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->ip_versions[UVD_HWIP][0]) {
90 	case IP_VERSION(1, 0, 0):
91 	case IP_VERSION(1, 0, 1):
92 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
93 			fw_name = FIRMWARE_RAVEN2;
94 		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
95 			fw_name = FIRMWARE_PICASSO;
96 		else
97 			fw_name = FIRMWARE_RAVEN;
98 		break;
99 	case IP_VERSION(2, 5, 0):
100 		fw_name = FIRMWARE_ARCTURUS;
101 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
102 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
103 			adev->vcn.indirect_sram = true;
104 		break;
105 	case IP_VERSION(2, 2, 0):
106 		if (adev->apu_flags & AMD_APU_IS_RENOIR)
107 			fw_name = FIRMWARE_RENOIR;
108 		else
109 			fw_name = FIRMWARE_GREEN_SARDINE;
110 
111 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
112 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
113 			adev->vcn.indirect_sram = true;
114 		break;
115 	case IP_VERSION(2, 6, 0):
116 		fw_name = FIRMWARE_ALDEBARAN;
117 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
118 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
119 			adev->vcn.indirect_sram = true;
120 		break;
121 	case IP_VERSION(2, 0, 0):
122 		fw_name = FIRMWARE_NAVI10;
123 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
124 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
125 			adev->vcn.indirect_sram = true;
126 		break;
127 	case IP_VERSION(2, 0, 2):
128 		if (adev->asic_type == CHIP_NAVI12)
129 			fw_name = FIRMWARE_NAVI12;
130 		else
131 			fw_name = FIRMWARE_NAVI14;
132 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
133 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
134 			adev->vcn.indirect_sram = true;
135 		break;
136 	case IP_VERSION(3, 0, 0):
137 	case IP_VERSION(3, 0, 64):
138 	case IP_VERSION(3, 0, 192):
139 		if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(10, 3, 0))
140 			fw_name = FIRMWARE_SIENNA_CICHLID;
141 		else
142 			fw_name = FIRMWARE_NAVY_FLOUNDER;
143 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
144 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
145 			adev->vcn.indirect_sram = true;
146 		break;
147 	case IP_VERSION(3, 0, 2):
148 		fw_name = FIRMWARE_VANGOGH;
149 		break;
150 	case IP_VERSION(3, 0, 16):
151 		fw_name = FIRMWARE_DIMGREY_CAVEFISH;
152 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
153 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
154 			adev->vcn.indirect_sram = true;
155 		break;
156 	case IP_VERSION(3, 0, 33):
157 		fw_name = FIRMWARE_BEIGE_GOBY;
158 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
159 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
160 			adev->vcn.indirect_sram = true;
161 		break;
162 	case IP_VERSION(3, 1, 1):
163 		fw_name = FIRMWARE_YELLOW_CARP;
164 		if ((adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) &&
165 		    (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG))
166 			adev->vcn.indirect_sram = true;
167 		break;
168 	default:
169 		return -EINVAL;
170 	}
171 
172 	r = request_firmware(&adev->vcn.fw, fw_name, adev->dev);
173 	if (r) {
174 		dev_err(adev->dev, "amdgpu_vcn: Can't load firmware \"%s\"\n",
175 			fw_name);
176 		return r;
177 	}
178 
179 	r = amdgpu_ucode_validate(adev->vcn.fw);
180 	if (r) {
181 		dev_err(adev->dev, "amdgpu_vcn: Can't validate firmware \"%s\"\n",
182 			fw_name);
183 		release_firmware(adev->vcn.fw);
184 		adev->vcn.fw = NULL;
185 		return r;
186 	}
187 
188 	hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
189 	adev->vcn.fw_version = le32_to_cpu(hdr->ucode_version);
190 
191 	/* Bit 20-23, it is encode major and non-zero for new naming convention.
192 	 * This field is part of version minor and DRM_DISABLED_FLAG in old naming
193 	 * convention. Since the l:wq!atest version minor is 0x5B and DRM_DISABLED_FLAG
194 	 * is zero in old naming convention, this field is always zero so far.
