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 
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	msecs_to_jiffies(1000)
40 
41 /* Firmware Names */
42 #ifdef CONFIG_DRM_AMDGPU_CIK
43 #define FIRMWARE_BONAIRE	"amdgpu/bonaire_vce.bin"
44 #define FIRMWARE_KABINI	"amdgpu/kabini_vce.bin"
45 #define FIRMWARE_KAVERI	"amdgpu/kaveri_vce.bin"
46 #define FIRMWARE_HAWAII	"amdgpu/hawaii_vce.bin"
47 #define FIRMWARE_MULLINS	"amdgpu/mullins_vce.bin"
48 #endif
49 #define FIRMWARE_TONGA		"amdgpu/tonga_vce.bin"
50 #define FIRMWARE_CARRIZO	"amdgpu/carrizo_vce.bin"
51 #define FIRMWARE_FIJI		"amdgpu/fiji_vce.bin"
52 #define FIRMWARE_STONEY		"amdgpu/stoney_vce.bin"
53 #define FIRMWARE_POLARIS10	"amdgpu/polaris10_vce.bin"
54 #define FIRMWARE_POLARIS11	"amdgpu/polaris11_vce.bin"
55 #define FIRMWARE_POLARIS12	"amdgpu/polaris12_vce.bin"
56 #define FIRMWARE_VEGAM		"amdgpu/vegam_vce.bin"
57 
58 #define FIRMWARE_VEGA10		"amdgpu/vega10_vce.bin"
59 #define FIRMWARE_VEGA12		"amdgpu/vega12_vce.bin"
60 #define FIRMWARE_VEGA20		"amdgpu/vega20_vce.bin"
61 
62 #ifdef CONFIG_DRM_AMDGPU_CIK
63 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
64 MODULE_FIRMWARE(FIRMWARE_KABINI);
65 MODULE_FIRMWARE(FIRMWARE_KAVERI);
66 MODULE_FIRMWARE(FIRMWARE_HAWAII);
67 MODULE_FIRMWARE(FIRMWARE_MULLINS);
68 #endif
69 MODULE_FIRMWARE(FIRMWARE_TONGA);
70 MODULE_FIRMWARE(FIRMWARE_CARRIZO);
71 MODULE_FIRMWARE(FIRMWARE_FIJI);
72 MODULE_FIRMWARE(FIRMWARE_STONEY);
73 MODULE_FIRMWARE(FIRMWARE_POLARIS10);
74 MODULE_FIRMWARE(FIRMWARE_POLARIS11);
75 MODULE_FIRMWARE(FIRMWARE_POLARIS12);
76 MODULE_FIRMWARE(FIRMWARE_VEGAM);
77 
78 MODULE_FIRMWARE(FIRMWARE_VEGA10);
79 MODULE_FIRMWARE(FIRMWARE_VEGA12);
80 MODULE_FIRMWARE(FIRMWARE_VEGA20);
81 
82 static void amdgpu_vce_idle_work_handler(struct work_struct *work);
83 static int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
84 				     struct amdgpu_bo *bo,
85 				     struct dma_fence **fence);
86 static int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
87 				      bool direct, struct dma_fence **fence);
88 
89 /**
90  * amdgpu_vce_init - allocate memory, load vce firmware
91  *
92  * @adev: amdgpu_device pointer
93  *
94  * First step to get VCE online, allocate memory and load the firmware
95  */
96 int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
97 {
98 	const char *fw_name;
99 	const struct common_firmware_header *hdr;
100 	unsigned ucode_version, version_major, version_minor, binary_id;
101 	int i, r;
102 
103 	switch (adev->asic_type) {
104 #ifdef CONFIG_DRM_AMDGPU_CIK
105 	case CHIP_BONAIRE:
106 		fw_name = FIRMWARE_BONAIRE;
107 		break;
108 	case CHIP_KAVERI:
109 		fw_name = FIRMWARE_KAVERI;
110 		break;
111 	case CHIP_KABINI:
112 		fw_name = FIRMWARE_KABINI;
113 		break;
114 	case CHIP_HAWAII:
115 		fw_name = FIRMWARE_HAWAII;
116 		break;
117 	case CHIP_MULLINS:
118 		fw_name = FIRMWARE_MULLINS;
119 		break;
120 #endif
121 	case CHIP_TONGA:
122 		fw_name = FIRMWARE_TONGA;
123 		break;
124 	case CHIP_CARRIZO:
125 		fw_name = FIRMWARE_CARRIZO;
126 		break;
127 	case CHIP_FIJI:
128 		fw_name = FIRMWARE_FIJI;
129 		break;
130 	case CHIP_STONEY:
131 		fw_name = FIRMWARE_STONEY;
132 		break;
133 	case CHIP_POLARIS10:
134 		fw_name = FIRMWARE_POLARIS10;
135 		break;
136 	case CHIP_POLARIS11:
137 		fw_name = FIRMWARE_POLARIS11;
138 		break;
139 	case CHIP_POLARIS12:
140 		fw_name = FIRMWARE_POLARIS12;
141 		break;
142 	case CHIP_VEGAM:
143 		fw_name = FIRMWARE_VEGAM;
144 		break;
145 	case CHIP_VEGA10:
146 		fw_name = FIRMWARE_VEGA10;
147 		break;
148 	case CHIP_VEGA12:
149 		fw_name = FIRMWARE_VEGA12;
150 		break;
151 	case CHIP_VEGA20:
152 		fw_name = FIRMWARE_VEGA20;
153 		break;
154 
155 	default:
156 		return -EINVAL;
157 	}
158 
159 	r = request_firmware(&adev->vce.fw, fw_name, adev->dev);
160 	if (r) {
161 		dev_err(adev->dev, "amdgpu_vce: Can't load firmware \"%s\"\n",
162 			fw_name);
163 		return r;
164 	}
165 
166 	r = amdgpu_ucode_validate(adev->vce.fw);
167 	if (r) {
168 		dev_err(adev->dev, "amdgpu_vce: Can't validate firmware \"%s\"\n",
