xref: /linux/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c (revision e08a1d97d33e2ac05cd368b955f9fdc2823f15fd)
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 #include <drm/drmP.h>
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_init_cg(struct amdgpu_device *adev);
44 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev);
45 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev);
46 static int uvd_v4_2_start(struct amdgpu_device *adev);
47 static void uvd_v4_2_stop(struct amdgpu_device *adev);
48 
49 /**
50  * uvd_v4_2_ring_get_rptr - get read pointer
51  *
52  * @ring: amdgpu_ring pointer
53  *
54  * Returns the current hardware read pointer
55  */
56 static uint32_t uvd_v4_2_ring_get_rptr(struct amdgpu_ring *ring)
57 {
58 	struct amdgpu_device *adev = ring->adev;
59 
60 	return RREG32(mmUVD_RBC_RB_RPTR);
61 }
62 
63 /**
64  * uvd_v4_2_ring_get_wptr - get write pointer
65  *
66  * @ring: amdgpu_ring pointer
67  *
68  * Returns the current hardware write pointer
69  */
70 static uint32_t uvd_v4_2_ring_get_wptr(struct amdgpu_ring *ring)
71 {
72 	struct amdgpu_device *adev = ring->adev;
73 
74 	return RREG32(mmUVD_RBC_RB_WPTR);
75 }
76 
77 /**
78  * uvd_v4_2_ring_set_wptr - set write pointer
79  *
80  * @ring: amdgpu_ring pointer
81  *
82  * Commits the write pointer to the hardware
83  */
84 static void uvd_v4_2_ring_set_wptr(struct amdgpu_ring *ring)
85 {
86 	struct amdgpu_device *adev = ring->adev;
87 
88 	WREG32(mmUVD_RBC_RB_WPTR, ring->wptr);
89 }
90 
91 static int uvd_v4_2_early_init(void *handle)
92 {
93 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
94 
95 	uvd_v4_2_set_ring_funcs(adev);
96 	uvd_v4_2_set_irq_funcs(adev);
97 
98 	return 0;
99 }
100 
101 static int uvd_v4_2_sw_init(void *handle)
102 {
103 	struct amdgpu_ring *ring;
104 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
105 	int r;
106 
107 	/* UVD TRAP */
108 	r = amdgpu_irq_add_id(adev, 124, &adev->uvd.irq);
109 	if (r)
110 		return r;
111 
112 	r = amdgpu_uvd_sw_init(adev);
113 	if (r)
114 		return r;
115 
116 	r = amdgpu_uvd_resume(adev);
117 	if (r)
118 		return r;
119 
120 	ring = &adev->uvd.ring;
121 	sprintf(ring->name, "uvd");
122 	r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.irq, 0);
123 
124 	return r;
125 }
126 
127 static int uvd_v4_2_sw_fini(void *handle)
128 {
129 	int r;
130 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
131 
132 	r = amdgpu_uvd_suspend(adev);
133 	if (r)
134 		return r;
135 
136 	r = amdgpu_uvd_sw_fini(adev);
137 	if (r)
138 		return r;
139 
140 	return r;
141 }
142 
143 /**
144  * uvd_v4_2_hw_init - start and test UVD block
145  *
146  * @adev: amdgpu_device pointer
147  *
148  * Initialize the hardware, boot up the VCPU and do some testing
149  */
150 static int uvd_v4_2_hw_init(void *handle)
151 {
152 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
153 	struct amdgpu_ring *ring = &adev->uvd.ring;
154 	uint32_t tmp;
155 	int r;
156 
157 	/* raise clocks while booting up the VCPU */
158 	amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
159 
160 	r = uvd_v4_2_start(adev);
161 	if (r)
162 		goto done;
163 
164 	ring->ready = true;
165 	r = amdgpu_ring_test_ring(ring);
166 	if (r) {
167 		ring->ready = false;
168 		goto done;
169 	}
170 
171 	r = amdgpu_ring_alloc(ring, 10);
172 	if (r) {
173 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
174 		goto done;
175 	}
176 
177 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
178 	amdgpu_ring_write(ring, tmp);
179 	amdgpu_ring_write(ring, 0xFFFFF);
180 
181 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
182 	amdgpu_ring_write(ring, tmp);
183 	amdgpu_ring_write(ring, 0xFFFFF);
184 
185 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
186 	amdgpu_ring_write(ring, tmp);
187 	amdgpu_ring_write(ring, 0xFFFFF);
188 
189 	/* Clear timeout status bits */
190 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
191 	amdgpu_ring_write(ring, 0x8);
192 
193 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
194 	amdgpu_ring_write(ring, 3);
195 
196 	amdgpu_ring_commit(ring);
197 
198 done:
199 	/* lower clocks again */
200 	amdgpu_asic_set_uvd_clocks(adev, 0, 0);
201 
202 	if (!r)
203 		DRM_INFO("UVD initialized successfully.\n");
204 
205 	return r;
206 }
207 
208 /**
209  * uvd_v4_2_hw_fini - stop the hardware block
210  *
211  * @adev: amdgpu_device pointer
212  *
213  * Stop the UVD block, mark ring as not ready any more
214  */
215 static int uvd_v4_2_hw_fini(void *handle)
216 {
217 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
218 	struct amdgpu_ring *ring = &adev->uvd.ring;
219 
220 	uvd_v4_2_stop(adev);
221 	ring->ready = false;
222 
223 	return 0;
224 }
225 
226 static int uvd_v4_2_suspend(void *handle)
227 {
228 	int r;
229 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
230 
231 	r = uvd_v4_2_hw_fini(adev);
232 	if (r)
233 		return r;
234 
235 	r = amdgpu_uvd_suspend(adev);
236 	if (r)
237 		return r;
238 
239 	return r;
240 }
241 
242 static int uvd_v4_2_resume(void *handle)
243 {
244 	int r;
245 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