195 	 * These four bits are used to tell which naming convention is present.
196 	 */
197 	fw_check = (le32_to_cpu(hdr->ucode_version) >> 20) & 0xf;
198 	if (fw_check) {
199 		unsigned int dec_ver, enc_major, enc_minor, vep, fw_rev;
200 
201 		fw_rev = le32_to_cpu(hdr->ucode_version) & 0xfff;
202 		enc_minor = (le32_to_cpu(hdr->ucode_version) >> 12) & 0xff;
203 		enc_major = fw_check;
204 		dec_ver = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xf;
205 		vep = (le32_to_cpu(hdr->ucode_version) >> 28) & 0xf;
206 		DRM_INFO("Found VCN firmware Version ENC: %u.%u DEC: %u VEP: %u Revision: %u\n",
207 			enc_major, enc_minor, dec_ver, vep, fw_rev);
208 	} else {
209 		unsigned int version_major, version_minor, family_id;
210 
211 		family_id = le32_to_cpu(hdr->ucode_version) & 0xff;
212 		version_major = (le32_to_cpu(hdr->ucode_version) >> 24) & 0xff;
213 		version_minor = (le32_to_cpu(hdr->ucode_version) >> 8) & 0xff;
214 		DRM_INFO("Found VCN firmware Version: %u.%u Family ID: %u\n",
215 			version_major, version_minor, family_id);
216 	}
217 
218 	bo_size = AMDGPU_VCN_STACK_SIZE + AMDGPU_VCN_CONTEXT_SIZE;
219 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
220 		bo_size += AMDGPU_GPU_PAGE_ALIGN(le32_to_cpu(hdr->ucode_size_bytes) + 8);
221 	bo_size += AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
222 
223 	for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
224 		if (adev->vcn.harvest_config & (1 << i))
225 			continue;
226 
227 		r = amdgpu_bo_create_kernel(adev, bo_size, PAGE_SIZE,
228 						AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].vcpu_bo,
229 						&adev->vcn.inst[i].gpu_addr, &adev->vcn.inst[i].cpu_addr);
230 		if (r) {
231 			dev_err(adev->dev, "(%d) failed to allocate vcn bo\n", r);
232 			return r;
233 		}
234 
235 		adev->vcn.inst[i].fw_shared_cpu_addr = adev->vcn.inst[i].cpu_addr +
236 				bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
237 		adev->vcn.inst[i].fw_shared_gpu_addr = adev->vcn.inst[i].gpu_addr +
238 				bo_size - AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared));
239 
240 		if (adev->vcn.indirect_sram) {
241 			r = amdgpu_bo_create_kernel(adev, 64 * 2 * 4, PAGE_SIZE,
242 					AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].dpg_sram_bo,
243 					&adev->vcn.inst[i].dpg_sram_gpu_addr, &adev->vcn.inst[i].dpg_sram_cpu_addr);
244 			if (r) {
245 				dev_err(adev->dev, "VCN %d (%d) failed to allocate DPG bo\n", i, r);
246 				return r;
247 			}
248 		}
249 	}
250 
251 	return 0;
252 }
253 
254 int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)
255 {
256 	int i, j;
257 
258 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
259 		if (adev->vcn.harvest_config & (1 << j))
260 			continue;
261 
262 		if (adev->vcn.indirect_sram) {
263 			amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo,
264 						  &adev->vcn.inst[j].dpg_sram_gpu_addr,
265 						  (void **)&adev->vcn.inst[j].dpg_sram_cpu_addr);
266 		}
267 		kvfree(adev->vcn.inst[j].saved_bo);
268 
269 		amdgpu_bo_free_kernel(&adev->vcn.inst[j].vcpu_bo,
270 					  &adev->vcn.inst[j].gpu_addr,
271 					  (void **)&adev->vcn.inst[j].cpu_addr);
272 
273 		amdgpu_ring_fini(&adev->vcn.inst[j].ring_dec);
274 
275 		for (i = 0; i < adev->vcn.num_enc_rings; ++i)
276 			amdgpu_ring_fini(&adev->vcn.inst[j].ring_enc[i]);
