1 /*
2  * Copyright 2016 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  */
26 
27 #include <linux/firmware.h>
28 #include <linux/module.h>
29 #include <linux/pci.h>
30 
31 #include "amdgpu.h"
32 #include "amdgpu_pm.h"
33 #include "amdgpu_vcn.h"
34 #include "soc15d.h"
35 
36 /* Firmware Names */
37 #define FIRMWARE_RAVEN		"amdgpu/raven_vcn.bin"
38 #define FIRMWARE_PICASSO	"amdgpu/picasso_vcn.bin"
39 #define FIRMWARE_RAVEN2		"amdgpu/raven2_vcn.bin"
40 #define FIRMWARE_ARCTURUS	"amdgpu/arcturus_vcn.bin"
41 #define FIRMWARE_RENOIR		"amdgpu/renoir_vcn.bin"
42 #define FIRMWARE_GREEN_SARDINE	"amdgpu/green_sardine_vcn.bin"
43 #define FIRMWARE_NAVI10		"amdgpu/navi10_vcn.bin"
44 #define FIRMWARE_NAVI14		"amdgpu/navi14_vcn.bin"
45 #define FIRMWARE_NAVI12		"amdgpu/navi12_vcn.bin"
46 #define FIRMWARE_SIENNA_CICHLID	"amdgpu/sienna_cichlid_vcn.bin"
47 #define FIRMWARE_NAVY_FLOUNDER	"amdgpu/navy_flounder_vcn.bin"
48 #define FIRMWARE_VANGOGH	"amdgpu/vangogh_vcn.bin"
49 #define FIRMWARE_DIMGREY_CAVEFISH	"amdgpu/dimgrey_cavefish_vcn.bin"
50 
51 MODULE_FIRMWARE(FIRMWARE_RAVEN);
52 MODULE_FIRMWARE(FIRMWARE_PICASSO);
53 MODULE_FIRMWARE(FIRMWARE_RAVEN2);
54 MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
55 MODULE_FIRMWARE(FIRMWARE_RENOIR);
56 MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
57 MODULE_FIRMWARE(FIRMWARE_NAVI10);
58 MODULE_FIRMWARE(FIRMWARE_NAVI14);
59 MODULE_FIRMWARE(FIRMWARE_NAVI12);
60 MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
61 MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
62 MODULE_FIRMWARE(FIRMWARE_VANGOGH);
63 MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
64 
65 static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
66 
67 int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
68 {
69 	unsigned long bo_size;
70 	const char *fw_name;
71 	const struct common_firmware_header *hdr;
72 	unsigned char fw_check;
73 	int i, r;
74 
75 	INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
76 	mutex_init(&adev->vcn.vcn_pg_lock);
77 	mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
78 	atomic_set(&adev->vcn.total_submission_cnt, 0);
79 	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
80 		atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
81 
82 	switch (adev->asic_type) {
83 	case CHIP_RAVEN:
84 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
85 			fw_name = FIRMWARE_RAVEN2;
86 		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
87 			fw_name = FIRMWARE_PICASSO;
88 		else
89 			fw_name = FIRMWARE_RAVEN;
90 		break;
91 	case CHIP_ARCTURUS:
92 		fw_name = FIRMWARE_ARCTURUS;
93 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
94 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
95 			adev->vcn.indirect_sram = true;
96 		break;
97 	case CHIP_RENOIR:
98 		if (adev->apu_flags & AMD_APU_IS_RENOIR)
99 			fw_name = FIRMWARE_RENOIR;
100 		else
101 			fw_name = FIRMWARE_GREEN_SARDINE;
102 
103 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
104 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
105 			adev->vcn.indirect_sram = true;
106 		break;
107 	case CHIP_NAVI10:
108 		fw_name = FIRMWARE_NAVI10;
109 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
110 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
111 			adev->vcn.indirect_sram = true;
112 		break;
113 	case CHIP_NAVI14:
114 		fw_name = FIRMWARE_NAVI14;
115 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
116 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
117 			adev->vcn.indirect_sram = true;
118 		break;
119 	case CHIP_NAVI12:
120 		fw_name = FIRMWARE_NAVI12;
121 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
122 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
123 			adev->vcn.indirect_sram = true;
124 		break;
125 	case CHIP_SIENNA_CICHLID:
126 		fw_name = FIRMWARE_SIENNA_CICHLID;
127 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
128 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
129 			adev->vcn.indirect_sram = true;
130 		break;
131 	case CHIP_NAVY_FLOUNDER:
132 		fw_name = FIRMWARE_NAVY_FLOUNDER;
133 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
134 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
135 			adev->vcn.indirect_sram = true;
136 		break;
137 	case CHIP_VANGOGH:
138 		fw_name = FIRMWARE_VANGOGH;
139 		break;
140 	case CHIP_DIMGREY_CAVEFISH:
141 		fw_name = FIRMWARE_DIMGREY_CAVEFISH;
142 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
143 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
144 			adev->vcn.indirect_sram = true;
145 		break;
146 	default:
147 		return -EINVAL;
148 	}
149 
150 	r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
151 	if (r) {
152 		dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
153 			fw_name);
154 		return r;
155 	}
156 
157 	r = amdgpu_ucode_validate(adev->vcn.fw);
158 	if (r) {
159 		dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
160 			fw_name);
161 		release_firmware(adev->vcn.fw);
162 		adev->vcn.fw = NULL;
163 		return r;
164 	}
165 
166 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
167 	adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
168 
169 	/* Bit 20-23, it is encode major and non-zero for new naming convention.
