]>
Commit | Line | Data |
---|---|---|
d38ceaf9 AD |
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 <christian.koenig@amd.com> | |
26 | */ | |
27 | ||
28 | #include <linux/firmware.h> | |
29 | #include <linux/module.h> | |
30 | #include <drm/drmP.h> | |
31 | #include <drm/drm.h> | |
32 | ||
33 | #include "amdgpu.h" | |
34 | #include "amdgpu_pm.h" | |
35 | #include "amdgpu_vce.h" | |
36 | #include "cikd.h" | |
37 | ||
38 | /* 1 second timeout */ | |
39 | #define VCE_IDLE_TIMEOUT_MS 1000 | |
40 | ||
41 | /* Firmware Names */ | |
42 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
43 | #define FIRMWARE_BONAIRE "radeon/bonaire_vce.bin" | |
44 | #define FIRMWARE_KABINI "radeon/kabini_vce.bin" | |
45 | #define FIRMWARE_KAVERI "radeon/kaveri_vce.bin" | |
46 | #define FIRMWARE_HAWAII "radeon/hawaii_vce.bin" | |
47 | #define FIRMWARE_MULLINS "radeon/mullins_vce.bin" | |
48 | #endif | |
c65444fe JZ |
49 | #define FIRMWARE_TONGA "amdgpu/tonga_vce.bin" |
50 | #define FIRMWARE_CARRIZO "amdgpu/carrizo_vce.bin" | |
188a9bcd | 51 | #define FIRMWARE_FIJI "amdgpu/fiji_vce.bin" |
cfaba566 | 52 | #define FIRMWARE_STONEY "amdgpu/stoney_vce.bin" |
d38ceaf9 AD |
53 | |
54 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
55 | MODULE_FIRMWARE(FIRMWARE_BONAIRE); | |
56 | MODULE_FIRMWARE(FIRMWARE_KABINI); | |
57 | MODULE_FIRMWARE(FIRMWARE_KAVERI); | |
58 | MODULE_FIRMWARE(FIRMWARE_HAWAII); | |
59 | MODULE_FIRMWARE(FIRMWARE_MULLINS); | |
60 | #endif | |
61 | MODULE_FIRMWARE(FIRMWARE_TONGA); | |
62 | MODULE_FIRMWARE(FIRMWARE_CARRIZO); | |
188a9bcd | 63 | MODULE_FIRMWARE(FIRMWARE_FIJI); |
cfaba566 | 64 | MODULE_FIRMWARE(FIRMWARE_STONEY); |
d38ceaf9 AD |
65 | |
66 | static void amdgpu_vce_idle_work_handler(struct work_struct *work); | |
67 | ||
68 | /** | |
69 | * amdgpu_vce_init - allocate memory, load vce firmware | |
70 | * | |
71 | * @adev: amdgpu_device pointer | |
72 | * | |
73 | * First step to get VCE online, allocate memory and load the firmware | |
74 | */ | |
e9822622 | 75 | int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size) |
d38ceaf9 | 76 | { |
d38ceaf9 AD |
77 | const char *fw_name; |
78 | const struct common_firmware_header *hdr; | |
79 | unsigned ucode_version, version_major, version_minor, binary_id; | |
80 | int i, r; | |
81 | ||
82 | INIT_DELAYED_WORK(&adev->vce.idle_work, amdgpu_vce_idle_work_handler); | |
83 | ||
84 | switch (adev->asic_type) { | |
85 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
86 | case CHIP_BONAIRE: | |
87 | fw_name = FIRMWARE_BONAIRE; | |
88 | break; | |
89 | case CHIP_KAVERI: | |
90 | fw_name = FIRMWARE_KAVERI; | |
91 | break; | |
92 | case CHIP_KABINI: | |
93 | fw_name = FIRMWARE_KABINI; | |
94 | break; | |
95 | case CHIP_HAWAII: | |
96 | fw_name = FIRMWARE_HAWAII; | |
97 | break; | |
98 | case CHIP_MULLINS: | |
99 | fw_name = FIRMWARE_MULLINS; | |
100 | break; | |
101 | #endif | |
102 | case CHIP_TONGA: | |
103 | fw_name = FIRMWARE_TONGA; | |
104 | break; | |
105 | case CHIP_CARRIZO: | |
106 | fw_name = FIRMWARE_CARRIZO; | |
107 | break; | |
188a9bcd AD |
108 | case CHIP_FIJI: |
109 | fw_name = FIRMWARE_FIJI; | |
110 | break; | |
cfaba566 SL |
111 | case CHIP_STONEY: |
112 | fw_name = FIRMWARE_STONEY; | |
113 | break; | |
d38ceaf9 AD |
114 | |
115 | default: | |
116 | return -EINVAL; | |
117 | } | |
118 | ||
119 | r = request_firmware(&adev->vce.fw, fw_name, adev->dev); | |
120 | if (r) { | |
121 | dev_err(adev->dev, "amdgpu_vce: Can't load firmware \"%s\"\n", | |
122 | fw_name); | |
123 | return r; | |
124 | } | |
125 | ||
126 | r = amdgpu_ucode_validate(adev->vce.fw); | |
127 | if (r) { | |
