xref: /linux/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c (revision 4b132aacb0768ac1e652cf517097ea6f237214b9)
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 	return r;
129 }
130 
131 static int uvd_v5_0_sw_fini(void *handle)
132 {
133 	int r;
134 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
135 
136 	r = amdgpu_uvd_suspend(adev);
137 	if (r)
138 		return r;
139 
140 	return amdgpu_uvd_sw_fini(adev);
141 }
142 
143 /**
144  * uvd_v5_0_hw_init - start and test UVD block
145  *
146  * @handle: handle used to pass amdgpu_device pointer
147  *
148  * Initialize the hardware, boot up the VCPU and do some testing
149  */
150 static int uvd_v5_0_hw_init(void *handle)
151 {
152 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
153 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
154 	uint32_t tmp;
155 	int r;
156 
157 	amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
158 	uvd_v5_0_set_clockgating_state(adev, AMD_CG_STATE_UNGATE);
159 	uvd_v5_0_enable_mgcg(adev, true);
160 
161 	r = amdgpu_ring_test_helper(ring);
162 	if (r)
163 		goto done;
164 
165 	r = amdgpu_ring_alloc(ring, 10);
166 	if (r) {
167 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
168 		goto done;
169 	}
170 
171 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
172 	amdgpu_ring_write(ring, tmp);
173 	amdgpu_ring_write(ring, 0xFFFFF);
174 
175 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
176 	amdgpu_ring_write(ring, tmp);
177 	amdgpu_ring_write(ring, 0xFFFFF);
178 
179 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
180 	amdgpu_ring_write(ring, tmp);
181 	amdgpu_ring_write(ring, 0xFFFFF);
182 
183 	/* Clear timeout status bits */
184 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
185 	amdgpu_ring_write(ring, 0x8);
186 
187 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
188 	amdgpu_ring_write(ring, 3);
189 
190 	amdgpu_ring_commit(ring);
191 
192 done:
193 	if (!r)
194 		DRM_INFO("UVD initialized successfully.\n");
195 
196 	return r;
197 
198 }
199 
200 /**
201  * uvd_v5_0_hw_fini - stop the hardware block
202  *
203  * @handle: handle used to pass amdgpu_device pointer
204  *
205  * Stop the UVD block, mark ring as not ready any more
206  */
207 static int uvd_v5_0_hw_fini(void *handle)
208 {
209 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
210 
211 	cancel_delayed_work_sync(&adev->uvd.idle_work);
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_prepare_suspend(void *handle)
220 {
221 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
222 
223 	return amdgpu_uvd_prepare_suspend(adev);
224 }
225 
226 static int uvd_v5_0_suspend(void *handle)
227 {
228 	int r;
229 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
230 
231 	/*
232 	 * Proper cleanups before halting the HW engine:
233 	 *   - cancel the delayed idle work
234 	 *   - enable powergating
235 	 *   - enable clockgating
236 	 *   - disable dpm
237 	 *
238 	 * TODO: to align with the VCN implementation, move the
239 	 * jobs for clockgating/powergating/dpm setting to
240 	 * ->set_powergating_state().
