xref: /linux/drivers/gpu/drm/amd/amdgpu/uvd_v5_0.c (revision 7f4f3b14e8079ecde096bd734af10e30d40c27b7)
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(struct amdgpu_ip_block *ip_block)
92 {
93 	struct amdgpu_device *adev = ip_block->adev;
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(struct amdgpu_ip_block *ip_block)
103 {
104 	struct amdgpu_ring *ring;
105 	struct amdgpu_device *adev = ip_block->adev;
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(struct amdgpu_ip_block *ip_block)
132 {
133 	int r;
134 	struct amdgpu_device *adev = ip_block->adev;
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  * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
147  *
148  * Initialize the hardware, boot up the VCPU and do some testing
149  */
150 static int uvd_v5_0_hw_init(struct amdgpu_ip_block *ip_block)
151 {
152 	struct amdgpu_device *adev = ip_block->adev;
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  * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
204  *
205  * Stop the UVD block, mark ring as not ready any more
206  */
207 static int uvd_v5_0_hw_fini(struct amdgpu_ip_block *ip_block)
208 {
209 	struct amdgpu_device *adev = ip_block->adev;
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(struct amdgpu_ip_block *ip_block)
220 {
221 	struct amdgpu_device *adev = ip_block->adev;
222 
223 	return amdgpu_uvd_prepare_suspend(adev);
224 }
225 
226 static int uvd_v5_0_suspend(struct amdgpu_ip_block *ip_block)
227 {
228 	int r;
229 	struct amdgpu_device *adev = ip_block->adev;
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(ip_block);
256 	if (r)
257 		return r;
258 
259 	return amdgpu_uvd_suspend(adev);
260 }
261 
262 static int uvd_v5_0_resume(struct amdgpu_ip_block *ip_block)
263 {
264 	int r;
265 
266 	r = amdgpu_uvd_resume(ip_block->adev);
267 	if (r)
268 		return r;
269 
270 	return uvd_v5_0_hw_init(ip_block);
271 }
272 
273 /**
274  * uvd_v5_0_mc_resume - memory controller programming
275  *
276  * @adev: amdgpu_device pointer
277  *
278  * Let the UVD memory controller know it's offsets
279  */
280 static void uvd_v5_0_mc_resume(struct amdgpu_device *adev)
281 {
282 	uint64_t offset;
283 	uint32_t size;
284 
285 	/* program memory controller bits 0-27 */
286 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
287 			lower_32_bits(adev->uvd.inst->gpu_addr));
288 	WREG32(mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
289 			upper_32_bits(adev->uvd.inst->gpu_addr));
290 
291 	offset = AMDGPU_UVD_FIRMWARE_OFFSET;
292 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev);
293 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, offset >> 3);
294 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
295 
296 	offset += size;
297 	size = AMDGPU_UVD_HEAP_SIZE;
298 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, offset >> 3);
299 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
300 
301 	offset += size;
302 	size = AMDGPU_UVD_STACK_SIZE +
303 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles);
304 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, offset >> 3);
305 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
306 
307 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
308 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
309 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
310 }
311 
312 /**
313  * uvd_v5_0_start - start UVD block
314  *
315  * @adev: amdgpu_device pointer
316  *
317  * Setup and start the UVD block
318  */
319 static int uvd_v5_0_start(struct amdgpu_device *adev)
320 {
321 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
322 	uint32_t rb_bufsz, tmp;
323 	uint32_t lmi_swap_cntl;
324 	uint32_t mp_swap_cntl;
325 	int i, j, r;
326 
327 	/*disable DPG */
328 	WREG32_P(mmUVD_POWER_STATUS, 0, ~(1 << 2));
329 
330 	/* disable byte swapping */
331 	lmi_swap_cntl = 0;
332 	mp_swap_cntl = 0;
333 
334 	uvd_v5_0_mc_resume(adev);
335 
336 	/* disable interupt */
337 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
338 
339 	/* stall UMC and register bus before resetting VCPU */
340 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
341 	mdelay(1);
342 
343 	/* put LMI, VCPU, RBC etc... into reset */
344 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
345 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
346 		UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
347 		UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
