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 u32 i915_gem_get_seqno(struct drm_device *dev)
39 drm_i915_private_t *dev_priv = dev->dev_private;
42 seqno = dev_priv->next_seqno;
44 /* reserve 0 for non-seqno */
45 if (++dev_priv->next_seqno == 0)
46 dev_priv->next_seqno = 1;
52 render_ring_flush(struct intel_ring_buffer *ring,
53 u32 invalidate_domains,
56 struct drm_device *dev = ring->dev;
57 drm_i915_private_t *dev_priv = dev->dev_private;
62 DRM_INFO("%s: invalidate %08x flush %08x\n", __func__,
63 invalidate_domains, flush_domains);
66 trace_i915_gem_request_flush(dev, dev_priv->next_seqno,
67 invalidate_domains, flush_domains);
69 if ((invalidate_domains | flush_domains) & I915_GEM_GPU_DOMAINS) {
73 * I915_GEM_DOMAIN_RENDER is always invalidated, but is
74 * only flushed if MI_NO_WRITE_FLUSH is unset. On 965, it is
75 * also flushed at 2d versus 3d pipeline switches.
79 * I915_GEM_DOMAIN_SAMPLER is flushed on pre-965 if
80 * MI_READ_FLUSH is set, and is always flushed on 965.
82 * I915_GEM_DOMAIN_COMMAND may not exist?
84 * I915_GEM_DOMAIN_INSTRUCTION, which exists on 965, is
85 * invalidated when MI_EXE_FLUSH is set.
87 * I915_GEM_DOMAIN_VERTEX, which exists on 965, is
88 * invalidated with every MI_FLUSH.
92 * On 965, TLBs associated with I915_GEM_DOMAIN_COMMAND
93 * and I915_GEM_DOMAIN_CPU in are invalidated at PTE write and
94 * I915_GEM_DOMAIN_RENDER and I915_GEM_DOMAIN_SAMPLER
95 * are flushed at any MI_FLUSH.
98 cmd = MI_FLUSH | MI_NO_WRITE_FLUSH;
99 if ((invalidate_domains|flush_domains) &
100 I915_GEM_DOMAIN_RENDER)
101 cmd &= ~MI_NO_WRITE_FLUSH;
102 if (INTEL_INFO(dev)->gen < 4) {
104 * On the 965, the sampler cache always gets flushed
105 * and this bit is reserved.
107 if (invalidate_domains & I915_GEM_DOMAIN_SAMPLER)
108 cmd |= MI_READ_FLUSH;
110 if (invalidate_domains & I915_GEM_DOMAIN_INSTRUCTION)
113 if (invalidate_domains & I915_GEM_DOMAIN_COMMAND &&
114 (IS_G4X(dev) || IS_GEN5(dev)))
115 cmd |= MI_INVALIDATE_ISP;
118 DRM_INFO("%s: queue flush %08x to ring\n", __func__, cmd);
120 ret = intel_ring_begin(ring, 2);
124 intel_ring_emit(ring, cmd);
125 intel_ring_emit(ring, MI_NOOP);
126 intel_ring_advance(ring);
132 static void ring_write_tail(struct intel_ring_buffer *ring,
135 drm_i915_private_t *dev_priv = ring->dev->dev_private;
136 I915_WRITE_TAIL(ring, value);
139 u32 intel_ring_get_active_head(struct intel_ring_buffer *ring)
141 drm_i915_private_t *dev_priv = ring->dev->dev_private;
142 u32 acthd_reg = INTEL_INFO(ring->dev)->gen >= 4 ?
