I915_WRITE(GEN6_RP_UP_THRESHOLD, 7600000 / 128); /* 76ms busyness per EI, 90% */
  4901.         I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 31300000 / 128); /* 313ms busyness per EI, 70%*/
  4902.         I915_WRITE(GEN6_RP_UP_EI, 66000); /* 84.48ms, XXX: random? */
  4903.         I915_WRITE(GEN6_RP_DOWN_EI, 350000); /* 448ms, XXX: random? */
  4904.  
  4905.         I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
  4906.  
  4907.         /* 5: Enable RPS */
  4908.         I915_WRITE(GEN6_RP_CONTROL,
  4909.                    GEN6_RP_MEDIA_TURBO |
  4910.                    GEN6_RP_MEDIA_HW_NORMAL_MODE |
  4911.                    GEN6_RP_MEDIA_IS_GFX |
  4912.                    GEN6_RP_ENABLE |
  4913.                    GEN6_RP_UP_BUSY_AVG |
  4914.                    GEN6_RP_DOWN_IDLE_AVG);
  4915.  
  4916.         /* 6: Ring frequency + overclocking (our driver does this later */
  4917.  
  4918.         dev_priv->rps.power = HIGH_POWER; /* force a reset */
  4919.         gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq);
  4920.  
  4921.         intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
  4922. }
  4923.  
  4924. static void gen6_enable_rps(struct drm_device *dev)
  4925. {
  4926.         struct drm_i915_private *dev_priv = dev->dev_private;
  4927.         struct intel_engine_cs *ring;
  4928.         u32 rc6vids, pcu_mbox = 0, rc6_mask = 0;
  4929.         u32 gtfifodbg;
  4930.         int rc6_mode;
  4931.         int i, ret;
  4932.  
  4933.         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
  4934.  
  4935.         /* Here begins a magic sequence of register writes to enable
  4936.          * auto-downclocking.
  4937.          *
  4938.          * Perhaps there might be some value in exposing these to
  4939.          * userspace...
  4940.          */
  4941.         I915_WRITE(GEN6_RC_STATE, 0);
  4942.  
  4943.         /* Clear the DBG now so we don't confuse earlier errors */
  4944.         if ((gtfifodbg = I915_READ(GTFIFODBG))) {
  4945.                 DRM_ERROR("GT fifo had a previous error %x\n", gtfifodbg);
  4946.                 I915_WRITE(GTFIFODBG, gtfifodbg);
  4947.         }
  4948.  
  4949.         intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
  4950.  
  4951.         /* Initialize rps frequencies */
  4952.         gen6_init_rps_frequencies(dev);
  4953.  
  4954.         /* disable the counters and set deterministic thresholds */
  4955.         I915_WRITE(GEN6_RC_CONTROL, 0);
  4956.  
  4957.         I915_WRITE(GEN6_RC1_WAKE_RATE_LIMIT, 1000 << 16);
  4958.         I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16 | 30);
  4959.         I915_WRITE(GEN6_RC6pp_WAKE_RATE_LIMIT, 30);
  4960.         I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
  4961.         I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
  4962.  
  4963.         for_each_ring(ring, dev_priv, i)
  4964.                 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
  4965.  
  4966.         I915_WRITE(GEN6_RC_SLEEP, 0);
  4967.         I915_WRITE(GEN6_RC1e_THRESHOLD, 1000);
  4968.         if (IS_IVYBRIDGE(dev))
  4969.                 I915_WRITE(GEN6_RC6_THRESHOLD, 125000);
  4970.         else
  4971.                 I915_WRITE(GEN6_RC6_THRESHOLD, 50000);
  4972.         I915_WRITE(GEN6_RC6p_THRESHOLD, 150000);
  4973.         I915_WRITE(GEN6_RC6pp_THRESHOLD, 64000); /* unused */
  4974.  
  4975.         /* Check if we are enabling RC6 */
  4976.         rc6_mode = intel_enable_rc6(dev_priv->dev);
  4977.         if (rc6_mode & INTEL_RC6_ENABLE)
  4978.                 rc6_mask |= GEN6_RC_CTL_RC6_ENABLE;
  4979.  
  4980.         /* We don't use those on Haswell */
  4981.         if (!IS_HASWELL(dev)) {
  4982.                 if (rc6_mode & INTEL_RC6p_ENABLE)
  4983.                         rc6_mask |= GEN6_RC_CTL_RC6p_ENABLE;
  4984.  
  4985.                 if (rc6_mode & INTEL_RC6pp_ENABLE)
  4986.                         rc6_mask |= GEN6_RC_CTL_RC6pp_ENABLE;
  4987.         }
  4988.  
  4989.         intel_print_rc6_info(dev, rc6_mask);
  4990.  
  4991.         I915_WRITE(GEN6_RC_CONTROL,
  4992.                    rc6_mask |
  4993.                    GEN6_RC_CTL_EI_MODE(1) |
  4994.                    GEN6_RC_CTL_HW_ENABLE);
  4995.  
  4996.         /* Power down if completely idle for over 50ms */
  4997.         I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 50000);
  4998.         I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
  4999.  
  5000.         ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_MIN_FREQ_TABLE, 0);
  5001.         if (ret)
  5002.                 DRM_DEBUG_DRIVER("Failed to set the min frequency\n");
  5003.  
  5004.         ret = sandybridge_pcode_read(dev_priv, GEN6_READ_OC_PARAMS, &pcu_mbox);
  5005.         if (!ret && (pcu_mbox & (1<<31))) { /* OC supported */
  5006.                 DRM_DEBUG_DRIVER("Overclocking supported. Max: %dMHz, Overclock max: %dMHz\n",
  5007.                                  (dev_priv->rps.max_freq_softlimit & 0xff) * 50,
  5008.                                  (pcu_mbox & 0xff) * 50);
  5009.                 dev_priv->rps.max_freq = pcu_mbox & 0xff;
  5010.         }
  5011.  
  5012.         dev_priv->rps.power = HIGH_POWER; /* force a reset */
  5013.         gen6_set_rps(dev_priv->dev, dev_priv->rps.idle_freq);
  5014.  
  5015.         rc6vids = 0;
  5016.         ret = sandybridge_pcode_read(dev_priv, GEN6_PCODE_READ_RC6VIDS, &rc6vids);
  5017.         if (IS_GEN6(dev) && ret) {
  5018.                 DRM_DEBUG_DRIVER("Couldn't check for BIOS workaround\n");
  5019.         } else if (IS_GEN6(dev) && (GEN6_DECODE_RC6_VID(rc6vids & 0xff) < 450)) {
  5020.                 DRM_DEBUG_DRIVER("You should update your BIOS. Correcting minimum rc6 voltage (%dmV->%dmV)\n",
  5021.                           GEN6_DECODE_RC6_VID(rc6vids & 0xff), 450);
  5022.                 rc6vids &= 0xffff00;
  5023.                 rc6vids |= GEN6_ENCODE_RC6_VID(450);
  5024.                 ret = sandybridge_pcode_write(dev_priv, GEN6_PCODE_WRITE_RC6VIDS, rc6vids);
  5025.                 if (ret)
  5026.                         DRM_ERROR("Couldn't fix incorrect rc6 voltage\n");
  5027.         }
  5028.  
  5029.         intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
  5030. }
  5031.  
  5032. static void __gen6_update_ring_freq(struct drm_device *dev)
  5033. {
  5034.         struct drm_i915_private *dev_priv = dev->dev_private;
  5035.         int min_freq = 15;
  5036.         unsigned int gpu_freq;
  5037.         unsigned int max_ia_freq, min_ring_freq;
  5038.         unsigned int max_gpu_freq, min_gpu_freq;
  5039.         int scaling_factor = 180;
  5040.         struct cpufreq_policy *policy;
  5041.  
  5042.         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
  5043.  
  5044.         max_ia_freq = cpufreq_quick_get_max(0);
  5045.         /*
  5046.                  * Default to measured freq if none found, PCU will ensure we
  5047.                  * don't go over
  5048.          */
  5049.                 max_ia_freq = tsc_khz;
  5050.  
  5051.         /* Convert from kHz to MHz */
  5052.         max_ia_freq /= 1000;
  5053.  
  5054.         min_ring_freq = I915_READ(DCLK) & 0xf;
  5055.         /* convert DDR frequency from units of 266.6MHz to bandwidth */
  5056.         min_ring_freq = mult_frac(min_ring_freq, 8, 3);
  5057.  
  5058.         if (IS_SKYLAKE(dev)) {
  5059.                 /* Convert GT frequency to 50 HZ units */
  5060.                 min_gpu_freq = dev_priv->rps.min_freq / GEN9_FREQ_SCALER;
  5061.                 max_gpu_freq = dev_priv->rps.max_freq / GEN9_FREQ_SCALER;
  5062.         } else {
  5063.                 min_gpu_freq = dev_priv->rps.min_freq;
  5064.                 max_gpu_freq = dev_priv->rps.max_freq;
  5065.         }
  5066.  
  5067.         /*
  5068.          * For each potential GPU frequency, load a ring frequency we'd like
  5069.          * to use for memory access.  We do this by specifying the IA frequency
  5070.          * the PCU should use as a reference to determine the ring frequency.
  5071.          */
  5072.         for (gpu_freq = max_gpu_freq; gpu_freq >= min_gpu_freq; gpu_freq--) {
  5073.                 int diff = max_gpu_freq - gpu_freq;
  5074.                 unsigned int ia_freq = 0, ring_freq = 0;
  5075.  
  5076.                 if (IS_SKYLAKE(dev)) {
  5077.                         /*
  5078.                          * ring_freq = 2 * GT. ring_freq is in 100MHz units
  5079.                          * No floor required for ring frequency on SKL.
  5080.                          */
  5081.                         ring_freq = gpu_freq;
  5082.                 } else if (INTEL_INFO(dev)->gen >= 8) {
  5083.                         /* max(2 * GT, DDR). NB: GT is 50MHz units */
  5084.                         ring_freq = max(min_ring_freq, gpu_freq);
  5085.                 } else if (IS_HASWELL(dev)) {
  5086.                         ring_freq = mult_frac(gpu_freq, 5, 4);
  5087.                         ring_freq = max(min_ring_freq, ring_freq);
  5088.                         /* leave ia_freq as the default, chosen by cpufreq */
  5089.                 } else {
  5090.                         /* On older processors, there is no separate ring
  5091.                          * clock domain, so in order to boost the bandwidth
  5092.                          * of the ring, we need to upclock the CPU (ia_freq).
  5093.                          *
  5094.                          * For GPU frequencies less than 750MHz,
  5095.                          * just use the lowest ring freq.
  5096.                          */
  5097.                         if (gpu_freq < min_freq)
  5098.                                 ia_freq = 800;
  5099.                         else
  5100.                                 ia_freq = max_ia_freq - ((diff * scaling_factor) / 2);
  5101.                         ia_freq = DIV_ROUND_CLOSEST(ia_freq, 100);
  5102.                 }
  5103.  
  5104.                 sandybridge_pcode_write(dev_priv,
  5105.                                         GEN6_PCODE_WRITE_MIN_FREQ_TABLE,
  5106.                                         ia_freq << GEN6_PCODE_FREQ_IA_RATIO_SHIFT |
  5107.                                         ring_freq << GEN6_PCODE_FREQ_RING_RATIO_SHIFT |
  5108.                                         gpu_freq);
  5109.         }
  5110. }
  5111.  
  5112. void gen6_update_ring_freq(struct drm_device *dev)
  5113. {
  5114.         struct drm_i915_private *dev_priv = dev->dev_private;
  5115.  
  5116.         if (!HAS_CORE_RING_FREQ(dev))
  5117.                 return;
  5118.  
  5119.         mutex_lock(&dev_priv->rps.hw_lock);
  5120.         __gen6_update_ring_freq(dev);
  5121.         mutex_unlock(&dev_priv->rps.hw_lock);
  5122. }
  5123.  
  5124. static int cherryview_rps_max_freq(struct drm_i915_private *dev_priv)
  5125. {
  5126.         struct drm_device *dev = dev_priv->dev;
  5127.         u32 val, rp0;
  5128.  
  5129.         val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE);
  5130.  
  5131.         switch (INTEL_INFO(dev)->eu_total) {
  5132.         case 8:
  5133.                 /* (2 * 4) config */
  5134.                 rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS4EU_FUSE_SHIFT);
  5135.                 break;
  5136.         case 12:
  5137.                 /* (2 * 6) config */
  5138.                 rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS6EU_FUSE_SHIFT);
  5139.                 break;
  5140.         case 16:
  5141.                 /* (2 * 8) config */
  5142.         default:
  5143.                 /* Setting (2 * 8) Min RP0 for any other combination */
  5144.                 rp0 = (val >> FB_GFX_FMAX_AT_VMAX_2SS8EU_FUSE_SHIFT);
  5145.                 break;
  5146.         }
  5147.  
  5148.         rp0 = (rp0 & FB_GFX_FREQ_FUSE_MASK);
  5149.  
  5150.         return rp0;
  5151. }
  5152.  
  5153. static int cherryview_rps_rpe_freq(struct drm_i915_private *dev_priv)
  5154. {
  5155.         u32 val, rpe;
  5156.  
  5157.         val = vlv_punit_read(dev_priv, PUNIT_GPU_DUTYCYCLE_REG);
  5158.         rpe = (val >> PUNIT_GPU_DUTYCYCLE_RPE_FREQ_SHIFT) & PUNIT_GPU_DUTYCYCLE_RPE_FREQ_MASK;
  5159.  
  5160.         return rpe;
  5161. }
  5162.  
  5163. static int cherryview_rps_guar_freq(struct drm_i915_private *dev_priv)
  5164. {
  5165.         u32 val, rp1;
  5166.  
  5167.         val = vlv_punit_read(dev_priv, FB_GFX_FMAX_AT_VMAX_FUSE);
  5168.         rp1 = (val & FB_GFX_FREQ_FUSE_MASK);
  5169.  
  5170.         return rp1;
  5171. }
  5172.  
  5173. static int valleyview_rps_guar_freq(struct drm_i915_private *dev_priv)
  5174. {
  5175.         u32 val, rp1;
  5176.  
  5177.         val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
  5178.  
  5179.         rp1 = (val & FB_GFX_FGUARANTEED_FREQ_FUSE_MASK) >> FB_GFX_FGUARANTEED_FREQ_FUSE_SHIFT;
  5180.  
  5181.         return rp1;
  5182. }
  5183.  
  5184. static int valleyview_rps_max_freq(struct drm_i915_private *dev_priv)
  5185. {
  5186.         u32 val, rp0;
  5187.  
  5188.         val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FREQ_FUSE);
  5189.  
