xref: /linux/drivers/gpu/drm/amd/amdgpu/uvd_v3_1.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /*
2  * Copyright 2020 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: Sonny Jiang <sonny.jiang@amd.com>
23  */
24 
25 #include <linux/firmware.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_uvd.h"
29 #include "sid.h"
30 
31 #include "uvd/uvd_3_1_d.h"
32 #include "uvd/uvd_3_1_sh_mask.h"
33 
34 #include "oss/oss_1_0_d.h"
35 #include "oss/oss_1_0_sh_mask.h"
36 
37 /**
38  * uvd_v3_1_ring_get_rptr - get read pointer
39  *
40  * @ring: amdgpu_ring pointer
41  *
42  * Returns the current hardware read pointer
43  */
44 static uint64_t uvd_v3_1_ring_get_rptr(struct amdgpu_ring *ring)
45 {
46 	struct amdgpu_device *adev = ring->adev;
47 
48 	return RREG32(mmUVD_RBC_RB_RPTR);
49 }
50 
51 /**
52  * uvd_v3_1_ring_get_wptr - get write pointer
53  *
54  * @ring: amdgpu_ring pointer
55  *
56  * Returns the current hardware write pointer
57  */
58 static uint64_t uvd_v3_1_ring_get_wptr(struct amdgpu_ring *ring)
59 {
60 	struct amdgpu_device *adev = ring->adev;
61 
62 	return RREG32(mmUVD_RBC_RB_WPTR);
63 }
64 
65 /**
66  * uvd_v3_1_ring_set_wptr - set write pointer
67  *
68  * @ring: amdgpu_ring pointer
69  *
70  * Commits the write pointer to the hardware
71  */
72 static void uvd_v3_1_ring_set_wptr(struct amdgpu_ring *ring)
73 {
74 	struct amdgpu_device *adev = ring->adev;
75 
76 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
77 }
78 
79 /**
80  * uvd_v3_1_ring_emit_ib - execute indirect buffer
81  *
82  * @ring: amdgpu_ring pointer
83  * @job: iob associated with the indirect buffer
84  * @ib: indirect buffer to execute
85  * @flags: flags associated with the indirect buffer
86  *
87  * Write ring commands to execute the indirect buffer
88  */
89 static void uvd_v3_1_ring_emit_ib(struct amdgpu_ring *ring,
90 				  struct amdgpu_job *job,
91 				  struct amdgpu_ib *ib,
92 				  uint32_t flags)
93 {
94 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
95 	amdgpu_ring_write(ring, ib->gpu_addr);
96 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
97 	amdgpu_ring_write(ring, ib->length_dw);
98 }
99 
100 /**
101  * uvd_v3_1_ring_emit_fence - emit an fence & trap command
102  *
103  * @ring: amdgpu_ring pointer
104  * @addr: address
105  * @seq: sequence number
106  * @flags: fence related flags
107  *
108  * Write a fence and a trap command to the ring.
