Merge remote-tracking branch 'lsk/v3.10/topic/gator' into linux-linaro-lsk
[firefly-linux-kernel-4.4.55.git] / drivers / gpu / drm / radeon / radeon_uvd.c
1 /*
2  * Copyright 2011 Advanced Micro Devices, Inc.
3  * All Rights Reserved.
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the
7  * "Software"), to deal in the Software without restriction, including
8  * without limitation the rights to use, copy, modify, merge, publish,
9  * distribute, sub license, and/or sell copies of the Software, and to
10  * permit persons to whom the Software is furnished to do so, subject to
11  * the following conditions:
12  *
13  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
16  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
17  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
18  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
19  * USE OR OTHER DEALINGS IN THE SOFTWARE.
20  *
21  * The above copyright notice and this permission notice (including the
22  * next paragraph) shall be included in all copies or substantial portions
23  * of the Software.
24  *
25  */
26 /*
27  * Authors:
28  *    Christian König <deathsimple@vodafone.de>
29  */
30
31 #include <linux/firmware.h>
32 #include <linux/module.h>
33 #include <drm/drmP.h>
34 #include <drm/drm.h>
35
36 #include "radeon.h"
37 #include "r600d.h"
38
39 /* 1 second timeout */
40 #define UVD_IDLE_TIMEOUT_MS     1000
41
42 /* Firmware Names */
43 #define FIRMWARE_RV710          "radeon/RV710_uvd.bin"
44 #define FIRMWARE_CYPRESS        "radeon/CYPRESS_uvd.bin"
45 #define FIRMWARE_SUMO           "radeon/SUMO_uvd.bin"
46 #define FIRMWARE_TAHITI         "radeon/TAHITI_uvd.bin"
47
48 MODULE_FIRMWARE(FIRMWARE_RV710);
49 MODULE_FIRMWARE(FIRMWARE_CYPRESS);
50 MODULE_FIRMWARE(FIRMWARE_SUMO);
51 MODULE_FIRMWARE(FIRMWARE_TAHITI);
52
53 static void radeon_uvd_idle_work_handler(struct work_struct *work);
54
55 int radeon_uvd_init(struct radeon_device *rdev)
56 {
57         struct platform_device *pdev;
58         unsigned long bo_size;
59         const char *fw_name;
60         int i, r;
61
62         INIT_DELAYED_WORK(&rdev->uvd.idle_work, radeon_uvd_idle_work_handler);
63
64         pdev = platform_device_register_simple("radeon_uvd", 0, NULL, 0);
65         r = IS_ERR(pdev);
66         if (r) {
67                 dev_err(rdev->dev, "radeon_uvd: Failed to register firmware\n");
68                 return -EINVAL;
69         }
70
71         switch (rdev->family) {
72         case CHIP_RV710:
73         case CHIP_RV730:
74         case CHIP_RV740:
75                 fw_name = FIRMWARE_RV710;
76                 break;
77
78         case CHIP_CYPRESS:
79         case CHIP_HEMLOCK:
80         case CHIP_JUNIPER:
81         case CHIP_REDWOOD:
82         case CHIP_CEDAR:
83                 fw_name = FIRMWARE_CYPRESS;
84                 break;
85
86         case CHIP_SUMO:
87         case CHIP_SUMO2:
88         case CHIP_PALM:
89         case CHIP_CAYMAN:
90         case CHIP_BARTS:
91         case CHIP_TURKS:
92         case CHIP_CAICOS:
93                 fw_name = FIRMWARE_SUMO;
94                 break;
95
96         case CHIP_TAHITI:
97         case CHIP_VERDE:
98         case CHIP_PITCAIRN:
99         case CHIP_ARUBA:
100                 fw_name = FIRMWARE_TAHITI;
101                 break;
102
103         default:
104                 return -EINVAL;
105         }
106
107         r = request_firmware(&rdev->uvd_fw, fw_name, &pdev->dev);
108         if (r) {
109                 dev_err(rdev->dev, "radeon_uvd: Can't load firmware \"%s\"\n",
110                         fw_name);
111                 platform_device_unregister(pdev);
112                 return r;
113         }
114
115         platform_device_unregister(pdev);
116
117         bo_size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 8) +
118                   RADEON_UVD_STACK_SIZE + RADEON_UVD_HEAP_SIZE;
119         r = radeon_bo_create(rdev, bo_size, PAGE_SIZE, true,
120                              RADEON_GEM_DOMAIN_VRAM, NULL, &rdev->uvd.vcpu_bo);
121         if (r) {
122                 dev_err(rdev->dev, "(%d) failed to allocate UVD bo\n", r);
123                 return r;
124         }
125
126         r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
127         if (r) {
128                 radeon_bo_unref(&rdev->uvd.vcpu_bo);
129                 dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r);
130                 return r;
131         }
132
133         r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
