xref: /linux/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
1 /*
2  * Copyright 2013 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  * Authors: Christian König <christian.koenig@amd.com>
23  */
24 
25 #include <linux/firmware.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_uvd.h"
29 #include "cikd.h"
30 
31 #include "uvd/uvd_4_2_d.h"
32 #include "uvd/uvd_4_2_sh_mask.h"
33 
34 #include "oss/oss_2_0_d.h"
35 #include "oss/oss_2_0_sh_mask.h"
36 
37 #include "bif/bif_4_1_d.h"
38 
39 #include "smu/smu_7_0_1_d.h"
40 #include "smu/smu_7_0_1_sh_mask.h"
41 
42 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev);
43 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev);
44 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev);
45 static int uvd_v4_2_start(struct amdgpu_device *adev);
46 static void uvd_v4_2_stop(struct amdgpu_device *adev);
47 static int uvd_v4_2_set_clockgating_state(void *handle,
48 				enum amd_clockgating_state state);
49 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
50 			     bool sw_mode);
51 /**
52  * uvd_v4_2_ring_get_rptr - get read pointer
53  *
54  * @ring: amdgpu_ring pointer
55  *
56  * Returns the current hardware read pointer
57  */
58 static uint64_t uvd_v4_2_ring_get_rptr(struct amdgpu_ring *ring)
59 {
60 	struct amdgpu_device *adev = ring->adev;
61 
62 	return RREG32(mmUVD_RBC_RB_RPTR);
63 }
64 
65 /**
66  * uvd_v4_2_ring_get_wptr - get write pointer
67  *
68  * @ring: amdgpu_ring pointer
69  *
70  * Returns the current hardware write pointer
71  */
72 static uint64_t uvd_v4_2_ring_get_wptr(struct amdgpu_ring *ring)
73 {
74 	struct amdgpu_device *adev = ring->adev;
75 
76 	return RREG32(mmUVD_RBC_RB_WPTR);
77 }
78 
79 /**
80  * uvd_v4_2_ring_set_wptr - set write pointer
81  *
82  * @ring: amdgpu_ring pointer
83  *
84  * Commits the write pointer to the hardware
85  */
86 static void uvd_v4_2_ring_set_wptr(struct amdgpu_ring *ring)
87 {
88 	struct amdgpu_device *adev = ring->adev;
89 
90 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
91 }
92 
93 static int uvd_v4_2_early_init(struct amdgpu_ip_block *ip_block)
94 {
95 	struct amdgpu_device *adev = ip_block->adev;
96 	adev->uvd.num_uvd_inst = 1;
97 
98 	uvd_v4_2_set_ring_funcs(adev);
99 	uvd_v4_2_set_irq_funcs(adev);
100 
101 	return 0;
102 }
103 
104 static int uvd_v4_2_sw_init(struct amdgpu_ip_block *ip_block)
105 {
106 	struct amdgpu_ring *ring;
107 	struct amdgpu_device *adev = ip_block->adev;
108 	int r;
109 
110 	/* UVD TRAP */
111 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 124, &adev->uvd.inst->irq);
112 	if (r)
113 		return r;
114 
115 	r = amdgpu_uvd_sw_init(adev);
116 	if (r)
117 		return r;
118 
119 	ring = &adev->uvd.inst->ring;
120 	sprintf(ring->name, "uvd");
121 	r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst->irq, 0,
122 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
123 	if (r)
124 		return r;
125 
126 	r = amdgpu_uvd_resume(adev);
127 	if (r)
128 		return r;
129 
130 	return r;
131 }
132 