143 RING_ACTHD(ring->mmio_base) : ACTHD;
145 return I915_READ(acthd_reg);
148 static int init_ring_common(struct intel_ring_buffer *ring)
150 drm_i915_private_t *dev_priv = ring->dev->dev_private;
151 struct drm_i915_gem_object *obj = ring->obj;
154 /* Stop the ring if it's running. */
155 I915_WRITE_CTL(ring, 0);
156 I915_WRITE_HEAD(ring, 0);
157 ring->write_tail(ring, 0);
159 /* Initialize the ring. */
160 I915_WRITE_START(ring, obj->gtt_offset);
161 head = I915_READ_HEAD(ring) & HEAD_ADDR;
163 /* G45 ring initialization fails to reset head to zero */
165 DRM_DEBUG_KMS("%s head not reset to zero "
166 "ctl %08x head %08x tail %08x start %08x\n",
169 I915_READ_HEAD(ring),
170 I915_READ_TAIL(ring),
171 I915_READ_START(ring));
173 I915_WRITE_HEAD(ring, 0);
175 if (I915_READ_HEAD(ring) & HEAD_ADDR) {
176 DRM_ERROR("failed to set %s head to zero "
177 "ctl %08x head %08x tail %08x start %08x\n",
180 I915_READ_HEAD(ring),
181 I915_READ_TAIL(ring),
182 I915_READ_START(ring));
187 ((ring->size - PAGE_SIZE) & RING_NR_PAGES)
188 | RING_REPORT_64K | RING_VALID);
190 /* If the head is still not zero, the ring is dead */
191 if ((I915_READ_CTL(ring) & RING_VALID) == 0 ||
192 I915_READ_START(ring) != obj->gtt_offset ||
193 (I915_READ_HEAD(ring) & HEAD_ADDR) != 0) {
194 DRM_ERROR("%s initialization failed "
195 "ctl %08x head %08x tail %08x start %08x\n",
198 I915_READ_HEAD(ring),
199 I915_READ_TAIL(ring),
200 I915_READ_START(ring));
204 if (!drm_core_check_feature(ring->dev, DRIVER_MODESET))
205 i915_kernel_lost_context(ring->dev);
207 ring->head = I915_READ_HEAD(ring) & HEAD_ADDR;
208 ring->tail = I915_READ_TAIL(ring) & TAIL_ADDR;
209 ring->space = ring->head - (ring->tail + 8);
211 ring->space += ring->size;
218 * 965+ support PIPE_CONTROL commands, which provide finer grained control
219 * over cache flushing.
221 struct pipe_control {
222 struct drm_i915_gem_object *obj;
223 volatile u32 *cpu_page;
228 init_pipe_control(struct intel_ring_buffer *ring)
230 struct pipe_control *pc;
231 struct drm_i915_gem_object *obj;
237 pc = kmalloc(sizeof(*pc), GFP_KERNEL);
241 obj = i915_gem_alloc_object(ring->dev, 4096);
243 DRM_ERROR("Failed to allocate seqno page\n");
247 obj->agp_type = AGP_USER_CACHED_MEMORY;
249 ret = i915_gem_object_pin(obj, 4096, true);
253 pc->gtt_offset = obj->gtt_offset;
254 pc->cpu_page = kmap(obj->pages[0]);
255 if (pc->cpu_page == NULL)
263 i915_gem_object_unpin(obj);
265 drm_gem_object_unreference(&obj->base);
272 cleanup_pipe_control(struct intel_ring_buffer *ring)
274 struct pipe_control *pc = ring->private;
275 struct drm_i915_gem_object *obj;
281 kunmap(obj->pages[0]);
282 i915_gem_object_unpin(obj);
283 drm_gem_object_unreference(&obj->base);
286 ring->private = NULL;
289 static int init_render_ring(struct intel_ring_buffer *ring)
291 struct drm_device *dev = ring->dev;
292 struct drm_i915_private *dev_priv = dev->dev_private;
293 int ret = init_ring_common(ring);
295 if (INTEL_INFO(dev)->gen > 3) {
296 int mode = VS_TIMER_DISPATCH << 16 | VS_TIMER_DISPATCH;
298 mode |= MI_FLUSH_ENABLE << 16 | MI_FLUSH_ENABLE;
299 I915_WRITE(MI_MODE, mode);
302 if (INTEL_INFO(dev)->gen >= 6) {
303 } else if (IS_GEN5(dev)) {
304 ret = init_pipe_control(ring);
312 static void render_ring_cleanup(struct intel_ring_buffer *ring)
317 cleanup_pipe_control(ring);
321 update_semaphore(struct intel_ring_buffer *ring, int i, u32 seqno)
323 struct drm_device *dev = ring->dev;
324 struct drm_i915_private *dev_priv = dev->dev_private;
328 * cs -> 1 = vcs, 0 = bcs
329 * vcs -> 1 = bcs, 0 = cs,
330 * bcs -> 1 = cs, 0 = vcs.
