Rev 5271 | Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
5078 | serge | 1 | /* |
2 | * Copyright 2013 Advanced Micro Devices, Inc. |
||
3 | * All Rights Reserved. |
||
4 | * |
||
5 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
6 | * copy of this software and associated documentation files (the |
||
7 | * "Software"), to deal in the Software without restriction, including |
||
8 | * without limitation the rights to use, copy, modify, merge, publish, |
||
9 | * distribute, sub license, and/or sell copies of the Software, and to |
||
10 | * permit persons to whom the Software is furnished to do so, subject to |
||
11 | * the following conditions: |
||
12 | * |
||
13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
15 | * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL |
||
16 | * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, |
||
17 | * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR |
||
18 | * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE |
||
19 | * USE OR OTHER DEALINGS IN THE SOFTWARE. |
||
20 | * |
||
21 | * The above copyright notice and this permission notice (including the |
||
22 | * next paragraph) shall be included in all copies or substantial portions |
||
23 | * of the Software. |
||
24 | * |
||
25 | * Authors: Christian König |
||
26 | */ |
||
27 | |||
28 | #include |
||
29 | #include |
||
30 | #include |
||
31 | #include |
||
32 | |||
33 | #include "radeon.h" |
||
34 | #include "radeon_asic.h" |
||
35 | #include "sid.h" |
||
36 | |||
37 | /* 1 second timeout */ |
||
38 | #define VCE_IDLE_TIMEOUT_MS 1000 |
||
39 | |||
40 | /* Firmware Names */ |
||
41 | #define FIRMWARE_BONAIRE "radeon/BONAIRE_vce.bin" |
||
42 | |||
43 | MODULE_FIRMWARE(FIRMWARE_BONAIRE); |
||
44 | |||
45 | static void radeon_vce_idle_work_handler(struct work_struct *work); |
||
46 | |||
47 | /** |
||
48 | * radeon_vce_init - allocate memory, load vce firmware |
||
49 | * |
||
50 | * @rdev: radeon_device pointer |
||
51 | * |
||
52 | * First step to get VCE online, allocate memory and load the firmware |
||
53 | */ |
||
54 | int radeon_vce_init(struct radeon_device *rdev) |
||
55 | { |
||
56 | static const char *fw_version = "[ATI LIB=VCEFW,"; |
||
57 | static const char *fb_version = "[ATI LIB=VCEFWSTATS,"; |
||
58 | unsigned long size; |
||
59 | const char *fw_name, *c; |
||
60 | uint8_t start, mid, end; |
||
61 | int i, r; |
||
62 | |||
63 | INIT_DELAYED_WORK(&rdev->vce.idle_work, radeon_vce_idle_work_handler); |
||
64 | |||
65 | switch (rdev->family) { |
||
66 | case CHIP_BONAIRE: |
||
67 | case CHIP_KAVERI: |
||
68 | case CHIP_KABINI: |
||
69 | case CHIP_HAWAII: |
||
70 | case CHIP_MULLINS: |
||
71 | fw_name = FIRMWARE_BONAIRE; |
||
72 | break; |
||
73 | |||
74 | default: |
||
75 | return -EINVAL; |
||
76 | } |
||
77 | |||
78 | r = request_firmware(&rdev->vce_fw, fw_name, rdev->dev); |
||
79 | if (r) { |
||
80 | dev_err(rdev->dev, "radeon_vce: Can't load firmware \"%s\"\n", |
||
81 | fw_name); |
||
82 | return r; |
||
83 | } |
||
84 | |||
85 | /* search for firmware version */ |
||
86 | |||
87 | size = rdev->vce_fw->size - strlen(fw_version) - 9; |
||
88 | c = rdev->vce_fw->data; |
||
89 | for (;size > 0; --size, ++c) |
||
90 | if (strncmp(c, fw_version, strlen(fw_version)) == 0) |
