xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c (revision 78560d41)
1 /*
2  * Copyright 2014 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/delay.h>
26 #include <linux/firmware.h>
27 
28 #include "amdgpu.h"
29 #include "amdgpu_uvd.h"
30 #include "vid.h"
31 #include "uvd/uvd_5_0_d.h"
32 #include "uvd/uvd_5_0_sh_mask.h"
33 #include "oss/oss_2_0_d.h"
34 #include "oss/oss_2_0_sh_mask.h"
35 #include "bif/bif_5_0_d.h"
36 #include "vi.h"
37 #include "smu/smu_7_1_2_d.h"
38 #include "smu/smu_7_1_2_sh_mask.h"
39 #include "ivsrcid/ivsrcid_vislands30.h"
40 
41 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev);
42 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev);
43 static int uvd_v5_0_start(struct amdgpu_device *adev);
44 static void uvd_v5_0_stop(struct amdgpu_device *adev);
45 static int uvd_v5_0_set_clockgating_state(void *handle,
46 					  enum amd_clockgating_state state);
47 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev,
48 				 bool enable);
49 /**
50  * uvd_v5_0_ring_get_rptr - get read pointer
51  *
52  * @ring: amdgpu_ring pointer
53  *
54  * Returns the current hardware read pointer
55  */
56 static uint64_t uvd_v5_0_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_v5_0_ring_get_wptr - get write pointer
65  *
66  * @ring: amdgpu_ring pointer
67  *
68  * Returns the current hardware write pointer
69  */
70 static uint64_t uvd_v5_0_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_v5_0_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_v5_0_ring_set_wptr(struct amdgpu_ring *ring)
85 {
86 	struct amdgpu_device *adev = ring->adev;
87 
88 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
89 }
90 
91 static int uvd_v5_0_early_init(void *handle)
92 {
93 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
94 	adev->uvd.num_uvd_inst = 1;
95 
96 	uvd_v5_0_set_ring_funcs(adev);
97 	uvd_v5_0_set_irq_funcs(adev);
98 
99 	return 0;
100 }
101 
102 static int uvd_v5_0_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, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_UVD_SYSTEM_MESSAGE, &adev->uvd.inst->irq);
110 	if (r)
111 		return r;
112 
113 	r = amdgpu_uvd_sw_init(adev);
114 	if (r)
115 		return r;
116 
117 	ring = &adev->uvd.inst->ring;
118 	sprintf(ring->name, "uvd");
119 	r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst->irq, 0);
120 	if (r)
121 		return r;
122 
123 	r = amdgpu_uvd_resume(adev);
124 	if (r)
125 		return r;
126 
127 	r = amdgpu_uvd_entity_init(adev);
128 
129 	return r;
130 }
131 
132 static int uvd_v5_0_sw_fini(void *handle)
133 {
134 	int r;
135 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
136 
137 	r = amdgpu_uvd_suspend(adev);
138 	if (r)
139 		return r;
140 
141 	return amdgpu_uvd_sw_fini(adev);
142 }
143 
144 /**
145  * uvd_v5_0_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_v5_0_hw_init(void *handle)
152 {
153 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
154 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
155 	uint32_t tmp;
156 	int r;
157 
158 	amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
159 	uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_UNGATE);
160 	uvd_v5_0_enable_mgcg(adev, true);
161 
162 	r = amdgpu_ring_test_helper(ring);
163 	if (r)
164 		goto done;
165 
166 	r = amdgpu_ring_alloc(ring, 10);
167 	if (r) {
168 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
169 		goto done;
170 	}
171 
172 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
173 	amdgpu_ring_write(ring, tmp);
174 	amdgpu_ring_write(ring, 0xFFFFF);
175 
176 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
177 	amdgpu_ring_write(ring, tmp);
178 	amdgpu_ring_write(ring, 0xFFFFF);
179 
180 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
181 	amdgpu_ring_write(ring, tmp);
182 	amdgpu_ring_write(ring, 0xFFFFF);
183 
184 	/* Clear timeout status bits */
185 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
186 	amdgpu_ring_write(ring, 0x8);
187 
188 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
189 	amdgpu_ring_write(ring, 3);
