2 * Copyright © 2008-2010 Intel Corporation
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:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Eric Anholt <eric@anholt.net>
25 * Zou Nan hai <nanhai.zou@intel.com>
26 * Xiang Hai hao<haihao.xiang@intel.com>
34 #include "i915_trace.h"
35 #include "intel_drv.h"
37 static inline int ring_space(struct intel_ring_buffer *ring)
39 int space = (ring->head & HEAD_ADDR) - (ring->tail + 8);
45 static u32 i915_gem_get_seqno(struct drm_device *dev)
47 drm_i915_private_t *dev_priv = dev->dev_private;
50 seqno = dev_priv->next_seqno;
52 /* reserve 0 for non-seqno */
53 if (++dev_priv->next_seqno == 0)
54 dev_priv->next_seqno = 1;
60 render_ring_flush(struct intel_ring_buffer *ring,
61 u32 invalidate_domains,
64 struct drm_device *dev = ring->dev;
65 drm_i915_private_t *dev_priv = dev->dev_private;
70 DRM_INFO("%s: invalidate %08x flush %08x\n", __func__,
71 invalidate_domains, flush_domains);
74 trace_i915_gem_request_flush(dev, dev_priv->next_seqno,
75 invalidate_domains, flush_domains);
77 if ((invalidate_domains | flush_domains) & I915_GEM_GPU_DOMAINS) {
81 * I915_GEM_DOMAIN_RENDER is always invalidated, but is
82 * only flushed if MI_NO_WRITE_FLUSH is unset. On 965, it is
83 * also flushed at 2d versus 3d pipeline switches.
87 * I915_GEM_DOMAIN_SAMPLER is flushed on pre-965 if
88 * MI_READ_FLUSH is set, and is always flushed on 965.
90 * I915_GEM_DOMAIN_COMMAND may not exist?
92 * I915_GEM_DOMAIN_INSTRUCTION, which exists on 965, is
93 * invalidated when MI_EXE_FLUSH is set.
95 * I915_GEM_DOMAIN_VERTEX, which exists on 965, is
96 * invalidated with every MI_FLUSH.
100 * On 965, TLBs associated with I915_GEM_DOMAIN_COMMAND
101 * and I915_GEM_DOMAIN_CPU in are invalidated at PTE write and
102 * I915_GEM_DOMAIN_RENDER and I915_GEM_DOMAIN_SAMPLER
103 * are flushed at any MI_FLUSH.
106 cmd = MI_FLUSH | MI_NO_WRITE_FLUSH;
107 if ((invalidate_domains|flush_domains) &
108 I915_GEM_DOMAIN_RENDER)
109 cmd &= ~MI_NO_WRITE_FLUSH;
110 if (INTEL_INFO(dev)->gen < 4) {
112 * On the 965, the sampler cache always gets flushed
113 * and this bit is reserved.
115 if (invalidate_domains & I915_GEM_DOMAIN_SAMPLER)
116 cmd |= MI_READ_FLUSH;
118 if (invalidate_domains & I915_GEM_DOMAIN_INSTRUCTION)
121 if (invalidate_domains & I915_GEM_DOMAIN_COMMAND &&
122 (IS_G4X(dev) || IS_GEN5(dev)))
123 cmd |= MI_INVALIDATE_ISP;
126 DRM_INFO("%s: queue flush %08x to ring\n", __func__, cmd);
128 ret = intel_ring_begin(ring, 2);
132 intel_ring_emit(ring, cmd);
133 intel_ring_emit(ring, MI_NOOP);
134 intel_ring_advance(ring);
140 static void ring_write_tail(struct intel_ring_buffer *ring,
143 drm_i915_private_t *dev_priv = ring->dev->dev_private;
144 I915_WRITE_TAIL(ring, value);
147 u32 intel_ring_get_active_head(struct intel_ring_buffer *ring)
149 drm_i915_private_t *dev_priv = ring->dev->dev_private;
150 u32 acthd_reg = INTEL_INFO(ring->dev)->gen >= 4 ?