277 	}
278 
279 	release_firmware(adev->vcn.fw);
280 	mutex_destroy(&adev->vcn.vcn1_jpeg1_workaround);
281 	mutex_destroy(&adev->vcn.vcn_pg_lock);
282 
283 	return 0;
284 }
285 
286 bool amdgpu_vcn_is_disabled_vcn(struct amdgpu_device *adev, enum vcn_ring_type type, uint32_t vcn_instance)
287 {
288 	bool ret = false;
289 
290 	int major;
291 	int minor;
292 	int revision;
293 
294 	/* if cannot find IP data, then this VCN does not exist */
295 	if (amdgpu_discovery_get_vcn_version(adev, vcn_instance, &major, &minor, &revision) != 0)
296 		return true;
297 
298 	if ((type == VCN_ENCODE_RING) && (revision & VCN_BLOCK_ENCODE_DISABLE_MASK)) {
299 		ret = true;
300 	} else if ((type == VCN_DECODE_RING) && (revision & VCN_BLOCK_DECODE_DISABLE_MASK)) {
301 		ret = true;
302 	} else if ((type == VCN_UNIFIED_RING) && (revision & VCN_BLOCK_QUEUE_DISABLE_MASK)) {
303 		ret = true;
304 	}
305 
306 	return ret;
307 }
308 
309 int amdgpu_vcn_suspend(struct amdgpu_device *adev)
310 {
311 	unsigned size;
312 	void *ptr;
313 	int i, idx;
314 
315 	cancel_delayed_work_sync(&adev->vcn.idle_work);
316 
317 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
318 		if (adev->vcn.harvest_config & (1 << i))
319 			continue;
320 		if (adev->vcn.inst[i].vcpu_bo == NULL)
321 			return 0;
322 
323 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
324 		ptr = adev->vcn.inst[i].cpu_addr;
325 
326 		adev->vcn.inst[i].saved_bo = kvmalloc(size, GFP_KERNEL);
327 		if (!adev->vcn.inst[i].saved_bo)
328 			return -ENOMEM;
329 
330 		if (drm_dev_enter(adev_to_drm(adev), &idx)) {
331 			memcpy_fromio(adev->vcn.inst[i].saved_bo, ptr, size);
332 			drm_dev_exit(idx);
333 		}
334 	}
335 	return 0;
336 }
337 
338 int amdgpu_vcn_resume(struct amdgpu_device *adev)
339 {
340 	unsigned size;
341 	void *ptr;
342 	int i, idx;
343 
344 	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
345 		if (adev->vcn.harvest_config & (1 << i))
346 			continue;
347 		if (adev->vcn.inst[i].vcpu_bo == NULL)
348 			return -EINVAL;
349 
350 		size = amdgpu_bo_size(adev->vcn.inst[i].vcpu_bo);
351 		ptr = adev->vcn.inst[i].cpu_addr;
352 
353 		if (adev->vcn.inst[i].saved_bo != NULL) {
354 			if (drm_dev_enter(adev_to_drm(adev), &idx)) {
355 				memcpy_toio(ptr, adev->vcn.inst[i].saved_bo, size);
356 				drm_dev_exit(idx);
357 			}
358 			kvfree(adev->vcn.inst[i].saved_bo);
359 			adev->vcn.inst[i].saved_bo = NULL;
360 		} else {
361 			const struct common_firmware_header *hdr;
362 			unsigned offset;
363 
364 			hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
365 			if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
366 				offset = le32_to_cpu(hdr->ucode_array_offset_bytes);
367 				if (drm_dev_enter(adev_to_drm(adev), &idx)) {
368 					memcpy_toio(adev->vcn.inst[i].cpu_addr, adev->vcn.fw->data + offset,
369 						    le32_to_cpu(hdr->ucode_size_bytes));
370 					drm_dev_exit(idx);
371 				}
372 				size -= le32_to_cpu(hdr->ucode_size_bytes);
373 				ptr += le32_to_cpu(hdr->ucode_size_bytes);
374 			}
375 			memset_io(ptr, 0, size);
376 		}
377 	}
378 	return 0;
379 }
380 
381 static void amdgpu_vcn_idle_work_handler(struct work_struct *work)
382 {
383 	struct amdgpu_device *adev =
384 		container_of(work, struct amdgpu_device, vcn.idle_work.work);
