xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/uvd_v7_0.c (revision 2359ccdd)
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 "soc15.h"
29 #include "soc15d.h"
30 #include "soc15_common.h"
31 #include "mmsch_v1_0.h"
32 
33 #include "uvd/uvd_7_0_offset.h"
34 #include "uvd/uvd_7_0_sh_mask.h"
35 #include "vce/vce_4_0_offset.h"
36 #include "vce/vce_4_0_default.h"
37 #include "vce/vce_4_0_sh_mask.h"
38 #include "nbif/nbif_6_1_offset.h"
39 #include "hdp/hdp_4_0_offset.h"
40 #include "mmhub/mmhub_1_0_offset.h"
41 #include "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_DEBUG("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_DEBUG("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 drm_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, SOC15_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, SOC15_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[DRM_SCHED_PRIORITY_NORMAL];
420 	r = drm_sched_entity_init(&ring->sched, &adev->uvd.entity_enc,
421 				  rq, amdgpu_sched_jobs, NULL);
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 	drm_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 	return amdgpu_uvd_suspend(adev);
596 }
597 
598 static int uvd_v7_0_resume(void *handle)
599 {
600 	int r;
601 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
602 
603 	r = amdgpu_uvd_resume(adev);
604 	if (r)
605 		return r;
606 
607 	return uvd_v7_0_hw_init(adev);
608 }
609 
610 /**
611  * uvd_v7_0_mc_resume - memory controller programming
612  *
613  * @adev: amdgpu_device pointer
614  *
615  * Let the UVD memory controller know it's offsets
616  */
617 static void uvd_v7_0_mc_resume(struct amdgpu_device *adev)
618 {
619 	uint32_t size = AMDGPU_UVD_FIRMWARE_SIZE(adev);
620 	uint32_t offset;
621 
622 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
623 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
624 			lower_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
625 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
626 			upper_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
627 		offset = 0;
628 	} else {
629 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
630 			lower_32_bits(adev->uvd.gpu_addr));
631 		WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
632 			upper_32_bits(adev->uvd.gpu_addr));
633 		offset = size;
634 	}
635 
636 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0,
637 				AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
638 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size);
639 
640 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
641 			lower_32_bits(adev->uvd.gpu_addr + offset));
642 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
643 			upper_32_bits(adev->uvd.gpu_addr + offset));
644 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, (1 << 21));
645 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_UVD_HEAP_SIZE);
646 
647 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
648 			lower_32_bits(adev->uvd.gpu_addr + offset + AMDGPU_UVD_HEAP_SIZE));
649 	WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
650 			upper_32_bits(adev->uvd.gpu_addr + offset + AMDGPU_UVD_HEAP_SIZE));
651 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, (2 << 21));
652 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2,
653 			AMDGPU_UVD_STACK_SIZE + (AMDGPU_UVD_SESSION_SIZE * 40));
654 
655 	WREG32_SOC15(UVD, 0, mmUVD_UDEC_ADDR_CONFIG,
656 			adev->gfx.config.gb_addr_config);
657 	WREG32_SOC15(UVD, 0, mmUVD_UDEC_DB_ADDR_CONFIG,
658 			adev->gfx.config.gb_addr_config);
659 	WREG32_SOC15(UVD, 0, mmUVD_UDEC_DBW_ADDR_CONFIG,
660 			adev->gfx.config.gb_addr_config);
661 
662 	WREG32_SOC15(UVD, 0, mmUVD_GP_SCRATCH4, adev->uvd.max_handles);
663 }
664 
665 static int uvd_v7_0_mmsch_start(struct amdgpu_device *adev,
666 				struct amdgpu_mm_table *table)
667 {
668 	uint32_t data = 0, loop;
669 	uint64_t addr = table->gpu_addr;
670 	struct mmsch_v1_0_init_header *header = (struct mmsch_v1_0_init_header *)table->cpu_addr;
671 	uint32_t size;
672 
673 	size = header->header_size + header->vce_table_size + header->uvd_table_size;
674 
675 	/* 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of memory descriptor location */
676 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
677 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
678 
679 	/* 2, update vmid of descriptor */
680 	data = RREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_VMID);
681 	data &= ~VCE_MMSCH_VF_VMID__VF_CTX_VMID_MASK;
682 	data |= (0 << VCE_MMSCH_VF_VMID__VF_CTX_VMID__SHIFT); /* use domain0 for MM scheduler */
683 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_VMID, data);
684 
685 	/* 3, notify mmsch about the size of this descriptor */
686 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_CTX_SIZE, size);
687 
688 	/* 4, set resp to zero */
689 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_RESP, 0);
690 
691 	WDOORBELL32(adev->uvd.ring_enc[0].doorbell_index, 0);
692 	adev->wb.wb[adev->uvd.ring_enc[0].wptr_offs] = 0;
693 	adev->uvd.ring_enc[0].wptr = 0;
694 	adev->uvd.ring_enc[0].wptr_old = 0;
695 
696 	/* 5, kick off the initialization and wait until VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero */
697 	WREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_HOST, 0x10000001);
698 
699 	data = RREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_RESP);
700 	loop = 1000;
701 	while ((data & 0x10000002) != 0x10000002) {
702 		udelay(10);
703 		data = RREG32_SOC15(VCE, 0, mmVCE_MMSCH_VF_MAILBOX_RESP);
704 		loop--;
705 		if (!loop)
706 			break;
707 	}
708 
709 	if (!loop) {
710 		dev_err(adev->dev, "failed to init MMSCH, mmVCE_MMSCH_VF_MAILBOX_RESP = %x\n", data);
711 		return -EBUSY;
712 	}
713 
714 	return 0;
715 }
716 
717 static int uvd_v7_0_sriov_start(struct amdgpu_device *adev)
718 {
719 	struct amdgpu_ring *ring;
720 	uint32_t offset, size, tmp;
721 	uint32_t table_size = 0;
722 	struct mmsch_v1_0_cmd_direct_write direct_wt = { {0} };
723 	struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { {0} };
724 	struct mmsch_v1_0_cmd_direct_polling direct_poll = { {0} };
725 	struct mmsch_v1_0_cmd_end end = { {0} };
726 	uint32_t *init_table = adev->virt.mm_table.cpu_addr;
727 	struct mmsch_v1_0_init_header *header = (struct mmsch_v1_0_init_header *)init_table;
728 
729 	direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
730 	direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
731 	direct_poll.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_POLLING;
732 	end.cmd_header.command_type = MMSCH_COMMAND__END;
733 
734 	if (header->uvd_table_offset == 0 && header->uvd_table_size == 0) {
735 		header->version = MMSCH_VERSION;
736 		header->header_size = sizeof(struct mmsch_v1_0_init_header) >> 2;
737 
738 		if (header->vce_table_offset == 0 && header->vce_table_size == 0)
739 			header->uvd_table_offset = header->header_size;
740 		else
741 			header->uvd_table_offset = header->vce_table_size + header->vce_table_offset;
