2 * Copyright 2011 Advanced Micro Devices, Inc.
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:
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.
21 * The above copyright notice and this permission notice (including the
22 * next paragraph) shall be included in all copies or substantial portions
28 * Christian König <deathsimple@vodafone.de>
31 #include <linux/firmware.h>
32 #include <linux/module.h>
39 /* 1 second timeout */
40 #define UVD_IDLE_TIMEOUT_MS 1000
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"
52 MODULE_FIRMWARE(FIRMWARE_R600);
53 MODULE_FIRMWARE(FIRMWARE_RS780);
54 MODULE_FIRMWARE(FIRMWARE_RV770);
55 MODULE_FIRMWARE(FIRMWARE_RV710);
56 MODULE_FIRMWARE(FIRMWARE_CYPRESS);
57 MODULE_FIRMWARE(FIRMWARE_SUMO);
58 MODULE_FIRMWARE(FIRMWARE_TAHITI);
59 MODULE_FIRMWARE(FIRMWARE_BONAIRE);
61 static void radeon_uvd_idle_work_handler(struct work_struct *work);
63 int radeon_uvd_init(struct radeon_device *rdev)
65 unsigned long bo_size;
69 INIT_DELAYED_WORK(&rdev->uvd.idle_work, radeon_uvd_idle_work_handler);
71 switch (rdev->family) {
77 fw_name = FIRMWARE_R600;
82 fw_name = FIRMWARE_RS780;
86 fw_name = FIRMWARE_RV770;
92 fw_name = FIRMWARE_RV710;
100 fw_name = FIRMWARE_CYPRESS;
110 fw_name = FIRMWARE_SUMO;
118 fw_name = FIRMWARE_TAHITI;
126 fw_name = FIRMWARE_BONAIRE;
133 r = request_firmware(&rdev->uvd_fw, fw_name, rdev->dev);
135 dev_err(rdev->dev, "radeon_uvd: Can't load firmware \"%s\"\n",
140 bo_size = RADEON_GPU_PAGE_ALIGN(rdev->uvd_fw->size + 8) +
141 RADEON_UVD_STACK_SIZE + RADEON_UVD_HEAP_SIZE;
142 r = radeon_bo_create(rdev, bo_size, PAGE_SIZE, true,
143 RADEON_GEM_DOMAIN_VRAM, 0, NULL, &rdev->uvd.vcpu_bo);
145 dev_err(rdev->dev, "(%d) failed to allocate UVD bo\n", r);
149 r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
151 radeon_bo_unref(&rdev->uvd.vcpu_bo);
152 dev_err(rdev->dev, "(%d) failed to reserve UVD bo\n", r);
156 r = radeon_bo_pin(rdev->uvd.vcpu_bo, RADEON_GEM_DOMAIN_VRAM,
157 &rdev->uvd.gpu_addr);
159 radeon_bo_unreserve(rdev->uvd.vcpu_bo);
160 radeon_bo_unref(&rdev->uvd.vcpu_bo);
161 dev_err(rdev->dev, "(%d) UVD bo pin failed\n", r);
165 r = radeon_bo_kmap(rdev->uvd.vcpu_bo, &rdev->uvd.cpu_addr);
167 dev_err(rdev->dev, "(%d) UVD map failed\n", r);
171 radeon_bo_unreserve(rdev->uvd.vcpu_bo);
173 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
174 atomic_set(&rdev->uvd.handles[i], 0);
175 rdev->uvd.filp[i] = NULL;
176 rdev->uvd.img_size[i] = 0;
182 void radeon_uvd_fini(struct radeon_device *rdev)
186 if (rdev->uvd.vcpu_bo == NULL)
189 r = radeon_bo_reserve(rdev->uvd.vcpu_bo, false);
191 radeon_bo_kunmap(rdev->uvd.vcpu_bo);
192 radeon_bo_unpin(rdev->uvd.vcpu_bo);
193 radeon_bo_unreserve(rdev->uvd.vcpu_bo);
196 radeon_bo_unref(&rdev->uvd.vcpu_bo);
198 radeon_ring_fini(rdev, &rdev->ring[R600_RING_TYPE_UVD_INDEX]);
200 release_firmware(rdev->uvd_fw);
203 int radeon_uvd_suspend(struct radeon_device *rdev)
209 if (rdev->uvd.vcpu_bo == NULL)
212 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i)
213 if (atomic_read(&rdev->uvd.handles[i]))
216 if (i == RADEON_MAX_UVD_HANDLES)
219 size = radeon_bo_size(rdev->uvd.vcpu_bo);
220 size -= rdev->uvd_fw->size;
222 ptr = rdev->uvd.cpu_addr;
223 ptr += rdev->uvd_fw->size;
225 rdev->uvd.saved_bo = kmalloc(size, GFP_KERNEL);
226 memcpy(rdev->uvd.saved_bo, ptr, size);
231 int radeon_uvd_resume(struct radeon_device *rdev)