169 			fw_name);
170 		release_firmware(adev->vce.fw);
171 		adev->vce.fw = NULL;
172 		return r;
173 	}
174 
175 	hdr = (const struct common_firmware_header *)adev->vce.fw->data;
176 
177 	ucode_version = le32_to_cpu(hdr->ucode_version);
178 	version_major = (ucode_version >> 20) & 0xfff;
179 	version_minor = (ucode_version >> 8) & 0xfff;
180 	binary_id = ucode_version & 0xff;
181 	DRM_INFO("Found VCE firmware Version: %hhd.%hhd Binary ID: %hhd\n",
182 		version_major, version_minor, binary_id);
183 	adev->vce.fw_version = ((version_major << 24) | (version_minor << 16) |
184 				(binary_id << 8));
185 
186 	r = amdgpu_bo_create_kernel(adev, size, PAGE_SIZE,
187 				    AMDGPU_GEM_DOMAIN_VRAM, &adev->vce.vcpu_bo,
188 				    &adev->vce.gpu_addr, &adev->vce.cpu_addr);
189 	if (r) {
190 		dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r);
191 		return r;
192 	}
193 
194 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
195 		atomic_set(&adev->vce.handles[i], 0);
196 		adev->vce.filp[i] = NULL;
197 	}
198 
199 	INIT_DELAYED_WORK(&adev->vce.idle_work, amdgpu_vce_idle_work_handler);
200 	mutex_init(&adev->vce.idle_mutex);
201 
202 	return 0;
203 }
204 
205 /**
206  * amdgpu_vce_fini - free memory
207  *
208  * @adev: amdgpu_device pointer
209  *
210  * Last step on VCE teardown, free firmware memory
211  */
212 int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
213 {
214 	unsigned i;
215 
216 	if (adev->vce.vcpu_bo == NULL)
217 		return 0;
218 
219 	cancel_delayed_work_sync(&adev->vce.idle_work);
220 	drm_sched_entity_destroy(&adev->vce.entity);
221 
222 	amdgpu_bo_free_kernel(&adev->vce.vcpu_bo, &adev->vce.gpu_addr,
223 		(void **)&adev->vce.cpu_addr);
224 
225 	for (i = 0; i < adev->vce.num_rings; i++)
226 		amdgpu_ring_fini(&adev->vce.ring[i]);
227 
228 	release_firmware(adev->vce.fw);
229 	mutex_destroy(&adev->vce.idle_mutex);
230 
231 	return 0;
232 }
233 
234 /**
235  * amdgpu_vce_entity_init - init entity
236  *
237  * @adev: amdgpu_device pointer
238  *
239  */
240 int amdgpu_vce_entity_init(struct amdgpu_device *adev)
241 {
242 	struct amdgpu_ring *ring;
243 	struct drm_gpu_scheduler *sched;
244 	int r;
245 
246 	ring = &adev->vce.ring[0];
247 	sched = &ring->sched;
248 	r = drm_sched_entity_init(&adev->vce.entity, DRM_SCHED_PRIORITY_NORMAL,
249 				  &sched, 1, NULL);
250 	if (r != 0) {
251 		DRM_ERROR("Failed setting up VCE run queue.\n");
252 		return r;
253 	}
254 
255 	return 0;
256 }
257 
258 /**
259  * amdgpu_vce_suspend - unpin VCE fw memory
260  *
261  * @adev: amdgpu_device pointer
262  *
263  */
264 int amdgpu_vce_suspend(struct amdgpu_device *adev)
265 {
266 	int i;
267 
268 	cancel_delayed_work_sync(&adev->vce.idle_work);
269 
270 	if (adev->vce.vcpu_bo == NULL)
271 		return 0;
272 
273 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
274 		if (atomic_read(&adev->vce.handles[i]))
275 			break;
276 
277 	if (i == AMDGPU_MAX_VCE_HANDLES)
278 		return 0;
279 
280 	/* TODO: suspending running encoding sessions isn't supported */
281 	return -EINVAL;
282 }
283 
284 /**
285  * amdgpu_vce_resume - pin VCE fw memory
286  *
287  * @adev: amdgpu_device pointer
288  *
289  */
290 int amdgpu_vce_resume(struct amdgpu_device *adev)
291 {
292 	void *cpu_addr;
293 	const struct common_firmware_header *hdr;
294 	unsigned offset;
295 	int r;
296 
297 	if (adev->vce.vcpu_bo == NULL)
298 		return -EINVAL;
299 
300 	r = amdgpu_bo_reserve(adev->vce.vcpu_bo, false);
301 	if (r) {
302 		dev_err(adev->dev, "(%d) failed to reserve VCE bo\n", r);
303 		return r;
304 	}
305 
306 	r = amdgpu_bo_kmap(adev->vce.vcpu_bo, &cpu_addr);
307 	if (r) {
308 		amdgpu_bo_unreserve(adev->vce.vcpu_bo);
309 		dev_err(adev->dev, "(%d) VCE map failed\n", r);
310 		return r;
311 	}
312 
313 	hdr = (const struct common_firmware_header *)adev->vce.fw->data;
314 	offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
315 	memcpy_toio(cpu_addr, adev->vce.fw->data + offset,
316 		    adev->vce.fw->size - offset);
317 
318 	amdgpu_bo_kunmap(adev->vce.vcpu_bo);
319 