170 	 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
171 	 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
172 	 * is zero in old naming convention, this field is always zero so far.
173 	 * These four bits are used to tell which naming convention is present.
174 	 */
175 	fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
176 	if (fw_check) {
177 		unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
178 
179 		fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
180 		enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
181 		enc_major = fw_check;
182 		dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
183 		vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
184 		DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
185 			enc_major, enc_minor, dec_ver, vep, fw_rev);
186 	} else {
187 		unsigned int version_major, version_minor, family_id;
188 
189 		family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
190 		version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
191 		version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
192 		DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
193 			version_major, version_minor, family_id);
194 	}
195 
196 	bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
197 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
198 		bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
199 	bo_size += AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
200 
201 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
202 		if (adev->vcn.harvest_config & (1 << i))
203 			continue;
204 
205 		r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
206 						AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].vcpu_bo,
207 						&adev->vcn.inst[i].gpu_addr, &adev->vcn.inst[i].cpu_addr);
208 		if (r) {
209 			dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
210 			return r;
211 		}
212 
213 		adev->vcn.inst[i].fw_shared_cpu_addr = adev->vcn.inst[i].cpu_addr +
214 				bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
215 		adev->vcn.inst[i].fw_shared_gpu_addr = adev->vcn.inst[i].gpu_addr +
216 				bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
217 
218 		if (adev->vcn.indirect_sram) {
219 			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
220 					AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
221 					&adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
222 			if (r) {
223 				dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
224 				return r;
225 			}
226 		}
227 	}
228 
229 	return 0;
230 }
231 
232 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
233 {
234 	int i, j;
235 
236 	cancel_delayed_work_sync(&adev->vcn.idle_work);
237 
238 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
239 		if (adev->vcn.harvest_config & (1 << j))
240 			continue;
241 
242 		if (adev->vcn.indirect_sram) {
243 			amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo,
244 						  &adev->vcn.inst[j].dpg_sram_gpu_addr,
245 						  (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
246 		}
247 		kvfree(adev->vcn.inst[j].saved_bo);
248 
249 		amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
250 					  &adev->vcn.inst[j].gpu_addr,
251 					  (void **)&adev->vcn.inst[j].cpu_addr);
252 
253 		amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
254 
255 		for (i = 0; i < adev->vcn.num_enc_rings; ++i)
256 			amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
257 	}
258 
259 	release_firmware(adev->vcn.fw);
260 	mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
261 	mutex_destroy(&adev->vcn.vcn_pg_lock);
262 
263 	return 0;
264 }
265 
266 int amdgpu_vcn_suspend(struct amdgpu_device *adev)
267 {
268 	unsigned size;
269 	void *ptr;
270 	int i;
271 
272 	cancel_delayed_work_sync(&adev->vcn.idle_work);
273 
274 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
275 		if (adev->vcn.harvest_config & (1 << i))
276 			continue;
277 		if (adev->vcn.inst[i].vcpu_bo == NULL)
278 			return 0;
279 
280 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
281 		ptr = adev->vcn.inst[i].cpu_addr;
282 
283 		adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
