xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c (revision 8730046c)
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 static int uvd_v4_2_set_clockgating_state(void *handle,
49 				enum amd_clockgating_state state);
50 /**
51  * uvd_v4_2_ring_get_rptr - get read pointer
52  *
53  * @ring: amdgpu_ring pointer
54  *
55  * Returns the current hardware read pointer
56  */
57 static uint32_t uvd_v4_2_ring_get_rptr(struct amdgpu_ring *ring)
58 {
59 	struct amdgpu_device *adev = ring->adev;
60 
61 	return RREG32(mmUVD_RBC_RB_RPTR);
62 }
63 
64 /**
65  * uvd_v4_2_ring_get_wptr - get write pointer
66  *
67  * @ring: amdgpu_ring pointer
68  *
69  * Returns the current hardware write pointer
70  */
71 static uint32_t uvd_v4_2_ring_get_wptr(struct amdgpu_ring *ring)
72 {
73 	struct amdgpu_device *adev = ring->adev;
74 
75 	return RREG32(mmUVD_RBC_RB_WPTR);
76 }
77 
78 /**
79  * uvd_v4_2_ring_set_wptr - set write pointer
80  *
81  * @ring: amdgpu_ring pointer
82  *
83  * Commits the write pointer to the hardware
84  */
85 static void uvd_v4_2_ring_set_wptr(struct amdgpu_ring *ring)
86 {
87 	struct amdgpu_device *adev = ring->adev;
88 
89 	WREG32(mmUVD_RBC_RB_WPTR, ring->wptr);
90 }
91 
92 static int uvd_v4_2_early_init(void *handle)
93 {
94 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
95 
96 	uvd_v4_2_set_ring_funcs(adev);
97 	uvd_v4_2_set_irq_funcs(adev);
98 
99 	return 0;
100 }
101 
102 static int uvd_v4_2_sw_init(void *handle)
103 {
104 	struct amdgpu_ring *ring;
105 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
106 	int r;
107 
108 	/* UVD TRAP */
109 	r = amdgpu_irq_add_id(adev, 124, &adev->uvd.irq);
110 	if (r)
111 		return r;
112 
113 	r = amdgpu_uvd_sw_init(adev);
114 	if (r)
115 		return r;
116 
117 	r = amdgpu_uvd_resume(adev);
118 	if (r)
119 		return r;
120 
121 	ring = &adev->uvd.ring;
122 	sprintf(ring->name, "uvd");
123 	r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.irq, 0);
124 
125 	return r;
126 }
127 
128 static int uvd_v4_2_sw_fini(void *handle)
129 {
130 	int r;
131 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
132 
133 	r = amdgpu_uvd_suspend(adev);
134 	if (r)
135 		return r;
136 
137 	r = amdgpu_uvd_sw_fini(adev);
138 	if (r)
139 		return r;
140 
141 	return r;
142 }
143 
144 /**
145  * uvd_v4_2_hw_init - start and test UVD block
146  *
147  * @adev: amdgpu_device pointer
148  *
149  * Initialize the hardware, boot up the VCPU and do some testing
150  */
151 static int uvd_v4_2_hw_init(void *handle)
152 {
153 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
154 	struct amdgpu_ring *ring = &adev->uvd.ring;
155 	uint32_t tmp;
156 	int r;
157 
158 	uvd_v4_2_init_cg(adev);
159 	uvd_v4_2_set_clockgating_state(adev, AMD_CG_STATE_GATE);
160 	amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
161 	r = uvd_v4_2_start(adev);
162 	if (r)
163 		goto done;
164 
165 	ring->ready = true;
166 	r = amdgpu_ring_test_ring(ring);
167 	if (r) {
168 		ring->ready = false;
169 		goto done;
170 	}
171 
172 	r = amdgpu_ring_alloc(ring, 10);
173 	if (r) {
174 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
175 		goto done;
176 	}
177 
178 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
179 	amdgpu_ring_write(ring, tmp);
180 	amdgpu_ring_write(ring, 0xFFFFF);
181 
182 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
183 	amdgpu_ring_write(ring, tmp);
184 	amdgpu_ring_write(ring, 0xFFFFF);
185 
186 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
187 	amdgpu_ring_write(ring, tmp);
188 	amdgpu_ring_write(ring, 0xFFFFF);
189 
190 	/* Clear timeout status bits */
191 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
192 	amdgpu_ring_write(ring, 0x8);
193 