385 	unsigned int fences = 0, fence[AMDGPU_MAX_VCN_INSTANCES] = {0};
386 	unsigned int i, j;
387 	int r = 0;
388 
389 	for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
390 		if (adev->vcn.harvest_config & (1 << j))
391 			continue;
392 
393 		for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
394 			fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_enc[i]);
395 		}
396 
397 		if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
398 			struct dpg_pause_state new_state;
399 
400 			if (fence[j] ||
401 				unlikely(atomic_read(&adev->vcn.inst[j].dpg_enc_submission_cnt)))
402 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
403 			else
404 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
405 
406 			adev->vcn.pause_dpg_mode(adev, j, &new_state);
407 		}
408 
409 		fence[j] += amdgpu_fence_count_emitted(&adev->vcn.inst[j].ring_dec);
410 		fences += fence[j];
411 	}
412 
413 	if (!fences && !atomic_read(&adev->vcn.total_submission_cnt)) {
414 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
415 		       AMD_PG_STATE_GATE);
416 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
417 				false);
418 		if (r)
419 			dev_warn(adev->dev, "(%d) failed to disable video power profile mode\n", r);
420 	} else {
421 		schedule_delayed_work(&adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
422 	}
423 }
424 
425 void amdgpu_vcn_ring_begin_use(struct amdgpu_ring *ring)
426 {
427 	struct amdgpu_device *adev = ring->adev;
428 	int r = 0;
429 
430 	atomic_inc(&adev->vcn.total_submission_cnt);
431 
432 	if (!cancel_delayed_work_sync(&adev->vcn.idle_work)) {
433 		r = amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_VIDEO,
434 				true);
435 		if (r)
436 			dev_warn(adev->dev, "(%d) failed to switch to video power profile mode\n", r);
437 	}
438 
439 	mutex_lock(&adev->vcn.vcn_pg_lock);
440 	amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_VCN,
441 	       AMD_PG_STATE_UNGATE);
442 
443 	if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)	{
444 		struct dpg_pause_state new_state;
445 
446 		if (ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC) {
447 			atomic_inc(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
448 			new_state.fw_based = VCN_DPG_STATE__PAUSE;
449 		} else {
450 			unsigned int fences = 0;
451 			unsigned int i;
452 
453 			for (i = 0; i < adev->vcn.num_enc_rings; ++i)
454 				fences += amdgpu_fence_count_emitted(&adev->vcn.inst[ring->me].ring_enc[i]);
455 
456 			if (fences || atomic_read(&adev->vcn.inst[ring->me].dpg_enc_submission_cnt))
457 				new_state.fw_based = VCN_DPG_STATE__PAUSE;
458 			else
459 				new_state.fw_based = VCN_DPG_STATE__UNPAUSE;
460 		}
461 
462 		adev->vcn.pause_dpg_mode(adev, ring->me, &new_state);
463 	}
464 	mutex_unlock(&adev->vcn.vcn_pg_lock);
465 }
466 
467 void amdgpu_vcn_ring_end_use(struct amdgpu_ring *ring)
468 {
469 	if (ring->adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG &&
470 		ring->funcs->type == AMDGPU_RING_TYPE_VCN_ENC)
471 		atomic_dec(&ring->adev->vcn.inst[ring->me].dpg_enc_submission_cnt);
472 
473 	atomic_dec(&ring->adev->vcn.total_submission_cnt);
474 
475 	schedule_delayed_work(&ring->adev->vcn.idle_work, VCN_IDLE_TIMEOUT);
476 }
477 
478 int amdgpu_vcn_dec_ring_test_ring(struct amdgpu_ring *ring)
479 {