190 
191 	amdgpu_ring_commit(ring);
192 
193 done:
194 	if (!r)
195 		DRM_INFO("UVD initialized successfully.\n");
196 
197 	return r;
198 
199 }
200 
201 /**
202  * uvd_v5_0_hw_fini - stop the hardware block
203  *
204  * @adev: amdgpu_device pointer
205  *
206  * Stop the UVD block, mark ring as not ready any more
207  */
208 static int uvd_v5_0_hw_fini(void *handle)
209 {
210 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
211 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
212 
213 	if (RREG32(mmUVD_STATUS) != 0)
214 		uvd_v5_0_stop(adev);
215 
216 	ring->sched.ready = false;
217 
218 	return 0;
219 }
220 
221 static int uvd_v5_0_suspend(void *handle)
222 {
223 	int r;
224 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
225 
226 	r = uvd_v5_0_hw_fini(adev);
227 	if (r)
228 		return r;
229 	uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_GATE);
230 
231 	return amdgpu_uvd_suspend(adev);
232 }
233 
234 static int uvd_v5_0_resume(void *handle)
235 {
236 	int r;
237 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
238 
239 	r = amdgpu_uvd_resume(adev);
240 	if (r)
241 		return r;
242 
243 	return uvd_v5_0_hw_init(adev);
244 }
245 
246 /**
247  * uvd_v5_0_mc_resume - memory controller programming
248  *
249  * @adev: amdgpu_device pointer
250  *
251  * Let the UVD memory controller know it's offsets
252  */
253 static void uvd_v5_0_mc_resume(struct amdgpu_device *adev)
254 {
255 	uint64_t offset;
256 	uint32_t size;
257 
258 	/* programm memory controller bits 0-27 */
259 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
260 			lower_32_bits(adev->uvd.inst->gpu_addr));
261 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
262 			upper_32_bits(adev->uvd.inst->gpu_addr));
263 
264 	offset = AMDGPU_UVD_FIRMWARE_OFFSET;
265 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev);
266 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, offset >> 3);
267 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
268 
269 	offset += size;
270 	size = AMDGPU_UVD_HEAP_SIZE;
271 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, offset >> 3);
272 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
273 
274 	offset += size;
275 	size = AMDGPU_UVD_STACK_SIZE +
276 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles);
277 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3);
278 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
279 
280 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
281 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
282 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
283 }
284 
285 /**
286  * uvd_v5_0_start - start UVD block
287  *
288  * @adev: amdgpu_device pointer
289  *
290  * Setup and start the UVD block
291  */
292 static int uvd_v5_0_start(struct amdgpu_device *adev)
293 {
294 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
295 	uint32_t rb_bufsz, tmp;
296 	uint32_t lmi_swap_cntl;
297 	uint32_t mp_swap_cntl;
298 	int i, j, r;
299 
300 	/*disable DPG */
301 	WREG32_P(mmUVD_POWER_STATUS, 0, ~(1 << 2));
302 
303 	/* disable byte swapping */
304 	lmi_swap_cntl = 0;
305 	mp_swap_cntl = 0;
306 
307 	uvd_v5_0_mc_resume(adev);
308 
309 	/* disable interupt */
310 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
311 
312 	/* stall UMC and register bus before resetting VCPU */
313 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
314 	mdelay(1);
315 
316 	/* put LMI, VCPU, RBC etc... into reset */
317 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
318 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
319 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
320 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
321 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
322 	mdelay(5);
323 
324 	/* take UVD block out of reset */
325 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
326 	mdelay(5);
327 
328 	/* initialize UVD memory controller */
329 	WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
330 			     (1 << 21) | (1 << 9) | (1 << 20));