742 
743 		init_table += header->uvd_table_offset;
744 
745 		ring = &adev->uvd.ring;
746 		ring->wptr = 0;
747 		size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4);
748 
749 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS),
750 						   0xFFFFFFFF, 0x00000004);
751 		/* mc resume*/
752 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
753 			MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
754 						    lower_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
755 			MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
756 						    upper_32_bits(adev->firmware.ucode[AMDGPU_UCODE_ID_UVD].mc_addr));
757 			offset = 0;
758 		} else {
759 			MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
760 						    lower_32_bits(adev->uvd.gpu_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->uvd.gpu_addr));
763 			offset = size;
764 		}
765 
766 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0),
767 					    AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
768 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size);
769 
770 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
771 					    lower_32_bits(adev->uvd.gpu_addr + offset));
772 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
773 					    upper_32_bits(adev->uvd.gpu_addr + offset));
774 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), (1 << 21));
775 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_UVD_HEAP_SIZE);
776 
777 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
778 					    lower_32_bits(adev->uvd.gpu_addr + offset + AMDGPU_UVD_HEAP_SIZE));
779 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
780 					    upper_32_bits(adev->uvd.gpu_addr + offset + AMDGPU_UVD_HEAP_SIZE));
781 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), (2 << 21));
782 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CACHE_SIZE2),
783 					    AMDGPU_UVD_STACK_SIZE + (AMDGPU_UVD_SESSION_SIZE * 40));
784 
785 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_GP_SCRATCH4), adev->uvd.max_handles);
786 		/* mc resume end*/
787 
788 		/* disable clock gating */
789 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_CGC_CTRL),
790 						   ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK, 0);
791 
792 		/* disable interupt */
793 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
794 						   ~UVD_MASTINT_EN__VCPU_EN_MASK, 0);
795 
796 		/* stall UMC and register bus before resetting VCPU */
797 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2),
798 						   ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK,
799 						   UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
800 
801 		/* put LMI, VCPU, RBC etc... into reset */
802 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
803 					    (uint32_t)(UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
804 						       UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
805 						       UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
806 						       UVD_SOFT_RESET__RBC_SOFT_RESET_MASK |
807 						       UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
808 						       UVD_SOFT_RESET__CXW_SOFT_RESET_MASK |
809 						       UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
810 						       UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK));
811 
812 		/* initialize UVD memory controller */
813 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL),
814 					    (uint32_t)((0x40 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
815 						       UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
816 						       UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
817 						       UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
818 						       UVD_LMI_CTRL__REQ_MODE_MASK |
819 						       0x00100000L));
820 
821 		/* take all subblocks out of reset, except VCPU */
822 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
823 					    UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
824 
825 		/* enable VCPU clock */
826 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL),
827 					    UVD_VCPU_CNTL__CLK_EN_MASK);
828 
829 		/* enable master interrupt */
830 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
831 						   ~(UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK),
832 						   (UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK));
833 
834 		/* clear the bit 4 of UVD_STATUS */
835 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS),
836 						   ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT), 0);
837 
838 		/* force RBC into idle state */
839 		size = order_base_2(ring->ring_size);
840 		tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, size);
841 		tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
842 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), tmp);
843 
844 		ring = &adev->uvd.ring_enc[0];
845 		ring->wptr = 0;
846 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_BASE_LO), ring->gpu_addr);
847 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_BASE_HI), upper_32_bits(ring->gpu_addr));
848 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_SIZE), ring->ring_size / 4);
849 
850 		/* boot up the VCPU */
851 		MMSCH_V1_0_INSERT_DIRECT_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0);
852 
853 		/* enable UMC */
854 		MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2),
855 										   ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK, 0);
856 
857 		MMSCH_V1_0_INSERT_DIRECT_POLL(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0x02, 0x02);
858 
859 		/* add end packet */
860 		memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
861 		table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
862 		header->uvd_table_size = table_size;
863 
864 	}
865 	return uvd_v7_0_mmsch_start(adev, &adev->virt.mm_table);
866 }
867 
868 /**
869  * uvd_v7_0_start - start UVD block
870  *
871  * @adev: amdgpu_device pointer
872  *
873  * Setup and start the UVD block
874  */
875 static int uvd_v7_0_start(struct amdgpu_device *adev)
876 {
877 	struct amdgpu_ring *ring = &adev->uvd.ring;
878 	uint32_t rb_bufsz, tmp;
879 	uint32_t lmi_swap_cntl;
880 	uint32_t mp_swap_cntl;
881 	int i, j, r;
882 
883 	/* disable DPG */
884 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0,
885 			~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
886 
887 	/* disable byte swapping */
888 	lmi_swap_cntl = 0;
889 	mp_swap_cntl = 0;
890 
891 	uvd_v7_0_mc_resume(adev);
892 
893 	/* disable clock gating */
894 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_CGC_CTRL), 0,
895 			~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK);
896 
897 	/* disable interupt */
898 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0,
899 			~UVD_MASTINT_EN__VCPU_EN_MASK);
900 
901 	/* stall UMC and register bus before resetting VCPU */
902 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2),
