Loading...
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#include <drm/drmP.h>
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
37static void uvd_v4_2_mc_resume(struct amdgpu_device *adev);
38static void uvd_v4_2_init_cg(struct amdgpu_device *adev);
39static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev);
40static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev);
41static int uvd_v4_2_start(struct amdgpu_device *adev);
42static void uvd_v4_2_stop(struct amdgpu_device *adev);
43
44/**
45 * uvd_v4_2_ring_get_rptr - get read pointer
46 *
47 * @ring: amdgpu_ring pointer
48 *
49 * Returns the current hardware read pointer
50 */
51static uint32_t uvd_v4_2_ring_get_rptr(struct amdgpu_ring *ring)
52{
53 struct amdgpu_device *adev = ring->adev;
54
55 return RREG32(mmUVD_RBC_RB_RPTR);
56}
57
58/**
59 * uvd_v4_2_ring_get_wptr - get write pointer
60 *
61 * @ring: amdgpu_ring pointer
62 *
63 * Returns the current hardware write pointer
64 */
65static uint32_t uvd_v4_2_ring_get_wptr(struct amdgpu_ring *ring)
66{
67 struct amdgpu_device *adev = ring->adev;
68
69 return RREG32(mmUVD_RBC_RB_WPTR);
70}
71
72/**
73 * uvd_v4_2_ring_set_wptr - set write pointer
74 *
75 * @ring: amdgpu_ring pointer
76 *
77 * Commits the write pointer to the hardware
78 */
79static void uvd_v4_2_ring_set_wptr(struct amdgpu_ring *ring)
80{
81 struct amdgpu_device *adev = ring->adev;
82
83 WREG32(mmUVD_RBC_RB_WPTR, ring->wptr);
84}
85
86static int uvd_v4_2_early_init(void *handle)
87{
88 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
89
90 uvd_v4_2_set_ring_funcs(adev);
91 uvd_v4_2_set_irq_funcs(adev);
92
93 return 0;
94}
95
96static int uvd_v4_2_sw_init(void *handle)
97{
98 struct amdgpu_ring *ring;
99 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
100 int r;
101
102 /* UVD TRAP */
103 r = amdgpu_irq_add_id(adev, 124, &adev->uvd.irq);
104 if (r)
105 return r;
106
107 r = amdgpu_uvd_sw_init(adev);
108 if (r)
109 return r;
110
111 r = amdgpu_uvd_resume(adev);
112 if (r)
113 return r;
114
115 ring = &adev->uvd.ring;
116 sprintf(ring->name, "uvd");
117 r = amdgpu_ring_init(adev, ring, 4096, CP_PACKET2, 0xf,
118 &adev->uvd.irq, 0, AMDGPU_RING_TYPE_UVD);
119
120 return r;
121}
122
123static int uvd_v4_2_sw_fini(void *handle)
124{
125 int r;
126 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
127
128 r = amdgpu_uvd_suspend(adev);
129 if (r)
130 return r;
131
132 r = amdgpu_uvd_sw_fini(adev);
133 if (r)
134 return r;
135
136 return r;
137}
138
139/**
140 * uvd_v4_2_hw_init - start and test UVD block
141 *
142 * @adev: amdgpu_device pointer
143 *
144 * Initialize the hardware, boot up the VCPU and do some testing
145 */
146static int uvd_v4_2_hw_init(void *handle)
147{
148 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
149 struct amdgpu_ring *ring = &adev->uvd.ring;
150 uint32_t tmp;
151 int r;
152
153 /* raise clocks while booting up the VCPU */
154 amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
155
156 r = uvd_v4_2_start(adev);
157 if (r)
158 goto done;
159
160 ring->ready = true;
161 r = amdgpu_ring_test_ring(ring);
162 if (r) {
163 ring->ready = false;
164 goto done;
165 }
166
167 r = amdgpu_ring_alloc(ring, 10);
168 if (r) {
169 DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
170 goto done;
171 }
172
173 tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
174 amdgpu_ring_write(ring, tmp);
175 amdgpu_ring_write(ring, 0xFFFFF);
176
177 tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
178 amdgpu_ring_write(ring, tmp);
179 amdgpu_ring_write(ring, 0xFFFFF);
180
181 tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
182 amdgpu_ring_write(ring, tmp);
183 amdgpu_ring_write(ring, 0xFFFFF);
184
185 /* Clear timeout status bits */
186 amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
187 amdgpu_ring_write(ring, 0x8);
188
189 amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
190 amdgpu_ring_write(ring, 3);
191
192 amdgpu_ring_commit(ring);
193
194done:
195 /* lower clocks again */
196 amdgpu_asic_set_uvd_clocks(adev, 0, 0);
197
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 * @adev: amdgpu_device pointer
208 *
209 * Stop the UVD block, mark ring as not ready any more
210 */
211static int uvd_v4_2_hw_fini(void *handle)
212{
213 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
214 struct amdgpu_ring *ring = &adev->uvd.ring;
215
216 uvd_v4_2_stop(adev);
217 ring->ready = false;
218
219 return 0;
220}
221
222static int uvd_v4_2_suspend(void *handle)
223{
224 int r;
225 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
226
227 r = uvd_v4_2_hw_fini(adev);
228 if (r)
229 return r;
230
231 r = amdgpu_uvd_suspend(adev);
232 if (r)
233 return r;
234
235 return r;
236}
237
238static int uvd_v4_2_resume(void *handle)
239{
240 int r;
241 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
242
243 r = amdgpu_uvd_resume(adev);
244 if (r)
245 return r;
246
247 r = uvd_v4_2_hw_init(adev);
248 if (r)
249 return r;
250
251 return r;
252}
253
254/**
255 * uvd_v4_2_start - start UVD block
256 *
257 * @adev: amdgpu_device pointer
258 *
259 * Setup and start the UVD block
260 */
261static int uvd_v4_2_start(struct amdgpu_device *adev)
262{
263 struct amdgpu_ring *ring = &adev->uvd.ring;
264 uint32_t rb_bufsz;
265 int i, j, r;
266
267 /* disable byte swapping */
268 u32 lmi_swap_cntl = 0;
269 u32 mp_swap_cntl = 0;
270
271 uvd_v4_2_mc_resume(adev);
272
273 /* disable clock gating */
274 WREG32(mmUVD_CGC_GATE, 0);
275
276 /* disable interupt */
277 WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
278
279 /* Stall UMC and register bus before resetting VCPU */
280 WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
281 mdelay(1);
282
283 /* put LMI, VCPU, RBC etc... into reset */
284 WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