348 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
349 	mdelay(5);
350 
351 	/* take UVD block out of reset */
352 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
353 	mdelay(5);
354 
355 	/* initialize UVD memory controller */
356 	WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
357 			     (1 << 21) | (1 << 9) | (1 << 20));
358 
359 #ifdef __BIG_ENDIAN
360 	/* swap (8 in 32) RB and IB */
361 	lmi_swap_cntl = 0xa;
362 	mp_swap_cntl = 0;
363 #endif
364 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
365 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
366 
367 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
368 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
369 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
370 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
371 	WREG32(mmUVD_MPC_SET_ALU, 0);
372 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
373 
374 	/* take all subblocks out of reset, except VCPU */
375 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
376 	mdelay(5);
377 
378 	/* enable VCPU clock */
379 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
380 
381 	/* enable UMC */
382 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
383 
384 	/* boot up the VCPU */
385 	WREG32(mmUVD_SOFT_RESET, 0);
386 	mdelay(10);
387 
388 	for (i = 0; i < 10; ++i) {
389 		uint32_t status;
390 		for (j = 0; j < 100; ++j) {
391 			status = RREG32(mmUVD_STATUS);
392 			if (status & 2)
393 				break;
394 			mdelay(10);
395 		}
396 		r = 0;
397 		if (status & 2)
398 			break;
399 
400 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
401 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
402 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
403 		mdelay(10);
404 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
405 		mdelay(10);
406 		r = -1;
407 	}
408 
409 	if (r) {
410 		DRM_ERROR("UVD not responding, giving up!!!\n");
411 		return r;
412 	}
413 	/* enable master interrupt */
414 	WREG32_P(mmUVD_MASTINT_EN, 3 << 1, ~(3 << 1));
415 
416 	/* clear the bit 4 of UVD_STATUS */
417 	WREG32_P(mmUVD_STATUS, 0, ~(2 << 1));
418 
419 	rb_bufsz = order_base_2(ring->ring_size);
420 	tmp = 0;
421 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
422 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
423 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
424 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_WPTR_POLL_EN, 0);
425 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
426 	tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
427 	/* force RBC into idle state */
428 	WREG32(mmUVD_RBC_RB_CNTL, tmp);
429 
430 	/* set the write pointer delay */
431 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
432 
433 	/* set the wb address */
434 	WREG32(mmUVD_RBC_RB_RPTR_ADDR, (upper_32_bits(ring->gpu_addr) >> 2));
435 
436 	/* program the RB_BASE for ring buffer */
437 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
438 			lower_32_bits(ring->gpu_addr));
439 	WREG32(mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
440 			upper_32_bits(ring->gpu_addr));
441 
442 	/* Initialize the ring buffer's read and write pointers */
443 	WREG32(mmUVD_RBC_RB_RPTR, 0);
444 
445 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
446 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
447 
448 	WREG32_P(mmUVD_RBC_RB_CNTL, 0, ~UVD_RBC_RB_CNTL__RB_NO_FETCH_MASK);
449 
450 	return 0;
451 }
452 
453 /**
454  * uvd_v5_0_stop - stop UVD block
455  *
456  * @adev: amdgpu_device pointer
457  *
458  * stop the UVD block
459  */
460 static void uvd_v5_0_stop(struct amdgpu_device *adev)
461 {
462 	/* force RBC into idle state */
463 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
464 
465 	/* Stall UMC and register bus before resetting VCPU */
466 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
467 	mdelay(1);
468 
469 	/* put VCPU into reset */
470 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
471 	mdelay(5);
472 
473 	/* disable VCPU clock */
474 	WREG32(mmUVD_VCPU_CNTL, 0x0);
475 
476 	/* Unstall UMC and register bus */
477 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
478 
479 	WREG32(mmUVD_STATUS, 0);
480 }
481 
482 /**
483  * uvd_v5_0_ring_emit_fence - emit an fence & trap command
484  *
485  * @ring: amdgpu_ring pointer
486  * @addr: address
487  * @seq: sequence number
488  * @flags: fence related flags
489  *
490  * Write a fence and a trap command to the ring.