109  */
110 static void uvd_v3_1_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
111 				 unsigned flags)
112 {
113 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
114 
115 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
116 	amdgpu_ring_write(ring, seq);
117 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
118 	amdgpu_ring_write(ring, addr & 0xffffffff);
119 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
120 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
121 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
122 	amdgpu_ring_write(ring, 0);
123 
124 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
125 	amdgpu_ring_write(ring, 0);
126 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
127 	amdgpu_ring_write(ring, 0);
128 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
129 	amdgpu_ring_write(ring, 2);
130 }
131 
132 /**
133  * uvd_v3_1_ring_test_ring - register write test
134  *
135  * @ring: amdgpu_ring pointer
136  *
137  * Test if we can successfully write to the context register
138  */
139 static int uvd_v3_1_ring_test_ring(struct amdgpu_ring *ring)
140 {
141 	struct amdgpu_device *adev = ring->adev;
142 	uint32_t tmp = 0;
143 	unsigned i;
144 	int r;
145 
146 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
147 	r = amdgpu_ring_alloc(ring, 3);
148 	if (r)
149 		return r;
150 
151 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
152 	amdgpu_ring_write(ring, 0xDEADBEEF);
153 	amdgpu_ring_commit(ring);
154 	for (i = 0; i < adev->usec_timeout; i++) {
155 		tmp = RREG32(mmUVD_CONTEXT_ID);
156 		if (tmp == 0xDEADBEEF)
157 			break;
158 		udelay(1);
159 	}
160 
161 	if (i >= adev->usec_timeout)
162 		r = -ETIMEDOUT;
163 
164 	return r;
165 }
166 
167 static void uvd_v3_1_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
168 {
169 	int i;
170 
171 	WARN_ON(ring->wptr % 2 || count % 2);
172 
173 	for (i = 0; i < count / 2; i++) {
174 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
175 		amdgpu_ring_write(ring, 0);
176 	}
177 }
178 
179 static const struct amdgpu_ring_funcs uvd_v3_1_ring_funcs = {
180 	.type = AMDGPU_RING_TYPE_UVD,
181 	.align_mask = 0xf,
182 	.support_64bit_ptrs = false,
183 	.no_user_fence = true,
184 	.get_rptr = uvd_v3_1_ring_get_rptr,
185 	.get_wptr = uvd_v3_1_ring_get_wptr,
186 	.set_wptr = uvd_v3_1_ring_set_wptr,
187 	.parse_cs = amdgpu_uvd_ring_parse_cs,
188 	.emit_frame_size =
189 		14, /* uvd_v3_1_ring_emit_fence  x1 no user fence */
190 	.emit_ib_size = 4, /* uvd_v3_1_ring_emit_ib */
191 	.emit_ib = uvd_v3_1_ring_emit_ib,
192 	.emit_fence = uvd_v3_1_ring_emit_fence,
193 	.test_ring = uvd_v3_1_ring_test_ring,
194 	.test_ib = amdgpu_uvd_ring_test_ib,
195 	.insert_nop = uvd_v3_1_ring_insert_nop,
196 	.pad_ib = amdgpu_ring_generic_pad_ib,
197 	.begin_use = amdgpu_uvd_ring_begin_use,
198 	.end_use = amdgpu_uvd_ring_end_use,
199 };
200 
201 static void uvd_v3_1_set_ring_funcs(struct amdgpu_device *adev)
202 {
203 	adev->uvd.inst->ring.funcs = &uvd_v3_1_ring_funcs;
204 }
205 
206 static void uvd_v3_1_set_dcm(struct amdgpu_device *adev,
207 							 bool sw_mode)
208 {
209 	u32 tmp, tmp2;
210 
211 	WREG32_FIELD(UVD_CGC_GATE, REGS, 0);
212 
213 	tmp = RREG32(mmUVD_CGC_CTRL);
214 	tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
215 	tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
216 		(1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
217 		(4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
218 
219 	if (sw_mode) {
220 		tmp &= ~0x7ffff800;
221 		tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
222 			UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
223 			(7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
224 	} else {
225 		tmp |= 0x7ffff800;
226 		tmp2 = 0;
227 	}
228 
229 	WREG32(mmUVD_CGC_CTRL, tmp);
230 	WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
231 }
232 
233 /**
234  * uvd_v3_1_mc_resume - memory controller programming
235  *
236  * @adev: amdgpu_device pointer
237  *
238  * Let the UVD memory controller know it's offsets
239  */
240 static void uvd_v3_1_mc_resume(struct amdgpu_device *adev)
241 {
242 	uint64_t addr;
243 	uint32_t size;
244 
245 	/* programm the VCPU memory controller bits 0-27 */
246 	addr = (adev->uvd.inst->gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
247 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev) >> 3;
248 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
249 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
250 
251 	addr += size;
252 	size = AMDGPU_UVD_HEAP_SIZE >> 3;
253 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
254 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
255 
256 	addr += size;
257 	size = (AMDGPU_UVD_STACK_SIZE +
258 		(AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles)) >> 3;
259 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
260 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
261 
262 	/* bits 28-31 */
263 	addr = (adev->uvd.inst->gpu_addr >> 28) & 0xF;
264 	WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
265 
266 	/* bits 32-39 */
267 	addr = (adev->uvd.inst->gpu_addr >> 32) & 0xFF;
268 	WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
269 
270 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
271 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
272 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
273 }
274 
275 /**
276  * uvd_v3_1_fw_validate - FW validation operation
277  *
278  * @adev: amdgpu_device pointer
279  *
280  * Initialate and check UVD validation.