285 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK | UVD_SOFT_RESET__LBSI_SOFT_RESET_MASK |
286 UVD_SOFT_RESET__RBC_SOFT_RESET_MASK | UVD_SOFT_RESET__CSM_SOFT_RESET_MASK |
287 UVD_SOFT_RESET__CXW_SOFT_RESET_MASK | UVD_SOFT_RESET__TAP_SOFT_RESET_MASK |
288 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
289 mdelay(5);
290
291 /* take UVD block out of reset */
292 WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
293 mdelay(5);
294
295 /* initialize UVD memory controller */
296 WREG32(mmUVD_LMI_CTRL, 0x40 | (1 << 8) | (1 << 13) |
297 (1 << 21) | (1 << 9) | (1 << 20));
298
299#ifdef __BIG_ENDIAN
300 /* swap (8 in 32) RB and IB */
301 lmi_swap_cntl = 0xa;
302 mp_swap_cntl = 0;
303#endif
304 WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
305 WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
306
307 WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
308 WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
309 WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
310 WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
311 WREG32(mmUVD_MPC_SET_ALU, 0);
312 WREG32(mmUVD_MPC_SET_MUX, 0x88);
313
314 /* take all subblocks out of reset, except VCPU */
315 WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
316 mdelay(5);
317
318 /* enable VCPU clock */
319 WREG32(mmUVD_VCPU_CNTL, 1 << 9);
320
321 /* enable UMC */
322 WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
323
324 /* boot up the VCPU */
325 WREG32(mmUVD_SOFT_RESET, 0);
326 mdelay(10);
327
328 for (i = 0; i < 10; ++i) {
329 uint32_t status;
330 for (j = 0; j < 100; ++j) {
331 status = RREG32(mmUVD_STATUS);
332 if (status & 2)
333 break;
334 mdelay(10);
335 }
336 r = 0;
337 if (status & 2)
338 break;
339
340 DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
341 WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
342 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
343 mdelay(10);
344 WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
345 mdelay(10);
346 r = -1;
347 }
348
349 if (r) {
350 DRM_ERROR("UVD not responding, giving up!!!\n");
351 return r;
352 }
353
354 /* enable interupt */
355 WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
356
357 /* force RBC into idle state */
358 WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
359
360 /* Set the write pointer delay */
361 WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
362
363 /* programm the 4GB memory segment for rptr and ring buffer */
364 WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
365 (0x7 << 16) | (0x1 << 31));
366
367 /* Initialize the ring buffer's read and write pointers */
368 WREG32(mmUVD_RBC_RB_RPTR, 0x0);
369
370 ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
371 WREG32(mmUVD_RBC_RB_WPTR, ring->wptr);
372
373 /* set the ring address */
374 WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
375
376 /* Set ring buffer size */
377 rb_bufsz = order_base_2(ring->ring_size);
378 rb_bufsz = (0x1 << 8) | rb_bufsz;
379 WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
380
381 return 0;
382}
383
384/**
385 * uvd_v4_2_stop - stop UVD block
386 *
387 * @adev: amdgpu_device pointer
388 *
389 * stop the UVD block
390 */
391static void uvd_v4_2_stop(struct amdgpu_device *adev)
392{
393 /* force RBC into idle state */
394 WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
395
396 /* Stall UMC and register bus before resetting VCPU */
397 WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
398 mdelay(1);
399
400 /* put VCPU into reset */
401 WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
402 mdelay(5);
403
404 /* disable VCPU clock */
405 WREG32(mmUVD_VCPU_CNTL, 0x0);
406
407 /* Unstall UMC and register bus */
408 WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
409}
410
411/**
412 * uvd_v4_2_ring_emit_fence - emit an fence & trap command
413 *
414 * @ring: amdgpu_ring pointer
415 * @fence: fence to emit
416 *
417 * Write a fence and a trap command to the ring.
418 */
419static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
420 unsigned flags)
421{
422 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
423
424 amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
425 amdgpu_ring_write(ring, seq);
426 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
427 amdgpu_ring_write(ring, addr & 0xffffffff);
428 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
429 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
430 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
431 amdgpu_ring_write(ring, 0);
432
433 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
434 amdgpu_ring_write(ring, 0);
435 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
436 amdgpu_ring_write(ring, 0);
437 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
438 amdgpu_ring_write(ring, 2);
439}
440
441/**
442 * uvd_v4_2_ring_test_ring - register write test
443 *
444 * @ring: amdgpu_ring pointer
445 *
446 * Test if we can successfully write to the context register
447 */
448static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring)
449{
450 struct amdgpu_device *adev = ring->adev;
451 uint32_t tmp = 0;
452 unsigned i;
453 int r;
454
455 WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
456 r = amdgpu_ring_alloc(ring, 3);
457 if (r) {
458 DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
459 ring->idx, r);
460 return r;
461 }
462 amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
463 amdgpu_ring_write(ring, 0xDEADBEEF);
464 amdgpu_ring_commit(ring);
465 for (i = 0; i < adev->usec_timeout; i++) {
466 tmp = RREG32(mmUVD_CONTEXT_ID);
467 if (tmp == 0xDEADBEEF)
468 break;
469 DRM_UDELAY(1);
470 }
471
472 if (i < adev->usec_timeout) {
473 DRM_INFO("ring test on %d succeeded in %d usecs\n",
474 ring->idx, i);
475 } else {
476 DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
