xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c (revision 9a29f5fc)
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, NULL);
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 	cancel_delayed_work_sync(&adev->uvd.idle_work);
214 
215 	if (RREG32(mmUVD_STATUS) != 0)
216 		uvd_v5_0_stop(adev);
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 	/*
227 	 * Proper cleanups before halting the HW engine:
228 	 *   - cancel the delayed idle work
229 	 *   - enable powergating
230 	 *   - enable clockgating
231 	 *   - disable dpm
232 	 *
233 	 * TODO: to align with the VCN implementation, move the
234 	 * jobs for clockgating/powergating/dpm setting to
235 	 * ->set_powergating_state().
236 	 */
237 	cancel_delayed_work_sync(&adev->uvd.idle_work);
238 
239 	if (adev->pm.dpm_enabled) {
240 		amdgpu_dpm_enable_uvd(adev, false);
241 	} else {
242 		amdgpu_asic_set_uvd_clocks(adev, 0, 0);
243 		/* shutdown the UVD block */
244 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
245 						       AMD_PG_STATE_GATE);
246 		amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
247 						       AMD_CG_STATE_GATE);
248 	}
249 
250 	r = uvd_v5_0_hw_fini(adev);
251 	if (r)
252 		return r;
253 
254 	return amdgpu_uvd_suspend(adev);
255 }
256 
257 static int uvd_v5_0_resume(void *handle)
258 {
259 	int r;
260 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
261 
262 	r = amdgpu_uvd_resume(adev);
263 	if (r)
264 		return r;
265 
266 	return uvd_v5_0_hw_init(adev);
267 }
268 
269 /**
270  * uvd_v5_0_mc_resume - memory controller programming
271  *
272  * @adev: amdgpu_device pointer
273  *
274  * Let the UVD memory controller know it's offsets
275  */
276 static void uvd_v5_0_mc_resume(struct amdgpu_device *adev)
277 {
278 	uint64_t offset;
279 	uint32_t size;
280 
281 	/* program memory controller bits 0-27 */
282 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
283 			lower_32_bits(adev->uvd.inst->gpu_addr));
284 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
285 			upper_32_bits(adev->uvd.inst->gpu_addr));
286 
287 	offset = AMDGPU_UVD_FIRMWARE_OFFSET;
288 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev);
289 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, offset >> 3);
290 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
291 
292 	offset += size;
293 	size = AMDGPU_UVD_HEAP_SIZE;
294 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, offset >> 3);
295 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
296 
297 	offset += size;
298 	size = AMDGPU_UVD_STACK_SIZE +
299 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles);
300 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3);
301 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
302 
303 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
304 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
305 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
306 }
307 
308 /**
309  * uvd_v5_0_start - start UVD block
310  *
311  * @adev: amdgpu_device pointer
312  *
313  * Setup and start the UVD block
314  */
315 static int uvd_v5_0_start(struct amdgpu_device *adev)
316 {
317 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
318 	uint32_t rb_bufsz, tmp;
319 	uint32_t lmi_swap_cntl;
320 	uint32_t mp_swap_cntl;
321 	int i, j, r;
322 
323 	/*disable DPG */
324 	WREG32_P(mmUVD_POWER_STATUS, 0, ~(1 << 2));
325 
326 	/* disable byte swapping */
327 	lmi_swap_cntl = 0;
328 	mp_swap_cntl = 0;
329 
330 	uvd_v5_0_mc_resume(adev);
331 
332 	/* disable interupt */
333 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
334 
335 	/* stall UMC and register bus before resetting VCPU */
336 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
337 	mdelay(1);
338 
339 	/* put LMI, VCPU, RBC etc... into reset */
340 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
341 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
342 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
343 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
344 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
345 	mdelay(5);
346 
347 	/* take UVD block out of reset */
348 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
349 	mdelay(5);
350 