134                           &rdev->uvd.gpu_addr);
135         if (r) {
136                 radeon_bo_unreserve(rdev->uvd.vcpu_bo);
137                 radeon_bo_unref(&rdev->uvd.vcpu_bo);
138                 dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r);
139                 return r;
140         }
141
142         r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr);
143         if (r) {
144                 dev_err(rdev->dev, "(%d) UVD map failed\n", r);
145                 return r;
146         }
147
148         radeon_bo_unreserve(rdev->uvd.vcpu_bo);
149
150         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
151                 atomic_set(&rdev->uvd.handles[i], 0);
152                 rdev->uvd.filp[i] = NULL;
153         }
154
155         return 0;
156 }
157
158 void radeon_uvd_fini(struct radeon_device *rdev)
159 {
160         int r;
161
162         if (rdev->uvd.vcpu_bo == NULL)
163                 return;
164
165         r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
166         if (!r) {
167                 radeon_bo_kunmap(rdev->uvd.vcpu_bo);
168                 radeon_bo_unpin(rdev->uvd.vcpu_bo);
169                 radeon_bo_unreserve(rdev->uvd.vcpu_bo);
170         }
171
172         radeon_bo_unref(&rdev->uvd.vcpu_bo);
173
174         release_firmware(rdev->uvd_fw);
175 }
176
177 int radeon_uvd_suspend(struct radeon_device *rdev)
178 {
179         unsigned size;
180         void *ptr;
181         int i;
182
183         if (rdev->uvd.vcpu_bo == NULL)
184                 return 0;
185
186         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i)
187                 if (atomic_read(&rdev->uvd.handles[i]))
188                         break;
189
190         if (i == RADEON_MAX_UVD_HANDLES)
191                 return 0;
192
193         size = radeon_bo_size(rdev->uvd.vcpu_bo);
194         size -= rdev->uvd_fw->size;
195
196         ptr = rdev->uvd.cpu_addr;
197         ptr += rdev->uvd_fw->size;
198
199         rdev->uvd.saved_bo = kmalloc(size, GFP_KERNEL);
200         memcpy(rdev->uvd.saved_bo, ptr, size);
201
202         return 0;
203 }
204
205 int radeon_uvd_resume(struct radeon_device *rdev)
206 {
207         unsigned size;
208         void *ptr;
209
210         if (rdev->uvd.vcpu_bo == NULL)
211                 return -EINVAL;
212
213         memcpy(rdev->uvd.cpu_addr, rdev->uvd_fw->data, rdev->uvd_fw->size);
214
215         size = radeon_bo_size(rdev->uvd.vcpu_bo);
216         size -= rdev->uvd_fw->size;
217
218         ptr = rdev->uvd.cpu_addr;
219         ptr += rdev->uvd_fw->size;
220
221         if (rdev->uvd.saved_bo != NULL) {
222                 memcpy(ptr, rdev->uvd.saved_bo, size);
223                 kfree(rdev->uvd.saved_bo);
224                 rdev->uvd.saved_bo = NULL;
225         } else
226                 memset(ptr, 0, size);
227
228         return 0;
229 }
230
231 void radeon_uvd_force_into_uvd_segment(struct radeon_bo *rbo)
232 {
233         rbo->placement.fpfn = 0 >> PAGE_SHIFT;
234         rbo->placement.lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT;
235 }
236
237 void radeon_uvd_free_handles(struct radeon_device *rdev, struct drm_file *filp)
238 {
239         int i, r;
240         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
241                 uint32_t handle = atomic_read(&rdev->uvd.handles[i]);
242                 if (handle != 0 && rdev->uvd.filp[i] == filp) {
243                         struct radeon_fence *fence;
244
245                         radeon_uvd_note_usage(rdev);
246
247                         r = radeon_uvd_get_destroy_msg(rdev,
248                                 R600_RING_TYPE_UVD_INDEX, handle, &fence);
249                         if (r) {
250                                 DRM_ERROR("Error destroying UVD (%d)!\n", r);
251                                 continue;
252                         }
253
254                         radeon_fence_wait(fence, false);
255                         radeon_fence_unref(&fence);
256
257                         rdev->uvd.filp[i] = NULL;
258                         atomic_set(&rdev->uvd.handles[i], 0);
259                 }
260         }
261 }
262
263 static int radeon_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[])
264 {
265         unsigned stream_type = msg[4];
266         unsigned width = msg[6];
267         unsigned height = msg[7];
268         unsigned dpb_size = msg[9];
269         unsigned pitch = msg[28];
270
271         unsigned width_in_mb = width / 16;