903 			UVD_LMI_CTRL2__STALL_ARB_UMC_MASK,
904 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
905 	mdelay(1);
906 
907 	/* put LMI, VCPU, RBC etc... into reset */
908 	WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET,
909 		UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
910 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
911 		UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
912 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK |
913 		UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
914 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK |
915 		UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
916 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
917 	mdelay(5);
918 
919 	/* initialize UVD memory controller */
920 	WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL,
921 		(0x40 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
922 		UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
923 		UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
924 		UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
925 		UVD_LMI_CTRL__REQ_MODE_MASK |
926 		0x00100000L);
927 
928 #ifdef __BIG_ENDIAN
929 	/* swap (8 in 32) RB and IB */
930 	lmi_swap_cntl = 0xa;
931 	mp_swap_cntl = 0;
932 #endif
933 	WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
934 	WREG32_SOC15(UVD, 0, mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
935 
936 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA0, 0x40c2040);
937 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXA1, 0x0);
938 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB0, 0x40c2040);
939 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUXB1, 0x0);
940 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_ALU, 0);
941 	WREG32_SOC15(UVD, 0, mmUVD_MPC_SET_MUX, 0x88);
942 
943 	/* take all subblocks out of reset, except VCPU */
944 	WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET,
945 			UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
946 	mdelay(5);
947 
948 	/* enable VCPU clock */
949 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL,
950 			UVD_VCPU_CNTL__CLK_EN_MASK);
951 
952 	/* enable UMC */
953 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0,
954 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
955 
956 	/* boot up the VCPU */
957 	WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET, 0);
958 	mdelay(10);
959 
960 	for (i = 0; i < 10; ++i) {
961 		uint32_t status;
962 
963 		for (j = 0; j < 100; ++j) {
964 			status = RREG32_SOC15(UVD, 0, mmUVD_STATUS);
965 			if (status & 2)
966 				break;
967 			mdelay(10);
968 		}
969 		r = 0;
970 		if (status & 2)
971 			break;
972 
973 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
974 		WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
975 				UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
976 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
977 		mdelay(10);
978 		WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0,
979 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
980 		mdelay(10);
981 		r = -1;
982 	}
983 
984 	if (r) {
985 		DRM_ERROR("UVD not responding, giving up!!!\n");
986 		return r;
987 	}
988 	/* enable master interrupt */
989 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
990 		(UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK),
991 		~(UVD_MASTINT_EN__VCPU_EN_MASK|UVD_MASTINT_EN__SYS_EN_MASK));
992 
993 	/* clear the bit 4 of UVD_STATUS */
994 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0,
995 			~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
996 
997 	/* force RBC into idle state */
998 	rb_bufsz = order_base_2(ring->ring_size);
999 	tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1000 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1001 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1002 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0);
1003 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1004 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1005 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp);
1006 
1007 	/* set the write pointer delay */
1008 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0);
1009 
1010 	/* set the wb address */
1011 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR,
1012 			(upper_32_bits(ring->gpu_addr) >> 2));
1013 
1014 	/* programm the RB_BASE for ring buffer */
1015 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1016 			lower_32_bits(ring->gpu_addr));
1017 	WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1018 			upper_32_bits(ring->gpu_addr));
1019 
1020 	/* Initialize the ring buffer's read and write pointers */
1021 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0);
1022 
1023 	ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1024 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1025 			lower_32_bits(ring->wptr));
1026 
1027 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_RB_CNTL), 0,
1028 			~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK);
1029 
1030 	ring = &adev->uvd.ring_enc[0];
1031 	WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1032 	WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1033 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1034 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1035 	WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1036 
1037 	ring = &adev->uvd.ring_enc[1];
1038 	WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1039 	WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1040 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1041 	WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1042 	WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1043 
1044 	return 0;
1045 }
1046 
1047 /**
1048  * uvd_v7_0_stop - stop UVD block
1049  *
1050  * @adev: amdgpu_device pointer
1051  *
1052  * stop the UVD block
1053  */
1054 static void uvd_v7_0_stop(struct amdgpu_device *adev)
1055 {
1056 	/* force RBC into idle state */
1057 	WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, 0x11010101);
1058 
1059 	/* Stall UMC and register bus before resetting VCPU */
1060 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2),
1061 			UVD_LMI_CTRL2__STALL_ARB_UMC_MASK,
1062 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1063 	mdelay(1);
1064 
1065 	/* put VCPU into reset */
1066 	WREG32_SOC15(UVD, 0, mmUVD_SOFT_RESET,
1067 			UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1068 	mdelay(5);
1069 
1070 	/* disable VCPU clock */
1071 	WREG32_SOC15(UVD, 0, mmUVD_VCPU_CNTL, 0x0);
1072 
1073 	/* Unstall UMC and register bus */
1074 	WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0,
1075 			~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1076 }
1077 
1078 /**
1079  * uvd_v7_0_ring_emit_fence - emit an fence & trap command
1080  *
1081  * @ring: amdgpu_ring pointer
1082  * @fence: fence to emit
1083  *
1084  * Write a fence and a trap command to the ring.