351 	/* initialize UVD memory controller */
352 	WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
353 			     (1 << 21) | (1 << 9) | (1 << 20));
354 
355 #ifdef __BIG_ENDIAN
356 	/* swap (8 in 32) RB and IB */
357 	lmi_swap_cntl = 0xa;
358 	mp_swap_cntl = 0;
359 #endif
360 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
361 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
362 
363 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
364 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
365 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
366 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
367 	WREG32(mmUVD_MPC_SET_ALU, 0);
368 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
369 
370 	/* take all subblocks out of reset, except VCPU */
371 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
372 	mdelay(5);
373 
374 	/* enable VCPU clock */
375 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
376 
377 	/* enable UMC */
378 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
379 
380 	/* boot up the VCPU */
381 	WREG32(mmUVD_SOFT_RESET, 0);
382 	mdelay(10);
383 
384 	for (i = 0; i < 10; ++i) {
385 		uint32_t status;
386 		for (j = 0; j < 100; ++j) {
387 			status = RREG32(mmUVD_STATUS);
388 			if (status & 2)
389 				break;
390 			mdelay(10);
391 		}
392 		r = 0;
393 		if (status & 2)
394 			break;
395 
396 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
397 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
398 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
399 		mdelay(10);
400 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
401 		mdelay(10);
402 		r = -1;
403 	}
404 
405 	if (r) {
406 		DRM_ERROR("UVD not responding, giving up!!!\n");
407 		return r;
408 	}
409 	/* enable master interrupt */
410 	WREG32_P(mmUVD_MASTINT_EN, 3 << 1, ~(3 << 1));
411 
412 	/* clear the bit 4 of UVD_STATUS */
413 	WREG32_P(mmUVD_STATUS, 0, ~(2 << 1));
414 
415 	rb_bufsz = order_base_2(ring->ring_size);
416 	tmp = 0;
417 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
418 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
419 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
420 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0);
421 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
422 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
423 	/* force RBC into idle state */
424 	WREG32(mmUVD_RBC_RB_CNTL, tmp);
425 
426 	/* set the write pointer delay */
427 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
428 
429 	/* set the wb address */
430 	WREG32(mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2));
431 
432 	/* program the RB_BASE for ring buffer */
433 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
434 			lower_32_bits(ring->gpu_addr));
435 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
436 			upper_32_bits(ring->gpu_addr));
437 
438 	/* Initialize the ring buffer's read and write pointers */
439 	WREG32(mmUVD_RBC_RB_RPTR, 0);
440 
441 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
442 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
443 
444 	WREG32_P(mmUVD_RBC_RB_CNTL, 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK);
445 
446 	return 0;
447 }
448 
449 /**
450  * uvd_v5_0_stop - stop UVD block
451  *
452  * @adev: amdgpu_device pointer
453  *
454  * stop the UVD block
455  */
456 static void uvd_v5_0_stop(struct amdgpu_device *adev)
457 {
458 	/* force RBC into idle state */
459 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
460 
461 	/* Stall UMC and register bus before resetting VCPU */
462 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
463 	mdelay(1);
464 
465 	/* put VCPU into reset */
466 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
467 	mdelay(5);
468 
469 	/* disable VCPU clock */
470 	WREG32(mmUVD_VCPU_CNTL, 0x0);
471 
472 	/* Unstall UMC and register bus */
473 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
474 
475 	WREG32(mmUVD_STATUS, 0);
476 }
477 
478 /**
479  * uvd_v5_0_ring_emit_fence - emit an fence & trap command
480  *
481  * @ring: amdgpu_ring pointer
482  * @addr: address
483  * @seq: sequence number
484  * @flags: fence related flags
485  *
486  * Write a fence and a trap command to the ring.
