Rev 6084 | Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5354 | serge | 1 | /* |
2 | * Copyright © 2014 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 | * Ben Widawsky |
||
25 | * Michel Thierry |
||
26 | * Thomas Daniel |
||
27 | * Oscar Mateo |
||
28 | * |
||
29 | */ |
||
30 | |||
31 | /** |
||
32 | * DOC: Logical Rings, Logical Ring Contexts and Execlists |
||
33 | * |
||
34 | * Motivation: |
||
35 | * GEN8 brings an expansion of the HW contexts: "Logical Ring Contexts". |
||
36 | * These expanded contexts enable a number of new abilities, especially |
||
37 | * "Execlists" (also implemented in this file). |
||
38 | * |
||
39 | * One of the main differences with the legacy HW contexts is that logical |
||
40 | * ring contexts incorporate many more things to the context's state, like |
||
41 | * PDPs or ringbuffer control registers: |
||
42 | * |
||
43 | * The reason why PDPs are included in the context is straightforward: as |
||
44 | * PPGTTs (per-process GTTs) are actually per-context, having the PDPs |
||
45 | * contained there mean you don't need to do a ppgtt->switch_mm yourself, |
||
46 | * instead, the GPU will do it for you on the context switch. |
||
47 | * |
||
48 | * But, what about the ringbuffer control registers (head, tail, etc..)? |
||
49 | * shouldn't we just need a set of those per engine command streamer? This is |
||
50 | * where the name "Logical Rings" starts to make sense: by virtualizing the |
||
51 | * rings, the engine cs shifts to a new "ring buffer" with every context |
||
52 | * switch. When you want to submit a workload to the GPU you: A) choose your |
||
53 | * context, B) find its appropriate virtualized ring, C) write commands to it |
||
54 | * and then, finally, D) tell the GPU to switch to that context. |
||
55 | * |
||
56 | * Instead of the legacy MI_SET_CONTEXT, the way you tell the GPU to switch |
||
57 | * to a contexts is via a context execution list, ergo "Execlists". |
||
58 | * |
||
59 | * LRC implementation: |
||
60 | * Regarding the creation of contexts, we have: |
||
61 | * |
||
62 | * - One global default context. |
||
63 | * - One local default context for each opened fd. |
||
64 | * - One local extra context for each context create ioctl call. |
||
65 | * |
||
66 | * Now that ringbuffers belong per-context (and not per-engine, like before) |
||
67 | * and that contexts are uniquely tied to a given engine (and not reusable, |
||
68 | * like before) we need: |
||
69 | * |
||
70 | * - One ringbuffer per-engine inside each context. |
||
71 | * - One backing object per-engine inside each context. |
||
72 | * |
||
73 | * The global default context starts its life with these new objects fully |
||
74 | * allocated and populated. The local default context for each opened fd is |
||
75 | * more complex, because we don't know at creation time which engine is going |
||
76 | * to use them. To handle this, we have implemented a deferred creation of LR |
||
77 | * contexts: |
||
78 | * |
||
79 | * The local context starts its life as a hollow or blank holder, that only |
||
80 | * gets populated for a given engine once we receive an execbuffer. If later |
||
81 | * on we receive another execbuffer ioctl for the same context but a different |
||
82 | * engine, we allocate/populate a new ringbuffer and context backing object and |
||
83 | * so on. |
||
84 | * |
||
85 | * Finally, regarding local contexts created using the ioctl call: as they are |
||
86 | * only allowed with the render ring, we can allocate & populate them right |
||
87 | * away (no need to defer anything, at least for now). |
||
88 | * |
||
89 | * Execlists implementation: |
||
90 | * Execlists are the new method by which, on gen8+ hardware, workloads are |
||
91 | * submitted for execution (as opposed to the legacy, ringbuffer-based, method). |
||
92 | * This method works as follows: |
||
93 | * |
||
94 | * When a request is committed, its commands (the BB start and any leading or |
||
95 | * trailing commands, like the seqno breadcrumbs) are placed in the ringbuffer |
||
96 | * for the appropriate context. The tail pointer in the hardware context is not |
||
97 | * updated at this time, but instead, kept by the driver in the ringbuffer |
||
98 | * structure. A structure representing this request is added to a request queue |
||
99 | * for the appropriate engine: this structure contains a copy of the context's |
||
100 | * tail after the request was written to the ring buffer and a pointer to the |
||
101 | * context itself. |
||
102 | * |
||
103 | * If the engine's request queue was empty before the request was added, the |
||
104 | * queue is processed immediately. Otherwise the queue will be processed during |
||
105 | * a context switch interrupt. In any case, elements on the queue will get sent |
||
106 | * (in pairs) to the GPU's ExecLists Submit Port (ELSP, for short) with a |
||
107 | * globally unique 20-bits submission ID. |
||
108 | * |
||
109 | * When execution of a request completes, the GPU updates the context status |
||
110 | * buffer with a context complete event and generates a context switch interrupt. |
||
111 | * During the interrupt handling, the driver examines the events in the buffer: |
||
112 | * for each context complete event, if the announced ID matches that on the head |
||
113 | * of the request queue, then that request is retired and removed from the queue. |
||
114 | * |
||
115 | * After processing, if any requests were retired and the queue is not empty |
||
116 | * then a new execution list can be submitted. The two requests at the front of |
||
117 | * the queue are next to be submitted but since a context may not occur twice in |
||
118 | * an execution list, if subsequent requests have the same ID as the first then |
||
119 | * the two requests must be combined. This is done simply by discarding requests |
||
120 | * at the head of the queue until either only one requests is left (in which case |
||
121 | * we use a NULL second context) or the first two requests have unique IDs. |
||
122 | * |
||
123 | * By always executing the first two requests in the queue the driver ensures |
||
124 | * that the GPU is kept as busy as possible. In the case where a single context |
||
125 | * completes but a second context is still executing, the request for this second |
||
126 | * context will be at the head of the queue when we remove the first one. This |
||
127 | * request will then be resubmitted along with a new request for a different context, |
||
128 | * which will cause the hardware to continue executing the second request and queue |
||
129 | * the new request (the GPU detects the condition of a context getting preempted |
||
130 | * with the same context and optimizes the context switch flow by not doing |
||
131 | * preemption, but just sampling the new tail pointer). |
||
132 | * |
||
133 | */ |
||
134 | |||
135 | #include |
||
136 | #include |
||
137 | #include "intel_drv.h" |
||
138 | #include "i915_drv.h" |
||
139 | |||
140 | #define GEN9_LR_CONTEXT_RENDER_SIZE (22 * PAGE_SIZE) |
||
141 | #define GEN8_LR_CONTEXT_RENDER_SIZE (20 * PAGE_SIZE) |
||
142 | #define GEN8_LR_CONTEXT_OTHER_SIZE (2 * PAGE_SIZE) |
||
143 | |||
144 | #define RING_EXECLIST_QFULL (1 << 0x2) |
||
145 | #define RING_EXECLIST1_VALID (1 << 0x3) |
||
146 | #define RING_EXECLIST0_VALID (1 << 0x4) |
||
147 | #define RING_EXECLIST_ACTIVE_STATUS (3 << 0xE) |
||
148 | #define RING_EXECLIST1_ACTIVE (1 << 0x11) |
||
149 | #define RING_EXECLIST0_ACTIVE (1 << 0x12) |
||
150 | |||
151 | #define GEN8_CTX_STATUS_IDLE_ACTIVE (1 << 0) |
||
152 | #define GEN8_CTX_STATUS_PREEMPTED (1 << 1) |
||
153 | #define GEN8_CTX_STATUS_ELEMENT_SWITCH (1 << 2) |
||
154 | #define GEN8_CTX_STATUS_ACTIVE_IDLE (1 << 3) |
||
155 | #define GEN8_CTX_STATUS_COMPLETE (1 << 4) |
||
156 | #define GEN8_CTX_STATUS_LITE_RESTORE (1 << 15) |
||
157 | |||
158 | #define CTX_LRI_HEADER_0 0x01 |
||
159 | #define CTX_CONTEXT_CONTROL 0x02 |
||
160 | #define CTX_RING_HEAD 0x04 |
||
161 | #define CTX_RING_TAIL 0x06 |
||
162 | #define CTX_RING_BUFFER_START 0x08 |
||
163 | #define CTX_RING_BUFFER_CONTROL 0x0a |
||
164 | #define CTX_BB_HEAD_U 0x0c |
||
165 | #define CTX_BB_HEAD_L 0x0e |
||
166 | #define CTX_BB_STATE 0x10 |
||
167 | #define CTX_SECOND_BB_HEAD_U 0x12 |
||
168 | #define CTX_SECOND_BB_HEAD_L 0x14 |
||
169 | #define CTX_SECOND_BB_STATE 0x16 |
||
170 | #define CTX_BB_PER_CTX_PTR 0x18 |
||
171 | #define CTX_RCS_INDIRECT_CTX 0x1a |
||
172 | #define CTX_RCS_INDIRECT_CTX_OFFSET 0x1c |
||
173 | #define CTX_LRI_HEADER_1 0x21 |
||
174 | #define CTX_CTX_TIMESTAMP 0x22 |
||
175 | #define CTX_PDP3_UDW 0x24 |
||
176 | #define CTX_PDP3_LDW 0x26 |
||
177 | #define CTX_PDP2_UDW 0x28 |
||
178 | #define CTX_PDP2_LDW 0x2a |
||
179 | #define CTX_PDP1_UDW 0x2c |
||
180 | #define CTX_PDP1_LDW 0x2e |
||
181 | #define CTX_PDP0_UDW 0x30 |
||
182 | #define CTX_PDP0_LDW 0x32 |
||
183 | #define CTX_LRI_HEADER_2 0x41 |
||
184 | #define CTX_R_PWR_CLK_STATE 0x42 |
||
185 | #define CTX_GPGPU_CSR_BASE_ADDRESS 0x44 |
||
186 | |||
187 | #define GEN8_CTX_VALID (1<<0) |
||
188 | #define GEN8_CTX_FORCE_PD_RESTORE (1<<1) |
||
189 | #define GEN8_CTX_FORCE_RESTORE (1<<2) |
||
190 | #define GEN8_CTX_L3LLC_COHERENT (1<<5) |
||
191 | #define GEN8_CTX_PRIVILEGE (1<<8) |
||
192 | enum { |
||
193 | ADVANCED_CONTEXT = 0, |
||
194 | LEGACY_CONTEXT, |
||
195 | ADVANCED_AD_CONTEXT, |
||
196 | LEGACY_64B_CONTEXT |
||
197 | }; |
||
198 | #define GEN8_CTX_MODE_SHIFT 3 |
