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 #include <drm/drm_drv.h>
31 
32 #include "amdgpu.h"
33 #include "amdgpu_pm.h"
34 #include "amdgpu_vcn.h"
35 #include "soc15d.h"
36 
37 /* Firmware Names */
38 #define FIRMWARE_RAVEN		"amdgpu/raven_vcn.bin"
39 #define FIRMWARE_PICASSO	"amdgpu/picasso_vcn.bin"
40 #define FIRMWARE_RAVEN2		"amdgpu/raven2_vcn.bin"
41 #define FIRMWARE_ARCTURUS	"amdgpu/arcturus_vcn.bin"
42 #define FIRMWARE_RENOIR		"amdgpu/renoir_vcn.bin"
43 #define FIRMWARE_GREEN_SARDINE	"amdgpu/green_sardine_vcn.bin"
44 #define FIRMWARE_NAVI10		"amdgpu/navi10_vcn.bin"
45 #define FIRMWARE_NAVI14		"amdgpu/navi14_vcn.bin"
46 #define FIRMWARE_NAVI12		"amdgpu/navi12_vcn.bin"
47 #define FIRMWARE_SIENNA_CICHLID	"amdgpu/sienna_cichlid_vcn.bin"
48 #define FIRMWARE_NAVY_FLOUNDER	"amdgpu/navy_flounder_vcn.bin"
49 #define FIRMWARE_VANGOGH	"amdgpu/vangogh_vcn.bin"
50 #define FIRMWARE_DIMGREY_CAVEFISH	"amdgpu/dimgrey_cavefish_vcn.bin"
51 #define FIRMWARE_ALDEBARAN	"amdgpu/aldebaran_vcn.bin"
52 #define FIRMWARE_BEIGE_GOBY	"amdgpu/beige_goby_vcn.bin"
53 #define FIRMWARE_YELLOW_CARP	"amdgpu/yellow_carp_vcn.bin"
54 
55 MODULE_FIRMWARE(FIRMWARE_RAVEN);
56 MODULE_FIRMWARE(FIRMWARE_PICASSO);
57 MODULE_FIRMWARE(FIRMWARE_RAVEN2);
58 MODULE_FIRMWARE(FIRMWARE_ARCTURUS);
59 MODULE_FIRMWARE(FIRMWARE_RENOIR);
60 MODULE_FIRMWARE(FIRMWARE_GREEN_SARDINE);
61 MODULE_FIRMWARE(FIRMWARE_ALDEBARAN);
62 MODULE_FIRMWARE(FIRMWARE_NAVI10);
63 MODULE_FIRMWARE(FIRMWARE_NAVI14);
64 MODULE_FIRMWARE(FIRMWARE_NAVI12);
65 MODULE_FIRMWARE(FIRMWARE_SIENNA_CICHLID);
66 MODULE_FIRMWARE(FIRMWARE_NAVY_FLOUNDER);
67 MODULE_FIRMWARE(FIRMWARE_VANGOGH);
68 MODULE_FIRMWARE(FIRMWARE_DIMGREY_CAVEFISH);
69 MODULE_FIRMWARE(FIRMWARE_BEIGE_GOBY);
70 MODULE_FIRMWARE(FIRMWARE_YELLOW_CARP);
71 
72 static void amdgpu_vcn_idle_work_handler(struct work_struct *work);
73 
74 int amdgpu_vcn_sw_init(struct amdgpu_device *adev)
75 {
76 	unsigned long bo_size;
77 	const char *fw_name;
78 	const struct common_firmware_header *hdr;
79 	unsigned char fw_check;
80 	int i, r;
81 
82 	INIT_DELAYED_WORK(&adev->vcn.idle_work, amdgpu_vcn_idle_work_handler);
83 	mutex_init(&adev->vcn.vcn_pg_lock);
84 	mutex_init(&adev->vcn.vcn1_jpeg1_workaround);
85 	atomic_set(&adev->vcn.total_submission_cnt, 0);
86 	for (i = 0; i < adev->vcn.num_vcn_inst; i++)
87 		atomic_set(&adev->vcn.inst[i].dpg_enc_submission_cnt, 0);
88 
89 	switch (adev->asic_type) {
90 	case CHIP_RAVEN:
91 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
92 			fw_name = FIRMWARE_RAVEN2;
93 		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
94 			fw_name = FIRMWARE_PICASSO;
95 		else
96 			fw_name = FIRMWARE_RAVEN;
97 		break;
98 	case CHIP_ARCTURUS:
99 		fw_name = FIRMWARE_ARCTURUS;
100 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
101 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
102 			adev->vcn.indirect_sram = true;
103 		break;
104 	case CHIP_RENOIR:
105 		if (adev->apu_flags & AMD_APU_IS_RENOIR)
106 			fw_name = FIRMWARE_RENOIR;
107 		else
108 			fw_name = FIRMWARE_GREEN_SARDINE;
109 
110 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
111 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