477 ring->idx, tmp);
478 r = -EINVAL;
479 }
480 return r;
481}
482
483/**
484 * uvd_v4_2_ring_emit_ib - execute indirect buffer
485 *
486 * @ring: amdgpu_ring pointer
487 * @ib: indirect buffer to execute
488 *
489 * Write ring commands to execute the indirect buffer
490 */
491static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
492 struct amdgpu_ib *ib)
493{
494 amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
495 amdgpu_ring_write(ring, ib->gpu_addr);
496 amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
497 amdgpu_ring_write(ring, ib->length_dw);
498}
499
500/**
501 * uvd_v4_2_ring_test_ib - test ib execution
502 *
503 * @ring: amdgpu_ring pointer
504 *
505 * Test if we can successfully execute an IB
506 */
507static int uvd_v4_2_ring_test_ib(struct amdgpu_ring *ring)
508{
509 struct amdgpu_device *adev = ring->adev;
510 struct fence *fence = NULL;
511 int r;
512
513 r = amdgpu_asic_set_uvd_clocks(adev, 53300, 40000);
514 if (r) {
515 DRM_ERROR("amdgpu: failed to raise UVD clocks (%d).\n", r);
516 return r;
517 }
518
519 r = amdgpu_uvd_get_create_msg(ring, 1, NULL);
520 if (r) {
521 DRM_ERROR("amdgpu: failed to get create msg (%d).\n", r);
522 goto error;
523 }
524
525 r = amdgpu_uvd_get_destroy_msg(ring, 1, true, &fence);
526 if (r) {
527 DRM_ERROR("amdgpu: failed to get destroy ib (%d).\n", r);
528 goto error;
529 }
530
531 r = fence_wait(fence, false);
532 if (r) {
533 DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
534 goto error;
535 }
536 DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
537error:
538 fence_put(fence);
539 amdgpu_asic_set_uvd_clocks(adev, 0, 0);
540 return r;
541}
542
543/**
544 * uvd_v4_2_mc_resume - memory controller programming
545 *
546 * @adev: amdgpu_device pointer
547 *
548 * Let the UVD memory controller know it's offsets
549 */
550static void uvd_v4_2_mc_resume(struct amdgpu_device *adev)
551{
552 uint64_t addr;
553 uint32_t size;
554
555 /* programm the VCPU memory controller bits 0-27 */
556 addr = (adev->uvd.gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
557 size = AMDGPU_GPU_PAGE_ALIGN(adev->uvd.fw->size + 4) >> 3;
558 WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
559 WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
560
561 addr += size;
562 size = AMDGPU_UVD_STACK_SIZE >> 3;
563 WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
564 WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
565
566 addr += size;
567 size = AMDGPU_UVD_HEAP_SIZE >> 3;
568 WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
569 WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
570
571 /* bits 28-31 */
572 addr = (adev->uvd.gpu_addr >> 28) & 0xF;
573 WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
574
575 /* bits 32-39 */
576 addr = (adev->uvd.gpu_addr >> 32) & 0xFF;
577 WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
578
579 WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
580 WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
581 WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
582
583 uvd_v4_2_init_cg(adev);
584}
585
586static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
587 bool enable)
588{
589 u32 orig, data;
590
591 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
592 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
593 data = 0xfff;
594 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
595
596 orig = data = RREG32(mmUVD_CGC_CTRL);
597 data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
598 if (orig != data)
599 WREG32(mmUVD_CGC_CTRL, data);
600 } else {
601 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
602 data &= ~0xfff;
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 }
610}
611
612static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
613 bool sw_mode)
614{
615 u32 tmp, tmp2;
616
617 tmp = RREG32(mmUVD_CGC_CTRL);
618 tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
619 tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
620 (1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
621 (4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
622
623 if (sw_mode) {
624 tmp &= ~0x7ffff800;
625 tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
626 UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
627 (7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
628 } else {
629 tmp |= 0x7ffff800;
630 tmp2 = 0;
631 }
632
633 WREG32(mmUVD_CGC_CTRL, tmp);
634 WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
635}
636
637static void uvd_v4_2_init_cg(struct amdgpu_device *adev)
638{
639 bool hw_mode = true;
640
641 if (hw_mode) {
642 uvd_v4_2_set_dcm(adev, false);
643 } else {
644 u32 tmp = RREG32(mmUVD_CGC_CTRL);
645 tmp &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
646 WREG32(mmUVD_CGC_CTRL, tmp);
647 }
648}
649
650static bool uvd_v4_2_is_idle(void *handle)
651{
652 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
653
654 return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
655}
656
657static int uvd_v4_2_wait_for_idle(void *handle)
658{
659 unsigned i;
660 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
661
662 for (i = 0; i < adev->usec_timeout; i++) {
663 if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
664 return 0;
665 }
666 return -ETIMEDOUT;
667}
668
669static int uvd_v4_2_soft_reset(void *handle)
670{
671 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
672
673 uvd_v4_2_stop(adev);
674
675 WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
676 ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
677 mdelay(5);
678
679 return uvd_v4_2_start(adev);
680}
681
682static void uvd_v4_2_print_status(void *handle)
683{
684 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
685 dev_info(adev->dev, "UVD 4.2 registers\n");
686 dev_info(adev->dev, " UVD_SEMA_ADDR_LOW=0x%08X\n",