||
91 | break; |
||
92 | |||
93 | if (size == 0) |
||
94 | return -EINVAL; |
||
95 | |||
96 | c += strlen(fw_version); |
||
97 | if (sscanf(c, "%2hhd.%2hhd.%2hhd]", &start, &mid, &end) != 3) |
||
98 | return -EINVAL; |
||
99 | |||
100 | /* search for feedback version */ |
||
101 | |||
102 | size = rdev->vce_fw->size - strlen(fb_version) - 3; |
||
103 | c = rdev->vce_fw->data; |
||
104 | for (;size > 0; --size, ++c) |
||
105 | if (strncmp(c, fb_version, strlen(fb_version)) == 0) |
||
106 | break; |
||
107 | |||
108 | if (size == 0) |
||
109 | return -EINVAL; |
||
110 | |||
111 | c += strlen(fb_version); |
||
112 | if (sscanf(c, "%2u]", &rdev->vce.fb_version) != 1) |
||
113 | return -EINVAL; |
||
114 | |||
115 | DRM_INFO("Found VCE firmware/feedback version %hhd.%hhd.%hhd / %d!\n", |
||
116 | start, mid, end, rdev->vce.fb_version); |
||
117 | |||
118 | rdev->vce.fw_version = (start << 24) | (mid << 16) | (end << 8); |
||
119 | |||
120 | /* we can only work with this fw version for now */ |
||
121 | if (rdev->vce.fw_version != ((40 << 24) | (2 << 16) | (2 << 8))) |
||
122 | return -EINVAL; |
||
123 | |||
124 | /* allocate firmware, stack and heap BO */ |
||
125 | |||
126 | size = RADEON_GPU_PAGE_ALIGN(rdev->vce_fw->size) + |
||
127 | RADEON_VCE_STACK_SIZE + RADEON_VCE_HEAP_SIZE; |
||
128 | r = radeon_bo_create(rdev, size, PAGE_SIZE, true, |
||
129 | RADEON_GEM_DOMAIN_VRAM, 0, NULL, &rdev->vce.vcpu_bo); |
||
130 | if (r) { |
||
131 | dev_err(rdev->dev, "(%d) failed to allocate VCE bo\n", r); |
||
132 | return r; |
||
133 | } |
||
134 | |||
135 | r = radeon_bo_reserve(rdev->vce.vcpu_bo, false); |
||
136 | if (r) { |
||
137 | radeon_bo_unref(&rdev->vce.vcpu_bo); |
||
138 | dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r); |
||
139 | return r; |
||
140 | } |
||
141 | |||
142 | r = radeon_bo_pin(rdev->vce.vcpu_bo, RADEON_GEM_DOMAIN_VRAM, |
||
143 | &rdev->vce.gpu_addr); |
||
144 | radeon_bo_unreserve(rdev->vce.vcpu_bo); |
||
145 | if (r) { |
||
146 | radeon_bo_unref(&rdev->vce.vcpu_bo); |
||
147 | dev_err(rdev->dev, "(%d) VCE bo pin failed\n", r); |
||
148 | return r; |
||
149 | } |
||
150 | |||
151 | for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) { |
||
152 | atomic_set(&rdev->vce.handles[i], 0); |
||
153 | rdev->vce.filp[i] = NULL; |
||
154 | } |
||
155 | |||
156 | return 0; |
||
157 | } |
||
158 | |||
159 | /** |
||
160 | * radeon_vce_fini - free memory |
||
161 | * |
||
162 | * @rdev: radeon_device pointer |
||
163 | * |
||
164 | * Last step on VCE teardown, free firmware memory |
||
165 | */ |
||
166 | void radeon_vce_fini(struct radeon_device *rdev) |
||
167 | { |
||
168 | if (rdev->vce.vcpu_bo == NULL) |
||
169 | return; |
||
170 | |||
171 | radeon_bo_unref(&rdev->vce.vcpu_bo); |
||
172 | |||
173 | release_firmware(rdev->vce_fw); |
||
174 | } |
||
175 | |||
176 | /** |
||
177 | * radeon_vce_suspend - unpin VCE fw memory |
||
178 | * |
||
179 | * @rdev: radeon_device pointer |
||
180 | * |
||
181 | */ |
||
182 | int radeon_vce_suspend(struct radeon_device *rdev) |
||
183 | { |
||
184 | int i; |
||
185 | |||
186 | if (rdev->vce.vcpu_bo == NULL) |
||
187 | return 0; |
||
188 | |||
189 | for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) |
||
190 | if (atomic_read(&rdev->vce.handles[i])) |