491  */
492 static void uvd_v5_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
493 				     unsigned flags)
494 {
495 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
496 
497 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
498 	amdgpu_ring_write(ring, seq);
499 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
500 	amdgpu_ring_write(ring, addr & 0xffffffff);
501 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
502 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
503 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
504 	amdgpu_ring_write(ring, 0);
505 
506 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
507 	amdgpu_ring_write(ring, 0);
508 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
509 	amdgpu_ring_write(ring, 0);
510 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
511 	amdgpu_ring_write(ring, 2);
512 }
513 
514 /**
515  * uvd_v5_0_ring_test_ring - register write test
516  *
517  * @ring: amdgpu_ring pointer
518  *
519  * Test if we can successfully write to the context register
520  */
521 static int uvd_v5_0_ring_test_ring(struct amdgpu_ring *ring)
522 {
523 	struct amdgpu_device *adev = ring->adev;
524 	uint32_t tmp = 0;
525 	unsigned i;
526 	int r;
527 
528 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
529 	r = amdgpu_ring_alloc(ring, 3);
530 	if (r)
531 		return r;
532 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
533 	amdgpu_ring_write(ring, 0xDEADBEEF);
534 	amdgpu_ring_commit(ring);
535 	for (i = 0; i < adev->usec_timeout; i++) {
536 		tmp = RREG32(mmUVD_CONTEXT_ID);
537 		if (tmp == 0xDEADBEEF)
538 			break;
539 		udelay(1);
540 	}
541 
542 	if (i >= adev->usec_timeout)
543 		r = -ETIMEDOUT;
544 
545 	return r;
546 }
547 
548 /**
549  * uvd_v5_0_ring_emit_ib - execute indirect buffer
550  *
551  * @ring: amdgpu_ring pointer
552  * @job: job to retrieve vmid from
553  * @ib: indirect buffer to execute
554  * @flags: unused
555  *
556  * Write ring commands to execute the indirect buffer
557  */
558 static void uvd_v5_0_ring_emit_ib(struct amdgpu_ring *ring,
559 				  struct amdgpu_job *job,
560 				  struct amdgpu_ib *ib,
561 				  uint32_t flags)
562 {
563 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
564 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
565 	amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
566 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
567 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
568 	amdgpu_ring_write(ring, ib->length_dw);
569 }
570 
571 static void uvd_v5_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
572 {
573 	int i;
574 
575 	WARN_ON(ring->wptr % 2 || count % 2);
576 
577 	for (i = 0; i < count / 2; i++) {
578 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
579 		amdgpu_ring_write(ring, 0);
580 	}
581 }
582 
583 static bool uvd_v5_0_is_idle(void *handle)
584 {
585 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
586 
587 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
588 }
589 
590 static int uvd_v5_0_wait_for_idle(struct amdgpu_ip_block *ip_block)
591 {
592 	unsigned i;
593 	struct amdgpu_device *adev = ip_block->adev;
594 
595 	for (i = 0; i < adev->usec_timeout; i++) {
596 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
597 			return 0;
598 	}
599 	return -ETIMEDOUT;
600 }
601 
602 static int uvd_v5_0_soft_reset(struct amdgpu_ip_block *ip_block)
603 {
604 	struct amdgpu_device *adev = ip_block->adev;
605 
606 	uvd_v5_0_stop(adev);
607 
608 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