687 RREG32(mmUVD_SEMA_ADDR_LOW));
688 dev_info(adev->dev, " UVD_SEMA_ADDR_HIGH=0x%08X\n",
689 RREG32(mmUVD_SEMA_ADDR_HIGH));
690 dev_info(adev->dev, " UVD_SEMA_CMD=0x%08X\n",
691 RREG32(mmUVD_SEMA_CMD));
692 dev_info(adev->dev, " UVD_GPCOM_VCPU_CMD=0x%08X\n",
693 RREG32(mmUVD_GPCOM_VCPU_CMD));
694 dev_info(adev->dev, " UVD_GPCOM_VCPU_DATA0=0x%08X\n",
695 RREG32(mmUVD_GPCOM_VCPU_DATA0));
696 dev_info(adev->dev, " UVD_GPCOM_VCPU_DATA1=0x%08X\n",
697 RREG32(mmUVD_GPCOM_VCPU_DATA1));
698 dev_info(adev->dev, " UVD_ENGINE_CNTL=0x%08X\n",
699 RREG32(mmUVD_ENGINE_CNTL));
700 dev_info(adev->dev, " UVD_UDEC_ADDR_CONFIG=0x%08X\n",
701 RREG32(mmUVD_UDEC_ADDR_CONFIG));
702 dev_info(adev->dev, " UVD_UDEC_DB_ADDR_CONFIG=0x%08X\n",
703 RREG32(mmUVD_UDEC_DB_ADDR_CONFIG));
704 dev_info(adev->dev, " UVD_UDEC_DBW_ADDR_CONFIG=0x%08X\n",
705 RREG32(mmUVD_UDEC_DBW_ADDR_CONFIG));
706 dev_info(adev->dev, " UVD_SEMA_CNTL=0x%08X\n",
707 RREG32(mmUVD_SEMA_CNTL));
708 dev_info(adev->dev, " UVD_LMI_EXT40_ADDR=0x%08X\n",
709 RREG32(mmUVD_LMI_EXT40_ADDR));
710 dev_info(adev->dev, " UVD_CTX_INDEX=0x%08X\n",
711 RREG32(mmUVD_CTX_INDEX));
712 dev_info(adev->dev, " UVD_CTX_DATA=0x%08X\n",
713 RREG32(mmUVD_CTX_DATA));
714 dev_info(adev->dev, " UVD_CGC_GATE=0x%08X\n",
715 RREG32(mmUVD_CGC_GATE));
716 dev_info(adev->dev, " UVD_CGC_CTRL=0x%08X\n",
717 RREG32(mmUVD_CGC_CTRL));
718 dev_info(adev->dev, " UVD_LMI_CTRL2=0x%08X\n",
719 RREG32(mmUVD_LMI_CTRL2));
720 dev_info(adev->dev, " UVD_MASTINT_EN=0x%08X\n",
721 RREG32(mmUVD_MASTINT_EN));
722 dev_info(adev->dev, " UVD_LMI_ADDR_EXT=0x%08X\n",
723 RREG32(mmUVD_LMI_ADDR_EXT));
724 dev_info(adev->dev, " UVD_LMI_CTRL=0x%08X\n",
725 RREG32(mmUVD_LMI_CTRL));
726 dev_info(adev->dev, " UVD_LMI_SWAP_CNTL=0x%08X\n",
727 RREG32(mmUVD_LMI_SWAP_CNTL));
728 dev_info(adev->dev, " UVD_MP_SWAP_CNTL=0x%08X\n",
729 RREG32(mmUVD_MP_SWAP_CNTL));
730 dev_info(adev->dev, " UVD_MPC_SET_MUXA0=0x%08X\n",
731 RREG32(mmUVD_MPC_SET_MUXA0));
732 dev_info(adev->dev, " UVD_MPC_SET_MUXA1=0x%08X\n",
733 RREG32(mmUVD_MPC_SET_MUXA1));
734 dev_info(adev->dev, " UVD_MPC_SET_MUXB0=0x%08X\n",
735 RREG32(mmUVD_MPC_SET_MUXB0));
736 dev_info(adev->dev, " UVD_MPC_SET_MUXB1=0x%08X\n",
737 RREG32(mmUVD_MPC_SET_MUXB1));
738 dev_info(adev->dev, " UVD_MPC_SET_MUX=0x%08X\n",
739 RREG32(mmUVD_MPC_SET_MUX));
740 dev_info(adev->dev, " UVD_MPC_SET_ALU=0x%08X\n",
741 RREG32(mmUVD_MPC_SET_ALU));
742 dev_info(adev->dev, " UVD_VCPU_CACHE_OFFSET0=0x%08X\n",
743 RREG32(mmUVD_VCPU_CACHE_OFFSET0));
744 dev_info(adev->dev, " UVD_VCPU_CACHE_SIZE0=0x%08X\n",
745 RREG32(mmUVD_VCPU_CACHE_SIZE0));
746 dev_info(adev->dev, " UVD_VCPU_CACHE_OFFSET1=0x%08X\n",
747 RREG32(mmUVD_VCPU_CACHE_OFFSET1));
748 dev_info(adev->dev, " UVD_VCPU_CACHE_SIZE1=0x%08X\n",
749 RREG32(mmUVD_VCPU_CACHE_SIZE1));
750 dev_info(adev->dev, " UVD_VCPU_CACHE_OFFSET2=0x%08X\n",
751 RREG32(mmUVD_VCPU_CACHE_OFFSET2));
752 dev_info(adev->dev, " UVD_VCPU_CACHE_SIZE2=0x%08X\n",
753 RREG32(mmUVD_VCPU_CACHE_SIZE2));
754 dev_info(adev->dev, " UVD_VCPU_CNTL=0x%08X\n",
755 RREG32(mmUVD_VCPU_CNTL));
756 dev_info(adev->dev, " UVD_SOFT_RESET=0x%08X\n",
757 RREG32(mmUVD_SOFT_RESET));
758 dev_info(adev->dev, " UVD_RBC_IB_BASE=0x%08X\n",
759 RREG32(mmUVD_RBC_IB_BASE));
760 dev_info(adev->dev, " UVD_RBC_IB_SIZE=0x%08X\n",
761 RREG32(mmUVD_RBC_IB_SIZE));
762 dev_info(adev->dev, " UVD_RBC_RB_BASE=0x%08X\n",
763 RREG32(mmUVD_RBC_RB_BASE));
764 dev_info(adev->dev, " UVD_RBC_RB_RPTR=0x%08X\n",
765 RREG32(mmUVD_RBC_RB_RPTR));
766 dev_info(adev->dev, " UVD_RBC_RB_WPTR=0x%08X\n",
767 RREG32(mmUVD_RBC_RB_WPTR));
768 dev_info(adev->dev, " UVD_RBC_RB_WPTR_CNTL=0x%08X\n",
769 RREG32(mmUVD_RBC_RB_WPTR_CNTL));
770 dev_info(adev->dev, " UVD_RBC_RB_CNTL=0x%08X\n",
771 RREG32(mmUVD_RBC_RB_CNTL));
772 dev_info(adev->dev, " UVD_STATUS=0x%08X\n",
773 RREG32(mmUVD_STATUS));
774 dev_info(adev->dev, " UVD_SEMA_TIMEOUT_STATUS=0x%08X\n",
775 RREG32(mmUVD_SEMA_TIMEOUT_STATUS));
776 dev_info(adev->dev, " UVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL=0x%08X\n",
777 RREG32(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL));
778 dev_info(adev->dev, " UVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL=0x%08X\n",
779 RREG32(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL));
780 dev_info(adev->dev, " UVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL=0x%08X\n",
781 RREG32(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL));
782 dev_info(adev->dev, " UVD_CONTEXT_ID=0x%08X\n",
783 RREG32(mmUVD_CONTEXT_ID));
784 dev_info(adev->dev, " UVD_UDEC_ADDR_CONFIG=0x%08X\n",
785 RREG32(mmUVD_UDEC_ADDR_CONFIG));
786 dev_info(adev->dev, " UVD_UDEC_DB_ADDR_CONFIG=0x%08X\n",
787 RREG32(mmUVD_UDEC_DB_ADDR_CONFIG));
788 dev_info(adev->dev, " UVD_UDEC_DBW_ADDR_CONFIG=0x%08X\n",
789 RREG32(mmUVD_UDEC_DBW_ADDR_CONFIG));
790
791}
792
793static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev,
794 struct amdgpu_irq_src *source,
795 unsigned type,
796 enum amdgpu_interrupt_state state)
797{
798 // TODO
799 return 0;
800}
801
802static int uvd_v4_2_process_interrupt(struct amdgpu_device *adev,
803 struct amdgpu_irq_src *source,
804 struct amdgpu_iv_entry *entry)
805{
806 DRM_DEBUG("IH: UVD TRAP\n");
807 amdgpu_fence_process(&adev->uvd.ring);
808 return 0;
809}
810
811static int uvd_v4_2_set_clockgating_state(void *handle,
812 enum amd_clockgating_state state)
813{
814 bool gate = false;
815 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
816
817 if (!(adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG))
818 return 0;
819
820 if (state == AMD_CG_STATE_GATE)
821 gate = true;
822
823 uvd_v4_2_enable_mgcg(adev, gate);
824
825 return 0;
826}
827
828static int uvd_v4_2_set_powergating_state(void *handle,
829 enum amd_powergating_state state)
830{
831 /* This doesn't actually powergate the UVD block.