112 			adev->vcn.indirect_sram = true;
113 		break;
114 	case CHIP_ALDEBARAN:
115 		fw_name = FIRMWARE_ALDEBARAN;
116 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
117 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
118 			adev->vcn.indirect_sram = true;
119 		break;
120 	case CHIP_NAVI10:
121 		fw_name = FIRMWARE_NAVI10;
122 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
123 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
124 			adev->vcn.indirect_sram = true;
125 		break;
126 	case CHIP_NAVI14:
127 		fw_name = FIRMWARE_NAVI14;
128 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
129 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
130 			adev->vcn.indirect_sram = true;
131 		break;
132 	case CHIP_NAVI12:
133 		fw_name = FIRMWARE_NAVI12;
134 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
135 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
136 			adev->vcn.indirect_sram = true;
137 		break;
138 	case CHIP_SIENNA_CICHLID:
139 		fw_name = FIRMWARE_SIENNA_CICHLID;
140 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
141 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
142 			adev->vcn.indirect_sram = true;
143 		break;
144 	case CHIP_NAVY_FLOUNDER:
145 		fw_name = FIRMWARE_NAVY_FLOUNDER;
146 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
147 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
148 			adev->vcn.indirect_sram = true;
149 		break;
150 	case CHIP_VANGOGH:
151 		fw_name = FIRMWARE_VANGOGH;
152 		break;
153 	case CHIP_DIMGREY_CAVEFISH:
154 		fw_name = FIRMWARE_DIMGREY_CAVEFISH;
155 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
156 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
157 			adev->vcn.indirect_sram = true;
158 		break;
159 	case CHIP_BEIGE_GOBY:
160 		fw_name = FIRMWARE_BEIGE_GOBY;
161 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
162 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
163 			adev->vcn.indirect_sram = true;
164 		break;
165 	case CHIP_YELLOW_CARP:
166 		fw_name = FIRMWARE_YELLOW_CARP;
167 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
168 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
169 			adev->vcn.indirect_sram = true;
170 		break;
171 	default:
172 		return -EINVAL;
173 	}
174 
175 	r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
176 	if (r) {
177 		dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
178 			fw_name);
179 		return r;
180 	}
181 
182 	r = amdgpu_ucode_validate(adev->vcn.fw);
183 	if (r) {
184 		dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
185 			fw_name);
186 		release_firmware(adev->vcn.fw);
187 		adev->vcn.fw = NULL;
188 		return r;
189 	}
190 
191 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
192 	adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
193 
194 	/* Bit 20-23, it is encode major and non-zero for new naming convention.
195 	 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
196 	 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
197 	 * is zero in old naming convention, this field is always zero so far.
198 	 * These four bits are used to tell which naming convention is present.