246 
247 	r = amdgpu_uvd_resume(adev);
248 	if (r)
249 		return r;
250 
251 	r = uvd_v4_2_hw_init(adev);
252 	if (r)
253 		return r;
254 
255 	return r;
256 }
257 
258 /**
259  * uvd_v4_2_start - start UVD block
260  *
261  * @adev: amdgpu_device pointer
262  *
263  * Setup and start the UVD block
264  */
265 static int uvd_v4_2_start(struct amdgpu_device *adev)
266 {
267 	struct amdgpu_ring *ring = &adev->uvd.ring;
268 	uint32_t rb_bufsz;
269 	int i, j, r;
270 
271 	/* disable byte swapping */
272 	u32 lmi_swap_cntl = 0;
273 	u32 mp_swap_cntl = 0;
274 
275 	uvd_v4_2_mc_resume(adev);
276 
277 	/* disable clock gating */
278 	WREG32(mmUVD_CGC_GATE, 0);
279 
280 	/* disable interupt */
281 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
282 
283 	/* Stall UMC and register bus before resetting VCPU */
284 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
285 	mdelay(1);
286 
287 	/* put LMI, VCPU, RBC etc... into reset */
288 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
289 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
290 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
291 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
292 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
293 	mdelay(5);
294 
295 	/* take UVD block out of reset */
296 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
297 	mdelay(5);
298 
299 	/* initialize UVD memory controller */
300 	WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
301 			     (1 << 21) | (1 << 9) | (1 << 20));
302 
303 #ifdef __BIG_ENDIAN
304 	/* swap (8 in 32) RB and IB */
305 	lmi_swap_cntl = 0xa;
306 	mp_swap_cntl = 0;
307 #endif
308 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
309 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
310 
311 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
312 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
313 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
314 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
315 	WREG32(mmUVD_MPC_SET_ALU, 0);
316 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
317 
318 	/* take all subblocks out of reset, except VCPU */
319 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
320 	mdelay(5);
321 
322 	/* enable VCPU clock */
323 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
324 
325 	/* enable UMC */
326 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
327 
328 	/* boot up the VCPU */
329 	WREG32(mmUVD_SOFT_RESET, 0);
330 	mdelay(10);
331 
332 	for (i = 0; i < 10; ++i) {
333 		uint32_t status;
334 		for (j = 0; j < 100; ++j) {
335 			status = RREG32(mmUVD_STATUS);
336 			if (status & 2)
337 				break;
338 			mdelay(10);
339 		}
340 		r = 0;
341 		if (status & 2)
342 			break;
343 
344 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
345 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
346 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
347 		mdelay(10);
348 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
349 		mdelay(10);
350 		r = -1;
351 	}
352 
353 	if (r) {
354 		DRM_ERROR("UVD not responding, giving up!!!\n");
355 		return r;
356 	}
357 
358 	/* enable interupt */
359 	WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
360 
361 	/* force RBC into idle state */
362 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
363 
364 	/* Set the write pointer delay */
365 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
366 
367 	/* programm the 4GB memory segment for rptr and ring buffer */
368 	WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
369 				   (0x7 << 16) | (0x1 << 31));
370 
371 	/* Initialize the ring buffer's read and write pointers */
372 	WREG32(mmUVD_RBC_RB_RPTR, 0x0);
373 
374 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
375 	WREG32(mmUVD_RBC_RB_WPTR, ring->wptr);
376 
377 	/* set the ring address */
378 	WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
379 
380 	/* Set ring buffer size */
381 	rb_bufsz = order_base_2(ring->ring_size);
382 	rb_bufsz = (0x1 << 8) | rb_bufsz;
383 	WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
384 
385 	return 0;
386 }
387 
388 /**
389  * uvd_v4_2_stop - stop UVD block
390  *
391  * @adev: amdgpu_device pointer
392  *
393  * stop the UVD block
394  */
395 static void uvd_v4_2_stop(struct amdgpu_device *adev)
396 {
397 	/* force RBC into idle state */
398 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
399 
400 	/* Stall UMC and register bus before resetting VCPU */
401 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
402 	mdelay(1);
403 
404 	/* put VCPU into reset */
405 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
406 	mdelay(5);
407 
408 	/* disable VCPU clock */
409 	WREG32(mmUVD_VCPU_CNTL, 0x0);
410 
411 	/* Unstall UMC and register bus */
412 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
413 }
414 
415 /**
416  * uvd_v4_2_ring_emit_fence - emit an fence & trap command
417  *
418  * @ring: amdgpu_ring pointer
419  * @fence: fence to emit
420  *
421  * Write a fence and a trap command to the ring.