272         unsigned height_in_mb = ALIGN(height / 16, 2);
273
274         unsigned image_size, tmp, min_dpb_size;
275
276         image_size = width * height;
277         image_size += image_size / 2;
278         image_size = ALIGN(image_size, 1024);
279
280         switch (stream_type) {
281         case 0: /* H264 */
282
283                 /* reference picture buffer */
284                 min_dpb_size = image_size * 17;
285
286                 /* macroblock context buffer */
287                 min_dpb_size += width_in_mb * height_in_mb * 17 * 192;
288
289                 /* IT surface buffer */
290                 min_dpb_size += width_in_mb * height_in_mb * 32;
291                 break;
292
293         case 1: /* VC1 */
294
295                 /* reference picture buffer */
296                 min_dpb_size = image_size * 3;
297
298                 /* CONTEXT_BUFFER */
299                 min_dpb_size += width_in_mb * height_in_mb * 128;
300
301                 /* IT surface buffer */
302                 min_dpb_size += width_in_mb * 64;
303
304                 /* DB surface buffer */
305                 min_dpb_size += width_in_mb * 128;
306
307                 /* BP */
308                 tmp = max(width_in_mb, height_in_mb);
309                 min_dpb_size += ALIGN(tmp * 7 * 16, 64);
310                 break;
311
312         case 3: /* MPEG2 */
313
314                 /* reference picture buffer */
315                 min_dpb_size = image_size * 3;
316                 break;
317
318         case 4: /* MPEG4 */
319
320                 /* reference picture buffer */
321                 min_dpb_size = image_size * 3;
322
323                 /* CM */
324                 min_dpb_size += width_in_mb * height_in_mb * 64;
325
326                 /* IT surface buffer */
327                 min_dpb_size += ALIGN(width_in_mb * height_in_mb * 32, 64);
328                 break;
329
330         default:
331                 DRM_ERROR("UVD codec not handled %d!\n", stream_type);
332                 return -EINVAL;
333         }
334
335         if (width > pitch) {
336                 DRM_ERROR("Invalid UVD decoding target pitch!\n");
337                 return -EINVAL;
338         }
339
340         if (dpb_size < min_dpb_size) {
341                 DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n",
342                           dpb_size, min_dpb_size);
343                 return -EINVAL;
344         }
345
346         buf_sizes[0x1] = dpb_size;
347         buf_sizes[0x2] = image_size;
348         return 0;
349 }
350
351 static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo,
352                              unsigned offset, unsigned buf_sizes[])
353 {
354         int32_t *msg, msg_type, handle;
355         void *ptr;
356
357         int i, r;
358
359         if (offset & 0x3F) {
360                 DRM_ERROR("UVD messages must be 64 byte aligned!\n");
361                 return -EINVAL;
362         }
363
364         if (bo->tbo.sync_obj) {
365                 r = radeon_fence_wait(bo->tbo.sync_obj, false);
366                 if (r) {
367                         DRM_ERROR("Failed waiting for UVD message (%d)!\n", r);
368                         return r;
369                 }
370         }
371
372         r = radeon_bo_kmap(bo, &ptr);
373         if (r)
374                 return r;
375
376         msg = ptr + offset;
377
378         msg_type = msg[1];
379         handle = msg[2];
380
381         if (handle == 0) {
382                 DRM_ERROR("Invalid UVD handle!\n");
383                 return -EINVAL;
384         }
385
386         if (msg_type == 1) {
387                 /* it's a decode msg, calc buffer sizes */
388                 r = radeon_uvd_cs_msg_decode(msg, buf_sizes);
389                 radeon_bo_kunmap(bo);
390                 if (r)
391                         return r;
392
393         } else if (msg_type == 2) {
394                 /* it's a destroy msg, free the handle */
395                 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i)
396                         atomic_cmpxchg(&p->rdev->uvd.handles[i], handle, 0);
397                 radeon_bo_kunmap(bo);
398                 return 0;
399         } else {
400                 /* it's a create msg, no special handling needed */
401                 radeon_bo_kunmap(bo);
402         }
403
404         /* create or decode, validate the handle */
405         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
406                 if (atomic_read(&p->rdev->uvd.handles[i]) == handle)
