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_R600		"radeon/R600_uvd.bin"
44#define FIRMWARE_RS780		"radeon/RS780_uvd.bin"
45#define FIRMWARE_RV770		"radeon/RV770_uvd.bin"
46#define FIRMWARE_RV710		"radeon/RV710_uvd.bin"
47#define FIRMWARE_CYPRESS	"radeon/CYPRESS_uvd.bin"
48#define FIRMWARE_SUMO		"radeon/SUMO_uvd.bin"
49#define FIRMWARE_TAHITI		"radeon/TAHITI_uvd.bin"
50#define FIRMWARE_BONAIRE	"radeon/BONAIRE_uvd.bin"
51
52MODULE_FIRMWARE(FIRMWARE_R600);
53MODULE_FIRMWARE(FIRMWARE_RS780);
54MODULE_FIRMWARE(FIRMWARE_RV770);
55MODULE_FIRMWARE(FIRMWARE_RV710);
56MODULE_FIRMWARE(FIRMWARE_CYPRESS);
57MODULE_FIRMWARE(FIRMWARE_SUMO);
58MODULE_FIRMWARE(FIRMWARE_TAHITI);
59MODULE_FIRMWARE(FIRMWARE_BONAIRE);
60
61static void radeon_uvd_idle_work_handler(struct work_struct *work);
62
63int radeon_uvd_init(struct radeon_device *rdev)
64{
65	unsigned long bo_size;
66	const char *fw_name;
67	int i, r;
68
69	INIT_DELAYED_WORK(&rdev->uvd.idle_work, radeon_uvd_idle_work_handler);
70
71	switch (rdev->family) {
72	case CHIP_RV610:
73	case CHIP_RV630:
74	case CHIP_RV670:
75	case CHIP_RV620:
76	case CHIP_RV635:
77		fw_name = FIRMWARE_R600;
78		break;
79
80	case CHIP_RS780:
81	case CHIP_RS880:
82		fw_name = FIRMWARE_RS780;
83		break;
84
85	case CHIP_RV770:
86		fw_name = FIRMWARE_RV770;
87		break;
88
89	case CHIP_RV710:
90	case CHIP_RV730:
91	case CHIP_RV740:
92		fw_name = FIRMWARE_RV710;
93		break;
94
95	case CHIP_CYPRESS:
96	case CHIP_HEMLOCK:
97	case CHIP_JUNIPER:
98	case CHIP_REDWOOD:
99	case CHIP_CEDAR:
100		fw_name = FIRMWARE_CYPRESS;
101		break;
102
103	case CHIP_SUMO:
104	case CHIP_SUMO2:
105	case CHIP_PALM:
106	case CHIP_CAYMAN:
107	case CHIP_BARTS:
108	case CHIP_TURKS:
109	case CHIP_CAICOS:
110		fw_name = FIRMWARE_SUMO;
111		break;
112
113	case CHIP_TAHITI:
114	case CHIP_VERDE:
115	case CHIP_PITCAIRN:
116	case CHIP_ARUBA:
117	case CHIP_OLAND:
118		fw_name = FIRMWARE_TAHITI;
119		break;
120
121	case CHIP_BONAIRE:
122	case CHIP_KABINI:
123	case CHIP_KAVERI:
124	case CHIP_HAWAII:
125	case CHIP_MULLINS:
126		fw_name = FIRMWARE_BONAIRE;
127		break;
128
129	default:
130		return -EINVAL;
131	}
132
133	r = request_firmware(&rdev->uvd_fw, fw_name, rdev->dev);
134	if (r) {
135		dev_err(rdev->dev, "radeon_uvd: Can't load firmware \"%s\"\n",
136			fw_name);
137		return r;
138	}
139
140	bo_size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 8) +
141		  RADEON_UVD_STACK_SIZE + RADEON_UVD_HEAP_SIZE +
142		  RADEON_GPU_PAGE_SIZE;
143	r = radeon_bo_create(rdev, bo_size, PAGE_SIZE, true,
144			     RADEON_GEM_DOMAIN_VRAM, 0, NULL,
145			     NULL, &rdev->uvd.vcpu_bo);
146	if (r) {
147		dev_err(rdev->dev, "(%d) failed to allocate UVD bo\n", r);
148		return r;
149	}
150
151	r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
152	if (r) {
153		radeon_bo_unref(&rdev->uvd.vcpu_bo);
154		dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r);
155		return r;
156	}
157
158	r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
159			  &rdev->uvd.gpu_addr);
160	if (r) {
161		radeon_bo_unreserve(rdev->uvd.vcpu_bo);
162		radeon_bo_unref(&rdev->uvd.vcpu_bo);
163		dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r);
164		return r;
165	}
166
167	r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr);
168	if (r) {