||
199 | enum { |
||
200 | FAULT_AND_HANG = 0, |
||
201 | FAULT_AND_HALT, /* Debug only */ |
||
202 | FAULT_AND_STREAM, |
||
203 | FAULT_AND_CONTINUE /* Unsupported */ |
||
204 | }; |
||
205 | #define GEN8_CTX_ID_SHIFT 32 |
||
206 | |||
207 | static int intel_lr_context_pin(struct intel_engine_cs *ring, |
||
208 | struct intel_context *ctx); |
||
209 | |||
210 | /** |
||
211 | * intel_sanitize_enable_execlists() - sanitize i915.enable_execlists |
||
212 | * @dev: DRM device. |
||
213 | * @enable_execlists: value of i915.enable_execlists module parameter. |
||
214 | * |
||
215 | * Only certain platforms support Execlists (the prerequisites being |
||
216 | * support for Logical Ring Contexts and Aliasing PPGTT or better), |
||
217 | * and only when enabled via module parameter. |
||
218 | * |
||
219 | * Return: 1 if Execlists is supported and has to be enabled. |
||
220 | */ |
||
221 | int intel_sanitize_enable_execlists(struct drm_device *dev, int enable_execlists) |
||
222 | { |
||
223 | WARN_ON(i915.enable_ppgtt == -1); |
||
224 | |||
225 | if (INTEL_INFO(dev)->gen >= 9) |
||
226 | return 1; |
||
227 | |||
228 | if (enable_execlists == 0) |
||
229 | return 0; |
||
230 | |||
231 | if (HAS_LOGICAL_RING_CONTEXTS(dev) && USES_PPGTT(dev) && |
||
232 | i915.use_mmio_flip >= 0) |
||
233 | return 1; |
||
234 | |||
235 | return 0; |
||
236 | } |
||
237 | |||
238 | /** |
||
239 | * intel_execlists_ctx_id() - get the Execlists Context ID |
||
240 | * @ctx_obj: Logical Ring Context backing object. |
||
241 | * |
||
242 | * Do not confuse with ctx->id! Unfortunately we have a name overload |
||
243 | * here: the old context ID we pass to userspace as a handler so that |
||
244 | * they can refer to a context, and the new context ID we pass to the |
||
245 | * ELSP so that the GPU can inform us of the context status via |
||
246 | * interrupts. |
||
247 | * |
||
248 | * Return: 20-bits globally unique context ID. |
||
249 | */ |
||
250 | u32 intel_execlists_ctx_id(struct drm_i915_gem_object *ctx_obj) |
||
251 | { |
||
252 | u32 lrca = i915_gem_obj_ggtt_offset(ctx_obj); |
||
253 | |||
254 | /* LRCA is required to be 4K aligned so the more significant 20 bits |
||
255 | * are globally unique */ |
||
256 | return lrca >> 12; |
||
257 | } |
||
258 | |||
259 | static uint64_t execlists_ctx_descriptor(struct drm_i915_gem_object *ctx_obj) |
||
260 | { |
||
261 | uint64_t desc; |
||
262 | uint64_t lrca = i915_gem_obj_ggtt_offset(ctx_obj); |
||
263 | |||
264 | WARN_ON(lrca & 0xFFFFFFFF00000FFFULL); |
||
265 | |||
266 | desc = GEN8_CTX_VALID; |
||
267 | desc |= LEGACY_CONTEXT << GEN8_CTX_MODE_SHIFT; |
||
268 | desc |= GEN8_CTX_L3LLC_COHERENT; |
||
269 | desc |= GEN8_CTX_PRIVILEGE; |
||
270 | desc |= lrca; |
||
271 | desc |= (u64)intel_execlists_ctx_id(ctx_obj) << GEN8_CTX_ID_SHIFT; |
||
272 | |||
273 | /* TODO: WaDisableLiteRestore when we start using semaphore |
||
274 | * signalling between Command Streamers */ |
||
275 | /* desc |= GEN8_CTX_FORCE_RESTORE; */ |
||
276 | |||
277 | return desc; |
||
278 | } |
||
279 | |||
280 | static void execlists_elsp_write(struct intel_engine_cs *ring, |
||
281 | struct drm_i915_gem_object *ctx_obj0, |
||
282 | struct drm_i915_gem_object *ctx_obj1) |
||
283 | { |
||
284 | struct drm_device *dev = ring->dev; |
||
285 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
286 | uint64_t temp = 0; |
||
287 | uint32_t desc[4]; |
||
288 | unsigned long flags; |
||
289 | |||
290 | /* XXX: You must always write both descriptors in the order below. */ |
||
291 | if (ctx_obj1) |
||
292 | temp = execlists_ctx_descriptor(ctx_obj1); |
||
293 | else |
||
294 | temp = 0; |
||
295 | desc[1] = (u32)(temp >> 32); |
||
296 | desc[0] = (u32)temp; |
||
297 | |||
298 | temp = execlists_ctx_descriptor(ctx_obj0); |
||
299 | desc[3] = (u32)(temp >> 32); |
||
300 | desc[2] = (u32)temp; |
||
301 | |||
302 | /* Set Force Wakeup bit to prevent GT from entering C6 while ELSP writes |
||
303 | * are in progress. |
||
304 | * |
||
305 | * The other problem is that we can't just call gen6_gt_force_wake_get() |
||
306 | * because that function calls intel_runtime_pm_get(), which might sleep. |
||
307 | * Instead, we do the runtime_pm_get/put when creating/destroying requests. |
||
308 | */ |
||
309 | spin_lock_irqsave(&dev_priv->uncore.lock, flags); |
||
310 | if (IS_CHERRYVIEW(dev) || INTEL_INFO(dev)->gen >= 9) { |
||
311 | if (dev_priv->uncore.fw_rendercount++ == 0) |
||
312 | dev_priv->uncore.funcs.force_wake_get(dev_priv, |
||
313 | FORCEWAKE_RENDER); |
||
314 | if (dev_priv->uncore.fw_mediacount++ == 0) |
||
315 | dev_priv->uncore.funcs.force_wake_get(dev_priv, |
||
316 | FORCEWAKE_MEDIA); |
||
317 | if (INTEL_INFO(dev)->gen >= 9) { |
||
318 | if (dev_priv->uncore.fw_blittercount++ == 0) |
||
319 | dev_priv->uncore.funcs.force_wake_get(dev_priv, |
||
320 | FORCEWAKE_BLITTER); |
||
321 | } |
||
322 | } else { |
||
323 | if (dev_priv->uncore.forcewake_count++ == 0) |
||
324 | dev_priv->uncore.funcs.force_wake_get(dev_priv, |
||
325 | FORCEWAKE_ALL); |
||
326 | } |
||
327 | spin_unlock_irqrestore(&dev_priv->uncore.lock, flags); |
||
328 | |||
329 | I915_WRITE(RING_ELSP(ring), desc[1]); |
||
330 | I915_WRITE(RING_ELSP(ring), desc[0]); |
||
331 | I915_WRITE(RING_ELSP(ring), desc[3]); |
||
332 | /* The context is automatically loaded after the following */ |
||
333 | I915_WRITE(RING_ELSP(ring), desc[2]); |
||
334 | |||
335 | /* ELSP is a wo register, so use another nearby reg for posting instead */ |
||
336 | POSTING_READ(RING_EXECLIST_STATUS(ring)); |
||
337 | |||
338 | /* Release Force Wakeup (see the big comment above). */ |
||
339 | spin_lock_irqsave(&dev_priv->uncore.lock, flags); |
||
340 | if (IS_CHERRYVIEW(dev) || INTEL_INFO(dev)->gen >= 9) { |
||
341 | if (--dev_priv->uncore.fw_rendercount == 0) |
||
342 | dev_priv->uncore.funcs.force_wake_put(dev_priv, |
||
343 | FORCEWAKE_RENDER); |
||
344 | if (--dev_priv->uncore.fw_mediacount == 0) |
||
345 | dev_priv->uncore.funcs.force_wake_put(dev_priv, |
||
346 | FORCEWAKE_MEDIA); |
||
347 | if (INTEL_INFO(dev)->gen >= 9) { |
||
348 | if (--dev_priv->uncore.fw_blittercount == 0) |
||
349 | dev_priv->uncore.funcs.force_wake_put(dev_priv, |
||
350 | FORCEWAKE_BLITTER); |
||
351 | } |
||
352 | } else { |
||
353 | if (--dev_priv->uncore.forcewake_count == 0) |
||
354 | dev_priv->uncore.funcs.force_wake_put(dev_priv, |
||
355 | FORCEWAKE_ALL); |
||
356 | } |
||
357 | |||
358 | spin_unlock_irqrestore(&dev_priv->uncore.lock, flags); |
||
359 | } |
||
360 | |||
361 | static int execlists_update_context(struct drm_i915_gem_object *ctx_obj, |
||
362 | struct drm_i915_gem_object *ring_obj, |
||
363 | u32 tail) |
||
364 | { |
||
365 | struct page *page; |
||
366 | uint32_t *reg_state; |
||
367 | |||
368 | page = i915_gem_object_get_page(ctx_obj, 1); |
||
369 | reg_state = kmap_atomic(page); |
||
370 | |||
371 | reg_state[CTX_RING_TAIL+1] = tail; |
||
372 | reg_state[CTX_RING_BUFFER_START+1] = i915_gem_obj_ggtt_offset(ring_obj); |
||
373 | |||
374 | kunmap_atomic(reg_state); |
||
375 | |||
376 | return 0; |
||
377 | } |
||
378 | |||
379 | static void execlists_submit_contexts(struct intel_engine_cs *ring, |
||
380 | struct intel_context *to0, u32 tail0, |
||
381 | struct intel_context *to1, u32 tail1) |
||
382 | { |
||
383 | struct drm_i915_gem_object *ctx_obj0 = to0->engine[ring->id].state; |
||
384 | struct intel_ringbuffer *ringbuf0 = to0->engine[ring->id].ringbuf; |
||
385 | struct drm_i915_gem_object *ctx_obj1 = NULL; |
||
386 | struct intel_ringbuffer *ringbuf1 = NULL; |
||
387 | |||
388 | BUG_ON(!ctx_obj0); |
||
389 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj0)); |
||
390 | WARN_ON(!i915_gem_obj_is_pinned(ringbuf0->obj)); |
||
391 | |||
392 | execlists_update_context(ctx_obj0, ringbuf0->obj, tail0); |
||
393 | |||
394 | if (to1) { |
||
395 | ringbuf1 = to1->engine[ring->id].ringbuf; |
||
396 | ctx_obj1 = to1->engine[ring->id].state; |
||
397 | BUG_ON(!ctx_obj1); |
||
398 | WARN_ON(!i915_gem_obj_is_pinned(ctx_obj1)); |
||
399 | WARN_ON(!i915_gem_obj_is_pinned(ringbuf1->obj)); |
||
400 | |||
401 | execlists_update_context(ctx_obj1, ringbuf1->obj, tail1); |
||
402 | } |
||
403 | |||
404 | execlists_elsp_write(ring, ctx_obj0, ctx_obj1); |
||
405 | } |
||
406 | |||
407 | static void execlists_context_unqueue(struct intel_engine_cs *ring) |
||
408 | { |
||
409 | struct intel_ctx_submit_request *req0 = NULL, *req1 = NULL; |
||
410 | struct intel_ctx_submit_request *cursor = NULL, *tmp = NULL; |
||
411 | |||
412 | assert_spin_locked(&ring->execlist_lock); |
||
413 | |||
414 | if (list_empty(&ring->execlist_queue)) |
||
415 | return; |
||
416 | |||
417 | /* Try to read in pairs */ |
||
418 | list_for_each_entry_safe(cursor, tmp, &ring->execlist_queue, |
||
419 | execlist_link) { |
||
420 | if (!req0) { |
||
421 | req0 = cursor; |
||
422 | } else if (req0->ctx == cursor->ctx) { |
||
423 | /* Same ctx: ignore first request, as second request |
||
424 | * will update tail past first request's workload */ |
||
425 | cursor->elsp_submitted = req0->elsp_submitted; |
||
426 | list_del(&req0->execlist_link); |
||
427 | list_add_tail(&req0->execlist_link, |
||
428 | &ring->execlist_retired_req_list); |
||
429 | req0 = cursor; |
||
430 | } else { |
||
431 | req1 = cursor; |
||
432 | break; |
||
433 | } |
||
434 | } |
||
435 | |||
436 | WARN_ON(req1 && req1->elsp_submitted); |
||
437 | |||
438 | execlists_submit_contexts(ring, req0->ctx, req0->tail, |