151 RING_ACTHD(ring->mmio_base) : ACTHD;
153 return I915_READ(acthd_reg);
156 static int init_ring_common(struct intel_ring_buffer *ring)
158 drm_i915_private_t *dev_priv = ring->dev->dev_private;
159 struct drm_i915_gem_object *obj = ring->obj;
162 /* Stop the ring if it's running. */
163 I915_WRITE_CTL(ring, 0);
164 I915_WRITE_HEAD(ring, 0);
165 ring->write_tail(ring, 0);
167 /* Initialize the ring. */
168 I915_WRITE_START(ring, obj->gtt_offset);
169 head = I915_READ_HEAD(ring) & HEAD_ADDR;
171 /* G45 ring initialization fails to reset head to zero */
173 DRM_DEBUG_KMS("%s head not reset to zero "
174 "ctl %08x head %08x tail %08x start %08x\n",
177 I915_READ_HEAD(ring),
178 I915_READ_TAIL(ring),
179 I915_READ_START(ring));
181 I915_WRITE_HEAD(ring, 0);
183 if (I915_READ_HEAD(ring) & HEAD_ADDR) {
184 DRM_ERROR("failed to set %s head to zero "
185 "ctl %08x head %08x tail %08x start %08x\n",
188 I915_READ_HEAD(ring),
189 I915_READ_TAIL(ring),
190 I915_READ_START(ring));
195 ((ring->size - PAGE_SIZE) & RING_NR_PAGES)
196 | RING_REPORT_64K | RING_VALID);
198 /* If the head is still not zero, the ring is dead */
199 if ((I915_READ_CTL(ring) & RING_VALID) == 0 ||
200 I915_READ_START(ring) != obj->gtt_offset ||
201 (I915_READ_HEAD(ring) & HEAD_ADDR) != 0) {
202 DRM_ERROR("%s initialization failed "
203 "ctl %08x head %08x tail %08x start %08x\n",
206 I915_READ_HEAD(ring),
207 I915_READ_TAIL(ring),
208 I915_READ_START(ring));
212 if (!drm_core_check_feature(ring->dev, DRIVER_MODESET))
213 i915_kernel_lost_context(ring->dev);
215 ring->head = I915_READ_HEAD(ring);
216 ring->tail = I915_READ_TAIL(ring) & TAIL_ADDR;
217 ring->space = ring_space(ring);
224 * 965+ support PIPE_CONTROL commands, which provide finer grained control
225 * over cache flushing.
227 struct pipe_control {
228 struct drm_i915_gem_object *obj;
229 volatile u32 *cpu_page;
234 init_pipe_control(struct intel_ring_buffer *ring)
236 struct pipe_control *pc;
237 struct drm_i915_gem_object *obj;
243 pc = kmalloc(sizeof(*pc), GFP_KERNEL);
247 obj = i915_gem_alloc_object(ring->dev, 4096);
249 DRM_ERROR("Failed to allocate seqno page\n");
253 obj->agp_type = AGP_USER_CACHED_MEMORY;
255 ret = i915_gem_object_pin(obj, 4096, true);
259 pc->gtt_offset = obj->gtt_offset;
260 pc->cpu_page = kmap(obj->pages[0]);
261 if (pc->cpu_page == NULL)
269 i915_gem_object_unpin(obj);
271 drm_gem_object_unreference(&obj->base);
278 cleanup_pipe_control(struct intel_ring_buffer *ring)
280 struct pipe_control *pc = ring->private;
281 struct drm_i915_gem_object *obj;
287 kunmap(obj->pages[0]);
288 i915_gem_object_unpin(obj);
289 drm_gem_object_unreference(&obj->base);
292 ring->private = NULL;
295 static int init_render_ring(struct intel_ring_buffer *ring)
297 struct drm_device *dev = ring->dev;
298 struct drm_i915_private *dev_priv = dev->dev_private;
299 int ret = init_ring_common(ring);
301 if (INTEL_INFO(dev)->gen > 3) {
302 int mode = VS_TIMER_DISPATCH << 16 | VS_TIMER_DISPATCH;
304 mode |= MI_FLUSH_ENABLE << 16 | MI_FLUSH_ENABLE;
305 I915_WRITE(MI_MODE, mode);
308 if (INTEL_INFO(dev)->gen >= 6) {
309 } else if (IS_GEN5(dev)) {
310 ret = init_pipe_control(ring);
318 static void render_ring_cleanup(struct intel_ring_buffer *ring)
323 cleanup_pipe_control(ring);
327 update_semaphore(struct intel_ring_buffer *ring, int i, u32 seqno)
329 struct drm_device *dev = ring->dev;
330 struct drm_i915_private *dev_priv = dev->dev_private;
334 * cs -> 1 = vcs, 0 = bcs
335 * vcs -> 1 = bcs, 0 = cs,
336 * bcs -> 1 = cs, 0 = vcs.