487  */
488 static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
489 				     unsigned flags)
490 {
491 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
492 
493 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
494 	amdgpu_ring_write(ring, seq);
495 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
496 	amdgpu_ring_write(ring, addr & 0xffffffff);
497 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
498 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
499 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
500 	amdgpu_ring_write(ring, 0);
501 
502 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
503 	amdgpu_ring_write(ring, 0);
504 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
505 	amdgpu_ring_write(ring, 0);
506 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
507 	amdgpu_ring_write(ring, 2);
508 }
509 
510 /**
511  * uvd_v5_0_ring_test_ring - register write test
512  *
513  * @ring: amdgpu_ring pointer
514  *
515  * Test if we can successfully write to the context register
516  */
517 static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring)
518 {
519 	struct amdgpu_device *adev = ring->adev;
520 	uint32_t tmp = 0;
521 	unsigned i;
522 	int r;
523 
524 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
525 	r = amdgpu_ring_alloc(ring, 3);
526 	if (r)
527 		return r;
528 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
529 	amdgpu_ring_write(ring, 0xDEADBEEF);
530 	amdgpu_ring_commit(ring);
531 	for (i = 0; i < adev->usec_timeout; i++) {
532 		tmp = RREG32(mmUVD_CONTEXT_ID);
533 		if (tmp == 0xDEADBEEF)
534 			break;
535 		udelay(1);
536 	}
537 
538 	if (i >= adev->usec_timeout)
539 		r = -ETIMEDOUT;
540 
541 	return r;
542 }
543 
544 /**
545  * uvd_v5_0_ring_emit_ib - execute indirect buffer
546  *
547  * @ring: amdgpu_ring pointer
548  * @job: job to retrieve vmid from
549  * @ib: indirect buffer to execute
550  * @flags: unused
551  *
552  * Write ring commands to execute the indirect buffer
553  */
554 static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
555 				  struct amdgpu_job *job,
556 				  struct amdgpu_ib *ib,
557 				  uint32_t flags)
558 {
559 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
560 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
561 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
562 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
563 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
564 	amdgpu_ring_write(ring, ib->length_dw);
565 }
566 
567 static void uvd_v5_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
568 {
569 	int i;
570 
571 	WARN_ON(ring->wptr % 2 || count % 2);
572 
573 	for (i = 0; i < count / 2; i++) {
574 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
575 		amdgpu_ring_write(ring, 0);
576 	}
577 }
578 
579 static bool uvd_v5_0_is_idle(void *handle)
580 {
581 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
582 
583 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
584 }
585 
586 static int uvd_v5_0_wait_for_idle(void *handle)
587 {
588 	unsigned i;
589 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
590 
591 	for (i = 0; i < adev->usec_timeout; i++) {
592 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
593 			return 0;
594 	}
595 	return -ETIMEDOUT;
596 }
597 
598 static int uvd_v5_0_soft_reset(void *handle)
599 {
600 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
601 
602 	uvd_v5_0_stop(adev);
603 
604 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
