xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/uvd_v4_2.c (revision 24b4d710)
1 /*
2  * Copyright 2013 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/firmware.h>
26 
27 #include "amdgpu.h"
28 #include "amdgpu_uvd.h"
29 #include "cikd.h"
30 
31 #include "uvd/uvd_4_2_d.h"
32 #include "uvd/uvd_4_2_sh_mask.h"
33 
34 #include "oss/oss_2_0_d.h"
35 #include "oss/oss_2_0_sh_mask.h"
36 
37 #include "bif/bif_4_1_d.h"
38 
39 #include "smu/smu_7_0_1_d.h"
40 #include "smu/smu_7_0_1_sh_mask.h"
41 
42 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev);
43 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev);
44 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev);
45 static int uvd_v4_2_start(struct amdgpu_device *adev);
46 static void uvd_v4_2_stop(struct amdgpu_device *adev);
47 static int uvd_v4_2_set_clockgating_state(void *handle,
48 				enum amd_clockgating_state state);
49 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
50 			     bool sw_mode);
51 /**
52  * uvd_v4_2_ring_get_rptr - get read pointer
53  *
54  * @ring: amdgpu_ring pointer
55  *
56  * Returns the current hardware read pointer
57  */
uvd_v4_2_ring_get_rptr(struct amdgpu_ring * ring)58 static uint64_t uvd_v4_2_ring_get_rptr(struct amdgpu_ring *ring)
59 {
60 	struct amdgpu_device *adev = ring->adev;
61 
62 	return RREG32(mmUVD_RBC_RB_RPTR);
63 }
64 
65 /**
66  * uvd_v4_2_ring_get_wptr - get write pointer
67  *
68  * @ring: amdgpu_ring pointer
69  *
70  * Returns the current hardware write pointer
71  */
uvd_v4_2_ring_get_wptr(struct amdgpu_ring * ring)72 static uint64_t uvd_v4_2_ring_get_wptr(struct amdgpu_ring *ring)
73 {
74 	struct amdgpu_device *adev = ring->adev;
75 
76 	return RREG32(mmUVD_RBC_RB_WPTR);
77 }
78 
79 /**
80  * uvd_v4_2_ring_set_wptr - set write pointer
81  *
82  * @ring: amdgpu_ring pointer
83  *
84  * Commits the write pointer to the hardware
85  */
uvd_v4_2_ring_set_wptr(struct amdgpu_ring * ring)86 static void uvd_v4_2_ring_set_wptr(struct amdgpu_ring *ring)
87 {
88 	struct amdgpu_device *adev = ring->adev;
89 
90 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
91 }
92 
uvd_v4_2_early_init(void * handle)93 static int uvd_v4_2_early_init(void *handle)
94 {
95 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
96 	adev->uvd.num_uvd_inst = 1;
97 
98 	uvd_v4_2_set_ring_funcs(adev);
99 	uvd_v4_2_set_irq_funcs(adev);
100 
101 	return 0;
102 }
103 
uvd_v4_2_sw_init(void * handle)104 static int uvd_v4_2_sw_init(void *handle)
105 {
106 	struct amdgpu_ring *ring;
107 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
108 	int r;
109 
110 	/* UVD TRAP */
111 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, 124, &adev->uvd.inst->irq);
112 	if (r)
113 		return r;
114 
115 	r = amdgpu_uvd_sw_init(adev);
116 	if (r)
117 		return r;
118 
119 	ring = &adev->uvd.inst->ring;
120 	sprintf(ring->name, "uvd");
121 	r = amdgpu_ring_init(adev, ring, 512, &adev->uvd.inst->irq, 0,
122 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
123 	if (r)
124 		return r;
125 
126 	r = amdgpu_uvd_resume(adev);
127 	if (r)
128 		return r;
129 
130 	r = amdgpu_uvd_entity_init(adev);
131 
132 	return r;
133 }
134 
uvd_v4_2_sw_fini(void * handle)135 static int uvd_v4_2_sw_fini(void *handle)
136 {
137 	int r;
138 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
139 
140 	r = amdgpu_uvd_suspend(adev);
141 	if (r)
142 		return r;
143 
144 	return amdgpu_uvd_sw_fini(adev);
145 }
146 
147 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
148 				 bool enable);
149 /**
150  * uvd_v4_2_hw_init - start and test UVD block
151  *
152  * @handle: handle used to pass amdgpu_device pointer
153  *
154  * Initialize the hardware, boot up the VCPU and do some testing
155  */
uvd_v4_2_hw_init(void * handle)156 static int uvd_v4_2_hw_init(void *handle)
157 {
158 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
159 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
160 	uint32_t tmp;
161 	int r;
162 
163 	uvd_v4_2_enable_mgcg(adev, true);
164 	amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
165 
166 	r = amdgpu_ring_test_helper(ring);
167 	if (r)
168 		goto done;
169 
170 	r = amdgpu_ring_alloc(ring, 10);
171 	if (r) {