832 * That's done in the dpm code via the SMC. This
833 * just re-inits the block as necessary. The actual
834 * gating still happens in the dpm code. We should
835 * revisit this when there is a cleaner line between
836 * the smc and the hw blocks
837 */
838 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
839
840 if (!(adev->pg_flags & AMD_PG_SUPPORT_UVD))
841 return 0;
842
843 if (state == AMD_PG_STATE_GATE) {
844 uvd_v4_2_stop(adev);
845 return 0;
846 } else {
847 return uvd_v4_2_start(adev);
848 }
849}
850
851const struct amd_ip_funcs uvd_v4_2_ip_funcs = {
852 .early_init = uvd_v4_2_early_init,
853 .late_init = NULL,
854 .sw_init = uvd_v4_2_sw_init,
855 .sw_fini = uvd_v4_2_sw_fini,
856 .hw_init = uvd_v4_2_hw_init,
857 .hw_fini = uvd_v4_2_hw_fini,
858 .suspend = uvd_v4_2_suspend,
859 .resume = uvd_v4_2_resume,
860 .is_idle = uvd_v4_2_is_idle,
861 .wait_for_idle = uvd_v4_2_wait_for_idle,
862 .soft_reset = uvd_v4_2_soft_reset,
863 .print_status = uvd_v4_2_print_status,
864 .set_clockgating_state = uvd_v4_2_set_clockgating_state,
865 .set_powergating_state = uvd_v4_2_set_powergating_state,
866};
867
868static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
869 .get_rptr = uvd_v4_2_ring_get_rptr,
870 .get_wptr = uvd_v4_2_ring_get_wptr,
871 .set_wptr = uvd_v4_2_ring_set_wptr,
872 .parse_cs = amdgpu_uvd_ring_parse_cs,
873 .emit_ib = uvd_v4_2_ring_emit_ib,
874 .emit_fence = uvd_v4_2_ring_emit_fence,
875 .test_ring = uvd_v4_2_ring_test_ring,
876 .test_ib = uvd_v4_2_ring_test_ib,
877 .insert_nop = amdgpu_ring_insert_nop,
878 .pad_ib = amdgpu_ring_generic_pad_ib,
879};
880
881static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)
882{
883 adev->uvd.ring.funcs = &uvd_v4_2_ring_funcs;
884}
885
886static const struct amdgpu_irq_src_funcs uvd_v4_2_irq_funcs = {
887 .set = uvd_v4_2_set_interrupt_state,
888 .process = uvd_v4_2_process_interrupt,
889};
890
891static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev)
892{
893 adev->uvd.irq.num_types = 1;
894 adev->uvd.irq.funcs = &uvd_v4_2_irq_funcs;
895}
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
42static void uvd_v4_2_mc_resume(struct amdgpu_device *adev);
43static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev);
44static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev);
45static int uvd_v4_2_start(struct amdgpu_device *adev);
46static void uvd_v4_2_stop(struct amdgpu_device *adev);
47static int uvd_v4_2_set_clockgating_state(void *handle,
48 enum amd_clockgating_state state);
49static 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 */
58static 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 */
72static 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 */
86static 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
93static 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
104static 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 if (r)
123 return r;
124
125 r = amdgpu_uvd_resume(adev);
126 if (r)
127 return r;
128
129 r = amdgpu_uvd_entity_init(adev);
130
131 return r;
132}
133
134static int uvd_v4_2_sw_fini(void *handle)
135{
136 int r;
137 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
138
139 r = amdgpu_uvd_suspend(adev);
140 if (r)
141 return r;
142
143 return amdgpu_uvd_sw_fini(adev);
144}
145
146static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
147 bool enable);
148/**
149 * uvd_v4_2_hw_init - start and test UVD block
150 *
151 * @adev: amdgpu_device pointer
152 *
153 * Initialize the hardware, boot up the VCPU and do some testing
154 */
155static int uvd_v4_2_hw_init(void *handle)
156{
157 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
158 struct amdgpu_ring *ring = &adev->uvd.inst->ring;
159 uint32_t tmp;
160 int r;
161
162 uvd_v4_2_enable_mgcg(adev, true);
163 amdgpu_asic_set_uvd_clocks(adev, 10000, 10000);
164
165 r = amdgpu_ring_test_helper(ring);
166 if (r)
167 goto done;
168
169 r = amdgpu_ring_alloc(ring, 10);
170 if (r) {
171 DRM_ERROR("amdgpu: ring failed to lock UVD ring (%d).\n", r);
172 goto done;
173 }
174
175 tmp = PACKET0(mmUVD_SEMA_WAIT_FAULT_TIMEOUT_CNTL, 0);
176 amdgpu_ring_write(ring, tmp);
177 amdgpu_ring_write(ring, 0xFFFFF);
178
179 tmp = PACKET0(mmUVD_SEMA_WAIT_INCOMPLETE_TIMEOUT_CNTL, 0);
180 amdgpu_ring_write(ring, tmp);
181 amdgpu_ring_write(ring, 0xFFFFF);
182
183 tmp = PACKET0(mmUVD_SEMA_SIGNAL_INCOMPLETE_TIMEOUT_CNTL, 0);
184 amdgpu_ring_write(ring, tmp);
185 amdgpu_ring_write(ring, 0xFFFFF);
186
187 /* Clear timeout status bits */
188 amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_TIMEOUT_STATUS, 0));
189 amdgpu_ring_write(ring, 0x8);
190
191 amdgpu_ring_write(ring, PACKET0(mmUVD_SEMA_CNTL, 0));
192 amdgpu_ring_write(ring, 3);
193
194 amdgpu_ring_commit(ring);
195
196done:
197 if (!r)
198 DRM_INFO("UVD initialized successfully.\n");
199
200 return r;
201}
202
203/**
204 * uvd_v4_2_hw_fini - stop the hardware block
205 *
206 * @adev: amdgpu_device pointer
207 *
208 * Stop the UVD block, mark ring as not ready any more
209 */
210static int uvd_v4_2_hw_fini(void *handle)
211{
212 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
213 struct amdgpu_ring *ring = &adev->uvd.inst->ring;
214
215 if (RREG32(mmUVD_STATUS) != 0)
216 uvd_v4_2_stop(adev);
217
218 ring->sched.ready = false;
219
220 return 0;
221}
222
223static int uvd_v4_2_suspend(void *handle)
224{
225 int r;
226 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
227