284 		if (!adev->vcn.inst[i].saved_bo)
285 			return -ENOMEM;
286 
287 		memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
288 	}
289 	return 0;
290 }
291 
292 int amdgpu_vcn_resume(struct amdgpu_device *adev)
293 {
294 	unsigned size;
295 	void *ptr;
296 	int i;
297 
298 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
299 		if (adev->vcn.harvest_config & (1 << i))
300 			continue;
301 		if (adev->vcn.inst[i].vcpu_bo == NULL)
302 			return -EINVAL;
303 
304 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
305 		ptr = adev->vcn.inst[i].cpu_addr;
306 
307 		if (adev->vcn.inst[i].saved_bo != NULL) {
308 			memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
309 			kvfree(adev->vcn.inst[i].saved_bo);
310 			adev->vcn.inst[i].saved_bo = NULL;
311 		} else {
312 			const struct common_firmware_header *hdr;
313 			unsigned offset;
314 
315 			hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
316 			if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
317 				offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
318 				memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
319 					    le32_to_cpu(hdr->ucode_size_bytes));
320 				size -= le32_to_cpu(hdr->ucode_size_bytes);
321 				ptr += le32_to_cpu(hdr->ucode_size_bytes);
322 			}
323 			memset_io(ptr, 0, size);
324 		}
325 	}
326 	return 0;
327 }
328 
329 static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
330 {
331 	struct amdgpu_device *adev =
332 		container_of(work, struct amdgpu_device, vcn.idle_work.work);
333 	unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
334 	unsigned int i, j;
335 	int r = 0;
336 
337 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
338 		if (adev->vcn.harvest_config & (1 << j))
339 			continue;
340 
341 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
342 			fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
343 		}
344 
345 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
346 			struct dpg_pause_state new_state;
347 
348 			if (fence[j] ||
349 				unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
350 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
351 			else
352 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
353 
354 			adev->vcn.pause_dpg_mode(adev, j, &new_state);
355 		}
356 
357 		fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
358 		fences += fence[j];
359 	}
360 
361 	if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
362 		amdgpu_gfx_off_ctrl(adev, true);
363 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
364 		       AMD_PG_STATE_GATE);
365 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
366 				false);
367 		if (r)
368 			dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
369 	} else {
370 		schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
371 	}
372 }
373 
374 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
375 {
376 	struct amdgpu_device *adev = ring->adev;
377 	int r = 0;
378 
379 	atomic_inc(&adev->vcn.total_submission_cnt);
380 
381 	if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
382 		amdgpu_gfx_off_ctrl(adev, false);
383 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
384 				true);
385 		if (r)
386 			dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
387 	}
388 
389 	mutex_lock(&adev->vcn.vcn_pg_lock);
390 	amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
391 	       AMD_PG_STATE_UNGATE);
392 
393 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
394 		struct dpg_pause_state new_state;
395 
396 		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
397 			atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
398 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
399 		} else {
400 			unsigned int fences = 0;
401 			unsigned int i;
402 
403 			for (i = 0; i < adev->vcn.num_enc_rings; ++i)
404 				fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
405 
406 			if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
407 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
408 			else
409 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
410 		}
411 
412 		adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