338 id = ring - dev_priv->ring;
342 intel_ring_emit(ring,
344 MI_SEMAPHORE_REGISTER |
345 MI_SEMAPHORE_UPDATE);
346 intel_ring_emit(ring, seqno);
347 intel_ring_emit(ring,
348 RING_SYNC_0(dev_priv->ring[id].mmio_base) + 4*i);
352 gen6_add_request(struct intel_ring_buffer *ring,
358 ret = intel_ring_begin(ring, 10);
362 seqno = i915_gem_get_seqno(ring->dev);
363 update_semaphore(ring, 0, seqno);
364 update_semaphore(ring, 1, seqno);
366 intel_ring_emit(ring, MI_STORE_DWORD_INDEX);
367 intel_ring_emit(ring, I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
368 intel_ring_emit(ring, seqno);
369 intel_ring_emit(ring, MI_USER_INTERRUPT);
370 intel_ring_advance(ring);
377 intel_ring_sync(struct intel_ring_buffer *ring,
378 struct intel_ring_buffer *to,
383 ret = intel_ring_begin(ring, 4);
387 intel_ring_emit(ring,
389 MI_SEMAPHORE_REGISTER |
390 intel_ring_sync_index(ring, to) << 17 |
391 MI_SEMAPHORE_COMPARE);
392 intel_ring_emit(ring, seqno);
393 intel_ring_emit(ring, 0);
394 intel_ring_emit(ring, MI_NOOP);
395 intel_ring_advance(ring);
400 #define PIPE_CONTROL_FLUSH(ring__, addr__) \
402 intel_ring_emit(ring__, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE | \
403 PIPE_CONTROL_DEPTH_STALL | 2); \
404 intel_ring_emit(ring__, (addr__) | PIPE_CONTROL_GLOBAL_GTT); \
405 intel_ring_emit(ring__, 0); \
406 intel_ring_emit(ring__, 0); \
410 pc_render_add_request(struct intel_ring_buffer *ring,
413 struct drm_device *dev = ring->dev;
414 u32 seqno = i915_gem_get_seqno(dev);
415 struct pipe_control *pc = ring->private;
416 u32 scratch_addr = pc->gtt_offset + 128;
419 /* For Ironlake, MI_USER_INTERRUPT was deprecated and apparently
420 * incoherent with writes to memory, i.e. completely fubar,
421 * so we need to use PIPE_NOTIFY instead.
423 * However, we also need to workaround the qword write
424 * incoherence by flushing the 6 PIPE_NOTIFY buffers out to
425 * memory before requesting an interrupt.
427 ret = intel_ring_begin(ring, 32);
431 intel_ring_emit(ring, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |
432 PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_TC_FLUSH);
433 intel_ring_emit(ring, pc->gtt_offset | PIPE_CONTROL_GLOBAL_GTT);
434 intel_ring_emit(ring, seqno);
435 intel_ring_emit(ring, 0);
436 PIPE_CONTROL_FLUSH(ring, scratch_addr);
437 scratch_addr += 128; /* write to separate cachelines */
438 PIPE_CONTROL_FLUSH(ring, scratch_addr);
440 PIPE_CONTROL_FLUSH(ring, scratch_addr);
442 PIPE_CONTROL_FLUSH(ring, scratch_addr);
444 PIPE_CONTROL_FLUSH(ring, scratch_addr);
446 PIPE_CONTROL_FLUSH(ring, scratch_addr);
447 intel_ring_emit(ring, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |
448 PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_TC_FLUSH |
449 PIPE_CONTROL_NOTIFY);
450 intel_ring_emit(ring, pc->gtt_offset | PIPE_CONTROL_GLOBAL_GTT);
451 intel_ring_emit(ring, seqno);
452 intel_ring_emit(ring, 0);
453 intel_ring_advance(ring);
460 render_ring_add_request(struct intel_ring_buffer *ring,
463 struct drm_device *dev = ring->dev;
464 u32 seqno = i915_gem_get_seqno(dev);
467 ret = intel_ring_begin(ring, 4);
471 intel_ring_emit(ring, MI_STORE_DWORD_INDEX);
472 intel_ring_emit(ring, I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
473 intel_ring_emit(ring, seqno);
474 intel_ring_emit(ring, MI_USER_INTERRUPT);
475 intel_ring_advance(ring);
482 ring_get_seqno(struct intel_ring_buffer *ring)
484 return intel_read_status_page(ring, I915_GEM_HWS_INDEX);
488 pc_render_get_seqno(struct intel_ring_buffer *ring)
490 struct pipe_control *pc = ring->private;
491 return pc->cpu_page[0];
495 ironlake_enable_irq(drm_i915_private_t *dev_priv, u32 mask)