609 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
610 	mdelay(5);
611 
612 	return uvd_v5_0_start(adev);
613 }
614 
615 static int uvd_v5_0_set_interrupt_state(struct amdgpu_device *adev,
616 					struct amdgpu_irq_src *source,
617 					unsigned type,
618 					enum amdgpu_interrupt_state state)
619 {
620 	// TODO
621 	return 0;
622 }
623 
624 static int uvd_v5_0_process_interrupt(struct amdgpu_device *adev,
625 				      struct amdgpu_irq_src *source,
626 				      struct amdgpu_iv_entry *entry)
627 {
628 	DRM_DEBUG("IH: UVD TRAP\n");
629 	amdgpu_fence_process(&adev->uvd.inst->ring);
630 	return 0;
631 }
632 
633 static void uvd_v5_0_enable_clock_gating(struct amdgpu_device *adev, bool enable)
634 {
635 	uint32_t data1, data3, suvd_flags;
636 
637 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
638 	data3 = RREG32(mmUVD_CGC_GATE);
639 
640 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
641 		     UVD_SUVD_CGC_GATE__SIT_MASK |
642 		     UVD_SUVD_CGC_GATE__SMP_MASK |
643 		     UVD_SUVD_CGC_GATE__SCM_MASK |
644 		     UVD_SUVD_CGC_GATE__SDB_MASK;
645 
646 	if (enable) {
647 		data3 |= (UVD_CGC_GATE__SYS_MASK     |
648 			UVD_CGC_GATE__UDEC_MASK      |
649 			UVD_CGC_GATE__MPEG2_MASK     |
650 			UVD_CGC_GATE__RBC_MASK       |
651 			UVD_CGC_GATE__LMI_MC_MASK    |
652 			UVD_CGC_GATE__IDCT_MASK      |
653 			UVD_CGC_GATE__MPRD_MASK      |
654 			UVD_CGC_GATE__MPC_MASK       |
655 			UVD_CGC_GATE__LBSI_MASK      |
656 			UVD_CGC_GATE__LRBBM_MASK     |
657 			UVD_CGC_GATE__UDEC_RE_MASK   |
658 			UVD_CGC_GATE__UDEC_CM_MASK   |
659 			UVD_CGC_GATE__UDEC_IT_MASK   |
660 			UVD_CGC_GATE__UDEC_DB_MASK   |
661 			UVD_CGC_GATE__UDEC_MP_MASK   |
662 			UVD_CGC_GATE__WCB_MASK       |
663 			UVD_CGC_GATE__JPEG_MASK      |
664 			UVD_CGC_GATE__SCPU_MASK);
665 		/* only in pg enabled, we can gate clock to vcpu*/
666 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD)
667 			data3 |= UVD_CGC_GATE__VCPU_MASK;
668 		data3 &= ~UVD_CGC_GATE__REGS_MASK;
669 		data1 |= suvd_flags;
670 	} else {
671 		data3 = 0;
672 		data1 = 0;
673 	}
674 
675 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
676 	WREG32(mmUVD_CGC_GATE, data3);
677 }
678 
679 static void uvd_v5_0_set_sw_clock_gating(struct amdgpu_device *adev)
680 {
681 	uint32_t data, data2;
682 
683 	data = RREG32(mmUVD_CGC_CTRL);
684 	data2 = RREG32(mmUVD_SUVD_CGC_CTRL);
685 
686 
687 	data &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK |
688 		  UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
689 
690 
691 	data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
692 		(1 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_GATE_DLY_TIMER)) |
693 		(4 << REG_FIELD_SHIFT(UVD_CGC_CTRL, CLK_OFF_DELAY));
694 
695 	data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
696 			UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
697 			UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
698 			UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
699 			UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
700 			UVD_CGC_CTRL__SYS_MODE_MASK |
701 			UVD_CGC_CTRL__UDEC_MODE_MASK |
702 			UVD_CGC_CTRL__MPEG2_MODE_MASK |
703 			UVD_CGC_CTRL__REGS_MODE_MASK |
704 			UVD_CGC_CTRL__RBC_MODE_MASK |
705 			UVD_CGC_CTRL__LMI_MC_MODE_MASK |
706 			UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
707 			UVD_CGC_CTRL__IDCT_MODE_MASK |
708 			UVD_CGC_CTRL__MPRD_MODE_MASK |
709 			UVD_CGC_CTRL__MPC_MODE_MASK |
710 			UVD_CGC_CTRL__LBSI_MODE_MASK |