480 	struct amdgpu_device *adev = ring->adev;
481 	uint32_t tmp = 0;
482 	unsigned i;
483 	int r;
484 
485 	/* VCN in SRIOV does not support direct register read/write */
486 	if (amdgpu_sriov_vf(adev))
487 		return 0;
488 
489 	WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
490 	r = amdgpu_ring_alloc(ring, 3);
491 	if (r)
492 		return r;
493 	amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
494 	amdgpu_ring_write(ring, 0xDEADBEEF);
495 	amdgpu_ring_commit(ring);
496 	for (i = 0; i < adev->usec_timeout; i++) {
497 		tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
498 		if (tmp == 0xDEADBEEF)
499 			break;
500 		udelay(1);
501 	}
502 
503 	if (i >= adev->usec_timeout)
504 		r = -ETIMEDOUT;
505 
506 	return r;
507 }
508 
509 int amdgpu_vcn_dec_sw_ring_test_ring(struct amdgpu_ring *ring)
510 {
511 	struct amdgpu_device *adev = ring->adev;
512 	uint32_t rptr;
513 	unsigned int i;
514 	int r;
515 
516 	if (amdgpu_sriov_vf(adev))
517 		return 0;
518 
519 	r = amdgpu_ring_alloc(ring, 16);
520 	if (r)
521 		return r;
522 
523 	rptr = amdgpu_ring_get_rptr(ring);
524 
525 	amdgpu_ring_write(ring, VCN_DEC_SW_CMD_END);
526 	amdgpu_ring_commit(ring);
527 
528 	for (i = 0; i < adev->usec_timeout; i++) {
529 		if (amdgpu_ring_get_rptr(ring) != rptr)
530 			break;
531 		udelay(1);
532 	}
533 
534 	if (i >= adev->usec_timeout)
535 		r = -ETIMEDOUT;
536 
537 	return r;
538 }
539 
540 static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
541 				   struct amdgpu_ib *ib_msg,
542 				   struct dma_fence **fence)
543 {
544 	struct amdgpu_device *adev = ring->adev;
545 	struct dma_fence *f = NULL;
546 	struct amdgpu_job *job;
547 	struct amdgpu_ib *ib;
548 	uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
549 	int i, r;
550 
551 	r = amdgpu_job_alloc_with_ib(adev, 64,
552 					AMDGPU_IB_POOL_DIRECT, &job);
553 	if (r)
554 		goto err;
555 
556 	ib = &job->ibs[0];
557 	ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
558 	ib->ptr[1] = addr;
559 	ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
560 	ib->ptr[3] = addr >> 32;
561 	ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
562 	ib->ptr[5] = 0;
563 	for (i = 6; i < 16; i += 2) {
564 		ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
565 		ib->ptr[i+1] = 0;
566 	}
567 	ib->length_dw = 16;
568 
569 	r = amdgpu_job_submit_direct(job, ring, &f);
570 	if (r)
571 		goto err_free;
572 
573 	amdgpu_ib_free(adev, ib_msg, f);
574 
575 	if (fence)
576 		*fence = dma_fence_get(f);
577 	dma_fence_put(f);
578 
579 	return 0;
580 
581 err_free:
582 	amdgpu_job_free(job);
583 err:
584 	amdgpu_ib_free(adev, ib_msg, f);
585 	return r;
586 }
587 
588 static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
589 		struct amdgpu_ib *ib)
590 {
591 	struct amdgpu_device *adev = ring->adev;
592 	uint32_t *msg;
593 	int r, i;
594 
595 	memset(ib, 0, sizeof(*ib));
596 	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
597 			AMDGPU_IB_POOL_DIRECT,
598 			ib);
599 	if (r)
600 		return r;
601 
602 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
603 	msg[0] = cpu_to_le32(0x00000028);
604 	msg[1] = cpu_to_le32(0x00000038);
605 	msg[2] = cpu_to_le32(0x00000001);
606 	msg[3] = cpu_to_le32(0x00000000);
607 	msg[4] = cpu_to_le32(handle);
608 	msg[5] = cpu_to_le32(0x00000000);
