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" |
d38ceaf9 AD |
52 | |
53 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
54 | MODULE_FIRMWARE(FIRMWARE_BONAIRE); | |
55 | MODULE_FIRMWARE(FIRMWARE_KABINI); | |
56 | MODULE_FIRMWARE(FIRMWARE_KAVERI); | |
57 | MODULE_FIRMWARE(FIRMWARE_HAWAII); | |
58 | MODULE_FIRMWARE(FIRMWARE_MULLINS); | |
59 | #endif | |
60 | MODULE_FIRMWARE(FIRMWARE_TONGA); | |
61 | MODULE_FIRMWARE(FIRMWARE_CARRIZO); | |
188a9bcd | 62 | MODULE_FIRMWARE(FIRMWARE_FIJI); |
d38ceaf9 AD |
63 | |
64 | static void amdgpu_vce_idle_work_handler(struct work_struct *work); | |
65 | ||
66 | /** | |
67 | * amdgpu_vce_init - allocate memory, load vce firmware | |
68 | * | |
69 | * @adev: amdgpu_device pointer | |
70 | * | |
71 | * First step to get VCE online, allocate memory and load the firmware | |
72 | */ | |
e9822622 | 73 | int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size) |
d38ceaf9 | 74 | { |
d38ceaf9 AD |
75 | const char *fw_name; |
76 | const struct common_firmware_header *hdr; | |
77 | unsigned ucode_version, version_major, version_minor, binary_id; | |
78 | int i, r; | |
79 | ||
80 | INIT_DELAYED_WORK(&adev->vce.idle_work, amdgpu_vce_idle_work_handler); | |
81 | ||
82 | switch (adev->asic_type) { | |
83 | #ifdef CONFIG_DRM_AMDGPU_CIK | |
84 | case CHIP_BONAIRE: | |
85 | fw_name = FIRMWARE_BONAIRE; | |
86 | break; | |
87 | case CHIP_KAVERI: | |
88 | fw_name = FIRMWARE_KAVERI; | |
89 | break; | |
90 | case CHIP_KABINI: | |
91 | fw_name = FIRMWARE_KABINI; | |
92 | break; | |
93 | case CHIP_HAWAII: | |
94 | fw_name = FIRMWARE_HAWAII; | |
95 | break; | |
96 | case CHIP_MULLINS: | |
97 | fw_name = FIRMWARE_MULLINS; | |
98 | break; | |
99 | #endif | |
100 | case CHIP_TONGA: | |
101 | fw_name = FIRMWARE_TONGA; | |
102 | break; | |
103 | case CHIP_CARRIZO: | |
104 | fw_name = FIRMWARE_CARRIZO; | |
105 | break; | |
188a9bcd AD |
106 | case CHIP_FIJI: |
107 | fw_name = FIRMWARE_FIJI; | |
108 | break; | |
d38ceaf9 AD |
109 | |
110 | default: | |
111 | return -EINVAL; | |
112 | } | |
113 | ||
114 | r = request_firmware(&adev->vce.fw, fw_name, adev->dev); | |
115 | if (r) { | |
116 | dev_err(adev->dev, "amdgpu_vce: Can't load firmware \"%s\"\n", | |
117 | fw_name); | |
118 | return r; | |
119 | } | |
120 | ||
121 | r = amdgpu_ucode_validate(adev->vce.fw); | |
122 | if (r) { | |
123 | dev_err(adev->dev, "amdgpu_vce: Can't validate firmware \"%s\"\n", | |
124 | fw_name); | |
125 | release_firmware(adev->vce.fw); | |
126 | adev->vce.fw = NULL; | |
127 | return r; | |
128 | } | |
129 | ||
130 | hdr = (const struct common_firmware_header *)adev->vce.fw->data; | |
131 | ||
132 | ucode_version = le32_to_cpu(hdr->ucode_version); | |
133 | version_major = (ucode_version >> 20) & 0xfff; | |
134 | version_minor = (ucode_version >> 8) & 0xfff; | |
135 | binary_id = ucode_version & 0xff; | |
136 | DRM_INFO("Found VCE firmware Version: %hhd.%hhd Binary ID: %hhd\n", | |
137 | version_major, version_minor, binary_id); | |
138 | adev->vce.fw_version = ((version_major << 24) | (version_minor << 16) | | |
139 | (binary_id << 8)); | |
140 | ||
141 | /* allocate firmware, stack and heap BO */ | |
142 | ||
d38ceaf9 | 143 | r = amdgpu_bo_create(adev, size, PAGE_SIZE, true, |
857d913d AD |
144 | AMDGPU_GEM_DOMAIN_VRAM, |
145 | AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED, | |
72d7668b | 146 | NULL, NULL, &adev->vce.vcpu_bo); |
d38ceaf9 AD |
147 | if (r) { |
148 | dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r); | |
149 | return r; | |