1085  */
1086 static void uvd_v7_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
1087 				     unsigned flags)
1088 {
1089 	struct amdgpu_device *adev = ring->adev;
1090 
1091 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1092 
1093 	amdgpu_ring_write(ring,
1094 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0));
1095 	amdgpu_ring_write(ring, seq);
1096 	amdgpu_ring_write(ring,
1097 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
1098 	amdgpu_ring_write(ring, addr & 0xffffffff);
1099 	amdgpu_ring_write(ring,
1100 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
1101 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
1102 	amdgpu_ring_write(ring,
1103 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
1104 	amdgpu_ring_write(ring, 0);
1105 
1106 	amdgpu_ring_write(ring,
1107 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
1108 	amdgpu_ring_write(ring, 0);
1109 	amdgpu_ring_write(ring,
1110 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
1111 	amdgpu_ring_write(ring, 0);
1112 	amdgpu_ring_write(ring,
1113 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
1114 	amdgpu_ring_write(ring, 2);
1115 }
1116 
1117 /**
1118  * uvd_v7_0_enc_ring_emit_fence - emit an enc fence & trap command
1119  *
1120  * @ring: amdgpu_ring pointer
1121  * @fence: fence to emit
1122  *
1123  * Write enc a fence and a trap command to the ring.
1124  */
1125 static void uvd_v7_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1126 			u64 seq, unsigned flags)
1127 {
1128 
1129 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1130 
1131 	amdgpu_ring_write(ring, HEVC_ENC_CMD_FENCE);
1132 	amdgpu_ring_write(ring, addr);
1133 	amdgpu_ring_write(ring, upper_32_bits(addr));
1134 	amdgpu_ring_write(ring, seq);
1135 	amdgpu_ring_write(ring, HEVC_ENC_CMD_TRAP);
1136 }
1137 
1138 /**
1139  * uvd_v7_0_ring_test_ring - register write test
1140  *
1141  * @ring: amdgpu_ring pointer
1142  *
1143  * Test if we can successfully write to the context register
1144  */
1145 static int uvd_v7_0_ring_test_ring(struct amdgpu_ring *ring)
1146 {
1147 	struct amdgpu_device *adev = ring->adev;
1148 	uint32_t tmp = 0;
1149 	unsigned i;
1150 	int r;
1151 
1152 	WREG32_SOC15(UVD, 0, mmUVD_CONTEXT_ID, 0xCAFEDEAD);
1153 	r = amdgpu_ring_alloc(ring, 3);
1154 	if (r) {
1155 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
1156 			  ring->idx, r);
1157 		return r;
1158 	}
1159 	amdgpu_ring_write(ring,
1160 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_CONTEXT_ID), 0));
1161 	amdgpu_ring_write(ring, 0xDEADBEEF);
1162 	amdgpu_ring_commit(ring);
1163 	for (i = 0; i < adev->usec_timeout; i++) {
1164 		tmp = RREG32_SOC15(UVD, 0, mmUVD_CONTEXT_ID);
1165 		if (tmp == 0xDEADBEEF)
1166 			break;
1167 		DRM_UDELAY(1);
1168 	}
1169 
1170 	if (i < adev->usec_timeout) {
1171 		DRM_DEBUG("ring test on %d succeeded in %d usecs\n",
1172 			 ring->idx, i);
1173 	} else {
1174 		DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
1175 			  ring->idx, tmp);
1176 		r = -EINVAL;
1177 	}
1178 	return r;
1179 }
1180 
1181 /**
1182  * uvd_v7_0_ring_emit_ib - execute indirect buffer
1183  *
1184  * @ring: amdgpu_ring pointer
1185  * @ib: indirect buffer to execute
1186  *
1187  * Write ring commands to execute the indirect buffer
1188  */
1189 static void uvd_v7_0_ring_emit_ib(struct amdgpu_ring *ring,
1190 				  struct amdgpu_ib *ib,
1191 				  unsigned vmid, bool ctx_switch)
1192 {
1193 	struct amdgpu_device *adev = ring->adev;
1194 
1195 	amdgpu_ring_write(ring,
1196 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_VMID), 0));
1197 	amdgpu_ring_write(ring, vmid);
1198 
1199 	amdgpu_ring_write(ring,
1200 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_LOW), 0));
1201 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1202 	amdgpu_ring_write(ring,
1203 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH), 0));
1204 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1205 	amdgpu_ring_write(ring,
1206 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_RBC_IB_SIZE), 0));
1207 	amdgpu_ring_write(ring, ib->length_dw);
1208 }
1209 
1210 /**
1211  * uvd_v7_0_enc_ring_emit_ib - enc execute indirect buffer
1212  *
1213  * @ring: amdgpu_ring pointer
1214  * @ib: indirect buffer to execute
1215  *
1216  * Write enc ring commands to execute the indirect buffer
1217  */
1218 static void uvd_v7_0_enc_ring_emit_ib(struct amdgpu_ring *ring,