228 r = uvd_v4_2_hw_fini(adev);
229 if (r)
230 return r;
231
232 return amdgpu_uvd_suspend(adev);
233}
234
235static int uvd_v4_2_resume(void *handle)
236{
237 int r;
238 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
239
240 r = amdgpu_uvd_resume(adev);
241 if (r)
242 return r;
243
244 return uvd_v4_2_hw_init(adev);
245}
246
247/**
248 * uvd_v4_2_start - start UVD block
249 *
250 * @adev: amdgpu_device pointer
251 *
252 * Setup and start the UVD block
253 */
254static int uvd_v4_2_start(struct amdgpu_device *adev)
255{
256 struct amdgpu_ring *ring = &adev->uvd.inst->ring;
257 uint32_t rb_bufsz;
258 int i, j, r;
259 u32 tmp;
260 /* disable byte swapping */
261 u32 lmi_swap_cntl = 0;
262 u32 mp_swap_cntl = 0;
263
264 /* set uvd busy */
265 WREG32_P(mmUVD_STATUS, 1<<2, ~(1<<2));
266
267 uvd_v4_2_set_dcm(adev, true);
268 WREG32(mmUVD_CGC_GATE, 0);
269
270 /* take UVD block out of reset */
271 WREG32_P(mmSRBM_SOFT_RESET, 0, ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
272 mdelay(5);
273
274 /* enable VCPU clock */
275 WREG32(mmUVD_VCPU_CNTL, 1 << 9);
276
277 /* disable interupt */
278 WREG32_P(mmUVD_MASTINT_EN, 0, ~(1 << 1));
279
280#ifdef __BIG_ENDIAN
281 /* swap (8 in 32) RB and IB */
282 lmi_swap_cntl = 0xa;
283 mp_swap_cntl = 0;
284#endif
285 WREG32(mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
286 WREG32(mmUVD_MP_SWAP_CNTL, mp_swap_cntl);
287 /* initialize UVD memory controller */
288 WREG32(mmUVD_LMI_CTRL, 0x203108);
289
290 tmp = RREG32(mmUVD_MPC_CNTL);
291 WREG32(mmUVD_MPC_CNTL, tmp | 0x10);
292
293 WREG32(mmUVD_MPC_SET_MUXA0, 0x40c2040);
294 WREG32(mmUVD_MPC_SET_MUXA1, 0x0);
295 WREG32(mmUVD_MPC_SET_MUXB0, 0x40c2040);
296 WREG32(mmUVD_MPC_SET_MUXB1, 0x0);
297 WREG32(mmUVD_MPC_SET_ALU, 0);
298 WREG32(mmUVD_MPC_SET_MUX, 0x88);
299
300 uvd_v4_2_mc_resume(adev);
301
302 tmp = RREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL);
303 WREG32_UVD_CTX(ixUVD_LMI_CACHE_CTRL, tmp & (~0x10));
304
305 /* enable UMC */
306 WREG32_P(mmUVD_LMI_CTRL2, 0, ~(1 << 8));
307
308 WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
309
310 WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
311
312 WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
313
314 mdelay(10);
315
316 for (i = 0; i < 10; ++i) {
317 uint32_t status;
318 for (j = 0; j < 100; ++j) {
319 status = RREG32(mmUVD_STATUS);
320 if (status & 2)
321 break;
322 mdelay(10);
323 }
324 r = 0;
325 if (status & 2)
326 break;
327
328 DRM_ERROR("UVD not responding, trying to reset the VCPU!!!\n");
329 WREG32_P(mmUVD_SOFT_RESET, UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
330 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
331 mdelay(10);
332 WREG32_P(mmUVD_SOFT_RESET, 0, ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
333 mdelay(10);
334 r = -1;
335 }
336
337 if (r) {
338 DRM_ERROR("UVD not responding, giving up!!!\n");
339 return r;
340 }
341
342 /* enable interupt */
343 WREG32_P(mmUVD_MASTINT_EN, 3<<1, ~(3 << 1));
344
345 WREG32_P(mmUVD_STATUS, 0, ~(1<<2));
346
347 /* force RBC into idle state */
348 WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
349
350 /* Set the write pointer delay */
351 WREG32(mmUVD_RBC_RB_WPTR_CNTL, 0);
352
353 /* programm the 4GB memory segment for rptr and ring buffer */
354 WREG32(mmUVD_LMI_EXT40_ADDR, upper_32_bits(ring->gpu_addr) |
355 (0x7 << 16) | (0x1 << 31));
356
357 /* Initialize the ring buffer's read and write pointers */
358 WREG32(mmUVD_RBC_RB_RPTR, 0x0);
359
360 ring->wptr = RREG32(mmUVD_RBC_RB_RPTR);
361 WREG32(mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
362
363 /* set the ring address */
364 WREG32(mmUVD_RBC_RB_BASE, ring->gpu_addr);
365
366 /* Set ring buffer size */
367 rb_bufsz = order_base_2(ring->ring_size);
368 rb_bufsz = (0x1 << 8) | rb_bufsz;
369 WREG32_P(mmUVD_RBC_RB_CNTL, rb_bufsz, ~0x11f1f);
370
371 return 0;
372}
373
374/**
375 * uvd_v4_2_stop - stop UVD block
376 *
377 * @adev: amdgpu_device pointer
378 *
379 * stop the UVD block
380 */
381static void uvd_v4_2_stop(struct amdgpu_device *adev)
382{
383 uint32_t i, j;
384 uint32_t status;
385
386 WREG32(mmUVD_RBC_RB_CNTL, 0x11010101);
387
388 for (i = 0; i < 10; ++i) {
389 for (j = 0; j < 100; ++j) {
390 status = RREG32(mmUVD_STATUS);
391 if (status & 2)
392 break;
393 mdelay(1);
394 }
395 if (status & 2)
396 break;
397 }
398
399 for (i = 0; i < 10; ++i) {
400 for (j = 0; j < 100; ++j) {
401 status = RREG32(mmUVD_LMI_STATUS);
402 if (status & 0xf)
403 break;
404 mdelay(1);
405 }
406 if (status & 0xf)
407 break;
408 }
409
410 /* Stall UMC and register bus before resetting VCPU */
411 WREG32_P(mmUVD_LMI_CTRL2, 1 << 8, ~(1 << 8));
412
413 for (i = 0; i < 10; ++i) {
414 for (j = 0; j < 100; ++j) {
415 status = RREG32(mmUVD_LMI_STATUS);
416 if (status & 0x240)
417 break;
418 mdelay(1);
419 }
420 if (status & 0x240)
421 break;
422 }
423
424 WREG32_P(0x3D49, 0, ~(1 << 2));
425
426 WREG32_P(mmUVD_VCPU_CNTL, 0, ~(1 << 9));
427
428 /* put LMI, VCPU, RBC etc... into reset */
429 WREG32(mmUVD_SOFT_RESET, UVD_SOFT_RESET__LMI_SOFT_RESET_MASK |
430 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK |
431 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
432
433 WREG32(mmUVD_STATUS, 0);
434
435 uvd_v4_2_set_dcm(adev, false);
436}
437
438/**
439 * uvd_v4_2_ring_emit_fence - emit an fence & trap command
440 *
441 * @ring: amdgpu_ring pointer
442 * @fence: fence to emit
443 *
444 * Write a fence and a trap command to the ring.