331 
332 #ifdef __BIG_ENDIAN
333 	/* swap (8 in 32) RB and IB */
334 	lmi_swap_cntl = 0xa;
335 	mp_swap_cntl = 0;
336 #endif
337 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
338 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
339 
340 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
341 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
342 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
343 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
344 	WREG32(mmUVD_MPC_SET_ALU, 0);
345 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
346 
347 	/* take all subblocks out of reset, except VCPU */
348 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
349 	mdelay(5);
350 
351 	/* enable VCPU clock */
352 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
353 
354 	/* enable UMC */
355 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
356 
357 	/* boot up the VCPU */
358 	WREG32(mmUVD_SOFT_RESET, 0);
359 	mdelay(10);
360 
361 	for (i = 0; i < 10; ++i) {
362 		uint32_t status;
363 		for (j = 0; j < 100; ++j) {
364 			status = RREG32(mmUVD_STATUS);
365 			if (status & 2)
366 				break;
367 			mdelay(10);
368 		}
369 		r = 0;
370 		if (status & 2)
371 			break;
372 
373 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
374 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
375 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
376 		mdelay(10);
377 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
378 		mdelay(10);
379 		r = -1;
380 	}
381 
382 	if (r) {
383 		DRM_ERROR("UVD not responding, giving up!!!\n");
384 		return r;
385 	}
386 	/* enable master interrupt */
387 	WREG32_P(mmUVD_MASTINT_EN, 3 << 1, ~(3 << 1));
388 
389 	/* clear the bit 4 of UVD_STATUS */
390 	WREG32_P(mmUVD_STATUS, 0, ~(2 << 1));
391 
392 	rb_bufsz = order_base_2(ring->ring_size);
393 	tmp = 0;
394 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
395 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
396 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
397 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0);
398 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
399 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
400 	/* force RBC into idle state */
401 	WREG32(mmUVD_RBC_RB_CNTL, tmp);
402 
403 	/* set the write pointer delay */
404 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
405 
406 	/* set the wb address */
407 	WREG32(mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2));
408 
409 	/* programm the RB_BASE for ring buffer */
410 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
411 			lower_32_bits(ring->gpu_addr));
412 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
413 			upper_32_bits(ring->gpu_addr));
414 
415 	/* Initialize the ring buffer's read and write pointers */
416 	WREG32(mmUVD_RBC_RB_RPTR, 0);
417 
418 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
419 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
420 
421 	WREG32_P(mmUVD_RBC_RB_CNTL, 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK);
422 
423 	return 0;
424 }
425 
426 /**
427  * uvd_v5_0_stop - stop UVD block
428  *
429  * @adev: amdgpu_device pointer
430  *
431  * stop the UVD block
432  */
433 static void uvd_v5_0_stop(struct amdgpu_device *adev)
434 {
435 	/* force RBC into idle state */
436 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
437 
438 	/* Stall UMC and register bus before resetting VCPU */
439 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
440 	mdelay(1);
441 
442 	/* put VCPU into reset */
443 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
444 	mdelay(5);
445 
446 	/* disable VCPU clock */
447 	WREG32(mmUVD_VCPU_CNTL, 0x0);
448 
449 	/* Unstall UMC and register bus */
450 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
451 
452 	WREG32(mmUVD_STATUS, 0);
453 }
454 
455 /**
456  * uvd_v5_0_ring_emit_fence - emit an fence & trap command
457  *
458  * @ring: amdgpu_ring pointer
459  * @fence: fence to emit
460  *
461  * Write a fence and a trap command to the ring.