605 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
606 	mdelay(5);
607 
608 	return uvd_v5_0_start(adev);
609 }
610 
611 static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev,
612 					struct amdgpu_irq_src *source,
613 					unsigned type,
614 					enum amdgpu_interrupt_state state)
615 {
616 	// TODO
617 	return 0;
618 }
619 
620 static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev,
621 				      struct amdgpu_irq_src *source,
622 				      struct amdgpu_iv_entry *entry)
623 {
624 	DRM_DEBUG("IH: UVD TRAP\n");
625 	amdgpu_fence_process(&adev->uvd.inst->ring);
626 	return 0;
627 }
628 
629 static void uvd_v5_0_enable_clock_gating(struct amdgpu_device *adev, bool enable)
630 {
631 	uint32_t data1, data3, suvd_flags;
632 
633 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
634 	data3 = RREG32(mmUVD_CGC_GATE);
635 
636 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
637 		     UVD_SUVD_CGC_GATE__SIT_MASK |
638 		     UVD_SUVD_CGC_GATE__SMP_MASK |
639 		     UVD_SUVD_CGC_GATE__SCM_MASK |
640 		     UVD_SUVD_CGC_GATE__SDB_MASK;
641 
642 	if (enable) {
643 		data3 |= (UVD_CGC_GATE__SYS_MASK     |
644 			UVD_CGC_GATE__UDEC_MASK      |
645 			UVD_CGC_GATE__MPEG2_MASK     |
646 			UVD_CGC_GATE__RBC_MASK       |
647 			UVD_CGC_GATE__LMI_MC_MASK    |
648 			UVD_CGC_GATE__IDCT_MASK      |
649 			UVD_CGC_GATE__MPRD_MASK      |
650 			UVD_CGC_GATE__MPC_MASK       |
651 			UVD_CGC_GATE__LBSI_MASK      |
652 			UVD_CGC_GATE__LRBBM_MASK     |
653 			UVD_CGC_GATE__UDEC_RE_MASK   |
654 			UVD_CGC_GATE__UDEC_CM_MASK   |
655 			UVD_CGC_GATE__UDEC_IT_MASK   |
656 			UVD_CGC_GATE__UDEC_DB_MASK   |
657 			UVD_CGC_GATE__UDEC_MP_MASK   |
658 			UVD_CGC_GATE__WCB_MASK       |
659 			UVD_CGC_GATE__JPEG_MASK      |
660 			UVD_CGC_GATE__SCPU_MASK);
661 		/* only in pg enabled, we can gate clock to vcpu*/
662 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD)
663 			data3 |= UVD_CGC_GATE__VCPU_MASK;
664 		data3 &= ~UVD_CGC_GATE__REGS_MASK;
665 		data1 |= suvd_flags;
666 	} else {
667 		data3 = 0;
668 		data1 = 0;
669 	}
670 
671 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
672 	WREG32(mmUVD_CGC_GATE, data3);
673 }
674 
675 static void uvd_v5_0_set_sw_clock_gating(struct amdgpu_device *adev)
676 {
677 	uint32_t data, data2;
678 
679 	data = RREG32(mmUVD_CGC_CTRL);
680 	data2 = RREG32(mmUVD_SUVD_CGC_CTRL);
681 
682 
683 	data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
684 		  UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
685 
686 
687 	data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
688 		(1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) |
689 		(4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY));
690 
691 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
692 			UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
693 			UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
694 			UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
695 			UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
696 			UVD_CGC_CTRL__SYS_MODE_MASK |
697 			UVD_CGC_CTRL__UDEC_MODE_MASK |
698 			UVD_CGC_CTRL__MPEG2_MODE_MASK |
699 			UVD_CGC_CTRL__REGS_MODE_MASK |
700 			UVD_CGC_CTRL__RBC_MODE_MASK |
701 			UVD_CGC_CTRL__LMI_MC_MODE_MASK |
702 			UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
703 			UVD_CGC_CTRL__IDCT_MODE_MASK |
704 			UVD_CGC_CTRL__MPRD_MODE_MASK |
705 			UVD_CGC_CTRL__MPC_MODE_MASK |
706 			UVD_CGC_CTRL__LBSI_MODE_MASK |