169		dev_err(rdev->dev, "(%d) UVD map failed\n", r);
170		return r;
171	}
172
173	radeon_bo_unreserve(rdev->uvd.vcpu_bo);
174
175	for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
176		atomic_set(&rdev->uvd.handles[i], 0);
177		rdev->uvd.filp[i] = NULL;
178		rdev->uvd.img_size[i] = 0;
179	}
180
181	return 0;
182}
183
184void radeon_uvd_fini(struct radeon_device *rdev)
185{
186	int r;
187
188	if (rdev->uvd.vcpu_bo == NULL)
189		return;
190
191	r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
192	if (!r) {
193		radeon_bo_kunmap(rdev->uvd.vcpu_bo);
194		radeon_bo_unpin(rdev->uvd.vcpu_bo);
195		radeon_bo_unreserve(rdev->uvd.vcpu_bo);
196	}
197
198	radeon_bo_unref(&rdev->uvd.vcpu_bo);
199
200	radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX]);
201
202	release_firmware(rdev->uvd_fw);
203}
204
205int radeon_uvd_suspend(struct radeon_device *rdev)
206{
207	int i, r;
208
209	if (rdev->uvd.vcpu_bo == NULL)
210		return 0;
211
212	for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
213		uint32_t handle = atomic_read(&rdev->uvd.handles[i]);
214		if (handle != 0) {
215			struct radeon_fence *fence;
216
217			radeon_uvd_note_usage(rdev);
218
219			r = radeon_uvd_get_destroy_msg(rdev,
220				R600_RING_TYPE_UVD_INDEX, handle, &fence);
221			if (r) {
222				DRM_ERROR("Error destroying UVD (%d)!\n", r);
223				continue;
224			}
225
226			radeon_fence_wait(fence, false);
227			radeon_fence_unref(&fence);
228
229			rdev->uvd.filp[i] = NULL;
230			atomic_set(&rdev->uvd.handles[i], 0);
231		}
232	}
233
234	return 0;
235}
236
237int radeon_uvd_resume(struct radeon_device *rdev)
238{
239	unsigned size;
240	void *ptr;
241
242	if (rdev->uvd.vcpu_bo == NULL)
243		return -EINVAL;
244
245	memcpy(rdev->uvd.cpu_addr, rdev->uvd_fw->data, rdev->uvd_fw->size);
246
247	size = radeon_bo_size(rdev->uvd.vcpu_bo);
248	size -= rdev->uvd_fw->size;
249
250	ptr = rdev->uvd.cpu_addr;
251	ptr += rdev->uvd_fw->size;
252
253	memset(ptr, 0, size);
254
255	return 0;
256}
257
258void radeon_uvd_force_into_uvd_segment(struct radeon_bo *rbo,
259				       uint32_t allowed_domains)
260{
261	int i;
262
263	for (i = 0; i < rbo->placement.num_placement; ++i) {
264		rbo->placements[i].fpfn = 0 >> PAGE_SHIFT;
265		rbo->placements[i].lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT;
266	}
267
268	/* If it must be in VRAM it must be in the first segment as well */
269	if (allowed_domains == RADEON_GEM_DOMAIN_VRAM)
270		return;
271
272	/* abort if we already have more than one placement */
273	if (rbo->placement.num_placement > 1)
274		return;
275
276	/* add another 256MB segment */
277	rbo->placements[1] = rbo->placements[0];
278	rbo->placements[1].fpfn += (256 * 1024 * 1024) >> PAGE_SHIFT;
279	rbo->placements[1].lpfn += (256 * 1024 * 1024) >> PAGE_SHIFT;
280	rbo->placement.num_placement++;
281	rbo->placement.num_busy_placement++;
282}
283
284void radeon_uvd_free_handles(struct radeon_device *rdev, struct drm_file *filp)
285{
286	int i, r;
287	for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
288		uint32_t handle = atomic_read(&rdev->uvd.handles[i]);
289		if (handle != 0 && rdev->uvd.filp[i] == filp) {
290			struct radeon_fence *fence;
291
292			radeon_uvd_note_usage(rdev);
293
294			r = radeon_uvd_get_destroy_msg(rdev,