320 	amdgpu_bo_unreserve(adev->vce.vcpu_bo);
321 
322 	return 0;
323 }
324 
325 /**
326  * amdgpu_vce_idle_work_handler - power off VCE
327  *
328  * @work: pointer to work structure
329  *
330  * power of VCE when it's not used any more
331  */
332 static void amdgpu_vce_idle_work_handler(struct work_struct *work)
333 {
334 	struct amdgpu_device *adev =
335 		container_of(work, struct amdgpu_device, vce.idle_work.work);
336 	unsigned i, count = 0;
337 
338 	for (i = 0; i < adev->vce.num_rings; i++)
339 		count += amdgpu_fence_count_emitted(&adev->vce.ring[i]);
340 
341 	if (count == 0) {
342 		if (adev->pm.dpm_enabled) {
343 			amdgpu_dpm_enable_vce(adev, false);
344 		} else {
345 			amdgpu_asic_set_vce_clocks(adev, 0, 0);
346 			amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
347 							       AMD_PG_STATE_GATE);
348 			amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
349 							       AMD_CG_STATE_GATE);
350 		}
351 	} else {
352 		schedule_delayed_work(&adev->vce.idle_work, VCE_IDLE_TIMEOUT);
353 	}
354 }
355 
356 /**
357  * amdgpu_vce_ring_begin_use - power up VCE
358  *
359  * @ring: amdgpu ring
360  *
361  * Make sure VCE is powerd up when we want to use it
362  */
363 void amdgpu_vce_ring_begin_use(struct amdgpu_ring *ring)
364 {
365 	struct amdgpu_device *adev = ring->adev;
366 	bool set_clocks;
367 
368 	if (amdgpu_sriov_vf(adev))
369 		return;
370 
371 	mutex_lock(&adev->vce.idle_mutex);
372 	set_clocks = !cancel_delayed_work_sync(&adev->vce.idle_work);
373 	if (set_clocks) {
374 		if (adev->pm.dpm_enabled) {
375 			amdgpu_dpm_enable_vce(adev, true);
376 		} else {
377 			amdgpu_asic_set_vce_clocks(adev, 53300, 40000);
378 			amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
379 							       AMD_CG_STATE_UNGATE);
380 			amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCE,
381 							       AMD_PG_STATE_UNGATE);
382 
383 		}
384 	}
385 	mutex_unlock(&adev->vce.idle_mutex);
386 }
387 
388 /**
389  * amdgpu_vce_ring_end_use - power VCE down
390  *
391  * @ring: amdgpu ring
392  *
393  * Schedule work to power VCE down again
394  */
395 void amdgpu_vce_ring_end_use(struct amdgpu_ring *ring)
396 {
397 	if (!amdgpu_sriov_vf(ring->adev))
398 		schedule_delayed_work(&ring->adev->vce.idle_work, VCE_IDLE_TIMEOUT);
399 }
400 
401 /**
402  * amdgpu_vce_free_handles - free still open VCE handles
403  *
404  * @adev: amdgpu_device pointer
405  * @filp: drm file pointer
406  *
407  * Close all VCE handles still open by this file pointer
408  */
409 void amdgpu_vce_free_handles(struct amdgpu_device *adev, struct drm_file *filp)
410 {
411 	struct amdgpu_ring *ring = &adev->vce.ring[0];
412 	int i, r;
413 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
414 		uint32_t handle = atomic_read(&adev->vce.handles[i]);
415 
416 		if (!handle || adev->vce.filp[i] != filp)
417 			continue;
418 
419 		r = amdgpu_vce_get_destroy_msg(ring, handle, false, NULL);
420 		if (r)
421 			DRM_ERROR("Error destroying VCE handle (%d)!\n", r);
422 
423 		adev->vce.filp[i] = NULL;
424 		atomic_set(&adev->vce.handles[i], 0);
425 	}
426 }
427 
428 /**
429  * amdgpu_vce_get_create_msg - generate a VCE create msg
430  *
431  * @adev: amdgpu_device pointer
432  * @ring: ring we should submit the msg to
433  * @handle: VCE session handle to use
434  * @fence: optional fence to return
435  *
436  * Open up a stream for HW test
437  */
438 static int amdgpu_vce_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
439 				     struct amdgpu_bo *bo,
440 				     struct dma_fence **fence)
441 {
442 	const unsigned ib_size_dw = 1024;
443 	struct amdgpu_job *job;
444 	struct amdgpu_ib *ib;
445 	struct dma_fence *f = NULL;
446 	uint64_t addr;
447 	int i, r;
448 
449 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
450 				     AMDGPU_IB_POOL_DIRECT, &job);
451 	if (r)
452 		return r;
453 
454 	ib = &job->ibs[0];
455 
456 	addr = amdgpu_bo_gpu_offset(bo);
457 
458 	/* stitch together an VCE create msg */
459 	ib->length_dw = 0;
460 	ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