150 | } | |
151 | ||
152 | r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false); | |
153 | if (r) { | |
154 | amdgpu_bo_unref(&adev->vce.vcpu_bo); | |
155 | dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r); | |
156 | return r; | |
157 | } | |
158 | ||
159 | r = amdgpu_bo_pin(adev->vce.vcpu_bo, AMDGPU_GEM_DOMAIN_VRAM, | |
160 | &adev->vce.gpu_addr); | |
161 | amdgpu_bo_unreserve(adev->vce.vcpu_bo); | |
162 | if (r) { | |
163 | amdgpu_bo_unref(&adev->vce.vcpu_bo); | |
164 | dev_err(adev->dev, "(%d) VCE bo pin failed\n", r); | |
165 | return r; | |
166 | } | |
167 | ||
168 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { | |
169 | atomic_set(&adev->vce.handles[i], 0); | |
170 | adev->vce.filp[i] = NULL; | |
171 | } | |
172 | ||
173 | return 0; | |
174 | } | |
175 | ||
176 | /** | |
177 | * amdgpu_vce_fini - free memory | |
178 | * | |
179 | * @adev: amdgpu_device pointer | |
180 | * | |
181 | * Last step on VCE teardown, free firmware memory | |
182 | */ | |
183 | int amdgpu_vce_sw_fini(struct amdgpu_device *adev) | |
184 | { | |
185 | if (adev->vce.vcpu_bo == NULL) | |
186 | return 0; | |
187 | ||
188 | amdgpu_bo_unref(&adev->vce.vcpu_bo); | |
189 | ||
190 | amdgpu_ring_fini(&adev->vce.ring[0]); | |
191 | amdgpu_ring_fini(&adev->vce.ring[1]); | |
192 | ||
193 | release_firmware(adev->vce.fw); | |
194 | ||
195 | return 0; | |
196 | } | |
197 | ||
198 | /** | |
199 | * amdgpu_vce_suspend - unpin VCE fw memory | |
200 | * | |
201 | * @adev: amdgpu_device pointer | |
202 | * | |
203 | */ | |
204 | int amdgpu_vce_suspend(struct amdgpu_device *adev) | |
205 | { | |
206 | int i; | |
207 | ||
208 | if (adev->vce.vcpu_bo == NULL) | |
209 | return 0; | |
210 | ||
211 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) | |
212 | if (atomic_read(&adev->vce.handles[i])) | |
213 | break; | |
214 | ||
215 | if (i == AMDGPU_MAX_VCE_HANDLES) | |
216 | return 0; | |
217 | ||
218 | /* TODO: suspending running encoding sessions isn't supported */ | |
219 | return -EINVAL; | |
220 | } | |
221 | ||
222 | /** | |
223 | * amdgpu_vce_resume - pin VCE fw memory | |
224 | * | |
225 | * @adev: amdgpu_device pointer | |
226 | * | |
227 | */ | |
228 | int amdgpu_vce_resume(struct amdgpu_device *adev) | |
229 | { | |
230 | void *cpu_addr; | |
231 | const struct common_firmware_header *hdr; | |
232 | unsigned offset; | |
233 | int r; | |
234 | ||
235 | if (adev->vce.vcpu_bo == NULL) | |
236 | return -EINVAL; | |
237 | ||
238 | r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false); | |
239 | if (r) { | |
240 | dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r); | |
241 | return r; | |
242 | } | |
243 | ||
244 | r = amdgpu_bo_kmap(adev->vce.vcpu_bo, &cpu_addr); | |
245 | if (r) { | |
246 | amdgpu_bo_unreserve(adev->vce.vcpu_bo); | |
247 | dev_err(adev->dev, "(%d) VCE map failed\n", r); | |
248 | return r; | |
249 | } | |
250 | ||
251 | hdr = (const struct common_firmware_header *)adev->vce.fw->data; | |
252 | offset = le32_to_cpu(hdr->ucode_array_offset_bytes); | |
253 | memcpy(cpu_addr, (adev->vce.fw->data) + offset, | |
254 | (adev->vce.fw->size) - offset); | |
255 | ||
256 | amdgpu_bo_kunmap(adev->vce.vcpu_bo); | |
257 | ||
258 | amdgpu_bo_unreserve(adev->vce.vcpu_bo); | |
259 | ||
260 | return 0; | |
261 | } | |
262 | ||
263 | /** | |
264 | * amdgpu_vce_idle_work_handler - power off VCE | |
265 | * | |
266 | * @work: pointer to work structure | |
267 | * | |
268 | * power of VCE when it's not used any more | |
269 | */ | |