128 | dev_err(adev->dev, "amdgpu_vce: Can't validate firmware \"%s\"\n", | |
129 | fw_name); | |
130 | release_firmware(adev->vce.fw); | |
131 | adev->vce.fw = NULL; | |
132 | return r; | |
133 | } | |
134 | ||
135 | hdr = (const struct common_firmware_header *)adev->vce.fw->data; | |
136 | ||
137 | ucode_version = le32_to_cpu(hdr->ucode_version); | |
138 | version_major = (ucode_version >> 20) & 0xfff; | |
139 | version_minor = (ucode_version >> 8) & 0xfff; | |
140 | binary_id = ucode_version & 0xff; | |
141 | DRM_INFO("Found VCE firmware Version: %hhd.%hhd Binary ID: %hhd\n", | |
142 | version_major, version_minor, binary_id); | |
143 | adev->vce.fw_version = ((version_major << 24) | (version_minor << 16) | | |
144 | (binary_id << 8)); | |
145 | ||
146 | /* allocate firmware, stack and heap BO */ | |
147 | ||
d38ceaf9 | 148 | r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, |
857d913d AD |
149 | AMDGPU_GEM_DOMAIN_VRAM, |
150 | AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED, | |
72d7668b | 151 | NULL, NULL, &adev->vce.vcpu_bo); |
d38ceaf9 AD |
152 | if (r) { |
153 | dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r); | |
154 | return r; | |
155 | } | |
156 | ||
157 | r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false); | |
158 | if (r) { | |
159 | amdgpu_bo_unref(&adev->vce.vcpu_bo); | |
160 | dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r); | |
161 | return r; | |
162 | } | |
163 | ||
164 | r = amdgpu_bo_pin(adev->vce.vcpu_bo, AMDGPU_GEM_DOMAIN_VRAM, | |
165 | &adev->vce.gpu_addr); | |
166 | amdgpu_bo_unreserve(adev->vce.vcpu_bo); | |
167 | if (r) { | |
168 | amdgpu_bo_unref(&adev->vce.vcpu_bo); | |
169 | dev_err(adev->dev, "(%d) VCE bo pin failed\n", r); | |
170 | return r; | |
171 | } | |
172 | ||
173 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { | |
174 | atomic_set(&adev->vce.handles[i], 0); | |
175 | adev->vce.filp[i] = NULL; | |
176 | } | |
177 | ||
178 | return 0; | |
179 | } | |
180 | ||
181 | /** | |
182 | * amdgpu_vce_fini - free memory | |
183 | * | |
184 | * @adev: amdgpu_device pointer | |
185 | * | |
186 | * Last step on VCE teardown, free firmware memory | |
187 | */ | |
188 | int amdgpu_vce_sw_fini(struct amdgpu_device *adev) | |
189 | { | |
190 | if (adev->vce.vcpu_bo == NULL) | |
191 | return 0; | |
192 | ||
193 | amdgpu_bo_unref(&adev->vce.vcpu_bo); | |
194 | ||
195 | amdgpu_ring_fini(&adev->vce.ring[0]); | |
196 | amdgpu_ring_fini(&adev->vce.ring[1]); | |
197 | ||
198 | release_firmware(adev->vce.fw); | |
199 | ||
200 | return 0; | |
201 | } | |
202 | ||
203 | /** | |
204 | * amdgpu_vce_suspend - unpin VCE fw memory | |
205 | * | |
206 | * @adev: amdgpu_device pointer | |
207 | * | |
208 | */ | |
209 | int amdgpu_vce_suspend(struct amdgpu_device *adev) | |
210 | { | |
211 | int i; | |
212 | ||
213 | if (adev->vce.vcpu_bo == NULL) | |
214 | return 0; | |
215 | ||
216 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) | |
217 | if (atomic_read(&adev->vce.handles[i])) | |
218 | break; | |
219 | ||
220 | if (i == AMDGPU_MAX_VCE_HANDLES) | |
221 | return 0; | |
222 | ||
223 | /* TODO: suspending running encoding sessions isn't supported */ | |
224 | return -EINVAL; | |
225 | } | |
226 | ||
227 | /** | |
228 | * amdgpu_vce_resume - pin VCE fw memory | |
229 | * | |
230 | * @adev: amdgpu_device pointer | |
231 | * | |
232 | */ | |
233 | int amdgpu_vce_resume(struct amdgpu_device *adev) | |
234 | { | |
235 | void *cpu_addr; | |
236 | const struct common_firmware_header *hdr; | |
237 | unsigned offset; | |
238 | int r; | |
239 | ||
240 | if (adev->vce.vcpu_bo == NULL) | |
241 | return -EINVAL; | |
242 | ||