609 	msg[6] = cpu_to_le32(0x00000001);
610 	msg[7] = cpu_to_le32(0x00000028);
611 	msg[8] = cpu_to_le32(0x00000010);
612 	msg[9] = cpu_to_le32(0x00000000);
613 	msg[10] = cpu_to_le32(0x00000007);
614 	msg[11] = cpu_to_le32(0x00000000);
615 	msg[12] = cpu_to_le32(0x00000780);
616 	msg[13] = cpu_to_le32(0x00000440);
617 	for (i = 14; i < 1024; ++i)
618 		msg[i] = cpu_to_le32(0x0);
619 
620 	return 0;
621 }
622 
623 static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
624 					  struct amdgpu_ib *ib)
625 {
626 	struct amdgpu_device *adev = ring->adev;
627 	uint32_t *msg;
628 	int r, i;
629 
630 	memset(ib, 0, sizeof(*ib));
631 	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
632 			AMDGPU_IB_POOL_DIRECT,
633 			ib);
634 	if (r)
635 		return r;
636 
637 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
638 	msg[0] = cpu_to_le32(0x00000028);
639 	msg[1] = cpu_to_le32(0x00000018);
640 	msg[2] = cpu_to_le32(0x00000000);
641 	msg[3] = cpu_to_le32(0x00000002);
642 	msg[4] = cpu_to_le32(handle);
643 	msg[5] = cpu_to_le32(0x00000000);
644 	for (i = 6; i < 1024; ++i)
645 		msg[i] = cpu_to_le32(0x0);
646 
647 	return 0;
648 }
649 
650 int amdgpu_vcn_dec_ring_test_ib(struct amdgpu_ring *ring, long timeout)
651 {
652 	struct dma_fence *fence = NULL;
653 	struct amdgpu_ib ib;
654 	long r;
655 
656 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
657 	if (r)
658 		goto error;
659 
660 	r = amdgpu_vcn_dec_send_msg(ring, &ib, NULL);
661 	if (r)
662 		goto error;
663 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
664 	if (r)
665 		goto error;
666 
667 	r = amdgpu_vcn_dec_send_msg(ring, &ib, &fence);
668 	if (r)
669 		goto error;
670 
671 	r = dma_fence_wait_timeout(fence, false, timeout);
672 	if (r == 0)
673 		r = -ETIMEDOUT;
674 	else if (r > 0)
675 		r = 0;
676 
677 	dma_fence_put(fence);
678 error:
679 	return r;
680 }
681 
682 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
683 				      struct amdgpu_ib *ib_msg,
684 				      struct dma_fence **fence)
685 {
686 	struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
687 	const unsigned int ib_size_dw = 64;
688 	struct amdgpu_device *adev = ring->adev;
689 	struct dma_fence *f = NULL;
690 	struct amdgpu_job *job;
691 	struct amdgpu_ib *ib;
692 	uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
693 	int i, r;
694 
695 	r = amdgpu_job_alloc_with_ib(adev, ib_size_dw * 4,
696 				AMDGPU_IB_POOL_DIRECT, &job);
697 	if (r)
698 		goto err;
699 
700 	ib = &job->ibs[0];
701 	ib->length_dw = 0;
702 
703 	ib->ptr[ib->length_dw++] = sizeof(struct amdgpu_vcn_decode_buffer) + 8;
704 	ib->ptr[ib->length_dw++] = cpu_to_le32(AMDGPU_VCN_IB_FLAG_DECODE_BUFFER);
705 	decode_buffer = (struct amdgpu_vcn_decode_buffer *)&(ib->ptr[ib->length_dw]);
706 	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
707 	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
708 
709 	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
710 	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
711 	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
712 
713 	for (i = ib->length_dw; i < ib_size_dw; ++i)
714 		ib->ptr[i] = 0x0;
715 
716 	r = amdgpu_job_submit_direct(job, ring, &f);
717 	if (r)
718 		goto err_free;
719 
720 	amdgpu_ib_free(adev, ib_msg, f);
721 
722 	if (fence)