270 | static void amdgpu_vce_idle_work_handler(struct work_struct *work) | |
271 | { | |
272 | struct amdgpu_device *adev = | |
273 | container_of(work, struct amdgpu_device, vce.idle_work.work); | |
274 | ||
275 | if ((amdgpu_fence_count_emitted(&adev->vce.ring[0]) == 0) && | |
276 | (amdgpu_fence_count_emitted(&adev->vce.ring[1]) == 0)) { | |
277 | if (adev->pm.dpm_enabled) { | |
278 | amdgpu_dpm_enable_vce(adev, false); | |
279 | } else { | |
280 | amdgpu_asic_set_vce_clocks(adev, 0, 0); | |
281 | } | |
282 | } else { | |
283 | schedule_delayed_work(&adev->vce.idle_work, | |
284 | msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS)); | |
285 | } | |
286 | } | |
287 | ||
288 | /** | |
289 | * amdgpu_vce_note_usage - power up VCE | |
290 | * | |
291 | * @adev: amdgpu_device pointer | |
292 | * | |
293 | * Make sure VCE is powerd up when we want to use it | |
294 | */ | |
295 | static void amdgpu_vce_note_usage(struct amdgpu_device *adev) | |
296 | { | |
297 | bool streams_changed = false; | |
298 | bool set_clocks = !cancel_delayed_work_sync(&adev->vce.idle_work); | |
299 | set_clocks &= schedule_delayed_work(&adev->vce.idle_work, | |
300 | msecs_to_jiffies(VCE_IDLE_TIMEOUT_MS)); | |
301 | ||
302 | if (adev->pm.dpm_enabled) { | |
303 | /* XXX figure out if the streams changed */ | |
304 | streams_changed = false; | |
305 | } | |
306 | ||
307 | if (set_clocks || streams_changed) { | |
308 | if (adev->pm.dpm_enabled) { | |
309 | amdgpu_dpm_enable_vce(adev, true); | |
310 | } else { | |
311 | amdgpu_asic_set_vce_clocks(adev, 53300, 40000); | |
312 | } | |
313 | } | |
314 | } | |
315 | ||
316 | /** | |
317 | * amdgpu_vce_free_handles - free still open VCE handles | |
318 | * | |
319 | * @adev: amdgpu_device pointer | |
320 | * @filp: drm file pointer | |
321 | * | |
322 | * Close all VCE handles still open by this file pointer | |
323 | */ | |
324 | void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp) | |
325 | { | |
326 | struct amdgpu_ring *ring = &adev->vce.ring[0]; | |
327 | int i, r; | |
328 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { | |
329 | uint32_t handle = atomic_read(&adev->vce.handles[i]); | |
330 | if (!handle || adev->vce.filp[i] != filp) | |
331 | continue; | |
332 | ||
333 | amdgpu_vce_note_usage(adev); | |
334 | ||
335 | r = amdgpu_vce_get_destroy_msg(ring, handle, NULL); | |
336 | if (r) | |
337 | DRM_ERROR("Error destroying VCE handle (%d)!\n", r); | |
338 | ||
339 | adev->vce.filp[i] = NULL; | |
340 | atomic_set(&adev->vce.handles[i], 0); | |
341 | } | |
342 | } | |
343 | ||
8128765c | 344 | static int amdgpu_vce_free_job( |
4c7eb91c | 345 | struct amdgpu_job *job) |
8128765c | 346 | { |
4c7eb91c JZ |
347 | amdgpu_ib_free(job->adev, job->ibs); |
348 | kfree(job->ibs); | |
8128765c CZ |
349 | return 0; |
350 | } | |
351 | ||
d38ceaf9 AD |
352 | /** |
353 | * amdgpu_vce_get_create_msg - generate a VCE create msg | |
354 | * | |
355 | * @adev: amdgpu_device pointer | |
356 | * @ring: ring we should submit the msg to | |
357 | * @handle: VCE session handle to use | |
358 | * @fence: optional fence to return | |
359 | * | |
360 | * Open up a stream for HW test | |
361 | */ | |
362 | int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle, | |
ed40bfb8 | 363 | struct fence **fence) |
d38ceaf9 AD |
364 | { |
365 | const unsigned ib_size_dw = 1024; | |
8128765c | 366 | struct amdgpu_ib *ib = NULL; |
1763552e | 367 | struct fence *f = NULL; |
8128765c | 368 | struct amdgpu_device *adev = ring->adev; |
d38ceaf9 AD |
369 | uint64_t dummy; |
370 | int i, r; | |
