xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c (revision 5a244f48)
1 /*
2  * Copyright 2016 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include <drm/drmP.h>
26 #include "amdgpu.h"
27 #include "amdgpu_uvd.h"
28 #include "soc15d.h"
29 #include "soc15_common.h"
30 #include "mmsch_v1_0.h"
31 
32 #include "vega10/soc15ip.h"
33 #include "vega10/UVD/uvd_7_0_offset.h"
34 #include "vega10/UVD/uvd_7_0_sh_mask.h"
35 #include "vega10/VCE/vce_4_0_offset.h"
36 #include "vega10/VCE/vce_4_0_default.h"
37 #include "vega10/VCE/vce_4_0_sh_mask.h"
38 #include "vega10/NBIF/nbif_6_1_offset.h"
39 #include "vega10/HDP/hdp_4_0_offset.h"
40 #include "vega10/MMHUB/mmhub_1_0_offset.h"
41 #include "vega10/MMHUB/mmhub_1_0_sh_mask.h"
42 
43 static void uvd_v7_0_set_ring_funcs(struct amdgpu_device *adev);
44 static void uvd_v7_0_set_enc_ring_funcs(struct amdgpu_device *adev);
45 static void uvd_v7_0_set_irq_funcs(struct amdgpu_device *adev);
46 static int uvd_v7_0_start(struct amdgpu_device *adev);
47 static void uvd_v7_0_stop(struct amdgpu_device *adev);
48 static int uvd_v7_0_sriov_start(struct amdgpu_device *adev);
49 
50 /**
51  * uvd_v7_0_ring_get_rptr - get read pointer
52  *
53  * @ring: amdgpu_ring pointer
54  *
55  * Returns the current hardware read pointer
56  */
57 static uint64_t uvd_v7_0_ring_get_rptr(struct amdgpu_ring *ring)
58 {
59 	struct amdgpu_device *adev = ring->adev;
60 
61 	return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
62 }
63 
64 /**
65  * uvd_v7_0_enc_ring_get_rptr - get enc read pointer
66  *
67  * @ring: amdgpu_ring pointer
68  *
69  * Returns the current hardware enc read pointer
70  */
71 static uint64_t uvd_v7_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
72 {
73 	struct amdgpu_device *adev = ring->adev;
74 
75 	if (ring == &adev->uvd.ring_enc[0])
76 		return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR);
77 	else
78 		return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2);
79 }
80 
81 /**
82  * uvd_v7_0_ring_get_wptr - get write pointer
83  *
84  * @ring: amdgpu_ring pointer
85  *
86  * Returns the current hardware write pointer
87  */
88 static uint64_t uvd_v7_0_ring_get_wptr(struct amdgpu_ring *ring)
89 {
90 	struct amdgpu_device *adev = ring->adev;
91 
92 	return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR);
93 }
94 
95 /**
96  * uvd_v7_0_enc_ring_get_wptr - get enc write pointer
97  *
98  * @ring: amdgpu_ring pointer
99  *
100  * Returns the current hardware enc write pointer
101  */
102 static uint64_t uvd_v7_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
103 {
104 	struct amdgpu_device *adev = ring->adev;
105 
106 	if (ring->use_doorbell)
107 		return adev->wb.wb[ring->wptr_offs];
108 
109 	if (ring == &adev->uvd.ring_enc[0])
110 		return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
111 	else
112 		return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
113 }
114 
115 /**
116  * uvd_v7_0_ring_set_wptr - set write pointer
117  *
118  * @ring: amdgpu_ring pointer
119  *
120  * Commits the write pointer to the hardware
121  */
122 static void uvd_v7_0_ring_set_wptr(struct amdgpu_ring *ring)
123 {
124 	struct amdgpu_device *adev = ring->adev;
125 
126 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
127 }
128 
129 /**
130  * uvd_v7_0_enc_ring_set_wptr - set enc write pointer
131  *
132  * @ring: amdgpu_ring pointer
133  *
134  * Commits the enc write pointer to the hardware
135  */
136 static void uvd_v7_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
137 {
138 	struct amdgpu_device *adev = ring->adev;
139 
140 	if (ring->use_doorbell) {
141 		/* XXX check if swapping is necessary on BE */
142 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
143 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
144 		return;
145 	}
146 
147 	if (ring == &adev->uvd.ring_enc[0])
148 		WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR,
149 			lower_32_bits(ring->wptr));
150 	else
151 		WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2,
152 			lower_32_bits(ring->wptr));
153 }
154 
155 /**
156  * uvd_v7_0_enc_ring_test_ring - test if UVD ENC ring is working
157  *
158  * @ring: the engine to test on
159  *
160  */
161 static int uvd_v7_0_enc_ring_test_ring(struct amdgpu_ring *ring)
162 {
163 	struct amdgpu_device *adev = ring->adev;
164 	uint32_t rptr = amdgpu_ring_get_rptr(ring);
165 	unsigned i;
166 	int r;
167 
168 	if (amdgpu_sriov_vf(adev))
169 		return 0;
170 
171 	r = amdgpu_ring_alloc(ring, 16);
172 	if (r) {
173 		DRM_ERROR("amdgpu: uvd enc failed to lock ring %d (%d).\n",
174 			  ring->idx, r);
175 		return r;
176 	}
177 	amdgpu_ring_write(ring, HEVC_ENC_CMD_END);
178 	amdgpu_ring_commit(ring);
179 
180 	for (i = 0; i < adev->usec_timeout; i++) {
181 		if (amdgpu_ring_get_rptr(ring) != rptr)
182 			break;
183 		DRM_UDELAY(1);
184 	}
185 
186 	if (i < adev->usec_timeout) {
187 		DRM_INFO("ring test on %d succeeded in %d usecs\n",
188 			 ring->idx, i);
189 	} else {
190 		DRM_ERROR("amdgpu: ring %d test failed\n",
191 			  ring->idx);
192 		r = -ETIMEDOUT;
193 	}
194 
195 	return r;
196 }
197 
198 /**
199  * uvd_v7_0_enc_get_create_msg - generate a UVD ENC create msg
200  *
201  * @adev: amdgpu_device pointer
202  * @ring: ring we should submit the msg to
203  * @handle: session handle to use
204  * @fence: optional fence to return
205  *
206  * Open up a stream for HW test
207  */
208 static int uvd_v7_0_enc_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
209 				       struct dma_fence **fence)
210 {
211 	const unsigned ib_size_dw = 16;
212 	struct amdgpu_job *job;
213 	struct amdgpu_ib *ib;
214 	struct dma_fence *f = NULL;
215 	uint64_t dummy;
216 	int i, r;
217 
218 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
219 	if (r)
220 		return r;
221 
222 	ib = &job->ibs[0];
223 	dummy = ib->gpu_addr + 1024;
224 
225 	ib->length_dw = 0;
226 	ib->ptr[ib->length_dw++] = 0x00000018;
227 	ib->ptr[ib->length_dw++] = 0x00000001; /* session info */
228 	ib->ptr[ib->length_dw++] = handle;
229 	ib->ptr[ib->length_dw++] = 0x00000000;
230 	ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
231 	ib->ptr[ib->length_dw++] = dummy;
