xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c (revision 73affd30)
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(void *handle)
94 {
95 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
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(void *handle)
105 {
106 	struct amdgpu_ring *ring;
107 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
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 	r = amdgpu_uvd_entity_init(adev);
131 
132 	return r;
133 }
134 
135 static int uvd_v4_2_sw_fini(void *handle)
136 {
137 	int r;
138 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
139 
140 	r = amdgpu_uvd_suspend(adev);
141 	if (r)
142 		return r;
143 
144 	return amdgpu_uvd_sw_fini(adev);
145 }
146 
147 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
148 				 bool enable);
149 /**
150  * uvd_v4_2_hw_init - start and test UVD block
151  *
152  * @handle: handle used to pass amdgpu_device pointer
153  *
154  * Initialize the hardware, boot up the VCPU and do some testing
155  */
156 static int uvd_v4_2_hw_init(void *handle)
157 {
158 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
159 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
160 	uint32_t tmp;
161 	int r;
162 
163 	uvd_v4_2_enable_mgcg(adev, true);
164 	amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
165 
166 	r = amdgpu_ring_test_helper(ring);
167 	if (r)
168 		goto done;
169 
170 	r = amdgpu_ring_alloc(ring, 10);
171 	if (r) {
172 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
173 		goto done;
174 	}
175 
176 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
177 	amdgpu_ring_write(ring, tmp);
178 	amdgpu_ring_write(ring, 0xFFFFF);
179 
180 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
181 	amdgpu_ring_write(ring, tmp);
182 	amdgpu_ring_write(ring, 0xFFFFF);
183 
184 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
185 	amdgpu_ring_write(ring, tmp);
186 	amdgpu_ring_write(ring, 0xFFFFF);
187 
188 	/* Clear timeout status bits */
189 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
190 	amdgpu_ring_write(ring, 0x8);
191 
192 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
193 	amdgpu_ring_write(ring, 3);
194 
195 	amdgpu_ring_commit(ring);
196 
197 done:
198 	if (!r)
199 		DRM_INFO("UVD initialized successfully.\n");
200 
201 	return r;
202 }
203 
204 /**
205  * uvd_v4_2_hw_fini - stop the hardware block
206  *
207  * @handle: handle used to pass amdgpu_device pointer
208  *
209  * Stop the UVD block, mark ring as not ready any more
210  */
211 static int uvd_v4_2_hw_fini(void *handle)
212 {
213 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
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_suspend(void *handle)
222 {
223 	int r;
224 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
225 
226 	r = uvd_v4_2_hw_fini(adev);
227 	if (r)
228 		return r;
229 
230 	return amdgpu_uvd_suspend(adev);
231 }
232 
233 static int uvd_v4_2_resume(void *handle)
234 {
235 	int r;
236 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
237 
238 	r = amdgpu_uvd_resume(adev);
239 	if (r)
240 		return r;
241 
242 	return uvd_v4_2_hw_init(adev);
243 }
244 
245 /**
246  * uvd_v4_2_start - start UVD block
247  *
248  * @adev: amdgpu_device pointer
249  *
250  * Setup and start the UVD block
251  */
252 static int uvd_v4_2_start(struct amdgpu_device *adev)
253 {
254 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
255 	uint32_t rb_bufsz;
256 	int i, j, r;
257 	u32 tmp;
258 	/* disable byte swapping */
259 	u32 lmi_swap_cntl = 0;
260 	u32 mp_swap_cntl = 0;
261 
262 	/* set uvd busy */
263 	WREG32_P(mmUVD_STATUS, 1<<2, ~(1<<2));
264 
265 	uvd_v4_2_set_dcm(adev, true);
266 	WREG32(mmUVD_CGC_GATE, 0);
267 
268 	/* take UVD block out of reset */
269 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
270 	mdelay(5);
271 
272 	/* enable VCPU clock */
273 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
274 
275 	/* disable interupt */
276 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
277 
278 #ifdef __BIG_ENDIAN
279 	/* swap (8 in 32) RB and IB */
280 	lmi_swap_cntl = 0xa;
281 	mp_swap_cntl = 0;
282 #endif
283 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
284 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
285 	/* initialize UVD memory controller */
286 	WREG32(mmUVD_LMI_CTRL, 0x203108);
287 