  5190.         rp0 = (val & FB_GFX_MAX_FREQ_FUSE_MASK) >> FB_GFX_MAX_FREQ_FUSE_SHIFT;
  5191.         /* Clamp to max */
  5192.         rp0 = min_t(u32, rp0, 0xea);
  5193.  
  5194.         return rp0;
  5195. }
  5196.  
  5197. static int valleyview_rps_rpe_freq(struct drm_i915_private *dev_priv)
  5198. {
  5199.         u32 val, rpe;
  5200.  
  5201.         val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_LO);
  5202.         rpe = (val & FB_FMAX_VMIN_FREQ_LO_MASK) >> FB_FMAX_VMIN_FREQ_LO_SHIFT;
  5203.         val = vlv_nc_read(dev_priv, IOSF_NC_FB_GFX_FMAX_FUSE_HI);
  5204.         rpe |= (val & FB_FMAX_VMIN_FREQ_HI_MASK) << 5;
  5205.  
  5206.         return rpe;
  5207. }
  5208.  
  5209. static int valleyview_rps_min_freq(struct drm_i915_private *dev_priv)
  5210. {
  5211.         return vlv_punit_read(dev_priv, PUNIT_REG_GPU_LFM) & 0xff;
  5212. }
  5213.  
  5214. /* Check that the pctx buffer wasn't move under us. */
  5215. static void valleyview_check_pctx(struct drm_i915_private *dev_priv)
  5216. {
  5217.         unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
  5218.  
  5219.         WARN_ON(pctx_addr != dev_priv->mm.stolen_base +
  5220.                              dev_priv->vlv_pctx->stolen->start);
  5221. }
  5222.  
  5223.  
  5224. /* Check that the pcbr address is not empty. */
  5225. static void cherryview_check_pctx(struct drm_i915_private *dev_priv)
  5226. {
  5227.         unsigned long pctx_addr = I915_READ(VLV_PCBR) & ~4095;
  5228.  
  5229.         WARN_ON((pctx_addr >> VLV_PCBR_ADDR_SHIFT) == 0);
  5230. }
  5231.  
  5232. static void cherryview_setup_pctx(struct drm_device *dev)
  5233. {
  5234.         struct drm_i915_private *dev_priv = dev->dev_private;
  5235.         unsigned long pctx_paddr, paddr;
  5236.         struct i915_gtt *gtt = &dev_priv->gtt;
  5237.         u32 pcbr;
  5238.         int pctx_size = 32*1024;
  5239.  
  5240.         WARN_ON(!mutex_is_locked(&dev->struct_mutex));
  5241.  
  5242.         pcbr = I915_READ(VLV_PCBR);
  5243.         if ((pcbr >> VLV_PCBR_ADDR_SHIFT) == 0) {
  5244.                 DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
  5245.                 paddr = (dev_priv->mm.stolen_base +
  5246.                          (gtt->stolen_size - pctx_size));
  5247.  
  5248.                 pctx_paddr = (paddr & (~4095));
  5249.                 I915_WRITE(VLV_PCBR, pctx_paddr);
  5250.         }
  5251.  
  5252.         DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR));
  5253. }
  5254.  
  5255. static void valleyview_setup_pctx(struct drm_device *dev)
  5256. {
  5257.         struct drm_i915_private *dev_priv = dev->dev_private;
  5258.         struct drm_i915_gem_object *pctx;
  5259.         unsigned long pctx_paddr;
  5260.         u32 pcbr;
  5261.         int pctx_size = 24*1024;
  5262.  
  5263.         WARN_ON(!mutex_is_locked(&dev->struct_mutex));
  5264.  
  5265.         pcbr = I915_READ(VLV_PCBR);
  5266.         if (pcbr) {
  5267.                 /* BIOS set it up already, grab the pre-alloc'd space */
  5268.                 int pcbr_offset;
  5269.  
  5270.                 pcbr_offset = (pcbr & (~4095)) - dev_priv->mm.stolen_base;
  5271.                 pctx = i915_gem_object_create_stolen_for_preallocated(dev_priv->dev,
  5272.                                                                       pcbr_offset,
  5273.                                                                       I915_GTT_OFFSET_NONE,
  5274.                                                                       pctx_size);
  5275.                 goto out;
  5276.         }
  5277.  
  5278.         DRM_DEBUG_DRIVER("BIOS didn't set up PCBR, fixing up\n");
  5279.  
  5280.         /*
  5281.          * From the Gunit register HAS:
  5282.          * The Gfx driver is expected to program this register and ensure
  5283.          * proper allocation within Gfx stolen memory.  For example, this
  5284.          * register should be programmed such than the PCBR range does not
  5285.          * overlap with other ranges, such as the frame buffer, protected
  5286.          * memory, or any other relevant ranges.
  5287.          */
  5288.         pctx = i915_gem_object_create_stolen(dev, pctx_size);
  5289.         if (!pctx) {
  5290.                 DRM_DEBUG("not enough stolen space for PCTX, disabling\n");
  5291.                 return;
  5292.         }
  5293.  
  5294.         pctx_paddr = dev_priv->mm.stolen_base + pctx->stolen->start;
  5295.         I915_WRITE(VLV_PCBR, pctx_paddr);
  5296.  
  5297. out:
  5298.         DRM_DEBUG_DRIVER("PCBR: 0x%08x\n", I915_READ(VLV_PCBR));
  5299.         dev_priv->vlv_pctx = pctx;
  5300. }
  5301.  
  5302. static void valleyview_cleanup_pctx(struct drm_device *dev)
  5303. {
  5304.         struct drm_i915_private *dev_priv = dev->dev_private;
  5305.  
  5306.         if (WARN_ON(!dev_priv->vlv_pctx))
  5307.                 return;
  5308.  
  5309.         drm_gem_object_unreference(&dev_priv->vlv_pctx->base);
  5310.         dev_priv->vlv_pctx = NULL;
  5311. }
  5312.  
  5313. static void valleyview_init_gt_powersave(struct drm_device *dev)
  5314. {
  5315.         struct drm_i915_private *dev_priv = dev->dev_private;
  5316.         u32 val;
  5317.  
  5318.         valleyview_setup_pctx(dev);
  5319.  
  5320.         mutex_lock(&dev_priv->rps.hw_lock);
  5321.  
  5322.         val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
  5323.         switch ((val >> 6) & 3) {
  5324.         case 0:
  5325.         case 1:
  5326.                 dev_priv->mem_freq = 800;
  5327.                 break;
  5328.         case 2:
  5329.                 dev_priv->mem_freq = 1066;
  5330.                 break;
  5331.         case 3:
  5332.                 dev_priv->mem_freq = 1333;
  5333.                 break;
  5334.         }
  5335.         DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq);
  5336.  
  5337.         dev_priv->rps.max_freq = valleyview_rps_max_freq(dev_priv);
  5338.         dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
  5339.         DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
  5340.                          intel_gpu_freq(dev_priv, dev_priv->rps.max_freq),
  5341.                          dev_priv->rps.max_freq);
  5342.  
  5343.         dev_priv->rps.efficient_freq = valleyview_rps_rpe_freq(dev_priv);
  5344.         DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
  5345.                          intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
  5346.                          dev_priv->rps.efficient_freq);
  5347.  
  5348.         dev_priv->rps.rp1_freq = valleyview_rps_guar_freq(dev_priv);
  5349.         DRM_DEBUG_DRIVER("RP1(Guar Freq) GPU freq: %d MHz (%u)\n",
  5350.                          intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq),
  5351.                          dev_priv->rps.rp1_freq);
  5352.  
  5353.         dev_priv->rps.min_freq = valleyview_rps_min_freq(dev_priv);
  5354.         DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
  5355.                          intel_gpu_freq(dev_priv, dev_priv->rps.min_freq),
  5356.                          dev_priv->rps.min_freq);
  5357.  
  5358.         dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
  5359.  
  5360.         /* Preserve min/max settings in case of re-init */
  5361.         if (dev_priv->rps.max_freq_softlimit == 0)
  5362.                 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
  5363.  
  5364.         if (dev_priv->rps.min_freq_softlimit == 0)
  5365.                 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
  5366.  
  5367.         mutex_unlock(&dev_priv->rps.hw_lock);
  5368. }
  5369.  
  5370. static void cherryview_init_gt_powersave(struct drm_device *dev)
  5371. {
  5372.         struct drm_i915_private *dev_priv = dev->dev_private;
  5373.         u32 val;
  5374.  
  5375.         cherryview_setup_pctx(dev);
  5376.  
  5377.         mutex_lock(&dev_priv->rps.hw_lock);
  5378.  
  5379.         mutex_lock(&dev_priv->sb_lock);
  5380.         val = vlv_cck_read(dev_priv, CCK_FUSE_REG);
  5381.         mutex_unlock(&dev_priv->sb_lock);
  5382.  
  5383.         switch ((val >> 2) & 0x7) {
  5384.         case 3:
  5385.                 dev_priv->mem_freq = 2000;
  5386.                 break;
  5387.         default:
  5388.                 dev_priv->mem_freq = 1600;
  5389.                 break;
  5390.         }
  5391.         DRM_DEBUG_DRIVER("DDR speed: %d MHz\n", dev_priv->mem_freq);
  5392.  
  5393.         dev_priv->rps.max_freq = cherryview_rps_max_freq(dev_priv);
  5394.         dev_priv->rps.rp0_freq = dev_priv->rps.max_freq;
  5395.         DRM_DEBUG_DRIVER("max GPU freq: %d MHz (%u)\n",
  5396.                          intel_gpu_freq(dev_priv, dev_priv->rps.max_freq),
  5397.                          dev_priv->rps.max_freq);
  5398.  
  5399.         dev_priv->rps.efficient_freq = cherryview_rps_rpe_freq(dev_priv);
  5400.         DRM_DEBUG_DRIVER("RPe GPU freq: %d MHz (%u)\n",
  5401.                          intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
  5402.                          dev_priv->rps.efficient_freq);
  5403.  
  5404.         dev_priv->rps.rp1_freq = cherryview_rps_guar_freq(dev_priv);
  5405.         DRM_DEBUG_DRIVER("RP1(Guar) GPU freq: %d MHz (%u)\n",
  5406.                          intel_gpu_freq(dev_priv, dev_priv->rps.rp1_freq),
  5407.                          dev_priv->rps.rp1_freq);
  5408.  
  5409.         /* PUnit validated range is only [RPe, RP0] */
  5410.         dev_priv->rps.min_freq = dev_priv->rps.efficient_freq;
  5411.         DRM_DEBUG_DRIVER("min GPU freq: %d MHz (%u)\n",
  5412.                          intel_gpu_freq(dev_priv, dev_priv->rps.min_freq),
  5413.                          dev_priv->rps.min_freq);
  5414.  
  5415.         WARN_ONCE((dev_priv->rps.max_freq |
  5416.                    dev_priv->rps.efficient_freq |
  5417.                    dev_priv->rps.rp1_freq |
  5418.                    dev_priv->rps.min_freq) & 1,
  5419.                   "Odd GPU freq values\n");
  5420.  
  5421.         dev_priv->rps.idle_freq = dev_priv->rps.min_freq;
  5422.  
  5423.         /* Preserve min/max settings in case of re-init */
  5424.         if (dev_priv->rps.max_freq_softlimit == 0)
  5425.                 dev_priv->rps.max_freq_softlimit = dev_priv->rps.max_freq;
  5426.  
  5427.         if (dev_priv->rps.min_freq_softlimit == 0)
  5428.                 dev_priv->rps.min_freq_softlimit = dev_priv->rps.min_freq;
  5429.  
  5430.         mutex_unlock(&dev_priv->rps.hw_lock);
  5431. }
  5432.  
  5433. static void valleyview_cleanup_gt_powersave(struct drm_device *dev)
  5434. {
  5435.         valleyview_cleanup_pctx(dev);
  5436. }
  5437.  
  5438. static void cherryview_enable_rps(struct drm_device *dev)
  5439. {
  5440.         struct drm_i915_private *dev_priv = dev->dev_private;
  5441.         struct intel_engine_cs *ring;
  5442.         u32 gtfifodbg, val, rc6_mode = 0, pcbr;
  5443.         int i;
  5444.  
  5445.         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
  5446.  
  5447.         gtfifodbg = I915_READ(GTFIFODBG);
  5448.         if (gtfifodbg) {
  5449.                 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
  5450.                                  gtfifodbg);
  5451.                 I915_WRITE(GTFIFODBG, gtfifodbg);
  5452.         }
  5453.  
  5454.         cherryview_check_pctx(dev_priv);
  5455.  
  5456.         /* 1a & 1b: Get forcewake during program sequence. Although the driver
  5457.          * hasn't enabled a state yet where we need forcewake, BIOS may have.*/
  5458.         intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
  5459.  
  5460.         /*  Disable RC states. */
  5461.         I915_WRITE(GEN6_RC_CONTROL, 0);
  5462.  
  5463.         /* 2a: Program RC6 thresholds.*/
  5464.         I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 40 << 16);
  5465.         I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000); /* 12500 * 1280ns */
  5466.         I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25); /* 25 * 1280ns */
  5467.  
  5468.         for_each_ring(ring, dev_priv, i)
  5469.                 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
  5470.         I915_WRITE(GEN6_RC_SLEEP, 0);
  5471.  
  5472.         /* TO threshold set to 500 us ( 0x186 * 1.28 us) */
  5473.         I915_WRITE(GEN6_RC6_THRESHOLD, 0x186);
  5474.  
  5475.         /* allows RC6 residency counter to work */
  5476.         I915_WRITE(VLV_COUNTER_CONTROL,
  5477.                    _MASKED_BIT_ENABLE(VLV_COUNT_RANGE_HIGH |
  5478.                                       VLV_MEDIA_RC6_COUNT_EN |
  5479.                                       VLV_RENDER_RC6_COUNT_EN));
  5480.  
  5481.         /* For now we assume BIOS is allocating and populating the PCBR  */
  5482.         pcbr = I915_READ(VLV_PCBR);
  5483.  
  5484.         /* 3: Enable RC6 */
  5485.         if ((intel_enable_rc6(dev) & INTEL_RC6_ENABLE) &&
  5486.                                                 (pcbr >> VLV_PCBR_ADDR_SHIFT))
  5487.                 rc6_mode = GEN7_RC_CTL_TO_MODE;
  5488.  
  5489.         I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
  5490.  
  5491.         /* 4 Program defaults and thresholds for RPS*/
  5492.         I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
  5493.         I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
  5494.         I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
  5495.         I915_WRITE(GEN6_RP_UP_EI, 66000);
  5496.         I915_WRITE(GEN6_RP_DOWN_EI, 350000);
  5497.  