707 			UVD_CGC_CTRL__LRBBM_MODE_MASK |
708 			UVD_CGC_CTRL__WCB_MODE_MASK |
709 			UVD_CGC_CTRL__VCPU_MODE_MASK |
710 			UVD_CGC_CTRL__JPEG_MODE_MASK |
711 			UVD_CGC_CTRL__SCPU_MODE_MASK);
712 	data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
713 			UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
714 			UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
715 			UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
716 			UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
717 
718 	WREG32(mmUVD_CGC_CTRL, data);
719 	WREG32(mmUVD_SUVD_CGC_CTRL, data2);
720 }
721 
722 #if 0
723 static void uvd_v5_0_set_hw_clock_gating(struct amdgpu_device *adev)
724 {
725 	uint32_t data, data1, cgc_flags, suvd_flags;
726 
727 	data = RREG32(mmUVD_CGC_GATE);
728 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
729 
730 	cgc_flags = UVD_CGC_GATE__SYS_MASK |
731 				UVD_CGC_GATE__UDEC_MASK |
732 				UVD_CGC_GATE__MPEG2_MASK |
733 				UVD_CGC_GATE__RBC_MASK |
734 				UVD_CGC_GATE__LMI_MC_MASK |
735 				UVD_CGC_GATE__IDCT_MASK |
736 				UVD_CGC_GATE__MPRD_MASK |
737 				UVD_CGC_GATE__MPC_MASK |
738 				UVD_CGC_GATE__LBSI_MASK |
739 				UVD_CGC_GATE__LRBBM_MASK |
740 				UVD_CGC_GATE__UDEC_RE_MASK |
741 				UVD_CGC_GATE__UDEC_CM_MASK |
742 				UVD_CGC_GATE__UDEC_IT_MASK |
743 				UVD_CGC_GATE__UDEC_DB_MASK |
744 				UVD_CGC_GATE__UDEC_MP_MASK |
745 				UVD_CGC_GATE__WCB_MASK |
746 				UVD_CGC_GATE__VCPU_MASK |
747 				UVD_CGC_GATE__SCPU_MASK;
748 
749 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
750 				UVD_SUVD_CGC_GATE__SIT_MASK |
751 				UVD_SUVD_CGC_GATE__SMP_MASK |
752 				UVD_SUVD_CGC_GATE__SCM_MASK |
753 				UVD_SUVD_CGC_GATE__SDB_MASK;
754 
755 	data |= cgc_flags;
756 	data1 |= suvd_flags;
757 
758 	WREG32(mmUVD_CGC_GATE, data);
759 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
760 }
761 #endif
762 
763 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev,
764 				 bool enable)
765 {
766 	u32 orig, data;
767 
768 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
769 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
770 		data |= 0xfff;
771 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
772 
773 		orig = data = RREG32(mmUVD_CGC_CTRL);
774 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
775 		if (orig != data)
776 			WREG32(mmUVD_CGC_CTRL, data);
777 	} else {
778 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
779 		data &= ~0xfff;
780 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
781 
782 		orig = data = RREG32(mmUVD_CGC_CTRL);
783 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
784 		if (orig != data)
785 			WREG32(mmUVD_CGC_CTRL, data);
786 	}
787 }
788 
789 static int uvd_v5_0_set_clockgating_state(void *handle,
790 					  enum amd_clockgating_state state)
791 {
792 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
793 	bool enable = (state == AMD_CG_STATE_GATE);
794 
795 	if (enable) {
796 		/* wait for STATUS to clear */
797 		if (uvd_v5_0_wait_for_idle(handle))
798 			return -EBUSY;
799 		uvd_v5_0_enable_clock_gating(adev, true);
800 
801 		/* enable HW gates because UVD is idle */
802 /*		uvd_v5_0_set_hw_clock_gating(adev); */
803 	} else {
804 		uvd_v5_0_enable_clock_gating(adev, false);
805 	}
806 
807 	uvd_v5_0_set_sw_clock_gating(adev);
808 	return 0;
809 }
810 
811 static int uvd_v5_0_set_powergating_state(void *handle,
812 					  enum amd_powergating_state state)
813 {
814 	/* This doesn't actually powergate the UVD block.