243 | r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false); | |
244 | if (r) { | |
245 | dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r); | |
246 | return r; | |
247 | } | |
248 | ||
249 | r = amdgpu_bo_kmap(adev->vce.vcpu_bo, &cpu_addr); | |
250 | if (r) { | |
251 | amdgpu_bo_unreserve(adev->vce.vcpu_bo); | |
252 | dev_err(adev->dev, "(%d) VCE map failed\n", r); | |
253 | return r; | |
254 | } | |
255 | ||
256 | hdr = (const struct common_firmware_header *)adev->vce.fw->data; | |
257 | offset = le32_to_cpu(hdr->ucode_array_offset_bytes); | |
258 | memcpy(cpu_addr, (adev->vce.fw->data) + offset, | |
259 | (adev->vce.fw->size) - offset); | |
260 | ||
261 | amdgpu_bo_kunmap(adev->vce.vcpu_bo); | |
262 | ||
263 | amdgpu_bo_unreserve(adev->vce.vcpu_bo); | |
264 | ||
265 | return 0; | |
266 | } | |
267 | ||
268 | /** | |
269 | * amdgpu_vce_idle_work_handler - power off VCE | |
270 | * | |
271 | * @work: pointer to work structure | |
272 | * | |
273 | * power of VCE when it's not used any more | |
274 | */ | |
275 | static void amdgpu_vce_idle_work_handler(struct work_struct *work) | |
276 | { | |
277 | struct amdgpu_device *adev = | |
278 | container_of(work, struct amdgpu_device, vce.idle_work.work); | |
279 | ||
280 | if ((amdgpu_fence_count_emitted(&adev->vce.ring[0]) == 0) && | |
281 | (amdgpu_fence_count_emitted(&adev->vce.ring[1]) == 0)) { | |
282 | if (adev->pm.dpm_enabled) { | |
283 | amdgpu_dpm_enable_vce(adev, false); | |
284 | } else { | |
285 | amdgpu_asic_set_vce_clocks(adev, 0, 0); | |
286 | } | |
287 | } else { | |
288 | schedule_delayed_work(&adev->vce.idle_work, | |
289 | msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS)); | |
290 | } | |
291 | } | |
292 | ||
293 | /** | |
294 | * amdgpu_vce_note_usage - power up VCE | |
295 | * | |
296 | * @adev: amdgpu_device pointer | |
297 | * | |
298 | * Make sure VCE is powerd up when we want to use it | |
299 | */ | |
300 | static void amdgpu_vce_note_usage(struct amdgpu_device *adev) | |
301 | { | |
302 | bool streams_changed = false; | |
303 | bool set_clocks = !cancel_delayed_work_sync(&adev->vce.idle_work); | |
304 | set_clocks &= schedule_delayed_work(&adev->vce.idle_work, | |
305 | msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS)); | |
306 | ||
307 | if (adev->pm.dpm_enabled) { | |
308 | /* XXX figure out if the streams changed */ | |
309 | streams_changed = false; | |
310 | } | |
311 | ||
312 | if (set_clocks || streams_changed) { | |
313 | if (adev->pm.dpm_enabled) { | |
314 | amdgpu_dpm_enable_vce(adev, true); | |
315 | } else { | |
316 | amdgpu_asic_set_vce_clocks(adev, 53300, 40000); | |
317 | } | |
318 | } | |
319 | } | |
320 | ||
321 | /** | |
322 | * amdgpu_vce_free_handles - free still open VCE handles | |
323 | * | |
324 | * @adev: amdgpu_device pointer | |
325 | * @filp: drm file pointer | |
326 | * | |
327 | * Close all VCE handles still open by this file pointer | |
328 | */ | |
329 | void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp) | |
330 | { | |
331 | struct amdgpu_ring *ring = &adev->vce.ring[0]; | |
332 | int i, r; | |
333 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { | |
334 | uint32_t handle = atomic_read(&adev->vce.handles[i]); | |
335 | if (!handle || adev->vce.filp[i] != filp) | |
336 | continue; | |
337 | ||
338 | amdgpu_vce_note_usage(adev); | |
339 | ||
340 | r = amdgpu_vce_get_destroy_msg(ring, handle, NULL); | |
341 | if (r) | |
342 | DRM_ERROR("Error destroying VCE handle (%d)!\n", r); | |
343 | ||
344 | adev->vce.filp[i] = NULL; | |
345 | atomic_set(&adev->vce.handles[i], 0); | |
346 | } | |
347 | } | |
348 | ||
8128765c | 349 | static int amdgpu_vce_free_job( |