  5498.         I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
  5499.  
  5500.         /* 5: Enable RPS */
  5501.         I915_WRITE(GEN6_RP_CONTROL,
  5502.                    GEN6_RP_MEDIA_HW_NORMAL_MODE |
  5503.                    GEN6_RP_MEDIA_IS_GFX |
  5504.                    GEN6_RP_ENABLE |
  5505.                    GEN6_RP_UP_BUSY_AVG |
  5506.                    GEN6_RP_DOWN_IDLE_AVG);
  5507.  
  5508.         /* Setting Fixed Bias */
  5509.         val = VLV_OVERRIDE_EN |
  5510.                   VLV_SOC_TDP_EN |
  5511.                   CHV_BIAS_CPU_50_SOC_50;
  5512.         vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val);
  5513.  
  5514.         val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
  5515.  
  5516.         /* RPS code assumes GPLL is used */
  5517.         WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n");
  5518.  
  5519.         DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE));
  5520.         DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
  5521.  
  5522.         dev_priv->rps.cur_freq = (val >> 8) & 0xff;
  5523.         DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
  5524.                          intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
  5525.                          dev_priv->rps.cur_freq);
  5526.  
  5527.         DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
  5528.                          intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
  5529.                          dev_priv->rps.efficient_freq);
  5530.  
  5531.         valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq);
  5532.  
  5533.         intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
  5534. }
  5535.  
  5536. static void valleyview_enable_rps(struct drm_device *dev)
  5537. {
  5538.         struct drm_i915_private *dev_priv = dev->dev_private;
  5539.         struct intel_engine_cs *ring;
  5540.         u32 gtfifodbg, val, rc6_mode = 0;
  5541.         int i;
  5542.  
  5543.         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
  5544.  
  5545.         valleyview_check_pctx(dev_priv);
  5546.  
  5547.         if ((gtfifodbg = I915_READ(GTFIFODBG))) {
  5548.                 DRM_DEBUG_DRIVER("GT fifo had a previous error %x\n",
  5549.                                  gtfifodbg);
  5550.                 I915_WRITE(GTFIFODBG, gtfifodbg);
  5551.         }
  5552.  
  5553.         /* If VLV, Forcewake all wells, else re-direct to regular path */
  5554.         intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
  5555.  
  5556.         /*  Disable RC states. */
  5557.         I915_WRITE(GEN6_RC_CONTROL, 0);
  5558.  
  5559.         I915_WRITE(GEN6_RP_DOWN_TIMEOUT, 1000000);
  5560.         I915_WRITE(GEN6_RP_UP_THRESHOLD, 59400);
  5561.         I915_WRITE(GEN6_RP_DOWN_THRESHOLD, 245000);
  5562.         I915_WRITE(GEN6_RP_UP_EI, 66000);
  5563.         I915_WRITE(GEN6_RP_DOWN_EI, 350000);
  5564.  
  5565.         I915_WRITE(GEN6_RP_IDLE_HYSTERSIS, 10);
  5566.  
  5567.         I915_WRITE(GEN6_RP_CONTROL,
  5568.                    GEN6_RP_MEDIA_TURBO |
  5569.                    GEN6_RP_MEDIA_HW_NORMAL_MODE |
  5570.                    GEN6_RP_MEDIA_IS_GFX |
  5571.                    GEN6_RP_ENABLE |
  5572.                    GEN6_RP_UP_BUSY_AVG |
  5573.                    GEN6_RP_DOWN_IDLE_CONT);
  5574.  
  5575.         I915_WRITE(GEN6_RC6_WAKE_RATE_LIMIT, 0x00280000);
  5576.         I915_WRITE(GEN6_RC_EVALUATION_INTERVAL, 125000);
  5577.         I915_WRITE(GEN6_RC_IDLE_HYSTERSIS, 25);
  5578.  
  5579.         for_each_ring(ring, dev_priv, i)
  5580.                 I915_WRITE(RING_MAX_IDLE(ring->mmio_base), 10);
  5581.  
  5582.         I915_WRITE(GEN6_RC6_THRESHOLD, 0x557);
  5583.  
  5584.         /* allows RC6 residency counter to work */
  5585.         I915_WRITE(VLV_COUNTER_CONTROL,
  5586.                    _MASKED_BIT_ENABLE(VLV_MEDIA_RC0_COUNT_EN |
  5587.                                       VLV_RENDER_RC0_COUNT_EN |
  5588.                                       VLV_MEDIA_RC6_COUNT_EN |
  5589.                                       VLV_RENDER_RC6_COUNT_EN));
  5590.  
  5591.         if (intel_enable_rc6(dev) & INTEL_RC6_ENABLE)
  5592.                 rc6_mode = GEN7_RC_CTL_TO_MODE | VLV_RC_CTL_CTX_RST_PARALLEL;
  5593.  
  5594.         intel_print_rc6_info(dev, rc6_mode);
  5595.  
  5596.         I915_WRITE(GEN6_RC_CONTROL, rc6_mode);
  5597.  
  5598.         /* Setting Fixed Bias */
  5599.         val = VLV_OVERRIDE_EN |
  5600.                   VLV_SOC_TDP_EN |
  5601.                   VLV_BIAS_CPU_125_SOC_875;
  5602.         vlv_punit_write(dev_priv, VLV_TURBO_SOC_OVERRIDE, val);
  5603.  
  5604.         val = vlv_punit_read(dev_priv, PUNIT_REG_GPU_FREQ_STS);
  5605.  
  5606.         /* RPS code assumes GPLL is used */
  5607.         WARN_ONCE((val & GPLLENABLE) == 0, "GPLL not enabled\n");
  5608.  
  5609.         DRM_DEBUG_DRIVER("GPLL enabled? %s\n", yesno(val & GPLLENABLE));
  5610.         DRM_DEBUG_DRIVER("GPU status: 0x%08x\n", val);
  5611.  
  5612.         dev_priv->rps.cur_freq = (val >> 8) & 0xff;
  5613.         DRM_DEBUG_DRIVER("current GPU freq: %d MHz (%u)\n",
  5614.                          intel_gpu_freq(dev_priv, dev_priv->rps.cur_freq),
  5615.                          dev_priv->rps.cur_freq);
  5616.  
  5617.         DRM_DEBUG_DRIVER("setting GPU freq to %d MHz (%u)\n",
  5618.                          intel_gpu_freq(dev_priv, dev_priv->rps.efficient_freq),
  5619.                          dev_priv->rps.efficient_freq);
  5620.  
  5621.         valleyview_set_rps(dev_priv->dev, dev_priv->rps.efficient_freq);
  5622.  
  5623.         intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL);
  5624. }
  5625.  
  5626. static unsigned long intel_pxfreq(u32 vidfreq)
  5627. {
  5628.         unsigned long freq;
  5629.         int div = (vidfreq & 0x3f0000) >> 16;
  5630.         int post = (vidfreq & 0x3000) >> 12;
  5631.         int pre = (vidfreq & 0x7);
  5632.  
  5633.         if (!pre)
  5634.                 return 0;
  5635.  
  5636.         freq = ((div * 133333) / ((1<<post) * pre));
  5637.  
  5638.         return freq;
  5639. }
  5640.  
  5641. static const struct cparams {
  5642.         u16 i;
  5643.         u16 t;
  5644.         u16 m;
  5645.         u16 c;
  5646. } cparams[] = {
  5647.         { 1, 1333, 301, 28664 },
  5648.         { 1, 1066, 294, 24460 },
  5649.         { 1, 800, 294, 25192 },
  5650.         { 0, 1333, 276, 27605 },
  5651.         { 0, 1066, 276, 27605 },
  5652.         { 0, 800, 231, 23784 },
  5653. };
  5654.  
  5655. static unsigned long __i915_chipset_val(struct drm_i915_private *dev_priv)
  5656. {
  5657.         u64 total_count, diff, ret;
  5658.         u32 count1, count2, count3, m = 0, c = 0;
  5659.         unsigned long now = jiffies_to_msecs(jiffies), diff1;
  5660.         int i;
  5661.  
  5662.         assert_spin_locked(&mchdev_lock);
  5663.  
  5664.         diff1 = now - dev_priv->ips.last_time1;
  5665.  
  5666.         /* Prevent division-by-zero if we are asking too fast.
  5667.          * Also, we don't get interesting results if we are polling
  5668.          * faster than once in 10ms, so just return the saved value
  5669.          * in such cases.
  5670.          */
  5671.         if (diff1 <= 10)
  5672.                 return dev_priv->ips.chipset_power;
  5673.  
  5674.         count1 = I915_READ(DMIEC);
  5675.         count2 = I915_READ(DDREC);
  5676.         count3 = I915_READ(CSIEC);
  5677.  
  5678.         total_count = count1 + count2 + count3;
  5679.  
  5680.         /* FIXME: handle per-counter overflow */
  5681.         if (total_count < dev_priv->ips.last_count1) {
  5682.                 diff = ~0UL - dev_priv->ips.last_count1;
  5683.                 diff += total_count;
  5684.         } else {
  5685.                 diff = total_count - dev_priv->ips.last_count1;
  5686.         }
  5687.  
  5688.         for (i = 0; i < ARRAY_SIZE(cparams); i++) {
  5689.                 if (cparams[i].i == dev_priv->ips.c_m &&
  5690.                     cparams[i].t == dev_priv->ips.r_t) {
  5691.                         m = cparams[i].m;
  5692.                         c = cparams[i].c;
  5693.                         break;
  5694.                 }
  5695.         }
  5696.  
  5697.         diff = div_u64(diff, diff1);
  5698.         ret = ((m * diff) + c);
  5699.         ret = div_u64(ret, 10);
  5700.  
  5701.         dev_priv->ips.last_count1 = total_count;
  5702.         dev_priv->ips.last_time1 = now;
  5703.  
  5704.         dev_priv->ips.chipset_power = ret;
  5705.  
  5706.         return ret;
  5707. }
  5708.  
  5709. unsigned long i915_chipset_val(struct drm_i915_private *dev_priv)
  5710. {
  5711.         struct drm_device *dev = dev_priv->dev;
  5712.         unsigned long val;
  5713.  
  5714.         if (INTEL_INFO(dev)->gen != 5)
  5715.                 return 0;
  5716.  
  5717.         spin_lock_irq(&mchdev_lock);
  5718.  
  5719.         val = __i915_chipset_val(dev_priv);
  5720.  
  5721.         spin_unlock_irq(&mchdev_lock);
  5722.  
  5723.         return val;
  5724. }
  5725.  
  5726. unsigned long i915_mch_val(struct drm_i915_private *dev_priv)
  5727. {
  5728.         unsigned long m, x, b;
  5729.         u32 tsfs;
  5730.  
  5731.         tsfs = I915_READ(TSFS);
  5732.  
  5733.         m = ((tsfs & TSFS_SLOPE_MASK) >> TSFS_SLOPE_SHIFT);
  5734.         x = I915_READ8(TR1);
  5735.  
  5736.         b = tsfs & TSFS_INTR_MASK;
  5737.  
  5738.         return ((m * x) / 127) - b;
  5739. }
  5740.  
  5741. static int _pxvid_to_vd(u8 pxvid)
  5742. {
  5743.         if (pxvid == 0)
  5744.                 return 0;
  5745.  
  5746.         if (pxvid >= 8 && pxvid < 31)
  5747.                 pxvid = 31;
  5748.  
  5749.         return (pxvid + 2) * 125;
  5750. }
  5751.  
  5752. static u32 pvid_to_extvid(struct drm_i915_private *dev_priv, u8 pxvid)
  5753. {
  5754.         struct drm_device *dev = dev_priv->dev;
  5755.         const int vd = _pxvid_to_vd(pxvid);
  5756.         const int vm = vd - 1125;
  5757.  
  5758.         if (INTEL_INFO(dev)->is_mobile)
  5759.                 return vm > 0 ? vm : 0;
  5760.  
  5761.         return vd;
  5762. }
  5763.  
  5764. static void __i915_update_gfx_val(struct drm_i915_private *dev_priv)
  5765. {
  5766.         u64 now, diff, diffms;
  5767.         u32 count;
  5768.  
  5769.         assert_spin_locked(&mchdev_lock);
  5770.  
  5771.         now = ktime_get_raw_ns();
  5772.         diffms = now - dev_priv->ips.last_time2;
  5773.         do_div(diffms, NSEC_PER_MSEC);
  5774.  
  5775.         /* Don't divide by 0 */
  5776.         if (!diffms)
  5777.                 return;
  5778.  
  5779.         count = I915_READ(GFXEC);
  5780.  
  5781.         if (count < dev_priv->ips.last_count2) {
  5782.                 diff = ~0UL - dev_priv->ips.last_count2;
  5783.                 diff += count;
  5784.         } else {
  5785.                 diff = count - dev_priv->ips.last_count2;
  5786.         }
  5787.  
  5788.         dev_priv->ips.last_count2 = count;
  5789.         dev_priv->ips.last_time2 = now;
  5790.  
  5791.         /* More magic constants... */
  5792.         diff = diff * 1181;
  5793.         diff = div_u64(diff, diffms * 10);
  5794.         dev_priv->ips.gfx_power = diff;
  5795. }
  5796.  
  5797. void i915_update_gfx_val(struct drm_i915_private *dev_priv)
  5798. {
  5799.         struct drm_device *dev = dev_priv->dev;
  5800.  
  5801.         if (INTEL_INFO(dev)->gen != 5)
  5802.                 return;
  5803.  
  5804.         spin_lock_irq(&mchdev_lock);
  5805.  
  5806.         __i915_update_gfx_val(dev_priv);
  5807.  
  5808.         spin_unlock_irq(&mchdev_lock);
  5809. }
  5810.  
  5811. static unsigned long __i915_gfx_val(struct drm_i915_private *dev_priv)
  5812. {
  5813.         unsigned long t, corr, state1, corr2, state2;
  5814.         u32 pxvid, ext_v;
  5815.  
  5816.         assert_spin_locked(&mchdev_lock);
  5817.  
  5818.         pxvid = I915_READ(PXVFREQ(dev_priv->rps.cur_freq));
  5819.         pxvid = (pxvid >> 24) & 0x7f;
  5820.         ext_v = pvid_to_extvid(dev_priv, pxvid);
  5821.  
  5822.         state1 = ext_v;
  5823.  
  5824.         t = i915_mch_val(dev_priv);
  5825.  
  5826.         /* Revel in the empirically derived constants */
  5827.  