1219 		struct amdgpu_ib *ib, unsigned int vmid, bool ctx_switch)
1220 {
1221 	amdgpu_ring_write(ring, HEVC_ENC_CMD_IB_VM);
1222 	amdgpu_ring_write(ring, vmid);
1223 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1224 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1225 	amdgpu_ring_write(ring, ib->length_dw);
1226 }
1227 
1228 static void uvd_v7_0_ring_emit_wreg(struct amdgpu_ring *ring,
1229 				    uint32_t reg, uint32_t val)
1230 {
1231 	struct amdgpu_device *adev = ring->adev;
1232 
1233 	amdgpu_ring_write(ring,
1234 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
1235 	amdgpu_ring_write(ring, reg << 2);
1236 	amdgpu_ring_write(ring,
1237 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
1238 	amdgpu_ring_write(ring, val);
1239 	amdgpu_ring_write(ring,
1240 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
1241 	amdgpu_ring_write(ring, 8);
1242 }
1243 
1244 static void uvd_v7_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1245 					uint32_t val, uint32_t mask)
1246 {
1247 	struct amdgpu_device *adev = ring->adev;
1248 
1249 	amdgpu_ring_write(ring,
1250 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0), 0));
1251 	amdgpu_ring_write(ring, reg << 2);
1252 	amdgpu_ring_write(ring,
1253 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1), 0));
1254 	amdgpu_ring_write(ring, val);
1255 	amdgpu_ring_write(ring,
1256 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GP_SCRATCH8), 0));
1257 	amdgpu_ring_write(ring, mask);
1258 	amdgpu_ring_write(ring,
1259 		PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD), 0));
1260 	amdgpu_ring_write(ring, 12);
1261 }
1262 
1263 static void uvd_v7_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
1264 					unsigned vmid, uint64_t pd_addr)
1265 {
1266 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1267 	uint32_t data0, data1, mask;
1268 
1269 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1270 
1271 	/* wait for reg writes */
1272 	data0 = hub->ctx0_ptb_addr_lo32 + vmid * 2;
1273 	data1 = lower_32_bits(pd_addr);
1274 	mask = 0xffffffff;
1275 	uvd_v7_0_ring_emit_reg_wait(ring, data0, data1, mask);
1276 }
1277 
1278 static void uvd_v7_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
1279 {
1280 	int i;
1281 	struct amdgpu_device *adev = ring->adev;
1282 
1283 	for (i = 0; i < count; i++)
1284 		amdgpu_ring_write(ring, PACKET0(SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP), 0));
1285 
1286 }
1287 
1288 static void uvd_v7_0_enc_ring_insert_end(struct amdgpu_ring *ring)
1289 {
1290 	amdgpu_ring_write(ring, HEVC_ENC_CMD_END);
1291 }
1292 
1293 static void uvd_v7_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring,
1294 					    uint32_t reg, uint32_t val,
1295 					    uint32_t mask)
1296 {
1297 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WAIT);
1298 	amdgpu_ring_write(ring,	reg << 2);
1299 	amdgpu_ring_write(ring, mask);
1300 	amdgpu_ring_write(ring, val);
1301 }
1302 
1303 static void uvd_v7_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
1304 					    unsigned int vmid, uint64_t pd_addr)
1305 {
1306 	struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1307 
1308 	pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1309 
1310 	/* wait for reg writes */
1311 	uvd_v7_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + vmid * 2,
1312 					lower_32_bits(pd_addr), 0xffffffff);
1313 }
1314 
1315 static void uvd_v7_0_enc_ring_emit_wreg(struct amdgpu_ring *ring,
1316 					uint32_t reg, uint32_t val)
1317 {
1318 	amdgpu_ring_write(ring, HEVC_ENC_CMD_REG_WRITE);
1319 	amdgpu_ring_write(ring,	reg << 2);
1320 	amdgpu_ring_write(ring, val);
1321 }
1322 
1323 #if 0
1324 static bool uvd_v7_0_is_idle(void *handle)
1325 {
1326 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1327 
1328 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
1329 }
1330 
1331 static int uvd_v7_0_wait_for_idle(void *handle)
1332 {
1333 	unsigned i;
1334 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1335 
1336 	for (i = 0; i < adev->usec_timeout; i++) {
1337 		if (uvd_v7_0_is_idle(handle))
1338 			return 0;
1339 	}
1340 	return -ETIMEDOUT;
1341 }
1342 
1343 #define AMDGPU_UVD_STATUS_BUSY_MASK    0xfd
1344 static bool uvd_v7_0_check_soft_reset(void *handle)
1345 {
1346 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1347 	u32 srbm_soft_reset = 0;
1348 	u32 tmp = RREG32(mmSRBM_STATUS);
1349 
1350 	if (REG_GET_FIELD(tmp, SRBM_STATUS, UVD_RQ_PENDING) ||