232 
233 	ib->ptr[ib->length_dw++] = 0x00000014;
234 	ib->ptr[ib->length_dw++] = 0x00000002; /* task info */
235 	ib->ptr[ib->length_dw++] = 0x0000001c;
236 	ib->ptr[ib->length_dw++] = 0x00000000;
237 	ib->ptr[ib->length_dw++] = 0x00000000;
238 
239 	ib->ptr[ib->length_dw++] = 0x00000008;
240 	ib->ptr[ib->length_dw++] = 0x08000001; /* op initialize */
241 
242 	for (i = ib->length_dw; i < ib_size_dw; ++i)
243 		ib->ptr[i] = 0x0;
244 
245 	r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f);
246 	job->fence = dma_fence_get(f);
247 	if (r)
248 		goto err;
249 
250 	amdgpu_job_free(job);
251 	if (fence)
252 		*fence = dma_fence_get(f);
253 	dma_fence_put(f);
254 	return 0;
255 
256 err:
257 	amdgpu_job_free(job);
258 	return r;
259 }
260 
261 /**
262  * uvd_v7_0_enc_get_destroy_msg - generate a UVD ENC destroy msg
263  *
264  * @adev: amdgpu_device pointer
265  * @ring: ring we should submit the msg to
266  * @handle: session handle to use
267  * @fence: optional fence to return
268  *
269  * Close up a stream for HW test or if userspace failed to do so
270  */
271 int uvd_v7_0_enc_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
272 				 bool direct, struct dma_fence **fence)
273 {
274 	const unsigned ib_size_dw = 16;
275 	struct amdgpu_job *job;
276 	struct amdgpu_ib *ib;
277 	struct dma_fence *f = NULL;
278 	uint64_t dummy;
279 	int i, r;
280 
281 	r = amdgpu_job_alloc_with_ib(ring->adev, ib_size_dw * 4, &job);
282 	if (r)
283 		return r;
284 
285 	ib = &job->ibs[0];
286 	dummy = ib->gpu_addr + 1024;
287 
288 	ib->length_dw = 0;
289 	ib->ptr[ib->length_dw++] = 0x00000018;
290 	ib->ptr[ib->length_dw++] = 0x00000001;
291 	ib->ptr[ib->length_dw++] = handle;
292 	ib->ptr[ib->length_dw++] = 0x00000000;
293 	ib->ptr[ib->length_dw++] = upper_32_bits(dummy);
294 	ib->ptr[ib->length_dw++] = dummy;
295 
296 	ib->ptr[ib->length_dw++] = 0x00000014;
297 	ib->ptr[ib->length_dw++] = 0x00000002;
298 	ib->ptr[ib->length_dw++] = 0x0000001c;
299 	ib->ptr[ib->length_dw++] = 0x00000000;
300 	ib->ptr[ib->length_dw++] = 0x00000000;
301 
302 	ib->ptr[ib->length_dw++] = 0x00000008;
303 	ib->ptr[ib->length_dw++] = 0x08000002; /* op close session */
304 
305 	for (i = ib->length_dw; i < ib_size_dw; ++i)
306 		ib->ptr[i] = 0x0;
307 
308 	if (direct) {
309 		r = amdgpu_ib_schedule(ring, 1, ib, NULL, &f);
310 		job->fence = dma_fence_get(f);
311 		if (r)
312 			goto err;
313 
314 		amdgpu_job_free(job);
315 	} else {
316 		r = amdgpu_job_submit(job, ring, &ring->adev->vce.entity,
317 				      AMDGPU_FENCE_OWNER_UNDEFINED, &f);
318 		if (r)
319 			goto err;
320 	}
321 
322 	if (fence)
323 		*fence = dma_fence_get(f);
324 	dma_fence_put(f);
325 	return 0;
326 
327 err:
328 	amdgpu_job_free(job);
329 	return r;
330 }
331 
332 /**
333  * uvd_v7_0_enc_ring_test_ib - test if UVD ENC IBs are working
334  *
335  * @ring: the engine to test on
336  *
337  */
338 static int uvd_v7_0_enc_ring_test_ib(struct amdgpu_ring *ring, long timeout)
339 {
340 	struct dma_fence *fence = NULL;
341 	long r;
342 
343 	r = uvd_v7_0_enc_get_create_msg(ring, 1, NULL);
344 	if (r) {
345 		DRM_ERROR("amdgpu: failed to get create msg (%ld).\n", r);
346 		goto error;
347 	}
348 
349 	r = uvd_v7_0_enc_get_destroy_msg(ring, 1, true, &fence);
350 	if (r) {
351 		DRM_ERROR("amdgpu: failed to get destroy ib (%ld).\n", r);
352 		goto error;
353 	}
354 
355 	r = dma_fence_wait_timeout(fence, false, timeout);
356 	if (r == 0) {
357 		DRM_ERROR("amdgpu: IB test timed out.\n");
358 		r = -ETIMEDOUT;
359 	} else if (r < 0) {
360 		DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
361 	} else {
362 		DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
363 		r = 0;
364 	}
365 error:
366 	dma_fence_put(fence);
367 	return r;
368 }
369 
370 static int uvd_v7_0_early_init(void *handle)
371 {
372 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
373 
374 	if (amdgpu_sriov_vf(adev))
375 		adev->uvd.num_enc_rings = 1;
376 	else
377 		adev->uvd.num_enc_rings = 2;
378 	uvd_v7_0_set_ring_funcs(adev);
379 	uvd_v7_0_set_enc_ring_funcs(adev);
380 	uvd_v7_0_set_irq_funcs(adev);
381 
382 	return 0;
383 }
384 
385 static int uvd_v7_0_sw_init(void *handle)
386 {
387 	struct amdgpu_ring *ring;
388 	struct amd_sched_rq *rq;
389 	int i, r;
390 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
391 
392 	/* UVD TRAP */
393 	r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_UVD, 124, &adev->uvd.irq);
394 	if (r)
395 		return r;
396 
397 	/* UVD ENC TRAP */
398 	for (i = 0; i < adev->uvd.num_enc_rings; ++i) {
399 		r = amdgpu_irq_add_id(adev, AMDGPU_IH_CLIENTID_UVD, i + 119, &adev->uvd.irq);
400 		if (r)
401 			return r;
402 	}
403 
404 	r = amdgpu_uvd_sw_init(adev);
405 	if (r)
406 		return r;
407 
408 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
409 		const struct common_firmware_header *hdr;
410 		hdr = (const struct common_firmware_header *)adev->uvd.fw->data;
411 		adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].ucode_id = AMDGPU_UCODE_ID_UVD;
412 		adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].fw = adev->uvd.fw;
413 		adev->firmware.fw_size +=
414 			ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
415 		DRM_INFO("PSP loading UVD firmware\n");
416 	}
417 
418 	ring = &adev->uvd.ring_enc[0];
419 	rq = &ring->sched.sched_rq[AMD_SCHED_PRIORITY_NORMAL];
420 	r = amd_sched_entity_init(&ring->sched, &adev->uvd.entity_enc,
421 				  rq, amdgpu_sched_jobs);
422 	if (r) {
423 		DRM_ERROR("Failed setting up UVD ENC run queue.\n");
424 		return r;
425 	}
426 
427 	r = amdgpu_uvd_resume(adev);
428 	if (r)
429 		return r;
430 	if (!amdgpu_sriov_vf(adev)) {
431 		ring = &adev->uvd.ring;
432 		sprintf(ring->name, "uvd");
433 		r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.irq, 0);
434 		if (r)
435 			return r;
436 	}
437 
438 	for (i = 0; i < adev->uvd.num_enc_rings; ++i) {
439 		ring = &adev->uvd.ring_enc[i];
440 		sprintf(ring->name, "uvd_enc%d", i);
441 		if (amdgpu_sriov_vf(adev)) {
442 			ring->use_doorbell = true;
443 
444 			/* currently only use the first enconding ring for
445 			 * sriov, so set unused location for other unused rings.