462  */
463 static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
464 				     unsigned flags)
465 {
466 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
467 
468 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
469 	amdgpu_ring_write(ring, seq);
470 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
471 	amdgpu_ring_write(ring, addr & 0xffffffff);
472 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
473 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
474 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
475 	amdgpu_ring_write(ring, 0);
476 
477 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
478 	amdgpu_ring_write(ring, 0);
479 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
480 	amdgpu_ring_write(ring, 0);
481 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
482 	amdgpu_ring_write(ring, 2);
483 }
484 
485 /**
486  * uvd_v5_0_ring_test_ring - register write test
487  *
488  * @ring: amdgpu_ring pointer
489  *
490  * Test if we can successfully write to the context register
491  */
492 static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring)
493 {
494 	struct amdgpu_device *adev = ring->adev;
495 	uint32_t tmp = 0;
496 	unsigned i;
497 	int r;
498 
499 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
500 	r = amdgpu_ring_alloc(ring, 3);
501 	if (r)
502 		return r;
503 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
504 	amdgpu_ring_write(ring, 0xDEADBEEF);
505 	amdgpu_ring_commit(ring);
506 	for (i = 0; i < adev->usec_timeout; i++) {
507 		tmp = RREG32(mmUVD_CONTEXT_ID);
508 		if (tmp == 0xDEADBEEF)
509 			break;
510 		udelay(1);
511 	}
512 
513 	if (i >= adev->usec_timeout)
514 		r = -ETIMEDOUT;
515 
516 	return r;
517 }
518 
519 /**
520  * uvd_v5_0_ring_emit_ib - execute indirect buffer
521  *
522  * @ring: amdgpu_ring pointer
523  * @ib: indirect buffer to execute
524  *
525  * Write ring commands to execute the indirect buffer
526  */
527 static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
528 				  struct amdgpu_job *job,
529 				  struct amdgpu_ib *ib,
530 				  uint32_t flags)
531 {
532 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
533 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
534 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
535 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
536 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
537 	amdgpu_ring_write(ring, ib->length_dw);
538 }
539 
540 static void uvd_v5_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
541 {
542 	int i;
543 
544 	WARN_ON(ring->wptr % 2 || count % 2);
545 
546 	for (i = 0; i < count / 2; i++) {
547 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
548 		amdgpu_ring_write(ring, 0);
549 	}
550 }
551 
552 static bool uvd_v5_0_is_idle(void *handle)
553 {
554 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
555 
556 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
557 }
558 
559 static int uvd_v5_0_wait_for_idle(void *handle)
560 {
561 	unsigned i;
562 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
563 
564 	for (i = 0; i < adev->usec_timeout; i++) {
565 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
566 			return 0;
567 	}
568 	return -ETIMEDOUT;
569 }
570 
571 static int uvd_v5_0_soft_reset(void *handle)
572 {
573 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
574 
575 	uvd_v5_0_stop(adev);
576 
577 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