497 dev_priv->gt_irq_mask &= ~mask;
498 I915_WRITE(GTIMR, dev_priv->gt_irq_mask);
503 ironlake_disable_irq(drm_i915_private_t *dev_priv, u32 mask)
505 dev_priv->gt_irq_mask |= mask;
506 I915_WRITE(GTIMR, dev_priv->gt_irq_mask);
511 i915_enable_irq(drm_i915_private_t *dev_priv, u32 mask)
513 dev_priv->irq_mask &= ~mask;
514 I915_WRITE(IMR, dev_priv->irq_mask);
519 i915_disable_irq(drm_i915_private_t *dev_priv, u32 mask)
521 dev_priv->irq_mask |= mask;
522 I915_WRITE(IMR, dev_priv->irq_mask);
527 render_ring_get_irq(struct intel_ring_buffer *ring)
529 struct drm_device *dev = ring->dev;
530 drm_i915_private_t *dev_priv = dev->dev_private;
532 if (!dev->irq_enabled)
535 spin_lock(&ring->irq_lock);
536 if (ring->irq_refcount++ == 0) {
537 if (HAS_PCH_SPLIT(dev))
538 ironlake_enable_irq(dev_priv,
539 GT_PIPE_NOTIFY | GT_USER_INTERRUPT);
541 i915_enable_irq(dev_priv, I915_USER_INTERRUPT);
543 spin_unlock(&ring->irq_lock);
549 render_ring_put_irq(struct intel_ring_buffer *ring)
551 struct drm_device *dev = ring->dev;
552 drm_i915_private_t *dev_priv = dev->dev_private;
554 spin_lock(&ring->irq_lock);
555 if (--ring->irq_refcount == 0) {
556 if (HAS_PCH_SPLIT(dev))
557 ironlake_disable_irq(dev_priv,
561 i915_disable_irq(dev_priv, I915_USER_INTERRUPT);
563 spin_unlock(&ring->irq_lock);
566 void intel_ring_setup_status_page(struct intel_ring_buffer *ring)
568 drm_i915_private_t *dev_priv = ring->dev->dev_private;
569 u32 mmio = IS_GEN6(ring->dev) ?
570 RING_HWS_PGA_GEN6(ring->mmio_base) :
571 RING_HWS_PGA(ring->mmio_base);
572 I915_WRITE(mmio, (u32)ring->status_page.gfx_addr);
577 bsd_ring_flush(struct intel_ring_buffer *ring,
578 u32 invalidate_domains,
583 if ((flush_domains & I915_GEM_DOMAIN_RENDER) == 0)
586 ret = intel_ring_begin(ring, 2);
590 intel_ring_emit(ring, MI_FLUSH);
591 intel_ring_emit(ring, MI_NOOP);
592 intel_ring_advance(ring);
597 ring_add_request(struct intel_ring_buffer *ring,
603 ret = intel_ring_begin(ring, 4);
607 seqno = i915_gem_get_seqno(ring->dev);
609 intel_ring_emit(ring, MI_STORE_DWORD_INDEX);
610 intel_ring_emit(ring, I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
611 intel_ring_emit(ring, seqno);
612 intel_ring_emit(ring, MI_USER_INTERRUPT);
613 intel_ring_advance(ring);
615 DRM_DEBUG_DRIVER("%s %d\n", ring->name, seqno);
621 ring_get_irq(struct intel_ring_buffer *ring, u32 flag)
623 struct drm_device *dev = ring->dev;
624 drm_i915_private_t *dev_priv = dev->dev_private;
626 if (!dev->irq_enabled)
629 spin_lock(&ring->irq_lock);
630 if (ring->irq_refcount++ == 0)
631 ironlake_enable_irq(dev_priv, flag);
632 spin_unlock(&ring->irq_lock);
638 ring_put_irq(struct intel_ring_buffer *ring, u32 flag)
640 struct drm_device *dev = ring->dev;
641 drm_i915_private_t *dev_priv = dev->dev_private;
643 spin_lock(&ring->irq_lock);
644 if (--ring->irq_refcount == 0)
645 ironlake_disable_irq(dev_priv, flag);
646 spin_unlock(&ring->irq_lock);
650 gen6_ring_get_irq(struct intel_ring_buffer *ring, u32 gflag, u32 rflag)
652 struct drm_device *dev = ring->dev;
653 drm_i915_private_t *dev_priv = dev->dev_private;
655 if (!dev->irq_enabled)
658 spin_lock(&ring->irq_lock);
659 if (ring->irq_refcount++ == 0) {
660 ring->irq_mask &= ~rflag;
661 I915_WRITE_IMR(ring, ring->irq_mask);
662 ironlake_enable_irq(dev_priv, gflag);
664 spin_unlock(&ring->irq_lock);
670 gen6_ring_put_irq(struct intel_ring_buffer *ring, u32 gflag, u32 rflag)
672 struct drm_device *dev = ring->dev;
673 drm_i915_private_t *dev_priv = dev->dev_private;
675 spin_lock(&ring->irq_lock);
676 if (--ring->irq_refcount == 0) {
677 ring->irq_mask |= rflag;