288 	tmp = RREG32(mmUVD_MPC_CNTL);
289 	WREG32(mmUVD_MPC_CNTL, tmp | 0x10);
290 
291 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
292 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
293 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
294 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
295 	WREG32(mmUVD_MPC_SET_ALU, 0);
296 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
297 
298 	uvd_v4_2_mc_resume(adev);
299 
300 	tmp = RREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL);
301 	WREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL, tmp & (~0x10));
302 
303 	/* enable UMC */
304 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
305 
306 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
307 
308 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
309 
310 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
311 
312 	mdelay(10);
313 
314 	for (i = 0; i < 10; ++i) {
315 		uint32_t status;
316 		for (j = 0; j < 100; ++j) {
317 			status = RREG32(mmUVD_STATUS);
318 			if (status & 2)
319 				break;
320 			mdelay(10);
321 		}
322 		r = 0;
323 		if (status & 2)
324 			break;
325 
326 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
327 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
328 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
329 		mdelay(10);
330 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
331 		mdelay(10);
332 		r = -1;
333 	}
334 
335 	if (r) {
336 		DRM_ERROR("UVD not responding, giving up!!!\n");
337 		return r;
338 	}
339 
340 	/* enable interupt */
341 	WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
342 
343 	WREG32_P(mmUVD_STATUS, 0, ~(1<<2));
344 
345 	/* force RBC into idle state */
346 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
347 
348 	/* Set the write pointer delay */
349 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
350 
351 	/* program the 4GB memory segment for rptr and ring buffer */
352 	WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
353 				   (0x7 << 16) | (0x1 << 31));
354 
355 	/* Initialize the ring buffer's read and write pointers */
356 	WREG32(mmUVD_RBC_RB_RPTR, 0x0);
357 
358 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
359 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
360 
361 	/* set the ring address */
362 	WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
363 
364 	/* Set ring buffer size */
365 	rb_bufsz = order_base_2(ring->ring_size);
366 	rb_bufsz = (0x1 << 8) | rb_bufsz;
367 	WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
368 
369 	return 0;
370 }
371 
372 /**
373  * uvd_v4_2_stop - stop UVD block
374  *
375  * @adev: amdgpu_device pointer
376  *
377  * stop the UVD block
378  */
379 static void uvd_v4_2_stop(struct amdgpu_device *adev)
380 {
381 	uint32_t i, j;
382 	uint32_t status;
383 
384 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
385 
386 	for (i = 0; i < 10; ++i) {
387 		for (j = 0; j < 100; ++j) {
388 			status = RREG32(mmUVD_STATUS);
389 			if (status & 2)
390 				break;
391 			mdelay(1);
392 		}
393 		if (status & 2)
394 			break;
395 	}
396 
397 	for (i = 0; i < 10; ++i) {
398 		for (j = 0; j < 100; ++j) {
399 			status = RREG32(mmUVD_LMI_STATUS);
400 			if (status & 0xf)
401 				break;
402 			mdelay(1);
403 		}
404 		if (status & 0xf)
405 			break;
406 	}
407 
408 	/* Stall UMC and register bus before resetting VCPU */
409 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
410 
411 	for (i = 0; i < 10; ++i) {
412 		for (j = 0; j < 100; ++j) {
413 			status = RREG32(mmUVD_LMI_STATUS);
414 			if (status & 0x240)
415 				break;
416 			mdelay(1);
417 		}
418 		if (status & 0x240)
419 			break;
420 	}
421 
422 	WREG32_P(0x3D49, 0, ~(1 << 2));
423 
424 	WREG32_P(mmUVD_VCPU_CNTL, 0, ~(1 << 9));
425 
426 	/* put LMI, VCPU, RBC etc... into reset */
427 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
428 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
429 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
430 
431 	WREG32(mmUVD_STATUS, 0);
432 
433 	uvd_v4_2_set_dcm(adev, false);
434 }
435 
436 /**
437  * uvd_v4_2_ring_emit_fence - emit an fence & trap command
438  *
439  * @ring: amdgpu_ring pointer
440  * @addr: address
441  * @seq: sequence number
442  * @flags: fence related flags
443  *
444  * Write a fence and a trap command to the ring.