241 	 */
242 	cancel_delayed_work_sync(&adev->uvd.idle_work);
243 
244 	if (adev->pm.dpm_enabled) {
245 		amdgpu_dpm_enable_uvd(adev, false);
246 	} else {
247 		amdgpu_asic_set_uvd_clocks(adev, 0, 0);
248 		/* shutdown the UVD block */
249 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
250 						       AMD_PG_STATE_GATE);
251 		amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
252 						       AMD_CG_STATE_GATE);
253 	}
254 
255 	r = uvd_v5_0_hw_fini(adev);
256 	if (r)
257 		return r;
258 
259 	return amdgpu_uvd_suspend(adev);
260 }
261 
262 static int uvd_v5_0_resume(void *handle)
263 {
264 	int r;
265 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
266 
267 	r = amdgpu_uvd_resume(adev);
268 	if (r)
269 		return r;
270 
271 	return uvd_v5_0_hw_init(adev);
272 }
273 
274 /**
275  * uvd_v5_0_mc_resume - memory controller programming
276  *
277  * @adev: amdgpu_device pointer
278  *
279  * Let the UVD memory controller know it's offsets
280  */
281 static void uvd_v5_0_mc_resume(struct amdgpu_device *adev)
282 {
283 	uint64_t offset;
284 	uint32_t size;
285 
286 	/* program memory controller bits 0-27 */
287 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
288 			lower_32_bits(adev->uvd.inst->gpu_addr));
289 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
290 			upper_32_bits(adev->uvd.inst->gpu_addr));
291 
292 	offset = AMDGPU_UVD_FIRMWARE_OFFSET;
293 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev);
294 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, offset >> 3);
295 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
296 
297 	offset += size;
298 	size = AMDGPU_UVD_HEAP_SIZE;
299 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, offset >> 3);
300 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
301 
302 	offset += size;
303 	size = AMDGPU_UVD_STACK_SIZE +
304 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles);
305 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3);
306 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
307 
308 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
309 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
310 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
311 }
312 
313 /**
314  * uvd_v5_0_start - start UVD block
315  *
316  * @adev: amdgpu_device pointer
317  *
318  * Setup and start the UVD block
319  */
320 static int uvd_v5_0_start(struct amdgpu_device *adev)
321 {
322 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
323 	uint32_t rb_bufsz, tmp;
324 	uint32_t lmi_swap_cntl;
325 	uint32_t mp_swap_cntl;
326 	int i, j, r;
327 
328 	/*disable DPG */
329 	WREG32_P(mmUVD_POWER_STATUS, 0, ~(1 << 2));
330 
331 	/* disable byte swapping */
332 	lmi_swap_cntl = 0;
333 	mp_swap_cntl = 0;
334 
335 	uvd_v5_0_mc_resume(adev);
336 
337 	/* disable interupt */
338 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
339 
340 	/* stall UMC and register bus before resetting VCPU */
341 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
342 	mdelay(1);
343 
344 	/* put LMI, VCPU, RBC etc... into reset */
345 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
346 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
347 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
348 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
349 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
350 	mdelay(5);
351 
352 	/* take UVD block out of reset */
353 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
354 	mdelay(5);
355 
356 	/* initialize UVD memory controller */
357 	WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
358 			     (1 << 21) | (1 << 9) | (1 << 20));
359 
360 #ifdef __BIG_ENDIAN
361 	/* swap (8 in 32) RB and IB */
362 	lmi_swap_cntl = 0xa;
363 	mp_swap_cntl = 0;
364 #endif
365 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
366 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
367 
368 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
369 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
370 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
371 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
372 	WREG32(mmUVD_MPC_SET_ALU, 0);
373 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
374 
375 	/* take all subblocks out of reset, except VCPU */
376 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
377 	mdelay(5);
378 
379 	/* enable VCPU clock */
380 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
381 
382 	/* enable UMC */
383 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
384 
385 	/* boot up the VCPU */
386 	WREG32(mmUVD_SOFT_RESET, 0);
387 	mdelay(10);
388 
389 	for (i = 0; i < 10; ++i) {
390 		uint32_t status;
391 		for (j = 0; j < 100; ++j) {
392 			status = RREG32(mmUVD_STATUS);
393 			if (status & 2)
394 				break;
395 			mdelay(10);
396 		}
397 		r = 0;
398 		if (status & 2)
399 			break;
400 
401 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
402 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
403 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
404 		mdelay(10);
405 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
406 		mdelay(10);
407 		r = -1;
408 	}
409 
410 	if (r) {
411 		DRM_ERROR("UVD not responding, giving up!!!\n");
412 		return r;
413 	}
414 	/* enable master interrupt */
415 	WREG32_P(mmUVD_MASTINT_EN, 3 << 1, ~(3 << 1));
416 
417 	/* clear the bit 4 of UVD_STATUS */
418 	WREG32_P(mmUVD_STATUS, 0, ~(2 << 1));
419 
420 	rb_bufsz = order_base_2(ring->ring_size);
421 	tmp = 0;
422 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
423 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
424 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
425 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0);
426 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
427 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
428 	/* force RBC into idle state */
429 	WREG32(mmUVD_RBC_RB_CNTL, tmp);
430 
431 	/* set the write pointer delay */
432 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
433 
434 	/* set the wb address */
435 	WREG32(mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2));
436 
437 	/* program the RB_BASE for ring buffer */
438 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
439 			lower_32_bits(ring->gpu_addr));
440 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
441 			upper_32_bits(ring->gpu_addr));
442 
443 	/* Initialize the ring buffer's read and write pointers */
444 	WREG32(mmUVD_RBC_RB_RPTR, 0);
445 
446 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
447 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
448 
449 	WREG32_P(mmUVD_RBC_RB_CNTL, 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK);
450 
451 	return 0;
452 }
453 
454 /**
455  * uvd_v5_0_stop - stop UVD block
456  *
457  * @adev: amdgpu_device pointer
458  *
459  * stop the UVD block
460  */
461 static void uvd_v5_0_stop(struct amdgpu_device *adev)
462 {
463 	/* force RBC into idle state */
464 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
465 
466 	/* Stall UMC and register bus before resetting VCPU */
467 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
468 	mdelay(1);
469 
470 	/* put VCPU into reset */
471 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
472 	mdelay(5);
473 
474 	/* disable VCPU clock */
475 	WREG32(mmUVD_VCPU_CNTL, 0x0);
476 
477 	/* Unstall UMC and register bus */
478 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
479 
480 	WREG32(mmUVD_STATUS, 0);
481 }
482 
483 /**
484  * uvd_v5_0_ring_emit_fence - emit an fence & trap command
485  *
486  * @ring: amdgpu_ring pointer
487  * @addr: address
488  * @seq: sequence number
489  * @flags: fence related flags
490  *
491  * Write a fence and a trap command to the ring.