||
191 | break; |
||
192 | |||
193 | if (i == RADEON_MAX_VCE_HANDLES) |
||
194 | return 0; |
||
195 | |||
196 | /* TODO: suspending running encoding sessions isn't supported */ |
||
197 | return -EINVAL; |
||
198 | } |
||
199 | |||
200 | /** |
||
201 | * radeon_vce_resume - pin VCE fw memory |
||
202 | * |
||
203 | * @rdev: radeon_device pointer |
||
204 | * |
||
205 | */ |
||
206 | int radeon_vce_resume(struct radeon_device *rdev) |
||
207 | { |
||
208 | void *cpu_addr; |
||
209 | int r; |
||
210 | |||
211 | if (rdev->vce.vcpu_bo == NULL) |
||
212 | return -EINVAL; |
||
213 | |||
214 | r = radeon_bo_reserve(rdev->vce.vcpu_bo, false); |
||
215 | if (r) { |
||
216 | dev_err(rdev->dev, "(%d) failed to reserve VCE bo\n", r); |
||
217 | return r; |
||
218 | } |
||
219 | |||
220 | r = radeon_bo_kmap(rdev->vce.vcpu_bo, &cpu_addr); |
||
221 | if (r) { |
||
222 | radeon_bo_unreserve(rdev->vce.vcpu_bo); |
||
223 | dev_err(rdev->dev, "(%d) VCE map failed\n", r); |
||
224 | return r; |
||
225 | } |
||
226 | |||
227 | memcpy(cpu_addr, rdev->vce_fw->data, rdev->vce_fw->size); |
||
228 | |||
229 | radeon_bo_kunmap(rdev->vce.vcpu_bo); |
||
230 | |||
231 | radeon_bo_unreserve(rdev->vce.vcpu_bo); |
||
232 | |||
233 | return 0; |
||
234 | } |
||
235 | |||
236 | /** |
||
237 | * radeon_vce_idle_work_handler - power off VCE |
||
238 | * |
||
239 | * @work: pointer to work structure |
||
240 | * |
||
241 | * power of VCE when it's not used any more |
||
242 | */ |
||
243 | static void radeon_vce_idle_work_handler(struct work_struct *work) |
||
244 | { |
||
245 | struct radeon_device *rdev = |
||
246 | container_of(work, struct radeon_device, vce.idle_work.work); |
||
247 | |||
248 | if ((radeon_fence_count_emitted(rdev, TN_RING_TYPE_VCE1_INDEX) == 0) && |
||
249 | (radeon_fence_count_emitted(rdev, TN_RING_TYPE_VCE2_INDEX) == 0)) { |
||
250 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
||
251 | radeon_dpm_enable_vce(rdev, false); |
||
252 | } else { |
||
253 | radeon_set_vce_clocks(rdev, 0, 0); |
||
254 | } |
||
255 | } else { |
||
256 | schedule_delayed_work(&rdev->vce.idle_work, |
||
257 | msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS)); |
||
258 | } |
||
259 | } |
||
260 | |||
261 | /** |
||
262 | * radeon_vce_note_usage - power up VCE |
||
263 | * |
||
264 | * @rdev: radeon_device pointer |
||
265 | * |
||
266 | * Make sure VCE is powerd up when we want to use it |
||
267 | */ |
||
268 | void radeon_vce_note_usage(struct radeon_device *rdev) |
||
269 | { |
||
270 | bool streams_changed = false; |
||
271 | bool set_clocks = !cancel_delayed_work_sync(&rdev->vce.idle_work); |
||
272 | set_clocks &= schedule_delayed_work(&rdev->vce.idle_work, |
||
273 | msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS)); |
||
274 | |||
275 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
||
276 | /* XXX figure out if the streams changed */ |
||
277 | streams_changed = false; |
||
278 | } |
||
279 | |||
280 | if (set_clocks || streams_changed) { |
||
281 | if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) { |
||
282 | radeon_dpm_enable_vce(rdev, true); |
||
283 | } else { |
||
284 | radeon_set_vce_clocks(rdev, 53300, 40000); |
||
285 | } |
||
286 | } |
||
287 | } |
||
288 | |||
289 | /** |
||