236 if (rdev->uvd.vcpu_bo == NULL)
239 memcpy(rdev->uvd.cpu_addr, rdev->uvd_fw->data, rdev->uvd_fw->size);
241 size = radeon_bo_size(rdev->uvd.vcpu_bo);
242 size -= rdev->uvd_fw->size;
244 ptr = rdev->uvd.cpu_addr;
245 ptr += rdev->uvd_fw->size;
247 if (rdev->uvd.saved_bo != NULL) {
248 memcpy(ptr, rdev->uvd.saved_bo, size);
249 kfree(rdev->uvd.saved_bo);
250 rdev->uvd.saved_bo = NULL;
252 memset(ptr, 0, size);
257 void radeon_uvd_force_into_uvd_segment(struct radeon_bo *rbo)
259 rbo->placement.fpfn = 0 >> PAGE_SHIFT;
260 rbo->placement.lpfn = (256 * 1024 * 1024) >> PAGE_SHIFT;
263 void radeon_uvd_free_handles(struct radeon_device *rdev, struct drm_file *filp)
266 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
267 uint32_t handle = atomic_read(&rdev->uvd.handles[i]);
268 if (handle != 0 && rdev->uvd.filp[i] == filp) {
269 struct radeon_fence *fence;
271 radeon_uvd_note_usage(rdev);
273 r = radeon_uvd_get_destroy_msg(rdev,
274 R600_RING_TYPE_UVD_INDEX, handle, &fence);
276 DRM_ERROR("Error destroying UVD (%d)!\n", r);
280 radeon_fence_wait(fence, false);
281 radeon_fence_unref(&fence);
283 rdev->uvd.filp[i] = NULL;
284 atomic_set(&rdev->uvd.handles[i], 0);
289 static int radeon_uvd_cs_msg_decode(uint32_t *msg, unsigned buf_sizes[])
291 unsigned stream_type = msg[4];
292 unsigned width = msg[6];
293 unsigned height = msg[7];
294 unsigned dpb_size = msg[9];
295 unsigned pitch = msg[28];
297 unsigned width_in_mb = width / 16;
298 unsigned height_in_mb = ALIGN(height / 16, 2);
300 unsigned image_size, tmp, min_dpb_size;
302 image_size = width * height;
303 image_size += image_size / 2;
304 image_size = ALIGN(image_size, 1024);
306 switch (stream_type) {
309 /* reference picture buffer */
310 min_dpb_size = image_size * 17;
312 /* macroblock context buffer */
313 min_dpb_size += width_in_mb * height_in_mb * 17 * 192;
315 /* IT surface buffer */
316 min_dpb_size += width_in_mb * height_in_mb * 32;
321 /* reference picture buffer */
322 min_dpb_size = image_size * 3;
325 min_dpb_size += width_in_mb * height_in_mb * 128;
327 /* IT surface buffer */
328 min_dpb_size += width_in_mb * 64;
330 /* DB surface buffer */
331 min_dpb_size += width_in_mb * 128;
334 tmp = max(width_in_mb, height_in_mb);
335 min_dpb_size += ALIGN(tmp * 7 * 16, 64);
340 /* reference picture buffer */
341 min_dpb_size = image_size * 3;
346 /* reference picture buffer */
347 min_dpb_size = image_size * 3;
350 min_dpb_size += width_in_mb * height_in_mb * 64;
352 /* IT surface buffer */
353 min_dpb_size += ALIGN(width_in_mb * height_in_mb * 32, 64);
357 DRM_ERROR("UVD codec not handled %d!\n", stream_type);
362 DRM_ERROR("Invalid UVD decoding target pitch!\n");
366 if (dpb_size < min_dpb_size) {
367 DRM_ERROR("Invalid dpb_size in UVD message (%d / %d)!\n",
368 dpb_size, min_dpb_size);
372 buf_sizes[0x1] = dpb_size;
373 buf_sizes[0x2] = image_size;
377 static int radeon_uvd_cs_msg(struct radeon_cs_parser *p, struct radeon_bo *bo,
378 unsigned offset, unsigned buf_sizes[])
380 int32_t *msg, msg_type, handle;
381 unsigned img_size = 0;
387 DRM_ERROR("UVD messages must be 64 byte aligned!\n");
391 if (bo->tbo.sync_obj) {
392 r = radeon_fence_wait(bo->tbo.sync_obj, false);
394 DRM_ERROR("Failed waiting for UVD message (%d)!\n", r);
399 r = radeon_bo_kmap(bo, &ptr);
401 DRM_ERROR("Failed mapping the UVD message (%d)!\n", r);