578 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
579 	mdelay(5);
580 
581 	return uvd_v5_0_start(adev);
582 }
583 
584 static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev,
585 					struct amdgpu_irq_src *source,
586 					unsigned type,
587 					enum amdgpu_interrupt_state state)
588 {
589 	// TODO
590 	return 0;
591 }
592 
593 static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev,
594 				      struct amdgpu_irq_src *source,
595 				      struct amdgpu_iv_entry *entry)
596 {
597 	DRM_DEBUG("IH: UVD TRAP\n");
598 	amdgpu_fence_process(&adev->uvd.inst->ring);
599 	return 0;
600 }
601 
602 static void uvd_v5_0_enable_clock_gating(struct amdgpu_device *adev, bool enable)
603 {
604 	uint32_t data1, data3, suvd_flags;
605 
606 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
607 	data3 = RREG32(mmUVD_CGC_GATE);
608 
609 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
610 		     UVD_SUVD_CGC_GATE__SIT_MASK |
611 		     UVD_SUVD_CGC_GATE__SMP_MASK |
612 		     UVD_SUVD_CGC_GATE__SCM_MASK |
613 		     UVD_SUVD_CGC_GATE__SDB_MASK;
614 
615 	if (enable) {
616 		data3 |= (UVD_CGC_GATE__SYS_MASK     |
617 			UVD_CGC_GATE__UDEC_MASK      |
618 			UVD_CGC_GATE__MPEG2_MASK     |
619 			UVD_CGC_GATE__RBC_MASK       |
620 			UVD_CGC_GATE__LMI_MC_MASK    |
621 			UVD_CGC_GATE__IDCT_MASK      |
622 			UVD_CGC_GATE__MPRD_MASK      |
623 			UVD_CGC_GATE__MPC_MASK       |
624 			UVD_CGC_GATE__LBSI_MASK      |
625 			UVD_CGC_GATE__LRBBM_MASK     |
626 			UVD_CGC_GATE__UDEC_RE_MASK   |
627 			UVD_CGC_GATE__UDEC_CM_MASK   |
628 			UVD_CGC_GATE__UDEC_IT_MASK   |
629 			UVD_CGC_GATE__UDEC_DB_MASK   |
630 			UVD_CGC_GATE__UDEC_MP_MASK   |
631 			UVD_CGC_GATE__WCB_MASK       |
632 			UVD_CGC_GATE__JPEG_MASK      |
633 			UVD_CGC_GATE__SCPU_MASK);
634 		/* only in pg enabled, we can gate clock to vcpu*/
635 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD)
636 			data3 |= UVD_CGC_GATE__VCPU_MASK;
637 		data3 &= ~UVD_CGC_GATE__REGS_MASK;
638 		data1 |= suvd_flags;
639 	} else {
640 		data3 = 0;
641 		data1 = 0;
642 	}
643 
644 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
645 	WREG32(mmUVD_CGC_GATE, data3);
646 }
647 
648 static void uvd_v5_0_set_sw_clock_gating(struct amdgpu_device *adev)
649 {
650 	uint32_t data, data2;
651 
652 	data = RREG32(mmUVD_CGC_CTRL);
653 	data2 = RREG32(mmUVD_SUVD_CGC_CTRL);
654 
655 
656 	data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
657 		  UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
658 
659 
660 	data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
661 		(1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) |
662 		(4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY));
663 
664 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
665 			UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
666 			UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
667 			UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
668 			UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
669 			UVD_CGC_CTRL__SYS_MODE_MASK |
670 			UVD_CGC_CTRL__UDEC_MODE_MASK |
671 			UVD_CGC_CTRL__MPEG2_MODE_MASK |
672 			UVD_CGC_CTRL__REGS_MODE_MASK |
673 			UVD_CGC_CTRL__RBC_MODE_MASK |
674 			UVD_CGC_CTRL__LMI_MC_MODE_MASK |
675 			UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
676 			UVD_CGC_CTRL__IDCT_MODE_MASK |
677 			UVD_CGC_CTRL__MPRD_MODE_MASK |
678 			UVD_CGC_CTRL__MPC_MODE_MASK |
679 			UVD_CGC_CTRL__LBSI_MODE_MASK |