413 	}
414 	mutex_unlock(&adev->vcn.vcn_pg_lock);
415 }
416 
417 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
418 {
419 	if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
420 		ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
421 		atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
422 
423 	atomic_dec(&ring->adev->vcn.total_submission_cnt);
424 
425 	schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
426 }
427 
428 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
429 {
430 	struct amdgpu_device *adev = ring->adev;
431 	uint32_t tmp = 0;
432 	unsigned i;
433 	int r;
434 
435 	/* VCN in SRIOV does not support direct register read/write */
436 	if (amdgpu_sriov_vf(adev))
437 		return 0;
438 
439 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
440 	r = amdgpu_ring_alloc(ring, 3);
441 	if (r)
442 		return r;
443 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
444 	amdgpu_ring_write(ring, 0xDEADBEEF);
445 	amdgpu_ring_commit(ring);
446 	for (i = 0; i < adev->usec_timeout; i++) {
447 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
448 		if (tmp == 0xDEADBEEF)
449 			break;
450 		udelay(1);
451 	}
452 
453 	if (i >= adev->usec_timeout)
454 		r = -ETIMEDOUT;
455 
456 	return r;
457 }
458 
459 int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
460 {
461 	struct amdgpu_device *adev = ring->adev;
462 	uint32_t rptr;
463 	unsigned int i;
464 	int r;
465 
466 	if (amdgpu_sriov_vf(adev))
467 		return 0;
468 
469 	r = amdgpu_ring_alloc(ring, 16);
470 	if (r)
471 		return r;
472 
473 	rptr = amdgpu_ring_get_rptr(ring);
474 
475 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
476 	amdgpu_ring_commit(ring);
477 
478 	for (i = 0; i < adev->usec_timeout; i++) {
479 		if (amdgpu_ring_get_rptr(ring) != rptr)
480 			break;
481 		udelay(1);
482 	}
483 
484 	if (i >= adev->usec_timeout)
485 		r = -ETIMEDOUT;
486 
487 	return r;
488 }
489 
490 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
491 				   struct amdgpu_bo *bo,
492 				   struct dma_fence **fence)
493 {
494 	struct amdgpu_device *adev = ring->adev;
495 	struct dma_fence *f = NULL;
496 	struct amdgpu_job *job;
497 	struct amdgpu_ib *ib;
498 	uint64_t addr;
499 	void *msg = NULL;
500 	int i, r;
501 
502 	r = amdgpu_job_alloc_with_ib(adev, 64,
503 					AMDGPU_IB_POOL_DIRECT, &job);
504 	if (r)
505 		goto err;
506 
507 	ib = &job->ibs[0];
508 	addr = amdgpu_bo_gpu_offset(bo);
509 	msg = amdgpu_bo_kptr(bo);
510 	ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
511 	ib->ptr[1] = addr;
512 	ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
513 	ib->ptr[3] = addr >> 32;
514 	ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
515 	ib->ptr[5] = 0;
516 	for (i = 6; i < 16; i += 2) {
517 		ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
518 		ib->ptr[i+1] = 0;
519 	}
520 	ib->length_dw = 16;
521 
522 	r = amdgpu_job_submit_direct(job, ring, &f);
523 	if (r)
524 		goto err_free;
525 
526 	amdgpu_bo_fence(bo, f, false);
527 	amdgpu_bo_unreserve(bo);
528 	amdgpu_bo_free_kernel(&bo, NULL, (void **)&msg);
529 
530 	if (fence)
531 		*fence = dma_fence_get(f);
532 	dma_fence_put(f);
533 
534 	return 0;
535 
536 err_free:
537 	amdgpu_job_free(job);
538 
539 err:
540 	amdgpu_bo_unreserve(bo);
541 	amdgpu_bo_free_kernel(&bo, NULL, (void **)&msg);
542 	return r;
543 }
544 
545 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
546 					 struct amdgpu_bo **bo)
547 {
548 	struct amdgpu_device *adev = ring->adev;
549 	uint32_t *msg;
550 	int r, i;
551 
552 	*bo = NULL;
553 	r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
554 				      AMDGPU_GEM_DOMAIN_VRAM,
555 				      bo, NULL, (void **)&msg);
556 	if (r)
557 		return r;
558 
559 	msg[0] = cpu_to_le32(0x00000028);
560 	msg[1] = cpu_to_le32(0x00000038);
561 	msg[2] = cpu_to_le32(0x00000001);
562 	msg[3] = cpu_to_le32(0x00000000);
563 	msg[4] = cpu_to_le32(handle);
564 	msg[5] = cpu_to_le32(0x00000000);
565 	msg[6] = cpu_to_le32(0x00000001);
566 	msg[7] = cpu_to_le32(0x00000028);
567 	msg[8] = cpu_to_le32(0x00000010);
568 	msg[9] = cpu_to_le32(0x00000000);
569 	msg[10] = cpu_to_le32(0x00000007);
570 	msg[11] = cpu_to_le32(0x00000000);