4c7eb91c | 350 | struct amdgpu_job *job) |
8128765c | 351 | { |
4c7eb91c JZ |
352 | amdgpu_ib_free(job->adev, job->ibs); |
353 | kfree(job->ibs); | |
8128765c CZ |
354 | return 0; |
355 | } | |
356 | ||
d38ceaf9 AD |
357 | /** |
358 | * amdgpu_vce_get_create_msg - generate a VCE create msg | |
359 | * | |
360 | * @adev: amdgpu_device pointer | |
361 | * @ring: ring we should submit the msg to | |
362 | * @handle: VCE session handle to use | |
363 | * @fence: optional fence to return | |
364 | * | |
365 | * Open up a stream for HW test | |
366 | */ | |
367 | int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, | |
ed40bfb8 | 368 | struct fence **fence) |
d38ceaf9 AD |
369 | { |
370 | const unsigned ib_size_dw = 1024; | |
8128765c | 371 | struct amdgpu_ib *ib = NULL; |
1763552e | 372 | struct fence *f = NULL; |
8128765c | 373 | struct amdgpu_device *adev = ring->adev; |
d38ceaf9 AD |
374 | uint64_t dummy; |
375 | int i, r; | |
376 | ||
8128765c CZ |
377 | ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); |
378 | if (!ib) | |
379 | return -ENOMEM; | |
b07c60c0 | 380 | r = amdgpu_ib_get(adev, NULL, ib_size_dw * 4, ib); |
d38ceaf9 AD |
381 | if (r) { |
382 | DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); | |
8128765c | 383 | kfree(ib); |
d38ceaf9 AD |
384 | return r; |
385 | } | |
386 | ||
8128765c | 387 | dummy = ib->gpu_addr + 1024; |
d38ceaf9 AD |
388 | |
389 | /* stitch together an VCE create msg */ | |
8128765c CZ |
390 | ib->length_dw = 0; |
391 | ib->ptr[ib->length_dw++] = 0x0000000c; /* len */ | |
392 | ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */ | |
393 | ib->ptr[ib->length_dw++] = handle; | |
394 | ||
d66f8e48 LL |
395 | if ((ring->adev->vce.fw_version >> 24) >= 52) |
396 | ib->ptr[ib->length_dw++] = 0x00000040; /* len */ | |
397 | else | |
398 | ib->ptr[ib->length_dw++] = 0x00000030; /* len */ | |
8128765c CZ |
399 | ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */ |
400 | ib->ptr[ib->length_dw++] = 0x00000000; | |
401 | ib->ptr[ib->length_dw++] = 0x00000042; | |
402 | ib->ptr[ib->length_dw++] = 0x0000000a; | |
403 | ib->ptr[ib->length_dw++] = 0x00000001; | |
404 | ib->ptr[ib->length_dw++] = 0x00000080; | |
405 | ib->ptr[ib->length_dw++] = 0x00000060; | |
406 | ib->ptr[ib->length_dw++] = 0x00000100; | |
407 | ib->ptr[ib->length_dw++] = 0x00000100; | |
408 | ib->ptr[ib->length_dw++] = 0x0000000c; | |
409 | ib->ptr[ib->length_dw++] = 0x00000000; | |
d66f8e48 LL |
410 | if ((ring->adev->vce.fw_version >> 24) >= 52) { |
411 | ib->ptr[ib->length_dw++] = 0x00000000; | |
412 | ib->ptr[ib->length_dw++] = 0x00000000; | |
413 | ib->ptr[ib->length_dw++] = 0x00000000; | |
414 | ib->ptr[ib->length_dw++] = 0x00000000; | |
415 | } | |
8128765c CZ |
416 | |
417 | ib->ptr[ib->length_dw++] = 0x00000014; /* len */ | |
418 | ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */ | |
419 | ib->ptr[ib->length_dw++] = upper_32_bits(dummy); | |
420 | ib->ptr[ib->length_dw++] = dummy; | |
421 | ib->ptr[ib->length_dw++] = 0x00000001; | |
422 | ||
423 | for (i = ib->length_dw; i < ib_size_dw; ++i) | |
424 | ib->ptr[i] = 0x0; | |
425 | ||
426 | r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, | |
427 | &amdgpu_vce_free_job, | |
1763552e CZ |
428 | AMDGPU_FENCE_OWNER_UNDEFINED, |
429 | &f); | |
8128765c CZ |
430 | if (r) |
431 | goto err; | |
d38ceaf9 | 432 | if (fence) |
1763552e | 433 | *fence = fence_get(f); |
281b4223 | 434 | fence_put(f); |
cadf97b1 | 435 | return 0; |
8128765c CZ |
436 | err: |
437 | amdgpu_ib_free(adev, ib); | |
438 | kfree(ib); | |
d38ceaf9 AD |
439 | return r; |
440 | } | |
441 | ||
442 | /** | |
