xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c (revision 79f382b9)
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 			     AMDGPU_RING_PRIO_DEFAULT);
121 	if (r)
122 		return r;
123 
124 	r = amdgpu_uvd_resume(adev);
125 	if (r)
126 		return r;
127 
128 	r = amdgpu_uvd_entity_init(adev);
129 
130 	return r;
131 }
132 
133 static int uvd_v5_0_sw_fini(void *handle)
134 {
135 	int r;
136 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
137 
138 	r = amdgpu_uvd_suspend(adev);
139 	if (r)
140 		return r;
141 
142 	return amdgpu_uvd_sw_fini(adev);
143 }
144 
145 /**
146  * uvd_v5_0_hw_init - start and test UVD block
147  *
148  * @handle: handle used to pass amdgpu_device pointer
149  *
150  * Initialize the hardware, boot up the VCPU and do some testing
151  */
152 static int uvd_v5_0_hw_init(void *handle)
153 {
154 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
155 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
156 	uint32_t tmp;
157 	int r;
158 
159 	amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
160 	uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_UNGATE);
161 	uvd_v5_0_enable_mgcg(adev, true);
162 
163 	r = amdgpu_ring_test_helper(ring);
164 	if (r)
165 		goto done;
166 
167 	r = amdgpu_ring_alloc(ring, 10);
168 	if (r) {
169 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
170 		goto done;
171 	}
172 
173 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
174 	amdgpu_ring_write(ring, tmp);
175 	amdgpu_ring_write(ring, 0xFFFFF);
176 
177 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
178 	amdgpu_ring_write(ring, tmp);
179 	amdgpu_ring_write(ring, 0xFFFFF);
180 
181 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
182 	amdgpu_ring_write(ring, tmp);
183 	amdgpu_ring_write(ring, 0xFFFFF);
184 
185 	/* Clear timeout status bits */
186 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
187 	amdgpu_ring_write(ring, 0x8);
188 
189 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
190 	amdgpu_ring_write(ring, 3);
191 
192 	amdgpu_ring_commit(ring);
193 
194 done:
195 	if (!r)
196 		DRM_INFO("UVD initialized successfully.\n");
197 
198 	return r;
199 
200 }
201 
202 /**
203  * uvd_v5_0_hw_fini - stop the hardware block
204  *
205  * @handle: handle used to pass amdgpu_device pointer
206  *
207  * Stop the UVD block, mark ring as not ready any more
208  */
209 static int uvd_v5_0_hw_fini(void *handle)
210 {
211 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
212 
213 	if (RREG32(mmUVD_STATUS) != 0)
214 		uvd_v5_0_stop(adev);
215 
216 	return 0;
217 }
218 
219 static int uvd_v5_0_suspend(void *handle)
220 {
221 	int r;
222 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
223 
224 	r = uvd_v5_0_hw_fini(adev);
225 	if (r)
226 		return r;
227 	uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_GATE);
228 
229 	return amdgpu_uvd_suspend(adev);
230 }
231 
232 static int uvd_v5_0_resume(void *handle)
233 {
234 	int r;
235 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
236 
237 	r = amdgpu_uvd_resume(adev);
238 	if (r)
239 		return r;
240 
241 	return uvd_v5_0_hw_init(adev);
242 }
243 
244 /**
245  * uvd_v5_0_mc_resume - memory controller programming
246  *
247  * @adev: amdgpu_device pointer
248  *
249  * Let the UVD memory controller know it's offsets
250  */
251 static void uvd_v5_0_mc_resume(struct amdgpu_device *adev)
252 {
253 	uint64_t offset;
254 	uint32_t size;
255 
256 	/* program memory controller bits 0-27 */
257 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
258 			lower_32_bits(adev->uvd.inst->gpu_addr));
259 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
260 			upper_32_bits(adev->uvd.inst->gpu_addr));
261 
262 	offset = AMDGPU_UVD_FIRMWARE_OFFSET;