295				R600_RING_TYPE_UVD_INDEX, handle, &fence);
296			if (r) {
297				DRM_ERROR("Error destroying UVD (%d)!\n", r);
298				continue;
299			}
300
301			radeon_fence_wait(fence, false);
302			radeon_fence_unref(&fence);
303
304			rdev->uvd.filp[i] = NULL;
305			atomic_set(&rdev->uvd.handles[i], 0);
306		}
307	}
308}
309
310static int radeon_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[])
311{
312	unsigned stream_type = msg[4];
313	unsigned width = msg[6];
314	unsigned height = msg[7];
315	unsigned dpb_size = msg[9];
316	unsigned pitch = msg[28];
317
318	unsigned width_in_mb = width / 16;
319	unsigned height_in_mb = ALIGN(height / 16, 2);
320
321	unsigned image_size, tmp, min_dpb_size;
322
323	image_size = width * height;
324	image_size += image_size / 2;
325	image_size = ALIGN(image_size, 1024);
326
327	switch (stream_type) {
328	case 0: /* H264 */
329
330		/* reference picture buffer */
331		min_dpb_size = image_size * 17;
332
333		/* macroblock context buffer */
334		min_dpb_size += width_in_mb * height_in_mb * 17 * 192;
335
336		/* IT surface buffer */
337		min_dpb_size += width_in_mb * height_in_mb * 32;
338		break;
339
340	case 1: /* VC1 */
341
342		/* reference picture buffer */
343		min_dpb_size = image_size * 3;
344
345		/* CONTEXT_BUFFER */
346		min_dpb_size += width_in_mb * height_in_mb * 128;
347
348		/* IT surface buffer */
349		min_dpb_size += width_in_mb * 64;
350
351		/* DB surface buffer */
352		min_dpb_size += width_in_mb * 128;
353
354		/* BP */
355		tmp = max(width_in_mb, height_in_mb);
356		min_dpb_size += ALIGN(tmp * 7 * 16, 64);
357		break;
358
359	case 3: /* MPEG2 */
360
361		/* reference picture buffer */
362		min_dpb_size = image_size * 3;
363		break;
364
365	case 4: /* MPEG4 */
366
367		/* reference picture buffer */
368		min_dpb_size = image_size * 3;
369
370		/* CM */
371		min_dpb_size += width_in_mb * height_in_mb * 64;
372
373		/* IT surface buffer */
374		min_dpb_size += ALIGN(width_in_mb * height_in_mb * 32, 64);
375		break;
376
377	default:
378		DRM_ERROR("UVD codec not handled %d!\n", stream_type);
379		return -EINVAL;
380	}
381
382	if (width > pitch) {
383		DRM_ERROR("Invalid UVD decoding target pitch!\n");
384		return -EINVAL;
385	}
386
387	if (dpb_size < min_dpb_size) {
388		DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n",
389			  dpb_size, min_dpb_size);
390		return -EINVAL;
391	}
392
393	buf_sizes[0x1] = dpb_size;
394	buf_sizes[0x2] = image_size;
395	return 0;
396}
397
398static int radeon_uvd_validate_codec(struct radeon_cs_parser *p,
399				     unsigned stream_type)
400{
401	switch (stream_type) {
402	case 0: /* H264 */
403	case 1: /* VC1 */
404		/* always supported */
405		return 0;
406
407	case 3: /* MPEG2 */
408	case 4: /* MPEG4 */
409		/* only since UVD 3 */
410		if (p->rdev->family >= CHIP_PALM)
411			return 0;
412
413		/* fall through */
414	default:
415		DRM_ERROR("UVD codec not supported by hardware %d!\n",
416			  stream_type);
417		return -EINVAL;
418	}
419}
420
421static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo,
422			     unsigned offset, unsigned buf_sizes[])
423{
424	int32_t *msg, msg_type, handle;
425	unsigned img_size = 0;
426	struct fence *f;
427	void *ptr;
428
429	int i, r;
430
431	if (offset & 0x3F) {