446 			 */
447 			if (i == 0)
448 				ring->doorbell_index = AMDGPU_DOORBELL64_UVD_RING0_1 * 2;
449 			else
450 				ring->doorbell_index = AMDGPU_DOORBELL64_UVD_RING2_3 * 2 + 1;
451 		}
452 		r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.irq, 0);
453 		if (r)
454 			return r;
455 	}
456 
457 	r = amdgpu_virt_alloc_mm_table(adev);
458 	if (r)
459 		return r;
460 
461 	return r;
462 }
463 
464 static int uvd_v7_0_sw_fini(void *handle)
465 {
466 	int i, r;
467 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
468 
469 	amdgpu_virt_free_mm_table(adev);
470 
471 	r = amdgpu_uvd_suspend(adev);
472 	if (r)
473 		return r;
474 
475 	amd_sched_entity_fini(&adev->uvd.ring_enc[0].sched, &adev->uvd.entity_enc);
476 
477 	for (i = 0; i < adev->uvd.num_enc_rings; ++i)
478 		amdgpu_ring_fini(&adev->uvd.ring_enc[i]);
479 
480 	return amdgpu_uvd_sw_fini(adev);
481 }
482 
483 /**
484  * uvd_v7_0_hw_init - start and test UVD block
485  *
486  * @adev: amdgpu_device pointer
487  *
488  * Initialize the hardware, boot up the VCPU and do some testing
489  */
490 static int uvd_v7_0_hw_init(void *handle)
491 {
492 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
493 	struct amdgpu_ring *ring = &adev->uvd.ring;
494 	uint32_t tmp;
495 	int i, r;
496 
497 	if (amdgpu_sriov_vf(adev))
498 		r = uvd_v7_0_sriov_start(adev);
499 	else
500 		r = uvd_v7_0_start(adev);
501 	if (r)
502 		goto done;
503 
504 	if (!amdgpu_sriov_vf(adev)) {
505 		ring->ready = true;
506 		r = amdgpu_ring_test_ring(ring);
507 		if (r) {
508 			ring->ready = false;
509 			goto done;
510 		}
511 
512 		r = amdgpu_ring_alloc(ring, 10);
513 		if (r) {
514 			DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
515 			goto done;
516 		}
517 
518 		tmp = PACKET0(SOC15_REG_OFFSET(UVD, 0,
519 			mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL), 0);
520 		amdgpu_ring_write(ring, tmp);
521 		amdgpu_ring_write(ring, 0xFFFFF);
522 
523 		tmp = PACKET0(SOC15_REG_OFFSET(UVD, 0,
524 			mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL), 0);
525 		amdgpu_ring_write(ring, tmp);
526 		amdgpu_ring_write(ring, 0xFFFFF);
527 
528 		tmp = PACKET0(SOC15_REG_OFFSET(UVD, 0,
529 			mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL), 0);
530 		amdgpu_ring_write(ring, tmp);
531 		amdgpu_ring_write(ring, 0xFFFFF);
532 
533 		/* Clear timeout status bits */
534 		amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0,
535 			mmUVD_SEMA_TIMEOUT_STATUS), 0));
536 		amdgpu_ring_write(ring, 0x8);
537 
538 		amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0,
539 			mmUVD_SEMA_CNTL), 0));
540 		amdgpu_ring_write(ring, 3);
541 
542 		amdgpu_ring_commit(ring);
543 	}
544 
545 	for (i = 0; i < adev->uvd.num_enc_rings; ++i) {
546 		ring = &adev->uvd.ring_enc[i];
547 		ring->ready = true;
548 		r = amdgpu_ring_test_ring(ring);
549 		if (r) {
550 			ring->ready = false;
551 			goto done;
552 		}
553 	}
554 
555 done:
556 	if (!r)
557 		DRM_INFO("UVD and UVD ENC initialized successfully.\n");
558 
559 	return r;
560 }
561 
562 /**
563  * uvd_v7_0_hw_fini - stop the hardware block
564  *
565  * @adev: amdgpu_device pointer
566  *
567  * Stop the UVD block, mark ring as not ready any more
568  */
569 static int uvd_v7_0_hw_fini(void *handle)
570 {
571 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
572 	struct amdgpu_ring *ring = &adev->uvd.ring;
573 
574 	if (!amdgpu_sriov_vf(adev))
575 		uvd_v7_0_stop(adev);
576 	else {
577 		/* full access mode, so don't touch any UVD register */
578 		DRM_DEBUG("For SRIOV client, shouldn't do anything.\n");
579 	}
580 
581 	ring->ready = false;
582 
583 	return 0;
584 }
585 
586 static int uvd_v7_0_suspend(void *handle)
587 {
588 	int r;
589 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
590 
591 	r = uvd_v7_0_hw_fini(adev);
592 	if (r)
593 		return r;
594 
595 	/* Skip this for APU for now */
596 	if (!(adev->flags & AMD_IS_APU))
597 		r = amdgpu_uvd_suspend(adev);
598 
599 	return r;
600 }
601 
602 static int uvd_v7_0_resume(void *handle)
603 {
604 	int r;
605 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
606 
607 	/* Skip this for APU for now */
608 	if (!(adev->flags & AMD_IS_APU)) {
609 		r = amdgpu_uvd_resume(adev);
610 		if (r)
611 			return r;
612 	}
613 	return uvd_v7_0_hw_init(adev);
614 }
615 
616 /**
617  * uvd_v7_0_mc_resume - memory controller programming
618  *
619  * @adev: amdgpu_device pointer
620  *
621  * Let the UVD memory controller know it's offsets
622  */
623 static void uvd_v7_0_mc_resume(struct amdgpu_device *adev)
624 {
625 	uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4);
626 	uint32_t offset;
627 
628 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
629 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
630 			lower_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
631 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
632 			upper_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
633 		offset = 0;
634 	} else {
635 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
636 			lower_32_bits(adev->uvd.gpu_addr));
637 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
638 			upper_32_bits(adev->uvd.gpu_addr));
639 		offset = size;
640 	}
641 
642 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0,
643 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
644 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size);
645 
646 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
647 			lower_32_bits(adev->uvd.gpu_addr + offset));
648 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
649 			upper_32_bits(adev->uvd.gpu_addr + offset));
650 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, (1 << 21));
651 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_UVD_HEAP_SIZE);
652 
653 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
654 			lower_32_bits(adev->uvd.gpu_addr + offset + AMDGPU_UVD_HEAP_SIZE));
655 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
656 			upper_32_bits(adev->uvd.gpu_addr + offset + AMDGPU_UVD_HEAP_SIZE));
657 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, (2 << 21));
658 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2,
659 			AMDGPU_UVD_STACK_SIZE + (AMDGPU_UVD_SESSION_SIZE * 40));
660 
661 	WREG32_SOC15(UVD, 0, mmUVD_UDEC_ADDR_CONFIG,
662 			adev->gfx.config.gb_addr_config);
663 	WREG32_SOC15(UVD, 0, mmUVD_UDEC_DB_ADDR_CONFIG,
664 			adev->gfx.config.gb_addr_config);
665 	WREG32_SOC15(UVD, 0, mmUVD_UDEC_DBW_ADDR_CONFIG,
666 			adev->gfx.config.gb_addr_config);
667 
668 	WREG32_SOC15(UVD, 0, mmUVD_GP_SCRATCH4, adev->uvd.max_handles);
669 }
670 
671 static int uvd_v7_0_mmsch_start(struct amdgpu_device *adev,
672 				struct amdgpu_mm_table *table)
673 {
674 	uint32_t data = 0, loop;
675 	uint64_t addr = table->gpu_addr;
676 	struct mmsch_v1_0_init_header *header = (struct mmsch_v1_0_init_header *)table->cpu_addr;
677 	uint32_t size;
678 
679 	size = header->header_size + header->vce_table_size + header->uvd_table_size;
680 
681 	/* 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of memory descriptor location */
682 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
683 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
684 
685 	/* 2, update vmid of descriptor */
686 	data = RREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_VMID);
687 	data &= ~VCE_MMSCH_VF_VMID__VF_CTX_VMID_MASK;
688 	data |= (0 << VCE_MMSCH_VF_VMID__VF_CTX_VMID__SHIFT); /* use domain0 for MM scheduler */
689 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_VMID, data);
690 
691 	/* 3, notify mmsch about the size of this descriptor */
692 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_CTX_SIZE, size);
693 
694 	/* 4, set resp to zero */
695 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_RESP, 0);
696 
697 	WDOORBELL32(adev->uvd.ring_enc[0].doorbell_index, 0);
698 	adev->wb.wb[adev->uvd.ring_enc[0].wptr_offs] = 0;
699 	adev->uvd.ring_enc[0].wptr = 0;
700 	adev->uvd.ring_enc[0].wptr_old = 0;
701 
702 	/* 5, kick off the initialization and wait until VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero */
703 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_HOST, 0x10000001);
704 
705 	data = RREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_RESP);
706 	loop = 1000;
707 	while ((data & 0x10000002) != 0x10000002) {
708 		udelay(10);
709 		data = RREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_RESP);