172 		DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
173 		goto done;
174 	}
175 
176 	tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
177 	amdgpu_ring_write(ring, tmp);
178 	amdgpu_ring_write(ring, 0xFFFFF);
179 
180 	tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
181 	amdgpu_ring_write(ring, tmp);
182 	amdgpu_ring_write(ring, 0xFFFFF);
183 
184 	tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
185 	amdgpu_ring_write(ring, tmp);
186 	amdgpu_ring_write(ring, 0xFFFFF);
187 
188 	/* Clear timeout status bits */
189 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
190 	amdgpu_ring_write(ring, 0x8);
191 
192 	amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
193 	amdgpu_ring_write(ring, 3);
194 
195 	amdgpu_ring_commit(ring);
196 
197 done:
198 	if (!r)
199 		DRM_INFO("UVD initialized successfully.\n");
200 
201 	return r;
202 }
203 
204 /**
205  * uvd_v4_2_hw_fini - stop the hardware block
206  *
207  * @handle: handle used to pass amdgpu_device pointer
208  *
209  * Stop the UVD block, mark ring as not ready any more
210  */
uvd_v4_2_hw_fini(void * handle)211 static int uvd_v4_2_hw_fini(void *handle)
212 {
213 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
214 
215 	cancel_delayed_work_sync(&adev->uvd.idle_work);
216 
217 	if (RREG32(mmUVD_STATUS) != 0)
218 		uvd_v4_2_stop(adev);
219 
220 	return 0;
221 }
222 
uvd_v4_2_suspend(void * handle)223 static int uvd_v4_2_suspend(void *handle)
224 {
225 	int r;
226 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
227 
228 	/*
229 	 * Proper cleanups before halting the HW engine:
230 	 *   - cancel the delayed idle work
231 	 *   - enable powergating
232 	 *   - enable clockgating
233 	 *   - disable dpm
234 	 *
235 	 * TODO: to align with the VCN implementation, move the
236 	 * jobs for clockgating/powergating/dpm setting to
237 	 * ->set_powergating_state().
238 	 */
239 	cancel_delayed_work_sync(&adev->uvd.idle_work);
240 
241 	if (adev->pm.dpm_enabled) {
242 		amdgpu_dpm_enable_uvd(adev, false);
243 	} else {
244 		amdgpu_asic_set_uvd_clocks(adev, 0, 0);
245 		/* shutdown the UVD block */
246 		amdgpu_device_ip_set_powergating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
247 						       AMD_PG_STATE_GATE);
248 		amdgpu_device_ip_set_clockgating_state(adev, AMD_IP_BLOCK_TYPE_UVD,
249 						       AMD_CG_STATE_GATE);
250 	}
251 
252 	r = uvd_v4_2_hw_fini(adev);
253 	if (r)
254 		return r;
255 
256 	return amdgpu_uvd_suspend(adev);
257 }
258 
uvd_v4_2_resume(void * handle)259 static int uvd_v4_2_resume(void *handle)
260 {
261 	int r;
262 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
263 
264 	r = amdgpu_uvd_resume(adev);
265 	if (r)
266 		return r;
267 
268 	return uvd_v4_2_hw_init(adev);
269 }
270 
271 /**
272  * uvd_v4_2_start - start UVD block
273  *
274  * @adev: amdgpu_device pointer
275  *
276  * Setup and start the UVD block
277  */
uvd_v4_2_start(struct amdgpu_device * adev)278 static int uvd_v4_2_start(struct amdgpu_device *adev)
279 {
280 	struct amdgpu_ring *ring = &adev->uvd.inst->ring;
281 	uint32_t rb_bufsz;
282 	int i, j, r;
283 	u32 tmp;
284 	/* disable byte swapping */
285 	u32 lmi_swap_cntl = 0;
286 	u32 mp_swap_cntl = 0;
287 
288 	/* set uvd busy */
289 	WREG32_P(mmUVD_STATUS, 1<<2, ~(1<<2));
290 
291 	uvd_v4_2_set_dcm(adev, true);
292 	WREG32(mmUVD_CGC_GATE, 0);
293 
294 	/* take UVD block out of reset */
295 	WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
296 	mdelay(5);
297 
298 	/* enable VCPU clock */
299 	WREG32(mmUVD_VCPU_CNTL,  1 << 9);
300 
301 	/* disable interupt */
302 	WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
303 
304 #ifdef __BIG_ENDIAN
305 	/* swap (8 in 32) RB and IB */
306 	lmi_swap_cntl = 0xa;
307 	mp_swap_cntl = 0;
308 #endif
309 	WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
310 	WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
311 	/* initialize UVD memory controller */
312 	WREG32(mmUVD_LMI_CTRL, 0x203108);
313 
314 	tmp = RREG32(mmUVD_MPC_CNTL);
315 	WREG32(mmUVD_MPC_CNTL, tmp | 0x10);
316 
317 	WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
318 	WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