1351 	    REG_GET_FIELD(tmp, SRBM_STATUS, UVD_BUSY) ||
1352 	    (RREG32_SOC15(UVD, 0, mmUVD_STATUS) &
1353 		    AMDGPU_UVD_STATUS_BUSY_MASK))
1354 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
1355 				SRBM_SOFT_RESET, SOFT_RESET_UVD, 1);
1356 
1357 	if (srbm_soft_reset) {
1358 		adev->uvd.srbm_soft_reset = srbm_soft_reset;
1359 		return true;
1360 	} else {
1361 		adev->uvd.srbm_soft_reset = 0;
1362 		return false;
1363 	}
1364 }
1365 
1366 static int uvd_v7_0_pre_soft_reset(void *handle)
1367 {
1368 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1369 
1370 	if (!adev->uvd.srbm_soft_reset)
1371 		return 0;
1372 
1373 	uvd_v7_0_stop(adev);
1374 	return 0;
1375 }
1376 
1377 static int uvd_v7_0_soft_reset(void *handle)
1378 {
1379 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1380 	u32 srbm_soft_reset;
1381 
1382 	if (!adev->uvd.srbm_soft_reset)
1383 		return 0;
1384 	srbm_soft_reset = adev->uvd.srbm_soft_reset;
1385 
1386 	if (srbm_soft_reset) {
1387 		u32 tmp;
1388 
1389 		tmp = RREG32(mmSRBM_SOFT_RESET);
1390 		tmp |= srbm_soft_reset;
1391 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
1392 		WREG32(mmSRBM_SOFT_RESET, tmp);
1393 		tmp = RREG32(mmSRBM_SOFT_RESET);
1394 
1395 		udelay(50);
1396 
1397 		tmp &= ~srbm_soft_reset;
1398 		WREG32(mmSRBM_SOFT_RESET, tmp);
1399 		tmp = RREG32(mmSRBM_SOFT_RESET);
1400 
1401 		/* Wait a little for things to settle down */
1402 		udelay(50);
1403 	}
1404 
1405 	return 0;
1406 }
1407 
1408 static int uvd_v7_0_post_soft_reset(void *handle)
1409 {
1410 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1411 
1412 	if (!adev->uvd.srbm_soft_reset)
1413 		return 0;
1414 
1415 	mdelay(5);
1416 
1417 	return uvd_v7_0_start(adev);
1418 }
1419 #endif
1420 
1421 static int uvd_v7_0_set_interrupt_state(struct amdgpu_device *adev,
1422 					struct amdgpu_irq_src *source,
1423 					unsigned type,
1424 					enum amdgpu_interrupt_state state)
1425 {
1426 	// TODO
1427 	return 0;
1428 }
1429 
1430 static int uvd_v7_0_process_interrupt(struct amdgpu_device *adev,
1431 				      struct amdgpu_irq_src *source,
1432 				      struct amdgpu_iv_entry *entry)
1433 {
1434 	DRM_DEBUG("IH: UVD TRAP\n");
1435 	switch (entry->src_id) {
1436 	case 124:
1437 		amdgpu_fence_process(&adev->uvd.ring);
1438 		break;
1439 	case 119:
1440 		amdgpu_fence_process(&adev->uvd.ring_enc[0]);
1441 		break;
1442 	case 120:
1443 		if (!amdgpu_sriov_vf(adev))
1444 			amdgpu_fence_process(&adev->uvd.ring_enc[1]);
1445 		break;
1446 	default:
1447 		DRM_ERROR("Unhandled interrupt: %d %d\n",
1448 			  entry->src_id, entry->src_data[0]);
1449 		break;
1450 	}
1451 
1452 	return 0;
1453 }
1454 
1455 #if 0
1456 static void uvd_v7_0_set_sw_clock_gating(struct amdgpu_device *adev)
1457 {
1458 	uint32_t data, data1, data2, suvd_flags;
1459 
1460 	data = RREG32_SOC15(UVD, 0, mmUVD_CGC_CTRL);
1461 	data1 = RREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_GATE);
1462 	data2 = RREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_CTRL);
1463 
1464 	data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
1465 		  UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
1466 
1467 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
1468 		     UVD_SUVD_CGC_GATE__SIT_MASK |
1469 		     UVD_SUVD_CGC_GATE__SMP_MASK |
1470 		     UVD_SUVD_CGC_GATE__SCM_MASK |
1471 		     UVD_SUVD_CGC_GATE__SDB_MASK;
1472 
1473 	data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
1474 		(1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) |
1475 		(4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY));
1476 
1477 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
1478 			UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
1479 			UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
1480 			UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
1481 			UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
1482 			UVD_CGC_CTRL__SYS_MODE_MASK |
1483 			UVD_CGC_CTRL__UDEC_MODE_MASK |
1484 			UVD_CGC_CTRL__MPEG2_MODE_MASK |
1485 			UVD_CGC_CTRL__REGS_MODE_MASK |
1486 			UVD_CGC_CTRL__RBC_MODE_MASK |
1487 			UVD_CGC_CTRL__LMI_MC_MODE_MASK |
1488 			UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
1489 			UVD_CGC_CTRL__IDCT_MODE_MASK |
1490 			UVD_CGC_CTRL__MPRD_MODE_MASK |
1491 			UVD_CGC_CTRL__MPC_MODE_MASK |
1492 			UVD_CGC_CTRL__LBSI_MODE_MASK |
1493 			UVD_CGC_CTRL__LRBBM_MODE_MASK |
1494 			UVD_CGC_CTRL__WCB_MODE_MASK |