||
439 | req1 ? req1->ctx : NULL, |
||
440 | req1 ? req1->tail : 0); |
||
441 | |||
442 | req0->elsp_submitted++; |
||
443 | if (req1) |
||
444 | req1->elsp_submitted++; |
||
445 | } |
||
446 | |||
447 | static bool execlists_check_remove_request(struct intel_engine_cs *ring, |
||
448 | u32 request_id) |
||
449 | { |
||
450 | struct intel_ctx_submit_request *head_req; |
||
451 | |||
452 | assert_spin_locked(&ring->execlist_lock); |
||
453 | |||
454 | head_req = list_first_entry_or_null(&ring->execlist_queue, |
||
455 | struct intel_ctx_submit_request, |
||
456 | execlist_link); |
||
457 | |||
458 | if (head_req != NULL) { |
||
459 | struct drm_i915_gem_object *ctx_obj = |
||
460 | head_req->ctx->engine[ring->id].state; |
||
461 | if (intel_execlists_ctx_id(ctx_obj) == request_id) { |
||
462 | WARN(head_req->elsp_submitted == 0, |
||
463 | "Never submitted head request\n"); |
||
464 | |||
465 | if (--head_req->elsp_submitted <= 0) { |
||
466 | list_del(&head_req->execlist_link); |
||
467 | list_add_tail(&head_req->execlist_link, |
||
468 | &ring->execlist_retired_req_list); |
||
469 | return true; |
||
470 | } |
||
471 | } |
||
472 | } |
||
473 | |||
474 | return false; |
||
475 | } |
||
476 | |||
477 | /** |
||
478 | * intel_execlists_handle_ctx_events() - handle Context Switch interrupts |
||
479 | * @ring: Engine Command Streamer to handle. |
||
480 | * |
||
481 | * Check the unread Context Status Buffers and manage the submission of new |
||
482 | * contexts to the ELSP accordingly. |
||
483 | */ |
||
484 | void intel_execlists_handle_ctx_events(struct intel_engine_cs *ring) |
||
485 | { |
||
486 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
||
487 | u32 status_pointer; |
||
488 | u8 read_pointer; |
||
489 | u8 write_pointer; |
||
490 | u32 status; |
||
491 | u32 status_id; |
||
492 | u32 submit_contexts = 0; |
||
493 | |||
494 | status_pointer = I915_READ(RING_CONTEXT_STATUS_PTR(ring)); |
||
495 | |||
496 | read_pointer = ring->next_context_status_buffer; |
||
497 | write_pointer = status_pointer & 0x07; |
||
498 | if (read_pointer > write_pointer) |
||
499 | write_pointer += 6; |
||
500 | |||
501 | spin_lock(&ring->execlist_lock); |
||
502 | |||
503 | while (read_pointer < write_pointer) { |
||
504 | read_pointer++; |
||
505 | status = I915_READ(RING_CONTEXT_STATUS_BUF(ring) + |
||
506 | (read_pointer % 6) * 8); |
||
507 | status_id = I915_READ(RING_CONTEXT_STATUS_BUF(ring) + |
||
508 | (read_pointer % 6) * 8 + 4); |
||
509 | |||
510 | if (status & GEN8_CTX_STATUS_PREEMPTED) { |
||
511 | if (status & GEN8_CTX_STATUS_LITE_RESTORE) { |
||
512 | if (execlists_check_remove_request(ring, status_id)) |
||
513 | WARN(1, "Lite Restored request removed from queue\n"); |
||
514 | } else |
||
515 | WARN(1, "Preemption without Lite Restore\n"); |
||
516 | } |
||
517 | |||
518 | if ((status & GEN8_CTX_STATUS_ACTIVE_IDLE) || |
||
519 | (status & GEN8_CTX_STATUS_ELEMENT_SWITCH)) { |
||
520 | if (execlists_check_remove_request(ring, status_id)) |
||
521 | submit_contexts++; |
||
522 | } |
||
523 | } |
||
524 | |||
525 | if (submit_contexts != 0) |
||
526 | execlists_context_unqueue(ring); |
||
527 | |||
528 | spin_unlock(&ring->execlist_lock); |
||
529 | |||
530 | WARN(submit_contexts > 2, "More than two context complete events?\n"); |
||
531 | ring->next_context_status_buffer = write_pointer % 6; |
||
532 | |||
533 | I915_WRITE(RING_CONTEXT_STATUS_PTR(ring), |
||
534 | ((u32)ring->next_context_status_buffer & 0x07) << 8); |
||
535 | } |
||
536 | |||
537 | static int execlists_context_queue(struct intel_engine_cs *ring, |
||
538 | struct intel_context *to, |
||
539 | u32 tail) |
||
540 | { |
||
541 | struct intel_ctx_submit_request *req = NULL, *cursor; |
||
542 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
||
543 | unsigned long flags; |
||
544 | int num_elements = 0; |
||
545 | |||
546 | req = kzalloc(sizeof(*req), GFP_KERNEL); |
||
547 | if (req == NULL) |
||
548 | return -ENOMEM; |
||
549 | req->ctx = to; |
||
550 | i915_gem_context_reference(req->ctx); |
||
551 | |||
552 | if (to != ring->default_context) |
||
553 | intel_lr_context_pin(ring, to); |
||
554 | |||
555 | req->ring = ring; |
||
556 | req->tail = tail; |
||
557 | |||
558 | intel_runtime_pm_get(dev_priv); |
||
559 | |||
560 | spin_lock_irqsave(&ring->execlist_lock, flags); |
||
561 | |||
562 | list_for_each_entry(cursor, &ring->execlist_queue, execlist_link) |
||
563 | if (++num_elements > 2) |
||
564 | break; |
||
565 | |||
566 | if (num_elements > 2) { |
||
567 | struct intel_ctx_submit_request *tail_req; |
||
568 | |||
569 | tail_req = list_last_entry(&ring->execlist_queue, |
||
570 | struct intel_ctx_submit_request, |
||
571 | execlist_link); |
||
572 | |||
573 | if (to == tail_req->ctx) { |
||
574 | WARN(tail_req->elsp_submitted != 0, |
||
575 | "More than 2 already-submitted reqs queued\n"); |
||
576 | list_del(&tail_req->execlist_link); |
||
577 | list_add_tail(&tail_req->execlist_link, |
||
578 | &ring->execlist_retired_req_list); |
||
579 | } |
||
580 | } |
||
581 | |||
582 | list_add_tail(&req->execlist_link, &ring->execlist_queue); |
||
583 | if (num_elements == 0) |
||
584 | execlists_context_unqueue(ring); |
||
585 | |||
586 | spin_unlock_irqrestore(&ring->execlist_lock, flags); |
||
587 | |||
588 | return 0; |
||
589 | } |
||
590 | |||
591 | static int logical_ring_invalidate_all_caches(struct intel_ringbuffer *ringbuf) |
||
592 | { |
||
593 | struct intel_engine_cs *ring = ringbuf->ring; |
||
594 | uint32_t flush_domains; |
||
595 | int ret; |
||
596 | |||
597 | flush_domains = 0; |
||
598 | if (ring->gpu_caches_dirty) |
||
599 | flush_domains = I915_GEM_GPU_DOMAINS; |
||
600 | |||
601 | ret = ring->emit_flush(ringbuf, I915_GEM_GPU_DOMAINS, flush_domains); |
||
602 | if (ret) |
||
603 | return ret; |
||
604 | |||
605 | ring->gpu_caches_dirty = false; |
||
606 | return 0; |
||
607 | } |
||
608 | |||
609 | static int execlists_move_to_gpu(struct intel_ringbuffer *ringbuf, |
||
610 | struct list_head *vmas) |
||
611 | { |
||
612 | struct intel_engine_cs *ring = ringbuf->ring; |
||
613 | struct i915_vma *vma; |
||
614 | uint32_t flush_domains = 0; |
||
615 | bool flush_chipset = false; |
||
616 | int ret; |
||
617 | |||
618 | list_for_each_entry(vma, vmas, exec_list) { |
||
619 | struct drm_i915_gem_object *obj = vma->obj; |
||
620 | |||
621 | ret = i915_gem_object_sync(obj, ring); |
||
622 | if (ret) |
||
623 | return ret; |
||
624 | |||
625 | if (obj->base.write_domain & I915_GEM_DOMAIN_CPU) |
||
626 | flush_chipset |= i915_gem_clflush_object(obj, false); |
||
627 | |||
628 | flush_domains |= obj->base.write_domain; |
||
629 | } |
||
630 | |||
631 | if (flush_domains & I915_GEM_DOMAIN_GTT) |
||
632 | wmb(); |
||
633 | |||
634 | /* Unconditionally invalidate gpu caches and ensure that we do flush |
||
635 | * any residual writes from the previous batch. |
||
636 | */ |
||
637 | return logical_ring_invalidate_all_caches(ringbuf); |
||
638 | } |
||
639 | |||
640 | /** |
||
641 | * execlists_submission() - submit a batchbuffer for execution, Execlists style |
||
642 | * @dev: DRM device. |
||
643 | * @file: DRM file. |
||
644 | * @ring: Engine Command Streamer to submit to. |
||
645 | * @ctx: Context to employ for this submission. |
||
646 | * @args: execbuffer call arguments. |
||
647 | * @vmas: list of vmas. |
||
648 | * @batch_obj: the batchbuffer to submit. |
||
649 | * @exec_start: batchbuffer start virtual address pointer. |
||
650 | * @flags: translated execbuffer call flags. |
||
651 | * |
||
652 | * This is the evil twin version of i915_gem_ringbuffer_submission. It abstracts |
||
653 | * away the submission details of the execbuffer ioctl call. |
||
654 | * |
||
655 | * Return: non-zero if the submission fails. |
||
656 | */ |
||
657 | int intel_execlists_submission(struct drm_device *dev, struct drm_file *file, |
||
658 | struct intel_engine_cs *ring, |
||
659 | struct intel_context *ctx, |
||
660 | struct drm_i915_gem_execbuffer2 *args, |
||
661 | struct list_head *vmas, |
||
662 | struct drm_i915_gem_object *batch_obj, |
||
663 | u64 exec_start, u32 flags) |
||
664 | { |
||
665 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
666 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
||
667 | int instp_mode; |
||
668 | u32 instp_mask; |
||
669 | int ret; |
||
670 | |||
671 | instp_mode = args->flags & I915_EXEC_CONSTANTS_MASK; |
||
672 | instp_mask = I915_EXEC_CONSTANTS_MASK; |
||
673 | switch (instp_mode) { |
||
674 | case I915_EXEC_CONSTANTS_REL_GENERAL: |
||
675 | case I915_EXEC_CONSTANTS_ABSOLUTE: |
||
676 | case I915_EXEC_CONSTANTS_REL_SURFACE: |
||
677 | if (instp_mode != 0 && ring != &dev_priv->ring[RCS]) { |
||
678 | DRM_DEBUG("non-0 rel constants mode on non-RCS\n"); |
||
679 | return -EINVAL; |
||
680 | } |
||
681 | |||
682 | if (instp_mode != dev_priv->relative_constants_mode) { |
||
683 | if (instp_mode == I915_EXEC_CONSTANTS_REL_SURFACE) { |
||
684 | DRM_DEBUG("rel surface constants mode invalid on gen5+\n"); |
||
685 | return -EINVAL; |
||
686 | } |
||
687 | |||
688 | /* The HW changed the meaning on this bit on gen6 */ |
||
689 | instp_mask &= ~I915_EXEC_CONSTANTS_REL_SURFACE; |
||
690 | } |
||
691 | break; |
||
692 | default: |
||
693 | DRM_DEBUG("execbuf with unknown constants: %d\n", instp_mode); |
||
694 | return -EINVAL; |
||
695 | } |
||
696 | |||
697 | if (args->num_cliprects != 0) { |
||
698 | DRM_DEBUG("clip rectangles are only valid on pre-gen5\n"); |
||