332 id = ring - dev_priv->ring;
336 intel_ring_emit(ring,
338 MI_SEMAPHORE_REGISTER |
339 MI_SEMAPHORE_UPDATE);
340 intel_ring_emit(ring, seqno);
341 intel_ring_emit(ring,
342 RING_SYNC_0(dev_priv->ring[id].mmio_base) + 4*i);
346 gen6_add_request(struct intel_ring_buffer *ring,
352 ret = intel_ring_begin(ring, 10);
356 seqno = i915_gem_get_seqno(ring->dev);
357 update_semaphore(ring, 0, seqno);
358 update_semaphore(ring, 1, seqno);
360 intel_ring_emit(ring, MI_STORE_DWORD_INDEX);
361 intel_ring_emit(ring, I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
362 intel_ring_emit(ring, seqno);
363 intel_ring_emit(ring, MI_USER_INTERRUPT);
364 intel_ring_advance(ring);
371 intel_ring_sync(struct intel_ring_buffer *ring,
372 struct intel_ring_buffer *to,
377 ret = intel_ring_begin(ring, 4);
381 intel_ring_emit(ring,
383 MI_SEMAPHORE_REGISTER |
384 intel_ring_sync_index(ring, to) << 17 |
385 MI_SEMAPHORE_COMPARE);
386 intel_ring_emit(ring, seqno);
387 intel_ring_emit(ring, 0);
388 intel_ring_emit(ring, MI_NOOP);
389 intel_ring_advance(ring);
394 #define PIPE_CONTROL_FLUSH(ring__, addr__) \
396 intel_ring_emit(ring__, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE | \
397 PIPE_CONTROL_DEPTH_STALL | 2); \
398 intel_ring_emit(ring__, (addr__) | PIPE_CONTROL_GLOBAL_GTT); \
399 intel_ring_emit(ring__, 0); \
400 intel_ring_emit(ring__, 0); \
404 pc_render_add_request(struct intel_ring_buffer *ring,
407 struct drm_device *dev = ring->dev;
408 u32 seqno = i915_gem_get_seqno(dev);
409 struct pipe_control *pc = ring->private;
410 u32 scratch_addr = pc->gtt_offset + 128;
413 /* For Ironlake, MI_USER_INTERRUPT was deprecated and apparently
414 * incoherent with writes to memory, i.e. completely fubar,
415 * so we need to use PIPE_NOTIFY instead.
417 * However, we also need to workaround the qword write
418 * incoherence by flushing the 6 PIPE_NOTIFY buffers out to
419 * memory before requesting an interrupt.
421 ret = intel_ring_begin(ring, 32);
425 intel_ring_emit(ring, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |
426 PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_TC_FLUSH);
427 intel_ring_emit(ring, pc->gtt_offset | PIPE_CONTROL_GLOBAL_GTT);
428 intel_ring_emit(ring, seqno);
429 intel_ring_emit(ring, 0);
430 PIPE_CONTROL_FLUSH(ring, scratch_addr);
431 scratch_addr += 128; /* write to separate cachelines */
432 PIPE_CONTROL_FLUSH(ring, scratch_addr);
434 PIPE_CONTROL_FLUSH(ring, scratch_addr);
436 PIPE_CONTROL_FLUSH(ring, scratch_addr);
438 PIPE_CONTROL_FLUSH(ring, scratch_addr);
440 PIPE_CONTROL_FLUSH(ring, scratch_addr);
441 intel_ring_emit(ring, GFX_OP_PIPE_CONTROL | PIPE_CONTROL_QW_WRITE |
442 PIPE_CONTROL_WC_FLUSH | PIPE_CONTROL_TC_FLUSH |
443 PIPE_CONTROL_NOTIFY);
444 intel_ring_emit(ring, pc->gtt_offset | PIPE_CONTROL_GLOBAL_GTT);
445 intel_ring_emit(ring, seqno);
446 intel_ring_emit(ring, 0);
447 intel_ring_advance(ring);
454 render_ring_add_request(struct intel_ring_buffer *ring,
457 struct drm_device *dev = ring->dev;
458 u32 seqno = i915_gem_get_seqno(dev);
461 ret = intel_ring_begin(ring, 4);
465 intel_ring_emit(ring, MI_STORE_DWORD_INDEX);
466 intel_ring_emit(ring, I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
467 intel_ring_emit(ring, seqno);
468 intel_ring_emit(ring, MI_USER_INTERRUPT);
469 intel_ring_advance(ring);
476 ring_get_seqno(struct intel_ring_buffer *ring)
478 return intel_read_status_page(ring, I915_GEM_HWS_INDEX);