290 | * radeon_vce_free_handles - free still open VCE handles |
||
291 | * |
||
292 | * @rdev: radeon_device pointer |
||
293 | * @filp: drm file pointer |
||
294 | * |
||
295 | * Close all VCE handles still open by this file pointer |
||
296 | */ |
||
297 | void radeon_vce_free_handles(struct radeon_device *rdev, struct drm_file *filp) |
||
298 | { |
||
299 | int i, r; |
||
300 | for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) { |
||
301 | uint32_t handle = atomic_read(&rdev->vce.handles[i]); |
||
302 | if (!handle || rdev->vce.filp[i] != filp) |
||
303 | continue; |
||
304 | |||
305 | radeon_vce_note_usage(rdev); |
||
306 | |||
307 | r = radeon_vce_get_destroy_msg(rdev, TN_RING_TYPE_VCE1_INDEX, |
||
308 | handle, NULL); |
||
309 | if (r) |
||
310 | DRM_ERROR("Error destroying VCE handle (%d)!\n", r); |
||
311 | |||
312 | rdev->vce.filp[i] = NULL; |
||
313 | atomic_set(&rdev->vce.handles[i], 0); |
||
314 | } |
||
315 | } |
||
316 | |||
317 | /** |
||
318 | * radeon_vce_get_create_msg - generate a VCE create msg |
||
319 | * |
||
320 | * @rdev: radeon_device pointer |
||
321 | * @ring: ring we should submit the msg to |
||
322 | * @handle: VCE session handle to use |
||
323 | * @fence: optional fence to return |
||
324 | * |
||
325 | * Open up a stream for HW test |
||
326 | */ |
||
327 | int radeon_vce_get_create_msg(struct radeon_device *rdev, int ring, |
||
328 | uint32_t handle, struct radeon_fence **fence) |
||
329 | { |
||
330 | const unsigned ib_size_dw = 1024; |
||
331 | struct radeon_ib ib; |
||
332 | uint64_t dummy; |
||
333 | int i, r; |
||
334 | |||
335 | r = radeon_ib_get(rdev, ring, &ib, NULL, ib_size_dw * 4); |
||
336 | if (r) { |
||
337 | DRM_ERROR("radeon: failed to get ib (%d).\n", r); |
||
338 | return r; |
||
339 | } |
||
340 | |||
341 | dummy = ib.gpu_addr + 1024; |
||
342 | |||
343 | /* stitch together an VCE create msg */ |
||
344 | ib.length_dw = 0; |
||
345 | ib.ptr[ib.length_dw++] = 0x0000000c; /* len */ |
||
346 | ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */ |
||
347 | ib.ptr[ib.length_dw++] = handle; |
||
348 | |||
349 | ib.ptr[ib.length_dw++] = 0x00000030; /* len */ |
||
350 | ib.ptr[ib.length_dw++] = 0x01000001; /* create cmd */ |
||
351 | ib.ptr[ib.length_dw++] = 0x00000000; |
||
352 | ib.ptr[ib.length_dw++] = 0x00000042; |
||
353 | ib.ptr[ib.length_dw++] = 0x0000000a; |
||
354 | ib.ptr[ib.length_dw++] = 0x00000001; |
||
355 | ib.ptr[ib.length_dw++] = 0x00000080; |
||
356 | ib.ptr[ib.length_dw++] = 0x00000060; |
||
357 | ib.ptr[ib.length_dw++] = 0x00000100; |
||
358 | ib.ptr[ib.length_dw++] = 0x00000100; |
||
359 | ib.ptr[ib.length_dw++] = 0x0000000c; |
||
360 | ib.ptr[ib.length_dw++] = 0x00000000; |
||
361 | |||
362 | ib.ptr[ib.length_dw++] = 0x00000014; /* len */ |
||
363 | ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */ |
||
364 | ib.ptr[ib.length_dw++] = upper_32_bits(dummy); |
||
365 | ib.ptr[ib.length_dw++] = dummy; |
||
366 | ib.ptr[ib.length_dw++] = 0x00000001; |
||
367 | |||
368 | for (i = ib.length_dw; i < ib_size_dw; ++i) |
||
369 | ib.ptr[i] = 0x0; |
||
370 | |||
371 | r = radeon_ib_schedule(rdev, &ib, NULL, false); |
||
372 | if (r) { |
||
373 | DRM_ERROR("radeon: failed to schedule ib (%d).\n", r); |
||
374 | } |
||
375 | |||
376 | if (fence) |