411 DRM_ERROR("Invalid UVD handle!\n");
416 /* it's a decode msg, calc buffer sizes */
417 r = radeon_uvd_cs_msg_decode(msg, buf_sizes);
418 /* calc image size (width * height) */
419 img_size = msg[6] * msg[7];
420 radeon_bo_kunmap(bo);
424 } else if (msg_type == 2) {
425 /* it's a destroy msg, free the handle */
426 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i)
427 atomic_cmpxchg(&p->rdev->uvd.handles[i], handle, 0);
428 radeon_bo_kunmap(bo);
431 /* it's a create msg, calc image size (width * height) */
432 img_size = msg[7] * msg[8];
433 radeon_bo_kunmap(bo);
436 DRM_ERROR("Illegal UVD message type (%d)!\n", msg_type);
440 /* it's a create msg, no special handling needed */
443 /* create or decode, validate the handle */
444 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
445 if (atomic_read(&p->rdev->uvd.handles[i]) == handle)
449 /* handle not found try to alloc a new one */
450 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
451 if (!atomic_cmpxchg(&p->rdev->uvd.handles[i], 0, handle)) {
452 p->rdev->uvd.filp[i] = p->filp;
453 p->rdev->uvd.img_size[i] = img_size;
458 DRM_ERROR("No more free UVD handles!\n");
462 static int radeon_uvd_cs_reloc(struct radeon_cs_parser *p,
463 int data0, int data1,
464 unsigned buf_sizes[], bool *has_msg_cmd)
466 struct radeon_cs_chunk *relocs_chunk;
467 struct radeon_cs_reloc *reloc;
468 unsigned idx, cmd, offset;
472 relocs_chunk = &p->chunks[p->chunk_relocs_idx];
473 offset = radeon_get_ib_value(p, data0);
474 idx = radeon_get_ib_value(p, data1);
475 if (idx >= relocs_chunk->length_dw) {
476 DRM_ERROR("Relocs at %d after relocations chunk end %d !\n",
477 idx, relocs_chunk->length_dw);
481 reloc = p->relocs_ptr[(idx / 4)];
482 start = reloc->gpu_offset;
483 end = start + radeon_bo_size(reloc->robj);
486 p->ib.ptr[data0] = start & 0xFFFFFFFF;
487 p->ib.ptr[data1] = start >> 32;
489 cmd = radeon_get_ib_value(p, p->idx) >> 1;
493 DRM_ERROR("invalid reloc offset %X!\n", offset);
496 if ((end - start) < buf_sizes[cmd]) {
497 DRM_ERROR("buffer (%d) to small (%d / %d)!\n", cmd,
498 (unsigned)(end - start), buf_sizes[cmd]);
502 } else if (cmd != 0x100) {
503 DRM_ERROR("invalid UVD command %X!\n", cmd);
507 if ((start >> 28) != ((end - 1) >> 28)) {
508 DRM_ERROR("reloc %LX-%LX crossing 256MB boundary!\n",
513 /* TODO: is this still necessary on NI+ ? */
514 if ((cmd == 0 || cmd == 0x3) &&
515 (start >> 28) != (p->rdev->uvd.gpu_addr >> 28)) {
516 DRM_ERROR("msg/fb buffer %LX-%LX out of 256MB segment!\n",
523 DRM_ERROR("More than one message in a UVD-IB!\n");
527 r = radeon_uvd_cs_msg(p, reloc->robj, offset, buf_sizes);
530 } else if (!*has_msg_cmd) {
531 DRM_ERROR("Message needed before other commands are send!\n");
538 static int radeon_uvd_cs_reg(struct radeon_cs_parser *p,
539 struct radeon_cs_packet *pkt,
540 int *data0, int *data1,
541 unsigned buf_sizes[],
547 for (i = 0; i <= pkt->count; ++i) {
548 switch (pkt->reg + i*4) {
549 case UVD_GPCOM_VCPU_DATA0:
552 case UVD_GPCOM_VCPU_DATA1:
555 case UVD_GPCOM_VCPU_CMD:
556 r = radeon_uvd_cs_reloc(p, *data0, *data1,
557 buf_sizes, has_msg_cmd);
561 case UVD_ENGINE_CNTL:
564 DRM_ERROR("Invalid reg 0x%X!\n",
573 int radeon_uvd_cs_parse(struct radeon_cs_parser *p)
575 struct radeon_cs_packet pkt;
576 int r, data0 = 0, data1 = 0;
578 /* does the IB has a msg command */
579 bool has_msg_cmd = false;