461 	ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
462 	ib->ptr[ib->length_dw++] = handle;
463 
464 	if ((ring->adev->vce.fw_version >> 24) >= 52)
465 		ib->ptr[ib->length_dw++] = 0x00000040; /* len */
466 	else
467 		ib->ptr[ib->length_dw++] = 0x00000030; /* len */
468 	ib->ptr[ib->length_dw++] = 0x01000001; /* create cmd */
469 	ib->ptr[ib->length_dw++] = 0x00000000;
470 	ib->ptr[ib->length_dw++] = 0x00000042;
471 	ib->ptr[ib->length_dw++] = 0x0000000a;
472 	ib->ptr[ib->length_dw++] = 0x00000001;
473 	ib->ptr[ib->length_dw++] = 0x00000080;
474 	ib->ptr[ib->length_dw++] = 0x00000060;
475 	ib->ptr[ib->length_dw++] = 0x00000100;
476 	ib->ptr[ib->length_dw++] = 0x00000100;
477 	ib->ptr[ib->length_dw++] = 0x0000000c;
478 	ib->ptr[ib->length_dw++] = 0x00000000;
479 	if ((ring->adev->vce.fw_version >> 24) >= 52) {
480 		ib->ptr[ib->length_dw++] = 0x00000000;
481 		ib->ptr[ib->length_dw++] = 0x00000000;
482 		ib->ptr[ib->length_dw++] = 0x00000000;
483 		ib->ptr[ib->length_dw++] = 0x00000000;
484 	}
485 
486 	ib->ptr[ib->length_dw++] = 0x00000014; /* len */
487 	ib->ptr[ib->length_dw++] = 0x05000005; /* feedback buffer */
488 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
489 	ib->ptr[ib->length_dw++] = addr;
490 	ib->ptr[ib->length_dw++] = 0x00000001;
491 
492 	for (i = ib->length_dw; i < ib_size_dw; ++i)
493 		ib->ptr[i] = 0x0;
494 
495 	r = amdgpu_job_submit_direct(job, ring, &f);
496 	if (r)
497 		goto err;
498 
499 	if (fence)
500 		*fence = dma_fence_get(f);
501 	dma_fence_put(f);
502 	return 0;
503 
504 err:
505 	amdgpu_job_free(job);
506 	return r;
507 }
508 
509 /**
510  * amdgpu_vce_get_destroy_msg - generate a VCE destroy msg
511  *
512  * @adev: amdgpu_device pointer
513  * @ring: ring we should submit the msg to
514  * @handle: VCE session handle to use
515  * @fence: optional fence to return
516  *
517  * Close up a stream for HW test or if userspace failed to do so
518  */
519 static int amdgpu_vce_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
520 				      bool direct, struct dma_fence **fence)
521 {
522 	const unsigned ib_size_dw = 1024;
523 	struct amdgpu_job *job;
524 	struct amdgpu_ib *ib;
525 	struct dma_fence *f = NULL;
526 	int i, r;
527 
528 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
529 				     direct ? AMDGPU_IB_POOL_DIRECT :
530 				     AMDGPU_IB_POOL_DELAYED, &job);
531 	if (r)
532 		return r;
533 
534 	ib = &job->ibs[0];
535 
536 	/* stitch together an VCE destroy msg */
537 	ib->length_dw = 0;
538 	ib->ptr[ib->length_dw++] = 0x0000000c; /* len */
539 	ib->ptr[ib->length_dw++] = 0x00000001; /* session cmd */
540 	ib->ptr[ib->length_dw++] = handle;
541 
542 	ib->ptr[ib->length_dw++] = 0x00000020; /* len */
543 	ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
544 	ib->ptr[ib->length_dw++] = 0xffffffff; /* next task info, set to 0xffffffff if no */
545 	ib->ptr[ib->length_dw++] = 0x00000001; /* destroy session */
546 	ib->ptr[ib->length_dw++] = 0x00000000;
547 	ib->ptr[ib->length_dw++] = 0x00000000;
548 	ib->ptr[ib->length_dw++] = 0xffffffff; /* feedback is not needed, set to 0xffffffff and firmware will not output feedback */
549 	ib->ptr[ib->length_dw++] = 0x00000000;
550 
551 	ib->ptr[ib->length_dw++] = 0x00000008; /* len */
552 	ib->ptr[ib->length_dw++] = 0x02000001; /* destroy cmd */
553 
554 	for (i = ib->length_dw; i < ib_size_dw; ++i)
555 		ib->ptr[i] = 0x0;
556 
557 	if (direct)
558 		r = amdgpu_job_submit_direct(job, ring, &f);
559 	else
560 		r = amdgpu_job_submit(job, &ring->adev->vce.entity,
561 				      AMDGPU_FENCE_OWNER_UNDEFINED, &f);
562 	if (r)
563 		goto err;
564 
565 	if (fence)
566 		*fence = dma_fence_get(f);
567 	dma_fence_put(f);
568 	return 0;
569 
570 err:
571 	amdgpu_job_free(job);
572 	return r;
573 }
574 
575 /**
576  * amdgpu_vce_cs_validate_bo - make sure not to cross 4GB boundary
577  *
578  * @p: parser context
579  * @lo: address of lower dword
580  * @hi: address of higher dword
581  * @size: minimum size
582  * @index: bs/fb index
583  *
584  * Make sure that no BO cross a 4GB boundary.