699 | return -EINVAL; |
||
700 | } else { |
||
701 | if (args->DR4 == 0xffffffff) { |
||
702 | DRM_DEBUG("UXA submitting garbage DR4, fixing up\n"); |
||
703 | args->DR4 = 0; |
||
704 | } |
||
705 | |||
706 | if (args->DR1 || args->DR4 || args->cliprects_ptr) { |
||
707 | DRM_DEBUG("0 cliprects but dirt in cliprects fields\n"); |
||
708 | return -EINVAL; |
||
709 | } |
||
710 | } |
||
711 | |||
712 | if (args->flags & I915_EXEC_GEN7_SOL_RESET) { |
||
713 | DRM_DEBUG("sol reset is gen7 only\n"); |
||
714 | return -EINVAL; |
||
715 | } |
||
716 | |||
717 | ret = execlists_move_to_gpu(ringbuf, vmas); |
||
718 | if (ret) |
||
719 | return ret; |
||
720 | |||
721 | if (ring == &dev_priv->ring[RCS] && |
||
722 | instp_mode != dev_priv->relative_constants_mode) { |
||
723 | ret = intel_logical_ring_begin(ringbuf, 4); |
||
724 | if (ret) |
||
725 | return ret; |
||
726 | |||
727 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
||
728 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(1)); |
||
729 | intel_logical_ring_emit(ringbuf, INSTPM); |
||
730 | intel_logical_ring_emit(ringbuf, instp_mask << 16 | instp_mode); |
||
731 | intel_logical_ring_advance(ringbuf); |
||
732 | |||
733 | dev_priv->relative_constants_mode = instp_mode; |
||
734 | } |
||
735 | |||
736 | ret = ring->emit_bb_start(ringbuf, exec_start, flags); |
||
737 | if (ret) |
||
738 | return ret; |
||
739 | |||
740 | i915_gem_execbuffer_move_to_active(vmas, ring); |
||
741 | i915_gem_execbuffer_retire_commands(dev, file, ring, batch_obj); |
||
742 | |||
743 | return 0; |
||
744 | } |
||
745 | |||
746 | void intel_execlists_retire_requests(struct intel_engine_cs *ring) |
||
747 | { |
||
748 | struct intel_ctx_submit_request *req, *tmp; |
||
749 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
||
750 | unsigned long flags; |
||
751 | struct list_head retired_list; |
||
752 | |||
753 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
||
754 | if (list_empty(&ring->execlist_retired_req_list)) |
||
755 | return; |
||
756 | |||
757 | INIT_LIST_HEAD(&retired_list); |
||
758 | spin_lock_irqsave(&ring->execlist_lock, flags); |
||
759 | list_replace_init(&ring->execlist_retired_req_list, &retired_list); |
||
760 | spin_unlock_irqrestore(&ring->execlist_lock, flags); |
||
761 | |||
762 | list_for_each_entry_safe(req, tmp, &retired_list, execlist_link) { |
||
763 | struct intel_context *ctx = req->ctx; |
||
764 | struct drm_i915_gem_object *ctx_obj = |
||
765 | ctx->engine[ring->id].state; |
||
766 | |||
767 | if (ctx_obj && (ctx != ring->default_context)) |
||
768 | intel_lr_context_unpin(ring, ctx); |
||
769 | intel_runtime_pm_put(dev_priv); |
||
770 | i915_gem_context_unreference(req->ctx); |
||
771 | list_del(&req->execlist_link); |
||
772 | kfree(req); |
||
773 | } |
||
774 | } |
||
775 | |||
776 | void intel_logical_ring_stop(struct intel_engine_cs *ring) |
||
777 | { |
||
778 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
||
779 | int ret; |
||
780 | |||
781 | if (!intel_ring_initialized(ring)) |
||
782 | return; |
||
783 | |||
784 | ret = intel_ring_idle(ring); |
||
785 | if (ret && !i915_reset_in_progress(&to_i915(ring->dev)->gpu_error)) |
||
786 | DRM_ERROR("failed to quiesce %s whilst cleaning up: %d\n", |
||
787 | ring->name, ret); |
||
788 | |||
789 | /* TODO: Is this correct with Execlists enabled? */ |
||
790 | I915_WRITE_MODE(ring, _MASKED_BIT_ENABLE(STOP_RING)); |
||
791 | if (wait_for_atomic((I915_READ_MODE(ring) & MODE_IDLE) != 0, 1000)) { |
||
792 | DRM_ERROR("%s :timed out trying to stop ring\n", ring->name); |
||
793 | return; |
||
794 | } |
||
795 | I915_WRITE_MODE(ring, _MASKED_BIT_DISABLE(STOP_RING)); |
||
796 | } |
||
797 | |||
798 | int logical_ring_flush_all_caches(struct intel_ringbuffer *ringbuf) |
||
799 | { |
||
800 | struct intel_engine_cs *ring = ringbuf->ring; |
||
801 | int ret; |
||
802 | |||
803 | if (!ring->gpu_caches_dirty) |
||
804 | return 0; |
||
805 | |||
806 | ret = ring->emit_flush(ringbuf, 0, I915_GEM_GPU_DOMAINS); |
||
807 | if (ret) |
||
808 | return ret; |
||
809 | |||
810 | ring->gpu_caches_dirty = false; |
||
811 | return 0; |
||
812 | } |
||
813 | |||
814 | /** |
||
815 | * intel_logical_ring_advance_and_submit() - advance the tail and submit the workload |
||
816 | * @ringbuf: Logical Ringbuffer to advance. |
||
817 | * |
||
818 | * The tail is updated in our logical ringbuffer struct, not in the actual context. What |
||
819 | * really happens during submission is that the context and current tail will be placed |
||
820 | * on a queue waiting for the ELSP to be ready to accept a new context submission. At that |
||
821 | * point, the tail *inside* the context is updated and the ELSP written to. |
||
822 | */ |
||
823 | void intel_logical_ring_advance_and_submit(struct intel_ringbuffer *ringbuf) |
||
824 | { |
||
825 | struct intel_engine_cs *ring = ringbuf->ring; |
||
826 | struct intel_context *ctx = ringbuf->FIXME_lrc_ctx; |
||
827 | |||
828 | intel_logical_ring_advance(ringbuf); |
||
829 | |||
830 | if (intel_ring_stopped(ring)) |
||
831 | return; |
||
832 | |||
833 | execlists_context_queue(ring, ctx, ringbuf->tail); |
||
834 | } |
||
835 | |||
836 | static int intel_lr_context_pin(struct intel_engine_cs *ring, |
||
837 | struct intel_context *ctx) |
||
838 | { |
||
839 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; |
||
840 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
||
841 | int ret = 0; |
||
842 | |||
843 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
||
844 | if (ctx->engine[ring->id].unpin_count++ == 0) { |
||
845 | ret = i915_gem_obj_ggtt_pin(ctx_obj, |
||
846 | GEN8_LR_CONTEXT_ALIGN, 0); |
||
847 | if (ret) |
||
848 | goto reset_unpin_count; |
||
849 | |||
850 | ret = intel_pin_and_map_ringbuffer_obj(ring->dev, ringbuf); |
||
851 | if (ret) |
||
852 | goto unpin_ctx_obj; |
||
853 | } |
||
854 | |||
855 | return ret; |
||
856 | |||
857 | unpin_ctx_obj: |
||
858 | i915_gem_object_ggtt_unpin(ctx_obj); |
||
859 | reset_unpin_count: |
||
860 | ctx->engine[ring->id].unpin_count = 0; |
||
861 | |||
862 | return ret; |
||
863 | } |
||
864 | |||
865 | void intel_lr_context_unpin(struct intel_engine_cs *ring, |
||
866 | struct intel_context *ctx) |
||
867 | { |
||
868 | struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; |
||
869 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
||
870 | |||
871 | if (ctx_obj) { |
||
872 | WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); |
||
873 | if (--ctx->engine[ring->id].unpin_count == 0) { |
||
874 | intel_unpin_ringbuffer_obj(ringbuf); |
||
875 | i915_gem_object_ggtt_unpin(ctx_obj); |
||
876 | } |
||
877 | } |
||
878 | } |
||
879 | |||
880 | static int logical_ring_alloc_seqno(struct intel_engine_cs *ring, |
||
881 | struct intel_context *ctx) |
||
882 | { |
||
883 | int ret; |
||
884 | |||
885 | if (ring->outstanding_lazy_seqno) |
||
886 | return 0; |
||
887 | |||
888 | if (ring->preallocated_lazy_request == NULL) { |
||
889 | struct drm_i915_gem_request *request; |
||
890 | |||
891 | request = kmalloc(sizeof(*request), GFP_KERNEL); |
||
892 | if (request == NULL) |
||
893 | return -ENOMEM; |
||
894 | |||
895 | if (ctx != ring->default_context) { |
||
896 | ret = intel_lr_context_pin(ring, ctx); |
||
897 | if (ret) { |
||
898 | kfree(request); |
||
899 | return ret; |
||
900 | } |
||
901 | } |
||
902 | |||
903 | /* Hold a reference to the context this request belongs to |
||
904 | * (we will need it when the time comes to emit/retire the |
||
905 | * request). |
||
906 | */ |
||
907 | request->ctx = ctx; |
||
908 | i915_gem_context_reference(request->ctx); |
||
909 | |||
910 | ring->preallocated_lazy_request = request; |
||
911 | } |
||
912 | |||
913 | return i915_gem_get_seqno(ring->dev, &ring->outstanding_lazy_seqno); |
||
914 | } |
||
915 | |||
916 | static int logical_ring_wait_request(struct intel_ringbuffer *ringbuf, |
||
917 | int bytes) |
||
918 | { |
||
919 | struct intel_engine_cs *ring = ringbuf->ring; |
||
920 | struct drm_i915_gem_request *request; |
||
921 | u32 seqno = 0; |
||
922 | int ret; |
||
923 | |||
924 | if (ringbuf->last_retired_head != -1) { |
||
925 | ringbuf->head = ringbuf->last_retired_head; |
||
926 | ringbuf->last_retired_head = -1; |
||
927 | |||
928 | ringbuf->space = intel_ring_space(ringbuf); |
||
929 | if (ringbuf->space >= bytes) |
||
930 | return 0; |
||
931 | } |
||
932 | |||
933 | list_for_each_entry(request, &ring->request_list, list) { |
||
934 | if (__intel_ring_space(request->tail, ringbuf->tail, |
||
935 | ringbuf->size) >= bytes) { |
||
936 | seqno = request->seqno; |
||
937 | break; |
||
938 | } |
||
939 | } |
||
940 | |||
941 | if (seqno == 0) |
||
942 | return -ENOSPC; |
||
943 | |||
944 | ret = i915_wait_seqno(ring, seqno); |
||
945 | if (ret) |
||
946 | return ret; |
||
947 | |||
948 | i915_gem_retire_requests_ring(ring); |
||
949 | ringbuf->head = ringbuf->last_retired_head; |
||
950 | ringbuf->last_retired_head = -1; |
||
951 | |||
952 | ringbuf->space = intel_ring_space(ringbuf); |
||
953 | return 0; |
||
954 | } |
||
955 | |||
956 | static int logical_ring_wait_for_space(struct intel_ringbuffer *ringbuf, |
||
957 | int bytes) |
||
958 | { |
||
959 | struct intel_engine_cs *ring = ringbuf->ring; |
||
960 | struct drm_device *dev = ring->dev; |
||
961 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
962 | unsigned long end; |
||
963 | int ret; |
||
964 | |||
965 | ret = logical_ring_wait_request(ringbuf, bytes); |
||