815 	 * That's done in the dpm code via the SMC.  This
816 	 * just re-inits the block as necessary.  The actual
817 	 * gating still happens in the dpm code.  We should
818 	 * revisit this when there is a cleaner line between
819 	 * the smc and the hw blocks
820 	 */
821 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
822 	int ret = 0;
823 
824 	if (state == AMD_PG_STATE_GATE) {
825 		uvd_v5_0_stop(adev);
826 	} else {
827 		ret = uvd_v5_0_start(adev);
828 		if (ret)
829 			goto out;
830 	}
831 
832 out:
833 	return ret;
834 }
835 
836 static void uvd_v5_0_get_clockgating_state(void *handle, u64 *flags)
837 {
838 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
839 	int data;
840 
841 	mutex_lock(&adev->pm.mutex);
842 
843 	if (RREG32_SMC(ixCURRENT_PG_STATUS) &
844 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
845 		DRM_INFO("Cannot get clockgating state when UVD is powergated.\n");
846 		goto out;
847 	}
848 
849 	/* AMD_CG_SUPPORT_UVD_MGCG */
850 	data = RREG32(mmUVD_CGC_CTRL);
851 	if (data & UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK)
852 		*flags |= AMD_CG_SUPPORT_UVD_MGCG;
853 
854 out:
855 	mutex_unlock(&adev->pm.mutex);
856 }
857 
858 static const struct amd_ip_funcs uvd_v5_0_ip_funcs = {
859 	.name = "uvd_v5_0",
860 	.early_init = uvd_v5_0_early_init,
861 	.late_init = NULL,
862 	.sw_init = uvd_v5_0_sw_init,
863 	.sw_fini = uvd_v5_0_sw_fini,
864 	.hw_init = uvd_v5_0_hw_init,
865 	.hw_fini = uvd_v5_0_hw_fini,
866 	.suspend = uvd_v5_0_suspend,
867 	.resume = uvd_v5_0_resume,
868 	.is_idle = uvd_v5_0_is_idle,
869 	.wait_for_idle = uvd_v5_0_wait_for_idle,
870 	.soft_reset = uvd_v5_0_soft_reset,
871 	.set_clockgating_state = uvd_v5_0_set_clockgating_state,
872 	.set_powergating_state = uvd_v5_0_set_powergating_state,
873 	.get_clockgating_state = uvd_v5_0_get_clockgating_state,
874 };
875 
876 static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
877 	.type = AMDGPU_RING_TYPE_UVD,
878 	.align_mask = 0xf,
879 	.support_64bit_ptrs = false,
880 	.no_user_fence = true,
881 	.get_rptr = uvd_v5_0_ring_get_rptr,
882 	.get_wptr = uvd_v5_0_ring_get_wptr,
883 	.set_wptr = uvd_v5_0_ring_set_wptr,
884 	.parse_cs = amdgpu_uvd_ring_parse_cs,
885 	.emit_frame_size =
886 		14, /* uvd_v5_0_ring_emit_fence  x1 no user fence */
887 	.emit_ib_size = 6, /* uvd_v5_0_ring_emit_ib */
888 	.emit_ib = uvd_v5_0_ring_emit_ib,
889 	.emit_fence = uvd_v5_0_ring_emit_fence,
890 	.test_ring = uvd_v5_0_ring_test_ring,
891 	.test_ib = amdgpu_uvd_ring_test_ib,
892 	.insert_nop = uvd_v5_0_ring_insert_nop,
893 	.pad_ib = amdgpu_ring_generic_pad_ib,
894 	.begin_use = amdgpu_uvd_ring_begin_use,
895 	.end_use = amdgpu_uvd_ring_end_use,
896 };
897 
898 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev)
899 {
900 	adev->uvd.inst->ring.funcs = &uvd_v5_0_ring_funcs;
901 }
902 
903 static const struct amdgpu_irq_src_funcs uvd_v5_0_irq_funcs = {
904 	.set = uvd_v5_0_set_interrupt_state,
905 	.process = uvd_v5_0_process_interrupt,
906 };
907 
908 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev)
909 {
910 	adev->uvd.inst->irq.num_types = 1;
911 	adev->uvd.inst->irq.funcs = &uvd_v5_0_irq_funcs;
912 }
913 
914 const struct amdgpu_ip_block_version uvd_v5_0_ip_block =
915 {
916 		.type = AMD_IP_BLOCK_TYPE_UVD,
917 		.major = 5,
918 		.minor = 0,
919 		.rev = 0,
920 		.funcs = &uvd_v5_0_ip_funcs,
921 };
922