422  */
423 static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
424 				     unsigned flags)
425 {
426 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
427 
428 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
429 	amdgpu_ring_write(ring, seq);
430 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
431 	amdgpu_ring_write(ring, addr & 0xffffffff);
432 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
433 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
434 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
435 	amdgpu_ring_write(ring, 0);
436 
437 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
438 	amdgpu_ring_write(ring, 0);
439 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
440 	amdgpu_ring_write(ring, 0);
441 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
442 	amdgpu_ring_write(ring, 2);
443 }
444 
445 /**
446  * uvd_v4_2_ring_emit_hdp_flush - emit an hdp flush
447  *
448  * @ring: amdgpu_ring pointer
449  *
450  * Emits an hdp flush.
451  */
452 static void uvd_v4_2_ring_emit_hdp_flush(struct amdgpu_ring *ring)
453 {
454 	amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0));
455 	amdgpu_ring_write(ring, 0);
456 }
457 
458 /**
459  * uvd_v4_2_ring_hdp_invalidate - emit an hdp invalidate
460  *
461  * @ring: amdgpu_ring pointer
462  *
463  * Emits an hdp invalidate.
464  */
465 static void uvd_v4_2_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
466 {
467 	amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0));
468 	amdgpu_ring_write(ring, 1);
469 }
470 
471 /**
472  * uvd_v4_2_ring_test_ring - register write test
473  *
474  * @ring: amdgpu_ring pointer
475  *
476  * Test if we can successfully write to the context register
477  */
478 static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring)
479 {
480 	struct amdgpu_device *adev = ring->adev;
481 	uint32_t tmp = 0;
482 	unsigned i;
483 	int r;
484 
485 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
486 	r = amdgpu_ring_alloc(ring, 3);
487 	if (r) {
488 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
489 			  ring->idx, r);
490 		return r;
491 	}
492 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
493 	amdgpu_ring_write(ring, 0xDEADBEEF);
494 	amdgpu_ring_commit(ring);
495 	for (i = 0; i < adev->usec_timeout; i++) {
496 		tmp = RREG32(mmUVD_CONTEXT_ID);
497 		if (tmp == 0xDEADBEEF)
498 			break;
499 		DRM_UDELAY(1);
500 	}
501 
502 	if (i < adev->usec_timeout) {
503 		DRM_INFO("ring test on %d succeeded in %d usecs\n",
504 			 ring->idx, i);
505 	} else {
506 		DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
507 			  ring->idx, tmp);
508 		r = -EINVAL;
509 	}
510 	return r;
511 }
512 
513 /**
514  * uvd_v4_2_ring_emit_ib - execute indirect buffer
515  *
516  * @ring: amdgpu_ring pointer
517  * @ib: indirect buffer to execute
518  *
519  * Write ring commands to execute the indirect buffer
520  */
521 static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
522 				  struct amdgpu_ib *ib,
523 				  unsigned vm_id, bool ctx_switch)
524 {
525 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
526 	amdgpu_ring_write(ring, ib->gpu_addr);
527 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
528 	amdgpu_ring_write(ring, ib->length_dw);
529 }
530 
531 /**
532  * uvd_v4_2_mc_resume - memory controller programming
533  *
534  * @adev: amdgpu_device pointer
535  *
536  * Let the UVD memory controller know it's offsets
537  */
538 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev)
539 {
540 	uint64_t addr;
541 	uint32_t size;
542 
543 	/* programm the VCPU memory controller bits 0-27 */
544 	addr = (adev->uvd.gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
545 	size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4) >> 3;
546 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
547 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
548 
549 	addr += size;
550 	size = AMDGPU_UVD_HEAP_SIZE >> 3;
551 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
552 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
553 
554 	addr += size;
555 	size = (AMDGPU_UVD_STACK_SIZE +
556 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles)) >> 3;
557 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
558 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
559 
560 	/* bits 28-31 */
561 	addr = (adev->uvd.gpu_addr >> 28) & 0xF;
562 	WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
563 
564 	/* bits 32-39 */
565 	addr = (adev->uvd.gpu_addr >> 32) & 0xFF;
566 	WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