||
377 | *fence = radeon_fence_ref(ib.fence); |
||
378 | |||
379 | radeon_ib_free(rdev, &ib); |
||
380 | |||
381 | return r; |
||
382 | } |
||
383 | |||
384 | /** |
||
385 | * radeon_vce_get_destroy_msg - generate a VCE destroy msg |
||
386 | * |
||
387 | * @rdev: radeon_device pointer |
||
388 | * @ring: ring we should submit the msg to |
||
389 | * @handle: VCE session handle to use |
||
390 | * @fence: optional fence to return |
||
391 | * |
||
392 | * Close up a stream for HW test or if userspace failed to do so |
||
393 | */ |
||
394 | int radeon_vce_get_destroy_msg(struct radeon_device *rdev, int ring, |
||
395 | uint32_t handle, struct radeon_fence **fence) |
||
396 | { |
||
397 | const unsigned ib_size_dw = 1024; |
||
398 | struct radeon_ib ib; |
||
399 | uint64_t dummy; |
||
400 | int i, r; |
||
401 | |||
402 | r = radeon_ib_get(rdev, ring, &ib, NULL, ib_size_dw * 4); |
||
403 | if (r) { |
||
404 | DRM_ERROR("radeon: failed to get ib (%d).\n", r); |
||
405 | return r; |
||
406 | } |
||
407 | |||
408 | dummy = ib.gpu_addr + 1024; |
||
409 | |||
410 | /* stitch together an VCE destroy msg */ |
||
411 | ib.length_dw = 0; |
||
412 | ib.ptr[ib.length_dw++] = 0x0000000c; /* len */ |
||
413 | ib.ptr[ib.length_dw++] = 0x00000001; /* session cmd */ |
||
414 | ib.ptr[ib.length_dw++] = handle; |
||
415 | |||
416 | ib.ptr[ib.length_dw++] = 0x00000014; /* len */ |
||
417 | ib.ptr[ib.length_dw++] = 0x05000005; /* feedback buffer */ |
||
418 | ib.ptr[ib.length_dw++] = upper_32_bits(dummy); |
||
419 | ib.ptr[ib.length_dw++] = dummy; |
||
420 | ib.ptr[ib.length_dw++] = 0x00000001; |
||
421 | |||
422 | ib.ptr[ib.length_dw++] = 0x00000008; /* len */ |
||
423 | ib.ptr[ib.length_dw++] = 0x02000001; /* destroy cmd */ |
||
424 | |||
425 | for (i = ib.length_dw; i < ib_size_dw; ++i) |
||
426 | ib.ptr[i] = 0x0; |
||
427 | |||
428 | r = radeon_ib_schedule(rdev, &ib, NULL, false); |
||
429 | if (r) { |
||
430 | DRM_ERROR("radeon: failed to schedule ib (%d).\n", r); |
||
431 | } |
||
432 | |||
433 | if (fence) |
||
434 | *fence = radeon_fence_ref(ib.fence); |
||
435 | |||
436 | radeon_ib_free(rdev, &ib); |
||
437 | |||
438 | return r; |
||
439 | } |
||
440 | |||
441 | /** |
||
442 | * radeon_vce_cs_reloc - command submission relocation |
||
443 | * |
||
444 | * @p: parser context |
||
445 | * @lo: address of lower dword |
||
446 | * @hi: address of higher dword |
||
447 | * @size: size of checker for relocation buffer |
||
448 | * |
||
449 | * Patch relocation inside command stream with real buffer address |
||
450 | */ |
||
451 | int radeon_vce_cs_reloc(struct radeon_cs_parser *p, int lo, int hi, |
||
452 | unsigned size) |
||
453 | { |
||
454 | struct radeon_cs_chunk *relocs_chunk; |
||
455 | struct radeon_cs_reloc *reloc; |
||
456 | uint64_t start, end, offset; |
||
457 | unsigned idx; |
||
458 | |||
459 | relocs_chunk = &p->chunks[p->chunk_relocs_idx]; |
||
460 | offset = radeon_get_ib_value(p, lo); |
||
461 | idx = radeon_get_ib_value(p, hi); |
||
462 | |||
463 | if (idx >= relocs_chunk->length_dw) { |
||
464 | DRM_ERROR("Relocs at %d after relocations chunk end %d !\n", |
||
465 | idx, relocs_chunk->length_dw); |
||
466 | return -EINVAL; |
||
467 | } |
||
468 | |||
469 | reloc = p->relocs_ptr[(idx / 4)]; |
||