133 static int uvd_v4_2_sw_fini(struct amdgpu_ip_block *ip_block)
134 {
135 	int r;
136 	struct amdgpu_device *adev = ip_block->adev;
137 
138 	r = amdgpu_uvd_suspend(adev);
139 	if (r)
140 		return r;
141 
142 	return amdgpu_uvd_sw_fini(adev);
143 }
144 
145 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
146 				 bool enable);
147 /**
148  * uvd_v4_2_hw_init - start and test UVD block
149  *
150  * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
151  *
152  * Initialize the hardware, boot up the VCPU and do some testing
153  */
154 static int uvd_v4_2_hw_init(struct amdgpu_ip_block *ip_block)
155 {
156 	struct amdgpu_device *adev = ip_block->adev;
157 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
158 	uint32_t tmp;
159 	int r;
160 
161 	uvd_v4_2_enable_mgcg(adev, true);
162 	amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
163 
164 	r = amdgpu_ring_test_helper(ring);
165 	if (r)
166 		goto done;
167 
168 	r = amdgpu_ring_alloc(ring, 10);
169 	if (r) {
170 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
171 		goto done;
172 	}
173 
174 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
175 	amdgpu_ring_write(ring, tmp);
176 	amdgpu_ring_write(ring, 0xFFFFF);
177 
178 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
179 	amdgpu_ring_write(ring, tmp);
180 	amdgpu_ring_write(ring, 0xFFFFF);
181 
182 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
183 	amdgpu_ring_write(ring, tmp);
184 	amdgpu_ring_write(ring, 0xFFFFF);
185 
186 	/* Clear timeout status bits */
187 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
188 	amdgpu_ring_write(ring, 0x8);
189 
190 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
191 	amdgpu_ring_write(ring, 3);
192 
193 	amdgpu_ring_commit(ring);
194 
195 done:
196 	if (!r)
197 		DRM_INFO("UVD initialized successfully.\n");
198 
199 	return r;
200 }
201 
202 /**
203  * uvd_v4_2_hw_fini - stop the hardware block
204  *
205  * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
206  *
207  * Stop the UVD block, mark ring as not ready any more
208  */
209 static int uvd_v4_2_hw_fini(struct amdgpu_ip_block *ip_block)
210 {
211 	struct amdgpu_device *adev = ip_block->adev;
212 
213 	cancel_delayed_work_sync(&adev->uvd.idle_work);
214 
215 	if (RREG32(mmUVD_STATUS) != 0)
216 		uvd_v4_2_stop(adev);
217 
218 	return 0;
219 }
220 
221 static int uvd_v4_2_prepare_suspend(struct amdgpu_ip_block *ip_block)
222 {
223 	struct amdgpu_device *adev = ip_block->adev;
224 
225 	return amdgpu_uvd_prepare_suspend(adev);
226 }
227 
228 static int uvd_v4_2_suspend(struct amdgpu_ip_block *ip_block)
229 {
230 	int r;
231 	struct amdgpu_device *adev = ip_block->adev;
232 
233 	/*
234 	 * Proper cleanups before halting the HW engine:
235 	 *   - cancel the delayed idle work
236 	 *   - enable powergating
237 	 *   - enable clockgating
238 	 *   - disable dpm
239 	 *
240 	 * TODO: to align with the VCN implementation, move the
241 	 * jobs for clockgating/powergating/dpm setting to
242 	 * ->set_powergating_state().