281  */
282 static int uvd_v3_1_fw_validate(struct amdgpu_device *adev)
283 {
284 	int i;
285 	uint32_t keysel = adev->uvd.keyselect;
286 
287 	WREG32(mmUVD_FW_START, keysel);
288 
289 	for (i = 0; i < 10; ++i) {
290 		mdelay(10);
291 		if (RREG32(mmUVD_FW_STATUS) & UVD_FW_STATUS__DONE_MASK)
292 			break;
293 	}
294 
295 	if (i == 10)
296 		return -ETIMEDOUT;
297 
298 	if (!(RREG32(mmUVD_FW_STATUS) & UVD_FW_STATUS__PASS_MASK))
299 		return -EINVAL;
300 
301 	for (i = 0; i < 10; ++i) {
302 		mdelay(10);
303 		if (!(RREG32(mmUVD_FW_STATUS) & UVD_FW_STATUS__BUSY_MASK))
304 			break;
305 	}
306 
307 	if (i == 10)
308 		return -ETIMEDOUT;
309 
310 	return 0;
311 }
312 
313 /**
314  * uvd_v3_1_start - start UVD block
315  *
316  * @adev: amdgpu_device pointer
317  *
318  * Setup and start the UVD block
319  */
320 static int uvd_v3_1_start(struct amdgpu_device *adev)
321 {
322 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
323 	uint32_t rb_bufsz;
324 	int i, j, r;
325 	u32 tmp;
326 	/* disable byte swapping */
327 	u32 lmi_swap_cntl = 0;
328 	u32 mp_swap_cntl = 0;
329 
330 	/* set uvd busy */
331 	WREG32_P(mmUVD_STATUS, 1<<2, ~(1<<2));
332 
333 	uvd_v3_1_set_dcm(adev, true);
334 	WREG32(mmUVD_CGC_GATE, 0);
335 
336 	/* take UVD block out of reset */
337 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
338 	mdelay(5);
339 
340 	/* enable VCPU clock */
341 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
342 
343 	/* disable interrupt */
344 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
345 
346 #ifdef __BIG_ENDIAN
347 	/* swap (8 in 32) RB and IB */
348 	lmi_swap_cntl = 0xa;
349 	mp_swap_cntl = 0;
350 #endif
351 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
352 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
353 
354 	/* initialize UVD memory controller */
355 	WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
356 		(1 << 21) | (1 << 9) | (1 << 20));
357 
358 	tmp = RREG32(mmUVD_MPC_CNTL);
359 	WREG32(mmUVD_MPC_CNTL, tmp | 0x10);
360 
361 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
362 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
363 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
364 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
365 	WREG32(mmUVD_MPC_SET_ALU, 0);
366 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
367 
368 	tmp = RREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL);
369 	WREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL, tmp & (~0x10));
370 
371 	/* enable UMC */
372 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
373 
374 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
375 
376 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
377 
378 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
379 
380 	mdelay(10);
381 
382 	for (i = 0; i < 10; ++i) {
383 		uint32_t status;
384 		for (j = 0; j < 100; ++j) {
385 			status = RREG32(mmUVD_STATUS);
386 			if (status & 2)
387 				break;
388 			mdelay(10);
389 		}
390 		r = 0;
391 		if (status & 2)
392 			break;
393 
394 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
395 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
396 				 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
397 		mdelay(10);
398 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
399 		mdelay(10);
400 		r = -1;
401 	}
402 
403 	if (r) {
404 		DRM_ERROR("UVD not responding, giving up!!!\n");
405 		return r;
406 	}
407 
408 	/* enable interrupt */
409 	WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
410 
411 	WREG32_P(mmUVD_STATUS, 0, ~(1<<2));
412 
413 	/* force RBC into idle state */
414 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
415 
416 	/* Set the write pointer delay */
417 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
418 
419 	/* programm the 4GB memory segment for rptr and ring buffer */
420 	WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
421 		   (0x7 << 16) | (0x1 << 31));
422 
423 	/* Initialize the ring buffer's read and write pointers */