371 | ||
8128765c CZ |
372 | ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); |
373 | if (!ib) | |
374 | return -ENOMEM; | |
375 | r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib); | |
d38ceaf9 AD |
376 | if (r) { |
377 | DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); | |
8128765c | 378 | kfree(ib); |
d38ceaf9 AD |
379 | return r; |
380 | } | |
381 | ||
8128765c | 382 | dummy = ib->gpu_addr + 1024; |
d38ceaf9 AD |
383 | |
384 | /* stitch together an VCE create msg */ | |
8128765c CZ |
385 | ib->length_dw = 0; |
386 | ib->ptr[ib->length_dw++] = 0x0000000c; /* len */ | |
387 | ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */ | |
388 | ib->ptr[ib->length_dw++] = handle; | |
389 | ||
390 | ib->ptr[ib->length_dw++] = 0x00000030; /* len */ | |
391 | ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */ | |
392 | ib->ptr[ib->length_dw++] = 0x00000000; | |
393 | ib->ptr[ib->length_dw++] = 0x00000042; | |
394 | ib->ptr[ib->length_dw++] = 0x0000000a; | |
395 | ib->ptr[ib->length_dw++] = 0x00000001; | |
396 | ib->ptr[ib->length_dw++] = 0x00000080; | |
397 | ib->ptr[ib->length_dw++] = 0x00000060; | |
398 | ib->ptr[ib->length_dw++] = 0x00000100; | |
399 | ib->ptr[ib->length_dw++] = 0x00000100; | |
400 | ib->ptr[ib->length_dw++] = 0x0000000c; | |
401 | ib->ptr[ib->length_dw++] = 0x00000000; | |
402 | ||
403 | ib->ptr[ib->length_dw++] = 0x00000014; /* len */ | |
404 | ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */ | |
405 | ib->ptr[ib->length_dw++] = upper_32_bits(dummy); | |
406 | ib->ptr[ib->length_dw++] = dummy; | |
407 | ib->ptr[ib->length_dw++] = 0x00000001; | |
408 | ||
409 | for (i = ib->length_dw; i < ib_size_dw; ++i) | |
410 | ib->ptr[i] = 0x0; | |
411 | ||
412 | r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, | |
413 | &amdgpu_vce_free_job, | |
1763552e CZ |
414 | AMDGPU_FENCE_OWNER_UNDEFINED, |
415 | &f); | |
8128765c CZ |
416 | if (r) |
417 | goto err; | |
d38ceaf9 | 418 | if (fence) |
1763552e | 419 | *fence = fence_get(f); |
281b4223 | 420 | fence_put(f); |
8128765c CZ |
421 | if (amdgpu_enable_scheduler) |
422 | return 0; | |
423 | err: | |
424 | amdgpu_ib_free(adev, ib); | |
425 | kfree(ib); | |
d38ceaf9 AD |
426 | return r; |
427 | } | |
428 | ||
429 | /** | |
430 | * amdgpu_vce_get_destroy_msg - generate a VCE destroy msg | |
431 | * | |
432 | * @adev: amdgpu_device pointer | |
433 | * @ring: ring we should submit the msg to | |
434 | * @handle: VCE session handle to use | |
435 | * @fence: optional fence to return | |
436 | * | |
437 | * Close up a stream for HW test or if userspace failed to do so | |
438 | */ | |
439 | int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle, | |
ed40bfb8 | 440 | struct fence **fence) |
d38ceaf9 AD |
441 | { |
442 | const unsigned ib_size_dw = 1024; | |
8128765c | 443 | struct amdgpu_ib *ib = NULL; |
1763552e | 444 | struct fence *f = NULL; |
8128765c | 445 | struct amdgpu_device *adev = ring->adev; |
d38ceaf9 AD |
446 | uint64_t dummy; |
447 | int i, r; | |
448 | ||
8128765c CZ |
449 | ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); |
450 | if (!ib) | |
451 | return -ENOMEM; | |
452 | ||
453 | r = amdgpu_ib_get(ring, NULL, ib_size_dw * 4, ib); | |
d38ceaf9 | 454 | if (r) { |
8128765c | 455 | kfree(ib); |
d38ceaf9 AD |
456 | DRM_ERROR("amdgpu: failed to get ib (%d).\n", r); |
457 | return r; | |
458 | } | |
459 | ||
8128765c | 460 | dummy = ib->gpu_addr + 1024; |
d38ceaf9 AD |
461 | |
462 | /* stitch together an VCE destroy msg */ | |
8128765c CZ |