470 | start = reloc->gpu_offset; |
||
471 | end = start + radeon_bo_size(reloc->robj); |
||
472 | start += offset; |
||
473 | |||
474 | p->ib.ptr[lo] = start & 0xFFFFFFFF; |
||
475 | p->ib.ptr[hi] = start >> 32; |
||
476 | |||
477 | if (end <= start) { |
||
478 | DRM_ERROR("invalid reloc offset %llX!\n", offset); |
||
479 | return -EINVAL; |
||
480 | } |
||
481 | if ((end - start) < size) { |
||
482 | DRM_ERROR("buffer to small (%d / %d)!\n", |
||
483 | (unsigned)(end - start), size); |
||
484 | return -EINVAL; |
||
485 | } |
||
486 | |||
487 | return 0; |
||
488 | } |
||
489 | |||
490 | /** |
||
491 | * radeon_vce_validate_handle - validate stream handle |
||
492 | * |
||
493 | * @p: parser context |
||
494 | * @handle: handle to validate |
||
495 | * |
||
496 | * Validates the handle and return the found session index or -EINVAL |
||
497 | * we we don't have another free session index. |
||
498 | */ |
||
499 | int radeon_vce_validate_handle(struct radeon_cs_parser *p, uint32_t handle) |
||
500 | { |
||
501 | unsigned i; |
||
502 | |||
503 | /* validate the handle */ |
||
504 | for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) { |
||
505 | if (atomic_read(&p->rdev->vce.handles[i]) == handle) |
||
506 | return i; |
||
507 | } |
||
508 | |||
509 | /* handle not found try to alloc a new one */ |
||
510 | for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) { |
||
511 | if (!atomic_cmpxchg(&p->rdev->vce.handles[i], 0, handle)) { |
||
512 | p->rdev->vce.filp[i] = p->filp; |
||
513 | p->rdev->vce.img_size[i] = 0; |
||
514 | return i; |
||
515 | } |
||
516 | } |
||
517 | |||
518 | DRM_ERROR("No more free VCE handles!\n"); |
||
519 | return -EINVAL; |
||
520 | } |
||
521 | |||
522 | /** |
||
523 | * radeon_vce_cs_parse - parse and validate the command stream |
||
524 | * |
||
525 | * @p: parser context |
||
526 | * |
||
527 | */ |
||
528 | int radeon_vce_cs_parse(struct radeon_cs_parser *p) |
||
529 | { |
||
530 | int session_idx = -1; |
||
531 | bool destroyed = false; |
||
532 | uint32_t tmp, handle = 0; |
||
533 | uint32_t *size = &tmp; |
||
534 | int i, r; |
||
535 | |||
536 | while (p->idx < p->chunks[p->chunk_ib_idx].length_dw) { |
||
537 | uint32_t len = radeon_get_ib_value(p, p->idx); |
||
538 | uint32_t cmd = radeon_get_ib_value(p, p->idx + 1); |
||
539 | |||
540 | if ((len < 8) || (len & 3)) { |
||
541 | DRM_ERROR("invalid VCE command length (%d)!\n", len); |
||
542 | return -EINVAL; |
||
543 | } |
||
544 | |||
545 | if (destroyed) { |
||
546 | DRM_ERROR("No other command allowed after destroy!\n"); |
||
547 | return -EINVAL; |
||
548 | } |
||
549 | |||
550 | switch (cmd) { |
||
551 | case 0x00000001: // session |
||
552 | handle = radeon_get_ib_value(p, p->idx + 2); |
||
553 | session_idx = radeon_vce_validate_handle(p, handle); |
||
554 | if (session_idx < 0) |
||
555 | return session_idx; |
||
556 | size = &p->rdev->vce.img_size[session_idx]; |
||
557 | break; |
||
558 | |||
559 | case 0x00000002: // task info |
||
560 | break; |
||
561 | |||
562 | case 0x01000001: // create |
||
563 | *size = radeon_get_ib_value(p, p->idx + 8) * |
||
564 | radeon_get_ib_value(p, p->idx + 10) * |
||
565 | 8 * 3 / 2; |
||
566 | break; |
||
567 | |||
568 | case 0x04000001: // config extension |
||