432		DRM_ERROR("UVD messages must be 64 byte aligned!\n");
433		return -EINVAL;
434	}
435
436	f = reservation_object_get_excl(bo->tbo.resv);
437	if (f) {
438		r = radeon_fence_wait((struct radeon_fence *)f, false);
439		if (r) {
440			DRM_ERROR("Failed waiting for UVD message (%d)!\n", r);
441			return r;
442		}
443	}
444
445	r = radeon_bo_kmap(bo, &ptr);
446	if (r) {
447		DRM_ERROR("Failed mapping the UVD message (%d)!\n", r);
448		return r;
449	}
450
451	msg = ptr + offset;
452
453	msg_type = msg[1];
454	handle = msg[2];
455
456	if (handle == 0) {
457		DRM_ERROR("Invalid UVD handle!\n");
458		return -EINVAL;
459	}
460
461	switch (msg_type) {
462	case 0:
463		/* it's a create msg, calc image size (width * height) */
464		img_size = msg[7] * msg[8];
465
466		r = radeon_uvd_validate_codec(p, msg[4]);
467		radeon_bo_kunmap(bo);
468		if (r)
469			return r;
470
471		/* try to alloc a new handle */
472		for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
473			if (atomic_read(&p->rdev->uvd.handles[i]) == handle) {
474				DRM_ERROR("Handle 0x%x already in use!\n", handle);
475				return -EINVAL;
476			}
477
478			if (!atomic_cmpxchg(&p->rdev->uvd.handles[i], 0, handle)) {
479				p->rdev->uvd.filp[i] = p->filp;
480				p->rdev->uvd.img_size[i] = img_size;
481				return 0;
482			}
483		}
484
485		DRM_ERROR("No more free UVD handles!\n");
486		return -EINVAL;
487
488	case 1:
489		/* it's a decode msg, validate codec and calc buffer sizes */
490		r = radeon_uvd_validate_codec(p, msg[4]);
491		if (!r)
492			r = radeon_uvd_cs_msg_decode(msg, buf_sizes);
493		radeon_bo_kunmap(bo);
494		if (r)
495			return r;
496
497		/* validate the handle */
498		for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
499			if (atomic_read(&p->rdev->uvd.handles[i]) == handle) {
500				if (p->rdev->uvd.filp[i] != p->filp) {
501					DRM_ERROR("UVD handle collision detected!\n");
502					return -EINVAL;
503				}
504				return 0;
505			}
506		}
507
508		DRM_ERROR("Invalid UVD handle 0x%x!\n", handle);
509		return -ENOENT;
510
511	case 2:
512		/* it's a destroy msg, free the handle */
513		for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i)
514			atomic_cmpxchg(&p->rdev->uvd.handles[i], handle, 0);
515		radeon_bo_kunmap(bo);
516		return 0;
517
518	default:
519
520		DRM_ERROR("Illegal UVD message type (%d)!\n", msg_type);
521		return -EINVAL;
522	}
523
524	BUG();
525	return -EINVAL;
526}
527
528static int radeon_uvd_cs_reloc(struct radeon_cs_parser *p,
529			       int data0, int data1,
530			       unsigned buf_sizes[], bool *has_msg_cmd)
531{
532	struct radeon_cs_chunk *relocs_chunk;
533	struct radeon_bo_list *reloc;
534	unsigned idx, cmd, offset;
535	uint64_t start, end;
536	int r;
537
538	relocs_chunk = p->chunk_relocs;
539	offset = radeon_get_ib_value(p, data0);
540	idx = radeon_get_ib_value(p, data1);
541	if (idx >= relocs_chunk->length_dw) {
542		DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
543			  idx, relocs_chunk->length_dw);
544		return -EINVAL;
545	}
546
547	reloc = &p->relocs[(idx / 4)];
548	start = reloc->gpu_offset;
549	end = start + radeon_bo_size(reloc->robj);
550	start += offset;
551
552	p->ib.ptr[data0] = start & 0xFFFFFFFF;
553	p->ib.ptr[data1] = start >> 32;
554
555	cmd = radeon_get_ib_value(p, p->idx) >> 1;