585  */
586 static int amdgpu_vce_validate_bo(struct amdgpu_cs_parser *p, uint32_t ib_idx,
587 				  int lo, int hi, unsigned size, int32_t index)
588 {
589 	int64_t offset = ((uint64_t)size) * ((int64_t)index);
590 	struct ttm_operation_ctx ctx = { false, false };
591 	struct amdgpu_bo_va_mapping *mapping;
592 	unsigned i, fpfn, lpfn;
593 	struct amdgpu_bo *bo;
594 	uint64_t addr;
595 	int r;
596 
597 	addr = ((uint64_t)amdgpu_get_ib_value(p, ib_idx, lo)) |
598 	       ((uint64_t)amdgpu_get_ib_value(p, ib_idx, hi)) << 32;
599 	if (index >= 0) {
600 		addr += offset;
601 		fpfn = PAGE_ALIGN(offset) >> PAGE_SHIFT;
602 		lpfn = 0x100000000ULL >> PAGE_SHIFT;
603 	} else {
604 		fpfn = 0;
605 		lpfn = (0x100000000ULL - PAGE_ALIGN(offset)) >> PAGE_SHIFT;
606 	}
607 
608 	r = amdgpu_cs_find_mapping(p, addr, &bo, &mapping);
609 	if (r) {
610 		DRM_ERROR("Can't find BO for addr 0x%010Lx %d %d %d %d\n",
611 			  addr, lo, hi, size, index);
612 		return r;
613 	}
614 
615 	for (i = 0; i < bo->placement.num_placement; ++i) {
616 		bo->placements[i].fpfn = max(bo->placements[i].fpfn, fpfn);
617 		bo->placements[i].lpfn = bo->placements[i].lpfn ?
618 			min(bo->placements[i].lpfn, lpfn) : lpfn;
619 	}
620 	return ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
621 }
622 
623 
624 /**
625  * amdgpu_vce_cs_reloc - command submission relocation
626  *
627  * @p: parser context
628  * @lo: address of lower dword
629  * @hi: address of higher dword
630  * @size: minimum size
631  *
632  * Patch relocation inside command stream with real buffer address
633  */
634 static int amdgpu_vce_cs_reloc(struct amdgpu_cs_parser *p, uint32_t ib_idx,
635 			       int lo, int hi, unsigned size, uint32_t index)
636 {
637 	struct amdgpu_bo_va_mapping *mapping;
638 	struct amdgpu_bo *bo;
639 	uint64_t addr;
640 	int r;
641 
642 	if (index == 0xffffffff)
643 		index = 0;
644 
645 	addr = ((uint64_t)amdgpu_get_ib_value(p, ib_idx, lo)) |
646 	       ((uint64_t)amdgpu_get_ib_value(p, ib_idx, hi)) << 32;
647 	addr += ((uint64_t)size) * ((uint64_t)index);
648 
649 	r = amdgpu_cs_find_mapping(p, addr, &bo, &mapping);
650 	if (r) {
651 		DRM_ERROR("Can't find BO for addr 0x%010Lx %d %d %d %d\n",
652 			  addr, lo, hi, size, index);
653 		return r;
654 	}
655 
656 	if ((addr + (uint64_t)size) >
657 	    (mapping->last + 1) * AMDGPU_GPU_PAGE_SIZE) {
658 		DRM_ERROR("BO too small for addr 0x%010Lx %d %d\n",
659 			  addr, lo, hi);
660 		return -EINVAL;
661 	}
662 
663 	addr -= mapping->start * AMDGPU_GPU_PAGE_SIZE;
664 	addr += amdgpu_bo_gpu_offset(bo);
665 	addr -= ((uint64_t)size) * ((uint64_t)index);
666 
667 	amdgpu_set_ib_value(p, ib_idx, lo, lower_32_bits(addr));
668 	amdgpu_set_ib_value(p, ib_idx, hi, upper_32_bits(addr));
669 
670 	return 0;
671 }
672 
673 /**
674  * amdgpu_vce_validate_handle - validate stream handle
675  *
676  * @p: parser context
677  * @handle: handle to validate
678  * @allocated: allocated a new handle?
679  *
680  * Validates the handle and return the found session index or -EINVAL
681  * we we don't have another free session index.