263 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev);
264 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, offset >> 3);
265 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
266 
267 	offset += size;
268 	size = AMDGPU_UVD_HEAP_SIZE;
269 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, offset >> 3);
270 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
271 
272 	offset += size;
273 	size = AMDGPU_UVD_STACK_SIZE +
274 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles);
275 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3);
276 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
277 
278 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
279 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
280 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
281 }
282 
283 /**
284  * uvd_v5_0_start - start UVD block
285  *
286  * @adev: amdgpu_device pointer
287  *
288  * Setup and start the UVD block
289  */
290 static int uvd_v5_0_start(struct amdgpu_device *adev)
291 {
292 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
293 	uint32_t rb_bufsz, tmp;
294 	uint32_t lmi_swap_cntl;
295 	uint32_t mp_swap_cntl;
296 	int i, j, r;
297 
298 	/*disable DPG */
299 	WREG32_P(mmUVD_POWER_STATUS, 0, ~(1 << 2));
300 
301 	/* disable byte swapping */
302 	lmi_swap_cntl = 0;
303 	mp_swap_cntl = 0;
304 
305 	uvd_v5_0_mc_resume(adev);
306 
307 	/* disable interupt */
308 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
309 
310 	/* stall UMC and register bus before resetting VCPU */
311 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
312 	mdelay(1);
313 
314 	/* put LMI, VCPU, RBC etc... into reset */
315 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
316 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
317 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
318 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
319 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
320 	mdelay(5);
321 
322 	/* take UVD block out of reset */
323 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
324 	mdelay(5);
325 
326 	/* initialize UVD memory controller */
327 	WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
328 			     (1 << 21) | (1 << 9) | (1 << 20));
329 
330 #ifdef __BIG_ENDIAN
331 	/* swap (8 in 32) RB and IB */
332 	lmi_swap_cntl = 0xa;
333 	mp_swap_cntl = 0;
334 #endif
335 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
336 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
337 
338 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
339 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
340 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
341 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
342 	WREG32(mmUVD_MPC_SET_ALU, 0);
343 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
344 
345 	/* take all subblocks out of reset, except VCPU */
346 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
347 	mdelay(5);
348 
349 	/* enable VCPU clock */
350 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
351 
352 	/* enable UMC */
353 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
354 
355 	/* boot up the VCPU */
356 	WREG32(mmUVD_SOFT_RESET, 0);
357 	mdelay(10);
358 
359 	for (i = 0; i < 10; ++i) {
360 		uint32_t status;
361 		for (j = 0; j < 100; ++j) {
362 			status = RREG32(mmUVD_STATUS);
363 			if (status & 2)
364 				break;
365 			mdelay(10);
366 		}
367 		r = 0;
368 		if (status & 2)
369 			break;
370 
371 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
372 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
373 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
374 		mdelay(10);
375 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
376 		mdelay(10);