199 	 */
200 	fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
201 	if (fw_check) {
202 		unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
203 
204 		fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
205 		enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
206 		enc_major = fw_check;
207 		dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
208 		vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
209 		DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
210 			enc_major, enc_minor, dec_ver, vep, fw_rev);
211 	} else {
212 		unsigned int version_major, version_minor, family_id;
213 
214 		family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
215 		version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
216 		version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
217 		DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
218 			version_major, version_minor, family_id);
219 	}
220 
221 	bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
222 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
223 		bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
224 	bo_size += AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
225 
226 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
227 		if (adev->vcn.harvest_config & (1 << i))
228 			continue;
229 
230 		r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
231 						AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].vcpu_bo,
232 						&adev->vcn.inst[i].gpu_addr, &adev->vcn.inst[i].cpu_addr);
233 		if (r) {
234 			dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
235 			return r;
236 		}
237 
238 		adev->vcn.inst[i].fw_shared_cpu_addr = adev->vcn.inst[i].cpu_addr +
239 				bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
240 		adev->vcn.inst[i].fw_shared_gpu_addr = adev->vcn.inst[i].gpu_addr +
241 				bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
242 
243 		if (adev->vcn.indirect_sram) {
244 			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
245 					AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
246 					&adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
247 			if (r) {
248 				dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
249 				return r;
250 			}
251 		}
252 	}
253 
254 	return 0;
255 }
256 
257 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
258 {
259 	int i, j;
260 
261 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
262 		if (adev->vcn.harvest_config & (1 << j))
263 			continue;
264 
265 		if (adev->vcn.indirect_sram) {
266 			amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo,
267 						  &adev->vcn.inst[j].dpg_sram_gpu_addr,
268 						  (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
269 		}
270 		kvfree(adev->vcn.inst[j].saved_bo);
271 
272 		amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
273 					  &adev->vcn.inst[j].gpu_addr,
274 					  (void **)&adev->vcn.inst[j].cpu_addr);
275 
276 		amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
277 
278 		for (i = 0; i < adev->vcn.num_enc_rings; ++i)
279 			amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
280 	}
281 
282 	release_firmware(adev->vcn.fw);
283 	mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
284 	mutex_destroy(&adev->vcn.vcn_pg_lock);
285 
286 	return 0;
287 }
288 
289 bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
290 {
291 	bool ret = false;
292 
293 	int major;
294 	int minor;
295 	int revision;
296 
297 	/* if cannot find IP data, then this VCN does not exist */
298 	if (amdgpu_discovery_get_vcn_version(adev, vcn_instance, &major, &minor, &revision) != 0)
299 		return true;
300 
301 	if ((type == VCN_ENCODE_RING) && (revision & VCN_BLOCK_ENCODE_DISABLE_MASK)) {
302 		ret = true;
303 	} else if ((type == VCN_DECODE_RING) && (revision & VCN_BLOCK_DECODE_DISABLE_MASK)) {
304 		ret = true;
305 	} else if ((type == VCN_UNIFIED_RING) && (revision & VCN_BLOCK_QUEUE_DISABLE_MASK)) {
306 		ret = true;
307 	}
308 
309 	return ret;
310 }
311 
312 int amdgpu_vcn_suspend(struct amdgpu_device *adev)
313 {
314 	unsigned size;
315 	void *ptr;
316 	int i, idx;
317 
318 	cancel_delayed_work_sync(&adev->vcn.idle_work);
319 
320 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
321 		if (adev->vcn.harvest_config & (1 << i))