424 	WREG32(mmUVD_RBC_RB_RPTR, 0x0);
425 
426 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
427 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
428 
429 	/* set the ring address */
430 	WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
431 
432 	/* Set ring buffer size */
433 	rb_bufsz = order_base_2(ring->ring_size);
434 	rb_bufsz = (0x1 << 8) | rb_bufsz;
435 	WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
436 
437 	return 0;
438 }
439 
440 /**
441  * uvd_v3_1_stop - stop UVD block
442  *
443  * @adev: amdgpu_device pointer
444  *
445  * stop the UVD block
446  */
447 static void uvd_v3_1_stop(struct amdgpu_device *adev)
448 {
449 	uint32_t i, j;
450 	uint32_t status;
451 
452 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
453 
454 	for (i = 0; i < 10; ++i) {
455 		for (j = 0; j < 100; ++j) {
456 			status = RREG32(mmUVD_STATUS);
457 			if (status & 2)
458 				break;
459 			mdelay(1);
460 		}
461 		if (status & 2)
462 			break;
463 	}
464 
465 	for (i = 0; i < 10; ++i) {
466 		for (j = 0; j < 100; ++j) {
467 			status = RREG32(mmUVD_LMI_STATUS);
468 			if (status & 0xf)
469 				break;
470 			mdelay(1);
471 		}
472 		if (status & 0xf)
473 			break;
474 	}
475 
476 	/* Stall UMC and register bus before resetting VCPU */
477 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
478 
479 	for (i = 0; i < 10; ++i) {
480 		for (j = 0; j < 100; ++j) {
481 			status = RREG32(mmUVD_LMI_STATUS);
482 			if (status & 0x240)
483 				break;
484 			mdelay(1);
485 		}
486 		if (status & 0x240)
487 			break;
488 	}
489 
490 	WREG32_P(0x3D49, 0, ~(1 << 2));
491 
492 	WREG32_P(mmUVD_VCPU_CNTL, 0, ~(1 << 9));
493 
494 	/* put LMI, VCPU, RBC etc... into reset */
495 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
496 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
497 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
498 
499 	WREG32(mmUVD_STATUS, 0);
500 
501 	uvd_v3_1_set_dcm(adev, false);
502 }
503 
504 static int uvd_v3_1_set_interrupt_state(struct amdgpu_device *adev,
505 					struct amdgpu_irq_src *source,
506 					unsigned type,
507 					enum amdgpu_interrupt_state state)
508 {
509 	return 0;
510 }
511 
512 static int uvd_v3_1_process_interrupt(struct amdgpu_device *adev,
513 				      struct amdgpu_irq_src *source,
514 				      struct amdgpu_iv_entry *entry)
515 {
516 	DRM_DEBUG("IH: UVD TRAP\n");
517 	amdgpu_fence_process(&adev->uvd.inst->ring);
518 	return 0;
519 }
520 
521 
522 static const struct amdgpu_irq_src_funcs uvd_v3_1_irq_funcs = {
523 	.set = uvd_v3_1_set_interrupt_state,
524 	.process = uvd_v3_1_process_interrupt,
525 };
526 
527 static void uvd_v3_1_set_irq_funcs(struct amdgpu_device *adev)
528 {
529 	adev->uvd.inst->irq.num_types = 1;
530 	adev->uvd.inst->irq.funcs = &uvd_v3_1_irq_funcs;
531 }
532 
533 
534 static int uvd_v3_1_early_init(void *handle)
535 {
536 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
537 	adev->uvd.num_uvd_inst = 1;
538 
539 	uvd_v3_1_set_ring_funcs(adev);
540 	uvd_v3_1_set_irq_funcs(adev);
541 
542 	return 0;
543 }
544 
545 static int uvd_v3_1_sw_init(void *handle)
546 {
547 	struct amdgpu_ring *ring;
548 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
549 	int r;
550 	void *ptr;
551 	uint32_t ucode_len;
552 
553 	/* UVD TRAP */
554 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 124, &adev->uvd.inst->irq);
555 	if (r)
556 		return r;
557 
558 	r = amdgpu_uvd_sw_init(adev);
559 	if (r)
560 		return r;
561 
562 	ring = &adev->uvd.inst->ring;
563 	sprintf(ring->name, "uvd");
564 	r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst->irq, 0,
565 			 AMDGPU_RING_PRIO_DEFAULT, NULL);
566 	if (r)
567 		return r;
568 
569 	r = amdgpu_uvd_resume(adev);
570 	if (r)
571 		return r;
572 
573 	/* Retrieval firmware validate key */
574 	ptr = adev->uvd.inst[0].cpu_addr;
575 	ptr += 192 + 16;
576 	memcpy(&ucode_len, ptr, 4);
577 	ptr += ucode_len;
578 	memcpy(&adev->uvd.keyselect, ptr, 4);
579 
580 	return r;
581 }