194 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
195 	amdgpu_ring_write(ring, 3);
196 
197 	amdgpu_ring_commit(ring);
198 
199 done:
200 
201 	if (!r)
202 		DRM_INFO("UVD initialized successfully.\n");
203 
204 	return r;
205 }
206 
207 /**
208  * uvd_v4_2_hw_fini - stop the hardware block
209  *
210  * @adev: amdgpu_device pointer
211  *
212  * Stop the UVD block, mark ring as not ready any more
213  */
214 static int uvd_v4_2_hw_fini(void *handle)
215 {
216 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
217 	struct amdgpu_ring *ring = &adev->uvd.ring;
218 
219 	uvd_v4_2_stop(adev);
220 	ring->ready = false;
221 
222 	return 0;
223 }
224 
225 static int uvd_v4_2_suspend(void *handle)
226 {
227 	int r;
228 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
229 
230 	r = uvd_v4_2_hw_fini(adev);
231 	if (r)
232 		return r;
233 
234 	r = amdgpu_uvd_suspend(adev);
235 	if (r)
236 		return r;
237 
238 	return r;
239 }
240 
241 static int uvd_v4_2_resume(void *handle)
242 {
243 	int r;
244 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
245 
246 	r = amdgpu_uvd_resume(adev);
247 	if (r)
248 		return r;
249 
250 	r = uvd_v4_2_hw_init(adev);
251 	if (r)
252 		return r;
253 
254 	return r;
255 }
256 
257 /**
258  * uvd_v4_2_start - start UVD block
259  *
260  * @adev: amdgpu_device pointer
261  *
262  * Setup and start the UVD block
263  */
264 static int uvd_v4_2_start(struct amdgpu_device *adev)
265 {
266 	struct amdgpu_ring *ring = &adev->uvd.ring;
267 	uint32_t rb_bufsz;
268 	int i, j, r;
269 
270 	/* disable byte swapping */
271 	u32 lmi_swap_cntl = 0;
272 	u32 mp_swap_cntl = 0;
273 
274 	uvd_v4_2_mc_resume(adev);
275 
276 	/* disable interupt */
277 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
278 
279 	/* Stall UMC and register bus before resetting VCPU */
280 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
281 	mdelay(1);
282 
283 	/* put LMI, VCPU, RBC etc... into reset */
284 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
285 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
286 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
287 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
288 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
289 	mdelay(5);
290 
291 	/* take UVD block out of reset */
292 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
293 	mdelay(5);
294 
295 	/* initialize UVD memory controller */
296 	WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
297 			     (1 << 21) | (1 << 9) | (1 << 20));
298 
299 #ifdef __BIG_ENDIAN
300 	/* swap (8 in 32) RB and IB */
301 	lmi_swap_cntl = 0xa;
302 	mp_swap_cntl = 0;
303 #endif
304 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
305 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
306 
307 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
308 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
309 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
310 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
311 	WREG32(mmUVD_MPC_SET_ALU, 0);
312 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
313 
314 	/* take all subblocks out of reset, except VCPU */
315 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
316 	mdelay(5);
317 
318 	/* enable VCPU clock */
319 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
320 
321 	/* enable UMC */
322 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
323 
324 	/* boot up the VCPU */
325 	WREG32(mmUVD_SOFT_RESET, 0);
326 	mdelay(10);
327 
328 	for (i = 0; i < 10; ++i) {
329 		uint32_t status;
330 		for (j = 0; j < 100; ++j) {
331 			status = RREG32(mmUVD_STATUS);
332 			if (status & 2)
333 				break;
334 			mdelay(10);
335 		}
336 		r = 0;