556
557	if (cmd < 0x4) {
558		if (end <= start) {
559			DRM_ERROR("invalid reloc offset %X!\n", offset);
560			return -EINVAL;
561		}
562		if ((end - start) < buf_sizes[cmd]) {
563			DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd,
564				  (unsigned)(end - start), buf_sizes[cmd]);
565			return -EINVAL;
566		}
567
568	} else if (cmd != 0x100) {
569		DRM_ERROR("invalid UVD command %X!\n", cmd);
570		return -EINVAL;
571	}
572
573	if ((start >> 28) != ((end - 1) >> 28)) {
574		DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n",
575			  start, end);
576		return -EINVAL;
577	}
578
579	/* TODO: is this still necessary on NI+ ? */
580	if ((cmd == 0 || cmd == 0x3) &&
581	    (start >> 28) != (p->rdev->uvd.gpu_addr >> 28)) {
582		DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n",
583			  start, end);
584		return -EINVAL;
585	}
586
587	if (cmd == 0) {
588		if (*has_msg_cmd) {
589			DRM_ERROR("More than one message in a UVD-IB!\n");
590			return -EINVAL;
591		}
592		*has_msg_cmd = true;
593		r = radeon_uvd_cs_msg(p, reloc->robj, offset, buf_sizes);
594		if (r)
595			return r;
596	} else if (!*has_msg_cmd) {
597		DRM_ERROR("Message needed before other commands are send!\n");
598		return -EINVAL;
599	}
600
601	return 0;
602}
603
604static int radeon_uvd_cs_reg(struct radeon_cs_parser *p,
605			     struct radeon_cs_packet *pkt,
606			     int *data0, int *data1,
607			     unsigned buf_sizes[],
608			     bool *has_msg_cmd)
609{
610	int i, r;
611
612	p->idx++;
613	for (i = 0; i <= pkt->count; ++i) {
614		switch (pkt->reg + i*4) {
615		case UVD_GPCOM_VCPU_DATA0:
616			*data0 = p->idx;
617			break;
618		case UVD_GPCOM_VCPU_DATA1:
619			*data1 = p->idx;
620			break;
621		case UVD_GPCOM_VCPU_CMD:
622			r = radeon_uvd_cs_reloc(p, *data0, *data1,
623						buf_sizes, has_msg_cmd);
624			if (r)
625				return r;
626			break;
627		case UVD_ENGINE_CNTL:
628			break;
629		default:
630			DRM_ERROR("Invalid reg 0x%X!\n",
631				  pkt->reg + i*4);
632			return -EINVAL;
633		}
634		p->idx++;
635	}
636	return 0;
637}
638
639int radeon_uvd_cs_parse(struct radeon_cs_parser *p)
640{
641	struct radeon_cs_packet pkt;
642	int r, data0 = 0, data1 = 0;
643
644	/* does the IB has a msg command */
645	bool has_msg_cmd = false;
646
647	/* minimum buffer sizes */
648	unsigned buf_sizes[] = {
649		[0x00000000]	=	2048,
650		[0x00000001]	=	32 * 1024 * 1024,
651		[0x00000002]	=	2048 * 1152 * 3,
652		[0x00000003]	=	2048,
653	};
654
655	if (p->chunk_ib->length_dw % 16) {
656		DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n",
657			  p->chunk_ib->length_dw);
658		return -EINVAL;
659	}
660
661	if (p->chunk_relocs == NULL) {
662		DRM_ERROR("No relocation chunk !\n");
663		return -EINVAL;
664	}
665
666
667	do {
668		r = radeon_cs_packet_parse(p, &pkt, p->idx);
669		if (r)
670			return r;
671		switch (pkt.type) {
672		case RADEON_PACKET_TYPE0:
673			r = radeon_uvd_cs_reg(p, &pkt, &data0, &data1,
674					      buf_sizes, &has_msg_cmd);
675			if (r)
676				return r;
677			break;
678		case RADEON_PACKET_TYPE2:
679			p->idx += pkt.count + 2;
680			break;
681		default:
682			DRM_ERROR("Unknown packet type %d !\n", pkt.type);
683			return -EINVAL;
684		}
685	} while (p->idx < p->chunk_ib->length_dw);
686
687	if (!has_msg_cmd) {