443 | * amdgpu_vce_get_destroy_msg - generate a VCE destroy msg | |
444 | * | |
445 | * @adev: amdgpu_device pointer | |
446 | * @ring: ring we should submit the msg to | |
447 | * @handle: VCE session handle to use | |
448 | * @fence: optional fence to return | |
449 | * | |
450 | * Close up a stream for HW test or if userspace failed to do so | |
451 | */ | |
452 | int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, | |
ed40bfb8 | 453 | struct fence **fence) |
d38ceaf9 AD |
454 | { |
455 | const unsigned ib_size_dw = 1024; | |
8128765c | 456 | struct amdgpu_ib *ib = NULL; |
1763552e | 457 | struct fence *f = NULL; |
8128765c | 458 | struct amdgpu_device *adev = ring->adev; |
d38ceaf9 AD |
459 | uint64_t dummy; |
460 | int i, r; | |
461 | ||
8128765c CZ |
462 | ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); |
463 | if (!ib) | |
464 | return -ENOMEM; | |
465 | ||
b07c60c0 | 466 | r = amdgpu_ib_get(adev, NULL, ib_size_dw * 4, ib); |
d38ceaf9 | 467 | if (r) { |
8128765c | 468 | kfree(ib); |
d38ceaf9 AD |
469 | DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); |
470 | return r; | |
471 | } | |
472 | ||
8128765c | 473 | dummy = ib->gpu_addr + 1024; |
d38ceaf9 AD |
474 | |
475 | /* stitch together an VCE destroy msg */ | |
8128765c CZ |
476 | ib->length_dw = 0; |
477 | ib->ptr[ib->length_dw++] = 0x0000000c; /* len */ | |
478 | ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */ | |
479 | ib->ptr[ib->length_dw++] = handle; | |
480 | ||
481 | ib->ptr[ib->length_dw++] = 0x00000014; /* len */ | |
482 | ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */ | |
483 | ib->ptr[ib->length_dw++] = upper_32_bits(dummy); | |
484 | ib->ptr[ib->length_dw++] = dummy; | |
485 | ib->ptr[ib->length_dw++] = 0x00000001; | |
486 | ||
487 | ib->ptr[ib->length_dw++] = 0x00000008; /* len */ | |
488 | ib->ptr[ib->length_dw++] = 0x02000001; /* destroy cmd */ | |
489 | ||
490 | for (i = ib->length_dw; i < ib_size_dw; ++i) | |
491 | ib->ptr[i] = 0x0; | |
492 | r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, | |
493 | &amdgpu_vce_free_job, | |
1763552e CZ |
494 | AMDGPU_FENCE_OWNER_UNDEFINED, |
495 | &f); | |
8128765c CZ |
496 | if (r) |
497 | goto err; | |
d38ceaf9 | 498 | if (fence) |
1763552e | 499 | *fence = fence_get(f); |
281b4223 | 500 | fence_put(f); |
cadf97b1 | 501 | return 0; |
8128765c CZ |
502 | err: |
503 | amdgpu_ib_free(adev, ib); | |
504 | kfree(ib); | |
d38ceaf9 AD |
505 | return r; |
506 | } | |
507 | ||
508 | /** | |
509 | * amdgpu_vce_cs_reloc - command submission relocation | |
510 | * | |
511 | * @p: parser context | |
512 | * @lo: address of lower dword | |
513 | * @hi: address of higher dword | |
f1689ec1 | 514 | * @size: minimum size |
d38ceaf9 AD |
515 | * |
516 | * Patch relocation inside command stream with real buffer address | |
517 | */ | |
f1689ec1 | 518 | static int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx, |
dc78330a | 519 | int lo, int hi, unsigned size, uint32_t index) |
d38ceaf9 AD |
520 | { |
521 | struct amdgpu_bo_va_mapping *mapping; | |
d38ceaf9 AD |
522 | struct amdgpu_bo *bo; |
523 | uint64_t addr; | |
524 | ||
dc78330a CK |
525 | if (index == 0xffffffff) |
526 | index = 0; | |
527 | ||
d38ceaf9 AD |
528 | addr = ((uint64_t)amdgpu_get_ib_value(p, ib_idx, lo)) | |
529 | ((uint64_t)amdgpu_get_ib_value(p, ib_idx, hi)) << 32; | |
dc78330a | 530 | addr += ((uint64_t)size) * ((uint64_t)index); |
d38ceaf9 AD |
531 | |
532 | mapping = amdgpu_cs_find_mapping(p, addr, &bo); | |
533 | if (mapping == NULL) { | |
dc78330a CK |
534 | DRM_ERROR("Can't find BO for addr 0x%010Lx %d %d %d %d\n", |
535 | addr, lo, hi, size, index); | |
d38ceaf9 AD |