337 		if (status & 2)
338 			break;
339 
340 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
341 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
342 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
343 		mdelay(10);
344 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
345 		mdelay(10);
346 		r = -1;
347 	}
348 
349 	if (r) {
350 		DRM_ERROR("UVD not responding, giving up!!!\n");
351 		return r;
352 	}
353 
354 	/* enable interupt */
355 	WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
356 
357 	/* force RBC into idle state */
358 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
359 
360 	/* Set the write pointer delay */
361 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
362 
363 	/* programm the 4GB memory segment for rptr and ring buffer */
364 	WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
365 				   (0x7 << 16) | (0x1 << 31));
366 
367 	/* Initialize the ring buffer's read and write pointers */
368 	WREG32(mmUVD_RBC_RB_RPTR, 0x0);
369 
370 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
371 	WREG32(mmUVD_RBC_RB_WPTR, ring->wptr);
372 
373 	/* set the ring address */
374 	WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
375 
376 	/* Set ring buffer size */
377 	rb_bufsz = order_base_2(ring->ring_size);
378 	rb_bufsz = (0x1 << 8) | rb_bufsz;
379 	WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
380 
381 	return 0;
382 }
383 
384 /**
385  * uvd_v4_2_stop - stop UVD block
386  *
387  * @adev: amdgpu_device pointer
388  *
389  * stop the UVD block
390  */
391 static void uvd_v4_2_stop(struct amdgpu_device *adev)
392 {
393 	/* force RBC into idle state */
394 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
395 
396 	/* Stall UMC and register bus before resetting VCPU */
397 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
398 	mdelay(1);
399 
400 	/* put VCPU into reset */
401 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
402 	mdelay(5);
403 
404 	/* disable VCPU clock */
405 	WREG32(mmUVD_VCPU_CNTL, 0x0);
406 
407 	/* Unstall UMC and register bus */
408 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
409 }
410 
411 /**
412  * uvd_v4_2_ring_emit_fence - emit an fence & trap command
413  *
414  * @ring: amdgpu_ring pointer
415  * @fence: fence to emit
416  *
417  * Write a fence and a trap command to the ring.
418  */
419 static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
420 				     unsigned flags)
421 {
422 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
423 
424 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
425 	amdgpu_ring_write(ring, seq);
426 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
427 	amdgpu_ring_write(ring, addr & 0xffffffff);
428 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
429 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
430 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
431 	amdgpu_ring_write(ring, 0);
432 
433 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
434 	amdgpu_ring_write(ring, 0);
435 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
436 	amdgpu_ring_write(ring, 0);
437 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
438 	amdgpu_ring_write(ring, 2);
439 }
440 
441 /**
442  * uvd_v4_2_ring_emit_hdp_flush - emit an hdp flush
443  *
444  * @ring: amdgpu_ring pointer
445  *
446  * Emits an hdp flush.
447  */
448 static void uvd_v4_2_ring_emit_hdp_flush(struct amdgpu_ring *ring)
449 {
450 	amdgpu_ring_write(ring, PACKET0(mmHDP_MEM_COHERENCY_FLUSH_CNTL, 0));
451 	amdgpu_ring_write(ring, 0);
452 }
453 
454 /**
455  * uvd_v4_2_ring_hdp_invalidate - emit an hdp invalidate
456  *
457  * @ring: amdgpu_ring pointer
458  *
459  * Emits an hdp invalidate.