571 	msg[12] = cpu_to_le32(0x00000780);
572 	msg[13] = cpu_to_le32(0x00000440);
573 	for (i = 14; i < 1024; ++i)
574 		msg[i] = cpu_to_le32(0x0);
575 
576 	return 0;
577 }
578 
579 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
580 					  struct amdgpu_bo **bo)
581 {
582 	struct amdgpu_device *adev = ring->adev;
583 	uint32_t *msg;
584 	int r, i;
585 
586 	*bo = NULL;
587 	r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
588 				      AMDGPU_GEM_DOMAIN_VRAM,
589 				      bo, NULL, (void **)&msg);
590 	if (r)
591 		return r;
592 
593 	msg[0] = cpu_to_le32(0x00000028);
594 	msg[1] = cpu_to_le32(0x00000018);
595 	msg[2] = cpu_to_le32(0x00000000);
596 	msg[3] = cpu_to_le32(0x00000002);
597 	msg[4] = cpu_to_le32(handle);
598 	msg[5] = cpu_to_le32(0x00000000);
599 	for (i = 6; i < 1024; ++i)
600 		msg[i] = cpu_to_le32(0x0);
601 
602 	return 0;
603 }
604 
605 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
606 {
607 	struct dma_fence *fence = NULL;
608 	struct amdgpu_bo *bo;
609 	long r;
610 
611 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &bo);
612 	if (r)
613 		goto error;
614 
615 	r = amdgpu_vcn_dec_send_msg(ring, bo, NULL);
616 	if (r)
617 		goto error;
618 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &bo);
619 	if (r)
620 		goto error;
621 
622 	r = amdgpu_vcn_dec_send_msg(ring, bo, &fence);
623 	if (r)
624 		goto error;
625 
626 	r = dma_fence_wait_timeout(fence, false, timeout);
627 	if (r == 0)
628 		r = -ETIMEDOUT;
629 	else if (r > 0)
630 		r = 0;
631 
632 	dma_fence_put(fence);
633 error:
634 	return r;
635 }
636 
637 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
638 				   struct amdgpu_bo *bo,
639 				   struct dma_fence **fence)
640 {
641 	struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
642 	const unsigned int ib_size_dw = 64;
643 	struct amdgpu_device *adev = ring->adev;
644 	struct dma_fence *f = NULL;
645 	struct amdgpu_job *job;
646 	struct amdgpu_ib *ib;
647 	uint64_t addr;
648 	int i, r;
649 
650 	r = amdgpu_job_alloc_with_ib(adev, ib_size_dw * 4,
651 				AMDGPU_IB_POOL_DIRECT, &job);
652 	if (r)
653 		goto err;
654 
655 	ib = &job->ibs[0];
656 	addr = amdgpu_bo_gpu_offset(bo);
657 	ib->length_dw = 0;
658 
659 	ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
660 	ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
661 	decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
662 	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
663 	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
664 
665 	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
666 	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
667 	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
668 
669 	for (i = ib->length_dw; i < ib_size_dw; ++i)
670 		ib->ptr[i] = 0x0;
671 
672 	r = amdgpu_job_submit_direct(job, ring, &f);
673 	if (r)
674 		goto err_free;
675 
676 	amdgpu_bo_fence(bo, f, false);
677 	amdgpu_bo_unreserve(bo);
678 	amdgpu_bo_unref(&bo);
679 
680 	if (fence)
681 		*fence = dma_fence_get(f);
682 	dma_fence_put(f);
683 
684 	return 0;
685 
686 err_free:
687 	amdgpu_job_free(job);
688 
689 err:
690 	amdgpu_bo_unreserve(bo);
691 	amdgpu_bo_unref(&bo);
692 	return r;
693 }
694 
695 int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
696 {
697 	struct dma_fence *fence = NULL;
698 	struct amdgpu_bo *bo;
699 	long r;
700 
701 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &bo);
702 	if (r)
703 		goto error;
704 
705 	r = amdgpu_vcn_dec_sw_send_msg(ring, bo, NULL);
706 	if (r)
707 		goto error;
708 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &bo);
709 	if (r)
710 		goto error;
711 
712 	r = amdgpu_vcn_dec_sw_send_msg(ring, bo, &fence);
713 	if (r)
714 		goto error;
715 
716 	r = dma_fence_wait_timeout(fence, false, timeout);
717 	if (r == 0)
718 		r = -ETIMEDOUT;
719 	else if (r > 0)
720 		r = 0;
721 
722 	dma_fence_put(fence);
723 error:
724 	return r;
725 }
726 
727 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
728 {
729 	struct amdgpu_device *adev = ring->adev;
730 	uint32_t rptr;
731 	unsigned i;
732 	int r;
733 
734 	if (amdgpu_sriov_vf(adev))
735 		return 0;
736 