377 		r = -1;
378 	}
379 
380 	if (r) {
381 		DRM_ERROR("UVD not responding, giving up!!!\n");
382 		return r;
383 	}
384 	/* enable master interrupt */
385 	WREG32_P(mmUVD_MASTINT_EN, 3 << 1, ~(3 << 1));
386 
387 	/* clear the bit 4 of UVD_STATUS */
388 	WREG32_P(mmUVD_STATUS, 0, ~(2 << 1));
389 
390 	rb_bufsz = order_base_2(ring->ring_size);
391 	tmp = 0;
392 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
393 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
394 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
395 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0);
396 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
397 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
398 	/* force RBC into idle state */
399 	WREG32(mmUVD_RBC_RB_CNTL, tmp);
400 
401 	/* set the write pointer delay */
402 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
403 
404 	/* set the wb address */
405 	WREG32(mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2));
406 
407 	/* program the RB_BASE for ring buffer */
408 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
409 			lower_32_bits(ring->gpu_addr));
410 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
411 			upper_32_bits(ring->gpu_addr));
412 
413 	/* Initialize the ring buffer's read and write pointers */
414 	WREG32(mmUVD_RBC_RB_RPTR, 0);
415 
416 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
417 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
418 
419 	WREG32_P(mmUVD_RBC_RB_CNTL, 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK);
420 
421 	return 0;
422 }
423 
424 /**
425  * uvd_v5_0_stop - stop UVD block
426  *
427  * @adev: amdgpu_device pointer
428  *
429  * stop the UVD block
430  */
431 static void uvd_v5_0_stop(struct amdgpu_device *adev)
432 {
433 	/* force RBC into idle state */
434 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
435 
436 	/* Stall UMC and register bus before resetting VCPU */
437 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
438 	mdelay(1);
439 
440 	/* put VCPU into reset */
441 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
442 	mdelay(5);
443 
444 	/* disable VCPU clock */
445 	WREG32(mmUVD_VCPU_CNTL, 0x0);
446 
447 	/* Unstall UMC and register bus */
448 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
449 
450 	WREG32(mmUVD_STATUS, 0);
451 }
452 
453 /**
454  * uvd_v5_0_ring_emit_fence - emit an fence & trap command
455  *
456  * @ring: amdgpu_ring pointer
457  * @addr: address
458  * @seq: sequence number
459  * @flags: fence related flags
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  * @job: job to retrieve vmid from
524  * @ib: indirect buffer to execute
525  * @flags: unused
526  *
527  * Write ring commands to execute the indirect buffer
528  */
529 static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
530 				  struct amdgpu_job *job,
531 				  struct amdgpu_ib *ib,
532 				  uint32_t flags)
533 {
534 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
535 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
536 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
537 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
538 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
539 	amdgpu_ring_write(ring, ib->length_dw);
540 }
541 
542 static void uvd_v5_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
543 {
544 	int i;
545 
546 	WARN_ON(ring->wptr % 2 || count % 2);
547 
548 	for (i = 0; i < count / 2; i++) {
549 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
550 		amdgpu_ring_write(ring, 0);
551 	}
552 }
553 
554 static bool uvd_v5_0_is_idle(void *handle)
555 {
556 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
557 
558 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