482 pc_render_get_seqno(struct intel_ring_buffer *ring)
484 struct pipe_control *pc = ring->private;
485 return pc->cpu_page[0];
489 ironlake_enable_irq(drm_i915_private_t *dev_priv, u32 mask)
491 dev_priv->gt_irq_mask &= ~mask;
492 I915_WRITE(GTIMR, dev_priv->gt_irq_mask);
497 ironlake_disable_irq(drm_i915_private_t *dev_priv, u32 mask)
499 dev_priv->gt_irq_mask |= mask;
500 I915_WRITE(GTIMR, dev_priv->gt_irq_mask);
505 i915_enable_irq(drm_i915_private_t *dev_priv, u32 mask)
507 dev_priv->irq_mask &= ~mask;
508 I915_WRITE(IMR, dev_priv->irq_mask);
513 i915_disable_irq(drm_i915_private_t *dev_priv, u32 mask)
515 dev_priv->irq_mask |= mask;
516 I915_WRITE(IMR, dev_priv->irq_mask);
521 render_ring_get_irq(struct intel_ring_buffer *ring)
523 struct drm_device *dev = ring->dev;
524 drm_i915_private_t *dev_priv = dev->dev_private;
526 if (!dev->irq_enabled)
529 spin_lock(&ring->irq_lock);
530 if (ring->irq_refcount++ == 0) {
531 if (HAS_PCH_SPLIT(dev))
532 ironlake_enable_irq(dev_priv,
533 GT_PIPE_NOTIFY | GT_USER_INTERRUPT);
535 i915_enable_irq(dev_priv, I915_USER_INTERRUPT);
537 spin_unlock(&ring->irq_lock);
543 render_ring_put_irq(struct intel_ring_buffer *ring)
545 struct drm_device *dev = ring->dev;
546 drm_i915_private_t *dev_priv = dev->dev_private;
548 spin_lock(&ring->irq_lock);
549 if (--ring->irq_refcount == 0) {
550 if (HAS_PCH_SPLIT(dev))
551 ironlake_disable_irq(dev_priv,
555 i915_disable_irq(dev_priv, I915_USER_INTERRUPT);
557 spin_unlock(&ring->irq_lock);
560 void intel_ring_setup_status_page(struct intel_ring_buffer *ring)
562 drm_i915_private_t *dev_priv = ring->dev->dev_private;
563 u32 mmio = IS_GEN6(ring->dev) ?
564 RING_HWS_PGA_GEN6(ring->mmio_base) :
565 RING_HWS_PGA(ring->mmio_base);
566 I915_WRITE(mmio, (u32)ring->status_page.gfx_addr);
571 bsd_ring_flush(struct intel_ring_buffer *ring,
572 u32 invalidate_domains,
577 if ((flush_domains & I915_GEM_DOMAIN_RENDER) == 0)
580 ret = intel_ring_begin(ring, 2);
584 intel_ring_emit(ring, MI_FLUSH);
585 intel_ring_emit(ring, MI_NOOP);
586 intel_ring_advance(ring);
591 ring_add_request(struct intel_ring_buffer *ring,
597 ret = intel_ring_begin(ring, 4);
601 seqno = i915_gem_get_seqno(ring->dev);
603 intel_ring_emit(ring, MI_STORE_DWORD_INDEX);
604 intel_ring_emit(ring, I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT);
605 intel_ring_emit(ring, seqno);
606 intel_ring_emit(ring, MI_USER_INTERRUPT);
607 intel_ring_advance(ring);
609 DRM_DEBUG_DRIVER("%s %d\n", ring->name, seqno);
615 ring_get_irq(struct intel_ring_buffer *ring, u32 flag)
617 struct drm_device *dev = ring->dev;
618 drm_i915_private_t *dev_priv = dev->dev_private;
620 if (!dev->irq_enabled)
623 spin_lock(&ring->irq_lock);
624 if (ring->irq_refcount++ == 0)
625 ironlake_enable_irq(dev_priv, flag);
626 spin_unlock(&ring->irq_lock);
632 ring_put_irq(struct intel_ring_buffer *ring, u32 flag)
634 struct drm_device *dev = ring->dev;
635 drm_i915_private_t *dev_priv = dev->dev_private;
637 spin_lock(&ring->irq_lock);
638 if (--ring->irq_refcount == 0)
639 ironlake_disable_irq(dev_priv, flag);
640 spin_unlock(&ring->irq_lock);
644 gen6_ring_get_irq(struct intel_ring_buffer *ring, u32 gflag, u32 rflag)
646 struct drm_device *dev = ring->dev;
647 drm_i915_private_t *dev_priv = dev->dev_private;
649 if (!dev->irq_enabled)
652 spin_lock(&ring->irq_lock);
653 if (ring->irq_refcount++ == 0) {
654 ring->irq_mask &= ~rflag;
655 I915_WRITE_IMR(ring, ring->irq_mask);