688		DRM_ERROR("UVD-IBs need a msg command!\n");
689		return -EINVAL;
690	}
691
692	return 0;
693}
694
695static int radeon_uvd_send_msg(struct radeon_device *rdev,
696			       int ring, uint64_t addr,
697			       struct radeon_fence **fence)
698{
699	struct radeon_ib ib;
700	int i, r;
701
702	r = radeon_ib_get(rdev, ring, &ib, NULL, 64);
703	if (r)
704		return r;
705
706	ib.ptr[0] = PACKET0(UVD_GPCOM_VCPU_DATA0, 0);
707	ib.ptr[1] = addr;
708	ib.ptr[2] = PACKET0(UVD_GPCOM_VCPU_DATA1, 0);
709	ib.ptr[3] = addr >> 32;
710	ib.ptr[4] = PACKET0(UVD_GPCOM_VCPU_CMD, 0);
711	ib.ptr[5] = 0;
712	for (i = 6; i < 16; ++i)
713		ib.ptr[i] = PACKET2(0);
714	ib.length_dw = 16;
715
716	r = radeon_ib_schedule(rdev, &ib, NULL, false);
717
718	if (fence)
719		*fence = radeon_fence_ref(ib.fence);
720
721	radeon_ib_free(rdev, &ib);
722	return r;
723}
724
725/* multiple fence commands without any stream commands in between can
726   crash the vcpu so just try to emmit a dummy create/destroy msg to
727   avoid this */
728int radeon_uvd_get_create_msg(struct radeon_device *rdev, int ring,
729			      uint32_t handle, struct radeon_fence **fence)
730{
731	/* we use the last page of the vcpu bo for the UVD message */
732	uint64_t offs = radeon_bo_size(rdev->uvd.vcpu_bo) -
733		RADEON_GPU_PAGE_SIZE;
734
735	uint32_t *msg = rdev->uvd.cpu_addr + offs;
736	uint64_t addr = rdev->uvd.gpu_addr + offs;
737
738	int r, i;
739
740	r = radeon_bo_reserve(rdev->uvd.vcpu_bo, true);
741	if (r)
742		return r;
743
744	/* stitch together an UVD create msg */
745	msg[0] = cpu_to_le32(0x00000de4);
746	msg[1] = cpu_to_le32(0x00000000);
747	msg[2] = cpu_to_le32(handle);
748	msg[3] = cpu_to_le32(0x00000000);
749	msg[4] = cpu_to_le32(0x00000000);
750	msg[5] = cpu_to_le32(0x00000000);
751	msg[6] = cpu_to_le32(0x00000000);
752	msg[7] = cpu_to_le32(0x00000780);
753	msg[8] = cpu_to_le32(0x00000440);
754	msg[9] = cpu_to_le32(0x00000000);
755	msg[10] = cpu_to_le32(0x01b37000);
756	for (i = 11; i < 1024; ++i)
757		msg[i] = cpu_to_le32(0x0);
758
759	r = radeon_uvd_send_msg(rdev, ring, addr, fence);
760	radeon_bo_unreserve(rdev->uvd.vcpu_bo);
761	return r;
762}
763
764int radeon_uvd_get_destroy_msg(struct radeon_device *rdev, int ring,
765			       uint32_t handle, struct radeon_fence **fence)
766{
767	/* we use the last page of the vcpu bo for the UVD message */
768	uint64_t offs = radeon_bo_size(rdev->uvd.vcpu_bo) -
769		RADEON_GPU_PAGE_SIZE;
770
771	uint32_t *msg = rdev->uvd.cpu_addr + offs;
772	uint64_t addr = rdev->uvd.gpu_addr + offs;
773
774	int r, i;
775
776	r = radeon_bo_reserve(rdev->uvd.vcpu_bo, true);
777	if (r)
778		return r;
779
780	/* stitch together an UVD destroy msg */
781	msg[0] = cpu_to_le32(0x00000de4);
782	msg[1] = cpu_to_le32(0x00000002);
783	msg[2] = cpu_to_le32(handle);
784	msg[3] = cpu_to_le32(0x00000000);
785	for (i = 4; i < 1024; ++i)
786		msg[i] = cpu_to_le32(0x0);
787
788	r = radeon_uvd_send_msg(rdev, ring, addr, fence);
789	radeon_bo_unreserve(rdev->uvd.vcpu_bo);
790	return r;
791}
792
793/**
794 * radeon_uvd_count_handles - count number of open streams
795 *
796 * @rdev: radeon_device pointer
797 * @sd: number of SD streams
798 * @hd: number of HD streams
799 *
800 * Count the number of open SD/HD streams as a hint for power mangement