536 | return -EINVAL; |
537 | } | |
538 | ||
f1689ec1 CK |
539 | if ((addr + (uint64_t)size) > |
540 | ((uint64_t)mapping->it.last + 1) * AMDGPU_GPU_PAGE_SIZE) { | |
541 | DRM_ERROR("BO to small for addr 0x%010Lx %d %d\n", | |
542 | addr, lo, hi); | |
543 | return -EINVAL; | |
544 | } | |
545 | ||
d38ceaf9 AD |
546 | addr -= ((uint64_t)mapping->it.start) * AMDGPU_GPU_PAGE_SIZE; |
547 | addr += amdgpu_bo_gpu_offset(bo); | |
dc78330a | 548 | addr -= ((uint64_t)size) * ((uint64_t)index); |
d38ceaf9 | 549 | |
7270f839 CK |
550 | amdgpu_set_ib_value(p, ib_idx, lo, lower_32_bits(addr)); |
551 | amdgpu_set_ib_value(p, ib_idx, hi, upper_32_bits(addr)); | |
d38ceaf9 AD |
552 | |
553 | return 0; | |
554 | } | |
555 | ||
f1689ec1 CK |
556 | /** |
557 | * amdgpu_vce_validate_handle - validate stream handle | |
558 | * | |
559 | * @p: parser context | |
560 | * @handle: handle to validate | |
2f4b9368 | 561 | * @allocated: allocated a new handle? |
f1689ec1 CK |
562 | * |
563 | * Validates the handle and return the found session index or -EINVAL | |
564 | * we we don't have another free session index. | |
565 | */ | |
566 | static int amdgpu_vce_validate_handle(struct amdgpu_cs_parser *p, | |
2f4b9368 | 567 | uint32_t handle, bool *allocated) |
f1689ec1 CK |
568 | { |
569 | unsigned i; | |
570 | ||
2f4b9368 CK |
571 | *allocated = false; |
572 | ||
f1689ec1 CK |
573 | /* validate the handle */ |
574 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { | |
2f4b9368 CK |
575 | if (atomic_read(&p->adev->vce.handles[i]) == handle) { |
576 | if (p->adev->vce.filp[i] != p->filp) { | |
577 | DRM_ERROR("VCE handle collision detected!\n"); | |
578 | return -EINVAL; | |
579 | } | |
f1689ec1 | 580 | return i; |
2f4b9368 | 581 | } |
f1689ec1 CK |
582 | } |
583 | ||
584 | /* handle not found try to alloc a new one */ | |
585 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { | |
586 | if (!atomic_cmpxchg(&p->adev->vce.handles[i], 0, handle)) { | |
587 | p->adev->vce.filp[i] = p->filp; | |
588 | p->adev->vce.img_size[i] = 0; | |
2f4b9368 | 589 | *allocated = true; |
f1689ec1 CK |
590 | return i; |
591 | } | |
592 | } | |
593 | ||
594 | DRM_ERROR("No more free VCE handles!\n"); | |
595 | return -EINVAL; | |
596 | } | |
597 | ||
d38ceaf9 AD |
598 | /** |
599 | * amdgpu_vce_cs_parse - parse and validate the command stream | |
600 | * | |
601 | * @p: parser context | |
602 | * | |
603 | */ | |
604 | int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx) | |
605 | { | |
50838c8c | 606 | struct amdgpu_ib *ib = &p->job->ibs[ib_idx]; |
dc78330a | 607 | unsigned fb_idx = 0, bs_idx = 0; |
f1689ec1 CK |
608 | int session_idx = -1; |
609 | bool destroyed = false; | |
2f4b9368 CK |
610 | bool created = false; |
611 | bool allocated = false; | |
f1689ec1 CK |
612 | uint32_t tmp, handle = 0; |
613 | uint32_t *size = &tmp; | |
2f4b9368 | 614 | int i, r = 0, idx = 0; |
d38ceaf9 AD |
615 | |
616 | amdgpu_vce_note_usage(p->adev); | |
617 | ||
618 | while (idx < ib->length_dw) { | |
619 | uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx); | |
620 | uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1); | |
621 | ||
622 | if ((len < 8) || (len & 3)) { | |
623 | DRM_ERROR("invalid VCE command length (%d)!\n", len); | |
2f4b9368 CK |
624 | r = -EINVAL; |
625 | goto out; | |
d38ceaf9 AD |
626 | } |
627 | ||
f1689ec1 CK |
628 | if (destroyed) { |
629 | DRM_ERROR("No other command allowed after destroy!\n"); | |
2f4b9368 CK |
630 | r = -EINVAL; |
631 | goto out; | |
f1689ec1 CK |
632 | } |
633 | ||
d38ceaf9 AD |
634 | switch (cmd) { |
635 | case 0x00000001: // session | |