463 | ib->length_dw = 0; |
464 | ib->ptr[ib->length_dw++] = 0x0000000c; /* len */ | |
465 | ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */ | |
466 | ib->ptr[ib->length_dw++] = handle; | |
467 | ||
468 | ib->ptr[ib->length_dw++] = 0x00000014; /* len */ | |
469 | ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */ | |
470 | ib->ptr[ib->length_dw++] = upper_32_bits(dummy); | |
471 | ib->ptr[ib->length_dw++] = dummy; | |
472 | ib->ptr[ib->length_dw++] = 0x00000001; | |
473 | ||
474 | ib->ptr[ib->length_dw++] = 0x00000008; /* len */ | |
475 | ib->ptr[ib->length_dw++] = 0x02000001; /* destroy cmd */ | |
476 | ||
477 | for (i = ib->length_dw; i < ib_size_dw; ++i) | |
478 | ib->ptr[i] = 0x0; | |
479 | r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1, | |
480 | &amdgpu_vce_free_job, | |
1763552e CZ |
481 | AMDGPU_FENCE_OWNER_UNDEFINED, |
482 | &f); | |
8128765c CZ |
483 | if (r) |
484 | goto err; | |
d38ceaf9 | 485 | if (fence) |
1763552e | 486 | *fence = fence_get(f); |
281b4223 | 487 | fence_put(f); |
8128765c CZ |
488 | if (amdgpu_enable_scheduler) |
489 | return 0; | |
490 | err: | |
491 | amdgpu_ib_free(adev, ib); | |
492 | kfree(ib); | |
d38ceaf9 AD |
493 | return r; |
494 | } | |
495 | ||
496 | /** | |
497 | * amdgpu_vce_cs_reloc - command submission relocation | |
498 | * | |
499 | * @p: parser context | |
500 | * @lo: address of lower dword | |
501 | * @hi: address of higher dword | |
f1689ec1 | 502 | * @size: minimum size |
d38ceaf9 AD |
503 | * |
504 | * Patch relocation inside command stream with real buffer address | |
505 | */ | |
f1689ec1 | 506 | static int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx, |
dc78330a | 507 | int lo, int hi, unsigned size, uint32_t index) |
d38ceaf9 AD |
508 | { |
509 | struct amdgpu_bo_va_mapping *mapping; | |
510 | struct amdgpu_ib *ib = &p->ibs[ib_idx]; | |
511 | struct amdgpu_bo *bo; | |
512 | uint64_t addr; | |
513 | ||
dc78330a CK |
514 | if (index == 0xffffffff) |
515 | index = 0; | |
516 | ||
d38ceaf9 AD |
517 | addr = ((uint64_t)amdgpu_get_ib_value(p, ib_idx, lo)) | |
518 | ((uint64_t)amdgpu_get_ib_value(p, ib_idx, hi)) << 32; | |
dc78330a | 519 | addr += ((uint64_t)size) * ((uint64_t)index); |
d38ceaf9 AD |
520 | |
521 | mapping = amdgpu_cs_find_mapping(p, addr, &bo); | |
522 | if (mapping == NULL) { | |
dc78330a CK |
523 | DRM_ERROR("Can't find BO for addr 0x%010Lx %d %d %d %d\n", |
524 | addr, lo, hi, size, index); | |
d38ceaf9 AD |
525 | return -EINVAL; |
526 | } | |
527 | ||
f1689ec1 CK |
528 | if ((addr + (uint64_t)size) > |
529 | ((uint64_t)mapping->it.last + 1) * AMDGPU_GPU_PAGE_SIZE) { | |
530 | DRM_ERROR("BO to small for addr 0x%010Lx %d %d\n", | |
531 | addr, lo, hi); | |
532 | return -EINVAL; | |
533 | } | |
534 | ||
d38ceaf9 AD |
535 | addr -= ((uint64_t)mapping->it.start) * AMDGPU_GPU_PAGE_SIZE; |
536 | addr += amdgpu_bo_gpu_offset(bo); | |
dc78330a | 537 | addr -= ((uint64_t)size) * ((uint64_t)index); |
d38ceaf9 AD |
538 | |
539 | ib->ptr[lo] = addr & 0xFFFFFFFF; | |
540 | ib->ptr[hi] = addr >> 32; | |
541 | ||
542 | return 0; | |
543 | } | |
544 | ||
f1689ec1 CK |
545 | /** |
546 | * amdgpu_vce_validate_handle - validate stream handle | |
547 | * | |
548 | * @p: parser context | |
549 | * @handle: handle to validate | |
2f4b9368 | 550 | * @allocated: allocated a new handle? |
f1689ec1 CK |
551 | * |
552 | * Validates the handle and return the found session index or -EINVAL | |
553 | * we we don't have another free session index. | |
554 | */ | |