678 I915_WRITE_IMR(ring, ring->irq_mask);
679 ironlake_disable_irq(dev_priv, gflag);
681 spin_unlock(&ring->irq_lock);
685 bsd_ring_get_irq(struct intel_ring_buffer *ring)
687 return ring_get_irq(ring, GT_BSD_USER_INTERRUPT);
690 bsd_ring_put_irq(struct intel_ring_buffer *ring)
692 ring_put_irq(ring, GT_BSD_USER_INTERRUPT);
696 ring_dispatch_execbuffer(struct intel_ring_buffer *ring, u32 offset, u32 length)
700 ret = intel_ring_begin(ring, 2);
704 intel_ring_emit(ring,
705 MI_BATCH_BUFFER_START | (2 << 6) |
706 MI_BATCH_NON_SECURE_I965);
707 intel_ring_emit(ring, offset);
708 intel_ring_advance(ring);
714 render_ring_dispatch_execbuffer(struct intel_ring_buffer *ring,
717 struct drm_device *dev = ring->dev;
718 drm_i915_private_t *dev_priv = dev->dev_private;
721 trace_i915_gem_request_submit(dev, dev_priv->next_seqno + 1);
723 if (IS_I830(dev) || IS_845G(dev)) {
724 ret = intel_ring_begin(ring, 4);
728 intel_ring_emit(ring, MI_BATCH_BUFFER);
729 intel_ring_emit(ring, offset | MI_BATCH_NON_SECURE);
730 intel_ring_emit(ring, offset + len - 8);
731 intel_ring_emit(ring, 0);
733 ret = intel_ring_begin(ring, 2);
737 if (INTEL_INFO(dev)->gen >= 4) {
738 intel_ring_emit(ring,
739 MI_BATCH_BUFFER_START | (2 << 6) |
740 MI_BATCH_NON_SECURE_I965);
741 intel_ring_emit(ring, offset);
743 intel_ring_emit(ring,
744 MI_BATCH_BUFFER_START | (2 << 6));
745 intel_ring_emit(ring, offset | MI_BATCH_NON_SECURE);
748 intel_ring_advance(ring);
753 static void cleanup_status_page(struct intel_ring_buffer *ring)
755 drm_i915_private_t *dev_priv = ring->dev->dev_private;
756 struct drm_i915_gem_object *obj;
758 obj = ring->status_page.obj;
762 kunmap(obj->pages[0]);
763 i915_gem_object_unpin(obj);
764 drm_gem_object_unreference(&obj->base);
765 ring->status_page.obj = NULL;
767 memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map));
770 static int init_status_page(struct intel_ring_buffer *ring)
772 struct drm_device *dev = ring->dev;
773 drm_i915_private_t *dev_priv = dev->dev_private;
774 struct drm_i915_gem_object *obj;
777 obj = i915_gem_alloc_object(dev, 4096);
779 DRM_ERROR("Failed to allocate status page\n");
783 obj->agp_type = AGP_USER_CACHED_MEMORY;
785 ret = i915_gem_object_pin(obj, 4096, true);
790 ring->status_page.gfx_addr = obj->gtt_offset;
791 ring->status_page.page_addr = kmap(obj->pages[0]);
792 if (ring->status_page.page_addr == NULL) {
793 memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map));
796 ring->status_page.obj = obj;
797 memset(ring->status_page.page_addr, 0, PAGE_SIZE);
799 intel_ring_setup_status_page(ring);
800 DRM_DEBUG_DRIVER("%s hws offset: 0x%08x\n",
801 ring->name, ring->status_page.gfx_addr);
806 i915_gem_object_unpin(obj);
808 drm_gem_object_unreference(&obj->base);
813 int intel_init_ring_buffer(struct drm_device *dev,
814 struct intel_ring_buffer *ring)
816 struct drm_i915_gem_object *obj;
820 INIT_LIST_HEAD(&ring->active_list);
821 INIT_LIST_HEAD(&ring->request_list);
822 INIT_LIST_HEAD(&ring->gpu_write_list);
824 spin_lock_init(&ring->irq_lock);
827 if (I915_NEED_GFX_HWS(dev)) {
828 ret = init_status_page(ring);
833 obj = i915_gem_alloc_object(dev, ring->size);
835 DRM_ERROR("Failed to allocate ringbuffer\n");
842 ret = i915_gem_object_pin(obj, PAGE_SIZE, true);
846 ring->map.size = ring->size;
847 ring->map.offset = dev->agp->base + obj->gtt_offset;
852 drm_core_ioremap_wc(&ring->map, dev);
853 if (ring->map.handle == NULL) {
854 DRM_ERROR("Failed to map ringbuffer.\n");
859 ring->virtual_start = ring->map.handle;
860 ret = ring->init(ring);
864 /* Workaround an erratum on the i830 which causes a hang if