581 /* minimum buffer sizes */
582 unsigned buf_sizes[] = {
584 [0x00000001] = 32 * 1024 * 1024,
585 [0x00000002] = 2048 * 1152 * 3,
589 if (p->chunks[p->chunk_ib_idx].length_dw % 16) {
590 DRM_ERROR("UVD IB length (%d) not 16 dwords aligned!\n",
591 p->chunks[p->chunk_ib_idx].length_dw);
595 if (p->chunk_relocs_idx == -1) {
596 DRM_ERROR("No relocation chunk !\n");
602 r = radeon_cs_packet_parse(p, &pkt, p->idx);
606 case RADEON_PACKET_TYPE0:
607 r = radeon_uvd_cs_reg(p, &pkt, &data0, &data1,
608 buf_sizes, &has_msg_cmd);
612 case RADEON_PACKET_TYPE2:
613 p->idx += pkt.count + 2;
616 DRM_ERROR("Unknown packet type %d !\n", pkt.type);
619 } while (p->idx < p->chunks[p->chunk_ib_idx].length_dw);
622 DRM_ERROR("UVD-IBs need a msg command!\n");
629 static int radeon_uvd_send_msg(struct radeon_device *rdev,
630 int ring, struct radeon_bo *bo,
631 struct radeon_fence **fence)
633 struct ttm_validate_buffer tv;
634 struct ww_acquire_ctx ticket;
635 struct list_head head;
640 memset(&tv, 0, sizeof(tv));
643 INIT_LIST_HEAD(&head);
644 list_add(&tv.head, &head);
646 r = ttm_eu_reserve_buffers(&ticket, &head);
650 radeon_ttm_placement_from_domain(bo, RADEON_GEM_DOMAIN_VRAM);
651 radeon_uvd_force_into_uvd_segment(bo);
653 r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false);
657 r = radeon_ib_get(rdev, ring, &ib, NULL, 64);
661 addr = radeon_bo_gpu_offset(bo);
662 ib.ptr[0] = PACKET0(UVD_GPCOM_VCPU_DATA0, 0);
664 ib.ptr[2] = PACKET0(UVD_GPCOM_VCPU_DATA1, 0);
665 ib.ptr[3] = addr >> 32;
666 ib.ptr[4] = PACKET0(UVD_GPCOM_VCPU_CMD, 0);
668 for (i = 6; i < 16; ++i)
669 ib.ptr[i] = PACKET2(0);
672 r = radeon_ib_schedule(rdev, &ib, NULL, false);
675 ttm_eu_fence_buffer_objects(&ticket, &head, ib.fence);
678 *fence = radeon_fence_ref(ib.fence);
680 radeon_ib_free(rdev, &ib);
681 radeon_bo_unref(&bo);
685 ttm_eu_backoff_reservation(&ticket, &head);
689 /* multiple fence commands without any stream commands in between can
690 crash the vcpu so just try to emmit a dummy create/destroy msg to
692 int radeon_uvd_get_create_msg(struct radeon_device *rdev, int ring,
693 uint32_t handle, struct radeon_fence **fence)
695 struct radeon_bo *bo;
699 r = radeon_bo_create(rdev, 1024, PAGE_SIZE, true,
700 RADEON_GEM_DOMAIN_VRAM, 0, NULL, &bo);
704 r = radeon_bo_reserve(bo, false);
706 radeon_bo_unref(&bo);
710 r = radeon_bo_kmap(bo, (void **)&msg);
712 radeon_bo_unreserve(bo);
713 radeon_bo_unref(&bo);
717 /* stitch together an UVD create msg */
718 msg[0] = cpu_to_le32(0x00000de4);
719 msg[1] = cpu_to_le32(0x00000000);
720 msg[2] = cpu_to_le32(handle);
721 msg[3] = cpu_to_le32(0x00000000);
722 msg[4] = cpu_to_le32(0x00000000);
723 msg[5] = cpu_to_le32(0x00000000);
724 msg[6] = cpu_to_le32(0x00000000);
725 msg[7] = cpu_to_le32(0x00000780);
726 msg[8] = cpu_to_le32(0x00000440);
727 msg[9] = cpu_to_le32(0x00000000);
728 msg[10] = cpu_to_le32(0x01b37000);
729 for (i = 11; i < 1024; ++i)
730 msg[i] = cpu_to_le32(0x0);
732 radeon_bo_kunmap(bo);
733 radeon_bo_unreserve(bo);
735 return radeon_uvd_send_msg(rdev, ring, bo, fence);
738 int radeon_uvd_get_destroy_msg(struct radeon_device *rdev, int ring,
739 uint32_t handle, struct radeon_fence **fence)
741 struct radeon_bo *bo;
745 r = radeon_bo_create(rdev, 1024, PAGE_SIZE, true,
746 RADEON_GEM_DOMAIN_VRAM, 0, NULL, &bo);
750 r = radeon_bo_reserve(bo, false);