555 | static int amdgpu_vce_validate_handle(struct amdgpu_cs_parser *p, | |
2f4b9368 | 556 | uint32_t handle, bool *allocated) |
f1689ec1 CK |
557 | { |
558 | unsigned i; | |
559 | ||
2f4b9368 CK |
560 | *allocated = false; |
561 | ||
f1689ec1 CK |
562 | /* validate the handle */ |
563 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { | |
2f4b9368 CK |
564 | if (atomic_read(&p->adev->vce.handles[i]) == handle) { |
565 | if (p->adev->vce.filp[i] != p->filp) { | |
566 | DRM_ERROR("VCE handle collision detected!\n"); | |
567 | return -EINVAL; | |
568 | } | |
f1689ec1 | 569 | return i; |
2f4b9368 | 570 | } |
f1689ec1 CK |
571 | } |
572 | ||
573 | /* handle not found try to alloc a new one */ | |
574 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) { | |
575 | if (!atomic_cmpxchg(&p->adev->vce.handles[i], 0, handle)) { | |
576 | p->adev->vce.filp[i] = p->filp; | |
577 | p->adev->vce.img_size[i] = 0; | |
2f4b9368 | 578 | *allocated = true; |
f1689ec1 CK |
579 | return i; |
580 | } | |
581 | } | |
582 | ||
583 | DRM_ERROR("No more free VCE handles!\n"); | |
584 | return -EINVAL; | |
585 | } | |
586 | ||
d38ceaf9 AD |
587 | /** |
588 | * amdgpu_vce_cs_parse - parse and validate the command stream | |
589 | * | |
590 | * @p: parser context | |
591 | * | |
592 | */ | |
593 | int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx) | |
594 | { | |
d38ceaf9 | 595 | struct amdgpu_ib *ib = &p->ibs[ib_idx]; |
dc78330a | 596 | unsigned fb_idx = 0, bs_idx = 0; |
f1689ec1 CK |
597 | int session_idx = -1; |
598 | bool destroyed = false; | |
2f4b9368 CK |
599 | bool created = false; |
600 | bool allocated = false; | |
f1689ec1 CK |
601 | uint32_t tmp, handle = 0; |
602 | uint32_t *size = &tmp; | |
2f4b9368 | 603 | int i, r = 0, idx = 0; |
d38ceaf9 AD |
604 | |
605 | amdgpu_vce_note_usage(p->adev); | |
606 | ||
607 | while (idx < ib->length_dw) { | |
608 | uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx); | |
609 | uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1); | |
610 | ||
611 | if ((len < 8) || (len & 3)) { | |
612 | DRM_ERROR("invalid VCE command length (%d)!\n", len); | |
2f4b9368 CK |
613 | r = -EINVAL; |
614 | goto out; | |
d38ceaf9 AD |
615 | } |
616 | ||
f1689ec1 CK |
617 | if (destroyed) { |
618 | DRM_ERROR("No other command allowed after destroy!\n"); | |
2f4b9368 CK |
619 | r = -EINVAL; |
620 | goto out; | |
f1689ec1 CK |
621 | } |
622 | ||
d38ceaf9 AD |
623 | switch (cmd) { |
624 | case 0x00000001: // session | |
625 | handle = amdgpu_get_ib_value(p, ib_idx, idx + 2); | |
2f4b9368 CK |
626 | session_idx = amdgpu_vce_validate_handle(p, handle, |
627 | &allocated); | |
f1689ec1 CK |
628 | if (session_idx < 0) |
629 | return session_idx; | |
630 | size = &p->adev->vce.img_size[session_idx]; | |
d38ceaf9 AD |
631 | break; |
632 | ||
633 | case 0x00000002: // task info | |
dc78330a CK |
634 | fb_idx = amdgpu_get_ib_value(p, ib_idx, idx + 6); |
635 | bs_idx = amdgpu_get_ib_value(p, ib_idx, idx + 7); | |
f1689ec1 CK |
636 | break; |
637 | ||
d38ceaf9 | 638 | case 0x01000001: // create |
2f4b9368 CK |
639 | created = true; |
640 | if (!allocated) { | |
641 | DRM_ERROR("Handle already in use!\n"); | |
642 | r = -EINVAL; | |
643 | goto out; | |
644 | } | |
645 | ||
f1689ec1 CK |
646 | *size = amdgpu_get_ib_value(p, ib_idx, idx + 8) * |
647 | amdgpu_get_ib_value(p, ib_idx, idx + 10) * | |
648 | 8 * 3 / 2; | |
649 | break; | |
650 | ||
d38ceaf9 AD |
651 | case 0x04000001: // config extension |
652 | case 0x04000002: // pic control | |
653 | case 0x04000005: // rate control | |