  5828.         /* Correction factor in 1/100000 units */
  5829.         if (t > 80)
  5830.                 corr = ((t * 2349) + 135940);
  5831.         else if (t >= 50)
  5832.                 corr = ((t * 964) + 29317);
  5833.         else /* < 50 */
  5834.                 corr = ((t * 301) + 1004);
  5835.  
  5836.         corr = corr * ((150142 * state1) / 10000 - 78642);
  5837.         corr /= 100000;
  5838.         corr2 = (corr * dev_priv->ips.corr);
  5839.  
  5840.         state2 = (corr2 * state1) / 10000;
  5841.         state2 /= 100; /* convert to mW */
  5842.  
  5843.         __i915_update_gfx_val(dev_priv);
  5844.  
  5845.         return dev_priv->ips.gfx_power + state2;
  5846. }
  5847.  
  5848. unsigned long i915_gfx_val(struct drm_i915_private *dev_priv)
  5849. {
  5850.         struct drm_device *dev = dev_priv->dev;
  5851.         unsigned long val;
  5852.  
  5853.         if (INTEL_INFO(dev)->gen != 5)
  5854.                 return 0;
  5855.  
  5856.         spin_lock_irq(&mchdev_lock);
  5857.  
  5858.         val = __i915_gfx_val(dev_priv);
  5859.  
  5860.         spin_unlock_irq(&mchdev_lock);
  5861.  
  5862.         return val;
  5863. }
  5864.  
  5865. /**
  5866.  * i915_read_mch_val - return value for IPS use
  5867.  *
  5868.  * Calculate and return a value for the IPS driver to use when deciding whether
  5869.  * we have thermal and power headroom to increase CPU or GPU power budget.
  5870.  */
  5871. unsigned long i915_read_mch_val(void)
  5872. {
  5873.         struct drm_i915_private *dev_priv;
  5874.         unsigned long chipset_val, graphics_val, ret = 0;
  5875.  
  5876.         spin_lock_irq(&mchdev_lock);
  5877.         if (!i915_mch_dev)
  5878.                 goto out_unlock;
  5879.         dev_priv = i915_mch_dev;
  5880.  
  5881.         chipset_val = __i915_chipset_val(dev_priv);
  5882.         graphics_val = __i915_gfx_val(dev_priv);
  5883.  
  5884.         ret = chipset_val + graphics_val;
  5885.  
  5886. out_unlock:
  5887.         spin_unlock_irq(&mchdev_lock);
  5888.  
  5889.         return ret;
  5890. }
  5891. EXPORT_SYMBOL_GPL(i915_read_mch_val);
  5892.  
  5893. /**
  5894.  * i915_gpu_raise - raise GPU frequency limit
  5895.  *
  5896.  * Raise the limit; IPS indicates we have thermal headroom.
  5897.  */
  5898. bool i915_gpu_raise(void)
  5899. {
  5900.         struct drm_i915_private *dev_priv;
  5901.         bool ret = true;
  5902.  
  5903.         spin_lock_irq(&mchdev_lock);
  5904.         if (!i915_mch_dev) {
  5905.                 ret = false;
  5906.                 goto out_unlock;
  5907.         }
  5908.         dev_priv = i915_mch_dev;
  5909.  
  5910.         if (dev_priv->ips.max_delay > dev_priv->ips.fmax)
  5911.                 dev_priv->ips.max_delay--;
  5912.  
  5913. out_unlock:
  5914.         spin_unlock_irq(&mchdev_lock);
  5915.  
  5916.         return ret;
  5917. }
  5918. EXPORT_SYMBOL_GPL(i915_gpu_raise);
  5919.  
  5920. /**
  5921.  * i915_gpu_lower - lower GPU frequency limit
  5922.  *
  5923.  * IPS indicates we're close to a thermal limit, so throttle back the GPU
  5924.  * frequency maximum.
  5925.  */
  5926. bool i915_gpu_lower(void)
  5927. {
  5928.         struct drm_i915_private *dev_priv;
  5929.         bool ret = true;
  5930.  
  5931.         spin_lock_irq(&mchdev_lock);
  5932.         if (!i915_mch_dev) {
  5933.                 ret = false;
  5934.                 goto out_unlock;
  5935.         }
  5936.         dev_priv = i915_mch_dev;
  5937.  
  5938.         if (dev_priv->ips.max_delay < dev_priv->ips.min_delay)
  5939.                 dev_priv->ips.max_delay++;
  5940.  
  5941. out_unlock:
  5942.         spin_unlock_irq(&mchdev_lock);
  5943.  
  5944.         return ret;
  5945. }
  5946. EXPORT_SYMBOL_GPL(i915_gpu_lower);
  5947.  
  5948. /**
  5949.  * i915_gpu_busy - indicate GPU business to IPS
  5950.  *
  5951.  * Tell the IPS driver whether or not the GPU is busy.
  5952.  */
  5953. bool i915_gpu_busy(void)
  5954. {
  5955.         struct drm_i915_private *dev_priv;
  5956.         struct intel_engine_cs *ring;
  5957.         bool ret = false;
  5958.         int i;
  5959.  
  5960.         spin_lock_irq(&mchdev_lock);
  5961.         if (!i915_mch_dev)
  5962.                 goto out_unlock;
  5963.         dev_priv = i915_mch_dev;
  5964.  
  5965.         for_each_ring(ring, dev_priv, i)
  5966.                 ret |= !list_empty(&ring->request_list);
  5967.  
  5968. out_unlock:
  5969.         spin_unlock_irq(&mchdev_lock);
  5970.  
  5971.         return ret;
  5972. }
  5973. EXPORT_SYMBOL_GPL(i915_gpu_busy);
  5974.  
  5975. /**
  5976.  * i915_gpu_turbo_disable - disable graphics turbo
  5977.  *
  5978.  * Disable graphics turbo by resetting the max frequency and setting the
  5979.  * current frequency to the default.
  5980.  */
  5981. bool i915_gpu_turbo_disable(void)
  5982. {
  5983.         struct drm_i915_private *dev_priv;
  5984.         bool ret = true;
  5985.  
  5986.         spin_lock_irq(&mchdev_lock);
  5987.         if (!i915_mch_dev) {
  5988.                 ret = false;
  5989.                 goto out_unlock;
  5990.         }
  5991.         dev_priv = i915_mch_dev;
  5992.  
  5993.         dev_priv->ips.max_delay = dev_priv->ips.fstart;
  5994.  
  5995.         if (!ironlake_set_drps(dev_priv->dev, dev_priv->ips.fstart))
  5996.                 ret = false;
  5997.  
  5998. out_unlock:
  5999.         spin_unlock_irq(&mchdev_lock);
  6000.  
  6001.         return ret;
  6002. }
  6003. EXPORT_SYMBOL_GPL(i915_gpu_turbo_disable);
  6004.  
  6005. /**
  6006.  * Tells the intel_ips driver that the i915 driver is now loaded, if
  6007.  * IPS got loaded first.
  6008.  *
  6009.  * This awkward dance is so that neither module has to depend on the
  6010.  * other in order for IPS to do the appropriate communication of
  6011.  * GPU turbo limits to i915.
  6012.  */
  6013. static void
  6014. ips_ping_for_i915_load(void)
  6015. {
  6016.         void (*link)(void);
  6017.  
  6018. //   link = symbol_get(ips_link_to_i915_driver);
  6019. //   if (link) {
  6020. //       link();
  6021. //       symbol_put(ips_link_to_i915_driver);
  6022. //   }
  6023. }
  6024.  
  6025. void intel_gpu_ips_init(struct drm_i915_private *dev_priv)
  6026. {
  6027.         /* We only register the i915 ips part with intel-ips once everything is
  6028.          * set up, to avoid intel-ips sneaking in and reading bogus values. */
  6029.         spin_lock_irq(&mchdev_lock);
  6030.         i915_mch_dev = dev_priv;
  6031.         spin_unlock_irq(&mchdev_lock);
  6032.  
  6033.         ips_ping_for_i915_load();
  6034. }
  6035.  
  6036. void intel_gpu_ips_teardown(void)
  6037. {
  6038.         spin_lock_irq(&mchdev_lock);
  6039.         i915_mch_dev = NULL;
  6040.         spin_unlock_irq(&mchdev_lock);
  6041. }
  6042.  
  6043. static void intel_init_emon(struct drm_device *dev)
  6044. {
  6045.         struct drm_i915_private *dev_priv = dev->dev_private;
  6046.         u32 lcfuse;
  6047.         u8 pxw[16];
  6048.         int i;
  6049.  
  6050.         /* Disable to program */
  6051.         I915_WRITE(ECR, 0);
  6052.         POSTING_READ(ECR);
  6053.  
  6054.         /* Program energy weights for various events */
  6055.         I915_WRITE(SDEW, 0x15040d00);
  6056.         I915_WRITE(CSIEW0, 0x007f0000);
  6057.         I915_WRITE(CSIEW1, 0x1e220004);
  6058.         I915_WRITE(CSIEW2, 0x04000004);
  6059.  
  6060.         for (i = 0; i < 5; i++)
  6061.                 I915_WRITE(PEW(i), 0);
  6062.         for (i = 0; i < 3; i++)
  6063.                 I915_WRITE(DEW(i), 0);
  6064.  
  6065.         /* Program P-state weights to account for frequency power adjustment */
  6066.         for (i = 0; i < 16; i++) {
  6067.                 u32 pxvidfreq = I915_READ(PXVFREQ(i));
  6068.                 unsigned long freq = intel_pxfreq(pxvidfreq);
  6069.                 unsigned long vid = (pxvidfreq & PXVFREQ_PX_MASK) >>
  6070.                         PXVFREQ_PX_SHIFT;
  6071.                 unsigned long val;
  6072.  
  6073.                 val = vid * vid;
  6074.                 val *= (freq / 1000);
  6075.                 val *= 255;
  6076.                 val /= (127*127*900);
  6077.                 if (val > 0xff)
  6078.                         DRM_ERROR("bad pxval: %ld\n", val);
  6079.                 pxw[i] = val;
  6080.         }
  6081.         /* Render standby states get 0 weight */
  6082.         pxw[14] = 0;
  6083.         pxw[15] = 0;
  6084.  
  6085.         for (i = 0; i < 4; i++) {
  6086.                 u32 val = (pxw[i*4] << 24) | (pxw[(i*4)+1] << 16) |
  6087.                         (pxw[(i*4)+2] << 8) | (pxw[(i*4)+3]);
  6088.                 I915_WRITE(PXW(i), val);
  6089.         }
  6090.  
  6091.         /* Adjust magic regs to magic values (more experimental results) */
  6092.         I915_WRITE(OGW0, 0);
  6093.         I915_WRITE(OGW1, 0);
  6094.         I915_WRITE(EG0, 0x00007f00);
  6095.         I915_WRITE(EG1, 0x0000000e);
  6096.         I915_WRITE(EG2, 0x000e0000);
  6097.         I915_WRITE(EG3, 0x68000300);
  6098.         I915_WRITE(EG4, 0x42000000);
  6099.         I915_WRITE(EG5, 0x00140031);
  6100.         I915_WRITE(EG6, 0);
  6101.         I915_WRITE(EG7, 0);
  6102.  
  6103.         for (i = 0; i < 8; i++)
  6104.                 I915_WRITE(PXWL(i), 0);
  6105.  
  6106.         /* Enable PMON + select events */
  6107.         I915_WRITE(ECR, 0x80000019);
  6108.  
  6109.         lcfuse = I915_READ(LCFUSE02);
  6110.  
  6111.         dev_priv->ips.corr = (lcfuse & LCFUSE_HIV_MASK);
  6112. }
  6113.  
  6114. void intel_init_gt_powersave(struct drm_device *dev)
  6115. {
  6116.         i915.enable_rc6 = sanitize_rc6_option(dev, i915.enable_rc6);
  6117.  
  6118.         if (IS_CHERRYVIEW(dev))
  6119.                 cherryview_init_gt_powersave(dev);
  6120.         else if (IS_VALLEYVIEW(dev))
  6121.                 valleyview_init_gt_powersave(dev);
  6122. }
  6123.  
  6124. void intel_cleanup_gt_powersave(struct drm_device *dev)
  6125. {
  6126.         if (IS_CHERRYVIEW(dev))
  6127.                 return;
  6128.         else if (IS_VALLEYVIEW(dev))
  6129.                 valleyview_cleanup_gt_powersave(dev);
  6130. }
  6131.  
  6132. static void gen6_suspend_rps(struct drm_device *dev)
  6133. {
  6134.         struct drm_i915_private *dev_priv = dev->dev_private;
  6135.  
  6136. //   flush_delayed_work(&dev_priv->rps.delayed_resume_work);
  6137.  
  6138.         gen6_disable_rps_interrupts(dev);
  6139. }
  6140.  
  6141. /**
  6142.  * intel_suspend_gt_powersave - suspend PM work and helper threads
  6143.  * @dev: drm device
  6144.  *
  6145.  * We don't want to disable RC6 or other features here, we just want
  6146.  * to make sure any work we've queued has finished and won't bother
  6147.  * us while we're suspended.
  6148.  */
  6149. void intel_suspend_gt_powersave(struct drm_device *dev)
  6150. {
  6151.         struct drm_i915_private *dev_priv = dev->dev_private;
  6152.  
  6153.         if (INTEL_INFO(dev)->gen < 6)
  6154.                 return;
  6155.  
  6156.         gen6_suspend_rps(dev);
  6157.  
  6158.         /* Force GPU to min freq during suspend */
  6159.         gen6_rps_idle(dev_priv);
  6160. }
  6161.  
  6162. void intel_disable_gt_powersave(struct drm_device *dev)
  6163. {
  6164.         struct drm_i915_private *dev_priv = dev->dev_private;
  6165.  
  6166.         if (IS_IRONLAKE_M(dev)) {
  6167.                 ironlake_disable_drps(dev);
  6168.         } else if (INTEL_INFO(dev)->gen >= 6) {
  6169.                 intel_suspend_gt_powersave(dev);
  6170.  
  6171.                 mutex_lock(&dev_priv->rps.hw_lock);
  6172.                 if (INTEL_INFO(dev)->gen >= 9)
  6173.                         gen9_disable_rps(dev);
  6174.                 else if (IS_CHERRYVIEW(dev))
  6175.                         cherryview_disable_rps(dev);
  6176.                 else if (IS_VALLEYVIEW(dev))
  6177.                         valleyview_disable_rps(dev);
  6178.                 else
  6179.                         gen6_disable_rps(dev);
  6180.  
  6181.                 dev_priv->rps.enabled = false;
  6182.                 mutex_unlock(&dev_priv->rps.hw_lock);
  6183.         }
  6184. }
  6185.  