636 | handle = amdgpu_get_ib_value(p, ib_idx, idx + 2); | |
2f4b9368 CK |
637 | session_idx = amdgpu_vce_validate_handle(p, handle, |
638 | &allocated); | |
f1689ec1 CK |
639 | if (session_idx < 0) |
640 | return session_idx; | |
641 | size = &p->adev->vce.img_size[session_idx]; | |
d38ceaf9 AD |
642 | break; |
643 | ||
644 | case 0x00000002: // task info | |
dc78330a CK |
645 | fb_idx = amdgpu_get_ib_value(p, ib_idx, idx + 6); |
646 | bs_idx = amdgpu_get_ib_value(p, ib_idx, idx + 7); | |
f1689ec1 CK |
647 | break; |
648 | ||
d38ceaf9 | 649 | case 0x01000001: // create |
2f4b9368 CK |
650 | created = true; |
651 | if (!allocated) { | |
652 | DRM_ERROR("Handle already in use!\n"); | |
653 | r = -EINVAL; | |
654 | goto out; | |
655 | } | |
656 | ||
f1689ec1 CK |
657 | *size = amdgpu_get_ib_value(p, ib_idx, idx + 8) * |
658 | amdgpu_get_ib_value(p, ib_idx, idx + 10) * | |
659 | 8 * 3 / 2; | |
660 | break; | |
661 | ||
d38ceaf9 AD |
662 | case 0x04000001: // config extension |
663 | case 0x04000002: // pic control | |
664 | case 0x04000005: // rate control | |
665 | case 0x04000007: // motion estimation | |
666 | case 0x04000008: // rdo | |
667 | case 0x04000009: // vui | |
668 | case 0x05000002: // auxiliary buffer | |
669 | break; | |
670 | ||
671 | case 0x03000001: // encode | |
f1689ec1 | 672 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 10, idx + 9, |
dc78330a | 673 | *size, 0); |
d38ceaf9 | 674 | if (r) |
2f4b9368 | 675 | goto out; |
d38ceaf9 | 676 | |
f1689ec1 | 677 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 12, idx + 11, |
dc78330a | 678 | *size / 3, 0); |
d38ceaf9 | 679 | if (r) |
2f4b9368 | 680 | goto out; |
d38ceaf9 AD |
681 | break; |
682 | ||
683 | case 0x02000001: // destroy | |
f1689ec1 | 684 | destroyed = true; |
d38ceaf9 AD |
685 | break; |
686 | ||
687 | case 0x05000001: // context buffer | |
f1689ec1 | 688 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2, |
dc78330a | 689 | *size * 2, 0); |
f1689ec1 | 690 | if (r) |
2f4b9368 | 691 | goto out; |
f1689ec1 CK |
692 | break; |
693 | ||
d38ceaf9 | 694 | case 0x05000004: // video bitstream buffer |
f1689ec1 CK |
695 | tmp = amdgpu_get_ib_value(p, ib_idx, idx + 4); |
696 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2, | |
dc78330a | 697 | tmp, bs_idx); |
f1689ec1 | 698 | if (r) |
2f4b9368 | 699 | goto out; |
f1689ec1 CK |
700 | break; |
701 | ||
d38ceaf9 | 702 | case 0x05000005: // feedback buffer |
f1689ec1 | 703 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2, |
dc78330a | 704 | 4096, fb_idx); |
d38ceaf9 | 705 | if (r) |
2f4b9368 | 706 | goto out; |
d38ceaf9 AD |
707 | break; |
708 | ||
709 | default: | |
710 | DRM_ERROR("invalid VCE command (0x%x)!\n", cmd); | |
2f4b9368 CK |
711 | r = -EINVAL; |
712 | goto out; | |
d38ceaf9 AD |
713 | } |
714 | ||
f1689ec1 CK |
715 | if (session_idx == -1) { |
716 | DRM_ERROR("no session command at start of IB\n"); | |
2f4b9368 CK |
717 | r = -EINVAL; |
718 | goto out; | |
f1689ec1 CK |
719 | } |
720 | ||
d38ceaf9 AD |
721 | idx += len / 4; |
722 | } | |
723 | ||
2f4b9368 CK |
724 | if (allocated && !created) { |
725 | DRM_ERROR("New session without create command!\n"); | |
726 | r = -ENOENT; | |
727 | } | |
728 | ||
729 | out: | |
730 | if ((!r && destroyed) || (r && allocated)) { | |
731 | /* | |
732 | * IB contains a destroy msg or we have allocated an | |
733 | * handle and got an error, anyway free the handle | |
734 | */ | |
d38ceaf9 AD |
735 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) |
736 | atomic_cmpxchg(&p->adev->vce.handles[i], handle, 0); | |
d38ceaf9 AD |
737 | } |
738 | ||
2f4b9368 | 739 | return r; |
d38ceaf9 AD |
740 | } |
741 | ||