322 			continue;
323 		if (adev->vcn.inst[i].vcpu_bo == NULL)
324 			return 0;
325 
326 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
327 		ptr = adev->vcn.inst[i].cpu_addr;
328 
329 		adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
330 		if (!adev->vcn.inst[i].saved_bo)
331 			return -ENOMEM;
332 
333 		if (drm_dev_enter(&adev->ddev, &idx)) {
334 			memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
335 			drm_dev_exit(idx);
336 		}
337 	}
338 	return 0;
339 }
340 
341 int amdgpu_vcn_resume(struct amdgpu_device *adev)
342 {
343 	unsigned size;
344 	void *ptr;
345 	int i, idx;
346 
347 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
348 		if (adev->vcn.harvest_config & (1 << i))
349 			continue;
350 		if (adev->vcn.inst[i].vcpu_bo == NULL)
351 			return -EINVAL;
352 
353 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
354 		ptr = adev->vcn.inst[i].cpu_addr;
355 
356 		if (adev->vcn.inst[i].saved_bo != NULL) {
357 			if (drm_dev_enter(&adev->ddev, &idx)) {
358 				memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
359 				drm_dev_exit(idx);
360 			}
361 			kvfree(adev->vcn.inst[i].saved_bo);
362 			adev->vcn.inst[i].saved_bo = NULL;
363 		} else {
364 			const struct common_firmware_header *hdr;
365 			unsigned offset;
366 
367 			hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
368 			if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
369 				offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
370 				if (drm_dev_enter(&adev->ddev, &idx)) {
371 					memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
372 						    le32_to_cpu(hdr->ucode_size_bytes));
373 					drm_dev_exit(idx);
374 				}
375 				size -= le32_to_cpu(hdr->ucode_size_bytes);
376 				ptr += le32_to_cpu(hdr->ucode_size_bytes);
377 			}
378 			memset_io(ptr, 0, size);
379 		}
380 	}
381 	return 0;
382 }
383 
384 static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
385 {
386 	struct amdgpu_device *adev =
387 		container_of(work, struct amdgpu_device, vcn.idle_work.work);
388 	unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
389 	unsigned int i, j;
390 	int r = 0;
391 
392 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
393 		if (adev->vcn.harvest_config & (1 << j))
394 			continue;
395 
396 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
397 			fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
398 		}
399 
400 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
401 			struct dpg_pause_state new_state;
402 
403 			if (fence[j] ||
404 				unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
405 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
406 			else
407 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
408 
409 			adev->vcn.pause_dpg_mode(adev, j, &new_state);
410 		}
411 
412 		fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
413 		fences += fence[j];
414 	}
415 
416 	if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
417 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
418 		       AMD_PG_STATE_GATE);
419 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
420 				false);
421 		if (r)
422 			dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
423 	} else {
424 		schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
425 	}
426 }
427 
428 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
429 {
430 	struct amdgpu_device *adev = ring->adev;
431 	int r = 0;
432 
433 	atomic_inc(&adev->vcn.total_submission_cnt);
434 
435 	if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
436 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
437 				true);
438 		if (r)
439 			dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
440 	}
441 
442 	mutex_lock(&adev->vcn.vcn_pg_lock);
443 	amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
444 	       AMD_PG_STATE_UNGATE);
445 
446 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
447 		struct dpg_pause_state new_state;
448 
449 		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
450 			atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
451 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
452 		} else {
453 			unsigned int fences = 0;
454 			unsigned int i;
455 
456 			for (i = 0; i < adev->vcn.num_enc_rings; ++i)