559 }
560 
561 static int uvd_v5_0_wait_for_idle(void *handle)
562 {
563 	unsigned i;
564 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
565 
566 	for (i = 0; i < adev->usec_timeout; i++) {
567 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
568 			return 0;
569 	}
570 	return -ETIMEDOUT;
571 }
572 
573 static int uvd_v5_0_soft_reset(void *handle)
574 {
575 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
576 
577 	uvd_v5_0_stop(adev);
578 
579 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
580 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
581 	mdelay(5);
582 
583 	return uvd_v5_0_start(adev);
584 }
585 
586 static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev,
587 					struct amdgpu_irq_src *source,
588 					unsigned type,
589 					enum amdgpu_interrupt_state state)
590 {
591 	// TODO
592 	return 0;
593 }
594 
595 static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev,
596 				      struct amdgpu_irq_src *source,
597 				      struct amdgpu_iv_entry *entry)
598 {
599 	DRM_DEBUG("IH: UVD TRAP\n");
600 	amdgpu_fence_process(&adev->uvd.inst->ring);
601 	return 0;
602 }
603 
604 static void uvd_v5_0_enable_clock_gating(struct amdgpu_device *adev, bool enable)
605 {
606 	uint32_t data1, data3, suvd_flags;
607 
608 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
609 	data3 = RREG32(mmUVD_CGC_GATE);
610 
611 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
612 		     UVD_SUVD_CGC_GATE__SIT_MASK |
613 		     UVD_SUVD_CGC_GATE__SMP_MASK |
614 		     UVD_SUVD_CGC_GATE__SCM_MASK |
615 		     UVD_SUVD_CGC_GATE__SDB_MASK;
616 
617 	if (enable) {
618 		data3 |= (UVD_CGC_GATE__SYS_MASK     |
619 			UVD_CGC_GATE__UDEC_MASK      |
620 			UVD_CGC_GATE__MPEG2_MASK     |
621 			UVD_CGC_GATE__RBC_MASK       |
622 			UVD_CGC_GATE__LMI_MC_MASK    |
623 			UVD_CGC_GATE__IDCT_MASK      |
624 			UVD_CGC_GATE__MPRD_MASK      |
625 			UVD_CGC_GATE__MPC_MASK       |
626 			UVD_CGC_GATE__LBSI_MASK      |
627 			UVD_CGC_GATE__LRBBM_MASK     |
628 			UVD_CGC_GATE__UDEC_RE_MASK   |
629 			UVD_CGC_GATE__UDEC_CM_MASK   |
630 			UVD_CGC_GATE__UDEC_IT_MASK   |
631 			UVD_CGC_GATE__UDEC_DB_MASK   |
632 			UVD_CGC_GATE__UDEC_MP_MASK   |
633 			UVD_CGC_GATE__WCB_MASK       |
634 			UVD_CGC_GATE__JPEG_MASK      |
635 			UVD_CGC_GATE__SCPU_MASK);
636 		/* only in pg enabled, we can gate clock to vcpu*/
637 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD)
638 			data3 |= UVD_CGC_GATE__VCPU_MASK;
639 		data3 &= ~UVD_CGC_GATE__REGS_MASK;
640 		data1 |= suvd_flags;
641 	} else {
642 		data3 = 0;
643 		data1 = 0;
644 	}
645 
646 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
647 	WREG32(mmUVD_CGC_GATE, data3);
648 }
649 
650 static void uvd_v5_0_set_sw_clock_gating(struct amdgpu_device *adev)
651 {
652 	uint32_t data, data2;
653 
654 	data = RREG32(mmUVD_CGC_CTRL);
655 	data2 = RREG32(mmUVD_SUVD_CGC_CTRL);
656 
657 
658 	data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
659 		  UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
660 
661 
662 	data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
663 		(1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) |
664 		(4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY));
665 
666 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
667 			UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
668 			UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
669 			UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
670 			UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
671 			UVD_CGC_CTRL__SYS_MODE_MASK |
672 			UVD_CGC_CTRL__UDEC_MODE_MASK |
673 			UVD_CGC_CTRL__MPEG2_MODE_MASK |
674 			UVD_CGC_CTRL__REGS_MODE_MASK |
675 			UVD_CGC_CTRL__RBC_MODE_MASK |