710 		loop--;
711 		if (!loop)
712 			break;
713 	}
714 
715 	if (!loop) {
716 		dev_err(adev->dev, "failed to init MMSCH, mmVCE_MMSCH_VF_MAILBOX_RESP = %x\n", data);
717 		return -EBUSY;
718 	}
719 
720 	return 0;
721 }
722 
723 static int uvd_v7_0_sriov_start(struct amdgpu_device *adev)
724 {
725 	struct amdgpu_ring *ring;
726 	uint32_t offset, size, tmp;
727 	uint32_t table_size = 0;
728 	struct mmsch_v1_0_cmd_direct_write direct_wt = { {0} };
729 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { {0} };
730 	struct mmsch_v1_0_cmd_direct_polling direct_poll = { {0} };
731 	struct mmsch_v1_0_cmd_end end = { {0} };
732 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
733 	struct mmsch_v1_0_init_header *header = (struct mmsch_v1_0_init_header *)init_table;
734 
735 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
736 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
737 	direct_poll.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_POLLING;
738 	end.cmd_header.command_type = MMSCH_COMMAND__END;
739 
740 	if (header->uvd_table_offset == 0 && header->uvd_table_size == 0) {
741 		header->version = MMSCH_VERSION;
742 		header->header_size = sizeof(struct mmsch_v1_0_init_header) >> 2;
743 
744 		if (header->vce_table_offset == 0 && header->vce_table_size == 0)
745 			header->uvd_table_offset = header->header_size;
746 		else
747 			header->uvd_table_offset = header->vce_table_size + header->vce_table_offset;
748 
749 		init_table += header->uvd_table_offset;
750 
751 		ring = &adev->uvd.ring;
752 		ring->wptr = 0;
753 		size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4);
754 
755 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS),
756 						   0xFFFFFFFF, 0x00000004);
757 		/* mc resume*/
758 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
759 			MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
760 						    lower_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
761 			MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
762 						    upper_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
763 			offset = 0;
764 		} else {
765 			MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
766 						    lower_32_bits(adev->uvd.gpu_addr));
767 			MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
768 						    upper_32_bits(adev->uvd.gpu_addr));
769 			offset = size;
770 		}
771 
772 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0),
773 					    AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
774 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size);
775 
776 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
777 					    lower_32_bits(adev->uvd.gpu_addr + offset));
778 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
779 					    upper_32_bits(adev->uvd.gpu_addr + offset));
780 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), (1 << 21));
781 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_UVD_HEAP_SIZE);
782 
783 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
784 					    lower_32_bits(adev->uvd.gpu_addr + offset + AMDGPU_UVD_HEAP_SIZE));
785 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
786 					    upper_32_bits(adev->uvd.gpu_addr + offset + AMDGPU_UVD_HEAP_SIZE));
787 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), (2 << 21));
788 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_SIZE2),
789 					    AMDGPU_UVD_STACK_SIZE + (AMDGPU_UVD_SESSION_SIZE * 40));
790 
791 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_GP_SCRATCH4), adev->uvd.max_handles);
792 		/* mc resume end*/
793 
794 		/* disable clock gating */
795 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_CGC_CTRL),
796 						   ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK, 0);
797 
798 		/* disable interupt */
799 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
800 						   ~UVD_MASTINT_EN__VCPU_EN_MASK, 0);
801 
802 		/* stall UMC and register bus before resetting VCPU */
803 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2),
804 						   ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK,
805 						   UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
806 
807 		/* put LMI, VCPU, RBC etc... into reset */
808 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
809 					    (uint32_t)(UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
810 						       UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
811 						       UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
812 						       UVD_SOFT_RESET__RBC_SOFT_RESET_MASK |
813 						       UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
814 						       UVD_SOFT_RESET__CXW_SOFT_RESET_MASK |
815 						       UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
816 						       UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK));
817 
818 		/* initialize UVD memory controller */
819 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL),
820 					    (uint32_t)((0x40 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
821 						       UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
822 						       UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
823 						       UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
824 						       UVD_LMI_CTRL__REQ_MODE_MASK |
825 						       0x00100000L));
826 
827 		/* take all subblocks out of reset, except VCPU */
828 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
829 					    UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
830 
831 		/* enable VCPU clock */
832 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL),
833 					    UVD_VCPU_CNTL__CLK_EN_MASK);
834 
835 		/* enable master interrupt */
836 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
837 						   ~(UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK),
838 						   (UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK));
839 
840 		/* clear the bit 4 of UVD_STATUS */
841 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS),
842 						   ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT), 0);
843 
844 		/* force RBC into idle state */
845 		size = order_base_2(ring->ring_size);
846 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, size);
847 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
848 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), tmp);
849 
850 		ring = &adev->uvd.ring_enc[0];
851 		ring->wptr = 0;
852 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_BASE_LO), ring->gpu_addr);
853 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_BASE_HI), upper_32_bits(ring->gpu_addr));
854 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_SIZE), ring->ring_size / 4);
855 
856 		/* boot up the VCPU */
857 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0);
858 
859 		/* enable UMC */
860 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2),
861 										   ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK, 0);
862 
863 		MMSCH_V1_0_INSERT_DIRECT_POLL(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0x02, 0x02);
864 
865 		/* add end packet */
866 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
867 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
868 		header->uvd_table_size = table_size;
869 
870 	}
871 	return uvd_v7_0_mmsch_start(adev, &adev->virt.mm_table);
872 }
873 
874 /**
875  * uvd_v7_0_start - start UVD block
876  *
877  * @adev: amdgpu_device pointer
878  *
879  * Setup and start the UVD block
880  */
881 static int uvd_v7_0_start(struct amdgpu_device *adev)
882 {
883 	struct amdgpu_ring *ring = &adev->uvd.ring;
884 	uint32_t rb_bufsz, tmp;
885 	uint32_t lmi_swap_cntl;
886 	uint32_t mp_swap_cntl;
887 	int i, j, r;
888 
889 	/* disable DPG */
890 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0,
891 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
892 
893 	/* disable byte swapping */
894 	lmi_swap_cntl = 0;
895 	mp_swap_cntl = 0;
896 
897 	uvd_v7_0_mc_resume(adev);
898 
899 	/* disable clock gating */
900 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_CGC_CTRL), 0,
901 			~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK);
902 
903 	/* disable interupt */
904 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0,
905 			~UVD_MASTINT_EN__VCPU_EN_MASK);
906 
907 	/* stall UMC and register bus before resetting VCPU */
908 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2),
909 			UVD_LMI_CTRL2__STALL_ARB_UMC_MASK,