682  */
683 static int amdgpu_vce_validate_handle(struct amdgpu_cs_parser *p,
684 				      uint32_t handle, uint32_t *allocated)
685 {
686 	unsigned i;
687 
688 	/* validate the handle */
689 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
690 		if (atomic_read(&p->adev->vce.handles[i]) == handle) {
691 			if (p->adev->vce.filp[i] != p->filp) {
692 				DRM_ERROR("VCE handle collision detected!\n");
693 				return -EINVAL;
694 			}
695 			return i;
696 		}
697 	}
698 
699 	/* handle not found try to alloc a new one */
700 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i) {
701 		if (!atomic_cmpxchg(&p->adev->vce.handles[i], 0, handle)) {
702 			p->adev->vce.filp[i] = p->filp;
703 			p->adev->vce.img_size[i] = 0;
704 			*allocated |= 1 << i;
705 			return i;
706 		}
707 	}
708 
709 	DRM_ERROR("No more free VCE handles!\n");
710 	return -EINVAL;
711 }
712 
713 /**
714  * amdgpu_vce_cs_parse - parse and validate the command stream
715  *
716  * @p: parser context
717  *
718  */
719 int amdgpu_vce_ring_parse_cs(struct amdgpu_cs_parser *p, uint32_t ib_idx)
720 {
721 	struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
722 	unsigned fb_idx = 0, bs_idx = 0;
723 	int session_idx = -1;
724 	uint32_t destroyed = 0;
725 	uint32_t created = 0;
726 	uint32_t allocated = 0;
727 	uint32_t tmp, handle = 0;
728 	uint32_t *size = &tmp;
729 	unsigned idx;
730 	int i, r = 0;
731 
732 	p->job->vm = NULL;
733 	ib->gpu_addr = amdgpu_sa_bo_gpu_addr(ib->sa_bo);
734 
735 	for (idx = 0; idx < ib->length_dw;) {
736 		uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx);
737 		uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1);
738 
739 		if ((len < 8) || (len & 3)) {
740 			DRM_ERROR("invalid VCE command length (%d)!\n", len);
741 			r = -EINVAL;
742 			goto out;
743 		}
744 
745 		switch (cmd) {
746 		case 0x00000002: /* task info */
747 			fb_idx = amdgpu_get_ib_value(p, ib_idx, idx + 6);
748 			bs_idx = amdgpu_get_ib_value(p, ib_idx, idx + 7);
749 			break;
750 
751 		case 0x03000001: /* encode */
752 			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 10,
753 						   idx + 9, 0, 0);
754 			if (r)
755 				goto out;
756 
757 			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 12,
758 						   idx + 11, 0, 0);
759 			if (r)
760 				goto out;
761 			break;
762 
763 		case 0x05000001: /* context buffer */
764 			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 3,
765 						   idx + 2, 0, 0);
766 			if (r)
767 				goto out;
768 			break;
769 
770 		case 0x05000004: /* video bitstream buffer */
771 			tmp = amdgpu_get_ib_value(p, ib_idx, idx + 4);
772 			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 3, idx + 2,
773 						   tmp, bs_idx);
774 			if (r)
775 				goto out;
776 			break;
777 
778 		case 0x05000005: /* feedback buffer */
779 			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 3, idx + 2,
780 						   4096, fb_idx);
781 			if (r)
782 				goto out;
783 			break;
784 
785 		case 0x0500000d: /* MV buffer */
786 			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 3,
787 							idx + 2, 0, 0);
788 			if (r)
789 				goto out;
790 
791 			r = amdgpu_vce_validate_bo(p, ib_idx, idx + 8,
792 							idx + 7, 0, 0);
793 			if (r)
794 				goto out;
795 			break;
796 		}
797 
798 		idx += len / 4;
799 	}
800 
801 	for (idx = 0; idx < ib->length_dw;) {
802 		uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx);
803 		uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1);
804 
805 		switch (cmd) {
806 		case 0x00000001: /* session */
807 			handle = amdgpu_get_ib_value(p, ib_idx, idx + 2);
808 			session_idx = amdgpu_vce_validate_handle(p, handle,
809 								 &allocated);
810 			if (session_idx < 0) {
811 				r = session_idx;
812 				goto out;
813 			}
814 			size = &p->adev->vce.img_size[session_idx];
815 			break;
816 
817 		case 0x00000002: /* task info */
818 			fb_idx = amdgpu_get_ib_value(p, ib_idx, idx + 6);
819 			bs_idx = amdgpu_get_ib_value(p, ib_idx, idx + 7);
820 			break;
821 
822 		case 0x01000001: /* create */
823 			created |= 1 << session_idx;
824 			if (destroyed & (1 << session_idx)) {
825 				destroyed &= ~(1 << session_idx);
826 				allocated |= 1 << session_idx;
827 