569 | case 0x04000002: // pic control |
||
570 | case 0x04000005: // rate control |
||
571 | case 0x04000007: // motion estimation |
||
572 | case 0x04000008: // rdo |
||
573 | break; |
||
574 | |||
575 | case 0x03000001: // encode |
||
576 | r = radeon_vce_cs_reloc(p, p->idx + 10, p->idx + 9, |
||
577 | *size); |
||
578 | if (r) |
||
579 | return r; |
||
580 | |||
581 | r = radeon_vce_cs_reloc(p, p->idx + 12, p->idx + 11, |
||
582 | *size / 3); |
||
583 | if (r) |
||
584 | return r; |
||
585 | break; |
||
586 | |||
587 | case 0x02000001: // destroy |
||
588 | destroyed = true; |
||
589 | break; |
||
590 | |||
591 | case 0x05000001: // context buffer |
||
592 | r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2, |
||
593 | *size * 2); |
||
594 | if (r) |
||
595 | return r; |
||
596 | break; |
||
597 | |||
598 | case 0x05000004: // video bitstream buffer |
||
599 | tmp = radeon_get_ib_value(p, p->idx + 4); |
||
600 | r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2, |
||
601 | tmp); |
||
602 | if (r) |
||
603 | return r; |
||
604 | break; |
||
605 | |||
606 | case 0x05000005: // feedback buffer |
||
607 | r = radeon_vce_cs_reloc(p, p->idx + 3, p->idx + 2, |
||
608 | 4096); |
||
609 | if (r) |
||
610 | return r; |
||
611 | break; |
||
612 | |||
613 | default: |
||
614 | DRM_ERROR("invalid VCE command (0x%x)!\n", cmd); |
||
615 | return -EINVAL; |
||
616 | } |
||
617 | |||
618 | if (session_idx == -1) { |
||
619 | DRM_ERROR("no session command at start of IB\n"); |
||
620 | return -EINVAL; |
||
621 | } |
||
622 | |||
623 | p->idx += len / 4; |
||
624 | } |
||
625 | |||
626 | if (destroyed) { |
||
627 | /* IB contains a destroy msg, free the handle */ |
||
628 | for (i = 0; i < RADEON_MAX_VCE_HANDLES; ++i) |
||
629 | atomic_cmpxchg(&p->rdev->vce.handles[i], handle, 0); |
||
630 | } |
||
631 | |||
632 | return 0; |
||
633 | } |
||
634 | |||
635 | /** |
||
636 | * radeon_vce_semaphore_emit - emit a semaphore command |
||
637 | * |
||
638 | * @rdev: radeon_device pointer |
||
639 | * @ring: engine to use |
||
640 | * @semaphore: address of semaphore |
||
641 | * @emit_wait: true=emit wait, false=emit signal |
||
642 | * |
||
643 | */ |
||
644 | bool radeon_vce_semaphore_emit(struct radeon_device *rdev, |
||
645 | struct radeon_ring *ring, |
||
646 | struct radeon_semaphore *semaphore, |
||
647 | bool emit_wait) |
||
648 | { |
||
649 | uint64_t addr = semaphore->gpu_addr; |
||
650 | |||
651 | radeon_ring_write(ring, VCE_CMD_SEMAPHORE); |
||
652 | radeon_ring_write(ring, (addr >> 3) & 0x000FFFFF); |
||
653 | radeon_ring_write(ring, (addr >> 23) & 0x000FFFFF); |
||
654 | radeon_ring_write(ring, 0x01003000 | (emit_wait ? 1 : 0)); |
||
655 | if (!emit_wait) |
||
656 | radeon_ring_write(ring, VCE_CMD_END); |
||
657 | |||
658 | return true; |
||
659 | } |
||
660 | |||
661 | /** |
||
662 | * radeon_vce_ib_execute - execute indirect buffer |
||
663 | * |
||
664 | * @rdev: radeon_device pointer |
||
665 | * @ib: the IB to execute |
||
666 | * |
||
667 | */ |
||
668 | void radeon_vce_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib) |
||
669 | { |
||
670 | struct radeon_ring *ring = &rdev->ring[ib->ring]; |
||
671 | radeon_ring_write(ring, VCE_CMD_IB); |
||
672 | radeon_ring_write(ring, ib->gpu_addr); |
||