319 	WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
320 	WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
321 	WREG32(mmUVD_MPC_SET_ALU, 0);
322 	WREG32(mmUVD_MPC_SET_MUX, 0x88);
323 
324 	uvd_v4_2_mc_resume(adev);
325 
326 	tmp = RREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL);
327 	WREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL, tmp & (~0x10));
328 
329 	/* enable UMC */
330 	WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
331 
332 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
333 
334 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
335 
336 	WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
337 
338 	mdelay(10);
339 
340 	for (i = 0; i < 10; ++i) {
341 		uint32_t status;
342 		for (j = 0; j < 100; ++j) {
343 			status = RREG32(mmUVD_STATUS);
344 			if (status & 2)
345 				break;
346 			mdelay(10);
347 		}
348 		r = 0;
349 		if (status & 2)
350 			break;
351 
352 		DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
353 		WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
354 				~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
355 		mdelay(10);
356 		WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
357 		mdelay(10);
358 		r = -1;
359 	}
360 
361 	if (r) {
362 		DRM_ERROR("UVD not responding, giving up!!!\n");
363 		return r;
364 	}
365 
366 	/* enable interupt */
367 	WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
368 
369 	WREG32_P(mmUVD_STATUS, 0, ~(1<<2));
370 
371 	/* force RBC into idle state */
372 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
373 
374 	/* Set the write pointer delay */
375 	WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
376 
377 	/* program the 4GB memory segment for rptr and ring buffer */
378 	WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
379 				   (0x7 << 16) | (0x1 << 31));
380 
381 	/* Initialize the ring buffer's read and write pointers */
382 	WREG32(mmUVD_RBC_RB_RPTR, 0x0);
383 
384 	ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
385 	WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
386 
387 	/* set the ring address */
388 	WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
389 
390 	/* Set ring buffer size */
391 	rb_bufsz = order_base_2(ring->ring_size);
392 	rb_bufsz = (0x1 << 8) | rb_bufsz;
393 	WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
394 
395 	return 0;
396 }
397 
398 /**
399  * uvd_v4_2_stop - stop UVD block
400  *
401  * @adev: amdgpu_device pointer
402  *
403  * stop the UVD block
404  */
uvd_v4_2_stop(struct amdgpu_device * adev)405 static void uvd_v4_2_stop(struct amdgpu_device *adev)
406 {
407 	uint32_t i, j;
408 	uint32_t status;
409 
410 	WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
411 
412 	for (i = 0; i < 10; ++i) {
413 		for (j = 0; j < 100; ++j) {
414 			status = RREG32(mmUVD_STATUS);
415 			if (status & 2)
416 				break;
417 			mdelay(1);
418 		}
419 		if (status & 2)
420 			break;
421 	}
422 
423 	for (i = 0; i < 10; ++i) {
424 		for (j = 0; j < 100; ++j) {
425 			status = RREG32(mmUVD_LMI_STATUS);
426 			if (status & 0xf)
427 				break;
428 			mdelay(1);
429 		}
430 		if (status & 0xf)
431 			break;
432 	}
433 
434 	/* Stall UMC and register bus before resetting VCPU */
435 	WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
436 
437 	for (i = 0; i < 10; ++i) {
438 		for (j = 0; j < 100; ++j) {
439 			status = RREG32(mmUVD_LMI_STATUS);
440 			if (status & 0x240)
441 				break;
442 			mdelay(1);
443 		}
444 		if (status & 0x240)
445 			break;
446 	}
447 
448 	WREG32_P(0x3D49, 0, ~(1 << 2));
449 
450 	WREG32_P(mmUVD_VCPU_CNTL, 0, ~(1 << 9));
451 
452 	/* put LMI, VCPU, RBC etc... into reset */
453 	WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
454 		UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
455 		UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
456 
457 	WREG32(mmUVD_STATUS, 0);
458 
459 	uvd_v4_2_set_dcm(adev, false);
460 }
461 
462 /**
463  * uvd_v4_2_ring_emit_fence - emit an fence & trap command
464  *
465  * @ring: amdgpu_ring pointer
466  * @addr: address
467  * @seq: sequence number
468  * @flags: fence related flags
469  *
470  * Write a fence and a trap command to the ring.