752 radeon_bo_unref(&bo);
756 r = radeon_bo_kmap(bo, (void **)&msg);
758 radeon_bo_unreserve(bo);
759 radeon_bo_unref(&bo);
763 /* stitch together an UVD destroy msg */
764 msg[0] = cpu_to_le32(0x00000de4);
765 msg[1] = cpu_to_le32(0x00000002);
766 msg[2] = cpu_to_le32(handle);
767 msg[3] = cpu_to_le32(0x00000000);
768 for (i = 4; i < 1024; ++i)
769 msg[i] = cpu_to_le32(0x0);
771 radeon_bo_kunmap(bo);
772 radeon_bo_unreserve(bo);
774 return radeon_uvd_send_msg(rdev, ring, bo, fence);
778 * radeon_uvd_count_handles - count number of open streams
780 * @rdev: radeon_device pointer
781 * @sd: number of SD streams
782 * @hd: number of HD streams
784 * Count the number of open SD/HD streams as a hint for power mangement
786 static void radeon_uvd_count_handles(struct radeon_device *rdev,
787 unsigned *sd, unsigned *hd)
794 for (i = 0; i < RADEON_MAX_UVD_HANDLES; ++i) {
795 if (!atomic_read(&rdev->uvd.handles[i]))
798 if (rdev->uvd.img_size[i] >= 720*576)
805 static void radeon_uvd_idle_work_handler(struct work_struct *work)
807 struct radeon_device *rdev =
808 container_of(work, struct radeon_device, uvd.idle_work.work);
810 if (radeon_fence_count_emitted(rdev, R600_RING_TYPE_UVD_INDEX) == 0) {
811 if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
812 radeon_uvd_count_handles(rdev, &rdev->pm.dpm.sd,
814 radeon_dpm_enable_uvd(rdev, false);
816 radeon_set_uvd_clocks(rdev, 0, 0);
819 schedule_delayed_work(&rdev->uvd.idle_work,
820 msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
824 void radeon_uvd_note_usage(struct radeon_device *rdev)
826 bool streams_changed = false;
827 bool set_clocks = !cancel_delayed_work_sync(&rdev->uvd.idle_work);
828 set_clocks &= schedule_delayed_work(&rdev->uvd.idle_work,
829 msecs_to_jiffies(UVD_IDLE_TIMEOUT_MS));
831 if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
832 unsigned hd = 0, sd = 0;
833 radeon_uvd_count_handles(rdev, &sd, &hd);
834 if ((rdev->pm.dpm.sd != sd) ||
835 (rdev->pm.dpm.hd != hd)) {
836 rdev->pm.dpm.sd = sd;
837 rdev->pm.dpm.hd = hd;
838 /* disable this for now */
839 /*streams_changed = true;*/
843 if (set_clocks || streams_changed) {
844 if ((rdev->pm.pm_method == PM_METHOD_DPM) && rdev->pm.dpm_enabled) {
845 radeon_dpm_enable_uvd(rdev, true);
847 radeon_set_uvd_clocks(rdev, 53300, 40000);
852 static unsigned radeon_uvd_calc_upll_post_div(unsigned vco_freq,
853 unsigned target_freq,
857 unsigned post_div = vco_freq / target_freq;
859 /* adjust to post divider minimum value */
860 if (post_div < pd_min)
863 /* we alway need a frequency less than or equal the target */
864 if ((vco_freq / post_div) > target_freq)
867 /* post dividers above a certain value must be even */
868 if (post_div > pd_even && post_div % 2)
875 * radeon_uvd_calc_upll_dividers - calc UPLL clock dividers
877 * @rdev: radeon_device pointer
880 * @vco_min: minimum VCO frequency
881 * @vco_max: maximum VCO frequency
882 * @fb_factor: factor to multiply vco freq with
883 * @fb_mask: limit and bitmask for feedback divider
884 * @pd_min: post divider minimum
885 * @pd_max: post divider maximum
886 * @pd_even: post divider must be even above this value
887 * @optimal_fb_div: resulting feedback divider
888 * @optimal_vclk_div: resulting vclk post divider
889 * @optimal_dclk_div: resulting dclk post divider
891 * Calculate dividers for UVDs UPLL (R6xx-SI, except APUs).