673 | radeon_ring_write(ring, upper_32_bits(ib->gpu_addr)); |
||
674 | radeon_ring_write(ring, ib->length_dw); |
||
675 | } |
||
676 | |||
677 | /** |
||
678 | * radeon_vce_fence_emit - add a fence command to the ring |
||
679 | * |
||
680 | * @rdev: radeon_device pointer |
||
681 | * @fence: the fence |
||
682 | * |
||
683 | */ |
||
684 | void radeon_vce_fence_emit(struct radeon_device *rdev, |
||
685 | struct radeon_fence *fence) |
||
686 | { |
||
687 | struct radeon_ring *ring = &rdev->ring[fence->ring]; |
||
688 | uint64_t addr = rdev->fence_drv[fence->ring].gpu_addr; |
||
689 | |||
690 | radeon_ring_write(ring, VCE_CMD_FENCE); |
||
691 | radeon_ring_write(ring, addr); |
||
692 | radeon_ring_write(ring, upper_32_bits(addr)); |
||
693 | radeon_ring_write(ring, fence->seq); |
||
694 | radeon_ring_write(ring, VCE_CMD_TRAP); |
||
695 | radeon_ring_write(ring, VCE_CMD_END); |
||
696 | } |
||
697 | |||
698 | /** |
||
699 | * radeon_vce_ring_test - test if VCE ring is working |
||
700 | * |
||
701 | * @rdev: radeon_device pointer |
||
702 | * @ring: the engine to test on |
||
703 | * |
||
704 | */ |
||
705 | int radeon_vce_ring_test(struct radeon_device *rdev, struct radeon_ring *ring) |
||
706 | { |
||
707 | uint32_t rptr = vce_v1_0_get_rptr(rdev, ring); |
||
708 | unsigned i; |
||
709 | int r; |
||
710 | |||
711 | r = radeon_ring_lock(rdev, ring, 16); |
||
712 | if (r) { |
||
713 | DRM_ERROR("radeon: vce failed to lock ring %d (%d).\n", |
||
714 | ring->idx, r); |
||
715 | return r; |
||
716 | } |
||
717 | radeon_ring_write(ring, VCE_CMD_END); |
||
718 | radeon_ring_unlock_commit(rdev, ring, false); |
||
719 | |||
720 | for (i = 0; i < rdev->usec_timeout; i++) { |
||
721 | if (vce_v1_0_get_rptr(rdev, ring) != rptr) |
||
722 | break; |
||
723 | DRM_UDELAY(1); |
||
724 | } |
||
725 | |||
726 | if (i < rdev->usec_timeout) { |
||
727 | DRM_INFO("ring test on %d succeeded in %d usecs\n", |
||
728 | ring->idx, i); |
||
729 | } else { |
||
730 | DRM_ERROR("radeon: ring %d test failed\n", |
||
731 | ring->idx); |
||
732 | r = -ETIMEDOUT; |
||
733 | } |
||
734 | |||
735 | return r; |
||
736 | } |
||
737 | |||
738 | /** |
||
739 | * radeon_vce_ib_test - test if VCE IBs are working |
||
740 | * |
||
741 | * @rdev: radeon_device pointer |
||
742 | * @ring: the engine to test on |
||
743 | * |
||
744 | */ |
||
745 | int radeon_vce_ib_test(struct radeon_device *rdev, struct radeon_ring *ring) |
||
746 | { |
||
747 | struct radeon_fence *fence = NULL; |
||
748 | int r; |
||
749 | |||
750 | r = radeon_vce_get_create_msg(rdev, ring->idx, 1, NULL); |
||
751 | if (r) { |
||
752 | DRM_ERROR("radeon: failed to get create msg (%d).\n", r); |
||
753 | goto error; |
||
754 | } |
||
755 | |||
756 | r = radeon_vce_get_destroy_msg(rdev, ring->idx, 1, &fence); |
||
757 | if (r) { |
||
758 | DRM_ERROR("radeon: failed to get destroy ib (%d).\n", r); |
||
759 | goto error; |
||
760 | } |
||
761 | |||
762 | r = radeon_fence_wait(fence, false); |
||
763 | if (r) { |
||
764 | DRM_ERROR("radeon: fence wait failed (%d).\n", r); |
||
765 | } else { |
||
766 | DRM_INFO("ib test on ring %d succeeded\n", ring->idx); |
||
767 | } |
||
768 | error: |
||
769 | radeon_fence_unref(&fence); |
||
770 | return r; |
||
771 | }>>>>>>>>>=>>>>>>><>><>><>><>><>><> |