492  */
493 static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
494 				     unsigned flags)
495 {
496 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
497 
498 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
499 	amdgpu_ring_write(ring, seq);
500 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
501 	amdgpu_ring_write(ring, addr & 0xffffffff);
502 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
503 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
504 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
505 	amdgpu_ring_write(ring, 0);
506 
507 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
508 	amdgpu_ring_write(ring, 0);
509 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
510 	amdgpu_ring_write(ring, 0);
511 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
512 	amdgpu_ring_write(ring, 2);
513 }
514 
515 /**
516  * uvd_v5_0_ring_test_ring - register write test
517  *
518  * @ring: amdgpu_ring pointer
519  *
520  * Test if we can successfully write to the context register
521  */
522 static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring)
523 {
524 	struct amdgpu_device *adev = ring->adev;
525 	uint32_t tmp = 0;
526 	unsigned i;
527 	int r;
528 
529 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
530 	r = amdgpu_ring_alloc(ring, 3);
531 	if (r)
532 		return r;
533 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
534 	amdgpu_ring_write(ring, 0xDEADBEEF);
535 	amdgpu_ring_commit(ring);
536 	for (i = 0; i < adev->usec_timeout; i++) {
537 		tmp = RREG32(mmUVD_CONTEXT_ID);
538 		if (tmp == 0xDEADBEEF)
539 			break;
540 		udelay(1);
541 	}
542 
543 	if (i >= adev->usec_timeout)
544 		r = -ETIMEDOUT;
545 
546 	return r;
547 }
548 
549 /**
550  * uvd_v5_0_ring_emit_ib - execute indirect buffer
551  *
552  * @ring: amdgpu_ring pointer
553  * @job: job to retrieve vmid from
554  * @ib: indirect buffer to execute
555  * @flags: unused
556  *
557  * Write ring commands to execute the indirect buffer
558  */
559 static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
560 				  struct amdgpu_job *job,
561 				  struct amdgpu_ib *ib,
562 				  uint32_t flags)
563 {
564 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
565 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
566 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
567 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
568 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
569 	amdgpu_ring_write(ring, ib->length_dw);
570 }
571 
572 static void uvd_v5_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
573 {
574 	int i;
575 
576 	WARN_ON(ring->wptr % 2 || count % 2);
577 
578 	for (i = 0; i < count / 2; i++) {
579 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
580 		amdgpu_ring_write(ring, 0);
581 	}
582 }
583 
584 static bool uvd_v5_0_is_idle(void *handle)
585 {
586 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
587 
588 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
589 }
590 
591 static int uvd_v5_0_wait_for_idle(void *handle)
592 {
593 	unsigned i;
594 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
595 
596 	for (i = 0; i < adev->usec_timeout; i++) {
597 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
598 			return 0;
599 	}
600 	return -ETIMEDOUT;
601 }
602 
603 static int uvd_v5_0_soft_reset(void *handle)
604 {
605 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
606 
607 	uvd_v5_0_stop(adev);
608 
609 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