d38ceaf9 AD |
742 | /** |
743 | * amdgpu_vce_ring_emit_ib - execute indirect buffer | |
744 | * | |
745 | * @ring: engine to use | |
746 | * @ib: the IB to execute | |
747 | * | |
748 | */ | |
749 | void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) | |
750 | { | |
751 | amdgpu_ring_write(ring, VCE_CMD_IB); | |
752 | amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); | |
753 | amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); | |
754 | amdgpu_ring_write(ring, ib->length_dw); | |
755 | } | |
756 | ||
757 | /** | |
758 | * amdgpu_vce_ring_emit_fence - add a fence command to the ring | |
759 | * | |
760 | * @ring: engine to use | |
761 | * @fence: the fence | |
762 | * | |
763 | */ | |
764 | void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, | |
890ee23f | 765 | unsigned flags) |
d38ceaf9 | 766 | { |
890ee23f | 767 | WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); |
d38ceaf9 AD |
768 | |
769 | amdgpu_ring_write(ring, VCE_CMD_FENCE); | |
770 | amdgpu_ring_write(ring, addr); | |
771 | amdgpu_ring_write(ring, upper_32_bits(addr)); | |
772 | amdgpu_ring_write(ring, seq); | |
773 | amdgpu_ring_write(ring, VCE_CMD_TRAP); | |
774 | amdgpu_ring_write(ring, VCE_CMD_END); | |
775 | } | |
776 | ||
777 | /** | |
778 | * amdgpu_vce_ring_test_ring - test if VCE ring is working | |
779 | * | |
780 | * @ring: the engine to test on | |
781 | * | |
782 | */ | |
783 | int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring) | |
784 | { | |
785 | struct amdgpu_device *adev = ring->adev; | |
786 | uint32_t rptr = amdgpu_ring_get_rptr(ring); | |
787 | unsigned i; | |
788 | int r; | |
789 | ||
a27de35c | 790 | r = amdgpu_ring_alloc(ring, 16); |
d38ceaf9 AD |
791 | if (r) { |
792 | DRM_ERROR("amdgpu: vce failed to lock ring %d (%d).\n", | |
793 | ring->idx, r); | |
794 | return r; | |
795 | } | |
796 | amdgpu_ring_write(ring, VCE_CMD_END); | |
a27de35c | 797 | amdgpu_ring_commit(ring); |
d38ceaf9 AD |
798 | |
799 | for (i = 0; i < adev->usec_timeout; i++) { | |
800 | if (amdgpu_ring_get_rptr(ring) != rptr) | |
801 | break; | |
802 | DRM_UDELAY(1); | |
803 | } | |
804 | ||
805 | if (i < adev->usec_timeout) { | |
806 | DRM_INFO("ring test on %d succeeded in %d usecs\n", | |
807 | ring->idx, i); | |
808 | } else { | |
809 | DRM_ERROR("amdgpu: ring %d test failed\n", | |
810 | ring->idx); | |
811 | r = -ETIMEDOUT; | |
812 | } | |
813 | ||
814 | return r; | |
815 | } | |
816 | ||
817 | /** | |
818 | * amdgpu_vce_ring_test_ib - test if VCE IBs are working | |
819 | * | |
820 | * @ring: the engine to test on | |
821 | * | |
822 | */ | |
823 | int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring) | |
824 | { | |
ed40bfb8 | 825 | struct fence *fence = NULL; |
d38ceaf9 AD |
826 | int r; |
827 | ||
898e50d4 LL |
828 | /* skip vce ring1 ib test for now, since it's not reliable */ |
829 | if (ring == &ring->adev->vce.ring[1]) | |
830 | return 0; | |
831 | ||
d38ceaf9 AD |
832 | r = amdgpu_vce_get_create_msg(ring, 1, NULL); |
833 | if (r) { | |
834 | DRM_ERROR("amdgpu: failed to get create msg (%d).\n", r); | |
835 | goto error; | |
836 | } | |
837 | ||
838 | r = amdgpu_vce_get_destroy_msg(ring, 1, &fence); | |
839 | if (r) { | |
840 | DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r); | |
841 | goto error; | |
842 | } | |
843 | ||
ed40bfb8 | 844 | r = fence_wait(fence, false); |
d38ceaf9 AD |
845 | if (r) { |
846 | DRM_ERROR("amdgpu: fence wait failed (%d).\n", r); | |
847 | } else { | |
848 | DRM_INFO("ib test on ring %d succeeded\n", ring->idx); | |
849 | } | |
850 | error: | |
ed40bfb8 | 851 | fence_put(fence); |
d38ceaf9 AD |
852 | return r; |
853 | } |