  6186. static void intel_gen6_powersave_work(struct work_struct *work)
  6187. {
  6188.         struct drm_i915_private *dev_priv =
  6189.                 container_of(work, struct drm_i915_private,
  6190.                              rps.delayed_resume_work.work);
  6191.         struct drm_device *dev = dev_priv->dev;
  6192.  
  6193.         mutex_lock(&dev_priv->rps.hw_lock);
  6194.  
  6195.         gen6_reset_rps_interrupts(dev);
  6196.  
  6197.         if (IS_CHERRYVIEW(dev)) {
  6198.                 cherryview_enable_rps(dev);
  6199.         } else if (IS_VALLEYVIEW(dev)) {
  6200.                 valleyview_enable_rps(dev);
  6201.         } else if (INTEL_INFO(dev)->gen >= 9) {
  6202.                 gen9_enable_rc6(dev);
  6203.                 gen9_enable_rps(dev);
  6204.                 if (IS_SKYLAKE(dev))
  6205.                         __gen6_update_ring_freq(dev);
  6206.         } else if (IS_BROADWELL(dev)) {
  6207.                 gen8_enable_rps(dev);
  6208.                 __gen6_update_ring_freq(dev);
  6209.         } else {
  6210.                 gen6_enable_rps(dev);
  6211.                 __gen6_update_ring_freq(dev);
  6212.         }
  6213.  
  6214.         WARN_ON(dev_priv->rps.max_freq < dev_priv->rps.min_freq);
  6215.         WARN_ON(dev_priv->rps.idle_freq > dev_priv->rps.max_freq);
  6216.  
  6217.         WARN_ON(dev_priv->rps.efficient_freq < dev_priv->rps.min_freq);
  6218.         WARN_ON(dev_priv->rps.efficient_freq > dev_priv->rps.max_freq);
  6219.  
  6220.         dev_priv->rps.enabled = true;
  6221.  
  6222.         gen6_enable_rps_interrupts(dev);
  6223.  
  6224.         mutex_unlock(&dev_priv->rps.hw_lock);
  6225.  
  6226.         intel_runtime_pm_put(dev_priv);
  6227. }
  6228.  
  6229. void intel_enable_gt_powersave(struct drm_device *dev)
  6230. {
  6231.         struct drm_i915_private *dev_priv = dev->dev_private;
  6232.  
  6233.         /* Powersaving is controlled by the host when inside a VM */
  6234.         if (intel_vgpu_active(dev))
  6235.                 return;
  6236.  
  6237.         if (IS_IRONLAKE_M(dev)) {
  6238.                 mutex_lock(&dev->struct_mutex);
  6239.                 ironlake_enable_drps(dev);
  6240.                 intel_init_emon(dev);
  6241.                 mutex_unlock(&dev->struct_mutex);
  6242.         } else if (INTEL_INFO(dev)->gen >= 6) {
  6243.                 /*
  6244.                  * PCU communication is slow and this doesn't need to be
  6245.                  * done at any specific time, so do this out of our fast path
  6246.                  * to make resume and init faster.
  6247.                  *
  6248.                  * We depend on the HW RC6 power context save/restore
  6249.                  * mechanism when entering D3 through runtime PM suspend. So
  6250.                  * disable RPM until RPS/RC6 is properly setup. We can only
  6251.                  * get here via the driver load/system resume/runtime resume
  6252.                  * paths, so the _noresume version is enough (and in case of
  6253.                  * runtime resume it's necessary).
  6254.                  */
  6255.                 if (schedule_delayed_work(&dev_priv->rps.delayed_resume_work,
  6256.                                            round_jiffies_up_relative(HZ)))
  6257.                         intel_runtime_pm_get_noresume(dev_priv);
  6258.         }
  6259. }
  6260.  
  6261. void intel_reset_gt_powersave(struct drm_device *dev)
  6262. {
  6263.         struct drm_i915_private *dev_priv = dev->dev_private;
  6264.  
  6265.         if (INTEL_INFO(dev)->gen < 6)
  6266.                 return;
  6267.  
  6268.         gen6_suspend_rps(dev);
  6269.         dev_priv->rps.enabled = false;
  6270. }
  6271.  
  6272. static void ibx_init_clock_gating(struct drm_device *dev)
  6273. {
  6274.         struct drm_i915_private *dev_priv = dev->dev_private;
  6275.  
  6276.         /*
  6277.          * On Ibex Peak and Cougar Point, we need to disable clock
  6278.          * gating for the panel power sequencer or it will fail to
  6279.          * start up when no ports are active.
  6280.          */
  6281.         I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE);
  6282. }
  6283.  
  6284. static void g4x_disable_trickle_feed(struct drm_device *dev)
  6285. {
  6286.         struct drm_i915_private *dev_priv = dev->dev_private;
  6287.         enum pipe pipe;
  6288.  
  6289.         for_each_pipe(dev_priv, pipe) {
  6290.                 I915_WRITE(DSPCNTR(pipe),
  6291.                            I915_READ(DSPCNTR(pipe)) |
  6292.                            DISPPLANE_TRICKLE_FEED_DISABLE);
  6293.  
  6294.                 I915_WRITE(DSPSURF(pipe), I915_READ(DSPSURF(pipe)));
  6295.                 POSTING_READ(DSPSURF(pipe));
  6296.         }
  6297. }
  6298.  
  6299. static void ilk_init_lp_watermarks(struct drm_device *dev)
  6300. {
  6301.         struct drm_i915_private *dev_priv = dev->dev_private;
  6302.  
  6303.         I915_WRITE(WM3_LP_ILK, I915_READ(WM3_LP_ILK) & ~WM1_LP_SR_EN);
  6304.         I915_WRITE(WM2_LP_ILK, I915_READ(WM2_LP_ILK) & ~WM1_LP_SR_EN);
  6305.         I915_WRITE(WM1_LP_ILK, I915_READ(WM1_LP_ILK) & ~WM1_LP_SR_EN);
  6306.  
  6307.         /*
  6308.          * Don't touch WM1S_LP_EN here.
  6309.          * Doing so could cause underruns.
  6310.          */
  6311. }
  6312.  
  6313. static void ironlake_init_clock_gating(struct drm_device *dev)
  6314. {
  6315.         struct drm_i915_private *dev_priv = dev->dev_private;
  6316.         uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
  6317.  
  6318.         /*
  6319.          * Required for FBC
  6320.          * WaFbcDisableDpfcClockGating:ilk
  6321.          */
  6322.         dspclk_gate |= ILK_DPFCRUNIT_CLOCK_GATE_DISABLE |
  6323.                    ILK_DPFCUNIT_CLOCK_GATE_DISABLE |
  6324.                    ILK_DPFDUNIT_CLOCK_GATE_ENABLE;
  6325.  
  6326.         I915_WRITE(PCH_3DCGDIS0,
  6327.                    MARIUNIT_CLOCK_GATE_DISABLE |
  6328.                    SVSMUNIT_CLOCK_GATE_DISABLE);
  6329.         I915_WRITE(PCH_3DCGDIS1,
  6330.                    VFMUNIT_CLOCK_GATE_DISABLE);
  6331.  
  6332.         /*
  6333.          * According to the spec the following bits should be set in
  6334.          * order to enable memory self-refresh
  6335.          * The bit 22/21 of 0x42004
  6336.          * The bit 5 of 0x42020
  6337.          * The bit 15 of 0x45000
  6338.          */
  6339.         I915_WRITE(ILK_DISPLAY_CHICKEN2,
  6340.                    (I915_READ(ILK_DISPLAY_CHICKEN2) |
  6341.                     ILK_DPARB_GATE | ILK_VSDPFD_FULL));
  6342.         dspclk_gate |= ILK_DPARBUNIT_CLOCK_GATE_ENABLE;
  6343.         I915_WRITE(DISP_ARB_CTL,
  6344.                    (I915_READ(DISP_ARB_CTL) |
  6345.                     DISP_FBC_WM_DIS));
  6346.  
  6347.         ilk_init_lp_watermarks(dev);
  6348.  
  6349.         /*
  6350.          * Based on the document from hardware guys the following bits
  6351.          * should be set unconditionally in order to enable FBC.
  6352.          * The bit 22 of 0x42000
  6353.          * The bit 22 of 0x42004
  6354.          * The bit 7,8,9 of 0x42020.
  6355.          */
  6356.         if (IS_IRONLAKE_M(dev)) {
  6357.                 /* WaFbcAsynchFlipDisableFbcQueue:ilk */
  6358.                 I915_WRITE(ILK_DISPLAY_CHICKEN1,
  6359.                            I915_READ(ILK_DISPLAY_CHICKEN1) |
  6360.                            ILK_FBCQ_DIS);
  6361.                 I915_WRITE(ILK_DISPLAY_CHICKEN2,
  6362.                            I915_READ(ILK_DISPLAY_CHICKEN2) |
  6363.                            ILK_DPARB_GATE);
  6364.         }
  6365.  
  6366.         I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
  6367.  
  6368.         I915_WRITE(ILK_DISPLAY_CHICKEN2,
  6369.                    I915_READ(ILK_DISPLAY_CHICKEN2) |
  6370.                    ILK_ELPIN_409_SELECT);
  6371.         I915_WRITE(_3D_CHICKEN2,
  6372.                    _3D_CHICKEN2_WM_READ_PIPELINED << 16 |
  6373.                    _3D_CHICKEN2_WM_READ_PIPELINED);
  6374.  
  6375.         /* WaDisableRenderCachePipelinedFlush:ilk */
  6376.         I915_WRITE(CACHE_MODE_0,
  6377.                    _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
  6378.  
  6379.         /* WaDisable_RenderCache_OperationalFlush:ilk */
  6380.         I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
  6381.  
  6382.         g4x_disable_trickle_feed(dev);
  6383.  
  6384.         ibx_init_clock_gating(dev);
  6385. }
  6386.  
  6387. static void cpt_init_clock_gating(struct drm_device *dev)
  6388. {
  6389.         struct drm_i915_private *dev_priv = dev->dev_private;
  6390.         int pipe;
  6391.         uint32_t val;
  6392.  
  6393.         /*
  6394.          * On Ibex Peak and Cougar Point, we need to disable clock
  6395.          * gating for the panel power sequencer or it will fail to
  6396.          * start up when no ports are active.
  6397.          */
  6398.         I915_WRITE(SOUTH_DSPCLK_GATE_D, PCH_DPLSUNIT_CLOCK_GATE_DISABLE |
  6399.                    PCH_DPLUNIT_CLOCK_GATE_DISABLE |
  6400.                    PCH_CPUNIT_CLOCK_GATE_DISABLE);
  6401.         I915_WRITE(SOUTH_CHICKEN2, I915_READ(SOUTH_CHICKEN2) |
  6402.                    DPLS_EDP_PPS_FIX_DIS);
  6403.         /* The below fixes the weird display corruption, a few pixels shifted
  6404.          * downward, on (only) LVDS of some HP laptops with IVY.
  6405.          */
  6406.         for_each_pipe(dev_priv, pipe) {
  6407.                 val = I915_READ(TRANS_CHICKEN2(pipe));
  6408.                 val |= TRANS_CHICKEN2_TIMING_OVERRIDE;
  6409.                 val &= ~TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
  6410.                 if (dev_priv->vbt.fdi_rx_polarity_inverted)
  6411.                         val |= TRANS_CHICKEN2_FDI_POLARITY_REVERSED;
  6412.                 val &= ~TRANS_CHICKEN2_FRAME_START_DELAY_MASK;
  6413.                 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_COUNTER;
  6414.                 val &= ~TRANS_CHICKEN2_DISABLE_DEEP_COLOR_MODESWITCH;
  6415.                 I915_WRITE(TRANS_CHICKEN2(pipe), val);
  6416.         }
  6417.         /* WADP0ClockGatingDisable */
  6418.         for_each_pipe(dev_priv, pipe) {
  6419.                 I915_WRITE(TRANS_CHICKEN1(pipe),
  6420.                            TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
  6421.         }
  6422. }
  6423.  
  6424. static void gen6_check_mch_setup(struct drm_device *dev)
  6425. {
  6426.         struct drm_i915_private *dev_priv = dev->dev_private;
  6427.         uint32_t tmp;
  6428.  
  6429.         tmp = I915_READ(MCH_SSKPD);
  6430.         if ((tmp & MCH_SSKPD_WM0_MASK) != MCH_SSKPD_WM0_VAL)
  6431.                 DRM_DEBUG_KMS("Wrong MCH_SSKPD value: 0x%08x This can cause underruns.\n",
  6432.                               tmp);
  6433. }
  6434.  
  6435. static void gen6_init_clock_gating(struct drm_device *dev)
  6436. {
  6437.         struct drm_i915_private *dev_priv = dev->dev_private;
  6438.         uint32_t dspclk_gate = ILK_VRHUNIT_CLOCK_GATE_DISABLE;
  6439.  
  6440.         I915_WRITE(ILK_DSPCLK_GATE_D, dspclk_gate);
  6441.  
  6442.         I915_WRITE(ILK_DISPLAY_CHICKEN2,
  6443.                    I915_READ(ILK_DISPLAY_CHICKEN2) |
  6444.                    ILK_ELPIN_409_SELECT);
  6445.  
  6446.         /* WaDisableHiZPlanesWhenMSAAEnabled:snb */
  6447.         I915_WRITE(_3D_CHICKEN,
  6448.                    _MASKED_BIT_ENABLE(_3D_CHICKEN_HIZ_PLANE_DISABLE_MSAA_4X_SNB));
  6449.  
  6450.         /* WaDisable_RenderCache_OperationalFlush:snb */
  6451.         I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
  6452.  
  6453.         /*
  6454.          * BSpec recoomends 8x4 when MSAA is used,
  6455.          * however in practice 16x4 seems fastest.
  6456.          *
  6457.          * Note that PS/WM thread counts depend on the WIZ hashing
  6458.          * disable bit, which we don't touch here, but it's good
  6459.          * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
  6460.          */
  6461.         I915_WRITE(GEN6_GT_MODE,
  6462.                    _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
  6463.  
  6464.         ilk_init_lp_watermarks(dev);
  6465.  
  6466.         I915_WRITE(CACHE_MODE_0,
  6467.                    _MASKED_BIT_DISABLE(CM0_STC_EVICT_DISABLE_LRA_SNB));
  6468.  
  6469.         I915_WRITE(GEN6_UCGCTL1,
  6470.                    I915_READ(GEN6_UCGCTL1) |
  6471.                    GEN6_BLBUNIT_CLOCK_GATE_DISABLE |
  6472.                    GEN6_CSUNIT_CLOCK_GATE_DISABLE);
  6473.  