471  */
uvd_v4_2_ring_emit_fence(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned flags)472 static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
473 				     unsigned flags)
474 {
475 	WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
476 
477 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
478 	amdgpu_ring_write(ring, seq);
479 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
480 	amdgpu_ring_write(ring, addr & 0xffffffff);
481 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
482 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
483 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
484 	amdgpu_ring_write(ring, 0);
485 
486 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
487 	amdgpu_ring_write(ring, 0);
488 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
489 	amdgpu_ring_write(ring, 0);
490 	amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
491 	amdgpu_ring_write(ring, 2);
492 }
493 
494 /**
495  * uvd_v4_2_ring_test_ring - register write test
496  *
497  * @ring: amdgpu_ring pointer
498  *
499  * Test if we can successfully write to the context register
500  */
uvd_v4_2_ring_test_ring(struct amdgpu_ring * ring)501 static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring)
502 {
503 	struct amdgpu_device *adev = ring->adev;
504 	uint32_t tmp = 0;
505 	unsigned i;
506 	int r;
507 
508 	WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
509 	r = amdgpu_ring_alloc(ring, 3);
510 	if (r)
511 		return r;
512 
513 	amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
514 	amdgpu_ring_write(ring, 0xDEADBEEF);
515 	amdgpu_ring_commit(ring);
516 	for (i = 0; i < adev->usec_timeout; i++) {
517 		tmp = RREG32(mmUVD_CONTEXT_ID);
518 		if (tmp == 0xDEADBEEF)
519 			break;
520 		udelay(1);
521 	}
522 
523 	if (i >= adev->usec_timeout)
524 		r = -ETIMEDOUT;
525 
526 	return r;
527 }
528 
529 /**
530  * uvd_v4_2_ring_emit_ib - execute indirect buffer
531  *
532  * @ring: amdgpu_ring pointer
533  * @job: iob associated with the indirect buffer
534  * @ib: indirect buffer to execute
535  * @flags: flags associated with the indirect buffer
536  *
537  * Write ring commands to execute the indirect buffer
538  */
uvd_v4_2_ring_emit_ib(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)539 static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
540 				  struct amdgpu_job *job,
541 				  struct amdgpu_ib *ib,
542 				  uint32_t flags)
543 {
544 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
545 	amdgpu_ring_write(ring, ib->gpu_addr);
546 	amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
547 	amdgpu_ring_write(ring, ib->length_dw);
548 }
549 
uvd_v4_2_ring_insert_nop(struct amdgpu_ring * ring,uint32_t count)550 static void uvd_v4_2_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
551 {
552 	int i;
553 
554 	WARN_ON(ring->wptr % 2 || count % 2);
555 
556 	for (i = 0; i < count / 2; i++) {
557 		amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
558 		amdgpu_ring_write(ring, 0);
559 	}
560 }
561 
562 /**
563  * uvd_v4_2_mc_resume - memory controller programming
564  *
565  * @adev: amdgpu_device pointer
566  *
567  * Let the UVD memory controller know it's offsets
568  */
uvd_v4_2_mc_resume(struct amdgpu_device * adev)569 static void uvd_v4_2_mc_resume(struct amdgpu_device *adev)
570 {
571 	uint64_t addr;
572 	uint32_t size;
573 
574 	/* program the VCPU memory controller bits 0-27 */
575 	addr = (adev->uvd.inst->gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
576 	size = AMDGPU_UVD_FIRMWARE_SIZE(adev) >> 3;
577 	WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
578 	WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
579 
580 	addr += size;
581 	size = AMDGPU_UVD_HEAP_SIZE >> 3;
582 	WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
583 	WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
584 
585 	addr += size;
586 	size = (AMDGPU_UVD_STACK_SIZE +
587 	       (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles)) >> 3;
588 	WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
589 	WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
590 