582 
583 static int uvd_v3_1_sw_fini(void *handle)
584 {
585 	int r;
586 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
587 
588 	r = amdgpu_uvd_suspend(adev);
589 	if (r)
590 		return r;
591 
592 	return amdgpu_uvd_sw_fini(adev);
593 }
594 
595 static void uvd_v3_1_enable_mgcg(struct amdgpu_device *adev,
596 				 bool enable)
597 {
598 	u32 orig, data;
599 
600 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
601 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
602 		data |= 0x3fff;
603 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
604 
605 		orig = data = RREG32(mmUVD_CGC_CTRL);
606 		data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
607 		if (orig != data)
608 			WREG32(mmUVD_CGC_CTRL, data);
609 	} else {
610 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
611 		data &= ~0x3fff;
612 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
613 
614 		orig = data = RREG32(mmUVD_CGC_CTRL);
615 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
616 		if (orig != data)
617 			WREG32(mmUVD_CGC_CTRL, data);
618 	}
619 }
620 
621 /**
622  * uvd_v3_1_hw_init - start and test UVD block
623  *
624  * @handle: handle used to pass amdgpu_device pointer
625  *
626  * Initialize the hardware, boot up the VCPU and do some testing
627  */
628 static int uvd_v3_1_hw_init(void *handle)
629 {
630 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
631 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
632 	uint32_t tmp;
633 	int r;
634 
635 	uvd_v3_1_mc_resume(adev);
636 
637 	r = uvd_v3_1_fw_validate(adev);
638 	if (r) {
639 		DRM_ERROR("amdgpu: UVD Firmware validate fail (%d).\n", r);
640 		return r;
641 	}
642 
643 	uvd_v3_1_enable_mgcg(adev, true);
644 	amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
645 
646 	uvd_v3_1_start(adev);
647 
648 	r = amdgpu_ring_test_helper(ring);
649 	if (r) {
650 		DRM_ERROR("amdgpu: UVD ring test fail (%d).\n", r);
651 		goto done;
652 	}
653 
654 	r = amdgpu_ring_alloc(ring, 10);
655 	if (r) {
656 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
657 		goto done;
658 	}
659 
660 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
661 	amdgpu_ring_write(ring, tmp);
662 	amdgpu_ring_write(ring, 0xFFFFF);
663 
664 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
665 	amdgpu_ring_write(ring, tmp);
666 	amdgpu_ring_write(ring, 0xFFFFF);
667 
668 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
669 	amdgpu_ring_write(ring, tmp);
670 	amdgpu_ring_write(ring, 0xFFFFF);
671 
672 	/* Clear timeout status bits */
673 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
674 	amdgpu_ring_write(ring, 0x8);
675 
676 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
677 	amdgpu_ring_write(ring, 3);
678 
679 	amdgpu_ring_commit(ring);
680 
681 done:
682 	if (!r)
683 		DRM_INFO("UVD initialized successfully.\n");
684 
685 	return r;
686 }
687 
688 /**
689  * uvd_v3_1_hw_fini - stop the hardware block
690  *
691  * @handle: handle used to pass amdgpu_device pointer
692  *
693  * Stop the UVD block, mark ring as not ready any more
694  */
695 static int uvd_v3_1_hw_fini(void *handle)
696 {
697 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
698 
699 	cancel_delayed_work_sync(&adev->uvd.idle_work);
700 
701 	if (RREG32(mmUVD_STATUS) != 0)
702 		uvd_v3_1_stop(adev);
703 
704 	return 0;
705 }
706 
707 static int uvd_v3_1_prepare_suspend(void *handle)
708 {
709 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
710 
711 	return amdgpu_uvd_prepare_suspend(adev);
712 }
713 
714 static int uvd_v3_1_suspend(void *handle)
715 {
716 	int r;
717 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
718 
719 	/*
720 	 * Proper cleanups before halting the HW engine:
721 	 *   - cancel the delayed idle work
722 	 *   - enable powergating
723 	 *   - enable clockgating
724 	 *   - disable dpm
725 	 *
726 	 * TODO: to align with the VCN implementation, move the
727 	 * jobs for clockgating/powergating/dpm setting to
728 	 * ->set_powergating_state().