567 
568 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
569 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
570 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
571 
572 	uvd_v4_2_init_cg(adev);
573 }
574 
575 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
576 				 bool enable)
577 {
578 	u32 orig, data;
579 
580 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
581 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
582 		data = 0xfff;
583 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
584 
585 		orig = data = RREG32(mmUVD_CGC_CTRL);
586 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
587 		if (orig != data)
588 			WREG32(mmUVD_CGC_CTRL, data);
589 	} else {
590 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
591 		data &= ~0xfff;
592 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
593 
594 		orig = data = RREG32(mmUVD_CGC_CTRL);
595 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
596 		if (orig != data)
597 			WREG32(mmUVD_CGC_CTRL, data);
598 	}
599 }
600 
601 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
602 			     bool sw_mode)
603 {
604 	u32 tmp, tmp2;
605 
606 	tmp = RREG32(mmUVD_CGC_CTRL);
607 	tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
608 	tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
609 		(1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
610 		(4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
611 
612 	if (sw_mode) {
613 		tmp &= ~0x7ffff800;
614 		tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
615 			UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
616 			(7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
617 	} else {
618 		tmp |= 0x7ffff800;
619 		tmp2 = 0;
620 	}
621 
622 	WREG32(mmUVD_CGC_CTRL, tmp);
623 	WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
624 }
625 
626 static void uvd_v4_2_init_cg(struct amdgpu_device *adev)
627 {
628 	bool hw_mode = true;
629 
630 	if (hw_mode) {
631 		uvd_v4_2_set_dcm(adev, false);
632 	} else {
633 		u32 tmp = RREG32(mmUVD_CGC_CTRL);
634 		tmp &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
635 		WREG32(mmUVD_CGC_CTRL, tmp);
636 	}
637 }
638 
639 static bool uvd_v4_2_is_idle(void *handle)
640 {
641 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
642 
643 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
644 }
645 
646 static int uvd_v4_2_wait_for_idle(void *handle)
647 {
648 	unsigned i;
649 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
650 
651 	for (i = 0; i < adev->usec_timeout; i++) {
652 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
653 			return 0;
654 	}
655 	return -ETIMEDOUT;
656 }
657 
658 static int uvd_v4_2_soft_reset(void *handle)
659 {
660 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
661 
662 	uvd_v4_2_stop(adev);
663 
664 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
665 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
666 	mdelay(5);
667 
668 	return uvd_v4_2_start(adev);
669 }
670 
671 static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev,
672 					struct amdgpu_irq_src *source,
673 					unsigned type,
674 					enum amdgpu_interrupt_state state)
675 {
676 	// TODO
677 	return 0;
678 }
679 
680 static int uvd_v4_2_process_interrupt(struct amdgpu_device *adev,
681 				      struct amdgpu_irq_src *source,
682 				      struct amdgpu_iv_entry *entry)
683 {
684 	DRM_DEBUG("IH: UVD TRAP\n");
685 	amdgpu_fence_process(&adev->uvd.ring);
686 	return 0;
687 }
688 
689 static void uvd_v5_0_set_bypass_mode(struct amdgpu_device *adev, bool enable)
690 {
691 	u32 tmp = RREG32_SMC(ixGCK_DFS_BYPASS_CNTL);
692 
693 	if (enable)
694 		tmp |= (GCK_DFS_BYPASS_CNTL__BYPASSDCLK_MASK |
695 			GCK_DFS_BYPASS_CNTL__BYPASSVCLK_MASK);
696 	else
697 		tmp &= ~(GCK_DFS_BYPASS_CNTL__BYPASSDCLK_MASK |
698 			 GCK_DFS_BYPASS_CNTL__BYPASSVCLK_MASK);
699 
700 	WREG32_SMC(ixGCK_DFS_BYPASS_CNTL, tmp);
701 }
702 
703 static int uvd_v4_2_set_clockgating_state(void *handle,
704 					  enum amd_clockgating_state state)
705 {
706 	bool gate = false;
707 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
708 
709 	if (state == AMD_CG_STATE_GATE)
710 		gate = true;
711 
712 	uvd_v5_0_set_bypass_mode(adev, gate);
713 
714 	if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG))
715 		return 0;
716 
717 	uvd_v4_2_enable_mgcg(adev, gate);
718 
719 	return 0;
720 }
721 
722 static int uvd_v4_2_set_powergating_state(void *handle,
723 					  enum amd_powergating_state state)
724 {
725 	/* This doesn't actually powergate the UVD block.