457 				fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
458 
459 			if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
460 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
461 			else
462 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
463 		}
464 
465 		adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
466 	}
467 	mutex_unlock(&adev->vcn.vcn_pg_lock);
468 }
469 
470 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
471 {
472 	if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
473 		ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
474 		atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
475 
476 	atomic_dec(&ring->adev->vcn.total_submission_cnt);
477 
478 	schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
479 }
480 
481 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
482 {
483 	struct amdgpu_device *adev = ring->adev;
484 	uint32_t tmp = 0;
485 	unsigned i;
486 	int r;
487 
488 	/* VCN in SRIOV does not support direct register read/write */
489 	if (amdgpu_sriov_vf(adev))
490 		return 0;
491 
492 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
493 	r = amdgpu_ring_alloc(ring, 3);
494 	if (r)
495 		return r;
496 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
497 	amdgpu_ring_write(ring, 0xDEADBEEF);
498 	amdgpu_ring_commit(ring);
499 	for (i = 0; i < adev->usec_timeout; i++) {
500 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
501 		if (tmp == 0xDEADBEEF)
502 			break;
503 		udelay(1);
504 	}
505 
506 	if (i >= adev->usec_timeout)
507 		r = -ETIMEDOUT;
508 
509 	return r;
510 }
511 
512 int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
513 {
514 	struct amdgpu_device *adev = ring->adev;
515 	uint32_t rptr;
516 	unsigned int i;
517 	int r;
518 
519 	if (amdgpu_sriov_vf(adev))
520 		return 0;
521 
522 	r = amdgpu_ring_alloc(ring, 16);
523 	if (r)
524 		return r;
525 
526 	rptr = amdgpu_ring_get_rptr(ring);
527 
528 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
529 	amdgpu_ring_commit(ring);
530 
531 	for (i = 0; i < adev->usec_timeout; i++) {
532 		if (amdgpu_ring_get_rptr(ring) != rptr)
533 			break;
534 		udelay(1);
535 	}
536 
537 	if (i >= adev->usec_timeout)
538 		r = -ETIMEDOUT;
539 
540 	return r;
541 }
542 
543 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
544 				   struct amdgpu_bo *bo,
545 				   struct dma_fence **fence)
546 {
547 	struct amdgpu_device *adev = ring->adev;
548 	struct dma_fence *f = NULL;
549 	struct amdgpu_job *job;
550 	struct amdgpu_ib *ib;
551 	uint64_t addr;
552 	void *msg = NULL;
553 	int i, r;
554 
555 	r = amdgpu_job_alloc_with_ib(adev, 64,
556 					AMDGPU_IB_POOL_DIRECT, &job);
557 	if (r)
558 		goto err;
559 
560 	ib = &job->ibs[0];
561 	addr = amdgpu_bo_gpu_offset(bo);
562 	msg = amdgpu_bo_kptr(bo);
563 	ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
564 	ib->ptr[1] = addr;
565 	ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
566 	ib->ptr[3] = addr >> 32;
567 	ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
568 	ib->ptr[5] = 0;
569 	for (i = 6; i < 16; i += 2) {
570 		ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
571 		ib->ptr[i+1] = 0;
572 	}
573 	ib->length_dw = 16;
574 
575 	r = amdgpu_job_submit_direct(job, ring, &f);
576 	if (r)
577 		goto err_free;
578 
579 	amdgpu_bo_fence(bo, f, false);
580 	amdgpu_bo_unreserve(bo);
581 	amdgpu_bo_free_kernel(&bo, NULL, (void **)&msg);
582 
583 	if (fence)
584 		*fence = dma_fence_get(f);
585 	dma_fence_put(f);
586 
587 	return 0;
588 
589 err_free:
590 	amdgpu_job_free(job);
591 
592 err:
593 	amdgpu_bo_unreserve(bo);
594 	amdgpu_bo_free_kernel(&bo, NULL, (void **)&msg);
595 	return r;
596 }
597 
598 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
599 					 struct amdgpu_bo **bo)
600 {
601 	struct amdgpu_device *adev = ring->adev;
602 	uint32_t *msg;
603 	int r, i;
604 
605 	*bo = NULL;
606 	r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
607 				      AMDGPU_GEM_DOMAIN_VRAM,
608 				      bo, NULL, (void **)&msg);
609 	if (r)
610 		return r;
611 
612 	msg[0] = cpu_to_le32(0x00000028);
613 	msg[1] = cpu_to_le32(0x00000038);
614 	msg[2] = cpu_to_le32(0x00000001);
615 	msg[3] = cpu_to_le32(0x00000000);
616 	msg[4] = cpu_to_le32(handle);
617 	msg[5] = cpu_to_le32(0x00000000);