910 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
911 	mdelay(1);
912 
913 	/* put LMI, VCPU, RBC etc... into reset */
914 	WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET,
915 		UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
916 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
917 		UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
918 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK |
919 		UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
920 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK |
921 		UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
922 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
923 	mdelay(5);
924 
925 	/* initialize UVD memory controller */
926 	WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL,
927 		(0x40 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
928 		UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
929 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
930 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
931 		UVD_LMI_CTRL__REQ_MODE_MASK |
932 		0x00100000L);
933 
934 #ifdef __BIG_ENDIAN
935 	/* swap (8 in 32) RB and IB */
936 	lmi_swap_cntl = 0xa;
937 	mp_swap_cntl = 0;
938 #endif
939 	WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
940 	WREG32_SOC15(UVD, 0, mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
941 
942 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0, 0x40c2040);
943 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA1, 0x0);
944 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0, 0x40c2040);
945 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB1, 0x0);
946 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_ALU, 0);
947 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX, 0x88);
948 
949 	/* take all subblocks out of reset, except VCPU */
950 	WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET,
951 			UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
952 	mdelay(5);
953 
954 	/* enable VCPU clock */
955 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL,
956 			UVD_VCPU_CNTL__CLK_EN_MASK);
957 
958 	/* enable UMC */
959 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0,
960 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
961 
962 	/* boot up the VCPU */
963 	WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, 0);
964 	mdelay(10);
965 
966 	for (i = 0; i < 10; ++i) {
967 		uint32_t status;
968 
969 		for (j = 0; j < 100; ++j) {
970 			status = RREG32_SOC15(UVD, 0, mmUVD_STATUS);
971 			if (status & 2)
972 				break;
973 			mdelay(10);
974 		}
975 		r = 0;
976 		if (status & 2)
977 			break;
978 
979 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
980 		WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
981 				UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
982 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
983 		mdelay(10);
984 		WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0,
985 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
986 		mdelay(10);
987 		r = -1;
988 	}
989 
990 	if (r) {
991 		DRM_ERROR("UVD not responding, giving up!!!\n");
992 		return r;
993 	}
994 	/* enable master interrupt */
995 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
996 		(UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK),
997 		~(UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK));
998 
999 	/* clear the bit 4 of UVD_STATUS */
1000 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0,
1001 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1002 
1003 	/* force RBC into idle state */
1004 	rb_bufsz = order_base_2(ring->ring_size);
1005 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1006 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1007 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1008 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0);
1009 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1010 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1011 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp);
1012 
1013 	/* set the write pointer delay */
1014 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0);
1015 
1016 	/* set the wb address */
1017 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR,
1018 			(upper_32_bits(ring->gpu_addr) >> 2));
1019 
1020 	/* programm the RB_BASE for ring buffer */
1021 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1022 			lower_32_bits(ring->gpu_addr));
1023 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1024 			upper_32_bits(ring->gpu_addr));
1025 
1026 	/* Initialize the ring buffer's read and write pointers */
1027 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0);
1028 
1029 	ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1030 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1031 			lower_32_bits(ring->wptr));
1032 
1033 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0,
1034 			~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK);
1035 
1036 	ring = &adev->uvd.ring_enc[0];
1037 	WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1038 	WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1039 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1040 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1041 	WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1042 
1043 	ring = &adev->uvd.ring_enc[1];
1044 	WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1045 	WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1046 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1047 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1048 	WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1049 
1050 	return 0;
1051 }
1052 
1053 /**
1054  * uvd_v7_0_stop - stop UVD block
1055  *
1056  * @adev: amdgpu_device pointer
1057  *
1058  * stop the UVD block
1059  */
1060 static void uvd_v7_0_stop(struct amdgpu_device *adev)
1061 {
1062 	/* force RBC into idle state */
1063 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, 0x11010101);
1064 
1065 	/* Stall UMC and register bus before resetting VCPU */
1066 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2),
1067 			UVD_LMI_CTRL2__STALL_ARB_UMC_MASK,
1068 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1069 	mdelay(1);
1070 
1071 	/* put VCPU into reset */
1072 	WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET,
1073 			UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1074 	mdelay(5);
1075 
1076 	/* disable VCPU clock */
1077 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL, 0x0);
1078 
1079 	/* Unstall UMC and register bus */
1080 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0,
1081 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1082 }
1083 
1084 /**
1085  * uvd_v7_0_ring_emit_fence - emit an fence & trap command
1086  *
1087  * @ring: amdgpu_ring pointer
1088  * @fence: fence to emit
1089  *
1090  * Write a fence and a trap command to the ring.
1091  */
1092 static void uvd_v7_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
1093 				     unsigned flags)
1094 {
1095 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1096 
1097 	amdgpu_ring_write(ring,
1098 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0));
1099 	amdgpu_ring_write(ring, seq);
1100 	amdgpu_ring_write(ring,
1101 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
1102 	amdgpu_ring_write(ring, addr & 0xffffffff);
1103 	amdgpu_ring_write(ring,
1104 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
1105 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
1106 	amdgpu_ring_write(ring,
1107 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
1108 	amdgpu_ring_write(ring, 0);
1109 
1110 	amdgpu_ring_write(ring,
1111 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
1112 	amdgpu_ring_write(ring, 0);
1113 	amdgpu_ring_write(ring,
1114 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
1115 	amdgpu_ring_write(ring, 0);
1116 	amdgpu_ring_write(ring,
1117 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
1118 	amdgpu_ring_write(ring, 2);
1119 }
1120 
1121 /**
1122  * uvd_v7_0_enc_ring_emit_fence - emit an enc fence & trap command
1123  *
1124  * @ring: amdgpu_ring pointer
1125  * @fence: fence to emit
1126  *
1127  * Write enc a fence and a trap command to the ring.
1128  */
1129 static void uvd_v7_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1130 			u64 seq, unsigned flags)
1131 {