1495 			UVD_CGC_CTRL__VCPU_MODE_MASK |
1496 			UVD_CGC_CTRL__JPEG_MODE_MASK |
1497 			UVD_CGC_CTRL__JPEG2_MODE_MASK |
1498 			UVD_CGC_CTRL__SCPU_MODE_MASK);
1499 	data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
1500 			UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
1501 			UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
1502 			UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
1503 			UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
1504 	data1 |= suvd_flags;
1505 
1506 	WREG32_SOC15(UVD, 0, mmUVD_CGC_CTRL, data);
1507 	WREG32_SOC15(UVD, 0, mmUVD_CGC_GATE, 0);
1508 	WREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_GATE, data1);
1509 	WREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_CTRL, data2);
1510 }
1511 
1512 static void uvd_v7_0_set_hw_clock_gating(struct amdgpu_device *adev)
1513 {
1514 	uint32_t data, data1, cgc_flags, suvd_flags;
1515 
1516 	data = RREG32_SOC15(UVD, 0, mmUVD_CGC_GATE);
1517 	data1 = RREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_GATE);
1518 
1519 	cgc_flags = UVD_CGC_GATE__SYS_MASK |
1520 		UVD_CGC_GATE__UDEC_MASK |
1521 		UVD_CGC_GATE__MPEG2_MASK |
1522 		UVD_CGC_GATE__RBC_MASK |
1523 		UVD_CGC_GATE__LMI_MC_MASK |
1524 		UVD_CGC_GATE__IDCT_MASK |
1525 		UVD_CGC_GATE__MPRD_MASK |
1526 		UVD_CGC_GATE__MPC_MASK |
1527 		UVD_CGC_GATE__LBSI_MASK |
1528 		UVD_CGC_GATE__LRBBM_MASK |
1529 		UVD_CGC_GATE__UDEC_RE_MASK |
1530 		UVD_CGC_GATE__UDEC_CM_MASK |
1531 		UVD_CGC_GATE__UDEC_IT_MASK |
1532 		UVD_CGC_GATE__UDEC_DB_MASK |
1533 		UVD_CGC_GATE__UDEC_MP_MASK |
1534 		UVD_CGC_GATE__WCB_MASK |
1535 		UVD_CGC_GATE__VCPU_MASK |
1536 		UVD_CGC_GATE__SCPU_MASK |
1537 		UVD_CGC_GATE__JPEG_MASK |
1538 		UVD_CGC_GATE__JPEG2_MASK;
1539 
1540 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
1541 				UVD_SUVD_CGC_GATE__SIT_MASK |
1542 				UVD_SUVD_CGC_GATE__SMP_MASK |
1543 				UVD_SUVD_CGC_GATE__SCM_MASK |
1544 				UVD_SUVD_CGC_GATE__SDB_MASK;
1545 
1546 	data |= cgc_flags;
1547 	data1 |= suvd_flags;
1548 
1549 	WREG32_SOC15(UVD, 0, mmUVD_CGC_GATE, data);
1550 	WREG32_SOC15(UVD, 0, mmUVD_SUVD_CGC_GATE, data1);
1551 }
1552 
1553 static void uvd_v7_0_set_bypass_mode(struct amdgpu_device *adev, bool enable)
1554 {
1555 	u32 tmp = RREG32_SMC(ixGCK_DFS_BYPASS_CNTL);
1556 
1557 	if (enable)
1558 		tmp |= (GCK_DFS_BYPASS_CNTL__BYPASSDCLK_MASK |
1559 			GCK_DFS_BYPASS_CNTL__BYPASSVCLK_MASK);
1560 	else
1561 		tmp &= ~(GCK_DFS_BYPASS_CNTL__BYPASSDCLK_MASK |
1562 			 GCK_DFS_BYPASS_CNTL__BYPASSVCLK_MASK);
1563 
1564 	WREG32_SMC(ixGCK_DFS_BYPASS_CNTL, tmp);
1565 }
1566 
1567 
1568 static int uvd_v7_0_set_clockgating_state(void *handle,
1569 					  enum amd_clockgating_state state)
1570 {
1571 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1572 	bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1573 
1574 	uvd_v7_0_set_bypass_mode(adev, enable);
1575 
1576 	if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG))
1577 		return 0;
1578 
1579 	if (enable) {
1580 		/* disable HW gating and enable Sw gating */
1581 		uvd_v7_0_set_sw_clock_gating(adev);
1582 	} else {
1583 		/* wait for STATUS to clear */
1584 		if (uvd_v7_0_wait_for_idle(handle))
1585 			return -EBUSY;
1586 
1587 		/* enable HW gates because UVD is idle */
1588 		/* uvd_v7_0_set_hw_clock_gating(adev); */
1589 	}
1590 
1591 	return 0;
1592 }
1593 
1594 static int uvd_v7_0_set_powergating_state(void *handle,
1595 					  enum amd_powergating_state state)
1596 {
1597 	/* This doesn't actually powergate the UVD block.
1598 	 * That's done in the dpm code via the SMC.  This
1599 	 * just re-inits the block as necessary.  The actual
1600 	 * gating still happens in the dpm code.  We should
1601 	 * revisit this when there is a cleaner line between
1602 	 * the smc and the hw blocks
1603 	 */
1604 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1605 
1606 	if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD))
1607 		return 0;
1608 
1609 	WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, UVD_POWER_STATUS__UVD_PG_EN_MASK);
1610 
1611 	if (state == AMD_PG_STATE_GATE) {
1612 		uvd_v7_0_stop(adev);
1613 		return 0;
1614 	} else {
1615 		return uvd_v7_0_start(adev);
1616 	}
1617 }
1618 #endif
1619 
1620 static int uvd_v7_0_set_clockgating_state(void *handle,
1621 					  enum amd_clockgating_state state)
1622 {
1623 	/* needed for driver unload*/
1624 	return 0;
1625 }
1626 
1627 const struct amd_ip_funcs uvd_v7_0_ip_funcs = {