618 	msg[6] = cpu_to_le32(0x00000001);
619 	msg[7] = cpu_to_le32(0x00000028);
620 	msg[8] = cpu_to_le32(0x00000010);
621 	msg[9] = cpu_to_le32(0x00000000);
622 	msg[10] = cpu_to_le32(0x00000007);
623 	msg[11] = cpu_to_le32(0x00000000);
624 	msg[12] = cpu_to_le32(0x00000780);
625 	msg[13] = cpu_to_le32(0x00000440);
626 	for (i = 14; i < 1024; ++i)
627 		msg[i] = cpu_to_le32(0x0);
628 
629 	return 0;
630 }
631 
632 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
633 					  struct amdgpu_bo **bo)
634 {
635 	struct amdgpu_device *adev = ring->adev;
636 	uint32_t *msg;
637 	int r, i;
638 
639 	*bo = NULL;
640 	r = amdgpu_bo_create_reserved(adev, 1024, PAGE_SIZE,
641 				      AMDGPU_GEM_DOMAIN_VRAM,
642 				      bo, NULL, (void **)&msg);
643 	if (r)
644 		return r;
645 
646 	msg[0] = cpu_to_le32(0x00000028);
647 	msg[1] = cpu_to_le32(0x00000018);
648 	msg[2] = cpu_to_le32(0x00000000);
649 	msg[3] = cpu_to_le32(0x00000002);
650 	msg[4] = cpu_to_le32(handle);
651 	msg[5] = cpu_to_le32(0x00000000);
652 	for (i = 6; i < 1024; ++i)
653 		msg[i] = cpu_to_le32(0x0);
654 
655 	return 0;
656 }
657 
658 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
659 {
660 	struct dma_fence *fence = NULL;
661 	struct amdgpu_bo *bo;
662 	long r;
663 
664 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &bo);
665 	if (r)
666 		goto error;
667 
668 	r = amdgpu_vcn_dec_send_msg(ring, bo, NULL);
669 	if (r)
670 		goto error;
671 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &bo);
672 	if (r)
673 		goto error;
674 
675 	r = amdgpu_vcn_dec_send_msg(ring, bo, &fence);
676 	if (r)
677 		goto error;
678 
679 	r = dma_fence_wait_timeout(fence, false, timeout);
680 	if (r == 0)
681 		r = -ETIMEDOUT;
682 	else if (r > 0)
683 		r = 0;
684 
685 	dma_fence_put(fence);
686 error:
687 	return r;
688 }
689 
690 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
691 				   struct amdgpu_bo *bo,
692 				   struct dma_fence **fence)
693 {
694 	struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
695 	const unsigned int ib_size_dw = 64;
696 	struct amdgpu_device *adev = ring->adev;
697 	struct dma_fence *f = NULL;
698 	struct amdgpu_job *job;
699 	struct amdgpu_ib *ib;
700 	uint64_t addr;
701 	int i, r;
702 
703 	r = amdgpu_job_alloc_with_ib(adev, ib_size_dw * 4,
704 				AMDGPU_IB_POOL_DIRECT, &job);
705 	if (r)
706 		goto err;
707 
708 	ib = &job->ibs[0];
709 	addr = amdgpu_bo_gpu_offset(bo);
710 	ib->length_dw = 0;
711 
712 	ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
713 	ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
714 	decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
715 	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
716 	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
717 
718 	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
719 	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
720 	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
721 
722 	for (i = ib->length_dw; i < ib_size_dw; ++i)
723 		ib->ptr[i] = 0x0;
724 
725 	r = amdgpu_job_submit_direct(job, ring, &f);
726 	if (r)
727 		goto err_free;
728 
729 	amdgpu_bo_fence(bo, f, false);
730 	amdgpu_bo_unreserve(bo);
731 	amdgpu_bo_unref(&bo);
732 
733 	if (fence)
734 		*fence = dma_fence_get(f);
735 	dma_fence_put(f);
736 
737 	return 0;
738 
739 err_free:
740 	amdgpu_job_free(job);
741 
742 err:
743 	amdgpu_bo_unreserve(bo);
744 	amdgpu_bo_unref(&bo);
745 	return r;
746 }
747 
748 int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
749 {
750 	struct dma_fence *fence = NULL;
751 	struct amdgpu_bo *bo;
752 	long r;
753 
754 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &bo);
755 	if (r)
756 		goto error;
757 
758 	r = amdgpu_vcn_dec_sw_send_msg(ring, bo, NULL);
759 	if (r)
760 		goto error;
761 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &bo);
762 	if (r)
763 		goto error;
764 
765 	r = amdgpu_vcn_dec_sw_send_msg(ring, bo, &fence);
766 	if (r)
767 		goto error;
768 
769 	r = dma_fence_wait_timeout(fence, false, timeout);
770 	if (r == 0)
771 		r = -ETIMEDOUT;
772 	else if (r > 0)
773 		r = 0;
774 
775 	dma_fence_put(fence);
776 error:
777 	return r;
778 }
779 
780 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
781 {
782 	struct amdgpu_device *adev = ring->adev;