1132 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1133 
1134 	amdgpu_ring_write(ring, HEVC_ENC_CMD_FENCE);
1135 	amdgpu_ring_write(ring, addr);
1136 	amdgpu_ring_write(ring, upper_32_bits(addr));
1137 	amdgpu_ring_write(ring, seq);
1138 	amdgpu_ring_write(ring, HEVC_ENC_CMD_TRAP);
1139 }
1140 
1141 /**
1142  * uvd_v7_0_ring_emit_hdp_flush - emit an hdp flush
1143  *
1144  * @ring: amdgpu_ring pointer
1145  *
1146  * Emits an hdp flush.
1147  */
1148 static void uvd_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
1149 {
1150 	amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(NBIF, 0,
1151 		mmHDP_MEM_COHERENCY_FLUSH_CNTL), 0));
1152 	amdgpu_ring_write(ring, 0);
1153 }
1154 
1155 /**
1156  * uvd_v7_0_ring_hdp_invalidate - emit an hdp invalidate
1157  *
1158  * @ring: amdgpu_ring pointer
1159  *
1160  * Emits an hdp invalidate.
1161  */
1162 static void uvd_v7_0_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
1163 {
1164 	amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(HDP, 0, mmHDP_DEBUG0), 0));
1165 	amdgpu_ring_write(ring, 1);
1166 }
1167 
1168 /**
1169  * uvd_v7_0_ring_test_ring - register write test
1170  *
1171  * @ring: amdgpu_ring pointer
1172  *
1173  * Test if we can successfully write to the context register
1174  */
1175 static int uvd_v7_0_ring_test_ring(struct amdgpu_ring *ring)
1176 {
1177 	struct amdgpu_device *adev = ring->adev;
1178 	uint32_t tmp = 0;
1179 	unsigned i;
1180 	int r;
1181 
1182 	WREG32_SOC15(UVD, 0, mmUVD_CONTEXT_ID, 0xCAFEDEAD);
1183 	r = amdgpu_ring_alloc(ring, 3);
1184 	if (r) {
1185 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
1186 			  ring->idx, r);
1187 		return r;
1188 	}
1189 	amdgpu_ring_write(ring,
1190 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0));
1191 	amdgpu_ring_write(ring, 0xDEADBEEF);
1192 	amdgpu_ring_commit(ring);
1193 	for (i = 0; i < adev->usec_timeout; i++) {
1194 		tmp = RREG32_SOC15(UVD, 0, mmUVD_CONTEXT_ID);
1195 		if (tmp == 0xDEADBEEF)
1196 			break;
1197 		DRM_UDELAY(1);
1198 	}
1199 
1200 	if (i < adev->usec_timeout) {
1201 		DRM_INFO("ring test on %d succeeded in %d usecs\n",
1202 			 ring->idx, i);
1203 	} else {
1204 		DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
1205 			  ring->idx, tmp);
1206 		r = -EINVAL;
1207 	}
1208 	return r;
1209 }
1210 
1211 /**
1212  * uvd_v7_0_ring_emit_ib - execute indirect buffer
1213  *
1214  * @ring: amdgpu_ring pointer
1215  * @ib: indirect buffer to execute
1216  *
1217  * Write ring commands to execute the indirect buffer
1218  */
1219 static void uvd_v7_0_ring_emit_ib(struct amdgpu_ring *ring,
1220 				  struct amdgpu_ib *ib,
1221 				  unsigned vm_id, bool ctx_switch)
1222 {
1223 	amdgpu_ring_write(ring,
1224 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_VMID), 0));
1225 	amdgpu_ring_write(ring, vm_id);
1226 
1227 	amdgpu_ring_write(ring,
1228 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_LOW), 0));
1229 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1230 	amdgpu_ring_write(ring,
1231 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH), 0));
1232 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1233 	amdgpu_ring_write(ring,
1234 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_IB_SIZE), 0));
1235 	amdgpu_ring_write(ring, ib->length_dw);
1236 }
1237 
1238 /**
1239  * uvd_v7_0_enc_ring_emit_ib - enc execute indirect buffer
1240  *
1241  * @ring: amdgpu_ring pointer
1242  * @ib: indirect buffer to execute
1243  *
1244  * Write enc ring commands to execute the indirect buffer
1245  */
1246 static void uvd_v7_0_enc_ring_emit_ib(struct amdgpu_ring *ring,
1247 		struct amdgpu_ib *ib, unsigned int vm_id, bool ctx_switch)
1248 {
1249 	amdgpu_ring_write(ring, HEVC_ENC_CMD_IB_VM);
1250 	amdgpu_ring_write(ring, vm_id);
1251 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1252 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1253 	amdgpu_ring_write(ring, ib->length_dw);
1254 }
1255 
1256 static void uvd_v7_0_vm_reg_write(struct amdgpu_ring *ring,
1257 				uint32_t data0, uint32_t data1)
1258 {
1259 	amdgpu_ring_write(ring,
1260 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
1261 	amdgpu_ring_write(ring, data0);
1262 	amdgpu_ring_write(ring,
1263 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
1264 	amdgpu_ring_write(ring, data1);
1265 	amdgpu_ring_write(ring,
1266 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
1267 	amdgpu_ring_write(ring, 8);
1268 }
1269 
1270 static void uvd_v7_0_vm_reg_wait(struct amdgpu_ring *ring,
1271 				uint32_t data0, uint32_t data1, uint32_t mask)
1272 {
1273 	amdgpu_ring_write(ring,
1274 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
1275 	amdgpu_ring_write(ring, data0);
1276 	amdgpu_ring_write(ring,
1277 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
1278 	amdgpu_ring_write(ring, data1);
1279 	amdgpu_ring_write(ring,
1280 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GP_SCRATCH8), 0));
1281 	amdgpu_ring_write(ring, mask);
1282 	amdgpu_ring_write(ring,
1283 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
1284 	amdgpu_ring_write(ring, 12);
1285 }
1286 
1287 static void uvd_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
1288 					unsigned vm_id, uint64_t pd_addr)
1289 {
1290 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1291 	uint32_t req = ring->adev->gart.gart_funcs->get_invalidate_req(vm_id);
1292 	uint32_t data0, data1, mask;
1293 	unsigned eng = ring->vm_inv_eng;
1294 
1295 	pd_addr = amdgpu_gart_get_vm_pde(ring->adev, pd_addr);
1296 	pd_addr |= AMDGPU_PTE_VALID;
1297 
1298 	data0 = (hub->ctx0_ptb_addr_hi32 + vm_id * 2) << 2;
1299 	data1 = upper_32_bits(pd_addr);
1300 	uvd_v7_0_vm_reg_write(ring, data0, data1);
1301 
1302 	data0 = (hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2;
1303 	data1 = lower_32_bits(pd_addr);
1304 	uvd_v7_0_vm_reg_write(ring, data0, data1);
1305 
1306 	data0 = (hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2;
1307 	data1 = lower_32_bits(pd_addr);
1308 	mask = 0xffffffff;
1309 	uvd_v7_0_vm_reg_wait(ring, data0, data1, mask);
1310 
1311 	/* flush TLB */
1312 	data0 = (hub->vm_inv_eng0_req + eng) << 2;
1313 	data1 = req;
1314 	uvd_v7_0_vm_reg_write(ring, data0, data1);
1315 
1316 	/* wait for flush */
1317 	data0 = (hub->vm_inv_eng0_ack + eng) << 2;
1318 	data1 = 1 << vm_id;
1319 	mask =  1 << vm_id;
1320 	uvd_v7_0_vm_reg_wait(ring, data0, data1, mask);
1321 }
1322 
1323 static void uvd_v7_0_enc_ring_insert_end(struct amdgpu_ring *ring)
1324 {
1325 	amdgpu_ring_write(ring, HEVC_ENC_CMD_END);
1326 }
1327 
1328 static void uvd_v7_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
1329 			 unsigned int vm_id, uint64_t pd_addr)
1330 {
1331 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1332 	uint32_t req = ring->adev->gart.gart_funcs->get_invalidate_req(vm_id);
1333 	unsigned eng = ring->vm_inv_eng;
1334 
1335 	pd_addr = amdgpu_gart_get_vm_pde(ring->adev, pd_addr);
1336 	pd_addr |= AMDGPU_PTE_VALID;
1337 
1338 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WRITE);
1339 	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_hi32 + vm_id * 2) << 2);
1340 	amdgpu_ring_write(ring, upper_32_bits(pd_addr));
1341 
1342 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WRITE);
1343 	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2);
1344 	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
1345 
1346 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WAIT);
1347 	amdgpu_ring_write(ring,	(hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2);
1348 	amdgpu_ring_write(ring, 0xffffffff);
1349 	amdgpu_ring_write(ring, lower_32_bits(pd_addr));
1350 
1351 	/* flush TLB */
1352 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WRITE);
1353 	amdgpu_ring_write(ring,	(hub->vm_inv_eng0_req + eng) << 2);
1354 	amdgpu_ring_write(ring, req);
1355 
1356 	/* wait for flush */
1357 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WAIT);
1358 	amdgpu_ring_write(ring, (hub->vm_inv_eng0_ack + eng) << 2);
1359 	amdgpu_ring_write(ring, 1 << vm_id);
1360 	amdgpu_ring_write(ring, 1 << vm_id);
1361 }
1362 
1363 #if 0
1364 static bool uvd_v7_0_is_idle(void *handle)
1365 {
1366 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1367 
1368 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
1369 }
1370 
1371 static int uvd_v7_0_wait_for_idle(void *handle)
1372 {
1373 	unsigned i;
1374 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1375 