460  */
461 static void uvd_v4_2_ring_emit_hdp_invalidate(struct amdgpu_ring *ring)
462 {
463 	amdgpu_ring_write(ring, PACKET0(mmHDP_DEBUG0, 0));
464 	amdgpu_ring_write(ring, 1);
465 }
466 
467 /**
468  * uvd_v4_2_ring_test_ring - register write test
469  *
470  * @ring: amdgpu_ring pointer
471  *
472  * Test if we can successfully write to the context register
473  */
474 static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring)
475 {
476 	struct amdgpu_device *adev = ring->adev;
477 	uint32_t tmp = 0;
478 	unsigned i;
479 	int r;
480 
481 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
482 	r = amdgpu_ring_alloc(ring, 3);
483 	if (r) {
484 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
485 			  ring->idx, r);
486 		return r;
487 	}
488 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
489 	amdgpu_ring_write(ring, 0xDEADBEEF);
490 	amdgpu_ring_commit(ring);
491 	for (i = 0; i < adev->usec_timeout; i++) {
492 		tmp = RREG32(mmUVD_CONTEXT_ID);
493 		if (tmp == 0xDEADBEEF)
494 			break;
495 		DRM_UDELAY(1);
496 	}
497 
498 	if (i < adev->usec_timeout) {
499 		DRM_INFO("ring test on %d succeeded in %d usecs\n",
500 			 ring->idx, i);
501 	} else {
502 		DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
503 			  ring->idx, tmp);
504 		r = -EINVAL;
505 	}
506 	return r;
507 }
508 
509 /**
510  * uvd_v4_2_ring_emit_ib - execute indirect buffer
511  *
512  * @ring: amdgpu_ring pointer
513  * @ib: indirect buffer to execute
514  *
515  * Write ring commands to execute the indirect buffer
516  */
517 static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
518 				  struct amdgpu_ib *ib,
519 				  unsigned vm_id, bool ctx_switch)
520 {
521 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
522 	amdgpu_ring_write(ring, ib->gpu_addr);
523 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
524 	amdgpu_ring_write(ring, ib->length_dw);
525 }
526 
527 /**
528  * uvd_v4_2_mc_resume - memory controller programming
529  *
530  * @adev: amdgpu_device pointer
531  *
532  * Let the UVD memory controller know it's offsets
533  */
534 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev)
535 {
536 	uint64_t addr;
537 	uint32_t size;
538 
539 	/* programm the VCPU memory controller bits 0-27 */
540 	addr = (adev->uvd.gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
541 	size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4) >> 3;
542 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
543 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
544 
545 	addr += size;
546 	size = AMDGPU_UVD_HEAP_SIZE >> 3;
547 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
548 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
549 
550 	addr += size;
551 	size = (AMDGPU_UVD_STACK_SIZE +
552 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles)) >> 3;
553 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
554 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
555 
556 	/* bits 28-31 */
557 	addr = (adev->uvd.gpu_addr >> 28) & 0xF;
558 	WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
559 
560 	/* bits 32-39 */
561 	addr = (adev->uvd.gpu_addr >> 32) & 0xFF;
562 	WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
563 
564 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
565 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
566 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
567 }
568 
569 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
570 				 bool enable)
571 {
572 	u32 orig, data;
573 
574 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
575 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
576 		data |= 0xfff;
577 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
578 
579 		orig = data = RREG32(mmUVD_CGC_CTRL);
580 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
581 		if (orig != data)
582 			WREG32(mmUVD_CGC_CTRL, data);
583 	} else {
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 	}
593 }
594 
595 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
596 			     bool sw_mode)
597 {
598 	u32 tmp, tmp2;
599 
600 	WREG32_FIELD(UVD_CGC_GATE, REGS, 0);
601 
602 	tmp = RREG32(mmUVD_CGC_CTRL);
603 	tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
604 	tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
605 		(1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
606 		(4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
607 
608 	if (sw_mode) {
609 		tmp &= ~0x7ffff800;
610 		tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
611 			UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