865 * the TAIL pointer points to within the last 2 cachelines
868 ring->effective_size = ring->size;
869 if (IS_I830(ring->dev))
870 ring->effective_size -= 128;
875 drm_core_ioremapfree(&ring->map, dev);
877 i915_gem_object_unpin(obj);
879 drm_gem_object_unreference(&obj->base);
882 cleanup_status_page(ring);
886 void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring)
888 struct drm_i915_private *dev_priv;
891 if (ring->obj == NULL)
894 /* Disable the ring buffer. The ring must be idle at this point */
895 dev_priv = ring->dev->dev_private;
896 ret = intel_wait_ring_buffer(ring, ring->size - 8);
897 I915_WRITE_CTL(ring, 0);
899 drm_core_ioremapfree(&ring->map, ring->dev);
901 i915_gem_object_unpin(ring->obj);
902 drm_gem_object_unreference(&ring->obj->base);
908 cleanup_status_page(ring);
911 static int intel_wrap_ring_buffer(struct intel_ring_buffer *ring)
914 int rem = ring->size - ring->tail;
916 if (ring->space < rem) {
917 int ret = intel_wait_ring_buffer(ring, rem);
922 virt = (unsigned int *)(ring->virtual_start + ring->tail);
930 ring->space = ring_space(ring);
935 int intel_wait_ring_buffer(struct intel_ring_buffer *ring, int n)
937 struct drm_device *dev = ring->dev;
938 struct drm_i915_private *dev_priv = dev->dev_private;
942 /* If the reported head position has wrapped or hasn't advanced,
943 * fallback to the slow and accurate path.
945 head = intel_read_status_page(ring, 4);
946 if (head > ring->head) {
948 ring->space = ring_space(ring);
949 if (ring->space >= n)
953 trace_i915_ring_wait_begin (dev);
954 end = jiffies + 3 * HZ;
956 ring->head = I915_READ_HEAD(ring);
957 ring->space = ring_space(ring);
958 if (ring->space >= n) {
959 trace_i915_ring_wait_end(dev);
963 if (dev->primary->master) {
964 struct drm_i915_master_private *master_priv = dev->primary->master->driver_priv;
965 if (master_priv->sarea_priv)
966 master_priv->sarea_priv->perf_boxes |= I915_BOX_WAIT;
970 if (atomic_read(&dev_priv->mm.wedged))
972 } while (!time_after(jiffies, end));
973 trace_i915_ring_wait_end (dev);
977 int intel_ring_begin(struct intel_ring_buffer *ring,
980 int n = 4*num_dwords;
983 if (unlikely(ring->tail + n > ring->effective_size)) {
984 ret = intel_wrap_ring_buffer(ring);
989 if (unlikely(ring->space < n)) {
990 ret = intel_wait_ring_buffer(ring, n);
999 void intel_ring_advance(struct intel_ring_buffer *ring)
1001 ring->tail &= ring->size - 1;
1002 ring->write_tail(ring, ring->tail);
1005 static const struct intel_ring_buffer render_ring = {
1006 .name = "render ring",
1008 .mmio_base = RENDER_RING_BASE,
1009 .size = 32 * PAGE_SIZE,
1010 .init = init_render_ring,
1011 .write_tail = ring_write_tail,
1012 .flush = render_ring_flush,
1013 .add_request = render_ring_add_request,
1014 .get_seqno = ring_get_seqno,
1015 .irq_get = render_ring_get_irq,
1016 .irq_put = render_ring_put_irq,
1017 .dispatch_execbuffer = render_ring_dispatch_execbuffer,
1018 .cleanup = render_ring_cleanup,
1021 /* ring buffer for bit-stream decoder */
1023 static const struct intel_ring_buffer bsd_ring = {
1026 .mmio_base = BSD_RING_BASE,
1027 .size = 32 * PAGE_SIZE,
1028 .init = init_ring_common,
1029 .write_tail = ring_write_tail,
1030 .flush = bsd_ring_flush,
1031 .add_request = ring_add_request,
1032 .get_seqno = ring_get_seqno,
1033 .irq_get = bsd_ring_get_irq,
1034 .irq_put = bsd_ring_put_irq,
1035 .dispatch_execbuffer = ring_dispatch_execbuffer,
1039 static void gen6_bsd_ring_write_tail(struct intel_ring_buffer *ring,
1042 drm_i915_private_t *dev_priv = ring->dev->dev_private;
1044 /* Every tail move must follow the sequence below */