828 			} else if (!(allocated & (1 << session_idx))) {
829 				DRM_ERROR("Handle already in use!\n");
830 				r = -EINVAL;
831 				goto out;
832 			}
833 
834 			*size = amdgpu_get_ib_value(p, ib_idx, idx + 8) *
835 				amdgpu_get_ib_value(p, ib_idx, idx + 10) *
836 				8 * 3 / 2;
837 			break;
838 
839 		case 0x04000001: /* config extension */
840 		case 0x04000002: /* pic control */
841 		case 0x04000005: /* rate control */
842 		case 0x04000007: /* motion estimation */
843 		case 0x04000008: /* rdo */
844 		case 0x04000009: /* vui */
845 		case 0x05000002: /* auxiliary buffer */
846 		case 0x05000009: /* clock table */
847 			break;
848 
849 		case 0x0500000c: /* hw config */
850 			switch (p->adev->asic_type) {
851 #ifdef CONFIG_DRM_AMDGPU_CIK
852 			case CHIP_KAVERI:
853 			case CHIP_MULLINS:
854 #endif
855 			case CHIP_CARRIZO:
856 				break;
857 			default:
858 				r = -EINVAL;
859 				goto out;
860 			}
861 			break;
862 
863 		case 0x03000001: /* encode */
864 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 10, idx + 9,
865 						*size, 0);
866 			if (r)
867 				goto out;
868 
869 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 12, idx + 11,
870 						*size / 3, 0);
871 			if (r)
872 				goto out;
873 			break;
874 
875 		case 0x02000001: /* destroy */
876 			destroyed |= 1 << session_idx;
877 			break;
878 
879 		case 0x05000001: /* context buffer */
880 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
881 						*size * 2, 0);
882 			if (r)
883 				goto out;
884 			break;
885 
886 		case 0x05000004: /* video bitstream buffer */
887 			tmp = amdgpu_get_ib_value(p, ib_idx, idx + 4);
888 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
889 						tmp, bs_idx);
890 			if (r)
891 				goto out;
892 			break;
893 
894 		case 0x05000005: /* feedback buffer */
895 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3, idx + 2,
896 						4096, fb_idx);
897 			if (r)
898 				goto out;
899 			break;
900 
901 		case 0x0500000d: /* MV buffer */
902 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 3,
903 							idx + 2, *size, 0);
904 			if (r)
905 				goto out;
906 
907 			r = amdgpu_vce_cs_reloc(p, ib_idx, idx + 8,
908 							idx + 7, *size / 12, 0);
909 			if (r)
910 				goto out;
911 			break;
912 
913 		default:
914 			DRM_ERROR("invalid VCE command (0x%x)!\n", cmd);
915 			r = -EINVAL;
916 			goto out;
917 		}
918 
919 		if (session_idx == -1) {
920 			DRM_ERROR("no session command at start of IB\n");
921 			r = -EINVAL;
922 			goto out;
923 		}
924 
925 		idx += len / 4;
926 	}
927 
928 	if (allocated & ~created) {
929 		DRM_ERROR("New session without create command!\n");
930 		r = -ENOENT;
931 	}
932 
933 out:
934 	if (!r) {
935 		/* No error, free all destroyed handle slots */
936 		tmp = destroyed;
937 	} else {
938 		/* Error during parsing, free all allocated handle slots */
939 		tmp = allocated;
940 	}
941 
942 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
943 		if (tmp & (1 << i))
944 			atomic_set(&p->adev->vce.handles[i], 0);
945 
946 	return r;
947 }
948 
949 /**
950  * amdgpu_vce_cs_parse_vm - parse the command stream in VM mode
951  *
952  * @p: parser context
953  *
954  */
955 int amdgpu_vce_ring_parse_cs_vm(struct amdgpu_cs_parser *p, uint32_t ib_idx)
956 {
957 	struct amdgpu_ib *ib = &p->job->ibs[ib_idx];
958 	int session_idx = -1;
959 	uint32_t destroyed = 0;
960 	uint32_t created = 0;
961 	uint32_t allocated = 0;
962 	uint32_t tmp, handle = 0;
963 	int i, r = 0, idx = 0;
964 
965 	while (idx < ib->length_dw) {
966 		uint32_t len = amdgpu_get_ib_value(p, ib_idx, idx);
967 		uint32_t cmd = amdgpu_get_ib_value(p, ib_idx, idx + 1);
968 
969 		if ((len < 8) || (len & 3)) {
970 			DRM_ERROR("invalid VCE command length (%d)!\n", len);
971 			r = -EINVAL;
972 			goto out;
973 		}
974 
975 		switch (cmd) {
976 		case 0x00000001: /* session */
977 			handle = amdgpu_get_ib_value(p, ib_idx, idx + 2);
978 			session_idx = amdgpu_vce_validate_handle(p, handle,
979 								 &allocated);
980 			if (session_idx < 0) {
981 				r = session_idx;
982 				goto out;
983 			}
984 			break;
985 
986 		case 0x01000001: /* create */
987 			created |= 1 << session_idx;
988 			if (destroyed & (1 << session_idx)) {