680 			UVD_CGC_CTRL__LRBBM_MODE_MASK |
681 			UVD_CGC_CTRL__WCB_MODE_MASK |
682 			UVD_CGC_CTRL__VCPU_MODE_MASK |
683 			UVD_CGC_CTRL__JPEG_MODE_MASK |
684 			UVD_CGC_CTRL__SCPU_MODE_MASK);
685 	data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
686 			UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
687 			UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
688 			UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
689 			UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
690 
691 	WREG32(mmUVD_CGC_CTRL, data);
692 	WREG32(mmUVD_SUVD_CGC_CTRL, data2);
693 }
694 
695 #if 0
696 static void uvd_v5_0_set_hw_clock_gating(struct amdgpu_device *adev)
697 {
698 	uint32_t data, data1, cgc_flags, suvd_flags;
699 
700 	data = RREG32(mmUVD_CGC_GATE);
701 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
702 
703 	cgc_flags = UVD_CGC_GATE__SYS_MASK |
704 				UVD_CGC_GATE__UDEC_MASK |
705 				UVD_CGC_GATE__MPEG2_MASK |
706 				UVD_CGC_GATE__RBC_MASK |
707 				UVD_CGC_GATE__LMI_MC_MASK |
708 				UVD_CGC_GATE__IDCT_MASK |
709 				UVD_CGC_GATE__MPRD_MASK |
710 				UVD_CGC_GATE__MPC_MASK |
711 				UVD_CGC_GATE__LBSI_MASK |
712 				UVD_CGC_GATE__LRBBM_MASK |
713 				UVD_CGC_GATE__UDEC_RE_MASK |
714 				UVD_CGC_GATE__UDEC_CM_MASK |
715 				UVD_CGC_GATE__UDEC_IT_MASK |
716 				UVD_CGC_GATE__UDEC_DB_MASK |
717 				UVD_CGC_GATE__UDEC_MP_MASK |
718 				UVD_CGC_GATE__WCB_MASK |
719 				UVD_CGC_GATE__VCPU_MASK |
720 				UVD_CGC_GATE__SCPU_MASK;
721 
722 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
723 				UVD_SUVD_CGC_GATE__SIT_MASK |
724 				UVD_SUVD_CGC_GATE__SMP_MASK |
725 				UVD_SUVD_CGC_GATE__SCM_MASK |
726 				UVD_SUVD_CGC_GATE__SDB_MASK;
727 
728 	data |= cgc_flags;
729 	data1 |= suvd_flags;
730 
731 	WREG32(mmUVD_CGC_GATE, data);
732 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
733 }
734 #endif
735 
736 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev,
737 				 bool enable)
738 {
739 	u32 orig, data;
740 
741 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
742 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
743 		data |= 0xfff;
744 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
745 
746 		orig = data = RREG32(mmUVD_CGC_CTRL);
747 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
748 		if (orig != data)
749 			WREG32(mmUVD_CGC_CTRL, data);
750 	} else {
751 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
752 		data &= ~0xfff;
753 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
754 
755 		orig = data = RREG32(mmUVD_CGC_CTRL);
756 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
757 		if (orig != data)
758 			WREG32(mmUVD_CGC_CTRL, data);
759 	}
760 }
761 
762 static int uvd_v5_0_set_clockgating_state(void *handle,
763 					  enum amd_clockgating_state state)
764 {
765 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
766 	bool enable = (state == AMD_CG_STATE_GATE);
767 
768 	if (enable) {
769 		/* wait for STATUS to clear */
770 		if (uvd_v5_0_wait_for_idle(handle))
771 			return -EBUSY;
772 		uvd_v5_0_enable_clock_gating(adev, true);
773 
774 		/* enable HW gates because UVD is idle */
775 /*		uvd_v5_0_set_hw_clock_gating(adev); */
776 	} else {
777 		uvd_v5_0_enable_clock_gating(adev, false);
778 	}
779 
780 	uvd_v5_0_set_sw_clock_gating(adev);
781 	return 0;
782 }
783 
784 static int uvd_v5_0_set_powergating_state(void *handle,
785 					  enum amd_powergating_state state)
786 {
787 	/* This doesn't actually powergate the UVD block.