966 | if (ret != -ENOSPC) |
||
967 | return ret; |
||
968 | |||
969 | /* Force the context submission in case we have been skipping it */ |
||
970 | intel_logical_ring_advance_and_submit(ringbuf); |
||
971 | |||
972 | /* With GEM the hangcheck timer should kick us out of the loop, |
||
973 | * leaving it early runs the risk of corrupting GEM state (due |
||
974 | * to running on almost untested codepaths). But on resume |
||
975 | * timers don't work yet, so prevent a complete hang in that |
||
976 | * case by choosing an insanely large timeout. */ |
||
977 | end = jiffies + 60 * HZ; |
||
978 | |||
979 | do { |
||
980 | ringbuf->head = I915_READ_HEAD(ring); |
||
981 | ringbuf->space = intel_ring_space(ringbuf); |
||
982 | if (ringbuf->space >= bytes) { |
||
983 | ret = 0; |
||
984 | break; |
||
985 | } |
||
986 | |||
987 | msleep(1); |
||
988 | |||
989 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, |
||
990 | dev_priv->mm.interruptible); |
||
991 | if (ret) |
||
992 | break; |
||
993 | |||
994 | if (time_after(jiffies, end)) { |
||
995 | ret = -EBUSY; |
||
996 | break; |
||
997 | } |
||
998 | } while (1); |
||
999 | |||
1000 | return ret; |
||
1001 | } |
||
1002 | |||
1003 | static int logical_ring_wrap_buffer(struct intel_ringbuffer *ringbuf) |
||
1004 | { |
||
1005 | uint32_t __iomem *virt; |
||
1006 | int rem = ringbuf->size - ringbuf->tail; |
||
1007 | |||
1008 | if (ringbuf->space < rem) { |
||
1009 | int ret = logical_ring_wait_for_space(ringbuf, rem); |
||
1010 | |||
1011 | if (ret) |
||
1012 | return ret; |
||
1013 | } |
||
1014 | |||
1015 | virt = ringbuf->virtual_start + ringbuf->tail; |
||
1016 | rem /= 4; |
||
1017 | while (rem--) |
||
1018 | iowrite32(MI_NOOP, virt++); |
||
1019 | |||
1020 | ringbuf->tail = 0; |
||
1021 | ringbuf->space = intel_ring_space(ringbuf); |
||
1022 | |||
1023 | return 0; |
||
1024 | } |
||
1025 | |||
1026 | static int logical_ring_prepare(struct intel_ringbuffer *ringbuf, int bytes) |
||
1027 | { |
||
1028 | int ret; |
||
1029 | |||
1030 | if (unlikely(ringbuf->tail + bytes > ringbuf->effective_size)) { |
||
1031 | ret = logical_ring_wrap_buffer(ringbuf); |
||
1032 | if (unlikely(ret)) |
||
1033 | return ret; |
||
1034 | } |
||
1035 | |||
1036 | if (unlikely(ringbuf->space < bytes)) { |
||
1037 | ret = logical_ring_wait_for_space(ringbuf, bytes); |
||
1038 | if (unlikely(ret)) |
||
1039 | return ret; |
||
1040 | } |
||
1041 | |||
1042 | return 0; |
||
1043 | } |
||
1044 | |||
1045 | /** |
||
1046 | * intel_logical_ring_begin() - prepare the logical ringbuffer to accept some commands |
||
1047 | * |
||
1048 | * @ringbuf: Logical ringbuffer. |
||
1049 | * @num_dwords: number of DWORDs that we plan to write to the ringbuffer. |
||
1050 | * |
||
1051 | * The ringbuffer might not be ready to accept the commands right away (maybe it needs to |
||
1052 | * be wrapped, or wait a bit for the tail to be updated). This function takes care of that |
||
1053 | * and also preallocates a request (every workload submission is still mediated through |
||
1054 | * requests, same as it did with legacy ringbuffer submission). |
||
1055 | * |
||
1056 | * Return: non-zero if the ringbuffer is not ready to be written to. |
||
1057 | */ |
||
1058 | int intel_logical_ring_begin(struct intel_ringbuffer *ringbuf, int num_dwords) |
||
1059 | { |
||
1060 | struct intel_engine_cs *ring = ringbuf->ring; |
||
1061 | struct drm_device *dev = ring->dev; |
||
1062 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1063 | int ret; |
||
1064 | |||
1065 | ret = i915_gem_check_wedge(&dev_priv->gpu_error, |
||
1066 | dev_priv->mm.interruptible); |
||
1067 | if (ret) |
||
1068 | return ret; |
||
1069 | |||
1070 | ret = logical_ring_prepare(ringbuf, num_dwords * sizeof(uint32_t)); |
||
1071 | if (ret) |
||
1072 | return ret; |
||
1073 | |||
1074 | /* Preallocate the olr before touching the ring */ |
||
1075 | ret = logical_ring_alloc_seqno(ring, ringbuf->FIXME_lrc_ctx); |
||
1076 | if (ret) |
||
1077 | return ret; |
||
1078 | |||
1079 | ringbuf->space -= num_dwords * sizeof(uint32_t); |
||
1080 | return 0; |
||
1081 | } |
||
1082 | |||
1083 | static int intel_logical_ring_workarounds_emit(struct intel_engine_cs *ring, |
||
1084 | struct intel_context *ctx) |
||
1085 | { |
||
1086 | int ret, i; |
||
1087 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
||
1088 | struct drm_device *dev = ring->dev; |
||
1089 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1090 | struct i915_workarounds *w = &dev_priv->workarounds; |
||
1091 | |||
1092 | if (WARN_ON(w->count == 0)) |
||
1093 | return 0; |
||
1094 | |||
1095 | ring->gpu_caches_dirty = true; |
||
1096 | ret = logical_ring_flush_all_caches(ringbuf); |
||
1097 | if (ret) |
||
1098 | return ret; |
||
1099 | |||
1100 | ret = intel_logical_ring_begin(ringbuf, w->count * 2 + 2); |
||
1101 | if (ret) |
||
1102 | return ret; |
||
1103 | |||
1104 | intel_logical_ring_emit(ringbuf, MI_LOAD_REGISTER_IMM(w->count)); |
||
1105 | for (i = 0; i < w->count; i++) { |
||
1106 | intel_logical_ring_emit(ringbuf, w->reg[i].addr); |
||
1107 | intel_logical_ring_emit(ringbuf, w->reg[i].value); |
||
1108 | } |
||
1109 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
||
1110 | |||
1111 | intel_logical_ring_advance(ringbuf); |
||
1112 | |||
1113 | ring->gpu_caches_dirty = true; |
||
1114 | ret = logical_ring_flush_all_caches(ringbuf); |
||
1115 | if (ret) |
||
1116 | return ret; |
||
1117 | |||
1118 | return 0; |
||
1119 | } |
||
1120 | |||
1121 | static int gen8_init_common_ring(struct intel_engine_cs *ring) |
||
1122 | { |
||
1123 | struct drm_device *dev = ring->dev; |
||
1124 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1125 | |||
1126 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); |
||
1127 | I915_WRITE(RING_HWSTAM(ring->mmio_base), 0xffffffff); |
||
1128 | |||
1129 | I915_WRITE(RING_MODE_GEN7(ring), |
||
1130 | _MASKED_BIT_DISABLE(GFX_REPLAY_MODE) | |
||
1131 | _MASKED_BIT_ENABLE(GFX_RUN_LIST_ENABLE)); |
||
1132 | POSTING_READ(RING_MODE_GEN7(ring)); |
||
1133 | DRM_DEBUG_DRIVER("Execlists enabled for %s\n", ring->name); |
||
1134 | |||
1135 | memset(&ring->hangcheck, 0, sizeof(ring->hangcheck)); |
||
1136 | |||
1137 | return 0; |
||
1138 | } |
||
1139 | |||
1140 | static int gen8_init_render_ring(struct intel_engine_cs *ring) |
||
1141 | { |
||
1142 | struct drm_device *dev = ring->dev; |
||
1143 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1144 | int ret; |
||
1145 | |||
1146 | ret = gen8_init_common_ring(ring); |
||
1147 | if (ret) |
||
1148 | return ret; |
||
1149 | |||
1150 | /* We need to disable the AsyncFlip performance optimisations in order |
||
1151 | * to use MI_WAIT_FOR_EVENT within the CS. It should already be |
||
1152 | * programmed to '1' on all products. |
||
1153 | * |
||
1154 | * WaDisableAsyncFlipPerfMode:snb,ivb,hsw,vlv,bdw,chv |
||
1155 | */ |
||
1156 | I915_WRITE(MI_MODE, _MASKED_BIT_ENABLE(ASYNC_FLIP_PERF_DISABLE)); |
||
1157 | |||
1158 | ret = intel_init_pipe_control(ring); |
||
1159 | if (ret) |
||
1160 | return ret; |
||
1161 | |||
1162 | I915_WRITE(INSTPM, _MASKED_BIT_ENABLE(INSTPM_FORCE_ORDERING)); |
||
1163 | |||
1164 | return init_workarounds_ring(ring); |
||
1165 | } |
||
1166 | |||
1167 | static int gen8_emit_bb_start(struct intel_ringbuffer *ringbuf, |
||
1168 | u64 offset, unsigned flags) |
||
1169 | { |
||
1170 | bool ppgtt = !(flags & I915_DISPATCH_SECURE); |
||
1171 | int ret; |
||
1172 | |||
1173 | ret = intel_logical_ring_begin(ringbuf, 4); |
||
1174 | if (ret) |
||
1175 | return ret; |
||
1176 | |||
1177 | /* FIXME(BDW): Address space and security selectors. */ |
||
1178 | intel_logical_ring_emit(ringbuf, MI_BATCH_BUFFER_START_GEN8 | (ppgtt<<8)); |
||
1179 | intel_logical_ring_emit(ringbuf, lower_32_bits(offset)); |
||
1180 | intel_logical_ring_emit(ringbuf, upper_32_bits(offset)); |
||
1181 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
||
1182 | intel_logical_ring_advance(ringbuf); |
||
1183 | |||
1184 | return 0; |
||
1185 | } |
||
1186 | |||
1187 | static bool gen8_logical_ring_get_irq(struct intel_engine_cs *ring) |
||
1188 | { |
||
1189 | struct drm_device *dev = ring->dev; |
||
1190 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1191 | unsigned long flags; |
||
1192 | |||
1193 | if (WARN_ON(!intel_irqs_enabled(dev_priv))) |
||
1194 | return false; |
||
1195 | |||
1196 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
||
1197 | if (ring->irq_refcount++ == 0) { |
||
1198 | I915_WRITE_IMR(ring, ~(ring->irq_enable_mask | ring->irq_keep_mask)); |
||
1199 | POSTING_READ(RING_IMR(ring->mmio_base)); |
||
1200 | } |
||
1201 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
||
1202 | |||
1203 | return true; |
||
1204 | } |
||
1205 | |||
1206 | static void gen8_logical_ring_put_irq(struct intel_engine_cs *ring) |
||
1207 | { |
||
1208 | struct drm_device *dev = ring->dev; |
||
1209 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1210 | unsigned long flags; |
||
1211 | |||
1212 | spin_lock_irqsave(&dev_priv->irq_lock, flags); |
||
1213 | if (--ring->irq_refcount == 0) { |
||
1214 | I915_WRITE_IMR(ring, ~ring->irq_keep_mask); |
||
1215 | POSTING_READ(RING_IMR(ring->mmio_base)); |
||
1216 | } |
||
1217 | spin_unlock_irqrestore(&dev_priv->irq_lock, flags); |
||
1218 | } |
||
1219 | |||
1220 | static int gen8_emit_flush(struct intel_ringbuffer *ringbuf, |
||
1221 | u32 invalidate_domains, |
||
1222 | u32 unused) |
||
1223 | { |
||
1224 | struct intel_engine_cs *ring = ringbuf->ring; |