654 | case 0x04000007: // motion estimation | |
655 | case 0x04000008: // rdo | |
656 | case 0x04000009: // vui | |
657 | case 0x05000002: // auxiliary buffer | |
658 | break; | |
659 | ||
660 | case 0x03000001: // encode | |
f1689ec1 | 661 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 10, idx + 9, |
dc78330a | 662 | *size, 0); |
d38ceaf9 | 663 | if (r) |
2f4b9368 | 664 | goto out; |
d38ceaf9 | 665 | |
f1689ec1 | 666 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 12, idx + 11, |
dc78330a | 667 | *size / 3, 0); |
d38ceaf9 | 668 | if (r) |
2f4b9368 | 669 | goto out; |
d38ceaf9 AD |
670 | break; |
671 | ||
672 | case 0x02000001: // destroy | |
f1689ec1 | 673 | destroyed = true; |
d38ceaf9 AD |
674 | break; |
675 | ||
676 | case 0x05000001: // context buffer | |
f1689ec1 | 677 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2, |
dc78330a | 678 | *size * 2, 0); |
f1689ec1 | 679 | if (r) |
2f4b9368 | 680 | goto out; |
f1689ec1 CK |
681 | break; |
682 | ||
d38ceaf9 | 683 | case 0x05000004: // video bitstream buffer |
f1689ec1 CK |
684 | tmp = amdgpu_get_ib_value(p, ib_idx, idx + 4); |
685 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2, | |
dc78330a | 686 | tmp, bs_idx); |
f1689ec1 | 687 | if (r) |
2f4b9368 | 688 | goto out; |
f1689ec1 CK |
689 | break; |
690 | ||
d38ceaf9 | 691 | case 0x05000005: // feedback buffer |
f1689ec1 | 692 | r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2, |
dc78330a | 693 | 4096, fb_idx); |
d38ceaf9 | 694 | if (r) |
2f4b9368 | 695 | goto out; |
d38ceaf9 AD |
696 | break; |
697 | ||
698 | default: | |
699 | DRM_ERROR("invalid VCE command (0x%x)!\n", cmd); | |
2f4b9368 CK |
700 | r = -EINVAL; |
701 | goto out; | |
d38ceaf9 AD |
702 | } |
703 | ||
f1689ec1 CK |
704 | if (session_idx == -1) { |
705 | DRM_ERROR("no session command at start of IB\n"); | |
2f4b9368 CK |
706 | r = -EINVAL; |
707 | goto out; | |
f1689ec1 CK |
708 | } |
709 | ||
d38ceaf9 AD |
710 | idx += len / 4; |
711 | } | |
712 | ||
2f4b9368 CK |
713 | if (allocated && !created) { |
714 | DRM_ERROR("New session without create command!\n"); | |
715 | r = -ENOENT; | |
716 | } | |
717 | ||
718 | out: | |
719 | if ((!r && destroyed) || (r && allocated)) { | |
720 | /* | |
721 | * IB contains a destroy msg or we have allocated an | |
722 | * handle and got an error, anyway free the handle | |
723 | */ | |
d38ceaf9 AD |
724 | for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) |
725 | atomic_cmpxchg(&p->adev->vce.handles[i], handle, 0); | |
d38ceaf9 AD |
726 | } |
727 | ||
2f4b9368 | 728 | return r; |
d38ceaf9 AD |
729 | } |
730 | ||
731 | /** | |
732 | * amdgpu_vce_ring_emit_semaphore - emit a semaphore command | |
733 | * | |
734 | * @ring: engine to use | |
735 | * @semaphore: address of semaphore | |
736 | * @emit_wait: true=emit wait, false=emit signal | |
737 | * | |
738 | */ | |
739 | bool amdgpu_vce_ring_emit_semaphore(struct amdgpu_ring *ring, | |
740 | struct amdgpu_semaphore *semaphore, | |
741 | bool emit_wait) | |
742 | { | |
743 | uint64_t addr = semaphore->gpu_addr; | |
744 | ||
745 | amdgpu_ring_write(ring, VCE_CMD_SEMAPHORE); | |
746 | amdgpu_ring_write(ring, (addr >> 3) & 0x000FFFFF); | |
747 | amdgpu_ring_write(ring, (addr >> 23) & 0x000FFFFF); | |
748 | amdgpu_ring_write(ring, 0x01003000 | (emit_wait ? 1 : 0)); | |
749 | if (!emit_wait) | |
750 | amdgpu_ring_write(ring, VCE_CMD_END); | |
751 | ||
752 | return true; | |
753 | } | |
754 | ||
755 | /** | |
756 | * amdgpu_vce_ring_emit_ib - execute indirect buffer | |