  6474.         /* According to the BSpec vol1g, bit 12 (RCPBUNIT) clock
  6475.          * gating disable must be set.  Failure to set it results in
  6476.          * flickering pixels due to Z write ordering failures after
  6477.          * some amount of runtime in the Mesa "fire" demo, and Unigine
  6478.          * Sanctuary and Tropics, and apparently anything else with
  6479.          * alpha test or pixel discard.
  6480.          *
  6481.          * According to the spec, bit 11 (RCCUNIT) must also be set,
  6482.          * but we didn't debug actual testcases to find it out.
  6483.          *
  6484.          * WaDisableRCCUnitClockGating:snb
  6485.          * WaDisableRCPBUnitClockGating:snb
  6486.          */
  6487.         I915_WRITE(GEN6_UCGCTL2,
  6488.                    GEN6_RCPBUNIT_CLOCK_GATE_DISABLE |
  6489.                    GEN6_RCCUNIT_CLOCK_GATE_DISABLE);
  6490.  
  6491.         /* WaStripsFansDisableFastClipPerformanceFix:snb */
  6492.         I915_WRITE(_3D_CHICKEN3,
  6493.                    _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_FASTCLIP_CULL));
  6494.  
  6495.         /*
  6496.          * Bspec says:
  6497.          * "This bit must be set if 3DSTATE_CLIP clip mode is set to normal and
  6498.          * 3DSTATE_SF number of SF output attributes is more than 16."
  6499.          */
  6500.         I915_WRITE(_3D_CHICKEN3,
  6501.                    _MASKED_BIT_ENABLE(_3D_CHICKEN3_SF_DISABLE_PIPELINED_ATTR_FETCH));
  6502.  
  6503.         /*
  6504.          * According to the spec the following bits should be
  6505.          * set in order to enable memory self-refresh and fbc:
  6506.          * The bit21 and bit22 of 0x42000
  6507.          * The bit21 and bit22 of 0x42004
  6508.          * The bit5 and bit7 of 0x42020
  6509.          * The bit14 of 0x70180
  6510.          * The bit14 of 0x71180
  6511.          *
  6512.          * WaFbcAsynchFlipDisableFbcQueue:snb
  6513.          */
  6514.         I915_WRITE(ILK_DISPLAY_CHICKEN1,
  6515.                    I915_READ(ILK_DISPLAY_CHICKEN1) |
  6516.                    ILK_FBCQ_DIS | ILK_PABSTRETCH_DIS);
  6517.         I915_WRITE(ILK_DISPLAY_CHICKEN2,
  6518.                    I915_READ(ILK_DISPLAY_CHICKEN2) |
  6519.                    ILK_DPARB_GATE | ILK_VSDPFD_FULL);
  6520.         I915_WRITE(ILK_DSPCLK_GATE_D,
  6521.                    I915_READ(ILK_DSPCLK_GATE_D) |
  6522.                    ILK_DPARBUNIT_CLOCK_GATE_ENABLE  |
  6523.                    ILK_DPFDUNIT_CLOCK_GATE_ENABLE);
  6524.  
  6525.         g4x_disable_trickle_feed(dev);
  6526.  
  6527.         cpt_init_clock_gating(dev);
  6528.  
  6529.         gen6_check_mch_setup(dev);
  6530. }
  6531.  
  6532. static void gen7_setup_fixed_func_scheduler(struct drm_i915_private *dev_priv)
  6533. {
  6534.         uint32_t reg = I915_READ(GEN7_FF_THREAD_MODE);
  6535.  
  6536.         /*
  6537.          * WaVSThreadDispatchOverride:ivb,vlv
  6538.          *
  6539.          * This actually overrides the dispatch
  6540.          * mode for all thread types.
  6541.          */
  6542.         reg &= ~GEN7_FF_SCHED_MASK;
  6543.         reg |= GEN7_FF_TS_SCHED_HW;
  6544.         reg |= GEN7_FF_VS_SCHED_HW;
  6545.         reg |= GEN7_FF_DS_SCHED_HW;
  6546.  
  6547.         I915_WRITE(GEN7_FF_THREAD_MODE, reg);
  6548. }
  6549.  
  6550. static void lpt_init_clock_gating(struct drm_device *dev)
  6551. {
  6552.         struct drm_i915_private *dev_priv = dev->dev_private;
  6553.  
  6554.         /*
  6555.          * TODO: this bit should only be enabled when really needed, then
  6556.          * disabled when not needed anymore in order to save power.
  6557.          */
  6558.         if (HAS_PCH_LPT_LP(dev))
  6559.                 I915_WRITE(SOUTH_DSPCLK_GATE_D,
  6560.                            I915_READ(SOUTH_DSPCLK_GATE_D) |
  6561.                            PCH_LP_PARTITION_LEVEL_DISABLE);
  6562.  
  6563.         /* WADPOClockGatingDisable:hsw */
  6564.         I915_WRITE(TRANS_CHICKEN1(PIPE_A),
  6565.                    I915_READ(TRANS_CHICKEN1(PIPE_A)) |
  6566.                    TRANS_CHICKEN1_DP0UNIT_GC_DISABLE);
  6567. }
  6568.  
  6569. static void lpt_suspend_hw(struct drm_device *dev)
  6570. {
  6571.         struct drm_i915_private *dev_priv = dev->dev_private;
  6572.  
  6573.         if (HAS_PCH_LPT_LP(dev)) {
  6574.                 uint32_t val = I915_READ(SOUTH_DSPCLK_GATE_D);
  6575.  
  6576.                 val &= ~PCH_LP_PARTITION_LEVEL_DISABLE;
  6577.                 I915_WRITE(SOUTH_DSPCLK_GATE_D, val);
  6578.         }
  6579. }
  6580.  
  6581. static void broadwell_init_clock_gating(struct drm_device *dev)
  6582. {
  6583.         struct drm_i915_private *dev_priv = dev->dev_private;
  6584.         enum pipe pipe;
  6585.         uint32_t misccpctl;
  6586.  
  6587.         ilk_init_lp_watermarks(dev);
  6588.  
  6589.         /* WaSwitchSolVfFArbitrationPriority:bdw */
  6590.         I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
  6591.  
  6592.         /* WaPsrDPAMaskVBlankInSRD:bdw */
  6593.         I915_WRITE(CHICKEN_PAR1_1,
  6594.                    I915_READ(CHICKEN_PAR1_1) | DPA_MASK_VBLANK_SRD);
  6595.  
  6596.         /* WaPsrDPRSUnmaskVBlankInSRD:bdw */
  6597.         for_each_pipe(dev_priv, pipe) {
  6598.                 I915_WRITE(CHICKEN_PIPESL_1(pipe),
  6599.                            I915_READ(CHICKEN_PIPESL_1(pipe)) |
  6600.                            BDW_DPRS_MASK_VBLANK_SRD);
  6601.         }
  6602.  
  6603.         /* WaVSRefCountFullforceMissDisable:bdw */
  6604.         /* WaDSRefCountFullforceMissDisable:bdw */
  6605.         I915_WRITE(GEN7_FF_THREAD_MODE,
  6606.                    I915_READ(GEN7_FF_THREAD_MODE) &
  6607.                    ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
  6608.  
  6609.         I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
  6610.                    _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
  6611.  
  6612.         /* WaDisableSDEUnitClockGating:bdw */
  6613.         I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
  6614.                    GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
  6615.  
  6616.         /*
  6617.          * WaProgramL3SqcReg1Default:bdw
  6618.          * WaTempDisableDOPClkGating:bdw
  6619.          */
  6620.         misccpctl = I915_READ(GEN7_MISCCPCTL);
  6621.         I915_WRITE(GEN7_MISCCPCTL, misccpctl & ~GEN7_DOP_CLOCK_GATE_ENABLE);
  6622.         I915_WRITE(GEN8_L3SQCREG1, BDW_WA_L3SQCREG1_DEFAULT);
  6623.         I915_WRITE(GEN7_MISCCPCTL, misccpctl);
  6624.  
  6625.         /*
  6626.          * WaGttCachingOffByDefault:bdw
  6627.          * GTT cache may not work with big pages, so if those
  6628.          * are ever enabled GTT cache may need to be disabled.
  6629.          */
  6630.         I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL);
  6631.  
  6632.         lpt_init_clock_gating(dev);
  6633. }
  6634.  
  6635. static void haswell_init_clock_gating(struct drm_device *dev)
  6636. {
  6637.         struct drm_i915_private *dev_priv = dev->dev_private;
  6638.  
  6639.         ilk_init_lp_watermarks(dev);
  6640.  
  6641.         /* L3 caching of data atomics doesn't work -- disable it. */
  6642.         I915_WRITE(HSW_SCRATCH1, HSW_SCRATCH1_L3_DATA_ATOMICS_DISABLE);
  6643.         I915_WRITE(HSW_ROW_CHICKEN3,
  6644.                    _MASKED_BIT_ENABLE(HSW_ROW_CHICKEN3_L3_GLOBAL_ATOMICS_DISABLE));
  6645.  
  6646.         /* This is required by WaCatErrorRejectionIssue:hsw */
  6647.         I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
  6648.                         I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
  6649.                         GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
  6650.  
  6651.         /* WaVSRefCountFullforceMissDisable:hsw */
  6652.         I915_WRITE(GEN7_FF_THREAD_MODE,
  6653.                    I915_READ(GEN7_FF_THREAD_MODE) & ~GEN7_FF_VS_REF_CNT_FFME);
  6654.  
  6655.         /* WaDisable_RenderCache_OperationalFlush:hsw */
  6656.         I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
  6657.  
  6658.         /* enable HiZ Raw Stall Optimization */
  6659.         I915_WRITE(CACHE_MODE_0_GEN7,
  6660.                    _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
  6661.  
  6662.         /* WaDisable4x2SubspanOptimization:hsw */
  6663.         I915_WRITE(CACHE_MODE_1,
  6664.                    _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
  6665.  
  6666.         /*
  6667.          * BSpec recommends 8x4 when MSAA is used,
  6668.          * however in practice 16x4 seems fastest.
  6669.          *
  6670.          * Note that PS/WM thread counts depend on the WIZ hashing
  6671.          * disable bit, which we don't touch here, but it's good
  6672.          * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
  6673.          */
  6674.         I915_WRITE(GEN7_GT_MODE,
  6675.                    _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
  6676.  
  6677.         /* WaSampleCChickenBitEnable:hsw */
  6678.         I915_WRITE(HALF_SLICE_CHICKEN3,
  6679.                    _MASKED_BIT_ENABLE(HSW_SAMPLE_C_PERFORMANCE));
  6680.  
  6681.         /* WaSwitchSolVfFArbitrationPriority:hsw */
  6682.         I915_WRITE(GAM_ECOCHK, I915_READ(GAM_ECOCHK) | HSW_ECOCHK_ARB_PRIO_SOL);
  6683.  
  6684.         /* WaRsPkgCStateDisplayPMReq:hsw */
  6685.         I915_WRITE(CHICKEN_PAR1_1,
  6686.                    I915_READ(CHICKEN_PAR1_1) | FORCE_ARB_IDLE_PLANES);
  6687.  
  6688.         lpt_init_clock_gating(dev);
  6689. }
  6690.  
  6691. static void ivybridge_init_clock_gating(struct drm_device *dev)
  6692. {
  6693.         struct drm_i915_private *dev_priv = dev->dev_private;
  6694.         uint32_t snpcr;
  6695.  
  6696.         ilk_init_lp_watermarks(dev);
  6697.  
  6698.         I915_WRITE(ILK_DSPCLK_GATE_D, ILK_VRHUNIT_CLOCK_GATE_DISABLE);
  6699.  
  6700.         /* WaDisableEarlyCull:ivb */
  6701.         I915_WRITE(_3D_CHICKEN3,
  6702.                    _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
  6703.  
  6704.         /* WaDisableBackToBackFlipFix:ivb */
  6705.         I915_WRITE(IVB_CHICKEN3,
  6706.                    CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
  6707.                    CHICKEN3_DGMG_DONE_FIX_DISABLE);
  6708.  
  6709.         /* WaDisablePSDDualDispatchEnable:ivb */
  6710.         if (IS_IVB_GT1(dev))
  6711.                 I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
  6712.                            _MASKED_BIT_ENABLE(GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
  6713.  
  6714.         /* WaDisable_RenderCache_OperationalFlush:ivb */
  6715.         I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
  6716.  
  6717.         /* Apply the WaDisableRHWOOptimizationForRenderHang:ivb workaround. */
  6718.         I915_WRITE(GEN7_COMMON_SLICE_CHICKEN1,
  6719.                    GEN7_CSC1_RHWO_OPT_DISABLE_IN_RCC);
  6720.  
  6721.         /* WaApplyL3ControlAndL3ChickenMode:ivb */
  6722.         I915_WRITE(GEN7_L3CNTLREG1,
  6723.                         GEN7_WA_FOR_GEN7_L3_CONTROL);
  6724.         I915_WRITE(GEN7_L3_CHICKEN_MODE_REGISTER,
  6725.                    GEN7_WA_L3_CHICKEN_MODE);
  6726.         if (IS_IVB_GT1(dev))
  6727.                 I915_WRITE(GEN7_ROW_CHICKEN2,
  6728.                            _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
  6729.         else {
  6730.                 /* must write both registers */
  6731.                 I915_WRITE(GEN7_ROW_CHICKEN2,
  6732.                            _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
  6733.                 I915_WRITE(GEN7_ROW_CHICKEN2_GT2,
  6734.                            _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
  6735.         }
  6736.  
  6737.         /* WaForceL3Serialization:ivb */
  6738.         I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
  6739.                    ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
  6740.  
  6741.         /*
  6742.          * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
  6743.          * This implements the WaDisableRCZUnitClockGating:ivb workaround.
  6744.          */
  6745.         I915_WRITE(GEN6_UCGCTL2,
  6746.                    GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
  6747.  
  6748.         /* This is required by WaCatErrorRejectionIssue:ivb */
  6749.         I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
  6750.                         I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
  6751.                         GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
  6752.  
  6753.         g4x_disable_trickle_feed(dev);
  6754.  
  6755.         gen7_setup_fixed_func_scheduler(dev_priv);
  6756.  
  6757.         if (0) { /* causes HiZ corruption on ivb:gt1 */
  6758.                 /* enable HiZ Raw Stall Optimization */
  6759.                 I915_WRITE(CACHE_MODE_0_GEN7,
  6760.                            _MASKED_BIT_DISABLE(HIZ_RAW_STALL_OPT_DISABLE));
  6761.         }
  6762.  
  6763.         /* WaDisable4x2SubspanOptimization:ivb */
  6764.         I915_WRITE(CACHE_MODE_1,
  6765.                    _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
  6766.  