445  */
446 static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
447 				     unsigned flags)
448 {
449 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
450 
451 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
452 	amdgpu_ring_write(ring, seq);
453 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
454 	amdgpu_ring_write(ring, addr & 0xffffffff);
455 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
456 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
457 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
458 	amdgpu_ring_write(ring, 0);
459 
460 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
461 	amdgpu_ring_write(ring, 0);
462 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
463 	amdgpu_ring_write(ring, 0);
464 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
465 	amdgpu_ring_write(ring, 2);
466 }
467 
468 /**
469  * uvd_v4_2_ring_test_ring - register write test
470  *
471  * @ring: amdgpu_ring pointer
472  *
473  * Test if we can successfully write to the context register
474  */
475 static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring)
476 {
477 	struct amdgpu_device *adev = ring->adev;
478 	uint32_t tmp = 0;
479 	unsigned i;
480 	int r;
481 
482 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
483 	r = amdgpu_ring_alloc(ring, 3);
484 	if (r)
485 		return r;
486 
487 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
488 	amdgpu_ring_write(ring, 0xDEADBEEF);
489 	amdgpu_ring_commit(ring);
490 	for (i = 0; i < adev->usec_timeout; i++) {
491 		tmp = RREG32(mmUVD_CONTEXT_ID);
492 		if (tmp == 0xDEADBEEF)
493 			break;
494 		udelay(1);
495 	}
496 
497 	if (i >= adev->usec_timeout)
498 		r = -ETIMEDOUT;
499 
500 	return r;
501 }
502 
503 /**
504  * uvd_v4_2_ring_emit_ib - execute indirect buffer
505  *
506  * @ring: amdgpu_ring pointer
507  * @job: iob associated with the indirect buffer
508  * @ib: indirect buffer to execute
509  * @flags: flags associated with the indirect buffer
510  *
511  * Write ring commands to execute the indirect buffer
512  */
513 static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
514 				  struct amdgpu_job *job,
515 				  struct amdgpu_ib *ib,
516 				  uint32_t flags)
517 {
518 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
519 	amdgpu_ring_write(ring, ib->gpu_addr);
520 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
521 	amdgpu_ring_write(ring, ib->length_dw);
522 }
523 
524 static void uvd_v4_2_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
525 {
526 	int i;
527 
528 	WARN_ON(ring->wptr % 2 || count % 2);
529 
530 	for (i = 0; i < count / 2; i++) {
531 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
532 		amdgpu_ring_write(ring, 0);
533 	}
534 }
535 
536 /**
537  * uvd_v4_2_mc_resume - memory controller programming
538  *
539  * @adev: amdgpu_device pointer
540  *
541  * Let the UVD memory controller know it's offsets
542  */
543 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev)
544 {
545 	uint64_t addr;
546 	uint32_t size;
547 
548 	/* program the VCPU memory controller bits 0-27 */
549 	addr = (adev->uvd.inst->gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
550 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev) >> 3;
551 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
552 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
553 
554 	addr += size;
555 	size = AMDGPU_UVD_HEAP_SIZE >> 3;
556 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
557 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
558 
559 	addr += size;
560 	size = (AMDGPU_UVD_STACK_SIZE +
561 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles)) >> 3;
562 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
563 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
564 
565 	/* bits 28-31 */
566 	addr = (adev->uvd.inst->gpu_addr >> 28) & 0xF;
567 	WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
568 
569 	/* bits 32-39 */
570 	addr = (adev->uvd.inst->gpu_addr >> 32) & 0xFF;
571 	WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
572 
573 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
574 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
575 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
576 }
577 
578 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
579 				 bool enable)
580 {
581 	u32 orig, data;
582 
583 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
584 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
585 		data |= 0xfff;
586 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
587 
588 		orig = data = RREG32(mmUVD_CGC_CTRL);
589 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
590 		if (orig != data)
591 			WREG32(mmUVD_CGC_CTRL, data);
592 	} else {
593 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
594 		data &= ~0xfff;
595 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
596 
597 		orig = data = RREG32(mmUVD_CGC_CTRL);
598 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
599 		if (orig != data)
600 			WREG32(mmUVD_CGC_CTRL, data);
601 	}
602 }
603 
604 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
605 			     bool sw_mode)
606 {
607 	u32 tmp, tmp2;
608 
609 	WREG32_FIELD(UVD_CGC_GATE, REGS, 0);
610 
611 	tmp = RREG32(mmUVD_CGC_CTRL);
612 	tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
613 	tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
614 		(1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
615 		(4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
616 
617 	if (sw_mode) {
618 		tmp &= ~0x7ffff800;
619 		tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
620 			UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
621 			(7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
622 	} else {
623 		tmp |= 0x7ffff800;
624 		tmp2 = 0;
625 	}
626 
627 	WREG32(mmUVD_CGC_CTRL, tmp);
628 	WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
629 }
630 
631 static bool uvd_v4_2_is_idle(void *handle)
632 {
633 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
634 
635 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
636 }
637 
638 static int uvd_v4_2_wait_for_idle(void *handle)
639 {
640 	unsigned i;
641 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
642 
643 	for (i = 0; i < adev->usec_timeout; i++) {
644 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
645 			return 0;
646 	}
647 	return -ETIMEDOUT;
648 }
649 
650 static int uvd_v4_2_soft_reset(void *handle)
651 {
652 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
653 
654 	uvd_v4_2_stop(adev);
655 
656 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
657 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
658 	mdelay(5);
659 
660 	return uvd_v4_2_start(adev);
661 }
662 
663 static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev,
664 					struct amdgpu_irq_src *source,
665 					unsigned type,
666 					enum amdgpu_interrupt_state state)
667 {
668 	// TODO
669 	return 0;
670 }
671 
672 static int uvd_v4_2_process_interrupt(struct amdgpu_device *adev,
673 				      struct amdgpu_irq_src *source,
674 				      struct amdgpu_iv_entry *entry)
675 {
676 	DRM_DEBUG("IH: UVD TRAP\n");
677 	amdgpu_fence_process(&adev->uvd.inst->ring);
678 	return 0;
679 }
680 
681 static int uvd_v4_2_set_clockgating_state(void *handle,
682 					  enum amd_clockgating_state state)
683 {
684 	return 0;
685 }
686 
687 static int uvd_v4_2_set_powergating_state(void *handle,
688 					  enum amd_powergating_state state)
689 {
690 	/* This doesn't actually powergate the UVD block.