407                         return 0;
408         }
409
410         /* handle not found try to alloc a new one */
411         for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
412                 if (!atomic_cmpxchg(&p->rdev->uvd.handles[i], 0, handle)) {
413                         p->rdev->uvd.filp[i] = p->filp;
414                         return 0;
415                 }
416         }
417
418         DRM_ERROR("No more free UVD handles!\n");
419         return -EINVAL;
420 }
421
422 static int radeon_uvd_cs_reloc(struct radeon_cs_parser *p,
423                                int data0, int data1,
424                                unsigned buf_sizes[])
425 {
426         struct radeon_cs_chunk *relocs_chunk;
427         struct radeon_cs_reloc *reloc;
428         unsigned idx, cmd, offset;
429         uint64_t start, end;
430         int r;
431
432         relocs_chunk = &p->chunks[p->chunk_relocs_idx];
433         offset = radeon_get_ib_value(p, data0);
434         idx = radeon_get_ib_value(p, data1);
435         if (idx >= relocs_chunk->length_dw) {
436                 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
437                           idx, relocs_chunk->length_dw);
438                 return -EINVAL;
439         }
440
441         reloc = p->relocs_ptr[(idx / 4)];
442         start = reloc->lobj.gpu_offset;
443         end = start + radeon_bo_size(reloc->robj);
444         start += offset;
445
446         p->ib.ptr[data0] = start & 0xFFFFFFFF;
447         p->ib.ptr[data1] = start >> 32;
448
449         cmd = radeon_get_ib_value(p, p->idx) >> 1;
450
451         if (cmd < 0x4) {
452                 if (end <= start) {
453                         DRM_ERROR("invalid reloc offset %X!\n", offset);
454                         return -EINVAL;
455                 }
456                 if ((end - start) < buf_sizes[cmd]) {
457                         DRM_ERROR("buffer to small (%d / %d)!\n",
458                                   (unsigned)(end - start), buf_sizes[cmd]);
459                         return -EINVAL;
460                 }
461
462         } else if (cmd != 0x100) {
463                 DRM_ERROR("invalid UVD command %X!\n", cmd);
464                 return -EINVAL;
465         }
466
467         if ((start >> 28) != ((end - 1) >> 28)) {
468                 DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n",
469                           start, end);
470                 return -EINVAL;
471         }
472
473         /* TODO: is this still necessary on NI+ ? */
474         if ((cmd == 0 || cmd == 0x3) &&
475             (start >> 28) != (p->rdev->uvd.gpu_addr >> 28)) {
476                 DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n",
477                           start, end);
478                 return -EINVAL;
479         }
480
481         if (cmd == 0) {
482                 r = radeon_uvd_cs_msg(p, reloc->robj, offset, buf_sizes);
483                 if (r)
484                         return r;
485         }
486
487         return 0;
488 }
489
490 static int radeon_uvd_cs_reg(struct radeon_cs_parser *p,
491                              struct radeon_cs_packet *pkt,
492                              int *data0, int *data1,
493                              unsigned buf_sizes[])
494 {
495         int i, r;
496
497         p->idx++;
498         for (i = 0; i <= pkt->count; ++i) {
499                 switch (pkt->reg + i*4) {
500                 case UVD_GPCOM_VCPU_DATA0:
501                         *data0 = p->idx;
502                         break;
503                 case UVD_GPCOM_VCPU_DATA1:
504                         *data1 = p->idx;
505                         break;
506                 case UVD_GPCOM_VCPU_CMD:
507                         r = radeon_uvd_cs_reloc(p, *data0, *data1, buf_sizes);
508                         if (r)
509                                 return r;
510                         break;
511                 case UVD_ENGINE_CNTL:
512                         break;
513                 default:
514                         DRM_ERROR("Invalid reg 0x%X!\n",
515                                   pkt->reg + i*4);
516                         return -EINVAL;
517                 }
518                 p->idx++;
519         }
520         return 0;
521 }
522
523 int radeon_uvd_cs_parse(struct radeon_cs_parser *p)
524 {
525         struct radeon_cs_packet pkt;
526         int r, data0 = 0, data1 = 0;
527
528         /* minimum buffer sizes */
529         unsigned buf_sizes[] = {