711 			UVD_CGC_CTRL__LRBBM_MODE_MASK |
712 			UVD_CGC_CTRL__WCB_MODE_MASK |
713 			UVD_CGC_CTRL__VCPU_MODE_MASK |
714 			UVD_CGC_CTRL__JPEG_MODE_MASK |
715 			UVD_CGC_CTRL__SCPU_MODE_MASK);
716 	data2 &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK |
717 			UVD_SUVD_CGC_CTRL__SIT_MODE_MASK |
718 			UVD_SUVD_CGC_CTRL__SMP_MODE_MASK |
719 			UVD_SUVD_CGC_CTRL__SCM_MODE_MASK |
720 			UVD_SUVD_CGC_CTRL__SDB_MODE_MASK);
721 
722 	WREG32(mmUVD_CGC_CTRL, data);
723 	WREG32(mmUVD_SUVD_CGC_CTRL, data2);
724 }
725 
726 #if 0
727 static void uvd_v5_0_set_hw_clock_gating(struct amdgpu_device *adev)
728 {
729 	uint32_t data, data1, cgc_flags, suvd_flags;
730 
731 	data = RREG32(mmUVD_CGC_GATE);
732 	data1 = RREG32(mmUVD_SUVD_CGC_GATE);
733 
734 	cgc_flags = UVD_CGC_GATE__SYS_MASK |
735 				UVD_CGC_GATE__UDEC_MASK |
736 				UVD_CGC_GATE__MPEG2_MASK |
737 				UVD_CGC_GATE__RBC_MASK |
738 				UVD_CGC_GATE__LMI_MC_MASK |
739 				UVD_CGC_GATE__IDCT_MASK |
740 				UVD_CGC_GATE__MPRD_MASK |
741 				UVD_CGC_GATE__MPC_MASK |
742 				UVD_CGC_GATE__LBSI_MASK |
743 				UVD_CGC_GATE__LRBBM_MASK |
744 				UVD_CGC_GATE__UDEC_RE_MASK |
745 				UVD_CGC_GATE__UDEC_CM_MASK |
746 				UVD_CGC_GATE__UDEC_IT_MASK |
747 				UVD_CGC_GATE__UDEC_DB_MASK |
748 				UVD_CGC_GATE__UDEC_MP_MASK |
749 				UVD_CGC_GATE__WCB_MASK |
750 				UVD_CGC_GATE__VCPU_MASK |
751 				UVD_CGC_GATE__SCPU_MASK;
752 
753 	suvd_flags = UVD_SUVD_CGC_GATE__SRE_MASK |
754 				UVD_SUVD_CGC_GATE__SIT_MASK |
755 				UVD_SUVD_CGC_GATE__SMP_MASK |
756 				UVD_SUVD_CGC_GATE__SCM_MASK |
757 				UVD_SUVD_CGC_GATE__SDB_MASK;
758 
759 	data |= cgc_flags;
760 	data1 |= suvd_flags;
761 
762 	WREG32(mmUVD_CGC_GATE, data);
763 	WREG32(mmUVD_SUVD_CGC_GATE, data1);
764 }
765 #endif
766 
767 static void uvd_v5_0_enable_mgcg(struct amdgpu_device *adev,
768 				 bool enable)
769 {
770 	u32 orig, data;
771 
772 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
773 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
774 		data |= 0xfff;
775 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
776 
777 		orig = data = RREG32(mmUVD_CGC_CTRL);
778 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
779 		if (orig != data)
780 			WREG32(mmUVD_CGC_CTRL, data);
781 	} else {
782 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
783 		data &= ~0xfff;
784 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
785 
786 		orig = data = RREG32(mmUVD_CGC_CTRL);
787 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
788 		if (orig != data)
789 			WREG32(mmUVD_CGC_CTRL, data);
790 	}
791 }
792 
793 static int uvd_v5_0_set_clockgating_state(void *handle,
794 					  enum amd_clockgating_state state)
795 {
796 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
797 	bool enable = (state == AMD_CG_STATE_GATE);
798 	struct amdgpu_ip_block *ip_block;
799 
800 	ip_block = amdgpu_device_ip_get_ip_block(adev, AMD_IP_BLOCK_TYPE_UVD);
801 	if (!ip_block)
802 		return -EINVAL;
803 
804 	if (enable) {
805 		/* wait for STATUS to clear */
806 		if (uvd_v5_0_wait_for_idle(ip_block))
807 			return -EBUSY;
808 		uvd_v5_0_enable_clock_gating(adev, true);
809 
810 		/* enable HW gates because UVD is idle */
811 /*		uvd_v5_0_set_hw_clock_gating(adev); */
812 	} else {
813 		uvd_v5_0_enable_clock_gating(adev, false);
814 	}
815 
816 	uvd_v5_0_set_sw_clock_gating(adev);
817 	return 0;
818 }
819 
820 static int uvd_v5_0_set_powergating_state(void *handle,
821 					  enum amd_powergating_state state)
822 {
823 	/* This doesn't actually powergate the UVD block.