691 	 * That's done in the dpm code via the SMC.  This
692 	 * just re-inits the block as necessary.  The actual
693 	 * gating still happens in the dpm code.  We should
694 	 * revisit this when there is a cleaner line between
695 	 * the smc and the hw blocks
696 	 */
697 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
698 
699 	if (state == AMD_PG_STATE_GATE) {
700 		uvd_v4_2_stop(adev);
701 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD && !adev->pm.dpm_enabled) {
702 			if (!(RREG32_SMC(ixCURRENT_PG_STATUS) &
703 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK)) {
704 				WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK   |
705 							UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_DOWN_MASK |
706 							UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
707 				mdelay(20);
708 			}
709 		}
710 		return 0;
711 	} else {
712 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD && !adev->pm.dpm_enabled) {
713 			if (RREG32_SMC(ixCURRENT_PG_STATUS) &
714 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
715 				WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK   |
716 						UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_UP_MASK |
717 						UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
718 				mdelay(30);
719 			}
720 		}
721 		return uvd_v4_2_start(adev);
722 	}
723 }
724 
725 static const struct amd_ip_funcs uvd_v4_2_ip_funcs = {
726 	.name = "uvd_v4_2",
727 	.early_init = uvd_v4_2_early_init,
728 	.late_init = NULL,
729 	.sw_init = uvd_v4_2_sw_init,
730 	.sw_fini = uvd_v4_2_sw_fini,
731 	.hw_init = uvd_v4_2_hw_init,
732 	.hw_fini = uvd_v4_2_hw_fini,
733 	.suspend = uvd_v4_2_suspend,
734 	.resume = uvd_v4_2_resume,
735 	.is_idle = uvd_v4_2_is_idle,
736 	.wait_for_idle = uvd_v4_2_wait_for_idle,
737 	.soft_reset = uvd_v4_2_soft_reset,
738 	.set_clockgating_state = uvd_v4_2_set_clockgating_state,
739 	.set_powergating_state = uvd_v4_2_set_powergating_state,
740 };
741 
742 static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
743 	.type = AMDGPU_RING_TYPE_UVD,
744 	.align_mask = 0xf,
745 	.support_64bit_ptrs = false,
746 	.no_user_fence = true,
747 	.get_rptr = uvd_v4_2_ring_get_rptr,
748 	.get_wptr = uvd_v4_2_ring_get_wptr,
749 	.set_wptr = uvd_v4_2_ring_set_wptr,
750 	.parse_cs = amdgpu_uvd_ring_parse_cs,
751 	.emit_frame_size =
752 		14, /* uvd_v4_2_ring_emit_fence  x1 no user fence */
753 	.emit_ib_size = 4, /* uvd_v4_2_ring_emit_ib */
754 	.emit_ib = uvd_v4_2_ring_emit_ib,
755 	.emit_fence = uvd_v4_2_ring_emit_fence,
756 	.test_ring = uvd_v4_2_ring_test_ring,
757 	.test_ib = amdgpu_uvd_ring_test_ib,
758 	.insert_nop = uvd_v4_2_ring_insert_nop,
759 	.pad_ib = amdgpu_ring_generic_pad_ib,
760 	.begin_use = amdgpu_uvd_ring_begin_use,
761 	.end_use = amdgpu_uvd_ring_end_use,
762 };
763 
764 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)
765 {
766 	adev->uvd.inst->ring.funcs = &uvd_v4_2_ring_funcs;
767 }
768 
769 static const struct amdgpu_irq_src_funcs uvd_v4_2_irq_funcs = {
770 	.set = uvd_v4_2_set_interrupt_state,
771 	.process = uvd_v4_2_process_interrupt,
772 };
773 
774 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev)
775 {
776 	adev->uvd.inst->irq.num_types = 1;
777 	adev->uvd.inst->irq.funcs = &uvd_v4_2_irq_funcs;
778 }
779 
780 const struct amdgpu_ip_block_version uvd_v4_2_ip_block =
781 {
782 		.type = AMD_IP_BLOCK_TYPE_UVD,
783 		.major = 4,
784 		.minor = 2,
785 		.rev = 0,
786 		.funcs = &uvd_v4_2_ip_funcs,
787 };
788