729 	 */
730 	cancel_delayed_work_sync(&adev->uvd.idle_work);
731 
732 	if (adev->pm.dpm_enabled) {
733 		amdgpu_dpm_enable_uvd(adev, false);
734 	} else {
735 		amdgpu_asic_set_uvd_clocks(adev, 0, 0);
736 		/* shutdown the UVD block */
737 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
738 						       AMD_PG_STATE_GATE);
739 		amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
740 						       AMD_CG_STATE_GATE);
741 	}
742 
743 	r = uvd_v3_1_hw_fini(adev);
744 	if (r)
745 		return r;
746 
747 	return amdgpu_uvd_suspend(adev);
748 }
749 
750 static int uvd_v3_1_resume(void *handle)
751 {
752 	int r;
753 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
754 
755 	r = amdgpu_uvd_resume(adev);
756 	if (r)
757 		return r;
758 
759 	return uvd_v3_1_hw_init(adev);
760 }
761 
762 static bool uvd_v3_1_is_idle(void *handle)
763 {
764 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
765 
766 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
767 }
768 
769 static int uvd_v3_1_wait_for_idle(void *handle)
770 {
771 	unsigned i;
772 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
773 
774 	for (i = 0; i < adev->usec_timeout; i++) {
775 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
776 			return 0;
777 	}
778 	return -ETIMEDOUT;
779 }
780 
781 static int uvd_v3_1_soft_reset(void *handle)
782 {
783 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
784 
785 	uvd_v3_1_stop(adev);
786 
787 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
788 			 ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
789 	mdelay(5);
790 
791 	return uvd_v3_1_start(adev);
792 }
793 
794 static int uvd_v3_1_set_clockgating_state(void *handle,
795 					  enum amd_clockgating_state state)
796 {
797 	return 0;
798 }
799 
800 static int uvd_v3_1_set_powergating_state(void *handle,
801 					  enum amd_powergating_state state)
802 {
803 	return 0;
804 }
805 
806 static const struct amd_ip_funcs uvd_v3_1_ip_funcs = {
807 	.name = "uvd_v3_1",
808 	.early_init = uvd_v3_1_early_init,
809 	.late_init = NULL,
810 	.sw_init = uvd_v3_1_sw_init,
811 	.sw_fini = uvd_v3_1_sw_fini,
812 	.hw_init = uvd_v3_1_hw_init,
813 	.hw_fini = uvd_v3_1_hw_fini,
814 	.prepare_suspend = uvd_v3_1_prepare_suspend,
815 	.suspend = uvd_v3_1_suspend,
816 	.resume = uvd_v3_1_resume,
817 	.is_idle = uvd_v3_1_is_idle,
818 	.wait_for_idle = uvd_v3_1_wait_for_idle,
819 	.soft_reset = uvd_v3_1_soft_reset,
820 	.set_clockgating_state = uvd_v3_1_set_clockgating_state,
821 	.set_powergating_state = uvd_v3_1_set_powergating_state,
822 	.dump_ip_state = NULL,
823 	.print_ip_state = NULL,
824 };
825 
826 const struct amdgpu_ip_block_version uvd_v3_1_ip_block = {
827 	.type = AMD_IP_BLOCK_TYPE_UVD,
828 	.major = 3,
829 	.minor = 1,
830 	.rev = 0,
831 	.funcs = &uvd_v3_1_ip_funcs,
832 };
833