737 	r = amdgpu_ring_alloc(ring, 16);
738 	if (r)
739 		return r;
740 
741 	rptr = amdgpu_ring_get_rptr(ring);
742 
743 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
744 	amdgpu_ring_commit(ring);
745 
746 	for (i = 0; i < adev->usec_timeout; i++) {
747 		if (amdgpu_ring_get_rptr(ring) != rptr)
748 			break;
749 		udelay(1);
750 	}
751 
752 	if (i >= adev->usec_timeout)
753 		r = -ETIMEDOUT;
754 
755 	return r;
756 }
757 
758 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
759 					 struct amdgpu_bo *bo,
760 					 struct dma_fence **fence)
761 {
762 	const unsigned ib_size_dw = 16;
763 	struct amdgpu_job *job;
764 	struct amdgpu_ib *ib;
765 	struct dma_fence *f = NULL;
766 	uint64_t addr;
767 	int i, r;
768 
769 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
770 					AMDGPU_IB_POOL_DIRECT, &job);
771 	if (r)
772 		return r;
773 
774 	ib = &job->ibs[0];
775 	addr = amdgpu_bo_gpu_offset(bo);
776 
777 	ib->length_dw = 0;
778 	ib->ptr[ib->length_dw++] = 0x00000018;
779 	ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
780 	ib->ptr[ib->length_dw++] = handle;
781 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
782 	ib->ptr[ib->length_dw++] = addr;
783 	ib->ptr[ib->length_dw++] = 0x0000000b;
784 
785 	ib->ptr[ib->length_dw++] = 0x00000014;
786 	ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
787 	ib->ptr[ib->length_dw++] = 0x0000001c;
788 	ib->ptr[ib->length_dw++] = 0x00000000;
789 	ib->ptr[ib->length_dw++] = 0x00000000;
790 
791 	ib->ptr[ib->length_dw++] = 0x00000008;
792 	ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
793 
794 	for (i = ib->length_dw; i < ib_size_dw; ++i)
795 		ib->ptr[i] = 0x0;
796 
797 	r = amdgpu_job_submit_direct(job, ring, &f);
798 	if (r)
799 		goto err;
800 
801 	if (fence)
802 		*fence = dma_fence_get(f);
803 	dma_fence_put(f);
804 
805 	return 0;
806 
807 err:
808 	amdgpu_job_free(job);
809 	return r;
810 }
811 
812 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
813 					  struct amdgpu_bo *bo,
814 					  struct dma_fence **fence)
815 {
816 	const unsigned ib_size_dw = 16;
817 	struct amdgpu_job *job;
818 	struct amdgpu_ib *ib;
819 	struct dma_fence *f = NULL;
820 	uint64_t addr;
821 	int i, r;
822 
823 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
824 					AMDGPU_IB_POOL_DIRECT, &job);
825 	if (r)
826 		return r;
827 
828 	ib = &job->ibs[0];
829 	addr = amdgpu_bo_gpu_offset(bo);
830 
831 	ib->length_dw = 0;
832 	ib->ptr[ib->length_dw++] = 0x00000018;
833 	ib->ptr[ib->length_dw++] = 0x00000001;
834 	ib->ptr[ib->length_dw++] = handle;
835 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
836 	ib->ptr[ib->length_dw++] = addr;
837 	ib->ptr[ib->length_dw++] = 0x0000000b;
838 
839 	ib->ptr[ib->length_dw++] = 0x00000014;
840 	ib->ptr[ib->length_dw++] = 0x00000002;
841 	ib->ptr[ib->length_dw++] = 0x0000001c;
842 	ib->ptr[ib->length_dw++] = 0x00000000;
843 	ib->ptr[ib->length_dw++] = 0x00000000;
844 
845 	ib->ptr[ib->length_dw++] = 0x00000008;
846 	ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
847 
848 	for (i = ib->length_dw; i < ib_size_dw; ++i)
849 		ib->ptr[i] = 0x0;
850 
851 	r = amdgpu_job_submit_direct(job, ring, &f);
852 	if (r)
853 		goto err;
854 
855 	if (fence)
856 		*fence = dma_fence_get(f);
857 	dma_fence_put(f);
858 
859 	return 0;
860 
861 err:
862 	amdgpu_job_free(job);
863 	return r;
864 }
865 
866 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
867 {
868 	struct dma_fence *fence = NULL;
869 	struct amdgpu_bo *bo = NULL;
870 	long r;
871 
872 	r = amdgpu_bo_create_reserved(ring->adev, 128 * 1024, PAGE_SIZE,
873 				      AMDGPU_GEM_DOMAIN_VRAM,
874 				      &bo, NULL, NULL);
875 	if (r)
876 		return r;
877 
878 	r = amdgpu_vcn_enc_get_create_msg(ring, 1, bo, NULL);
879 	if (r)
880 		goto error;
881 
882 	r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, bo, &fence);
883 	if (r)
884 		goto error;
885 
886 	r = dma_fence_wait_timeout(fence, false, timeout);
887 	if (r == 0)
888 		r = -ETIMEDOUT;
889 	else if (r > 0)
890 		r = 0;
891 
892 error:
893 	dma_fence_put(fence);
894 	amdgpu_bo_unreserve(bo);
895 	amdgpu_bo_free_kernel(&bo, NULL, NULL);
896 
897 	return r;
898 }
899