1376 	for (i = 0; i < adev->usec_timeout; i++) {
1377 		if (uvd_v7_0_is_idle(handle))
1378 			return 0;
1379 	}
1380 	return -ETIMEDOUT;
1381 }
1382 
1383 #define AMDGPU_UVD_STATUS_BUSY_MASK    0xfd
1384 static bool uvd_v7_0_check_soft_reset(void *handle)
1385 {
1386 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1387 	u32 srbm_soft_reset = 0;
1388 	u32 tmp = RREG32(mmSRBM_STATUS);
1389 
1390 	if (REG_GET_FIELD(tmp, SRBM_STATUS, UVD_RQ_PENDING) ||
1391 	    REG_GET_FIELD(tmp, SRBM_STATUS, UVD_BUSY) ||
1392 	    (RREG32_SOC15(UVD, 0, mmUVD_STATUS) &
1393 		    AMDGPU_UVD_STATUS_BUSY_MASK))
1394 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1395 				SRBM_SOFT_RESET, SOFT_RESET_UVD, 1);
1396 
1397 	if (srbm_soft_reset) {
1398 		adev->uvd.srbm_soft_reset = srbm_soft_reset;
1399 		return true;
1400 	} else {
1401 		adev->uvd.srbm_soft_reset = 0;
1402 		return false;
1403 	}
1404 }
1405 
1406 static int uvd_v7_0_pre_soft_reset(void *handle)
1407 {
1408 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1409 
1410 	if (!adev->uvd.srbm_soft_reset)
1411 		return 0;
1412 
1413 	uvd_v7_0_stop(adev);
1414 	return 0;
1415 }
1416 
1417 static int uvd_v7_0_soft_reset(void *handle)
1418 {
1419 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1420 	u32 srbm_soft_reset;
1421 
1422 	if (!adev->uvd.srbm_soft_reset)
1423 		return 0;
1424 	srbm_soft_reset = adev->uvd.srbm_soft_reset;
1425 
1426 	if (srbm_soft_reset) {
1427 		u32 tmp;
1428 
1429 		tmp = RREG32(mmSRBM_SOFT_RESET);
1430 		tmp |= srbm_soft_reset;
1431 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1432 		WREG32(mmSRBM_SOFT_RESET, tmp);
1433 		tmp = RREG32(mmSRBM_SOFT_RESET);
1434 
1435 		udelay(50);
1436 
1437 		tmp &= ~srbm_soft_reset;
1438 		WREG32(mmSRBM_SOFT_RESET, tmp);
1439 		tmp = RREG32(mmSRBM_SOFT_RESET);
1440 
1441 		/* Wait a little for things to settle down */
1442 		udelay(50);
1443 	}
1444 
1445 	return 0;
1446 }
1447 
1448 static int uvd_v7_0_post_soft_reset(void *handle)
1449 {
1450 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1451 
1452 	if (!adev->uvd.srbm_soft_reset)
1453 		return 0;
1454 
1455 	mdelay(5);
1456 
1457 	return uvd_v7_0_start(adev);
1458 }
1459 #endif
1460 
1461 static int uvd_v7_0_set_interrupt_state(struct amdgpu_device *adev,
1462 					struct amdgpu_irq_src *source,
1463 					unsigned type,
1464 					enum amdgpu_interrupt_state state)
1465 {
1466 	// TODO
1467 	return 0;
1468 }
1469 
1470 static int uvd_v7_0_process_interrupt(struct amdgpu_device *adev,
1471 				      struct amdgpu_irq_src *source,
1472 				      struct amdgpu_iv_entry *entry)
1473 {
1474 	DRM_DEBUG("IH: UVD TRAP\n");
1475 	switch (entry->src_id) {
1476 	case 124:
1477 		amdgpu_fence_process(&adev->uvd.ring);
1478 		break;
1479 	case 119:
1480 		amdgpu_fence_process(&adev->uvd.ring_enc[0]);
1481 		break;
1482 	case 120:
1483 		if (!amdgpu_sriov_vf(adev))
1484 			amdgpu_fence_process(&adev->uvd.ring_enc[1]);
1485 		break;
1486 	default:
1487 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1488 			  entry->src_id, entry->src_data[0]);
1489 		break;
1490 	}
1491 
1492 	return 0;
1493 }
1494 
1495 #if 0
1496 static void uvd_v7_0_set_sw_clock_gating(struct amdgpu_device *adev)
1497 {
1498 	uint32_t data, data1, data2, suvd_flags;
1499 
1500 	data = RREG32_SOC15(UVD, 0, mmUVD_CGC_CTRL);
1501 	data1 = RREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_GATE);
1502 	data2 = RREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_CTRL);
1503 
1504 	data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
1505 		  UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
1506 
1507 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
1508 		     UVD_SUVD_CGC_GATE__SIT_MASK |
1509 		     UVD_SUVD_CGC_GATE__SMP_MASK |
1510 		     UVD_SUVD_CGC_GATE__SCM_MASK |
1511 		     UVD_SUVD_CGC_GATE__SDB_MASK;
1512 
1513 	data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
1514 		(1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) |
1515 		(4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY));
1516 
1517 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
1518 			UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
1519 			UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
1520 			UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
1521 			UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
1522 			UVD_CGC_CTRL__SYS_MODE_MASK |
1523 			UVD_CGC_CTRL__UDEC_MODE_MASK |
1524 			UVD_CGC_CTRL__MPEG2_MODE_MASK |
1525 			UVD_CGC_CTRL__REGS_MODE_MASK |
1526 			UVD_CGC_CTRL__RBC_MODE_MASK |
1527 			UVD_CGC_CTRL__LMI_MC_MODE_MASK |
1528 			UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
1529 			UVD_CGC_CTRL__IDCT_MODE_MASK |
1530 			UVD_CGC_CTRL__MPRD_MODE_MASK |
1531 			UVD_CGC_CTRL__MPC_MODE_MASK |
1532 			UVD_CGC_CTRL__LBSI_MODE_MASK |
1533 			UVD_CGC_CTRL__LRBBM_MODE_MASK |
1534 			UVD_CGC_CTRL__WCB_MODE_MASK |
1535 			UVD_CGC_CTRL__VCPU_MODE_MASK |
1536 			UVD_CGC_CTRL__JPEG_MODE_MASK |
1537 			UVD_CGC_CTRL__JPEG2_MODE_MASK |
1538 			UVD_CGC_CTRL__SCPU_MODE_MASK);
1539 	data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
1540 			UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
1541 			UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
1542 			UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
1543 			UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
1544 	data1 |= suvd_flags;
1545 
1546 	WREG32_SOC15(UVD, 0, mmUVD_CGC_CTRL, data);
1547 	WREG32_SOC15(UVD, 0, mmUVD_CGC_GATE, 0);
1548 	WREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_GATE, data1);
1549 	WREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_CTRL, data2);
1550 }
1551 
1552 static void uvd_v7_0_set_hw_clock_gating(struct amdgpu_device *adev)
1553 {
1554 	uint32_t data, data1, cgc_flags, suvd_flags;
1555 
1556 	data = RREG32_SOC15(UVD, 0, mmUVD_CGC_GATE);
1557 	data1 = RREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_GATE);
1558 
1559 	cgc_flags = UVD_CGC_GATE__SYS_MASK |
1560 		UVD_CGC_GATE__UDEC_MASK |
1561 		UVD_CGC_GATE__MPEG2_MASK |
1562 		UVD_CGC_GATE__RBC_MASK |
1563 		UVD_CGC_GATE__LMI_MC_MASK |
1564 		UVD_CGC_GATE__IDCT_MASK |
1565 		UVD_CGC_GATE__MPRD_MASK |
1566 		UVD_CGC_GATE__MPC_MASK |
1567 		UVD_CGC_GATE__LBSI_MASK |
1568 		UVD_CGC_GATE__LRBBM_MASK |
1569 		UVD_CGC_GATE__UDEC_RE_MASK |
1570 		UVD_CGC_GATE__UDEC_CM_MASK |
1571 		UVD_CGC_GATE__UDEC_IT_MASK |
1572 		UVD_CGC_GATE__UDEC_DB_MASK |
1573 		UVD_CGC_GATE__UDEC_MP_MASK |
1574 		UVD_CGC_GATE__WCB_MASK |
1575 		UVD_CGC_GATE__VCPU_MASK |
1576 		UVD_CGC_GATE__SCPU_MASK |
1577 		UVD_CGC_GATE__JPEG_MASK |
1578 		UVD_CGC_GATE__JPEG2_MASK;
1579 
1580 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
1581 				UVD_SUVD_CGC_GATE__SIT_MASK |
1582 				UVD_SUVD_CGC_GATE__SMP_MASK |
1583 				UVD_SUVD_CGC_GATE__SCM_MASK |
1584 				UVD_SUVD_CGC_GATE__SDB_MASK;
1585 
1586 	data |= cgc_flags;
1587 	data1 |= suvd_flags;
1588 
1589 	WREG32_SOC15(UVD, 0, mmUVD_CGC_GATE, data);
1590 	WREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_GATE, data1);
1591 }
1592 
1593 static void uvd_v7_0_set_bypass_mode(struct amdgpu_device *adev, bool enable)
1594 {
1595 	u32 tmp = RREG32_SMC(ixGCK_DFS_BYPASS_CNTL);
1596 
1597 	if (enable)
1598 		tmp |= (GCK_DFS_BYPASS_CNTL__BYPASSDCLK_MASK |
1599 			GCK_DFS_BYPASS_CNTL__BYPASSVCLK_MASK);
1600 	else
1601 		tmp &= ~(GCK_DFS_BYPASS_CNTL__BYPASSDCLK_MASK |
1602 			 GCK_DFS_BYPASS_CNTL__BYPASSVCLK_MASK);
1603 
1604 	WREG32_SMC(ixGCK_DFS_BYPASS_CNTL, tmp);
1605 }
1606 
1607 
1608 static int uvd_v7_0_set_clockgating_state(void *handle,
1609 					  enum amd_clockgating_state state)
1610 {
1611 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1612 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1613 
1614 	uvd_v7_0_set_bypass_mode(adev, enable);
1615 
1616 	if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG))
1617 		return 0;
1618 
1619 	if (enable) {
1620 		/* disable HW gating and enable Sw gating */
1621 		uvd_v7_0_set_sw_clock_gating(adev);
1622 	} else {
1623 		/* wait for STATUS to clear */
1624 		if (uvd_v7_0_wait_for_idle(handle))
1625 			return -EBUSY;
1626 
1627 		/* enable HW gates because UVD is idle */
1628 		/* uvd_v7_0_set_hw_clock_gating(adev); */
1629 	}
1630 
1631 	return 0;
1632 }
1633 
1634 static int uvd_v7_0_set_powergating_state(void *handle,
1635 					  enum amd_powergating_state state)
1636 {
1637 	/* This doesn't actually powergate the UVD block.