656 ironlake_enable_irq(dev_priv, gflag);
658 spin_unlock(&ring->irq_lock);
664 gen6_ring_put_irq(struct intel_ring_buffer *ring, u32 gflag, u32 rflag)
666 struct drm_device *dev = ring->dev;
667 drm_i915_private_t *dev_priv = dev->dev_private;
669 spin_lock(&ring->irq_lock);
670 if (--ring->irq_refcount == 0) {
671 ring->irq_mask |= rflag;
672 I915_WRITE_IMR(ring, ring->irq_mask);
673 ironlake_disable_irq(dev_priv, gflag);
675 spin_unlock(&ring->irq_lock);
679 bsd_ring_get_irq(struct intel_ring_buffer *ring)
681 return ring_get_irq(ring, GT_BSD_USER_INTERRUPT);
684 bsd_ring_put_irq(struct intel_ring_buffer *ring)
686 ring_put_irq(ring, GT_BSD_USER_INTERRUPT);
690 ring_dispatch_execbuffer(struct intel_ring_buffer *ring, u32 offset, u32 length)
694 ret = intel_ring_begin(ring, 2);
698 intel_ring_emit(ring,
699 MI_BATCH_BUFFER_START | (2 << 6) |
700 MI_BATCH_NON_SECURE_I965);
701 intel_ring_emit(ring, offset);
702 intel_ring_advance(ring);
708 render_ring_dispatch_execbuffer(struct intel_ring_buffer *ring,
711 struct drm_device *dev = ring->dev;
712 drm_i915_private_t *dev_priv = dev->dev_private;
715 trace_i915_gem_request_submit(dev, dev_priv->next_seqno + 1);
717 if (IS_I830(dev) || IS_845G(dev)) {
718 ret = intel_ring_begin(ring, 4);
722 intel_ring_emit(ring, MI_BATCH_BUFFER);
723 intel_ring_emit(ring, offset | MI_BATCH_NON_SECURE);
724 intel_ring_emit(ring, offset + len - 8);
725 intel_ring_emit(ring, 0);
727 ret = intel_ring_begin(ring, 2);
731 if (INTEL_INFO(dev)->gen >= 4) {
732 intel_ring_emit(ring,
733 MI_BATCH_BUFFER_START | (2 << 6) |
734 MI_BATCH_NON_SECURE_I965);
735 intel_ring_emit(ring, offset);
737 intel_ring_emit(ring,
738 MI_BATCH_BUFFER_START | (2 << 6));
739 intel_ring_emit(ring, offset | MI_BATCH_NON_SECURE);
742 intel_ring_advance(ring);
747 static void cleanup_status_page(struct intel_ring_buffer *ring)
749 drm_i915_private_t *dev_priv = ring->dev->dev_private;
750 struct drm_i915_gem_object *obj;
752 obj = ring->status_page.obj;
756 kunmap(obj->pages[0]);
757 i915_gem_object_unpin(obj);
758 drm_gem_object_unreference(&obj->base);
759 ring->status_page.obj = NULL;
761 memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map));
764 static int init_status_page(struct intel_ring_buffer *ring)
766 struct drm_device *dev = ring->dev;
767 drm_i915_private_t *dev_priv = dev->dev_private;
768 struct drm_i915_gem_object *obj;
771 obj = i915_gem_alloc_object(dev, 4096);
773 DRM_ERROR("Failed to allocate status page\n");
777 obj->agp_type = AGP_USER_CACHED_MEMORY;
779 ret = i915_gem_object_pin(obj, 4096, true);
784 ring->status_page.gfx_addr = obj->gtt_offset;
785 ring->status_page.page_addr = kmap(obj->pages[0]);
786 if (ring->status_page.page_addr == NULL) {
787 memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map));
790 ring->status_page.obj = obj;
791 memset(ring->status_page.page_addr, 0, PAGE_SIZE);
793 intel_ring_setup_status_page(ring);
794 DRM_DEBUG_DRIVER("%s hws offset: 0x%08x\n",
795 ring->name, ring->status_page.gfx_addr);
800 i915_gem_object_unpin(obj);
802 drm_gem_object_unreference(&obj->base);
807 int intel_init_ring_buffer(struct drm_device *dev,
808 struct intel_ring_buffer *ring)
810 struct drm_i915_gem_object *obj;
814 INIT_LIST_HEAD(&ring->active_list);
815 INIT_LIST_HEAD(&ring->request_list);
816 INIT_LIST_HEAD(&ring->gpu_write_list);
818 spin_lock_init(&ring->irq_lock);
821 if (I915_NEED_GFX_HWS(dev)) {
822 ret = init_status_page(ring);