676 			UVD_CGC_CTRL__LMI_MC_MODE_MASK |
677 			UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
678 			UVD_CGC_CTRL__IDCT_MODE_MASK |
679 			UVD_CGC_CTRL__MPRD_MODE_MASK |
680 			UVD_CGC_CTRL__MPC_MODE_MASK |
681 			UVD_CGC_CTRL__LBSI_MODE_MASK |
682 			UVD_CGC_CTRL__LRBBM_MODE_MASK |
683 			UVD_CGC_CTRL__WCB_MODE_MASK |
684 			UVD_CGC_CTRL__VCPU_MODE_MASK |
685 			UVD_CGC_CTRL__JPEG_MODE_MASK |
686 			UVD_CGC_CTRL__SCPU_MODE_MASK);
687 	data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
688 			UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
689 			UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
690 			UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
691 			UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
692 
693 	WREG32(mmUVD_CGC_CTRL, data);
694 	WREG32(mmUVD_SUVD_CGC_CTRL, data2);
695 }
696 
697 #if 0
698 static void uvd_v5_0_set_hw_clock_gating(struct amdgpu_device *adev)
699 {
700 	uint32_t data, data1, cgc_flags, suvd_flags;
701 
702 	data = RREG32(mmUVD_CGC_GATE);
703 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
704 
705 	cgc_flags = UVD_CGC_GATE__SYS_MASK |
706 				UVD_CGC_GATE__UDEC_MASK |
707 				UVD_CGC_GATE__MPEG2_MASK |
708 				UVD_CGC_GATE__RBC_MASK |
709 				UVD_CGC_GATE__LMI_MC_MASK |
710 				UVD_CGC_GATE__IDCT_MASK |
711 				UVD_CGC_GATE__MPRD_MASK |
712 				UVD_CGC_GATE__MPC_MASK |
713 				UVD_CGC_GATE__LBSI_MASK |
714 				UVD_CGC_GATE__LRBBM_MASK |
715 				UVD_CGC_GATE__UDEC_RE_MASK |
716 				UVD_CGC_GATE__UDEC_CM_MASK |
717 				UVD_CGC_GATE__UDEC_IT_MASK |
718 				UVD_CGC_GATE__UDEC_DB_MASK |
719 				UVD_CGC_GATE__UDEC_MP_MASK |
720 				UVD_CGC_GATE__WCB_MASK |
721 				UVD_CGC_GATE__VCPU_MASK |
722 				UVD_CGC_GATE__SCPU_MASK;
723 
724 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
725 				UVD_SUVD_CGC_GATE__SIT_MASK |
726 				UVD_SUVD_CGC_GATE__SMP_MASK |
727 				UVD_SUVD_CGC_GATE__SCM_MASK |
728 				UVD_SUVD_CGC_GATE__SDB_MASK;
729 
730 	data |= cgc_flags;
731 	data1 |= suvd_flags;
732 
733 	WREG32(mmUVD_CGC_GATE, data);
734 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
735 }
736 #endif
737 
738 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev,
739 				 bool enable)
740 {
741 	u32 orig, data;
742 
743 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
744 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
745 		data |= 0xfff;
746 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
747 
748 		orig = data = RREG32(mmUVD_CGC_CTRL);
749 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
750 		if (orig != data)
751 			WREG32(mmUVD_CGC_CTRL, data);
752 	} else {
753 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
754 		data &= ~0xfff;
755 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
756 
757 		orig = data = RREG32(mmUVD_CGC_CTRL);
758 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
759 		if (orig != data)
760 			WREG32(mmUVD_CGC_CTRL, data);
761 	}
762 }
763 
764 static int uvd_v5_0_set_clockgating_state(void *handle,
765 					  enum amd_clockgating_state state)
766 {
767 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
768 	bool enable = (state == AMD_CG_STATE_GATE);
769 
770 	if (enable) {
771 		/* wait for STATUS to clear */
772 		if (uvd_v5_0_wait_for_idle(handle))
773 			return -EBUSY;
774 		uvd_v5_0_enable_clock_gating(adev, true);
775 
776 		/* enable HW gates because UVD is idle */
777 /*		uvd_v5_0_set_hw_clock_gating(adev); */
778 	} else {
779 		uvd_v5_0_enable_clock_gating(adev, false);
780 	}
781 
782 	uvd_v5_0_set_sw_clock_gating(adev);
783 	return 0;
784 }
785 
786 static int uvd_v5_0_set_powergating_state(void *handle,
787 					  enum amd_powergating_state state)
788 {
789 	/* This doesn't actually powergate the UVD block.