757 | * | |
758 | * @ring: engine to use | |
759 | * @ib: the IB to execute | |
760 | * | |
761 | */ | |
762 | void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib) | |
763 | { | |
764 | amdgpu_ring_write(ring, VCE_CMD_IB); | |
765 | amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr)); | |
766 | amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr)); | |
767 | amdgpu_ring_write(ring, ib->length_dw); | |
768 | } | |
769 | ||
770 | /** | |
771 | * amdgpu_vce_ring_emit_fence - add a fence command to the ring | |
772 | * | |
773 | * @ring: engine to use | |
774 | * @fence: the fence | |
775 | * | |
776 | */ | |
777 | void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq, | |
890ee23f | 778 | unsigned flags) |
d38ceaf9 | 779 | { |
890ee23f | 780 | WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT); |
d38ceaf9 AD |
781 | |
782 | amdgpu_ring_write(ring, VCE_CMD_FENCE); | |
783 | amdgpu_ring_write(ring, addr); | |
784 | amdgpu_ring_write(ring, upper_32_bits(addr)); | |
785 | amdgpu_ring_write(ring, seq); | |
786 | amdgpu_ring_write(ring, VCE_CMD_TRAP); | |
787 | amdgpu_ring_write(ring, VCE_CMD_END); | |
788 | } | |
789 | ||
790 | /** | |
791 | * amdgpu_vce_ring_test_ring - test if VCE ring is working | |
792 | * | |
793 | * @ring: the engine to test on | |
794 | * | |
795 | */ | |
796 | int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring) | |
797 | { | |
798 | struct amdgpu_device *adev = ring->adev; | |
799 | uint32_t rptr = amdgpu_ring_get_rptr(ring); | |
800 | unsigned i; | |
801 | int r; | |
802 | ||
803 | r = amdgpu_ring_lock(ring, 16); | |
804 | if (r) { | |
805 | DRM_ERROR("amdgpu: vce failed to lock ring %d (%d).\n", | |
806 | ring->idx, r); | |
807 | return r; | |
808 | } | |
809 | amdgpu_ring_write(ring, VCE_CMD_END); | |
810 | amdgpu_ring_unlock_commit(ring); | |
811 | ||
812 | for (i = 0; i < adev->usec_timeout; i++) { | |
813 | if (amdgpu_ring_get_rptr(ring) != rptr) | |
814 | break; | |
815 | DRM_UDELAY(1); | |
816 | } | |
817 | ||
818 | if (i < adev->usec_timeout) { | |
819 | DRM_INFO("ring test on %d succeeded in %d usecs\n", | |
820 | ring->idx, i); | |
821 | } else { | |
822 | DRM_ERROR("amdgpu: ring %d test failed\n", | |
823 | ring->idx); | |
824 | r = -ETIMEDOUT; | |
825 | } | |
826 | ||
827 | return r; | |
828 | } | |
829 | ||
830 | /** | |
831 | * amdgpu_vce_ring_test_ib - test if VCE IBs are working | |
832 | * | |
833 | * @ring: the engine to test on | |
834 | * | |
835 | */ | |
836 | int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring) | |
837 | { | |
ed40bfb8 | 838 | struct fence *fence = NULL; |
d38ceaf9 AD |
839 | int r; |
840 | ||
898e50d4 LL |
841 | /* skip vce ring1 ib test for now, since it's not reliable */ |
842 | if (ring == &ring->adev->vce.ring[1]) | |
843 | return 0; | |
844 | ||
d38ceaf9 AD |
845 | r = amdgpu_vce_get_create_msg(ring, 1, NULL); |
846 | if (r) { | |
847 | DRM_ERROR("amdgpu: failed to get create msg (%d).\n", r); | |
848 | goto error; | |
849 | } | |
850 | ||
851 | r = amdgpu_vce_get_destroy_msg(ring, 1, &fence); | |
852 | if (r) { | |
853 | DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r); | |
854 | goto error; | |
855 | } | |
856 | ||
ed40bfb8 | 857 | r = fence_wait(fence, false); |
d38ceaf9 AD |
858 | if (r) { |
859 | DRM_ERROR("amdgpu: fence wait failed (%d).\n", r); | |
860 | } else { | |
861 | DRM_INFO("ib test on ring %d succeeded\n", ring->idx); | |
862 | } | |
863 | error: | |
ed40bfb8 | 864 | fence_put(fence); |
d38ceaf9 AD |
865 | return r; |
866 | } |