892 * Returns zero on success -EINVAL on error.
894 int radeon_uvd_calc_upll_dividers(struct radeon_device *rdev,
895 unsigned vclk, unsigned dclk,
896 unsigned vco_min, unsigned vco_max,
897 unsigned fb_factor, unsigned fb_mask,
898 unsigned pd_min, unsigned pd_max,
900 unsigned *optimal_fb_div,
901 unsigned *optimal_vclk_div,
902 unsigned *optimal_dclk_div)
904 unsigned vco_freq, ref_freq = rdev->clock.spll.reference_freq;
906 /* start off with something large */
907 unsigned optimal_score = ~0;
909 /* loop through vco from low to high */
910 vco_min = max(max(vco_min, vclk), dclk);
911 for (vco_freq = vco_min; vco_freq <= vco_max; vco_freq += 100) {
913 uint64_t fb_div = (uint64_t)vco_freq * fb_factor;
914 unsigned vclk_div, dclk_div, score;
916 do_div(fb_div, ref_freq);
918 /* fb div out of range ? */
919 if (fb_div > fb_mask)
920 break; /* it can oly get worse */
924 /* calc vclk divider with current vco freq */
925 vclk_div = radeon_uvd_calc_upll_post_div(vco_freq, vclk,
927 if (vclk_div > pd_max)
928 break; /* vco is too big, it has to stop */
930 /* calc dclk divider with current vco freq */
931 dclk_div = radeon_uvd_calc_upll_post_div(vco_freq, dclk,
933 if (vclk_div > pd_max)
934 break; /* vco is too big, it has to stop */
936 /* calc score with current vco freq */
937 score = vclk - (vco_freq / vclk_div) + dclk - (vco_freq / dclk_div);
939 /* determine if this vco setting is better than current optimal settings */
940 if (score < optimal_score) {
941 *optimal_fb_div = fb_div;
942 *optimal_vclk_div = vclk_div;
943 *optimal_dclk_div = dclk_div;
944 optimal_score = score;
945 if (optimal_score == 0)
946 break; /* it can't get better than this */
950 /* did we found a valid setup ? */
951 if (optimal_score == ~0)
957 int radeon_uvd_send_upll_ctlreq(struct radeon_device *rdev,
958 unsigned cg_upll_func_cntl)
962 /* make sure UPLL_CTLREQ is deasserted */
963 WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK);
967 /* assert UPLL_CTLREQ */
968 WREG32_P(cg_upll_func_cntl, UPLL_CTLREQ_MASK, ~UPLL_CTLREQ_MASK);
970 /* wait for CTLACK and CTLACK2 to get asserted */
971 for (i = 0; i < 100; ++i) {
972 uint32_t mask = UPLL_CTLACK_MASK | UPLL_CTLACK2_MASK;
973 if ((RREG32(cg_upll_func_cntl) & mask) == mask)
978 /* deassert UPLL_CTLREQ */
979 WREG32_P(cg_upll_func_cntl, 0, ~UPLL_CTLREQ_MASK);
982 DRM_ERROR("Timeout setting UVD clocks!\n");