530                 [0x00000000]    =       2048,
531                 [0x00000001]    =       32 * 1024 * 1024,
532                 [0x00000002]    =       2048 * 1152 * 3,
533                 [0x00000003]    =       2048,
534         };
535
536         if (p->chunks[p->chunk_ib_idx].length_dw % 16) {
537                 DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n",
538                           p->chunks[p->chunk_ib_idx].length_dw);
539                 return -EINVAL;
540         }
541
542         if (p->chunk_relocs_idx == -1) {
543                 DRM_ERROR("No relocation chunk !\n");
544                 return -EINVAL;
545         }
546
547
548         do {
549                 r = radeon_cs_packet_parse(p, &pkt, p->idx);
550                 if (r)
551                         return r;
552                 switch (pkt.type) {
553                 case RADEON_PACKET_TYPE0:
554                         r = radeon_uvd_cs_reg(p, &pkt, &data0,
555                                               &data1, buf_sizes);
556                         if (r)
557                                 return r;
558                         break;
559                 case RADEON_PACKET_TYPE2:
560                         p->idx += pkt.count + 2;
561                         break;
562                 default:
563                         DRM_ERROR("Unknown packet type %d !\n", pkt.type);
564                         return -EINVAL;
565                 }
566         } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
567         return 0;
568 }
569
570 static int radeon_uvd_send_msg(struct radeon_device *rdev,
571                                int ring, struct radeon_bo *bo,
572                                struct radeon_fence **fence)
573 {
574         struct ttm_validate_buffer tv;
575         struct list_head head;
576         struct radeon_ib ib;
577         uint64_t addr;
578         int i, r;
579
580         memset(&tv, 0, sizeof(tv));
581         tv.bo = &bo->tbo;
582
583         INIT_LIST_HEAD(&head);
584         list_add(&tv.head, &head);
585
586         r = ttm_eu_reserve_buffers(&head);
587         if (r)
588                 return r;
589
590         radeon_ttm_placement_from_domain(bo, RADEON_GEM_DOMAIN_VRAM);
591         radeon_uvd_force_into_uvd_segment(bo);
592
593         r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
594         if (r) {
595                 ttm_eu_backoff_reservation(&head);
596                 return r;
597         }
598
599         r = radeon_ib_get(rdev, ring, &ib, NULL, 16);
600         if (r) {
601                 ttm_eu_backoff_reservation(&head);
602                 return r;
603         }
604
605         addr = radeon_bo_gpu_offset(bo);
606         ib.ptr[0] = PACKET0(UVD_GPCOM_VCPU_DATA0, 0);
607         ib.ptr[1] = addr;
608         ib.ptr[2] = PACKET0(UVD_GPCOM_VCPU_DATA1, 0);
609         ib.ptr[3] = addr >> 32;
610         ib.ptr[4] = PACKET0(UVD_GPCOM_VCPU_CMD, 0);
611         ib.ptr[5] = 0;
612         for (i = 6; i < 16; ++i)
613                 ib.ptr[i] = PACKET2(0);
614         ib.length_dw = 16;
615
616         r = radeon_ib_schedule(rdev, &ib, NULL);
617         if (r) {
618                 ttm_eu_backoff_reservation(&head);
619                 return r;
620         }
621         ttm_eu_fence_buffer_objects(&head, ib.fence);
622
623         if (fence)
624                 *fence = radeon_fence_ref(ib.fence);
625
626         radeon_ib_free(rdev, &ib);
627         radeon_bo_unref(&bo);
628         return 0;
629 }
630
631 /* multiple fence commands without any stream commands in between can
632    crash the vcpu so just try to emmit a dummy create/destroy msg to
633    avoid this */
634 int radeon_uvd_get_create_msg(struct radeon_device *rdev, int ring,
635                               uint32_t handle, struct radeon_fence **fence)
636 {
637         struct radeon_bo *bo;
638         uint32_t *msg;
639         int r, i;
640
641         r = radeon_bo_create(rdev, 1024, PAGE_SIZE, true,
642                              RADEON_GEM_DOMAIN_VRAM, NULL, &bo);
643         if (r)
644                 return r;
645
646         r = radeon_bo_reserve(bo, false);
647         if (r) {
648                 radeon_bo_unref(&bo);
649                 return r;
650         }
651
652         r = radeon_bo_kmap(bo, (void **)&msg);
653         if (r) {
654                 radeon_bo_unreserve(bo);
655                 radeon_bo_unref(&bo);
656                 return r;
657         }
658
659         /* stitch together an UVD create msg */