591 	/* bits 28-31 */
592 	addr = (adev->uvd.inst->gpu_addr >> 28) & 0xF;
593 	WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
594 
595 	/* bits 32-39 */
596 	addr = (adev->uvd.inst->gpu_addr >> 32) & 0xFF;
597 	WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
598 
599 	WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
600 	WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
601 	WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
602 }
603 
uvd_v4_2_enable_mgcg(struct amdgpu_device * adev,bool enable)604 static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
605 				 bool enable)
606 {
607 	u32 orig, data;
608 
609 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
610 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
611 		data |= 0xfff;
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 	} else {
619 		data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
620 		data &= ~0xfff;
621 		WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
622 
623 		orig = data = RREG32(mmUVD_CGC_CTRL);
624 		data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
625 		if (orig != data)
626 			WREG32(mmUVD_CGC_CTRL, data);
627 	}
628 }
629 
uvd_v4_2_set_dcm(struct amdgpu_device * adev,bool sw_mode)630 static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
631 			     bool sw_mode)
632 {
633 	u32 tmp, tmp2;
634 
635 	WREG32_FIELD(UVD_CGC_GATE, REGS, 0);
636 
637 	tmp = RREG32(mmUVD_CGC_CTRL);
638 	tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
639 	tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
640 		(1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
641 		(4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
642 
643 	if (sw_mode) {
644 		tmp &= ~0x7ffff800;
645 		tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
646 			UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
647 			(7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
648 	} else {
649 		tmp |= 0x7ffff800;
650 		tmp2 = 0;
651 	}
652 
653 	WREG32(mmUVD_CGC_CTRL, tmp);
654 	WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
655 }
656 
uvd_v4_2_is_idle(void * handle)657 static bool uvd_v4_2_is_idle(void *handle)
658 {
659 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
660 
661 	return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
662 }
663 
uvd_v4_2_wait_for_idle(void * handle)664 static int uvd_v4_2_wait_for_idle(void *handle)
665 {
666 	unsigned i;
667 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
668 
669 	for (i = 0; i < adev->usec_timeout; i++) {
670 		if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
671 			return 0;
672 	}
673 	return -ETIMEDOUT;
674 }
675 
uvd_v4_2_soft_reset(void * handle)676 static int uvd_v4_2_soft_reset(void *handle)
677 {
678 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
679 
680 	uvd_v4_2_stop(adev);
681 
682 	WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
683 			~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
684 	mdelay(5);
685 
686 	return uvd_v4_2_start(adev);
687 }
688 
uvd_v4_2_set_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)689 static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev,
690 					struct amdgpu_irq_src *source,
691 					unsigned type,
692 					enum amdgpu_interrupt_state state)
693 {
694 	// TODO
695 	return 0;
696 }
697 
uvd_v4_2_process_interrupt(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)698 static int uvd_v4_2_process_interrupt(struct amdgpu_device *adev,
699 				      struct amdgpu_irq_src *source,
700 				      struct amdgpu_iv_entry *entry)
701 {
702 	DRM_DEBUG("IH: UVD TRAP\n");
703 	amdgpu_fence_process(&adev->uvd.inst->ring);
704 	return 0;
705 }
706 
uvd_v4_2_set_clockgating_state(void * handle,enum amd_clockgating_state state)707 static int uvd_v4_2_set_clockgating_state(void *handle,
708 					  enum amd_clockgating_state state)
709 {
710 	return 0;
711 }
712 
uvd_v4_2_set_powergating_state(void * handle,enum amd_powergating_state state)713 static int uvd_v4_2_set_powergating_state(void *handle,
714 					  enum amd_powergating_state state)
715 {
716 	/* This doesn't actually powergate the UVD block.