445 */
446static void uvd_v4_2_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
447 unsigned flags)
448{
449 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
450
451 amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
452 amdgpu_ring_write(ring, seq);
453 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
454 amdgpu_ring_write(ring, addr & 0xffffffff);
455 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
456 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
457 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
458 amdgpu_ring_write(ring, 0);
459
460 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
461 amdgpu_ring_write(ring, 0);
462 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
463 amdgpu_ring_write(ring, 0);
464 amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
465 amdgpu_ring_write(ring, 2);
466}
467
468/**
469 * uvd_v4_2_ring_test_ring - register write test
470 *
471 * @ring: amdgpu_ring pointer
472 *
473 * Test if we can successfully write to the context register
474 */
475static int uvd_v4_2_ring_test_ring(struct amdgpu_ring *ring)
476{
477 struct amdgpu_device *adev = ring->adev;
478 uint32_t tmp = 0;
479 unsigned i;
480 int r;
481
482 WREG32(mmUVD_CONTEXT_ID, 0xCAFEDEAD);
483 r = amdgpu_ring_alloc(ring, 3);
484 if (r)
485 return r;
486
487 amdgpu_ring_write(ring, PACKET0(mmUVD_CONTEXT_ID, 0));
488 amdgpu_ring_write(ring, 0xDEADBEEF);
489 amdgpu_ring_commit(ring);
490 for (i = 0; i < adev->usec_timeout; i++) {
491 tmp = RREG32(mmUVD_CONTEXT_ID);
492 if (tmp == 0xDEADBEEF)
493 break;
494 udelay(1);
495 }
496
497 if (i >= adev->usec_timeout)
498 r = -ETIMEDOUT;
499
500 return r;
501}
502
503/**
504 * uvd_v4_2_ring_emit_ib - execute indirect buffer
505 *
506 * @ring: amdgpu_ring pointer
507 * @ib: indirect buffer to execute
508 *
509 * Write ring commands to execute the indirect buffer
510 */
511static void uvd_v4_2_ring_emit_ib(struct amdgpu_ring *ring,
512 struct amdgpu_job *job,
513 struct amdgpu_ib *ib,
514 uint32_t flags)
515{
516 amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_BASE, 0));
517 amdgpu_ring_write(ring, ib->gpu_addr);
518 amdgpu_ring_write(ring, PACKET0(mmUVD_RBC_IB_SIZE, 0));
519 amdgpu_ring_write(ring, ib->length_dw);
520}
521
522static void uvd_v4_2_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
523{
524 int i;
525
526 WARN_ON(ring->wptr % 2 || count % 2);
527
528 for (i = 0; i < count / 2; i++) {
529 amdgpu_ring_write(ring, PACKET0(mmUVD_NO_OP, 0));
530 amdgpu_ring_write(ring, 0);
531 }
532}
533
534/**
535 * uvd_v4_2_mc_resume - memory controller programming
536 *
537 * @adev: amdgpu_device pointer
538 *
539 * Let the UVD memory controller know it's offsets
540 */
541static void uvd_v4_2_mc_resume(struct amdgpu_device *adev)
542{
543 uint64_t addr;
544 uint32_t size;
545
546 /* programm the VCPU memory controller bits 0-27 */
547 addr = (adev->uvd.inst->gpu_addr + AMDGPU_UVD_FIRMWARE_OFFSET) >> 3;
548 size = AMDGPU_UVD_FIRMWARE_SIZE(adev) >> 3;
549 WREG32(mmUVD_VCPU_CACHE_OFFSET0, addr);
550 WREG32(mmUVD_VCPU_CACHE_SIZE0, size);
551
552 addr += size;
553 size = AMDGPU_UVD_HEAP_SIZE >> 3;
554 WREG32(mmUVD_VCPU_CACHE_OFFSET1, addr);
555 WREG32(mmUVD_VCPU_CACHE_SIZE1, size);
556
557 addr += size;
558 size = (AMDGPU_UVD_STACK_SIZE +
559 (AMDGPU_UVD_SESSION_SIZE * adev->uvd.max_handles)) >> 3;
560 WREG32(mmUVD_VCPU_CACHE_OFFSET2, addr);
561 WREG32(mmUVD_VCPU_CACHE_SIZE2, size);
562
563 /* bits 28-31 */
564 addr = (adev->uvd.inst->gpu_addr >> 28) & 0xF;
565 WREG32(mmUVD_LMI_ADDR_EXT, (addr << 12) | (addr << 0));
566
567 /* bits 32-39 */
568 addr = (adev->uvd.inst->gpu_addr >> 32) & 0xFF;
569 WREG32(mmUVD_LMI_EXT40_ADDR, addr | (0x9 << 16) | (0x1 << 31));
570
571 WREG32(mmUVD_UDEC_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
572 WREG32(mmUVD_UDEC_DB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
573 WREG32(mmUVD_UDEC_DBW_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
574}
575
576static void uvd_v4_2_enable_mgcg(struct amdgpu_device *adev,
577 bool enable)
578{
579 u32 orig, data;
580
581 if (enable && (adev->cg_flags & AMD_CG_SUPPORT_UVD_MGCG)) {
582 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
583 data |= 0xfff;
584 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
585
586 orig = data = RREG32(mmUVD_CGC_CTRL);
587 data |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
588 if (orig != data)
589 WREG32(mmUVD_CGC_CTRL, data);
590 } else {
591 data = RREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL);
592 data &= ~0xfff;
593 WREG32_UVD_CTX(ixUVD_CGC_MEM_CTRL, data);
594
595 orig = data = RREG32(mmUVD_CGC_CTRL);
596 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
597 if (orig != data)
598 WREG32(mmUVD_CGC_CTRL, data);
599 }
600}
601
602static void uvd_v4_2_set_dcm(struct amdgpu_device *adev,
603 bool sw_mode)
604{
605 u32 tmp, tmp2;
606
607 WREG32_FIELD(UVD_CGC_GATE, REGS, 0);
608
609 tmp = RREG32(mmUVD_CGC_CTRL);
610 tmp &= ~(UVD_CGC_CTRL__CLK_OFF_DELAY_MASK | UVD_CGC_CTRL__CLK_GATE_DLY_TIMER_MASK);
611 tmp |= UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK |
612 (1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT) |