801 */
802static void radeon_uvd_count_handles(struct radeon_device *rdev,
803				     unsigned *sd, unsigned *hd)
804{
805	unsigned i;
806
807	*sd = 0;
808	*hd = 0;
809
810	for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
811		if (!atomic_read(&rdev->uvd.handles[i]))
812			continue;
813
814		if (rdev->uvd.img_size[i] >= 720*576)
815			++(*hd);
816		else
817			++(*sd);
818	}
819}
820
821static void radeon_uvd_idle_work_handler(struct work_struct *work)
822{
823	struct radeon_device *rdev =
824		container_of(work, struct radeon_device, uvd.idle_work.work);
825
826	if (radeon_fence_count_emitted(rdev, R600_RING_TYPE_UVD_INDEX) == 0) {
827		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
828			radeon_uvd_count_handles(rdev, &rdev->pm.dpm.sd,
829						 &rdev->pm.dpm.hd);
830			radeon_dpm_enable_uvd(rdev, false);
831		} else {
832			radeon_set_uvd_clocks(rdev, 0, 0);
833		}
834	} else {
835		schedule_delayed_work(&rdev->uvd.idle_work,
836				      msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
837	}
838}
839
840void radeon_uvd_note_usage(struct radeon_device *rdev)
841{
842	bool streams_changed = false;
843	bool set_clocks = !cancel_delayed_work_sync(&rdev->uvd.idle_work);
844	set_clocks &= schedule_delayed_work(&rdev->uvd.idle_work,
845					    msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
846
847	if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
848		unsigned hd = 0, sd = 0;
849		radeon_uvd_count_handles(rdev, &sd, &hd);
850		if ((rdev->pm.dpm.sd != sd) ||
851		    (rdev->pm.dpm.hd != hd)) {
852			rdev->pm.dpm.sd = sd;
853			rdev->pm.dpm.hd = hd;
854			/* disable this for now */
855			/*streams_changed = true;*/
856		}
857	}
858
859	if (set_clocks || streams_changed) {
860		if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
861			radeon_dpm_enable_uvd(rdev, true);
862		} else {
863			radeon_set_uvd_clocks(rdev, 53300, 40000);
864		}
865	}
866}
867
868static unsigned radeon_uvd_calc_upll_post_div(unsigned vco_freq,
869					      unsigned target_freq,
870					      unsigned pd_min,
871					      unsigned pd_even)
872{
873	unsigned post_div = vco_freq / target_freq;
874
875	/* adjust to post divider minimum value */
876	if (post_div < pd_min)
877		post_div = pd_min;
878
879	/* we alway need a frequency less than or equal the target */
880	if ((vco_freq / post_div) > target_freq)
881		post_div += 1;
882
883	/* post dividers above a certain value must be even */
884	if (post_div > pd_even && post_div % 2)
885		post_div += 1;
886
887	return post_div;
888}
889
890/**
891 * radeon_uvd_calc_upll_dividers - calc UPLL clock dividers
892 *
893 * @rdev: radeon_device pointer
894 * @vclk: wanted VCLK
895 * @dclk: wanted DCLK
896 * @vco_min: minimum VCO frequency
897 * @vco_max: maximum VCO frequency
898 * @fb_factor: factor to multiply vco freq with
899 * @fb_mask: limit and bitmask for feedback divider
900 * @pd_min: post divider minimum
901 * @pd_max: post divider maximum
902 * @pd_even: post divider must be even above this value
903 * @optimal_fb_div: resulting feedback divider
904 * @optimal_vclk_div: resulting vclk post divider
905 * @optimal_dclk_div: resulting dclk post divider
906 *
907 * Calculate dividers for UVDs UPLL (R6xx-SI, except APUs).