1628 	.name = "uvd_v7_0",
1629 	.early_init = uvd_v7_0_early_init,
1630 	.late_init = NULL,
1631 	.sw_init = uvd_v7_0_sw_init,
1632 	.sw_fini = uvd_v7_0_sw_fini,
1633 	.hw_init = uvd_v7_0_hw_init,
1634 	.hw_fini = uvd_v7_0_hw_fini,
1635 	.suspend = uvd_v7_0_suspend,
1636 	.resume = uvd_v7_0_resume,
1637 	.is_idle = NULL /* uvd_v7_0_is_idle */,
1638 	.wait_for_idle = NULL /* uvd_v7_0_wait_for_idle */,
1639 	.check_soft_reset = NULL /* uvd_v7_0_check_soft_reset */,
1640 	.pre_soft_reset = NULL /* uvd_v7_0_pre_soft_reset */,
1641 	.soft_reset = NULL /* uvd_v7_0_soft_reset */,
1642 	.post_soft_reset = NULL /* uvd_v7_0_post_soft_reset */,
1643 	.set_clockgating_state = uvd_v7_0_set_clockgating_state,
1644 	.set_powergating_state = NULL /* uvd_v7_0_set_powergating_state */,
1645 };
1646 
1647 static const struct amdgpu_ring_funcs uvd_v7_0_ring_vm_funcs = {
1648 	.type = AMDGPU_RING_TYPE_UVD,
1649 	.align_mask = 0xf,
1650 	.nop = PACKET0(0x81ff, 0),
1651 	.support_64bit_ptrs = false,
1652 	.vmhub = AMDGPU_MMHUB,
1653 	.get_rptr = uvd_v7_0_ring_get_rptr,
1654 	.get_wptr = uvd_v7_0_ring_get_wptr,
1655 	.set_wptr = uvd_v7_0_ring_set_wptr,
1656 	.emit_frame_size =
1657 		6 + 6 + /* hdp flush / invalidate */
1658 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1659 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1660 		8 + /* uvd_v7_0_ring_emit_vm_flush */
1661 		14 + 14, /* uvd_v7_0_ring_emit_fence x2 vm fence */
1662 	.emit_ib_size = 8, /* uvd_v7_0_ring_emit_ib */
1663 	.emit_ib = uvd_v7_0_ring_emit_ib,
1664 	.emit_fence = uvd_v7_0_ring_emit_fence,
1665 	.emit_vm_flush = uvd_v7_0_ring_emit_vm_flush,
1666 	.test_ring = uvd_v7_0_ring_test_ring,
1667 	.test_ib = amdgpu_uvd_ring_test_ib,
1668 	.insert_nop = uvd_v7_0_ring_insert_nop,
1669 	.pad_ib = amdgpu_ring_generic_pad_ib,
1670 	.begin_use = amdgpu_uvd_ring_begin_use,
1671 	.end_use = amdgpu_uvd_ring_end_use,
1672 	.emit_wreg = uvd_v7_0_ring_emit_wreg,
1673 	.emit_reg_wait = uvd_v7_0_ring_emit_reg_wait,
1674 };
1675 
1676 static const struct amdgpu_ring_funcs uvd_v7_0_enc_ring_vm_funcs = {
1677 	.type = AMDGPU_RING_TYPE_UVD_ENC,
1678 	.align_mask = 0x3f,
1679 	.nop = HEVC_ENC_CMD_NO_OP,
1680 	.support_64bit_ptrs = false,
1681 	.vmhub = AMDGPU_MMHUB,
1682 	.get_rptr = uvd_v7_0_enc_ring_get_rptr,
1683 	.get_wptr = uvd_v7_0_enc_ring_get_wptr,
1684 	.set_wptr = uvd_v7_0_enc_ring_set_wptr,
1685 	.emit_frame_size =
1686 		3 + 3 + /* hdp flush / invalidate */
1687 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1688 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1689 		4 + /* uvd_v7_0_enc_ring_emit_vm_flush */
1690 		5 + 5 + /* uvd_v7_0_enc_ring_emit_fence x2 vm fence */
1691 		1, /* uvd_v7_0_enc_ring_insert_end */
1692 	.emit_ib_size = 5, /* uvd_v7_0_enc_ring_emit_ib */
1693 	.emit_ib = uvd_v7_0_enc_ring_emit_ib,
1694 	.emit_fence = uvd_v7_0_enc_ring_emit_fence,
1695 	.emit_vm_flush = uvd_v7_0_enc_ring_emit_vm_flush,
1696 	.test_ring = uvd_v7_0_enc_ring_test_ring,
1697 	.test_ib = uvd_v7_0_enc_ring_test_ib,
1698 	.insert_nop = amdgpu_ring_insert_nop,
1699 	.insert_end = uvd_v7_0_enc_ring_insert_end,
1700 	.pad_ib = amdgpu_ring_generic_pad_ib,
1701 	.begin_use = amdgpu_uvd_ring_begin_use,
1702 	.end_use = amdgpu_uvd_ring_end_use,
1703 	.emit_wreg = uvd_v7_0_enc_ring_emit_wreg,
1704 	.emit_reg_wait = uvd_v7_0_enc_ring_emit_reg_wait,
1705 };
1706 
1707 static void uvd_v7_0_set_ring_funcs(struct amdgpu_device *adev)
1708 {
1709 	adev->uvd.ring.funcs = &uvd_v7_0_ring_vm_funcs;
1710 	DRM_INFO("UVD is enabled in VM mode\n");
1711 }
1712 
1713 static void uvd_v7_0_set_enc_ring_funcs(struct amdgpu_device *adev)
1714 {
1715 	int i;
1716 
1717 	for (i = 0; i < adev->uvd.num_enc_rings; ++i)
1718 		adev->uvd.ring_enc[i].funcs = &uvd_v7_0_enc_ring_vm_funcs;
1719 
1720 	DRM_INFO("UVD ENC is enabled in VM mode\n");
1721 }
1722 
1723 static const struct amdgpu_irq_src_funcs uvd_v7_0_irq_funcs = {
1724 	.set = uvd_v7_0_set_interrupt_state,
1725 	.process = uvd_v7_0_process_interrupt,
1726 };
1727 
1728 static void uvd_v7_0_set_irq_funcs(struct amdgpu_device *adev)
1729 {
1730 	adev->uvd.irq.num_types = adev->uvd.num_enc_rings + 1;
1731 	adev->uvd.irq.funcs = &uvd_v7_0_irq_funcs;
1732 }
1733 
1734 const struct amdgpu_ip_block_version uvd_v7_0_ip_block =
1735 {
1736 		.type = AMD_IP_BLOCK_TYPE_UVD,
1737 		.major = 7,
1738 		.minor = 0,
1739 		.rev = 0,
1740 		.funcs = &uvd_v7_0_ip_funcs,
1741 };
1742