827 obj = i915_gem_alloc_object(dev, ring->size);
829 DRM_ERROR("Failed to allocate ringbuffer\n");
836 ret = i915_gem_object_pin(obj, PAGE_SIZE, true);
840 ring->map.size = ring->size;
841 ring->map.offset = dev->agp->base + obj->gtt_offset;
846 drm_core_ioremap_wc(&ring->map, dev);
847 if (ring->map.handle == NULL) {
848 DRM_ERROR("Failed to map ringbuffer.\n");
853 ring->virtual_start = ring->map.handle;
854 ret = ring->init(ring);
858 /* Workaround an erratum on the i830 which causes a hang if
859 * the TAIL pointer points to within the last 2 cachelines
862 ring->effective_size = ring->size;
863 if (IS_I830(ring->dev))
864 ring->effective_size -= 128;
869 drm_core_ioremapfree(&ring->map, dev);
871 i915_gem_object_unpin(obj);
873 drm_gem_object_unreference(&obj->base);
876 cleanup_status_page(ring);
880 void intel_cleanup_ring_buffer(struct intel_ring_buffer *ring)
882 struct drm_i915_private *dev_priv;
885 if (ring->obj == NULL)
888 /* Disable the ring buffer. The ring must be idle at this point */
889 dev_priv = ring->dev->dev_private;
890 ret = intel_wait_ring_buffer(ring, ring->size - 8);
891 I915_WRITE_CTL(ring, 0);
893 drm_core_ioremapfree(&ring->map, ring->dev);
895 i915_gem_object_unpin(ring->obj);
896 drm_gem_object_unreference(&ring->obj->base);
902 cleanup_status_page(ring);
905 static int intel_wrap_ring_buffer(struct intel_ring_buffer *ring)
908 int rem = ring->size - ring->tail;
910 if (ring->space < rem) {
911 int ret = intel_wait_ring_buffer(ring, rem);
916 virt = (unsigned int *)(ring->virtual_start + ring->tail);
924 ring->space = ring->head - 8;
929 int intel_wait_ring_buffer(struct intel_ring_buffer *ring, int n)
931 struct drm_device *dev = ring->dev;
932 struct drm_i915_private *dev_priv = dev->dev_private;
936 trace_i915_ring_wait_begin (dev);
937 end = jiffies + 3 * HZ;
939 /* If the reported head position has wrapped or hasn't advanced,
940 * fallback to the slow and accurate path.
942 head = intel_read_status_page(ring, 4);
943 if (head < ring->actual_head)
944 head = I915_READ_HEAD(ring);
945 ring->actual_head = head;
946 ring->head = head & HEAD_ADDR;
947 ring->space = ring->head - (ring->tail + 8);
949 ring->space += ring->size;
950 if (ring->space >= n) {
951 trace_i915_ring_wait_end(dev);
955 if (dev->primary->master) {
956 struct drm_i915_master_private *master_priv = dev->primary->master->driver_priv;
957 if (master_priv->sarea_priv)
958 master_priv->sarea_priv->perf_boxes |= I915_BOX_WAIT;
962 if (atomic_read(&dev_priv->mm.wedged))
964 } while (!time_after(jiffies, end));
965 trace_i915_ring_wait_end (dev);
969 int intel_ring_begin(struct intel_ring_buffer *ring,
972 int n = 4*num_dwords;
975 if (unlikely(ring->tail + n > ring->effective_size)) {
976 ret = intel_wrap_ring_buffer(ring);
981 if (unlikely(ring->space < n)) {
982 ret = intel_wait_ring_buffer(ring, n);
991 void intel_ring_advance(struct intel_ring_buffer *ring)
993 ring->tail &= ring->size - 1;
994 ring->write_tail(ring, ring->tail);
997 static const struct intel_ring_buffer render_ring = {
998 .name = "render ring",
1000 .mmio_base = RENDER_RING_BASE,
1001 .size = 32 * PAGE_SIZE,
1002 .init = init_render_ring,
1003 .write_tail = ring_write_tail,
1004 .flush = render_ring_flush,
1005 .add_request = render_ring_add_request,
1006 .get_seqno = ring_get_seqno,
1007 .irq_get = render_ring_get_irq,
1008 .irq_put = render_ring_put_irq,
1009 .dispatch_execbuffer = render_ring_dispatch_execbuffer,