243 	 */
244 	cancel_delayed_work_sync(&adev->uvd.idle_work);
245 
246 	if (adev->pm.dpm_enabled) {
247 		amdgpu_dpm_enable_uvd(adev, false);
248 	} else {
249 		amdgpu_asic_set_uvd_clocks(adev, 0, 0);
250 		/* shutdown the UVD block */
251 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
252 						       AMD_PG_STATE_GATE);
253 		amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
254 						       AMD_CG_STATE_GATE);
255 	}
256 
257 	r = uvd_v4_2_hw_fini(ip_block);
258 	if (r)
259 		return r;
260 
261 	return amdgpu_uvd_suspend(adev);
262 }
263 
264 static int uvd_v4_2_resume(struct amdgpu_ip_block *ip_block)
265 {
266 	int r;
267 
268 	r = amdgpu_uvd_resume(ip_block->adev);
269 	if (r)
270 		return r;
271 
272 	return uvd_v4_2_hw_init(ip_block);
273 }
274 
275 /**
276  * uvd_v4_2_start - start UVD block
277  *
278  * @adev: amdgpu_device pointer
279  *
280  * Setup and start the UVD block
281  */
282 static int uvd_v4_2_start(struct amdgpu_device *adev)
283 {
284 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
285 	uint32_t rb_bufsz;
286 	int i, j, r;
287 	u32 tmp;
288 	/* disable byte swapping */
289 	u32 lmi_swap_cntl = 0;
290 	u32 mp_swap_cntl = 0;
291 
292 	/* set uvd busy */
293 	WREG32_P(mmUVD_STATUS, 1<<2, ~(1<<2));
294 
295 	uvd_v4_2_set_dcm(adev, true);
296 	WREG32(mmUVD_CGC_GATE, 0);
297 
298 	/* take UVD block out of reset */
299 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
300 	mdelay(5);
301 
302 	/* enable VCPU clock */
303 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
304 
305 	/* disable interupt */
306 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
307 
308 #ifdef __BIG_ENDIAN
309 	/* swap (8 in 32) RB and IB */
310 	lmi_swap_cntl = 0xa;
311 	mp_swap_cntl = 0;
312 #endif
313 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
314 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
315 	/* initialize UVD memory controller */
316 	WREG32(mmUVD_LMI_CTRL, 0x203108);
317 
318 	tmp = RREG32(mmUVD_MPC_CNTL);
319 	WREG32(mmUVD_MPC_CNTL, tmp | 0x10);
320 
321 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
322 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
323 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
324 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
325 	WREG32(mmUVD_MPC_SET_ALU, 0);
326 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
327 
328 	uvd_v4_2_mc_resume(adev);
329 
330 	tmp = RREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL);
331 	WREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL, tmp & (~0x10));
332 
333 	/* enable UMC */
334 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
335 
336 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
337 
338 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
339 
340 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
341 
342 	mdelay(10);
343 
344 	for (i = 0; i < 10; ++i) {
345 		uint32_t status;
346 		for (j = 0; j < 100; ++j) {
347 			status = RREG32(mmUVD_STATUS);
348 			if (status & 2)
349 				break;
350 			mdelay(10);
351 		}
352 		r = 0;
353 		if (status & 2)
354 			break;
355 
356 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
357 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
358 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
359 		mdelay(10);
360 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
361 		mdelay(10);
362 		r = -1;
363 	}
364 
365 	if (r) {
366 		DRM_ERROR("UVD not responding, giving up!!!\n");
367 		return r;
368 	}
369 
370 	/* enable interupt */
371 	WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
372 
373 	WREG32_P(mmUVD_STATUS, 0, ~(1<<2));
374 
375 	/* force RBC into idle state */
376 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
377 
378 	/* Set the write pointer delay */
379 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
380 
381 	/* program the 4GB memory segment for rptr and ring buffer */
382 	WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
383 				   (0x7 << 16) | (0x1 << 31));
384 
385 	/* Initialize the ring buffer's read and write pointers */
386 	WREG32(mmUVD_RBC_RB_RPTR, 0x0);
387 
388 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
389 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
390 
391 	/* set the ring address */
392 	WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
393 
394 	/* Set ring buffer size */
395 	rb_bufsz = order_base_2(ring->ring_size);
396 	rb_bufsz = (0x1 << 8) | rb_bufsz;
397 	WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
398 
399 	return 0;
400 }
401 
402 /**
403  * uvd_v4_2_stop - stop UVD block
404  *
405  * @adev: amdgpu_device pointer