  6767.         /*
  6768.          * BSpec recommends 8x4 when MSAA is used,
  6769.          * however in practice 16x4 seems fastest.
  6770.          *
  6771.          * Note that PS/WM thread counts depend on the WIZ hashing
  6772.          * disable bit, which we don't touch here, but it's good
  6773.          * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
  6774.          */
  6775.         I915_WRITE(GEN7_GT_MODE,
  6776.                    _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
  6777.  
  6778.         snpcr = I915_READ(GEN6_MBCUNIT_SNPCR);
  6779.         snpcr &= ~GEN6_MBC_SNPCR_MASK;
  6780.         snpcr |= GEN6_MBC_SNPCR_MED;
  6781.         I915_WRITE(GEN6_MBCUNIT_SNPCR, snpcr);
  6782.  
  6783.         if (!HAS_PCH_NOP(dev))
  6784.                 cpt_init_clock_gating(dev);
  6785.  
  6786.         gen6_check_mch_setup(dev);
  6787. }
  6788.  
  6789. static void vlv_init_display_clock_gating(struct drm_i915_private *dev_priv)
  6790. {
  6791.         I915_WRITE(DSPCLK_GATE_D, VRHUNIT_CLOCK_GATE_DISABLE);
  6792.  
  6793.         /*
  6794.          * Disable trickle feed and enable pnd deadline calculation
  6795.          */
  6796.         I915_WRITE(MI_ARB_VLV, MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE);
  6797.         I915_WRITE(CBR1_VLV, 0);
  6798. }
  6799.  
  6800. static void valleyview_init_clock_gating(struct drm_device *dev)
  6801. {
  6802.         struct drm_i915_private *dev_priv = dev->dev_private;
  6803.  
  6804.         vlv_init_display_clock_gating(dev_priv);
  6805.  
  6806.         /* WaDisableEarlyCull:vlv */
  6807.         I915_WRITE(_3D_CHICKEN3,
  6808.                    _MASKED_BIT_ENABLE(_3D_CHICKEN_SF_DISABLE_OBJEND_CULL));
  6809.  
  6810.         /* WaDisableBackToBackFlipFix:vlv */
  6811.         I915_WRITE(IVB_CHICKEN3,
  6812.                    CHICKEN3_DGMG_REQ_OUT_FIX_DISABLE |
  6813.                    CHICKEN3_DGMG_DONE_FIX_DISABLE);
  6814.  
  6815.         /* WaPsdDispatchEnable:vlv */
  6816.         /* WaDisablePSDDualDispatchEnable:vlv */
  6817.         I915_WRITE(GEN7_HALF_SLICE_CHICKEN1,
  6818.                    _MASKED_BIT_ENABLE(GEN7_MAX_PS_THREAD_DEP |
  6819.                                       GEN7_PSD_SINGLE_PORT_DISPATCH_ENABLE));
  6820.  
  6821.         /* WaDisable_RenderCache_OperationalFlush:vlv */
  6822.         I915_WRITE(CACHE_MODE_0_GEN7, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
  6823.  
  6824.         /* WaForceL3Serialization:vlv */
  6825.         I915_WRITE(GEN7_L3SQCREG4, I915_READ(GEN7_L3SQCREG4) &
  6826.                    ~L3SQ_URB_READ_CAM_MATCH_DISABLE);
  6827.  
  6828.         /* WaDisableDopClockGating:vlv */
  6829.         I915_WRITE(GEN7_ROW_CHICKEN2,
  6830.                    _MASKED_BIT_ENABLE(DOP_CLOCK_GATING_DISABLE));
  6831.  
  6832.         /* This is required by WaCatErrorRejectionIssue:vlv */
  6833.         I915_WRITE(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG,
  6834.                    I915_READ(GEN7_SQ_CHICKEN_MBCUNIT_CONFIG) |
  6835.                    GEN7_SQ_CHICKEN_MBCUNIT_SQINTMOB);
  6836.  
  6837.         gen7_setup_fixed_func_scheduler(dev_priv);
  6838.  
  6839.         /*
  6840.          * According to the spec, bit 13 (RCZUNIT) must be set on IVB.
  6841.          * This implements the WaDisableRCZUnitClockGating:vlv workaround.
  6842.          */
  6843.         I915_WRITE(GEN6_UCGCTL2,
  6844.                    GEN6_RCZUNIT_CLOCK_GATE_DISABLE);
  6845.  
  6846.         /* WaDisableL3Bank2xClockGate:vlv
  6847.          * Disabling L3 clock gating- MMIO 940c[25] = 1
  6848.          * Set bit 25, to disable L3_BANK_2x_CLK_GATING */
  6849.         I915_WRITE(GEN7_UCGCTL4,
  6850.                    I915_READ(GEN7_UCGCTL4) | GEN7_L3BANK2X_CLOCK_GATE_DISABLE);
  6851.  
  6852.         /*
  6853.          * BSpec says this must be set, even though
  6854.          * WaDisable4x2SubspanOptimization isn't listed for VLV.
  6855.          */
  6856.         I915_WRITE(CACHE_MODE_1,
  6857.                    _MASKED_BIT_ENABLE(PIXEL_SUBSPAN_COLLECT_OPT_DISABLE));
  6858.  
  6859.         /*
  6860.          * BSpec recommends 8x4 when MSAA is used,
  6861.          * however in practice 16x4 seems fastest.
  6862.          *
  6863.          * Note that PS/WM thread counts depend on the WIZ hashing
  6864.          * disable bit, which we don't touch here, but it's good
  6865.          * to keep in mind (see 3DSTATE_PS and 3DSTATE_WM).
  6866.          */
  6867.         I915_WRITE(GEN7_GT_MODE,
  6868.                    _MASKED_FIELD(GEN6_WIZ_HASHING_MASK, GEN6_WIZ_HASHING_16x4));
  6869.  
  6870.         /*
  6871.          * WaIncreaseL3CreditsForVLVB0:vlv
  6872.          * This is the hardware default actually.
  6873.          */
  6874.         I915_WRITE(GEN7_L3SQCREG1, VLV_B0_WA_L3SQCREG1_VALUE);
  6875.  
  6876.         /*
  6877.          * WaDisableVLVClockGating_VBIIssue:vlv
  6878.          * Disable clock gating on th GCFG unit to prevent a delay
  6879.          * in the reporting of vblank events.
  6880.          */
  6881.         I915_WRITE(VLV_GUNIT_CLOCK_GATE, GCFG_DIS);
  6882. }
  6883.  
  6884. static void cherryview_init_clock_gating(struct drm_device *dev)
  6885. {
  6886.         struct drm_i915_private *dev_priv = dev->dev_private;
  6887.  
  6888.         vlv_init_display_clock_gating(dev_priv);
  6889.  
  6890.         /* WaVSRefCountFullforceMissDisable:chv */
  6891.         /* WaDSRefCountFullforceMissDisable:chv */
  6892.         I915_WRITE(GEN7_FF_THREAD_MODE,
  6893.                    I915_READ(GEN7_FF_THREAD_MODE) &
  6894.                    ~(GEN8_FF_DS_REF_CNT_FFME | GEN7_FF_VS_REF_CNT_FFME));
  6895.  
  6896.         /* WaDisableSemaphoreAndSyncFlipWait:chv */
  6897.         I915_WRITE(GEN6_RC_SLEEP_PSMI_CONTROL,
  6898.                    _MASKED_BIT_ENABLE(GEN8_RC_SEMA_IDLE_MSG_DISABLE));
  6899.  
  6900.         /* WaDisableCSUnitClockGating:chv */
  6901.         I915_WRITE(GEN6_UCGCTL1, I915_READ(GEN6_UCGCTL1) |
  6902.                    GEN6_CSUNIT_CLOCK_GATE_DISABLE);
  6903.  
  6904.         /* WaDisableSDEUnitClockGating:chv */
  6905.         I915_WRITE(GEN8_UCGCTL6, I915_READ(GEN8_UCGCTL6) |
  6906.                    GEN8_SDEUNIT_CLOCK_GATE_DISABLE);
  6907.  
  6908.         /*
  6909.          * GTT cache may not work with big pages, so if those
  6910.          * are ever enabled GTT cache may need to be disabled.
  6911.          */
  6912.         I915_WRITE(HSW_GTT_CACHE_EN, GTT_CACHE_EN_ALL);
  6913. }
  6914.  
  6915. static void g4x_init_clock_gating(struct drm_device *dev)
  6916. {
  6917.         struct drm_i915_private *dev_priv = dev->dev_private;
  6918.         uint32_t dspclk_gate;
  6919.  
  6920.         I915_WRITE(RENCLK_GATE_D1, 0);
  6921.         I915_WRITE(RENCLK_GATE_D2, VF_UNIT_CLOCK_GATE_DISABLE |
  6922.                    GS_UNIT_CLOCK_GATE_DISABLE |
  6923.                    CL_UNIT_CLOCK_GATE_DISABLE);
  6924.         I915_WRITE(RAMCLK_GATE_D, 0);
  6925.         dspclk_gate = VRHUNIT_CLOCK_GATE_DISABLE |
  6926.                 OVRUNIT_CLOCK_GATE_DISABLE |
  6927.                 OVCUNIT_CLOCK_GATE_DISABLE;
  6928.         if (IS_GM45(dev))
  6929.                 dspclk_gate |= DSSUNIT_CLOCK_GATE_DISABLE;
  6930.         I915_WRITE(DSPCLK_GATE_D, dspclk_gate);
  6931.  
  6932.         /* WaDisableRenderCachePipelinedFlush */
  6933.         I915_WRITE(CACHE_MODE_0,
  6934.                    _MASKED_BIT_ENABLE(CM0_PIPELINED_RENDER_FLUSH_DISABLE));
  6935.  
  6936.         /* WaDisable_RenderCache_OperationalFlush:g4x */
  6937.         I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
  6938.  
  6939.         g4x_disable_trickle_feed(dev);
  6940. }
  6941.  
  6942. static void crestline_init_clock_gating(struct drm_device *dev)
  6943. {
  6944.         struct drm_i915_private *dev_priv = dev->dev_private;
  6945.  
  6946.         I915_WRITE(RENCLK_GATE_D1, I965_RCC_CLOCK_GATE_DISABLE);
  6947.         I915_WRITE(RENCLK_GATE_D2, 0);
  6948.         I915_WRITE(DSPCLK_GATE_D, 0);
  6949.         I915_WRITE(RAMCLK_GATE_D, 0);
  6950.         I915_WRITE16(DEUC, 0);
  6951.         I915_WRITE(MI_ARB_STATE,
  6952.                    _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
  6953.  
  6954.         /* WaDisable_RenderCache_OperationalFlush:gen4 */
  6955.         I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
  6956. }
  6957.  
  6958. static void broadwater_init_clock_gating(struct drm_device *dev)
  6959. {
  6960.         struct drm_i915_private *dev_priv = dev->dev_private;
  6961.  
  6962.         I915_WRITE(RENCLK_GATE_D1, I965_RCZ_CLOCK_GATE_DISABLE |
  6963.                    I965_RCC_CLOCK_GATE_DISABLE |
  6964.                    I965_RCPB_CLOCK_GATE_DISABLE |
  6965.                    I965_ISC_CLOCK_GATE_DISABLE |
  6966.                    I965_FBC_CLOCK_GATE_DISABLE);
  6967.         I915_WRITE(RENCLK_GATE_D2, 0);
  6968.         I915_WRITE(MI_ARB_STATE,
  6969.                    _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
  6970.  
  6971.         /* WaDisable_RenderCache_OperationalFlush:gen4 */
  6972.         I915_WRITE(CACHE_MODE_0, _MASKED_BIT_DISABLE(RC_OP_FLUSH_ENABLE));
  6973. }
  6974.  
  6975. static void gen3_init_clock_gating(struct drm_device *dev)
  6976. {
  6977.         struct drm_i915_private *dev_priv = dev->dev_private;
  6978.         u32 dstate = I915_READ(D_STATE);
  6979.  
  6980.         dstate |= DSTATE_PLL_D3_OFF | DSTATE_GFX_CLOCK_GATING |
  6981.                 DSTATE_DOT_CLOCK_GATING;
  6982.         I915_WRITE(D_STATE, dstate);
  6983.  
  6984.         if (IS_PINEVIEW(dev))
  6985.                 I915_WRITE(ECOSKPD, _MASKED_BIT_ENABLE(ECO_GATING_CX_ONLY));
  6986.  
  6987.         /* IIR "flip pending" means done if this bit is set */
  6988.         I915_WRITE(ECOSKPD, _MASKED_BIT_DISABLE(ECO_FLIP_DONE));
  6989.  
  6990.         /* interrupts should cause a wake up from C3 */
  6991.         I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_AGPBUSY_INT_EN));
  6992.  
  6993.         /* On GEN3 we really need to make sure the ARB C3 LP bit is set */
  6994.         I915_WRITE(MI_ARB_STATE, _MASKED_BIT_ENABLE(MI_ARB_C3_LP_WRITE_ENABLE));
  6995.  
  6996.         I915_WRITE(MI_ARB_STATE,
  6997.                    _MASKED_BIT_ENABLE(MI_ARB_DISPLAY_TRICKLE_FEED_DISABLE));
  6998. }
  6999.  
  7000. static void i85x_init_clock_gating(struct drm_device *dev)
  7001. {
  7002.         struct drm_i915_private *dev_priv = dev->dev_private;
  7003.  
  7004.         I915_WRITE(RENCLK_GATE_D1, SV_CLOCK_GATE_DISABLE);
  7005.  
  7006.         /* interrupts should cause a wake up from C3 */
  7007.         I915_WRITE(MI_STATE, _MASKED_BIT_ENABLE(MI_AGPBUSY_INT_EN) |
  7008.                    _MASKED_BIT_DISABLE(MI_AGPBUSY_830_MODE));
  7009.  
  7010.         I915_WRITE(MEM_MODE,
  7011.                    _MASKED_BIT_ENABLE(MEM_DISPLAY_TRICKLE_FEED_DISABLE));
  7012. }
  7013.  
  7014. static void i830_init_clock_gating(struct drm_device *dev)
  7015. {
  7016.         struct drm_i915_private *dev_priv = dev->dev_private;
  7017.  
  7018.         I915_WRITE(DSPCLK_GATE_D, OVRUNIT_CLOCK_GATE_DISABLE);
  7019.  
  7020.         I915_WRITE(MEM_MODE,
  7021.                    _MASKED_BIT_ENABLE(MEM_DISPLAY_A_TRICKLE_FEED_DISABLE) |
  7022.                    _MASKED_BIT_ENABLE(MEM_DISPLAY_B_TRICKLE_FEED_DISABLE));
  7023. }
  7024.  