613 (4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT);
614
615 if (sw_mode) {
616 tmp &= ~0x7ffff800;
617 tmp2 = UVD_CGC_CTRL2__DYN_OCLK_RAMP_EN_MASK |
618 UVD_CGC_CTRL2__DYN_RCLK_RAMP_EN_MASK |
619 (7 << UVD_CGC_CTRL2__GATER_DIV_ID__SHIFT);
620 } else {
621 tmp |= 0x7ffff800;
622 tmp2 = 0;
623 }
624
625 WREG32(mmUVD_CGC_CTRL, tmp);
626 WREG32_UVD_CTX(ixUVD_CGC_CTRL2, tmp2);
627}
628
629static bool uvd_v4_2_is_idle(void *handle)
630{
631 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
632
633 return !(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK);
634}
635
636static int uvd_v4_2_wait_for_idle(void *handle)
637{
638 unsigned i;
639 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
640
641 for (i = 0; i < adev->usec_timeout; i++) {
642 if (!(RREG32(mmSRBM_STATUS) & SRBM_STATUS__UVD_BUSY_MASK))
643 return 0;
644 }
645 return -ETIMEDOUT;
646}
647
648static int uvd_v4_2_soft_reset(void *handle)
649{
650 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
651
652 uvd_v4_2_stop(adev);
653
654 WREG32_P(mmSRBM_SOFT_RESET, SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK,
655 ~SRBM_SOFT_RESET__SOFT_RESET_UVD_MASK);
656 mdelay(5);
657
658 return uvd_v4_2_start(adev);
659}
660
661static int uvd_v4_2_set_interrupt_state(struct amdgpu_device *adev,
662 struct amdgpu_irq_src *source,
663 unsigned type,
664 enum amdgpu_interrupt_state state)
665{
666 // TODO
667 return 0;
668}
669
670static int uvd_v4_2_process_interrupt(struct amdgpu_device *adev,
671 struct amdgpu_irq_src *source,
672 struct amdgpu_iv_entry *entry)
673{
674 DRM_DEBUG("IH: UVD TRAP\n");
675 amdgpu_fence_process(&adev->uvd.inst->ring);
676 return 0;
677}
678
679static int uvd_v4_2_set_clockgating_state(void *handle,
680 enum amd_clockgating_state state)
681{
682 return 0;
683}
684
685static int uvd_v4_2_set_powergating_state(void *handle,
686 enum amd_powergating_state state)
687{
688 /* This doesn't actually powergate the UVD block.
689 * That's done in the dpm code via the SMC. This
690 * just re-inits the block as necessary. The actual
691 * gating still happens in the dpm code. We should
692 * revisit this when there is a cleaner line between
693 * the smc and the hw blocks
694 */
695 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
696
697 if (state == AMD_PG_STATE_GATE) {
698 uvd_v4_2_stop(adev);
699 if (adev->pg_flags & AMD_PG_SUPPORT_UVD && !adev->pm.dpm_enabled) {
700 if (!(RREG32_SMC(ixCURRENT_PG_STATUS) &
701 CURRENT_PG_STATUS__UVD_PG_STATUS_MASK)) {
702 WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK |
703 UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_DOWN_MASK |
704 UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
705 mdelay(20);
706 }
707 }
708 return 0;
709 } else {
710 if (adev->pg_flags & AMD_PG_SUPPORT_UVD && !adev->pm.dpm_enabled) {
711 if (RREG32_SMC(ixCURRENT_PG_STATUS) &
712 CURRENT_PG_STATUS__UVD_PG_STATUS_MASK) {
713 WREG32(mmUVD_PGFSM_CONFIG, (UVD_PGFSM_CONFIG__UVD_PGFSM_FSM_ADDR_MASK |
714 UVD_PGFSM_CONFIG__UVD_PGFSM_POWER_UP_MASK |
715 UVD_PGFSM_CONFIG__UVD_PGFSM_P1_SELECT_MASK));
716 mdelay(30);
717 }
718 }
719 return uvd_v4_2_start(adev);
720 }
721}
722
723static const struct amd_ip_funcs uvd_v4_2_ip_funcs = {
724 .name = "uvd_v4_2",
725 .early_init = uvd_v4_2_early_init,
726 .late_init = NULL,
727 .sw_init = uvd_v4_2_sw_init,
728 .sw_fini = uvd_v4_2_sw_fini,
729 .hw_init = uvd_v4_2_hw_init,
730 .hw_fini = uvd_v4_2_hw_fini,
731 .suspend = uvd_v4_2_suspend,
732 .resume = uvd_v4_2_resume,
733 .is_idle = uvd_v4_2_is_idle,
734 .wait_for_idle = uvd_v4_2_wait_for_idle,
735 .soft_reset = uvd_v4_2_soft_reset,
736 .set_clockgating_state = uvd_v4_2_set_clockgating_state,
737 .set_powergating_state = uvd_v4_2_set_powergating_state,
738};
739
740static const struct amdgpu_ring_funcs uvd_v4_2_ring_funcs = {
741 .type = AMDGPU_RING_TYPE_UVD,
742 .align_mask = 0xf,
743 .support_64bit_ptrs = false,
744 .no_user_fence = true,
745 .get_rptr = uvd_v4_2_ring_get_rptr,
746 .get_wptr = uvd_v4_2_ring_get_wptr,
747 .set_wptr = uvd_v4_2_ring_set_wptr,
748 .parse_cs = amdgpu_uvd_ring_parse_cs,
749 .emit_frame_size =
750 14, /* uvd_v4_2_ring_emit_fence x1 no user fence */
751 .emit_ib_size = 4, /* uvd_v4_2_ring_emit_ib */
752 .emit_ib = uvd_v4_2_ring_emit_ib,
753 .emit_fence = uvd_v4_2_ring_emit_fence,
754 .test_ring = uvd_v4_2_ring_test_ring,
755 .test_ib = amdgpu_uvd_ring_test_ib,
756 .insert_nop = uvd_v4_2_ring_insert_nop,
757 .pad_ib = amdgpu_ring_generic_pad_ib,
758 .begin_use = amdgpu_uvd_ring_begin_use,
759 .end_use = amdgpu_uvd_ring_end_use,
760};
761
762static void uvd_v4_2_set_ring_funcs(struct amdgpu_device *adev)
763{
764 adev->uvd.inst->ring.funcs = &uvd_v4_2_ring_funcs;
765}
766
767static const struct amdgpu_irq_src_funcs uvd_v4_2_irq_funcs = {
768 .set = uvd_v4_2_set_interrupt_state,
769 .process = uvd_v4_2_process_interrupt,
770};
771
772static void uvd_v4_2_set_irq_funcs(struct amdgpu_device *adev)
773{
774 adev->uvd.inst->irq.num_types = 1;
775 adev->uvd.inst->irq.funcs = &uvd_v4_2_irq_funcs;
776}
777
778const struct amdgpu_ip_block_version uvd_v4_2_ip_block =
779{
780 .type = AMD_IP_BLOCK_TYPE_UVD,
781 .major = 4,
782 .minor = 2,
783 .rev = 0,
784 .funcs = &uvd_v4_2_ip_funcs,
785};