824 	 * That's done in the dpm code via the SMC.  This
825 	 * just re-inits the block as necessary.  The actual
826 	 * gating still happens in the dpm code.  We should
827 	 * revisit this when there is a cleaner line between
828 	 * the smc and the hw blocks
829 	 */
830 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
831 	int ret = 0;
832 
833 	if (state == AMD_PG_STATE_GATE) {
834 		uvd_v5_0_stop(adev);
835 	} else {
836 		ret = uvd_v5_0_start(adev);
837 		if (ret)
838 			goto out;
839 	}
840 
841 out:
842 	return ret;
843 }
844 
845 static void uvd_v5_0_get_clockgating_state(void *handle, u64 *flags)
846 {
847 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
848 	int data;
849 
850 	mutex_lock(&adev->pm.mutex);
851 
852 	if (RREG32_SMC(ixCURRENT_PG_STATUS) &
853 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
854 		DRM_INFO("Cannot get clockgating state when UVD is powergated.\n");
855 		goto out;
856 	}
857 
858 	/* AMD_CG_SUPPORT_UVD_MGCG */
859 	data = RREG32(mmUVD_CGC_CTRL);
860 	if (data & UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK)
861 		*flags |= AMD_CG_SUPPORT_UVD_MGCG;
862 
863 out:
864 	mutex_unlock(&adev->pm.mutex);
865 }
866 
867 static const struct amd_ip_funcs uvd_v5_0_ip_funcs = {
868 	.name = "uvd_v5_0",
869 	.early_init = uvd_v5_0_early_init,
870 	.sw_init = uvd_v5_0_sw_init,
871 	.sw_fini = uvd_v5_0_sw_fini,
872 	.hw_init = uvd_v5_0_hw_init,
873 	.hw_fini = uvd_v5_0_hw_fini,
874 	.prepare_suspend = uvd_v5_0_prepare_suspend,
875 	.suspend = uvd_v5_0_suspend,
876 	.resume = uvd_v5_0_resume,
877 	.is_idle = uvd_v5_0_is_idle,
878 	.wait_for_idle = uvd_v5_0_wait_for_idle,
879 	.soft_reset = uvd_v5_0_soft_reset,
880 	.set_clockgating_state = uvd_v5_0_set_clockgating_state,
881 	.set_powergating_state = uvd_v5_0_set_powergating_state,
882 	.get_clockgating_state = uvd_v5_0_get_clockgating_state,
883 };
884 
885 static const struct amdgpu_ring_funcs uvd_v5_0_ring_funcs = {
886 	.type = AMDGPU_RING_TYPE_UVD,
887 	.align_mask = 0xf,
888 	.support_64bit_ptrs = false,
889 	.no_user_fence = true,
890 	.get_rptr = uvd_v5_0_ring_get_rptr,
891 	.get_wptr = uvd_v5_0_ring_get_wptr,
892 	.set_wptr = uvd_v5_0_ring_set_wptr,
893 	.parse_cs = amdgpu_uvd_ring_parse_cs,
894 	.emit_frame_size =
895 		14, /* uvd_v5_0_ring_emit_fence  x1 no user fence */
896 	.emit_ib_size = 6, /* uvd_v5_0_ring_emit_ib */
897 	.emit_ib = uvd_v5_0_ring_emit_ib,
898 	.emit_fence = uvd_v5_0_ring_emit_fence,
899 	.test_ring = uvd_v5_0_ring_test_ring,
900 	.test_ib = amdgpu_uvd_ring_test_ib,
901 	.insert_nop = uvd_v5_0_ring_insert_nop,
902 	.pad_ib = amdgpu_ring_generic_pad_ib,
903 	.begin_use = amdgpu_uvd_ring_begin_use,
904 	.end_use = amdgpu_uvd_ring_end_use,
905 };
906 
907 static void uvd_v5_0_set_ring_funcs(struct amdgpu_device *adev)
908 {
909 	adev->uvd.inst->ring.funcs = &uvd_v5_0_ring_funcs;
910 }
911 
912 static const struct amdgpu_irq_src_funcs uvd_v5_0_irq_funcs = {
913 	.set = uvd_v5_0_set_interrupt_state,
914 	.process = uvd_v5_0_process_interrupt,
915 };
916 
917 static void uvd_v5_0_set_irq_funcs(struct amdgpu_device *adev)
918 {
919 	adev->uvd.inst->irq.num_types = 1;
920 	adev->uvd.inst->irq.funcs = &uvd_v5_0_irq_funcs;
921 }
922 
923 const struct amdgpu_ip_block_version uvd_v5_0_ip_block =
924 {
925 		.type = AMD_IP_BLOCK_TYPE_UVD,
926 		.major = 5,
927 		.minor = 0,
928 		.rev = 0,
929 		.funcs = &uvd_v5_0_ip_funcs,
930 };
931