406  *
407  * stop the UVD block
408  */
409 static void uvd_v4_2_stop(struct amdgpu_device *adev)
410 {
411 	uint32_t i, j;
412 	uint32_t status;
413 
414 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
415 
416 	for (i = 0; i < 10; ++i) {
417 		for (j = 0; j < 100; ++j) {
418 			status = RREG32(mmUVD_STATUS);
419 			if (status & 2)
420 				break;
421 			mdelay(1);
422 		}
423 		if (status & 2)
424 			break;
425 	}
426 
427 	for (i = 0; i < 10; ++i) {
428 		for (j = 0; j < 100; ++j) {
429 			status = RREG32(mmUVD_LMI_STATUS);
430 			if (status & 0xf)
431 				break;
432 			mdelay(1);
433 		}
434 		if (status & 0xf)
435 			break;
436 	}
437 
438 	/* Stall UMC and register bus before resetting VCPU */
439 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
440 
441 	for (i = 0; i < 10; ++i) {
442 		for (j = 0; j < 100; ++j) {
443 			status = RREG32(mmUVD_LMI_STATUS);
444 			if (status & 0x240)
445 				break;
446 			mdelay(1);
447 		}
448 		if (status & 0x240)
449 			break;
450 	}
451 
452 	WREG32_P(0x3D49, 0, ~(1 << 2));
453 
454 	WREG32_P(mmUVD_VCPU_CNTL, 0, ~(1 << 9));
455 
456 	/* put LMI, VCPU, RBC etc... into reset */
457 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
458 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
459 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
460 
461 	WREG32(mmUVD_STATUS, 0);
462 
463 	uvd_v4_2_set_dcm(adev, false);
464 }
465 
466 /**
467  * uvd_v4_2_ring_emit_fence - emit an fence & trap command
468  *
469  * @ring: amdgpu_ring pointer
470  * @addr: address
471  * @seq: sequence number
472  * @flags: fence related flags
473  *
474  * Write a fence and a trap command to the ring.
475  */
476 static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
477 				     unsigned flags)
478 {
479 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
480 
481 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
482 	amdgpu_ring_write(ring, seq);
483 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
484 	amdgpu_ring_write(ring, addr & 0xffffffff);
485 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
486 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
487 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
488 	amdgpu_ring_write(ring, 0);
489 
490 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
491 	amdgpu_ring_write(ring, 0);
492 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
493 	amdgpu_ring_write(ring, 0);
494 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
495 	amdgpu_ring_write(ring, 2);
496 }
497 
498 /**
499  * uvd_v4_2_ring_test_ring - register write test
500  *
501  * @ring: amdgpu_ring pointer
502  *
503  * Test if we can successfully write to the context register
504  */
505 static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring)
506 {
507 	struct amdgpu_device *adev = ring->adev;
508 	uint32_t tmp = 0;
509 	unsigned i;
510 	int r;
511 
512 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
513 	r = amdgpu_ring_alloc(ring, 3);
514 	if (r)
515 		return r;
516 
517 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
518 	amdgpu_ring_write(ring, 0xDEADBEEF);
519 	amdgpu_ring_commit(ring);
520 	for (i = 0; i < adev->usec_timeout; i++) {
521 		tmp = RREG32(mmUVD_CONTEXT_ID);
522 		if (tmp == 0xDEADBEEF)
523 			break;
524 		udelay(1);
525 	}
526 
527 	if (i >= adev->usec_timeout)
528 		r = -ETIMEDOUT;
529 
530 	return r;
531 }
532 
533 /**
534  * uvd_v4_2_ring_emit_ib - execute indirect buffer
535  *
536  * @ring: amdgpu_ring pointer
537  * @job: iob associated with the indirect buffer
538  * @ib: indirect buffer to execute
539  * @flags: flags associated with the indirect buffer
540  *
541  * Write ring commands to execute the indirect buffer
542  */
543 static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
544 				  struct amdgpu_job *job,
545 				  struct amdgpu_ib *ib,
546 				  uint32_t flags)
547 {
548 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
549 	amdgpu_ring_write(ring, ib->gpu_addr);
550 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
551 	amdgpu_ring_write(ring, ib->length_dw);
552 }
553 
554 static void uvd_v4_2_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
555 {
556 	int i;
557 
558 	WARN_ON(ring->wptr % 2 || count % 2);
559 
560 	for (i = 0; i < count / 2; i++) {
561 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
562 		amdgpu_ring_write(ring, 0);
563 	}
564 }
565 
566 /**
567  * uvd_v4_2_mc_resume - memory controller programming
568  *
569  * @adev: amdgpu_device pointer
570  *
571  * Let the UVD memory controller know it's offsets
572  */