790 	 * That's done in the dpm code via the SMC.  This
791 	 * just re-inits the block as necessary.  The actual
792 	 * gating still happens in the dpm code.  We should
793 	 * revisit this when there is a cleaner line between
794 	 * the smc and the hw blocks
795 	 */
796 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
797 	int ret = 0;
798 
799 	if (state == AMD_PG_STATE_GATE) {
800 		uvd_v5_0_stop(adev);
801 	} else {
802 		ret = uvd_v5_0_start(adev);
803 		if (ret)
804 			goto out;
805 	}
806 
807 out:
808 	return ret;
809 }
810 
811 static void uvd_v5_0_get_clockgating_state(void *handle, u32 *flags)
812 {
813 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
814 	int data;
815 
816 	mutex_lock(&adev->pm.mutex);
817 
818 	if (RREG32_SMC(ixCURRENT_PG_STATUS) &
819 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
820 		DRM_INFO("Cannot get clockgating state when UVD is powergated.\n");
821 		goto out;
822 	}
823 
824 	/* AMD_CG_SUPPORT_UVD_MGCG */
825 	data = RREG32(mmUVD_CGC_CTRL);
826 	if (data & UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK)
827 		*flags |= AMD_CG_SUPPORT_UVD_MGCG;
828 
829 out:
830 	mutex_unlock(&adev->pm.mutex);
831 }
832 
833 static const struct amd_ip_funcs uvd_v5_0_ip_funcs = {
834 	.name = "uvd_v5_0",
835 	.early_init = uvd_v5_0_early_init,
836 	.late_init = NULL,
837 	.sw_init = uvd_v5_0_sw_init,
838 	.sw_fini = uvd_v5_0_sw_fini,
839 	.hw_init = uvd_v5_0_hw_init,
840 	.hw_fini = uvd_v5_0_hw_fini,
841 	.suspend = uvd_v5_0_suspend,
842 	.resume = uvd_v5_0_resume,
843 	.is_idle = uvd_v5_0_is_idle,
844 	.wait_for_idle = uvd_v5_0_wait_for_idle,
845 	.soft_reset = uvd_v5_0_soft_reset,
846 	.set_clockgating_state = uvd_v5_0_set_clockgating_state,
847 	.set_powergating_state = uvd_v5_0_set_powergating_state,
848 	.get_clockgating_state = uvd_v5_0_get_clockgating_state,
849 };
850 
851 static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
852 	.type = AMDGPU_RING_TYPE_UVD,
853 	.align_mask = 0xf,
854 	.support_64bit_ptrs = false,
855 	.no_user_fence = true,
856 	.get_rptr = uvd_v5_0_ring_get_rptr,
857 	.get_wptr = uvd_v5_0_ring_get_wptr,
858 	.set_wptr = uvd_v5_0_ring_set_wptr,
859 	.parse_cs = amdgpu_uvd_ring_parse_cs,
860 	.emit_frame_size =
861 		14, /* uvd_v5_0_ring_emit_fence  x1 no user fence */
862 	.emit_ib_size = 6, /* uvd_v5_0_ring_emit_ib */
863 	.emit_ib = uvd_v5_0_ring_emit_ib,
864 	.emit_fence = uvd_v5_0_ring_emit_fence,
865 	.test_ring = uvd_v5_0_ring_test_ring,
866 	.test_ib = amdgpu_uvd_ring_test_ib,
867 	.insert_nop = uvd_v5_0_ring_insert_nop,
868 	.pad_ib = amdgpu_ring_generic_pad_ib,
869 	.begin_use = amdgpu_uvd_ring_begin_use,
870 	.end_use = amdgpu_uvd_ring_end_use,
871 };
872 
873 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev)
874 {
875 	adev->uvd.inst->ring.funcs = &uvd_v5_0_ring_funcs;
876 }
877 
878 static const struct amdgpu_irq_src_funcs uvd_v5_0_irq_funcs = {
879 	.set = uvd_v5_0_set_interrupt_state,
880 	.process = uvd_v5_0_process_interrupt,
881 };
882 
883 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev)
884 {
885 	adev->uvd.inst->irq.num_types = 1;
886 	adev->uvd.inst->irq.funcs = &uvd_v5_0_irq_funcs;
887 }
888 
889 const struct amdgpu_ip_block_version uvd_v5_0_ip_block =
890 {
891 		.type = AMD_IP_BLOCK_TYPE_UVD,
892 		.major = 5,
893 		.minor = 0,
894 		.rev = 0,
895 		.funcs = &uvd_v5_0_ip_funcs,
896 };
897