989 				destroyed &= ~(1 << session_idx);
990 				allocated |= 1 << session_idx;
991 
992 			} else if (!(allocated & (1 << session_idx))) {
993 				DRM_ERROR("Handle already in use!\n");
994 				r = -EINVAL;
995 				goto out;
996 			}
997 
998 			break;
999 
1000 		case 0x02000001: /* destroy */
1001 			destroyed |= 1 << session_idx;
1002 			break;
1003 
1004 		default:
1005 			break;
1006 		}
1007 
1008 		if (session_idx == -1) {
1009 			DRM_ERROR("no session command at start of IB\n");
1010 			r = -EINVAL;
1011 			goto out;
1012 		}
1013 
1014 		idx += len / 4;
1015 	}
1016 
1017 	if (allocated & ~created) {
1018 		DRM_ERROR("New session without create command!\n");
1019 		r = -ENOENT;
1020 	}
1021 
1022 out:
1023 	if (!r) {
1024 		/* No error, free all destroyed handle slots */
1025 		tmp = destroyed;
1026 		amdgpu_ib_free(p->adev, ib, NULL);
1027 	} else {
1028 		/* Error during parsing, free all allocated handle slots */
1029 		tmp = allocated;
1030 	}
1031 
1032 	for (i = 0; i < AMDGPU_MAX_VCE_HANDLES; ++i)
1033 		if (tmp & (1 << i))
1034 			atomic_set(&p->adev->vce.handles[i], 0);
1035 
1036 	return r;
1037 }
1038 
1039 /**
1040  * amdgpu_vce_ring_emit_ib - execute indirect buffer
1041  *
1042  * @ring: engine to use
1043  * @ib: the IB to execute
1044  *
1045  */
1046 void amdgpu_vce_ring_emit_ib(struct amdgpu_ring *ring,
1047 				struct amdgpu_job *job,
1048 				struct amdgpu_ib *ib,
1049 				uint32_t flags)
1050 {
1051 	amdgpu_ring_write(ring, VCE_CMD_IB);
1052 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1053 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1054 	amdgpu_ring_write(ring, ib->length_dw);
1055 }
1056 
1057 /**
1058  * amdgpu_vce_ring_emit_fence - add a fence command to the ring
1059  *
1060  * @ring: engine to use
1061  * @fence: the fence
1062  *
1063  */
1064 void amdgpu_vce_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
1065 				unsigned flags)
1066 {
1067 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1068 
1069 	amdgpu_ring_write(ring, VCE_CMD_FENCE);
1070 	amdgpu_ring_write(ring, addr);
1071 	amdgpu_ring_write(ring, upper_32_bits(addr));
1072 	amdgpu_ring_write(ring, seq);
1073 	amdgpu_ring_write(ring, VCE_CMD_TRAP);
1074 	amdgpu_ring_write(ring, VCE_CMD_END);
1075 }
1076 
1077 /**
1078  * amdgpu_vce_ring_test_ring - test if VCE ring is working
1079  *
1080  * @ring: the engine to test on
1081  *
1082  */
1083 int amdgpu_vce_ring_test_ring(struct amdgpu_ring *ring)
1084 {
1085 	struct amdgpu_device *adev = ring->adev;
1086 	uint32_t rptr;
1087 	unsigned i;
1088 	int r, timeout = adev->usec_timeout;
1089 
1090 	/* skip ring test for sriov*/
1091 	if (amdgpu_sriov_vf(adev))
1092 		return 0;
1093 
1094 	r = amdgpu_ring_alloc(ring, 16);
1095 	if (r)
1096 		return r;
1097 
1098 	rptr = amdgpu_ring_get_rptr(ring);
1099 
1100 	amdgpu_ring_write(ring, VCE_CMD_END);
1101 	amdgpu_ring_commit(ring);
1102 
1103 	for (i = 0; i < timeout; i++) {
1104 		if (amdgpu_ring_get_rptr(ring) != rptr)
1105 			break;
1106 		udelay(1);
1107 	}
1108 
1109 	if (i >= timeout)
1110 		r = -ETIMEDOUT;
1111 
1112 	return r;
1113 }
1114 
1115 /**
1116  * amdgpu_vce_ring_test_ib - test if VCE IBs are working
1117  *
1118  * @ring: the engine to test on
1119  *
1120  */
1121 int amdgpu_vce_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1122 {
1123 	struct dma_fence *fence = NULL;
1124 	struct amdgpu_bo *bo = NULL;
1125 	long r;
1126 
1127 	/* skip vce ring1/2 ib test for now, since it's not reliable */
1128 	if (ring != &ring->adev->vce.ring[0])
1129 		return 0;
1130 
1131 	r = amdgpu_bo_create_reserved(ring->adev, 512, PAGE_SIZE,
1132 				      AMDGPU_GEM_DOMAIN_VRAM,
1133 				      &bo, NULL, NULL);
1134 	if (r)
1135 		return r;
1136 
1137 	r = amdgpu_vce_get_create_msg(ring, 1, bo, NULL);
1138 	if (r)
1139 		goto error;
1140 
1141 	r = amdgpu_vce_get_destroy_msg(ring, 1, true, &fence);
1142 	if (r)
1143 		goto error;
1144 
1145 	r = dma_fence_wait_timeout(fence, false, timeout);
1146 	if (r == 0)
1147 		r = -ETIMEDOUT;
1148 	else if (r > 0)
1149 		r = 0;
1150 
1151 error:
1152 	dma_fence_put(fence);
1153 	amdgpu_bo_unreserve(bo);
1154 	amdgpu_bo_unref(&bo);
1155 	return r;
1156 }
1157