610 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
611 	mdelay(5);
612 
613 	return uvd_v5_0_start(adev);
614 }
615 
616 static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev,
617 					struct amdgpu_irq_src *source,
618 					unsigned type,
619 					enum amdgpu_interrupt_state state)
620 {
621 	// TODO
622 	return 0;
623 }
624 
625 static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev,
626 				      struct amdgpu_irq_src *source,
627 				      struct amdgpu_iv_entry *entry)
628 {
629 	DRM_DEBUG("IH: UVD TRAP\n");
630 	amdgpu_fence_process(&adev->uvd.inst->ring);
631 	return 0;
632 }
633 
634 static void uvd_v5_0_enable_clock_gating(struct amdgpu_device *adev, bool enable)
635 {
636 	uint32_t data1, data3, suvd_flags;
637 
638 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
639 	data3 = RREG32(mmUVD_CGC_GATE);
640 
641 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
642 		     UVD_SUVD_CGC_GATE__SIT_MASK |
643 		     UVD_SUVD_CGC_GATE__SMP_MASK |
644 		     UVD_SUVD_CGC_GATE__SCM_MASK |
645 		     UVD_SUVD_CGC_GATE__SDB_MASK;
646 
647 	if (enable) {
648 		data3 |= (UVD_CGC_GATE__SYS_MASK     |
649 			UVD_CGC_GATE__UDEC_MASK      |
650 			UVD_CGC_GATE__MPEG2_MASK     |
651 			UVD_CGC_GATE__RBC_MASK       |
652 			UVD_CGC_GATE__LMI_MC_MASK    |
653 			UVD_CGC_GATE__IDCT_MASK      |
654 			UVD_CGC_GATE__MPRD_MASK      |
655 			UVD_CGC_GATE__MPC_MASK       |
656 			UVD_CGC_GATE__LBSI_MASK      |
657 			UVD_CGC_GATE__LRBBM_MASK     |
658 			UVD_CGC_GATE__UDEC_RE_MASK   |
659 			UVD_CGC_GATE__UDEC_CM_MASK   |
660 			UVD_CGC_GATE__UDEC_IT_MASK   |
661 			UVD_CGC_GATE__UDEC_DB_MASK   |
662 			UVD_CGC_GATE__UDEC_MP_MASK   |
663 			UVD_CGC_GATE__WCB_MASK       |
664 			UVD_CGC_GATE__JPEG_MASK      |
665 			UVD_CGC_GATE__SCPU_MASK);
666 		/* only in pg enabled, we can gate clock to vcpu*/
667 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD)
668 			data3 |= UVD_CGC_GATE__VCPU_MASK;
669 		data3 &= ~UVD_CGC_GATE__REGS_MASK;
670 		data1 |= suvd_flags;
671 	} else {
672 		data3 = 0;
673 		data1 = 0;
674 	}
675 
676 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
677 	WREG32(mmUVD_CGC_GATE, data3);
678 }
679 
680 static void uvd_v5_0_set_sw_clock_gating(struct amdgpu_device *adev)
681 {
682 	uint32_t data, data2;
683 
684 	data = RREG32(mmUVD_CGC_CTRL);
685 	data2 = RREG32(mmUVD_SUVD_CGC_CTRL);
686 
687 
688 	data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
689 		  UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
690 
691 
692 	data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
693 		(1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) |
694 		(4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY));
695 
696 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
697 			UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
698 			UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
699 			UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
700 			UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
701 			UVD_CGC_CTRL__SYS_MODE_MASK |
702 			UVD_CGC_CTRL__UDEC_MODE_MASK |
703 			UVD_CGC_CTRL__MPEG2_MODE_MASK |
704 			UVD_CGC_CTRL__REGS_MODE_MASK |
705 			UVD_CGC_CTRL__RBC_MODE_MASK |
706 			UVD_CGC_CTRL__LMI_MC_MODE_MASK |
707 			UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
708 			UVD_CGC_CTRL__IDCT_MODE_MASK |
709 			UVD_CGC_CTRL__MPRD_MODE_MASK |
710 			UVD_CGC_CTRL__MPC_MODE_MASK |
711 			UVD_CGC_CTRL__LBSI_MODE_MASK |