723 		*fence = dma_fence_get(f);
724 	dma_fence_put(f);
725 
726 	return 0;
727 
728 err_free:
729 	amdgpu_job_free(job);
730 err:
731 	amdgpu_ib_free(adev, ib_msg, f);
732 	return r;
733 }
734 
735 int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
736 {
737 	struct dma_fence *fence = NULL;
738 	struct amdgpu_ib ib;
739 	long r;
740 
741 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
742 	if (r)
743 		goto error;
744 
745 	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
746 	if (r)
747 		goto error;
748 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
749 	if (r)
750 		goto error;
751 
752 	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
753 	if (r)
754 		goto error;
755 
756 	r = dma_fence_wait_timeout(fence, false, timeout);
757 	if (r == 0)
758 		r = -ETIMEDOUT;
759 	else if (r > 0)
760 		r = 0;
761 
762 	dma_fence_put(fence);
763 error:
764 	return r;
765 }
766 
767 int amdgpu_vcn_enc_ring_test_ring(struct amdgpu_ring *ring)
768 {
769 	struct amdgpu_device *adev = ring->adev;
770 	uint32_t rptr;
771 	unsigned i;
772 	int r;
773 
774 	if (amdgpu_sriov_vf(adev))
775 		return 0;
776 
777 	r = amdgpu_ring_alloc(ring, 16);
778 	if (r)
779 		return r;
780 
781 	rptr = amdgpu_ring_get_rptr(ring);
782 
783 	amdgpu_ring_write(ring, VCN_ENC_CMD_END);
784 	amdgpu_ring_commit(ring);
785 
786 	for (i = 0; i < adev->usec_timeout; i++) {
787 		if (amdgpu_ring_get_rptr(ring) != rptr)
788 			break;
789 		udelay(1);
790 	}
791 
792 	if (i >= adev->usec_timeout)
793 		r = -ETIMEDOUT;
794 
795 	return r;
796 }
797 
798 static int amdgpu_vcn_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
799 					 struct amdgpu_ib *ib_msg,
800 					 struct dma_fence **fence)
801 {
802 	const unsigned ib_size_dw = 16;
803 	struct amdgpu_job *job;
804 	struct amdgpu_ib *ib;
805 	struct dma_fence *f = NULL;
806 	uint64_t addr;
807 	int i, r;
808 
809 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
810 					AMDGPU_IB_POOL_DIRECT, &job);
811 	if (r)
812 		return r;
813 
814 	ib = &job->ibs[0];
815 	addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
816 
817 	ib->length_dw = 0;
818 	ib->ptr[ib->length_dw++] = 0x00000018;
819 	ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
820 	ib->ptr[ib->length_dw++] = handle;
821 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
822 	ib->ptr[ib->length_dw++] = addr;
823 	ib->ptr[ib->length_dw++] = 0x0000000b;
824 
825 	ib->ptr[ib->length_dw++] = 0x00000014;
826 	ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
827 	ib->ptr[ib->length_dw++] = 0x0000001c;
828 	ib->ptr[ib->length_dw++] = 0x00000000;
829 	ib->ptr[ib->length_dw++] = 0x00000000;
830 
831 	ib->ptr[ib->length_dw++] = 0x00000008;
832 	ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
833 
834 	for (i = ib->length_dw; i < ib_size_dw; ++i)
835 		ib->ptr[i] = 0x0;
836 
837 	r = amdgpu_job_submit_direct(job, ring, &f);
838 	if (r)
839 		goto err;
840 
841 	if (fence)
842 		*fence = dma_fence_get(f);
843 	dma_fence_put(f);
844 
845 	return 0;
846 
847 err:
848 	amdgpu_job_free(job);
849 	return r;
850 }
851 
852 static int amdgpu_vcn_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
853 					  struct amdgpu_ib *ib_msg,
854 					  struct dma_fence **fence)
855 {
856 	const unsigned ib_size_dw = 16;
857 	struct amdgpu_job *job;