660         msg[0] = cpu_to_le32(0x00000de4);
661         msg[1] = cpu_to_le32(0x00000000);
662         msg[2] = cpu_to_le32(handle);
663         msg[3] = cpu_to_le32(0x00000000);
664         msg[4] = cpu_to_le32(0x00000000);
665         msg[5] = cpu_to_le32(0x00000000);
666         msg[6] = cpu_to_le32(0x00000000);
667         msg[7] = cpu_to_le32(0x00000780);
668         msg[8] = cpu_to_le32(0x00000440);
669         msg[9] = cpu_to_le32(0x00000000);
670         msg[10] = cpu_to_le32(0x01b37000);
671         for (i = 11; i < 1024; ++i)
672                 msg[i] = cpu_to_le32(0x0);
673
674         radeon_bo_kunmap(bo);
675         radeon_bo_unreserve(bo);
676
677         return radeon_uvd_send_msg(rdev, ring, bo, fence);
678 }
679
680 int radeon_uvd_get_destroy_msg(struct radeon_device *rdev, int ring,
681                                uint32_t handle, struct radeon_fence **fence)
682 {
683         struct radeon_bo *bo;
684         uint32_t *msg;
685         int r, i;
686
687         r = radeon_bo_create(rdev, 1024, PAGE_SIZE, true,
688                              RADEON_GEM_DOMAIN_VRAM, NULL, &bo);
689         if (r)
690                 return r;
691
692         r = radeon_bo_reserve(bo, false);
693         if (r) {
694                 radeon_bo_unref(&bo);
695                 return r;
696         }
697
698         r = radeon_bo_kmap(bo, (void **)&msg);
699         if (r) {
700                 radeon_bo_unreserve(bo);
701                 radeon_bo_unref(&bo);
702                 return r;
703         }
704
705         /* stitch together an UVD destroy msg */
706         msg[0] = cpu_to_le32(0x00000de4);
707         msg[1] = cpu_to_le32(0x00000002);
708         msg[2] = cpu_to_le32(handle);
709         msg[3] = cpu_to_le32(0x00000000);
710         for (i = 4; i < 1024; ++i)
711                 msg[i] = cpu_to_le32(0x0);
712
713         radeon_bo_kunmap(bo);
714         radeon_bo_unreserve(bo);
715
716         return radeon_uvd_send_msg(rdev, ring, bo, fence);
717 }
718
719 static void radeon_uvd_idle_work_handler(struct work_struct *work)
720 {
721         struct radeon_device *rdev =
722                 container_of(work, struct radeon_device, uvd.idle_work.work);
723
724         if (radeon_fence_count_emitted(rdev, R600_RING_TYPE_UVD_INDEX) == 0)
725                 radeon_set_uvd_clocks(rdev, 0, 0);
726         else
727                 schedule_delayed_work(&rdev->uvd.idle_work,
728                                       msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
729 }
730
731 void radeon_uvd_note_usage(struct radeon_device *rdev)
732 {
733         bool set_clocks = !cancel_delayed_work_sync(&rdev->uvd.idle_work);
734         set_clocks &= schedule_delayed_work(&rdev->uvd.idle_work,
735                                             msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
736         if (set_clocks)
737                 radeon_set_uvd_clocks(rdev, 53300, 40000);
738 }
739
740 static unsigned radeon_uvd_calc_upll_post_div(unsigned vco_freq,
741                                               unsigned target_freq,
742                                               unsigned pd_min,
743                                               unsigned pd_even)
744 {
745         unsigned post_div = vco_freq / target_freq;
746
747         /* adjust to post divider minimum value */
748         if (post_div < pd_min)
749                 post_div = pd_min;
750
751         /* we alway need a frequency less than or equal the target */
752         if ((vco_freq / post_div) > target_freq)
753                 post_div += 1;
754
755         /* post dividers above a certain value must be even */
756         if (post_div > pd_even && post_div % 2)
757                 post_div += 1;
758
759         return post_div;
760 }
761
762 /**
763  * radeon_uvd_calc_upll_dividers - calc UPLL clock dividers
764  *
765  * @rdev: radeon_device pointer
766  * @vclk: wanted VCLK
767  * @dclk: wanted DCLK
768  * @vco_min: minimum VCO frequency
769  * @vco_max: maximum VCO frequency
770  * @fb_factor: factor to multiply vco freq with
771  * @fb_mask: limit and bitmask for feedback divider
772  * @pd_min: post divider minimum
773  * @pd_max: post divider maximum
774  * @pd_even: post divider must be even above this value
775  * @optimal_fb_div: resulting feedback divider
776  * @optimal_vclk_div: resulting vclk post divider
777  * @optimal_dclk_div: resulting dclk post divider
778  *
779  * Calculate dividers for UVDs UPLL (R6xx-SI, except APUs).