||
1225 | struct drm_device *dev = ring->dev; |
||
1226 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1227 | uint32_t cmd; |
||
1228 | int ret; |
||
1229 | |||
1230 | ret = intel_logical_ring_begin(ringbuf, 4); |
||
1231 | if (ret) |
||
1232 | return ret; |
||
1233 | |||
1234 | cmd = MI_FLUSH_DW + 1; |
||
1235 | |||
1236 | if (ring == &dev_priv->ring[VCS]) { |
||
1237 | if (invalidate_domains & I915_GEM_GPU_DOMAINS) |
||
1238 | cmd |= MI_INVALIDATE_TLB | MI_INVALIDATE_BSD | |
||
1239 | MI_FLUSH_DW_STORE_INDEX | |
||
1240 | MI_FLUSH_DW_OP_STOREDW; |
||
1241 | } else { |
||
1242 | if (invalidate_domains & I915_GEM_DOMAIN_RENDER) |
||
1243 | cmd |= MI_INVALIDATE_TLB | MI_FLUSH_DW_STORE_INDEX | |
||
1244 | MI_FLUSH_DW_OP_STOREDW; |
||
1245 | } |
||
1246 | |||
1247 | intel_logical_ring_emit(ringbuf, cmd); |
||
1248 | intel_logical_ring_emit(ringbuf, |
||
1249 | I915_GEM_HWS_SCRATCH_ADDR | |
||
1250 | MI_FLUSH_DW_USE_GTT); |
||
1251 | intel_logical_ring_emit(ringbuf, 0); /* upper addr */ |
||
1252 | intel_logical_ring_emit(ringbuf, 0); /* value */ |
||
1253 | intel_logical_ring_advance(ringbuf); |
||
1254 | |||
1255 | return 0; |
||
1256 | } |
||
1257 | |||
1258 | static int gen8_emit_flush_render(struct intel_ringbuffer *ringbuf, |
||
1259 | u32 invalidate_domains, |
||
1260 | u32 flush_domains) |
||
1261 | { |
||
1262 | struct intel_engine_cs *ring = ringbuf->ring; |
||
1263 | u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES; |
||
1264 | u32 flags = 0; |
||
1265 | int ret; |
||
1266 | |||
1267 | flags |= PIPE_CONTROL_CS_STALL; |
||
1268 | |||
1269 | if (flush_domains) { |
||
1270 | flags |= PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH; |
||
1271 | flags |= PIPE_CONTROL_DEPTH_CACHE_FLUSH; |
||
1272 | } |
||
1273 | |||
1274 | if (invalidate_domains) { |
||
1275 | flags |= PIPE_CONTROL_TLB_INVALIDATE; |
||
1276 | flags |= PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE; |
||
1277 | flags |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE; |
||
1278 | flags |= PIPE_CONTROL_VF_CACHE_INVALIDATE; |
||
1279 | flags |= PIPE_CONTROL_CONST_CACHE_INVALIDATE; |
||
1280 | flags |= PIPE_CONTROL_STATE_CACHE_INVALIDATE; |
||
1281 | flags |= PIPE_CONTROL_QW_WRITE; |
||
1282 | flags |= PIPE_CONTROL_GLOBAL_GTT_IVB; |
||
1283 | } |
||
1284 | |||
1285 | ret = intel_logical_ring_begin(ringbuf, 6); |
||
1286 | if (ret) |
||
1287 | return ret; |
||
1288 | |||
1289 | intel_logical_ring_emit(ringbuf, GFX_OP_PIPE_CONTROL(6)); |
||
1290 | intel_logical_ring_emit(ringbuf, flags); |
||
1291 | intel_logical_ring_emit(ringbuf, scratch_addr); |
||
1292 | intel_logical_ring_emit(ringbuf, 0); |
||
1293 | intel_logical_ring_emit(ringbuf, 0); |
||
1294 | intel_logical_ring_emit(ringbuf, 0); |
||
1295 | intel_logical_ring_advance(ringbuf); |
||
1296 | |||
1297 | return 0; |
||
1298 | } |
||
1299 | |||
1300 | static u32 gen8_get_seqno(struct intel_engine_cs *ring, bool lazy_coherency) |
||
1301 | { |
||
1302 | return intel_read_status_page(ring, I915_GEM_HWS_INDEX); |
||
1303 | } |
||
1304 | |||
1305 | static void gen8_set_seqno(struct intel_engine_cs *ring, u32 seqno) |
||
1306 | { |
||
1307 | intel_write_status_page(ring, I915_GEM_HWS_INDEX, seqno); |
||
1308 | } |
||
1309 | |||
1310 | static int gen8_emit_request(struct intel_ringbuffer *ringbuf) |
||
1311 | { |
||
1312 | struct intel_engine_cs *ring = ringbuf->ring; |
||
1313 | u32 cmd; |
||
1314 | int ret; |
||
1315 | |||
1316 | ret = intel_logical_ring_begin(ringbuf, 6); |
||
1317 | if (ret) |
||
1318 | return ret; |
||
1319 | |||
1320 | cmd = MI_STORE_DWORD_IMM_GEN8; |
||
1321 | cmd |= MI_GLOBAL_GTT; |
||
1322 | |||
1323 | intel_logical_ring_emit(ringbuf, cmd); |
||
1324 | intel_logical_ring_emit(ringbuf, |
||
1325 | (ring->status_page.gfx_addr + |
||
1326 | (I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT))); |
||
1327 | intel_logical_ring_emit(ringbuf, 0); |
||
1328 | intel_logical_ring_emit(ringbuf, ring->outstanding_lazy_seqno); |
||
1329 | intel_logical_ring_emit(ringbuf, MI_USER_INTERRUPT); |
||
1330 | intel_logical_ring_emit(ringbuf, MI_NOOP); |
||
1331 | intel_logical_ring_advance_and_submit(ringbuf); |
||
1332 | |||
1333 | return 0; |
||
1334 | } |
||
1335 | |||
1336 | /** |
||
1337 | * intel_logical_ring_cleanup() - deallocate the Engine Command Streamer |
||
1338 | * |
||
1339 | * @ring: Engine Command Streamer. |
||
1340 | * |
||
1341 | */ |
||
1342 | void intel_logical_ring_cleanup(struct intel_engine_cs *ring) |
||
1343 | { |
||
1344 | struct drm_i915_private *dev_priv; |
||
1345 | |||
1346 | if (!intel_ring_initialized(ring)) |
||
1347 | return; |
||
1348 | |||
1349 | dev_priv = ring->dev->dev_private; |
||
1350 | |||
1351 | intel_logical_ring_stop(ring); |
||
1352 | WARN_ON((I915_READ_MODE(ring) & MODE_IDLE) == 0); |
||
1353 | ring->preallocated_lazy_request = NULL; |
||
1354 | ring->outstanding_lazy_seqno = 0; |
||
1355 | |||
1356 | if (ring->cleanup) |
||
1357 | ring->cleanup(ring); |
||
1358 | |||
1359 | i915_cmd_parser_fini_ring(ring); |
||
1360 | |||
1361 | if (ring->status_page.obj) { |
||
1362 | kunmap(sg_page(ring->status_page.obj->pages->sgl)); |
||
1363 | ring->status_page.obj = NULL; |
||
1364 | } |
||
1365 | } |
||
1366 | |||
1367 | static int logical_ring_init(struct drm_device *dev, struct intel_engine_cs *ring) |
||
1368 | { |
||
1369 | int ret; |
||
1370 | |||
1371 | /* Intentionally left blank. */ |
||
1372 | ring->buffer = NULL; |
||
1373 | |||
1374 | ring->dev = dev; |
||
1375 | INIT_LIST_HEAD(&ring->active_list); |
||
1376 | INIT_LIST_HEAD(&ring->request_list); |
||
1377 | init_waitqueue_head(&ring->irq_queue); |
||
1378 | |||
1379 | INIT_LIST_HEAD(&ring->execlist_queue); |
||
1380 | INIT_LIST_HEAD(&ring->execlist_retired_req_list); |
||
1381 | spin_lock_init(&ring->execlist_lock); |
||
1382 | ring->next_context_status_buffer = 0; |
||
1383 | |||
1384 | ret = i915_cmd_parser_init_ring(ring); |
||
1385 | if (ret) |
||
1386 | return ret; |
||
1387 | |||
1388 | if (ring->init) { |
||
1389 | ret = ring->init(ring); |
||
1390 | if (ret) |
||
1391 | return ret; |
||
1392 | } |
||
1393 | |||
1394 | ret = intel_lr_context_deferred_create(ring->default_context, ring); |
||
1395 | |||
1396 | return ret; |
||
1397 | } |
||
1398 | |||
1399 | static int logical_render_ring_init(struct drm_device *dev) |
||
1400 | { |
||
1401 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1402 | struct intel_engine_cs *ring = &dev_priv->ring[RCS]; |
||
1403 | |||
1404 | ring->name = "render ring"; |
||
1405 | ring->id = RCS; |
||
1406 | ring->mmio_base = RENDER_RING_BASE; |
||
1407 | ring->irq_enable_mask = |
||
1408 | GT_RENDER_USER_INTERRUPT << GEN8_RCS_IRQ_SHIFT; |
||
1409 | ring->irq_keep_mask = |
||
1410 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_RCS_IRQ_SHIFT; |
||
1411 | if (HAS_L3_DPF(dev)) |
||
1412 | ring->irq_keep_mask |= GT_RENDER_L3_PARITY_ERROR_INTERRUPT; |
||
1413 | |||
1414 | ring->init = gen8_init_render_ring; |
||
1415 | ring->init_context = intel_logical_ring_workarounds_emit; |
||
1416 | ring->cleanup = intel_fini_pipe_control; |
||
1417 | ring->get_seqno = gen8_get_seqno; |
||
1418 | ring->set_seqno = gen8_set_seqno; |
||
1419 | ring->emit_request = gen8_emit_request; |
||
1420 | ring->emit_flush = gen8_emit_flush_render; |
||
1421 | ring->irq_get = gen8_logical_ring_get_irq; |
||
1422 | ring->irq_put = gen8_logical_ring_put_irq; |
||
1423 | ring->emit_bb_start = gen8_emit_bb_start; |
||
1424 | |||
1425 | return logical_ring_init(dev, ring); |
||
1426 | } |
||
1427 | |||
1428 | static int logical_bsd_ring_init(struct drm_device *dev) |
||
1429 | { |
||
1430 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1431 | struct intel_engine_cs *ring = &dev_priv->ring[VCS]; |
||
1432 | |||
1433 | ring->name = "bsd ring"; |
||
1434 | ring->id = VCS; |
||
1435 | ring->mmio_base = GEN6_BSD_RING_BASE; |
||
1436 | ring->irq_enable_mask = |
||
1437 | GT_RENDER_USER_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; |
||
1438 | ring->irq_keep_mask = |
||
1439 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS1_IRQ_SHIFT; |
||
1440 | |||
1441 | ring->init = gen8_init_common_ring; |
||
1442 | ring->get_seqno = gen8_get_seqno; |
||
1443 | ring->set_seqno = gen8_set_seqno; |
||
1444 | ring->emit_request = gen8_emit_request; |
||
1445 | ring->emit_flush = gen8_emit_flush; |
||
1446 | ring->irq_get = gen8_logical_ring_get_irq; |
||
1447 | ring->irq_put = gen8_logical_ring_put_irq; |
||
1448 | ring->emit_bb_start = gen8_emit_bb_start; |
||
1449 | |||
1450 | return logical_ring_init(dev, ring); |
||
1451 | } |
||
1452 | |||
1453 | static int logical_bsd2_ring_init(struct drm_device *dev) |
||
1454 | { |
||
1455 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1456 | struct intel_engine_cs *ring = &dev_priv->ring[VCS2]; |
||
1457 | |||
1458 | ring->name = "bds2 ring"; |
||
1459 | ring->id = VCS2; |
||
1460 | ring->mmio_base = GEN8_BSD2_RING_BASE; |
||
1461 | ring->irq_enable_mask = |
||
1462 | GT_RENDER_USER_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; |
||
1463 | ring->irq_keep_mask = |
||
1464 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VCS2_IRQ_SHIFT; |
||
1465 | |||
1466 | ring->init = gen8_init_common_ring; |
||
1467 | ring->get_seqno = gen8_get_seqno; |
||
1468 | ring->set_seqno = gen8_set_seqno; |
||
1469 | ring->emit_request = gen8_emit_request; |
||
1470 | ring->emit_flush = gen8_emit_flush; |
||
1471 | ring->irq_get = gen8_logical_ring_get_irq; |