726 	 * That's done in the dpm code via the SMC.  This
727 	 * just re-inits the block as necessary.  The actual
728 	 * gating still happens in the dpm code.  We should
729 	 * revisit this when there is a cleaner line between
730 	 * the smc and the hw blocks
731 	 */
732 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
733 
734 	if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD))
735 		return 0;
736 
737 	if (state == AMD_PG_STATE_GATE) {
738 		uvd_v4_2_stop(adev);
739 		return 0;
740 	} else {
741 		return uvd_v4_2_start(adev);
742 	}
743 }
744 
745 static const struct amd_ip_funcs uvd_v4_2_ip_funcs = {
746 	.name = "uvd_v4_2",
747 	.early_init = uvd_v4_2_early_init,
748 	.late_init = NULL,
749 	.sw_init = uvd_v4_2_sw_init,
750 	.sw_fini = uvd_v4_2_sw_fini,
751 	.hw_init = uvd_v4_2_hw_init,
752 	.hw_fini = uvd_v4_2_hw_fini,
753 	.suspend = uvd_v4_2_suspend,
754 	.resume = uvd_v4_2_resume,
755 	.is_idle = uvd_v4_2_is_idle,
756 	.wait_for_idle = uvd_v4_2_wait_for_idle,
757 	.soft_reset = uvd_v4_2_soft_reset,
758 	.set_clockgating_state = uvd_v4_2_set_clockgating_state,
759 	.set_powergating_state = uvd_v4_2_set_powergating_state,
760 };
761 
762 static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
763 	.type = AMDGPU_RING_TYPE_UVD,
764 	.align_mask = 0xf,
765 	.nop = PACKET0(mmUVD_NO_OP, 0),
766 	.get_rptr = uvd_v4_2_ring_get_rptr,
767 	.get_wptr = uvd_v4_2_ring_get_wptr,
768 	.set_wptr = uvd_v4_2_ring_set_wptr,
769 	.parse_cs = amdgpu_uvd_ring_parse_cs,
770 	.emit_frame_size =
771 		2 + /* uvd_v4_2_ring_emit_hdp_flush */
772 		2 + /* uvd_v4_2_ring_emit_hdp_invalidate */
773 		14, /* uvd_v4_2_ring_emit_fence  x1 no user fence */
774 	.emit_ib_size = 4, /* uvd_v4_2_ring_emit_ib */
775 	.emit_ib = uvd_v4_2_ring_emit_ib,
776 	.emit_fence = uvd_v4_2_ring_emit_fence,
777 	.emit_hdp_flush = uvd_v4_2_ring_emit_hdp_flush,
778 	.emit_hdp_invalidate = uvd_v4_2_ring_emit_hdp_invalidate,
779 	.test_ring = uvd_v4_2_ring_test_ring,
780 	.test_ib = amdgpu_uvd_ring_test_ib,
781 	.insert_nop = amdgpu_ring_insert_nop,
782 	.pad_ib = amdgpu_ring_generic_pad_ib,
783 	.begin_use = amdgpu_uvd_ring_begin_use,
784 	.end_use = amdgpu_uvd_ring_end_use,
785 };
786 
787 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)
788 {
789 	adev->uvd.ring.funcs = &uvd_v4_2_ring_funcs;
790 }
791 
792 static const struct amdgpu_irq_src_funcs uvd_v4_2_irq_funcs = {
793 	.set = uvd_v4_2_set_interrupt_state,
794 	.process = uvd_v4_2_process_interrupt,
795 };
796 
797 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev)
798 {
799 	adev->uvd.irq.num_types = 1;
800 	adev->uvd.irq.funcs = &uvd_v4_2_irq_funcs;
801 }
802 
803 const struct amdgpu_ip_block_version uvd_v4_2_ip_block =
804 {
805 		.type = AMD_IP_BLOCK_TYPE_UVD,
806 		.major = 4,
807 		.minor = 2,
808 		.rev = 0,
809 		.funcs = &uvd_v4_2_ip_funcs,
810 };
811