1638 	 * That's done in the dpm code via the SMC.  This
1639 	 * just re-inits the block as necessary.  The actual
1640 	 * gating still happens in the dpm code.  We should
1641 	 * revisit this when there is a cleaner line between
1642 	 * the smc and the hw blocks
1643 	 */
1644 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1645 
1646 	if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD))
1647 		return 0;
1648 
1649 	WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, UVD_POWER_STATUS__UVD_PG_EN_MASK);
1650 
1651 	if (state == AMD_PG_STATE_GATE) {
1652 		uvd_v7_0_stop(adev);
1653 		return 0;
1654 	} else {
1655 		return uvd_v7_0_start(adev);
1656 	}
1657 }
1658 #endif
1659 
1660 static int uvd_v7_0_set_clockgating_state(void *handle,
1661 					  enum amd_clockgating_state state)
1662 {
1663 	/* needed for driver unload*/
1664 	return 0;
1665 }
1666 
1667 const struct amd_ip_funcs uvd_v7_0_ip_funcs = {
1668 	.name = "uvd_v7_0",
1669 	.early_init = uvd_v7_0_early_init,
1670 	.late_init = NULL,
1671 	.sw_init = uvd_v7_0_sw_init,
1672 	.sw_fini = uvd_v7_0_sw_fini,
1673 	.hw_init = uvd_v7_0_hw_init,
1674 	.hw_fini = uvd_v7_0_hw_fini,
1675 	.suspend = uvd_v7_0_suspend,
1676 	.resume = uvd_v7_0_resume,
1677 	.is_idle = NULL /* uvd_v7_0_is_idle */,
1678 	.wait_for_idle = NULL /* uvd_v7_0_wait_for_idle */,
1679 	.check_soft_reset = NULL /* uvd_v7_0_check_soft_reset */,
1680 	.pre_soft_reset = NULL /* uvd_v7_0_pre_soft_reset */,
1681 	.soft_reset = NULL /* uvd_v7_0_soft_reset */,
1682 	.post_soft_reset = NULL /* uvd_v7_0_post_soft_reset */,
1683 	.set_clockgating_state = uvd_v7_0_set_clockgating_state,
1684 	.set_powergating_state = NULL /* uvd_v7_0_set_powergating_state */,
1685 };
1686 
1687 static const struct amdgpu_ring_funcs uvd_v7_0_ring_vm_funcs = {
1688 	.type = AMDGPU_RING_TYPE_UVD,
1689 	.align_mask = 0xf,
1690 	.nop = PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP), 0),
1691 	.support_64bit_ptrs = false,
1692 	.vmhub = AMDGPU_MMHUB,
1693 	.get_rptr = uvd_v7_0_ring_get_rptr,
1694 	.get_wptr = uvd_v7_0_ring_get_wptr,
1695 	.set_wptr = uvd_v7_0_ring_set_wptr,
1696 	.emit_frame_size =
1697 		2 + /* uvd_v7_0_ring_emit_hdp_flush */
1698 		2 + /* uvd_v7_0_ring_emit_hdp_invalidate */
1699 		34 + /* uvd_v7_0_ring_emit_vm_flush */
1700 		14 + 14, /* uvd_v7_0_ring_emit_fence x2 vm fence */
1701 	.emit_ib_size = 8, /* uvd_v7_0_ring_emit_ib */
1702 	.emit_ib = uvd_v7_0_ring_emit_ib,
1703 	.emit_fence = uvd_v7_0_ring_emit_fence,
1704 	.emit_vm_flush = uvd_v7_0_ring_emit_vm_flush,
1705 	.emit_hdp_flush = uvd_v7_0_ring_emit_hdp_flush,
1706 	.emit_hdp_invalidate = uvd_v7_0_ring_emit_hdp_invalidate,
1707 	.test_ring = uvd_v7_0_ring_test_ring,
1708 	.test_ib = amdgpu_uvd_ring_test_ib,
1709 	.insert_nop = amdgpu_ring_insert_nop,
1710 	.pad_ib = amdgpu_ring_generic_pad_ib,
1711 	.begin_use = amdgpu_uvd_ring_begin_use,
1712 	.end_use = amdgpu_uvd_ring_end_use,
1713 };
1714 
1715 static const struct amdgpu_ring_funcs uvd_v7_0_enc_ring_vm_funcs = {
1716 	.type = AMDGPU_RING_TYPE_UVD_ENC,
1717 	.align_mask = 0x3f,
1718 	.nop = HEVC_ENC_CMD_NO_OP,
1719 	.support_64bit_ptrs = false,
1720 	.vmhub = AMDGPU_MMHUB,
1721 	.get_rptr = uvd_v7_0_enc_ring_get_rptr,
1722 	.get_wptr = uvd_v7_0_enc_ring_get_wptr,
1723 	.set_wptr = uvd_v7_0_enc_ring_set_wptr,
1724 	.emit_frame_size =
1725 		17 + /* uvd_v7_0_enc_ring_emit_vm_flush */
1726 		5 + 5 + /* uvd_v7_0_enc_ring_emit_fence x2 vm fence */
1727 		1, /* uvd_v7_0_enc_ring_insert_end */
1728 	.emit_ib_size = 5, /* uvd_v7_0_enc_ring_emit_ib */
1729 	.emit_ib = uvd_v7_0_enc_ring_emit_ib,
1730 	.emit_fence = uvd_v7_0_enc_ring_emit_fence,
1731 	.emit_vm_flush = uvd_v7_0_enc_ring_emit_vm_flush,
1732 	.test_ring = uvd_v7_0_enc_ring_test_ring,
1733 	.test_ib = uvd_v7_0_enc_ring_test_ib,
1734 	.insert_nop = amdgpu_ring_insert_nop,
1735 	.insert_end = uvd_v7_0_enc_ring_insert_end,
1736 	.pad_ib = amdgpu_ring_generic_pad_ib,
1737 	.begin_use = amdgpu_uvd_ring_begin_use,
1738 	.end_use = amdgpu_uvd_ring_end_use,
1739 };
1740 
1741 static void uvd_v7_0_set_ring_funcs(struct amdgpu_device *adev)
1742 {
1743 	adev->uvd.ring.funcs = &uvd_v7_0_ring_vm_funcs;
1744 	DRM_INFO("UVD is enabled in VM mode\n");
1745 }
1746 
1747 static void uvd_v7_0_set_enc_ring_funcs(struct amdgpu_device *adev)
1748 {
1749 	int i;
1750 
1751 	for (i = 0; i < adev->uvd.num_enc_rings; ++i)
1752 		adev->uvd.ring_enc[i].funcs = &uvd_v7_0_enc_ring_vm_funcs;
1753 
1754 	DRM_INFO("UVD ENC is enabled in VM mode\n");
1755 }
1756 
1757 static const struct amdgpu_irq_src_funcs uvd_v7_0_irq_funcs = {
1758 	.set = uvd_v7_0_set_interrupt_state,
1759 	.process = uvd_v7_0_process_interrupt,
1760 };
1761 
1762 static void uvd_v7_0_set_irq_funcs(struct amdgpu_device *adev)
1763 {
1764 	adev->uvd.irq.num_types = adev->uvd.num_enc_rings + 1;
1765 	adev->uvd.irq.funcs = &uvd_v7_0_irq_funcs;
1766 }
1767 
1768 const struct amdgpu_ip_block_version uvd_v7_0_ip_block =
1769 {
1770 		.type = AMD_IP_BLOCK_TYPE_UVD,
1771 		.major = 7,
1772 		.minor = 0,
1773 		.rev = 0,
1774 		.funcs = &uvd_v7_0_ip_funcs,
1775 };
1776