717 	 * That's done in the dpm code via the SMC.  This
718 	 * just re-inits the block as necessary.  The actual
719 	 * gating still happens in the dpm code.  We should
720 	 * revisit this when there is a cleaner line between
721 	 * the smc and the hw blocks
722 	 */
723 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
724 
725 	if (state == AMD_PG_STATE_GATE) {
726 		uvd_v4_2_stop(adev);
727 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD && !adev->pm.dpm_enabled) {
728 			if (!(RREG32_SMC(ixCURRENT_PG_STATUS) &
729 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK)) {
730 				WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK   |
731 							UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_DOWN_MASK |
732 							UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
733 				mdelay(20);
734 			}
735 		}
736 		return 0;
737 	} else {
738 		if (adev->pg_flags & AMD_PG_SUPPORT_UVD && !adev->pm.dpm_enabled) {
739 			if (RREG32_SMC(ixCURRENT_PG_STATUS) &
740 				CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
741 				WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK   |
742 						UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_UP_MASK |
743 						UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
744 				mdelay(30);
745 			}
746 		}
747 		return uvd_v4_2_start(adev);
748 	}
749 }
750 
751 static const struct amd_ip_funcs uvd_v4_2_ip_funcs = {
752 	.name = "uvd_v4_2",
753 	.early_init = uvd_v4_2_early_init,
754 	.late_init = NULL,
755 	.sw_init = uvd_v4_2_sw_init,
756 	.sw_fini = uvd_v4_2_sw_fini,
757 	.hw_init = uvd_v4_2_hw_init,
758 	.hw_fini = uvd_v4_2_hw_fini,
759 	.suspend = uvd_v4_2_suspend,
760 	.resume = uvd_v4_2_resume,
761 	.is_idle = uvd_v4_2_is_idle,
762 	.wait_for_idle = uvd_v4_2_wait_for_idle,
763 	.soft_reset = uvd_v4_2_soft_reset,
764 	.set_clockgating_state = uvd_v4_2_set_clockgating_state,
765 	.set_powergating_state = uvd_v4_2_set_powergating_state,
766 };
767 
768 static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
769 	.type = AMDGPU_RING_TYPE_UVD,
770 	.align_mask = 0xf,
771 	.support_64bit_ptrs = false,
772 	.no_user_fence = true,
773 	.get_rptr = uvd_v4_2_ring_get_rptr,
774 	.get_wptr = uvd_v4_2_ring_get_wptr,
775 	.set_wptr = uvd_v4_2_ring_set_wptr,
776 	.parse_cs = amdgpu_uvd_ring_parse_cs,
777 	.emit_frame_size =
778 		14, /* uvd_v4_2_ring_emit_fence  x1 no user fence */
779 	.emit_ib_size = 4, /* uvd_v4_2_ring_emit_ib */
780 	.emit_ib = uvd_v4_2_ring_emit_ib,
781 	.emit_fence = uvd_v4_2_ring_emit_fence,
782 	.test_ring = uvd_v4_2_ring_test_ring,
783 	.test_ib = amdgpu_uvd_ring_test_ib,
784 	.insert_nop = uvd_v4_2_ring_insert_nop,
785 	.pad_ib = amdgpu_ring_generic_pad_ib,
786 	.begin_use = amdgpu_uvd_ring_begin_use,
787 	.end_use = amdgpu_uvd_ring_end_use,
788 };
789 
uvd_v4_2_set_ring_funcs(struct amdgpu_device * adev)790 static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)
791 {
792 	adev->uvd.inst->ring.funcs = &uvd_v4_2_ring_funcs;
793 }
794 
795 static const struct amdgpu_irq_src_funcs uvd_v4_2_irq_funcs = {
796 	.set = uvd_v4_2_set_interrupt_state,
797 	.process = uvd_v4_2_process_interrupt,
798 };
799 
uvd_v4_2_set_irq_funcs(struct amdgpu_device * adev)800 static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev)
801 {
802 	adev->uvd.inst->irq.num_types = 1;
803 	adev->uvd.inst->irq.funcs = &uvd_v4_2_irq_funcs;
804 }
805 
806 const struct amdgpu_ip_block_version uvd_v4_2_ip_block =
807 {
808 		.type = AMD_IP_BLOCK_TYPE_UVD,
809 		.major = 4,
810 		.minor = 2,
811 		.rev = 0,
812 		.funcs = &uvd_v4_2_ip_funcs,
813 };
814