573 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev)
574 {
575 	uint64_t addr;
576 	uint32_t size;
577 
578 	/* program the VCPU memory controller bits 0-27 */
579 	addr = (adev->uvd.inst->gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
580 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev) >> 3;
581 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
582 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
583 
584 	addr += size;
585 	size = AMDGPU_UVD_HEAP_SIZE >> 3;
586 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
587 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
588 
589 	addr += size;
590 	size = (AMDGPU_UVD_STACK_SIZE +
591 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles)) >> 3;
592 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
593 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
594 
595 	/* bits 28-31 */
596 	addr = (adev->uvd.inst->gpu_addr >> 28) & 0xF;
597 	WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
598 
599 	/* bits 32-39 */
600 	addr = (adev->uvd.inst->gpu_addr >> 32) & 0xFF;
601 	WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
602 
603 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
604 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
605 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
606 }
607 
608 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
609 				 bool enable)
610 {
611 	u32 orig, data;
612 
613 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
614 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
615 		data |= 0xfff;
616 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
617 
618 		orig = data = RREG32(mmUVD_CGC_CTRL);
619 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
620 		if (orig != data)
621 			WREG32(mmUVD_CGC_CTRL, data);
622 	} else {
623 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
624 		data &= ~0xfff;
625 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
626 
627 		orig = data = RREG32(mmUVD_CGC_CTRL);
628 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
629 		if (orig != data)
630 			WREG32(mmUVD_CGC_CTRL, data);
631 	}
632 }
633 
634 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
635 			     bool sw_mode)
636 {
637 	u32 tmp, tmp2;
638 
639 	WREG32_FIELD(UVD_CGC_GATE, REGS, 0);
640 
641 	tmp = RREG32(mmUVD_CGC_CTRL);
642 	tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
643 	tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
644 		(1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
645 		(4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
646 
647 	if (sw_mode) {
648 		tmp &= ~0x7ffff800;
649 		tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
650 			UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
651 			(7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
652 	} else {
653 		tmp |= 0x7ffff800;
654 		tmp2 = 0;
655 	}
656 
657 	WREG32(mmUVD_CGC_CTRL, tmp);
658 	WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
659 }
660 
661 static bool uvd_v4_2_is_idle(void *handle)
662 {
663 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
664 
665 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
666 }
667 
668 static int uvd_v4_2_wait_for_idle(struct amdgpu_ip_block *ip_block)
669 {
670 	unsigned i;
671 	struct amdgpu_device *adev = ip_block->adev;
672 
673 	for (i = 0; i < adev->usec_timeout; i++) {
674 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
675 			return 0;
676 	}
677 	return -ETIMEDOUT;
678 }
679 
680 static int uvd_v4_2_soft_reset(struct amdgpu_ip_block *ip_block)
681 {
682 	struct amdgpu_device *adev = ip_block->adev;
683 
684 	uvd_v4_2_stop(adev);
685 
686 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
687 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
688 	mdelay(5);
689 
690 	return uvd_v4_2_start(adev);
691 }
692 
693 static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev,
694 					struct amdgpu_irq_src *source,
695 					unsigned type,
696 					enum amdgpu_interrupt_state state)
697 {
698 	// TODO
699 	return 0;
700 }
701 
702 static int uvd_v4_2_process_interrupt(struct amdgpu_device *adev,
703 				      struct amdgpu_irq_src *source,
704 				      struct amdgpu_iv_entry *entry)
705 {
706 	DRM_DEBUG("IH: UVD TRAP\n");
707 	amdgpu_fence_process(&adev->uvd.inst->ring);
708 	return 0;
709 }
710 
711 static int uvd_v4_2_set_clockgating_state(void *handle,
712 					  enum amd_clockgating_state state)
713 {
714 	return 0;
715 }
716 
717 static int uvd_v4_2_set_powergating_state(void *handle,
718 					  enum amd_powergating_state state)
719 {
720 	/* This doesn't actually powergate the UVD block.