||
1472 | ring->irq_put = gen8_logical_ring_put_irq; |
||
1473 | ring->emit_bb_start = gen8_emit_bb_start; |
||
1474 | |||
1475 | return logical_ring_init(dev, ring); |
||
1476 | } |
||
1477 | |||
1478 | static int logical_blt_ring_init(struct drm_device *dev) |
||
1479 | { |
||
1480 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1481 | struct intel_engine_cs *ring = &dev_priv->ring[BCS]; |
||
1482 | |||
1483 | ring->name = "blitter ring"; |
||
1484 | ring->id = BCS; |
||
1485 | ring->mmio_base = BLT_RING_BASE; |
||
1486 | ring->irq_enable_mask = |
||
1487 | GT_RENDER_USER_INTERRUPT << GEN8_BCS_IRQ_SHIFT; |
||
1488 | ring->irq_keep_mask = |
||
1489 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_BCS_IRQ_SHIFT; |
||
1490 | |||
1491 | ring->init = gen8_init_common_ring; |
||
1492 | ring->get_seqno = gen8_get_seqno; |
||
1493 | ring->set_seqno = gen8_set_seqno; |
||
1494 | ring->emit_request = gen8_emit_request; |
||
1495 | ring->emit_flush = gen8_emit_flush; |
||
1496 | ring->irq_get = gen8_logical_ring_get_irq; |
||
1497 | ring->irq_put = gen8_logical_ring_put_irq; |
||
1498 | ring->emit_bb_start = gen8_emit_bb_start; |
||
1499 | |||
1500 | return logical_ring_init(dev, ring); |
||
1501 | } |
||
1502 | |||
1503 | static int logical_vebox_ring_init(struct drm_device *dev) |
||
1504 | { |
||
1505 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1506 | struct intel_engine_cs *ring = &dev_priv->ring[VECS]; |
||
1507 | |||
1508 | ring->name = "video enhancement ring"; |
||
1509 | ring->id = VECS; |
||
1510 | ring->mmio_base = VEBOX_RING_BASE; |
||
1511 | ring->irq_enable_mask = |
||
1512 | GT_RENDER_USER_INTERRUPT << GEN8_VECS_IRQ_SHIFT; |
||
1513 | ring->irq_keep_mask = |
||
1514 | GT_CONTEXT_SWITCH_INTERRUPT << GEN8_VECS_IRQ_SHIFT; |
||
1515 | |||
1516 | ring->init = gen8_init_common_ring; |
||
1517 | ring->get_seqno = gen8_get_seqno; |
||
1518 | ring->set_seqno = gen8_set_seqno; |
||
1519 | ring->emit_request = gen8_emit_request; |
||
1520 | ring->emit_flush = gen8_emit_flush; |
||
1521 | ring->irq_get = gen8_logical_ring_get_irq; |
||
1522 | ring->irq_put = gen8_logical_ring_put_irq; |
||
1523 | ring->emit_bb_start = gen8_emit_bb_start; |
||
1524 | |||
1525 | return logical_ring_init(dev, ring); |
||
1526 | } |
||
1527 | |||
1528 | /** |
||
1529 | * intel_logical_rings_init() - allocate, populate and init the Engine Command Streamers |
||
1530 | * @dev: DRM device. |
||
1531 | * |
||
1532 | * This function inits the engines for an Execlists submission style (the equivalent in the |
||
1533 | * legacy ringbuffer submission world would be i915_gem_init_rings). It does it only for |
||
1534 | * those engines that are present in the hardware. |
||
1535 | * |
||
1536 | * Return: non-zero if the initialization failed. |
||
1537 | */ |
||
1538 | int intel_logical_rings_init(struct drm_device *dev) |
||
1539 | { |
||
1540 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1541 | int ret; |
||
1542 | |||
1543 | ret = logical_render_ring_init(dev); |
||
1544 | if (ret) |
||
1545 | return ret; |
||
1546 | |||
1547 | if (HAS_BSD(dev)) { |
||
1548 | ret = logical_bsd_ring_init(dev); |
||
1549 | if (ret) |
||
1550 | goto cleanup_render_ring; |
||
1551 | } |
||
1552 | |||
1553 | if (HAS_BLT(dev)) { |
||
1554 | ret = logical_blt_ring_init(dev); |
||
1555 | if (ret) |
||
1556 | goto cleanup_bsd_ring; |
||
1557 | } |
||
1558 | |||
1559 | if (HAS_VEBOX(dev)) { |
||
1560 | ret = logical_vebox_ring_init(dev); |
||
1561 | if (ret) |
||
1562 | goto cleanup_blt_ring; |
||
1563 | } |
||
1564 | |||
1565 | if (HAS_BSD2(dev)) { |
||
1566 | ret = logical_bsd2_ring_init(dev); |
||
1567 | if (ret) |
||
1568 | goto cleanup_vebox_ring; |
||
1569 | } |
||
1570 | |||
1571 | ret = i915_gem_set_seqno(dev, ((u32)~0 - 0x1000)); |
||
1572 | if (ret) |
||
1573 | goto cleanup_bsd2_ring; |
||
1574 | |||
1575 | return 0; |
||
1576 | |||
1577 | cleanup_bsd2_ring: |
||
1578 | intel_logical_ring_cleanup(&dev_priv->ring[VCS2]); |
||
1579 | cleanup_vebox_ring: |
||
1580 | intel_logical_ring_cleanup(&dev_priv->ring[VECS]); |
||
1581 | cleanup_blt_ring: |
||
1582 | intel_logical_ring_cleanup(&dev_priv->ring[BCS]); |
||
1583 | cleanup_bsd_ring: |
||
1584 | intel_logical_ring_cleanup(&dev_priv->ring[VCS]); |
||
1585 | cleanup_render_ring: |
||
1586 | intel_logical_ring_cleanup(&dev_priv->ring[RCS]); |
||
1587 | |||
1588 | return ret; |
||
1589 | } |
||
1590 | |||
1591 | int intel_lr_context_render_state_init(struct intel_engine_cs *ring, |
||
1592 | struct intel_context *ctx) |
||
1593 | { |
||
1594 | struct intel_ringbuffer *ringbuf = ctx->engine[ring->id].ringbuf; |
||
1595 | struct render_state so; |
||
1596 | struct drm_i915_file_private *file_priv = ctx->file_priv; |
||
1597 | struct drm_file *file = file_priv ? file_priv->file : NULL; |
||
1598 | int ret; |
||
1599 | |||
1600 | ret = i915_gem_render_state_prepare(ring, &so); |
||
1601 | if (ret) |
||
1602 | return ret; |
||
1603 | |||
1604 | if (so.rodata == NULL) |
||
1605 | return 0; |
||
1606 | |||
1607 | ret = ring->emit_bb_start(ringbuf, |
||
1608 | so.ggtt_offset, |
||
1609 | I915_DISPATCH_SECURE); |
||
1610 | if (ret) |
||
1611 | goto out; |
||
1612 | |||
1613 | i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), ring); |
||
1614 | |||
1615 | ret = __i915_add_request(ring, file, so.obj, NULL); |
||
1616 | /* intel_logical_ring_add_request moves object to inactive if it |
||
1617 | * fails */ |
||
1618 | out: |
||
1619 | i915_gem_render_state_fini(&so); |
||
1620 | return ret; |
||
1621 | } |
||
1622 | |||
1623 | static int |
||
1624 | populate_lr_context(struct intel_context *ctx, struct drm_i915_gem_object *ctx_obj, |
||
1625 | struct intel_engine_cs *ring, struct intel_ringbuffer *ringbuf) |
||
1626 | { |
||
1627 | struct drm_device *dev = ring->dev; |
||
1628 | struct drm_i915_private *dev_priv = dev->dev_private; |
||
1629 | struct i915_hw_ppgtt *ppgtt = ctx->ppgtt; |
||
1630 | struct page *page; |
||
1631 | uint32_t *reg_state; |
||
1632 | int ret; |
||
1633 | |||
1634 | if (!ppgtt) |
||
1635 | ppgtt = dev_priv->mm.aliasing_ppgtt; |
||
1636 | |||
1637 | ret = i915_gem_object_set_to_cpu_domain(ctx_obj, true); |
||
1638 | if (ret) { |
||
1639 | DRM_DEBUG_DRIVER("Could not set to CPU domain\n"); |
||
1640 | return ret; |
||
1641 | } |
||
1642 | |||
1643 | ret = i915_gem_object_get_pages(ctx_obj); |
||
1644 | if (ret) { |
||
1645 | DRM_DEBUG_DRIVER("Could not get object pages\n"); |
||
1646 | return ret; |
||
1647 | } |
||
1648 | |||
1649 | i915_gem_object_pin_pages(ctx_obj); |
||
1650 | |||
1651 | /* The second page of the context object contains some fields which must |
||
1652 | * be set up prior to the first execution. */ |
||
1653 | page = i915_gem_object_get_page(ctx_obj, 1); |
||
1654 | reg_state = kmap_atomic(page); |
||
1655 | |||
1656 | /* A context is actually a big batch buffer with several MI_LOAD_REGISTER_IMM |
||
1657 | * commands followed by (reg, value) pairs. The values we are setting here are |
||
1658 | * only for the first context restore: on a subsequent save, the GPU will |
||
1659 | * recreate this batchbuffer with new values (including all the missing |
||
1660 | * MI_LOAD_REGISTER_IMM commands that we are not initializing here). */ |
||
1661 | if (ring->id == RCS) |
||
1662 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(14); |
||
1663 | else |
||
1664 | reg_state[CTX_LRI_HEADER_0] = MI_LOAD_REGISTER_IMM(11); |
||
1665 | reg_state[CTX_LRI_HEADER_0] |= MI_LRI_FORCE_POSTED; |
||
1666 | reg_state[CTX_CONTEXT_CONTROL] = RING_CONTEXT_CONTROL(ring); |
||
1667 | reg_state[CTX_CONTEXT_CONTROL+1] = |
||
1668 | _MASKED_BIT_ENABLE((1<<3) | MI_RESTORE_INHIBIT); |
||
1669 | reg_state[CTX_RING_HEAD] = RING_HEAD(ring->mmio_base); |
||
1670 | reg_state[CTX_RING_HEAD+1] = 0; |
||
1671 | reg_state[CTX_RING_TAIL] = RING_TAIL(ring->mmio_base); |
||
1672 | reg_state[CTX_RING_TAIL+1] = 0; |
||
1673 | reg_state[CTX_RING_BUFFER_START] = RING_START(ring->mmio_base); |
||
1674 | /* Ring buffer start address is not known until the buffer is pinned. |
||
1675 | * It is written to the context image in execlists_update_context() |
||
1676 | */ |
||
1677 | reg_state[CTX_RING_BUFFER_CONTROL] = RING_CTL(ring->mmio_base); |
||
1678 | reg_state[CTX_RING_BUFFER_CONTROL+1] = |
||
1679 | ((ringbuf->size - PAGE_SIZE) & RING_NR_PAGES) | RING_VALID; |
||
1680 | reg_state[CTX_BB_HEAD_U] = ring->mmio_base + 0x168; |
||
1681 | reg_state[CTX_BB_HEAD_U+1] = 0; |
||
1682 | reg_state[CTX_BB_HEAD_L] = ring->mmio_base + 0x140; |
||
1683 | reg_state[CTX_BB_HEAD_L+1] = 0; |
||
1684 | reg_state[CTX_BB_STATE] = ring->mmio_base + 0x110; |
||
1685 | reg_state[CTX_BB_STATE+1] = (1<<5); |
||
1686 | reg_state[CTX_SECOND_BB_HEAD_U] = ring->mmio_base + 0x11c; |
||
1687 | reg_state[CTX_SECOND_BB_HEAD_U+1] = 0; |
||
1688 | reg_state[CTX_SECOND_BB_HEAD_L] = ring->mmio_base + 0x114; |
||
1689 | reg_state[CTX_SECOND_BB_HEAD_L+1] = 0; |
||
1690 | reg_state[CTX_SECOND_BB_STATE] = ring->mmio_base + 0x118; |
||
1691 | reg_state[CTX_SECOND_BB_STATE+1] = 0; |
||
1692 | if (ring->id == RCS) { |
||
1693 | /* TODO: according to BSpec, the register state context |
||
1694 | * for CHV does not have these. OTOH, these registers do |
||
1695 | * exist in CHV. I'm waiting for a clarification */ |
||
1696 | reg_state[CTX_BB_PER_CTX_PTR] = ring->mmio_base + 0x1c0; |
||
1697 | reg_state[CTX_BB_PER_CTX_PTR+1] = 0; |
||
1698 | reg_state[CTX_RCS_INDIRECT_CTX] = ring->mmio_base + 0x1c4; |
||