712 			UVD_CGC_CTRL__LRBBM_MODE_MASK |
713 			UVD_CGC_CTRL__WCB_MODE_MASK |
714 			UVD_CGC_CTRL__VCPU_MODE_MASK |
715 			UVD_CGC_CTRL__JPEG_MODE_MASK |
716 			UVD_CGC_CTRL__SCPU_MODE_MASK);
717 	data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
718 			UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
719 			UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
720 			UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
721 			UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
722 
723 	WREG32(mmUVD_CGC_CTRL, data);
724 	WREG32(mmUVD_SUVD_CGC_CTRL, data2);
725 }
726 
727 #if 0
728 static void uvd_v5_0_set_hw_clock_gating(struct amdgpu_device *adev)
729 {
730 	uint32_t data, data1, cgc_flags, suvd_flags;
731 
732 	data = RREG32(mmUVD_CGC_GATE);
733 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
734 
735 	cgc_flags = UVD_CGC_GATE__SYS_MASK |
736 				UVD_CGC_GATE__UDEC_MASK |
737 				UVD_CGC_GATE__MPEG2_MASK |
738 				UVD_CGC_GATE__RBC_MASK |
739 				UVD_CGC_GATE__LMI_MC_MASK |
740 				UVD_CGC_GATE__IDCT_MASK |
741 				UVD_CGC_GATE__MPRD_MASK |
742 				UVD_CGC_GATE__MPC_MASK |
743 				UVD_CGC_GATE__LBSI_MASK |
744 				UVD_CGC_GATE__LRBBM_MASK |
745 				UVD_CGC_GATE__UDEC_RE_MASK |
746 				UVD_CGC_GATE__UDEC_CM_MASK |
747 				UVD_CGC_GATE__UDEC_IT_MASK |
748 				UVD_CGC_GATE__UDEC_DB_MASK |
749 				UVD_CGC_GATE__UDEC_MP_MASK |
750 				UVD_CGC_GATE__WCB_MASK |
751 				UVD_CGC_GATE__VCPU_MASK |
752 				UVD_CGC_GATE__SCPU_MASK;
753 
754 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
755 				UVD_SUVD_CGC_GATE__SIT_MASK |
756 				UVD_SUVD_CGC_GATE__SMP_MASK |
757 				UVD_SUVD_CGC_GATE__SCM_MASK |
758 				UVD_SUVD_CGC_GATE__SDB_MASK;
759 
760 	data |= cgc_flags;
761 	data1 |= suvd_flags;
762 
763 	WREG32(mmUVD_CGC_GATE, data);
764 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
765 }
766 #endif
767 
768 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev,
769 				 bool enable)
770 {
771 	u32 orig, data;
772 
773 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
774 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
775 		data |= 0xfff;
776 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
777 
778 		orig = data = RREG32(mmUVD_CGC_CTRL);
779 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
780 		if (orig != data)
781 			WREG32(mmUVD_CGC_CTRL, data);
782 	} else {
783 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
784 		data &= ~0xfff;
785 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
786 
787 		orig = data = RREG32(mmUVD_CGC_CTRL);
788 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
789 		if (orig != data)
790 			WREG32(mmUVD_CGC_CTRL, data);
791 	}
792 }
793 
794 static int uvd_v5_0_set_clockgating_state(void *handle,
795 					  enum amd_clockgating_state state)
796 {
797 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
798 	bool enable = (state == AMD_CG_STATE_GATE);
799 
800 	if (enable) {
801 		/* wait for STATUS to clear */
802 		if (uvd_v5_0_wait_for_idle(handle))
803 			return -EBUSY;
804 		uvd_v5_0_enable_clock_gating(adev, true);
805 
806 		/* enable HW gates because UVD is idle */
807 /*		uvd_v5_0_set_hw_clock_gating(adev); */
808 	} else {
809 		uvd_v5_0_enable_clock_gating(adev, false);
810 	}
811 
812 	uvd_v5_0_set_sw_clock_gating(adev);
813 	return 0;
814 }
815 
816 static int uvd_v5_0_set_powergating_state(void *handle,
817 					  enum amd_powergating_state state)
818 {
819 	/* This doesn't actually powergate the UVD block.