1045 I915_WRITE(GEN6_BSD_SLEEP_PSMI_CONTROL,
1046 GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_MODIFY_MASK |
1047 GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_DISABLE);
1048 I915_WRITE(GEN6_BSD_RNCID, 0x0);
1050 if (wait_for((I915_READ(GEN6_BSD_SLEEP_PSMI_CONTROL) &
1051 GEN6_BSD_SLEEP_PSMI_CONTROL_IDLE_INDICATOR) == 0,
1053 DRM_ERROR("timed out waiting for IDLE Indicator\n");
1055 I915_WRITE_TAIL(ring, value);
1056 I915_WRITE(GEN6_BSD_SLEEP_PSMI_CONTROL,
1057 GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_MODIFY_MASK |
1058 GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_ENABLE);
1061 static int gen6_ring_flush(struct intel_ring_buffer *ring,
1062 u32 invalidate_domains,
1067 if ((flush_domains & I915_GEM_DOMAIN_RENDER) == 0)
1070 ret = intel_ring_begin(ring, 4);
1074 intel_ring_emit(ring, MI_FLUSH_DW);
1075 intel_ring_emit(ring, 0);
1076 intel_ring_emit(ring, 0);
1077 intel_ring_emit(ring, 0);
1078 intel_ring_advance(ring);
1083 gen6_ring_dispatch_execbuffer(struct intel_ring_buffer *ring,
1084 u32 offset, u32 len)
1088 ret = intel_ring_begin(ring, 2);
1092 intel_ring_emit(ring, MI_BATCH_BUFFER_START | MI_BATCH_NON_SECURE_I965);
1093 /* bit0-7 is the length on GEN6+ */
1094 intel_ring_emit(ring, offset);
1095 intel_ring_advance(ring);
1101 gen6_render_ring_get_irq(struct intel_ring_buffer *ring)
1103 return gen6_ring_get_irq(ring,
1105 GEN6_RENDER_USER_INTERRUPT);
1109 gen6_render_ring_put_irq(struct intel_ring_buffer *ring)
1111 return gen6_ring_put_irq(ring,
1113 GEN6_RENDER_USER_INTERRUPT);
1117 gen6_bsd_ring_get_irq(struct intel_ring_buffer *ring)
1119 return gen6_ring_get_irq(ring,
1120 GT_GEN6_BSD_USER_INTERRUPT,
1121 GEN6_BSD_USER_INTERRUPT);
1125 gen6_bsd_ring_put_irq(struct intel_ring_buffer *ring)
1127 return gen6_ring_put_irq(ring,
1128 GT_GEN6_BSD_USER_INTERRUPT,
1129 GEN6_BSD_USER_INTERRUPT);
1132 /* ring buffer for Video Codec for Gen6+ */
1133 static const struct intel_ring_buffer gen6_bsd_ring = {
1134 .name = "gen6 bsd ring",
1136 .mmio_base = GEN6_BSD_RING_BASE,
1137 .size = 32 * PAGE_SIZE,
1138 .init = init_ring_common,
1139 .write_tail = gen6_bsd_ring_write_tail,
1140 .flush = gen6_ring_flush,
1141 .add_request = gen6_add_request,
1142 .get_seqno = ring_get_seqno,
1143 .irq_get = gen6_bsd_ring_get_irq,
1144 .irq_put = gen6_bsd_ring_put_irq,
1145 .dispatch_execbuffer = gen6_ring_dispatch_execbuffer,
1148 /* Blitter support (SandyBridge+) */
1151 blt_ring_get_irq(struct intel_ring_buffer *ring)
1153 return gen6_ring_get_irq(ring,
1154 GT_BLT_USER_INTERRUPT,
1155 GEN6_BLITTER_USER_INTERRUPT);
1159 blt_ring_put_irq(struct intel_ring_buffer *ring)
1161 gen6_ring_put_irq(ring,
1162 GT_BLT_USER_INTERRUPT,
1163 GEN6_BLITTER_USER_INTERRUPT);
1167 /* Workaround for some stepping of SNB,
1168 * each time when BLT engine ring tail moved,
1169 * the first command in the ring to be parsed
1170 * should be MI_BATCH_BUFFER_START
1172 #define NEED_BLT_WORKAROUND(dev) \
1173 (IS_GEN6(dev) && (dev->pdev->revision < 8))
1175 static inline struct drm_i915_gem_object *
1176 to_blt_workaround(struct intel_ring_buffer *ring)
1178 return ring->private;
1181 static int blt_ring_init(struct intel_ring_buffer *ring)
1183 if (NEED_BLT_WORKAROUND(ring->dev)) {
1184 struct drm_i915_gem_object *obj;
1188 obj = i915_gem_alloc_object(ring->dev, 4096);
1192 ret = i915_gem_object_pin(obj, 4096, true);
1194 drm_gem_object_unreference(&obj->base);
1198 ptr = kmap(obj->pages[0]);
1199 *ptr++ = MI_BATCH_BUFFER_END;
1201 kunmap(obj->pages[0]);
1203 ret = i915_gem_object_set_to_gtt_domain(obj, false);
1205 i915_gem_object_unpin(obj);