612 			(7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
613 	} else {
614 		tmp |= 0x7ffff800;
615 		tmp2 = 0;
616 	}
617 
618 	WREG32(mmUVD_CGC_CTRL, tmp);
619 	WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
620 }
621 
622 static void uvd_v4_2_init_cg(struct amdgpu_device *adev)
623 {
624 	bool hw_mode = true;
625 
626 	if (hw_mode) {
627 		uvd_v4_2_set_dcm(adev, false);
628 	} else {
629 		u32 tmp = RREG32(mmUVD_CGC_CTRL);
630 		tmp &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
631 		WREG32(mmUVD_CGC_CTRL, tmp);
632 	}
633 }
634 
635 static bool uvd_v4_2_is_idle(void *handle)
636 {
637 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
638 
639 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
640 }
641 
642 static int uvd_v4_2_wait_for_idle(void *handle)
643 {
644 	unsigned i;
645 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
646 
647 	for (i = 0; i < adev->usec_timeout; i++) {
648 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
649 			return 0;
650 	}
651 	return -ETIMEDOUT;
652 }
653 
654 static int uvd_v4_2_soft_reset(void *handle)
655 {
656 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
657 
658 	uvd_v4_2_stop(adev);
659 
660 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
661 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
662 	mdelay(5);
663 
664 	return uvd_v4_2_start(adev);
665 }
666 
667 static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev,
668 					struct amdgpu_irq_src *source,
669 					unsigned type,
670 					enum amdgpu_interrupt_state state)
671 {
672 	// TODO
673 	return 0;
674 }
675 
676 static int uvd_v4_2_process_interrupt(struct amdgpu_device *adev,
677 				      struct amdgpu_irq_src *source,
678 				      struct amdgpu_iv_entry *entry)
679 {
680 	DRM_DEBUG("IH: UVD TRAP\n");
681 	amdgpu_fence_process(&adev->uvd.ring);
682 	return 0;
683 }
684 
685 static int uvd_v4_2_set_clockgating_state(void *handle,
686 					  enum amd_clockgating_state state)
687 {
688 	bool gate = false;
689 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
690 
691 	if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG))
692 		return 0;
693 
694 	if (state == AMD_CG_STATE_GATE)
695 		gate = true;
696 
697 	uvd_v4_2_enable_mgcg(adev, gate);
698 
699 	return 0;
700 }
701 
702 static int uvd_v4_2_set_powergating_state(void *handle,
703 					  enum amd_powergating_state state)
704 {
705 	/* This doesn't actually powergate the UVD block.
706 	 * That's done in the dpm code via the SMC.  This
707 	 * just re-inits the block as necessary.  The actual
708 	 * gating still happens in the dpm code.  We should
709 	 * revisit this when there is a cleaner line between
710 	 * the smc and the hw blocks
711 	 */
712 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
713 
714 	if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD))
715 		return 0;
716 
717 	if (state == AMD_PG_STATE_GATE) {
718 		uvd_v4_2_stop(adev);
719 		return 0;
720 	} else {
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 	.nop = PACKET0(mmUVD_NO_OP, 0),
746 	.get_rptr = uvd_v4_2_ring_get_rptr,
747 	.get_wptr = uvd_v4_2_ring_get_wptr,
748 	.set_wptr = uvd_v4_2_ring_set_wptr,
749 	.parse_cs = amdgpu_uvd_ring_parse_cs,
750 	.emit_frame_size =
751 		2 + /* uvd_v4_2_ring_emit_hdp_flush */
752 		2 + /* uvd_v4_2_ring_emit_hdp_invalidate */
753 		14, /* uvd_v4_2_ring_emit_fence  x1 no user fence */
754 	.emit_ib_size = 4, /* uvd_v4_2_ring_emit_ib */
755 	.emit_ib = uvd_v4_2_ring_emit_ib,
756 	.emit_fence = uvd_v4_2_ring_emit_fence,
757 	.emit_hdp_flush = uvd_v4_2_ring_emit_hdp_flush,
758 	.emit_hdp_invalidate = uvd_v4_2_ring_emit_hdp_invalidate,
759 	.test_ring = uvd_v4_2_ring_test_ring,
760 	.test_ib = amdgpu_uvd_ring_test_ib,
761 	.insert_nop = amdgpu_ring_insert_nop,
762 	.pad_ib = amdgpu_ring_generic_pad_ib,
763 	.begin_use = amdgpu_uvd_ring_begin_use,
764 	.end_use = amdgpu_uvd_ring_end_use,
765 };
766 
767 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)
768 {
769 	adev->uvd.ring.funcs = &uvd_v4_2_ring_funcs;
770 }
771 
772 static const struct amdgpu_irq_src_funcs uvd_v4_2_irq_funcs = {
773 	.set = uvd_v4_2_set_interrupt_state,
774 	.process = uvd_v4_2_process_interrupt,
775 };
776 
777 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev)
778 {
779 	adev->uvd.irq.num_types = 1;
780 	adev->uvd.irq.funcs = &uvd_v4_2_irq_funcs;
781 }
782 
783 const struct amdgpu_ip_block_version uvd_v4_2_ip_block =
784 {
785 		.type = AMD_IP_BLOCK_TYPE_UVD,
786 		.major = 4,
787 		.minor = 2,
788 		.rev = 0,
789 		.funcs = &uvd_v4_2_ip_funcs,
790 };
791