820 	 * That's done in the dpm code via the SMC.  This
821 	 * just re-inits the block as necessary.  The actual
822 	 * gating still happens in the dpm code.  We should
823 	 * revisit this when there is a cleaner line between
824 	 * the smc and the hw blocks
825 	 */
826 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
827 	int ret = 0;
828 
829 	if (state == AMD_PG_STATE_GATE) {
830 		uvd_v5_0_stop(adev);
831 	} else {
832 		ret = uvd_v5_0_start(adev);
833 		if (ret)
834 			goto out;
835 	}
836 
837 out:
838 	return ret;
839 }
840 
841 static void uvd_v5_0_get_clockgating_state(void *handle, u64 *flags)
842 {
843 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
844 	int data;
845 
846 	mutex_lock(&adev->pm.mutex);
847 
848 	if (RREG32_SMC(ixCURRENT_PG_STATUS) &
849 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
850 		DRM_INFO("Cannot get clockgating state when UVD is powergated.\n");
851 		goto out;
852 	}
853 
854 	/* AMD_CG_SUPPORT_UVD_MGCG */
855 	data = RREG32(mmUVD_CGC_CTRL);
856 	if (data & UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK)
857 		*flags |= AMD_CG_SUPPORT_UVD_MGCG;
858 
859 out:
860 	mutex_unlock(&adev->pm.mutex);
861 }
862 
863 static const struct amd_ip_funcs uvd_v5_0_ip_funcs = {
864 	.name = "uvd_v5_0",
865 	.early_init = uvd_v5_0_early_init,
866 	.late_init = NULL,
867 	.sw_init = uvd_v5_0_sw_init,
868 	.sw_fini = uvd_v5_0_sw_fini,
869 	.hw_init = uvd_v5_0_hw_init,
870 	.hw_fini = uvd_v5_0_hw_fini,
871 	.prepare_suspend = uvd_v5_0_prepare_suspend,
872 	.suspend = uvd_v5_0_suspend,
873 	.resume = uvd_v5_0_resume,
874 	.is_idle = uvd_v5_0_is_idle,
875 	.wait_for_idle = uvd_v5_0_wait_for_idle,
876 	.soft_reset = uvd_v5_0_soft_reset,
877 	.set_clockgating_state = uvd_v5_0_set_clockgating_state,
878 	.set_powergating_state = uvd_v5_0_set_powergating_state,
879 	.get_clockgating_state = uvd_v5_0_get_clockgating_state,
880 	.dump_ip_state = NULL,
881 	.print_ip_state = NULL,
882 };
883 
884 static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
885 	.type = AMDGPU_RING_TYPE_UVD,
886 	.align_mask = 0xf,
887 	.support_64bit_ptrs = false,
888 	.no_user_fence = true,
889 	.get_rptr = uvd_v5_0_ring_get_rptr,
890 	.get_wptr = uvd_v5_0_ring_get_wptr,
891 	.set_wptr = uvd_v5_0_ring_set_wptr,
892 	.parse_cs = amdgpu_uvd_ring_parse_cs,
893 	.emit_frame_size =
894 		14, /* uvd_v5_0_ring_emit_fence  x1 no user fence */
895 	.emit_ib_size = 6, /* uvd_v5_0_ring_emit_ib */
896 	.emit_ib = uvd_v5_0_ring_emit_ib,
897 	.emit_fence = uvd_v5_0_ring_emit_fence,
898 	.test_ring = uvd_v5_0_ring_test_ring,
899 	.test_ib = amdgpu_uvd_ring_test_ib,
900 	.insert_nop = uvd_v5_0_ring_insert_nop,
901 	.pad_ib = amdgpu_ring_generic_pad_ib,
902 	.begin_use = amdgpu_uvd_ring_begin_use,
903 	.end_use = amdgpu_uvd_ring_end_use,
904 };
905 
906 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev)
907 {
908 	adev->uvd.inst->ring.funcs = &uvd_v5_0_ring_funcs;
909 }
910 
911 static const struct amdgpu_irq_src_funcs uvd_v5_0_irq_funcs = {
912 	.set = uvd_v5_0_set_interrupt_state,
913 	.process = uvd_v5_0_process_interrupt,
914 };
915 
916 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev)
917 {
918 	adev->uvd.inst->irq.num_types = 1;
919 	adev->uvd.inst->irq.funcs = &uvd_v5_0_irq_funcs;
920 }
921 
922 const struct amdgpu_ip_block_version uvd_v5_0_ip_block =
923 {
924 		.type = AMD_IP_BLOCK_TYPE_UVD,
925 		.major = 5,
926 		.minor = 0,
927 		.rev = 0,
928 		.funcs = &uvd_v5_0_ip_funcs,
929 };
930