858 	struct amdgpu_ib *ib;
859 	struct dma_fence *f = NULL;
860 	uint64_t addr;
861 	int i, r;
862 
863 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4,
864 					AMDGPU_IB_POOL_DIRECT, &job);
865 	if (r)
866 		return r;
867 
868 	ib = &job->ibs[0];
869 	addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
870 
871 	ib->length_dw = 0;
872 	ib->ptr[ib->length_dw++] = 0x00000018;
873 	ib->ptr[ib->length_dw++] = 0x00000001;
874 	ib->ptr[ib->length_dw++] = handle;
875 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
876 	ib->ptr[ib->length_dw++] = addr;
877 	ib->ptr[ib->length_dw++] = 0x0000000b;
878 
879 	ib->ptr[ib->length_dw++] = 0x00000014;
880 	ib->ptr[ib->length_dw++] = 0x00000002;
881 	ib->ptr[ib->length_dw++] = 0x0000001c;
882 	ib->ptr[ib->length_dw++] = 0x00000000;
883 	ib->ptr[ib->length_dw++] = 0x00000000;
884 
885 	ib->ptr[ib->length_dw++] = 0x00000008;
886 	ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
887 
888 	for (i = ib->length_dw; i < ib_size_dw; ++i)
889 		ib->ptr[i] = 0x0;
890 
891 	r = amdgpu_job_submit_direct(job, ring, &f);
892 	if (r)
893 		goto err;
894 
895 	if (fence)
896 		*fence = dma_fence_get(f);
897 	dma_fence_put(f);
898 
899 	return 0;
900 
901 err:
902 	amdgpu_job_free(job);
903 	return r;
904 }
905 
906 int amdgpu_vcn_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
907 {
908 	struct amdgpu_device *adev = ring->adev;
909 	struct dma_fence *fence = NULL;
910 	struct amdgpu_ib ib;
911 	long r;
912 
913 	memset(&ib, 0, sizeof(ib));
914 	r = amdgpu_ib_get(adev, NULL, (128 << 10) + AMDGPU_GPU_PAGE_SIZE,
915 			AMDGPU_IB_POOL_DIRECT,
916 			&ib);
917 	if (r)
918 		return r;
919 
920 	r = amdgpu_vcn_enc_get_create_msg(ring, 1, &ib, NULL);
921 	if (r)
922 		goto error;
923 
924 	r = amdgpu_vcn_enc_get_destroy_msg(ring, 1, &ib, &fence);
925 	if (r)
926 		goto error;
927 
928 	r = dma_fence_wait_timeout(fence, false, timeout);
929 	if (r == 0)
930 		r = -ETIMEDOUT;
931 	else if (r > 0)
932 		r = 0;
933 
934 error:
935 	amdgpu_ib_free(adev, &ib, fence);
936 	dma_fence_put(fence);
937 
938 	return r;
939 }
940 
941 enum amdgpu_ring_priority_level amdgpu_vcn_get_enc_ring_prio(int ring)
942 {
943 	switch(ring) {
944 	case 0:
945 		return AMDGPU_RING_PRIO_0;
946 	case 1:
947 		return AMDGPU_RING_PRIO_1;
948 	case 2:
949 		return AMDGPU_RING_PRIO_2;
950 	default:
951 		return AMDGPU_RING_PRIO_0;
952 	}
953 }
954 
955 void amdgpu_vcn_setup_ucode(struct amdgpu_device *adev)
956 {
957 	int i;
958 	unsigned int idx;
959 
960 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
961 		const struct common_firmware_header *hdr;
962 		hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
963 
964 		for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
965 			if (adev->vcn.harvest_config & (1 << i))
966 				continue;
967 			/* currently only support 2 FW instances */
968 			if (i >= 2) {
969 				dev_info(adev->dev, "More then 2 VCN FW instances!\n");
970 				break;
971 			}
972 			idx = AMDGPU_UCODE_ID_VCN + i;
973 			adev->firmware.ucode[idx].ucode_id = idx;
974 			adev->firmware.ucode[idx].fw = adev->vcn.fw;
975 			adev->firmware.fw_size +=
976 				ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
977 		}
978 		dev_info(adev->dev, "Will use PSP to load VCN firmware\n");
979 	}
980 }
981