1010 .cleanup = render_ring_cleanup,
1013 /* ring buffer for bit-stream decoder */
1015 static const struct intel_ring_buffer bsd_ring = {
1018 .mmio_base = BSD_RING_BASE,
1019 .size = 32 * PAGE_SIZE,
1020 .init = init_ring_common,
1021 .write_tail = ring_write_tail,
1022 .flush = bsd_ring_flush,
1023 .add_request = ring_add_request,
1024 .get_seqno = ring_get_seqno,
1025 .irq_get = bsd_ring_get_irq,
1026 .irq_put = bsd_ring_put_irq,
1027 .dispatch_execbuffer = ring_dispatch_execbuffer,
1031 static void gen6_bsd_ring_write_tail(struct intel_ring_buffer *ring,
1034 drm_i915_private_t *dev_priv = ring->dev->dev_private;
1036 /* Every tail move must follow the sequence below */
1037 I915_WRITE(GEN6_BSD_SLEEP_PSMI_CONTROL,
1038 GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_MODIFY_MASK |
1039 GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_DISABLE);
1040 I915_WRITE(GEN6_BSD_RNCID, 0x0);
1042 if (wait_for((I915_READ(GEN6_BSD_SLEEP_PSMI_CONTROL) &
1043 GEN6_BSD_SLEEP_PSMI_CONTROL_IDLE_INDICATOR) == 0,
1045 DRM_ERROR("timed out waiting for IDLE Indicator\n");
1047 I915_WRITE_TAIL(ring, value);
1048 I915_WRITE(GEN6_BSD_SLEEP_PSMI_CONTROL,
1049 GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_MODIFY_MASK |
1050 GEN6_BSD_SLEEP_PSMI_CONTROL_RC_ILDL_MESSAGE_ENABLE);
1053 static int gen6_ring_flush(struct intel_ring_buffer *ring,
1054 u32 invalidate_domains,
1059 if ((flush_domains & I915_GEM_DOMAIN_RENDER) == 0)
1062 ret = intel_ring_begin(ring, 4);
1066 intel_ring_emit(ring, MI_FLUSH_DW);
1067 intel_ring_emit(ring, 0);
1068 intel_ring_emit(ring, 0);
1069 intel_ring_emit(ring, 0);
1070 intel_ring_advance(ring);
1075 gen6_ring_dispatch_execbuffer(struct intel_ring_buffer *ring,
1076 u32 offset, u32 len)
1080 ret = intel_ring_begin(ring, 2);
1084 intel_ring_emit(ring, MI_BATCH_BUFFER_START | MI_BATCH_NON_SECURE_I965);
1085 /* bit0-7 is the length on GEN6+ */
1086 intel_ring_emit(ring, offset);
1087 intel_ring_advance(ring);
1093 gen6_render_ring_get_irq(struct intel_ring_buffer *ring)
1095 return gen6_ring_get_irq(ring,
1097 GEN6_RENDER_USER_INTERRUPT);
1101 gen6_render_ring_put_irq(struct intel_ring_buffer *ring)
1103 return gen6_ring_put_irq(ring,
1105 GEN6_RENDER_USER_INTERRUPT);
1109 gen6_bsd_ring_get_irq(struct intel_ring_buffer *ring)
1111 return gen6_ring_get_irq(ring,
1112 GT_GEN6_BSD_USER_INTERRUPT,
1113 GEN6_BSD_USER_INTERRUPT);
1117 gen6_bsd_ring_put_irq(struct intel_ring_buffer *ring)
1119 return gen6_ring_put_irq(ring,
1120 GT_GEN6_BSD_USER_INTERRUPT,
1121 GEN6_BSD_USER_INTERRUPT);
1124 /* ring buffer for Video Codec for Gen6+ */
1125 static const struct intel_ring_buffer gen6_bsd_ring = {
1126 .name = "gen6 bsd ring",
1128 .mmio_base = GEN6_BSD_RING_BASE,
1129 .size = 32 * PAGE_SIZE,
1130 .init = init_ring_common,
1131 .write_tail = gen6_bsd_ring_write_tail,
1132 .flush = gen6_ring_flush,
1133 .add_request = gen6_add_request,
1134 .get_seqno = ring_get_seqno,
1135 .irq_get = gen6_bsd_ring_get_irq,
1136 .irq_put = gen6_bsd_ring_put_irq,
1137 .dispatch_execbuffer = gen6_ring_dispatch_execbuffer,
1140 /* Blitter support (SandyBridge+) */
1143 blt_ring_get_irq(struct intel_ring_buffer *ring)
1145 return gen6_ring_get_irq(ring,
1146 GT_BLT_USER_INTERRUPT,
1147 GEN6_BLITTER_USER_INTERRUPT);
1151 blt_ring_put_irq(struct intel_ring_buffer *ring)
1153 gen6_ring_put_irq(ring,
1154 GT_BLT_USER_INTERRUPT,
1155 GEN6_BLITTER_USER_INTERRUPT);
1159 /* Workaround for some stepping of SNB,