783 	uint32_t rptr;
784 	unsigned i;
785 	int r;
786 
787 	if (amdgpu_sriov_vf(adev))
788 		return 0;
789 
790 	r = amdgpu_ring_alloc(ring, 16);
791 	if (r)
792 		return r;
793 
794 	rptr = amdgpu_ring_get_rptr(ring);
795 
796 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
797 	amdgpu_ring_commit(ring);
798 
799 	for (i = 0; i < adev->usec_timeout; i++) {
800 		if (amdgpu_ring_get_rptr(ring) != rptr)
801 			break;
802 		udelay(1);
803 	}
804 
805 	if (i >= adev->usec_timeout)
806 		r = -ETIMEDOUT;
807 
808 	return r;
809 }
810 
811 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
812 					 struct amdgpu_bo *bo,
813 					 struct dma_fence **fence)
814 {
815 	const unsigned ib_size_dw = 16;
816 	struct amdgpu_job *job;
817 	struct amdgpu_ib *ib;
818 	struct dma_fence *f = NULL;
819 	uint64_t addr;
820 	int i, r;
821 
822 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
823 					AMDGPU_IB_POOL_DIRECT, &job);
824 	if (r)
825 		return r;
826 
827 	ib = &job->ibs[0];
828 	addr = amdgpu_bo_gpu_offset(bo);
829 
830 	ib->length_dw = 0;
831 	ib->ptr[ib->length_dw++] = 0x00000018;
832 	ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
833 	ib->ptr[ib->length_dw++] = handle;
834 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
835 	ib->ptr[ib->length_dw++] = addr;
836 	ib->ptr[ib->length_dw++] = 0x0000000b;
837 
838 	ib->ptr[ib->length_dw++] = 0x00000014;
839 	ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
840 	ib->ptr[ib->length_dw++] = 0x0000001c;
841 	ib->ptr[ib->length_dw++] = 0x00000000;
842 	ib->ptr[ib->length_dw++] = 0x00000000;
843 
844 	ib->ptr[ib->length_dw++] = 0x00000008;
845 	ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
846 
847 	for (i = ib->length_dw; i < ib_size_dw; ++i)
848 		ib->ptr[i] = 0x0;
849 
850 	r = amdgpu_job_submit_direct(job, ring, &f);
851 	if (r)
852 		goto err;
853 
854 	if (fence)
855 		*fence = dma_fence_get(f);
856 	dma_fence_put(f);
857 
858 	return 0;
859 
860 err:
861 	amdgpu_job_free(job);
862 	return r;
863 }
864 
865 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
866 					  struct amdgpu_bo *bo,
867 					  struct dma_fence **fence)
868 {
869 	const unsigned ib_size_dw = 16;
870 	struct amdgpu_job *job;
871 	struct amdgpu_ib *ib;
872 	struct dma_fence *f = NULL;
873 	uint64_t addr;
874 	int i, r;
875 
876 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
877 					AMDGPU_IB_POOL_DIRECT, &job);
878 	if (r)
879 		return r;
880 
881 	ib = &job->ibs[0];
882 	addr = amdgpu_bo_gpu_offset(bo);
883 
884 	ib->length_dw = 0;
885 	ib->ptr[ib->length_dw++] = 0x00000018;
886 	ib->ptr[ib->length_dw++] = 0x00000001;
887 	ib->ptr[ib->length_dw++] = handle;
888 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
889 	ib->ptr[ib->length_dw++] = addr;
890 	ib->ptr[ib->length_dw++] = 0x0000000b;
891 
892 	ib->ptr[ib->length_dw++] = 0x00000014;
893 	ib->ptr[ib->length_dw++] = 0x00000002;
894 	ib->ptr[ib->length_dw++] = 0x0000001c;
895 	ib->ptr[ib->length_dw++] = 0x00000000;
896 	ib->ptr[ib->length_dw++] = 0x00000000;
897 
898 	ib->ptr[ib->length_dw++] = 0x00000008;
899 	ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
900 
901 	for (i = ib->length_dw; i < ib_size_dw; ++i)
902 		ib->ptr[i] = 0x0;
903 
904 	r = amdgpu_job_submit_direct(job, ring, &f);
905 	if (r)
906 		goto err;
907 
908 	if (fence)
909 		*fence = dma_fence_get(f);
910 	dma_fence_put(f);
911 
912 	return 0;
913 
914 err:
915 	amdgpu_job_free(job);
916 	return r;
917 }
918 
919 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
920 {
921 	struct dma_fence *fence = NULL;
922 	struct amdgpu_bo *bo = NULL;
923 	long r;
924 
925 	r = amdgpu_bo_create_reserved(ring->adev, 128 * 1024, PAGE_SIZE,
926 				      AMDGPU_GEM_DOMAIN_VRAM,
927 				      &bo, NULL, NULL);
928 	if (r)
929 		return r;
930 
931 	r = amdgpu_vcn_enc_get_create_msg(ring, 1, bo, NULL);
932 	if (r)
933 		goto error;
934 
935 	r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, bo, &fence);
936 	if (r)
937 		goto error;
938 
939 	r = dma_fence_wait_timeout(fence, false, timeout);
940 	if (r == 0)
941 		r = -ETIMEDOUT;
942 	else if (r > 0)
943 		r = 0;
944 
945 error:
946 	dma_fence_put(fence);
947 	amdgpu_bo_unreserve(bo);
948 	amdgpu_bo_free_kernel(&bo, NULL, NULL);
949 
950 	return r;
951 }
952