908 * Returns zero on success -EINVAL on error.
909 */
910int radeon_uvd_calc_upll_dividers(struct radeon_device *rdev,
911				  unsigned vclk, unsigned dclk,
912				  unsigned vco_min, unsigned vco_max,
913				  unsigned fb_factor, unsigned fb_mask,
914				  unsigned pd_min, unsigned pd_max,
915				  unsigned pd_even,
916				  unsigned *optimal_fb_div,
917				  unsigned *optimal_vclk_div,
918				  unsigned *optimal_dclk_div)
919{
920	unsigned vco_freq, ref_freq = rdev->clock.spll.reference_freq;
921
922	/* start off with something large */
923	unsigned optimal_score = ~0;
924
925	/* loop through vco from low to high */
926	vco_min = max(max(vco_min, vclk), dclk);
927	for (vco_freq = vco_min; vco_freq <= vco_max; vco_freq += 100) {
928
929		uint64_t fb_div = (uint64_t)vco_freq * fb_factor;
930		unsigned vclk_div, dclk_div, score;
931
932		do_div(fb_div, ref_freq);
933
934		/* fb div out of range ? */
935		if (fb_div > fb_mask)
936			break; /* it can oly get worse */
937
938		fb_div &= fb_mask;
939
940		/* calc vclk divider with current vco freq */
941		vclk_div = radeon_uvd_calc_upll_post_div(vco_freq, vclk,
942							 pd_min, pd_even);
943		if (vclk_div > pd_max)
944			break; /* vco is too big, it has to stop */
945
946		/* calc dclk divider with current vco freq */
947		dclk_div = radeon_uvd_calc_upll_post_div(vco_freq, dclk,
948							 pd_min, pd_even);
949		if (vclk_div > pd_max)
950			break; /* vco is too big, it has to stop */
951
952		/* calc score with current vco freq */
953		score = vclk - (vco_freq / vclk_div) + dclk - (vco_freq / dclk_div);
954
955		/* determine if this vco setting is better than current optimal settings */
956		if (score < optimal_score) {
957			*optimal_fb_div = fb_div;
958			*optimal_vclk_div = vclk_div;
959			*optimal_dclk_div = dclk_div;
960			optimal_score = score;
961			if (optimal_score == 0)
962				break; /* it can't get better than this */
963		}
964	}
965
966	/* did we found a valid setup ? */
967	if (optimal_score == ~0)
968		return -EINVAL;
969
970	return 0;
971}
972
973int radeon_uvd_send_upll_ctlreq(struct radeon_device *rdev,
974				unsigned cg_upll_func_cntl)
975{
976	unsigned i;
977
978	/* make sure UPLL_CTLREQ is deasserted */
979	WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK);
980
981	mdelay(10);
982
983	/* assert UPLL_CTLREQ */
984	WREG32_P(cg_upll_func_cntl, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK);
985
986	/* wait for CTLACK and CTLACK2 to get asserted */
987	for (i = 0; i < 100; ++i) {
988		uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK;
989		if ((RREG32(cg_upll_func_cntl) & mask) == mask)
990			break;
991		mdelay(10);
992	}
993
994	/* deassert UPLL_CTLREQ */
995	WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK);
996
997	if (i == 100) {
998		DRM_ERROR("Timeout setting UVD clocks!\n");
999		return -ETIMEDOUT;
1000	}
1001
1002	return 0;
1003}
1004