1699 | reg_state[CTX_RCS_INDIRECT_CTX+1] = 0; |
||
1700 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET] = ring->mmio_base + 0x1c8; |
||
1701 | reg_state[CTX_RCS_INDIRECT_CTX_OFFSET+1] = 0; |
||
1702 | } |
||
1703 | reg_state[CTX_LRI_HEADER_1] = MI_LOAD_REGISTER_IMM(9); |
||
1704 | reg_state[CTX_LRI_HEADER_1] |= MI_LRI_FORCE_POSTED; |
||
1705 | reg_state[CTX_CTX_TIMESTAMP] = ring->mmio_base + 0x3a8; |
||
1706 | reg_state[CTX_CTX_TIMESTAMP+1] = 0; |
||
1707 | reg_state[CTX_PDP3_UDW] = GEN8_RING_PDP_UDW(ring, 3); |
||
1708 | reg_state[CTX_PDP3_LDW] = GEN8_RING_PDP_LDW(ring, 3); |
||
1709 | reg_state[CTX_PDP2_UDW] = GEN8_RING_PDP_UDW(ring, 2); |
||
1710 | reg_state[CTX_PDP2_LDW] = GEN8_RING_PDP_LDW(ring, 2); |
||
1711 | reg_state[CTX_PDP1_UDW] = GEN8_RING_PDP_UDW(ring, 1); |
||
1712 | reg_state[CTX_PDP1_LDW] = GEN8_RING_PDP_LDW(ring, 1); |
||
1713 | reg_state[CTX_PDP0_UDW] = GEN8_RING_PDP_UDW(ring, 0); |
||
1714 | reg_state[CTX_PDP0_LDW] = GEN8_RING_PDP_LDW(ring, 0); |
||
1715 | reg_state[CTX_PDP3_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[3]); |
||
1716 | reg_state[CTX_PDP3_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[3]); |
||
1717 | reg_state[CTX_PDP2_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[2]); |
||
1718 | reg_state[CTX_PDP2_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[2]); |
||
1719 | reg_state[CTX_PDP1_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[1]); |
||
1720 | reg_state[CTX_PDP1_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[1]); |
||
1721 | reg_state[CTX_PDP0_UDW+1] = upper_32_bits(ppgtt->pd_dma_addr[0]); |
||
1722 | reg_state[CTX_PDP0_LDW+1] = lower_32_bits(ppgtt->pd_dma_addr[0]); |
||
1723 | if (ring->id == RCS) { |
||
1724 | reg_state[CTX_LRI_HEADER_2] = MI_LOAD_REGISTER_IMM(1); |
||
1725 | reg_state[CTX_R_PWR_CLK_STATE] = 0x20c8; |
||
1726 | reg_state[CTX_R_PWR_CLK_STATE+1] = 0; |
||
1727 | } |
||
1728 | |||
1729 | kunmap_atomic(reg_state); |
||
1730 | |||
1731 | ctx_obj->dirty = 1; |
||
1732 | // set_page_dirty(page); |
||
1733 | i915_gem_object_unpin_pages(ctx_obj); |
||
1734 | |||
1735 | return 0; |
||
1736 | } |
||
1737 | |||
1738 | /** |
||
1739 | * intel_lr_context_free() - free the LRC specific bits of a context |
||
1740 | * @ctx: the LR context to free. |
||
1741 | * |
||
1742 | * The real context freeing is done in i915_gem_context_free: this only |
||
1743 | * takes care of the bits that are LRC related: the per-engine backing |
||
1744 | * objects and the logical ringbuffer. |
||
1745 | */ |
||
1746 | void intel_lr_context_free(struct intel_context *ctx) |
||
1747 | { |
||
1748 | int i; |
||
1749 | |||
1750 | for (i = 0; i < I915_NUM_RINGS; i++) { |
||
1751 | struct drm_i915_gem_object *ctx_obj = ctx->engine[i].state; |
||
1752 | |||
1753 | if (ctx_obj) { |
||
1754 | struct intel_ringbuffer *ringbuf = |
||
1755 | ctx->engine[i].ringbuf; |
||
1756 | struct intel_engine_cs *ring = ringbuf->ring; |
||
1757 | |||
1758 | if (ctx == ring->default_context) { |
||
1759 | intel_unpin_ringbuffer_obj(ringbuf); |
||
1760 | i915_gem_object_ggtt_unpin(ctx_obj); |
||
1761 | } |
||
1762 | intel_destroy_ringbuffer_obj(ringbuf); |
||
1763 | kfree(ringbuf); |
||
1764 | drm_gem_object_unreference(&ctx_obj->base); |
||
1765 | } |
||
1766 | } |
||
1767 | } |
||
1768 | |||
1769 | static uint32_t get_lr_context_size(struct intel_engine_cs *ring) |
||
1770 | { |
||
1771 | int ret = 0; |
||
1772 | |||
1773 | WARN_ON(INTEL_INFO(ring->dev)->gen < 8); |
||
1774 | |||
1775 | switch (ring->id) { |
||
1776 | case RCS: |
||
1777 | if (INTEL_INFO(ring->dev)->gen >= 9) |
||
1778 | ret = GEN9_LR_CONTEXT_RENDER_SIZE; |
||
1779 | else |
||
1780 | ret = GEN8_LR_CONTEXT_RENDER_SIZE; |
||
1781 | break; |
||
1782 | case VCS: |
||
1783 | case BCS: |
||
1784 | case VECS: |
||
1785 | case VCS2: |
||
1786 | ret = GEN8_LR_CONTEXT_OTHER_SIZE; |
||
1787 | break; |
||
1788 | } |
||
1789 | |||
1790 | return ret; |
||
1791 | } |
||
1792 | |||
1793 | static void lrc_setup_hardware_status_page(struct intel_engine_cs *ring, |
||
1794 | struct drm_i915_gem_object *default_ctx_obj) |
||
1795 | { |
||
1796 | struct drm_i915_private *dev_priv = ring->dev->dev_private; |
||
1797 | |||
1798 | /* The status page is offset 0 from the default context object |
||
1799 | * in LRC mode. */ |
||
1800 | ring->status_page.gfx_addr = i915_gem_obj_ggtt_offset(default_ctx_obj); |
||
1801 | ring->status_page.page_addr = |
||
1802 | kmap(sg_page(default_ctx_obj->pages->sgl)); |
||
1803 | ring->status_page.obj = default_ctx_obj; |
||
1804 | |||
1805 | I915_WRITE(RING_HWS_PGA(ring->mmio_base), |
||
1806 | (u32)ring->status_page.gfx_addr); |
||
1807 | POSTING_READ(RING_HWS_PGA(ring->mmio_base)); |
||
1808 | } |
||
1809 | |||
1810 | /** |
||
1811 | * intel_lr_context_deferred_create() - create the LRC specific bits of a context |
||
1812 | * @ctx: LR context to create. |
||
1813 | * @ring: engine to be used with the context. |
||
1814 | * |
||
1815 | * This function can be called more than once, with different engines, if we plan |
||
1816 | * to use the context with them. The context backing objects and the ringbuffers |
||
1817 | * (specially the ringbuffer backing objects) suck a lot of memory up, and that's why |
||
1818 | * the creation is a deferred call: it's better to make sure first that we need to use |
||
1819 | * a given ring with the context. |
||
1820 | * |
||
1821 | * Return: non-zero on error. |
||
1822 | */ |
||
1823 | int intel_lr_context_deferred_create(struct intel_context *ctx, |
||
1824 | struct intel_engine_cs *ring) |
||
1825 | { |
||
1826 | const bool is_global_default_ctx = (ctx == ring->default_context); |
||
1827 | struct drm_device *dev = ring->dev; |
||
1828 | struct drm_i915_gem_object *ctx_obj; |
||
1829 | uint32_t context_size; |
||
1830 | struct intel_ringbuffer *ringbuf; |
||
1831 | int ret; |
||
1832 | |||
1833 | WARN_ON(ctx->legacy_hw_ctx.rcs_state != NULL); |
||
1834 | if (ctx->engine[ring->id].state) |
||
1835 | return 0; |
||
1836 | |||
1837 | context_size = round_up(get_lr_context_size(ring), 4096); |
||
1838 | |||
1839 | ctx_obj = i915_gem_alloc_context_obj(dev, context_size); |
||
1840 | if (IS_ERR(ctx_obj)) { |
||
1841 | ret = PTR_ERR(ctx_obj); |
||
1842 | DRM_DEBUG_DRIVER("Alloc LRC backing obj failed: %d\n", ret); |
||
1843 | return ret; |
||
1844 | } |
||
1845 | |||
1846 | if (is_global_default_ctx) { |
||
1847 | ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, 0); |
||
1848 | if (ret) { |
||
1849 | DRM_DEBUG_DRIVER("Pin LRC backing obj failed: %d\n", |
||
1850 | ret); |
||
1851 | drm_gem_object_unreference(&ctx_obj->base); |
||
1852 | return ret; |
||
1853 | } |
||
1854 | } |
||
1855 | |||
1856 | ringbuf = kzalloc(sizeof(*ringbuf), GFP_KERNEL); |
||
1857 | if (!ringbuf) { |
||
1858 | DRM_DEBUG_DRIVER("Failed to allocate ringbuffer %s\n", |
||
1859 | ring->name); |
||
1860 | ret = -ENOMEM; |
||
1861 | goto error_unpin_ctx; |
||
1862 | } |
||
1863 | |||
1864 | ringbuf->ring = ring; |
||
1865 | ringbuf->FIXME_lrc_ctx = ctx; |
||
1866 | |||
1867 | ringbuf->size = 32 * PAGE_SIZE; |
||
1868 | ringbuf->effective_size = ringbuf->size; |
||
1869 | ringbuf->head = 0; |
||
1870 | ringbuf->tail = 0; |
||
1871 | ringbuf->space = ringbuf->size; |
||
1872 | ringbuf->last_retired_head = -1; |
||
1873 | |||
1874 | if (ringbuf->obj == NULL) { |
||
1875 | ret = intel_alloc_ringbuffer_obj(dev, ringbuf); |
||
1876 | if (ret) { |
||
1877 | DRM_DEBUG_DRIVER( |
||
1878 | "Failed to allocate ringbuffer obj %s: %d\n", |
||
1879 | ring->name, ret); |
||
1880 | goto error_free_rbuf; |
||
1881 | } |
||
1882 | |||
1883 | if (is_global_default_ctx) { |
||
1884 | ret = intel_pin_and_map_ringbuffer_obj(dev, ringbuf); |
||
1885 | if (ret) { |
||
1886 | DRM_ERROR( |
||
1887 | "Failed to pin and map ringbuffer %s: %d\n", |
||
1888 | ring->name, ret); |
||
1889 | goto error_destroy_rbuf; |
||
1890 | } |
||
1891 | } |
||
1892 | |||
1893 | } |
||
1894 | |||
1895 | ret = populate_lr_context(ctx, ctx_obj, ring, ringbuf); |
||
1896 | if (ret) { |
||
1897 | DRM_DEBUG_DRIVER("Failed to populate LRC: %d\n", ret); |
||
1898 | goto error; |
||
1899 | } |
||
1900 | |||
1901 | ctx->engine[ring->id].ringbuf = ringbuf; |
||
1902 | ctx->engine[ring->id].state = ctx_obj; |
||
1903 | |||
1904 | if (ctx == ring->default_context) |
||
1905 | lrc_setup_hardware_status_page(ring, ctx_obj); |
||
1906 | |||
1907 | if (ring->id == RCS && !ctx->rcs_initialized) { |
||
1908 | if (ring->init_context) { |
||
1909 | ret = ring->init_context(ring, ctx); |
||
1910 | if (ret) |
||
1911 | DRM_ERROR("ring init context: %d\n", ret); |
||
1912 | } |
||
1913 | |||
1914 | ret = intel_lr_context_render_state_init(ring, ctx); |
||
1915 | if (ret) { |
||
1916 | DRM_ERROR("Init render state failed: %d\n", ret); |
||
1917 | ctx->engine[ring->id].ringbuf = NULL; |
||
1918 | ctx->engine[ring->id].state = NULL; |
||
1919 | goto error; |
||
1920 | } |
||
1921 | ctx->rcs_initialized = true; |
||
1922 | } |
||
1923 | |||
1924 | return 0; |
||
1925 | |||
1926 | error: |
||
1927 | if (is_global_default_ctx) |
||
1928 | intel_unpin_ringbuffer_obj(ringbuf); |
||
1929 | error_destroy_rbuf: |
||
1930 | intel_destroy_ringbuffer_obj(ringbuf); |
||
1931 | error_free_rbuf: |
||
1932 | kfree(ringbuf); |
||
1933 | error_unpin_ctx: |
||
1934 | if (is_global_default_ctx) |
||
1935 | i915_gem_object_ggtt_unpin(ctx_obj); |
||
1936 | drm_gem_object_unreference(&ctx_obj->base); |
||
1937 | return ret; |
||
1938 | }>>5); |