  7025. void intel_init_clock_gating(struct drm_device *dev)
  7026. {
  7027.         struct drm_i915_private *dev_priv = dev->dev_private;
  7028.  
  7029.         if (dev_priv->display.init_clock_gating)
  7030.                 dev_priv->display.init_clock_gating(dev);
  7031. }
  7032.  
  7033. void intel_suspend_hw(struct drm_device *dev)
  7034. {
  7035.         if (HAS_PCH_LPT(dev))
  7036.                 lpt_suspend_hw(dev);
  7037. }
  7038.  
  7039. /* Set up chip specific power management-related functions */
  7040. void intel_init_pm(struct drm_device *dev)
  7041. {
  7042.         struct drm_i915_private *dev_priv = dev->dev_private;
  7043.  
  7044.         intel_fbc_init(dev_priv);
  7045.  
  7046.         /* For cxsr */
  7047.         if (IS_PINEVIEW(dev))
  7048.                 i915_pineview_get_mem_freq(dev);
  7049.         else if (IS_GEN5(dev))
  7050.                 i915_ironlake_get_mem_freq(dev);
  7051.  
  7052.         /* For FIFO watermark updates */
  7053.         if (INTEL_INFO(dev)->gen >= 9) {
  7054.                 skl_setup_wm_latency(dev);
  7055.  
  7056.                 if (IS_BROXTON(dev))
  7057.                         dev_priv->display.init_clock_gating =
  7058.                                 bxt_init_clock_gating;
  7059.                 dev_priv->display.update_wm = skl_update_wm;
  7060.                 dev_priv->display.update_sprite_wm = skl_update_sprite_wm;
  7061.         } else if (HAS_PCH_SPLIT(dev)) {
  7062.                 ilk_setup_wm_latency(dev);
  7063.  
  7064.                 if ((IS_GEN5(dev) && dev_priv->wm.pri_latency[1] &&
  7065.                      dev_priv->wm.spr_latency[1] && dev_priv->wm.cur_latency[1]) ||
  7066.                     (!IS_GEN5(dev) && dev_priv->wm.pri_latency[0] &&
  7067.                      dev_priv->wm.spr_latency[0] && dev_priv->wm.cur_latency[0])) {
  7068.                         dev_priv->display.update_wm = ilk_update_wm;
  7069.                         dev_priv->display.update_sprite_wm = ilk_update_sprite_wm;
  7070.                 } else {
  7071.                         DRM_DEBUG_KMS("Failed to read display plane latency. "
  7072.                                       "Disable CxSR\n");
  7073.                 }
  7074.  
  7075.                 if (IS_GEN5(dev))
  7076.                         dev_priv->display.init_clock_gating = ironlake_init_clock_gating;
  7077.                 else if (IS_GEN6(dev))
  7078.                         dev_priv->display.init_clock_gating = gen6_init_clock_gating;
  7079.                 else if (IS_IVYBRIDGE(dev))
  7080.                         dev_priv->display.init_clock_gating = ivybridge_init_clock_gating;
  7081.                 else if (IS_HASWELL(dev))
  7082.                         dev_priv->display.init_clock_gating = haswell_init_clock_gating;
  7083.                 else if (INTEL_INFO(dev)->gen == 8)
  7084.                         dev_priv->display.init_clock_gating = broadwell_init_clock_gating;
  7085.         } else if (IS_CHERRYVIEW(dev)) {
  7086.                 vlv_setup_wm_latency(dev);
  7087.  
  7088.                 dev_priv->display.update_wm = vlv_update_wm;
  7089.                 dev_priv->display.init_clock_gating =
  7090.                         cherryview_init_clock_gating;
  7091.         } else if (IS_VALLEYVIEW(dev)) {
  7092.                 vlv_setup_wm_latency(dev);
  7093.  
  7094.                 dev_priv->display.update_wm = vlv_update_wm;
  7095.                 dev_priv->display.init_clock_gating =
  7096.                         valleyview_init_clock_gating;
  7097.         } else if (IS_PINEVIEW(dev)) {
  7098.                 if (!intel_get_cxsr_latency(IS_PINEVIEW_G(dev),
  7099.                                             dev_priv->is_ddr3,
  7100.                                             dev_priv->fsb_freq,
  7101.                                             dev_priv->mem_freq)) {
  7102.                         DRM_INFO("failed to find known CxSR latency "
  7103.                                  "(found ddr%s fsb freq %d, mem freq %d), "
  7104.                                  "disabling CxSR\n",
  7105.                                  (dev_priv->is_ddr3 == 1) ? "3" : "2",
  7106.                                  dev_priv->fsb_freq, dev_priv->mem_freq);
  7107.                         /* Disable CxSR and never update its watermark again */
  7108.                         intel_set_memory_cxsr(dev_priv, false);
  7109.                         dev_priv->display.update_wm = NULL;
  7110.                 } else
  7111.                         dev_priv->display.update_wm = pineview_update_wm;
  7112.                 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
  7113.         } else if (IS_G4X(dev)) {
  7114.                 dev_priv->display.update_wm = g4x_update_wm;
  7115.                 dev_priv->display.init_clock_gating = g4x_init_clock_gating;
  7116.         } else if (IS_GEN4(dev)) {
  7117.                 dev_priv->display.update_wm = i965_update_wm;
  7118.                 if (IS_CRESTLINE(dev))
  7119.                         dev_priv->display.init_clock_gating = crestline_init_clock_gating;
  7120.                 else if (IS_BROADWATER(dev))
  7121.                         dev_priv->display.init_clock_gating = broadwater_init_clock_gating;
  7122.         } else if (IS_GEN3(dev)) {
  7123.                 dev_priv->display.update_wm = i9xx_update_wm;
  7124.                 dev_priv->display.get_fifo_size = i9xx_get_fifo_size;
  7125.                 dev_priv->display.init_clock_gating = gen3_init_clock_gating;
  7126.         } else if (IS_GEN2(dev)) {
  7127.                 if (INTEL_INFO(dev)->num_pipes == 1) {
  7128.                         dev_priv->display.update_wm = i845_update_wm;
  7129.                         dev_priv->display.get_fifo_size = i845_get_fifo_size;
  7130.                 } else {
  7131.                         dev_priv->display.update_wm = i9xx_update_wm;
  7132.                         dev_priv->display.get_fifo_size = i830_get_fifo_size;
  7133.                 }
  7134.  
  7135.                 if (IS_I85X(dev) || IS_I865G(dev))
  7136.                         dev_priv->display.init_clock_gating = i85x_init_clock_gating;
  7137.                 else
  7138.                         dev_priv->display.init_clock_gating = i830_init_clock_gating;
  7139.         } else {
  7140.                 DRM_ERROR("unexpected fall-through in intel_init_pm\n");
  7141.         }
  7142. }
  7143.  
  7144. int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u32 mbox, u32 *val)
  7145. {
  7146.         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
  7147.  
  7148.         if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
  7149.                 DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
  7150.                 return -EAGAIN;
  7151.         }
  7152.  
  7153.         I915_WRITE(GEN6_PCODE_DATA, *val);
  7154.         I915_WRITE(GEN6_PCODE_DATA1, 0);
  7155.         I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
  7156.  
  7157.         if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
  7158.                      500)) {
  7159.                 DRM_ERROR("timeout waiting for pcode read (%d) to finish\n", mbox);
  7160.                 return -ETIMEDOUT;
  7161.         }
  7162.  
  7163.         *val = I915_READ(GEN6_PCODE_DATA);
  7164.         I915_WRITE(GEN6_PCODE_DATA, 0);
  7165.  
  7166.         return 0;
  7167. }
  7168.  
  7169. int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u32 mbox, u32 val)
  7170. {
  7171.         WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
  7172.  
  7173.         if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
  7174.                 DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");
  7175.                 return -EAGAIN;
  7176.         }
  7177.  
  7178.         I915_WRITE(GEN6_PCODE_DATA, val);
  7179.         I915_WRITE(GEN6_PCODE_MAILBOX, GEN6_PCODE_READY | mbox);
  7180.  
  7181.         if (wait_for((I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) == 0,
  7182.                      500)) {
  7183.                 DRM_ERROR("timeout waiting for pcode write (%d) to finish\n", mbox);
  7184.                 return -ETIMEDOUT;
  7185.         }
  7186.  
  7187.         I915_WRITE(GEN6_PCODE_DATA, 0);
  7188.  
  7189.         return 0;
  7190. }
  7191.  
  7192. static int vlv_gpu_freq_div(unsigned int czclk_freq)
  7193. {
  7194.         switch (czclk_freq) {
  7195.         case 200:
  7196.                 return 10;
  7197.         case 267:
  7198.                 return 12;
  7199.         case 320:
  7200.         case 333:
  7201.                 return 16;
  7202.         case 400:
  7203.                 return 20;
  7204.         default:
  7205.                 return -1;
  7206.         }
  7207. }
  7208.  
  7209. static int byt_gpu_freq(struct drm_i915_private *dev_priv, int val)
  7210. {
  7211.         int div, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000);
  7212.  
  7213.         div = vlv_gpu_freq_div(czclk_freq);
  7214.         if (div < 0)
  7215.                 return div;
  7216.  
  7217.         return DIV_ROUND_CLOSEST(czclk_freq * (val + 6 - 0xbd), div);
  7218. }
  7219.  
  7220. static int byt_freq_opcode(struct drm_i915_private *dev_priv, int val)
  7221. {
  7222.         int mul, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000);
  7223.  
  7224.         mul = vlv_gpu_freq_div(czclk_freq);
  7225.         if (mul < 0)
  7226.                 return mul;
  7227.  
  7228.         return DIV_ROUND_CLOSEST(mul * val, czclk_freq) + 0xbd - 6;
  7229. }
  7230.  
  7231. static int chv_gpu_freq(struct drm_i915_private *dev_priv, int val)
  7232. {
  7233.         int div, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000);
  7234.  
  7235.         div = vlv_gpu_freq_div(czclk_freq) / 2;
  7236.         if (div < 0)
  7237.                 return div;
  7238.  
  7239.         return DIV_ROUND_CLOSEST(czclk_freq * val, 2 * div) / 2;
  7240. }
  7241.  
  7242. static int chv_freq_opcode(struct drm_i915_private *dev_priv, int val)
  7243. {
  7244.         int mul, czclk_freq = DIV_ROUND_CLOSEST(dev_priv->czclk_freq, 1000);
  7245.  
  7246.         mul = vlv_gpu_freq_div(czclk_freq) / 2;
  7247.         if (mul < 0)
  7248.                 return mul;
  7249.  
  7250.         /* CHV needs even values */
  7251.         return DIV_ROUND_CLOSEST(val * 2 * mul, czclk_freq) * 2;
  7252. }
  7253.  
  7254. int intel_gpu_freq(struct drm_i915_private *dev_priv, int val)
  7255. {
  7256.         if (IS_GEN9(dev_priv->dev))
  7257.                 return DIV_ROUND_CLOSEST(val * GT_FREQUENCY_MULTIPLIER,
  7258.                                          GEN9_FREQ_SCALER);
  7259.         else if (IS_CHERRYVIEW(dev_priv->dev))
  7260.                 return chv_gpu_freq(dev_priv, val);
  7261.         else if (IS_VALLEYVIEW(dev_priv->dev))
  7262.                 return byt_gpu_freq(dev_priv, val);
  7263.         else
  7264.                 return val * GT_FREQUENCY_MULTIPLIER;
  7265. }
  7266.  
  7267. int intel_freq_opcode(struct drm_i915_private *dev_priv, int val)
  7268. {
  7269.         if (IS_GEN9(dev_priv->dev))
  7270.                 return DIV_ROUND_CLOSEST(val * GEN9_FREQ_SCALER,
  7271.                                          GT_FREQUENCY_MULTIPLIER);
  7272.         else if (IS_CHERRYVIEW(dev_priv->dev))
  7273.                 return chv_freq_opcode(dev_priv, val);
  7274.         else if (IS_VALLEYVIEW(dev_priv->dev))
  7275.                 return byt_freq_opcode(dev_priv, val);
  7276.         else
  7277.                 return DIV_ROUND_CLOSEST(val, GT_FREQUENCY_MULTIPLIER);
  7278. }
  7279.  
  7280. struct request_boost {
  7281.         struct work_struct work;
  7282.         struct drm_i915_gem_request *req;
  7283. };
  7284.  
  7285. static void __intel_rps_boost_work(struct work_struct *work)
  7286. {
  7287.         struct request_boost *boost = container_of(work, struct request_boost, work);
  7288.         struct drm_i915_gem_request *req = boost->req;
  7289.  
  7290.         if (!i915_gem_request_completed(req, true))
  7291.                 gen6_rps_boost(to_i915(req->ring->dev), NULL,
  7292.                                req->emitted_jiffies);
  7293.  
  7294.         i915_gem_request_unreference__unlocked(req);
  7295.         kfree(boost);
  7296. }
  7297.  
  7298. void intel_queue_rps_boost_for_request(struct drm_device *dev,
  7299.                                        struct drm_i915_gem_request *req)
  7300. {
  7301.         struct request_boost *boost;
  7302.  
  7303.         if (req == NULL || INTEL_INFO(dev)->gen < 6)
  7304.                 return;
  7305.  
  7306.         if (i915_gem_request_completed(req, true))
  7307.                 return;
  7308.  
  7309.         boost = kmalloc(sizeof(*boost), GFP_ATOMIC);
  7310.         if (boost == NULL)
  7311.                 return;
  7312.  
  7313.         i915_gem_request_reference(req);
  7314.         boost->req = req;
  7315.  
  7316.         INIT_WORK(&boost->work, __intel_rps_boost_work);
  7317.         queue_work(to_i915(dev)->wq, &boost->work);
  7318. }
  7319.  
  7320. void intel_pm_setup(struct drm_device *dev)
  7321. {
  7322.         struct drm_i915_private *dev_priv = dev->dev_private;
  7323.  
  7324.         mutex_init(&dev_priv->rps.hw_lock);
  7325.         spin_lock_init(&dev_priv->rps.client_lock);
  7326.  
  7327.         INIT_DELAYED_WORK(&dev_priv->rps.delayed_resume_work,
  7328.                           intel_gen6_powersave_work);
  7329.         INIT_LIST_HEAD(&dev_priv->rps.clients);
  7330.         INIT_LIST_HEAD(&dev_priv->rps.semaphores.link);
  7331.         INIT_LIST_HEAD(&dev_priv->rps.mmioflips.link);
  7332.  
  7333.         dev_priv->pm.suspended = false;
  7334. }
  7335.