1206 drm_gem_object_unreference(&obj->base);
1210 ring->private = obj;
1213 return init_ring_common(ring);
1216 static int blt_ring_begin(struct intel_ring_buffer *ring,
1219 if (ring->private) {
1220 int ret = intel_ring_begin(ring, num_dwords+2);
1224 intel_ring_emit(ring, MI_BATCH_BUFFER_START);
1225 intel_ring_emit(ring, to_blt_workaround(ring)->gtt_offset);
1229 return intel_ring_begin(ring, 4);
1232 static int blt_ring_flush(struct intel_ring_buffer *ring,
1233 u32 invalidate_domains,
1238 if ((flush_domains & I915_GEM_DOMAIN_RENDER) == 0)
1241 ret = blt_ring_begin(ring, 4);
1245 intel_ring_emit(ring, MI_FLUSH_DW);
1246 intel_ring_emit(ring, 0);
1247 intel_ring_emit(ring, 0);
1248 intel_ring_emit(ring, 0);
1249 intel_ring_advance(ring);
1253 static void blt_ring_cleanup(struct intel_ring_buffer *ring)
1258 i915_gem_object_unpin(ring->private);
1259 drm_gem_object_unreference(ring->private);
1260 ring->private = NULL;
1263 static const struct intel_ring_buffer gen6_blt_ring = {
1266 .mmio_base = BLT_RING_BASE,
1267 .size = 32 * PAGE_SIZE,
1268 .init = blt_ring_init,
1269 .write_tail = ring_write_tail,
1270 .flush = blt_ring_flush,
1271 .add_request = gen6_add_request,
1272 .get_seqno = ring_get_seqno,
1273 .irq_get = blt_ring_get_irq,
1274 .irq_put = blt_ring_put_irq,
1275 .dispatch_execbuffer = gen6_ring_dispatch_execbuffer,
1276 .cleanup = blt_ring_cleanup,
1279 int intel_init_render_ring_buffer(struct drm_device *dev)
1281 drm_i915_private_t *dev_priv = dev->dev_private;
1282 struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
1284 *ring = render_ring;
1285 if (INTEL_INFO(dev)->gen >= 6) {
1286 ring->add_request = gen6_add_request;
1287 ring->irq_get = gen6_render_ring_get_irq;
1288 ring->irq_put = gen6_render_ring_put_irq;
1289 } else if (IS_GEN5(dev)) {
1290 ring->add_request = pc_render_add_request;
1291 ring->get_seqno = pc_render_get_seqno;
1294 if (!I915_NEED_GFX_HWS(dev)) {
1295 ring->status_page.page_addr = dev_priv->status_page_dmah->vaddr;
1296 memset(ring->status_page.page_addr, 0, PAGE_SIZE);
1299 return intel_init_ring_buffer(dev, ring);
1302 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size)
1304 drm_i915_private_t *dev_priv = dev->dev_private;
1305 struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
1307 *ring = render_ring;
1308 if (INTEL_INFO(dev)->gen >= 6) {
1309 ring->add_request = gen6_add_request;
1310 ring->irq_get = gen6_render_ring_get_irq;
1311 ring->irq_put = gen6_render_ring_put_irq;
1312 } else if (IS_GEN5(dev)) {
1313 ring->add_request = pc_render_add_request;
1314 ring->get_seqno = pc_render_get_seqno;
1318 INIT_LIST_HEAD(&ring->active_list);
1319 INIT_LIST_HEAD(&ring->request_list);
1320 INIT_LIST_HEAD(&ring->gpu_write_list);
1323 ring->effective_size = ring->size;
1324 if (IS_I830(ring->dev))
1325 ring->effective_size -= 128;
1327 ring->map.offset = start;
1328 ring->map.size = size;
1330 ring->map.flags = 0;
1333 drm_core_ioremap_wc(&ring->map, dev);
1334 if (ring->map.handle == NULL) {
1335 DRM_ERROR("can not ioremap virtual address for"
1340 ring->virtual_start = (void __force __iomem *)ring->map.handle;
1344 int intel_init_bsd_ring_buffer(struct drm_device *dev)
1346 drm_i915_private_t *dev_priv = dev->dev_private;
1347 struct intel_ring_buffer *ring = &dev_priv->ring[VCS];
1350 *ring = gen6_bsd_ring;
1354 return intel_init_ring_buffer(dev, ring);
1357 int intel_init_blt_ring_buffer(struct drm_device *dev)
1359 drm_i915_private_t *dev_priv = dev->dev_private;
1360 struct intel_ring_buffer *ring = &dev_priv->ring[BCS];
1362 *ring = gen6_blt_ring;
1364 return intel_init_ring_buffer(dev, ring);