780  * Returns zero on success -EINVAL on error.
781  */
782 int radeon_uvd_calc_upll_dividers(struct radeon_device *rdev,
783                                   unsigned vclk, unsigned dclk,
784                                   unsigned vco_min, unsigned vco_max,
785                                   unsigned fb_factor, unsigned fb_mask,
786                                   unsigned pd_min, unsigned pd_max,
787                                   unsigned pd_even,
788                                   unsigned *optimal_fb_div,
789                                   unsigned *optimal_vclk_div,
790                                   unsigned *optimal_dclk_div)
791 {
792         unsigned vco_freq, ref_freq = rdev->clock.spll.reference_freq;
793
794         /* start off with something large */
795         unsigned optimal_score = ~0;
796
797         /* loop through vco from low to high */
798         vco_min = max(max(vco_min, vclk), dclk);
799         for (vco_freq = vco_min; vco_freq <= vco_max; vco_freq += 100) {
800
801                 uint64_t fb_div = (uint64_t)vco_freq * fb_factor;
802                 unsigned vclk_div, dclk_div, score;
803
804                 do_div(fb_div, ref_freq);
805
806                 /* fb div out of range ? */
807                 if (fb_div > fb_mask)
808                         break; /* it can oly get worse */
809
810                 fb_div &= fb_mask;
811
812                 /* calc vclk divider with current vco freq */
813                 vclk_div = radeon_uvd_calc_upll_post_div(vco_freq, vclk,
814                                                          pd_min, pd_even);
815                 if (vclk_div > pd_max)
816                         break; /* vco is too big, it has to stop */
817
818                 /* calc dclk divider with current vco freq */
819                 dclk_div = radeon_uvd_calc_upll_post_div(vco_freq, dclk,
820                                                          pd_min, pd_even);
821                 if (vclk_div > pd_max)
822                         break; /* vco is too big, it has to stop */
823
824                 /* calc score with current vco freq */
825                 score = vclk - (vco_freq / vclk_div) + dclk - (vco_freq / dclk_div);
826
827                 /* determine if this vco setting is better than current optimal settings */
828                 if (score < optimal_score) {
829                         *optimal_fb_div = fb_div;
830                         *optimal_vclk_div = vclk_div;
831                         *optimal_dclk_div = dclk_div;
832                         optimal_score = score;
833                         if (optimal_score == 0)
834                                 break; /* it can't get better than this */
835                 }
836         }
837
838         /* did we found a valid setup ? */
839         if (optimal_score == ~0)
840                 return -EINVAL;
841
842         return 0;
843 }
844
845 int radeon_uvd_send_upll_ctlreq(struct radeon_device *rdev,
846                                 unsigned cg_upll_func_cntl)
847 {
848         unsigned i;
849
850         /* make sure UPLL_CTLREQ is deasserted */
851         WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK);
852
853         mdelay(10);
854
855         /* assert UPLL_CTLREQ */
856         WREG32_P(cg_upll_func_cntl, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK);
857
858         /* wait for CTLACK and CTLACK2 to get asserted */
859         for (i = 0; i < 100; ++i) {
860                 uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK;
861                 if ((RREG32(cg_upll_func_cntl) & mask) == mask)
862                         break;
863                 mdelay(10);
864         }
865
866         /* deassert UPLL_CTLREQ */
867         WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK);
868
869         if (i == 100) {
870                 DRM_ERROR("Timeout setting UVD clocks!\n");
871                 return -ETIMEDOUT;
872         }
873
874         return 0;
875 }