788 	 * That's done in the dpm code via the SMC.  This
789 	 * just re-inits the block as necessary.  The actual
790 	 * gating still happens in the dpm code.  We should
791 	 * revisit this when there is a cleaner line between
792 	 * the smc and the hw blocks
793 	 */
794 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
795 	int ret = 0;
796 
797 	if (state == AMD_PG_STATE_GATE) {
798 		uvd_v5_0_stop(adev);
799 	} else {
800 		ret = uvd_v5_0_start(adev);
801 		if (ret)
802 			goto out;
803 	}
804 
805 out:
806 	return ret;
807 }
808 
809 static void uvd_v5_0_get_clockgating_state(void *handle, u32 *flags)
810 {
811 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
812 	int data;
813 
814 	mutex_lock(&adev->pm.mutex);
815 
816 	if (RREG32_SMC(ixCURRENT_PG_STATUS) &
817 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
818 		DRM_INFO("Cannot get clockgating state when UVD is powergated.\n");
819 		goto out;
820 	}
821 
822 	/* AMD_CG_SUPPORT_UVD_MGCG */
823 	data = RREG32(mmUVD_CGC_CTRL);
824 	if (data & UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK)
825 		*flags |= AMD_CG_SUPPORT_UVD_MGCG;
826 
827 out:
828 	mutex_unlock(&adev->pm.mutex);
829 }
830 
831 static const struct amd_ip_funcs uvd_v5_0_ip_funcs = {
832 	.name = "uvd_v5_0",
833 	.early_init = uvd_v5_0_early_init,
834 	.late_init = NULL,
835 	.sw_init = uvd_v5_0_sw_init,
836 	.sw_fini = uvd_v5_0_sw_fini,
837 	.hw_init = uvd_v5_0_hw_init,
838 	.hw_fini = uvd_v5_0_hw_fini,
839 	.suspend = uvd_v5_0_suspend,
840 	.resume = uvd_v5_0_resume,
841 	.is_idle = uvd_v5_0_is_idle,
842 	.wait_for_idle = uvd_v5_0_wait_for_idle,
843 	.soft_reset = uvd_v5_0_soft_reset,
844 	.set_clockgating_state = uvd_v5_0_set_clockgating_state,
845 	.set_powergating_state = uvd_v5_0_set_powergating_state,
846 	.get_clockgating_state = uvd_v5_0_get_clockgating_state,
847 };
848 
849 static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
850 	.type = AMDGPU_RING_TYPE_UVD,
851 	.align_mask = 0xf,
852 	.support_64bit_ptrs = false,
853 	.no_user_fence = true,
854 	.get_rptr = uvd_v5_0_ring_get_rptr,
855 	.get_wptr = uvd_v5_0_ring_get_wptr,
856 	.set_wptr = uvd_v5_0_ring_set_wptr,
857 	.parse_cs = amdgpu_uvd_ring_parse_cs,
858 	.emit_frame_size =
859 		14, /* uvd_v5_0_ring_emit_fence  x1 no user fence */
860 	.emit_ib_size = 6, /* uvd_v5_0_ring_emit_ib */
861 	.emit_ib = uvd_v5_0_ring_emit_ib,
862 	.emit_fence = uvd_v5_0_ring_emit_fence,
863 	.test_ring = uvd_v5_0_ring_test_ring,
864 	.test_ib = amdgpu_uvd_ring_test_ib,
865 	.insert_nop = uvd_v5_0_ring_insert_nop,
866 	.pad_ib = amdgpu_ring_generic_pad_ib,
867 	.begin_use = amdgpu_uvd_ring_begin_use,
868 	.end_use = amdgpu_uvd_ring_end_use,
869 };
870 
871 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev)
872 {
873 	adev->uvd.inst->ring.funcs = &uvd_v5_0_ring_funcs;
874 }
875 
876 static const struct amdgpu_irq_src_funcs uvd_v5_0_irq_funcs = {
877 	.set = uvd_v5_0_set_interrupt_state,
878 	.process = uvd_v5_0_process_interrupt,
879 };
880 
881 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev)
882 {
883 	adev->uvd.inst->irq.num_types = 1;
884 	adev->uvd.inst->irq.funcs = &uvd_v5_0_irq_funcs;
885 }
886 
887 const struct amdgpu_ip_block_version uvd_v5_0_ip_block =
888 {
889 		.type = AMD_IP_BLOCK_TYPE_UVD,
890 		.major = 5,
891 		.minor = 0,
892 		.rev = 0,
893 		.funcs = &uvd_v5_0_ip_funcs,
894 };
895