1160 * each time when BLT engine ring tail moved,
1161 * the first command in the ring to be parsed
1162 * should be MI_BATCH_BUFFER_START
1164 #define NEED_BLT_WORKAROUND(dev) \
1165 (IS_GEN6(dev) && (dev->pdev->revision < 8))
1167 static inline struct drm_i915_gem_object *
1168 to_blt_workaround(struct intel_ring_buffer *ring)
1170 return ring->private;
1173 static int blt_ring_init(struct intel_ring_buffer *ring)
1175 if (NEED_BLT_WORKAROUND(ring->dev)) {
1176 struct drm_i915_gem_object *obj;
1180 obj = i915_gem_alloc_object(ring->dev, 4096);
1184 ret = i915_gem_object_pin(obj, 4096, true);
1186 drm_gem_object_unreference(&obj->base);
1190 ptr = kmap(obj->pages[0]);
1191 *ptr++ = MI_BATCH_BUFFER_END;
1193 kunmap(obj->pages[0]);
1195 ret = i915_gem_object_set_to_gtt_domain(obj, false);
1197 i915_gem_object_unpin(obj);
1198 drm_gem_object_unreference(&obj->base);
1202 ring->private = obj;
1205 return init_ring_common(ring);
1208 static int blt_ring_begin(struct intel_ring_buffer *ring,
1211 if (ring->private) {
1212 int ret = intel_ring_begin(ring, num_dwords+2);
1216 intel_ring_emit(ring, MI_BATCH_BUFFER_START);
1217 intel_ring_emit(ring, to_blt_workaround(ring)->gtt_offset);
1221 return intel_ring_begin(ring, 4);
1224 static int blt_ring_flush(struct intel_ring_buffer *ring,
1225 u32 invalidate_domains,
1230 if ((flush_domains & I915_GEM_DOMAIN_RENDER) == 0)
1233 ret = blt_ring_begin(ring, 4);
1237 intel_ring_emit(ring, MI_FLUSH_DW);
1238 intel_ring_emit(ring, 0);
1239 intel_ring_emit(ring, 0);
1240 intel_ring_emit(ring, 0);
1241 intel_ring_advance(ring);
1245 static void blt_ring_cleanup(struct intel_ring_buffer *ring)
1250 i915_gem_object_unpin(ring->private);
1251 drm_gem_object_unreference(ring->private);
1252 ring->private = NULL;
1255 static const struct intel_ring_buffer gen6_blt_ring = {
1258 .mmio_base = BLT_RING_BASE,
1259 .size = 32 * PAGE_SIZE,
1260 .init = blt_ring_init,
1261 .write_tail = ring_write_tail,
1262 .flush = blt_ring_flush,
1263 .add_request = gen6_add_request,
1264 .get_seqno = ring_get_seqno,
1265 .irq_get = blt_ring_get_irq,
1266 .irq_put = blt_ring_put_irq,
1267 .dispatch_execbuffer = gen6_ring_dispatch_execbuffer,
1268 .cleanup = blt_ring_cleanup,
1271 int intel_init_render_ring_buffer(struct drm_device *dev)
1273 drm_i915_private_t *dev_priv = dev->dev_private;
1274 struct intel_ring_buffer *ring = &dev_priv->ring[RCS];
1276 *ring = render_ring;
1277 if (INTEL_INFO(dev)->gen >= 6) {
1278 ring->add_request = gen6_add_request;
1279 ring->irq_get = gen6_render_ring_get_irq;
1280 ring->irq_put = gen6_render_ring_put_irq;
1281 } else if (IS_GEN5(dev)) {
1282 ring->add_request = pc_render_add_request;
1283 ring->get_seqno = pc_render_get_seqno;
1286 if (!I915_NEED_GFX_HWS(dev)) {
1287 ring->status_page.page_addr = dev_priv->status_page_dmah->vaddr;
1288 memset(ring->status_page.page_addr, 0, PAGE_SIZE);
1291 return intel_init_ring_buffer(dev, ring);
1294 int intel_init_bsd_ring_buffer(struct drm_device *dev)
1296 drm_i915_private_t *dev_priv = dev->dev_private;
1297 struct intel_ring_buffer *ring = &dev_priv->ring[VCS];
1300 *ring = gen6_bsd_ring;
1304 return intel_init_ring_buffer(dev, ring);
1307 int intel_init_blt_ring_buffer(struct drm_device *dev)
1309 drm_i915_private_t *dev_priv = dev->dev_private;
1310 struct intel_ring_buffer *ring = &dev_priv->ring[BCS];
1312 *ring = gen6_blt_ring;
1314 return intel_init_ring_buffer(dev, ring);