721 	 * That's done in the dpm code via the SMC.  This
722 	 * just re-inits the block as necessary.  The actual
723 	 * gating still happens in the dpm code.  We should
724 	 * revisit this when there is a cleaner line between
725 	 * the smc and the hw blocks
726 	 */
727 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
728 
729 	if (state == AMD_PG_STATE_GATE) {
730 		uvd_v4_2_stop(adev);
731 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD && !adev->pm.dpm_enabled) {
732 			if (!(RREG32_SMC(ixCURRENT_PG_STATUS) &
733 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK)) {
734 				WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK   |
735 							UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_DOWN_MASK |
736 							UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
737 				mdelay(20);
738 			}
739 		}
740 		return 0;
741 	} else {
742 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD && !adev->pm.dpm_enabled) {
743 			if (RREG32_SMC(ixCURRENT_PG_STATUS) &
744 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
745 				WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK   |
746 						UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_UP_MASK |
747 						UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
748 				mdelay(30);
749 			}
750 		}
751 		return uvd_v4_2_start(adev);
752 	}
753 }
754 
755 static const struct amd_ip_funcs uvd_v4_2_ip_funcs = {
756 	.name = "uvd_v4_2",
757 	.early_init = uvd_v4_2_early_init,
758 	.sw_init = uvd_v4_2_sw_init,
759 	.sw_fini = uvd_v4_2_sw_fini,
760 	.hw_init = uvd_v4_2_hw_init,
761 	.hw_fini = uvd_v4_2_hw_fini,
762 	.prepare_suspend = uvd_v4_2_prepare_suspend,
763 	.suspend = uvd_v4_2_suspend,
764 	.resume = uvd_v4_2_resume,
765 	.is_idle = uvd_v4_2_is_idle,
766 	.wait_for_idle = uvd_v4_2_wait_for_idle,
767 	.soft_reset = uvd_v4_2_soft_reset,
768 	.set_clockgating_state = uvd_v4_2_set_clockgating_state,
769 	.set_powergating_state = uvd_v4_2_set_powergating_state,
770 };
771 
772 static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
773 	.type = AMDGPU_RING_TYPE_UVD,
774 	.align_mask = 0xf,
775 	.support_64bit_ptrs = false,
776 	.no_user_fence = true,
777 	.get_rptr = uvd_v4_2_ring_get_rptr,
778 	.get_wptr = uvd_v4_2_ring_get_wptr,
779 	.set_wptr = uvd_v4_2_ring_set_wptr,
780 	.parse_cs = amdgpu_uvd_ring_parse_cs,
781 	.emit_frame_size =
782 		14, /* uvd_v4_2_ring_emit_fence  x1 no user fence */
783 	.emit_ib_size = 4, /* uvd_v4_2_ring_emit_ib */
784 	.emit_ib = uvd_v4_2_ring_emit_ib,
785 	.emit_fence = uvd_v4_2_ring_emit_fence,
786 	.test_ring = uvd_v4_2_ring_test_ring,
787 	.test_ib = amdgpu_uvd_ring_test_ib,
788 	.insert_nop = uvd_v4_2_ring_insert_nop,
789 	.pad_ib = amdgpu_ring_generic_pad_ib,
790 	.begin_use = amdgpu_uvd_ring_begin_use,
791 	.end_use = amdgpu_uvd_ring_end_use,
792 };
793 
794 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)
795 {
796 	adev->uvd.inst->ring.funcs = &uvd_v4_2_ring_funcs;
797 }
798 
799 static const struct amdgpu_irq_src_funcs uvd_v4_2_irq_funcs = {
800 	.set = uvd_v4_2_set_interrupt_state,
801 	.process = uvd_v4_2_process_interrupt,
802 };
803 
804 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev)
805 {
806 	adev->uvd.inst->irq.num_types = 1;
807 	adev->uvd.inst->irq.funcs = &uvd_v4_2_irq_funcs;
808 }
809 
810 const struct amdgpu_ip_block_version uvd_v4_2_ip_block =
811 {
812 		.type = AMD_IP_BLOCK_TYPE_UVD,
813 		.major = 4,
814 		.minor = 2,
815 		.rev = 0,
816 		.funcs = &uvd_v4_2_ip_funcs,
817 };
818