[Intel-gfx] [PATCH 4/5] drm/i915: Implement fair lru eviction across both rings.
Chris Wilson
chris at chris-wilson.co.uk
Wed Aug 4 13:20:36 CEST 2010
Based in a large part upon Daniel Vetter's implementation and adapted
for handling multiple rings in a single pass.
This should lead to better gtt usage and fixes the page-fault-of-doom
triggered. The fairness is provided by scanning through the GTT space
amalgamating space in rendering order. As soon as we have a contiguous
space in the GTT large enough for the new object (and its alignment),
evict any object which lies within that space. This should keep more
objects resident in the GTT.
Doing throughput testing on a PineView machine with cairo-perf-trace
indicates that there is very little difference with the new LRU scan,
perhaps a small improvement... Except oddly for the poppler trace.
Reference:
Bug 15911 - Intermittent X crash (freeze)
https://bugzilla.kernel.org/show_bug.cgi?id=15911
Bug 20152 - cannot view JPG in firefox when running UXA
https://bugs.freedesktop.org/show_bug.cgi?id=20152
Bug 24369 - Hang when scrolling firefox page with window in front
https://bugs.freedesktop.org/show_bug.cgi?id=24369
Bug 28478 - Intermittent graphics lockups due to overflow/loop
https://bugs.freedesktop.org/show_bug.cgi?id=28478
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/i915/i915_drv.h | 2 +
drivers/gpu/drm/i915/i915_gem_evict.c | 253 +++++++++++++++------------------
2 files changed, 118 insertions(+), 137 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 75fe397..61a8c90 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -668,6 +668,8 @@ struct drm_i915_gem_object {
struct list_head list;
/** This object's place on GPU write list */
struct list_head gpu_write_list;
+ /** This object's place on eviction list */
+ struct list_head evict_list;
/**
* This is set if the object is on the active or flushing lists
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 479e450..4e5af69 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -31,167 +31,146 @@
#include "i915_drv.h"
#include "i915_drm.h"
-static inline int
-i915_gem_object_is_purgeable(struct drm_i915_gem_object *obj_priv)
-{
- return obj_priv->madv == I915_MADV_DONTNEED;
-}
-
-static int
-i915_gem_scan_inactive_list_and_evict(struct drm_device *dev, int min_size,
- unsigned alignment, int *found)
+static struct drm_i915_gem_object *
+i915_gem_next_active_object(struct drm_device *dev,
+ struct list_head **render_iter,
+ struct list_head **bsd_iter)
{
drm_i915_private_t *dev_priv = dev->dev_private;
- struct drm_gem_object *obj;
- struct drm_i915_gem_object *obj_priv;
- struct drm_gem_object *best = NULL;
- struct drm_gem_object *first = NULL;
+ struct drm_i915_gem_object *render_obj = NULL, *bsd_obj = NULL;
- /* Try to find the smallest clean object */
- list_for_each_entry(obj_priv, &dev_priv->mm.inactive_list, list) {
- struct drm_gem_object *obj = &obj_priv->base;
- if (obj->size >= min_size) {
- if ((!obj_priv->dirty ||
- i915_gem_object_is_purgeable(obj_priv)) &&
- (!best || obj->size < best->size)) {
- best = obj;
- if (best->size == min_size)
- break;
- }
- if (!first)
- first = obj;
- }
- }
-
- obj = best ? best : first;
-
- if (!obj) {
- *found = 0;
- return 0;
- }
+ if (*render_iter != &dev_priv->render_ring.active_list)
+ render_obj = list_entry(*render_iter,
+ struct drm_i915_gem_object,
+ list);
- *found = 1;
+ if (HAS_BSD(dev)) {
+ if (*bsd_iter != &dev_priv->bsd_ring.active_list)
+ bsd_obj = list_entry(*bsd_iter,
+ struct drm_i915_gem_object,
+ list);
-#if WATCH_LRU
- DRM_INFO("%s: evicting %p\n", __func__, obj);
-#endif
- obj_priv = to_intel_bo(obj);
- BUG_ON(obj_priv->pin_count != 0);
- BUG_ON(obj_priv->active);
+ if (render_obj == NULL) {
+ *bsd_iter = (*bsd_iter)->next;
+ return bsd_obj;
+ }
- /* Wait on the rendering and unbind the buffer. */
- return i915_gem_object_unbind(obj);
-}
+ if (bsd_obj == NULL) {
+ *render_iter = (*render_iter)->next;
+ return render_obj;
+ }
-static void
-i915_gem_flush_ring(struct drm_device *dev,
- uint32_t invalidate_domains,
- uint32_t flush_domains,
- struct intel_ring_buffer *ring)
-{
- if (flush_domains & I915_GEM_DOMAIN_CPU)
- drm_agp_chipset_flush(dev);
- ring->flush(dev, ring,
- invalidate_domains,
- flush_domains);
+ /* XXX can we handle seqno wrapping? */
+ if (render_obj->last_rendering_seqno < bsd_obj->last_rendering_seqno) {
+ *render_iter = (*render_iter)->next;
+ return render_obj;
+ } else {
+ *bsd_iter = (*bsd_iter)->next;
+ return bsd_obj;
+ }
+ } else {
+ *render_iter = (*render_iter)->next;
+ return render_obj;
+ }
}
int
-i915_gem_evict_something(struct drm_device *dev,
- int min_size, unsigned alignment)
+i915_gem_evict_something(struct drm_device *dev, int min_size, unsigned alignment)
{
drm_i915_private_t *dev_priv = dev->dev_private;
- int ret, found;
-
- struct intel_ring_buffer *render_ring = &dev_priv->render_ring;
- struct intel_ring_buffer *bsd_ring = &dev_priv->bsd_ring;
- for (;;) {
- i915_gem_retire_requests(dev);
-
- /* If there's an inactive buffer available now, grab it
- * and be done.
- */
- ret = i915_gem_scan_inactive_list_and_evict(dev, min_size,
- alignment,
- &found);
- if (found)
- return ret;
+ struct list_head eviction_list, unwind_list;
+ struct drm_i915_gem_object *obj_priv, *tmp_obj_priv;
+ struct list_head *render_iter, *bsd_iter;
+ int ret = 0;
+
+ i915_gem_retire_requests(dev);
- /* If we didn't get anything, but the ring is still processing
- * things, wait for the next to finish and hopefully leave us
- * a buffer to evict.
- */
- if (!list_empty(&render_ring->request_list)) {
- struct drm_i915_gem_request *request;
+ /* Re-check for free space after retiring requests */
+ if (drm_mm_search_free(&dev_priv->mm.gtt_space,
+ min_size, alignment, 0))
+ return 0;
- request = list_first_entry(&render_ring->request_list,
- struct drm_i915_gem_request,
- list);
+ INIT_LIST_HEAD(&unwind_list);
+ drm_mm_init_scan(&dev_priv->mm.gtt_space, min_size, alignment);
- ret = i915_do_wait_request(dev, request->seqno, true, request->ring);
- if (ret)
- return ret;
+ /* First see if there is a large enough contiguous idle region... */
+ list_for_each_entry(obj_priv, &dev_priv->mm.inactive_list, list) {
+ list_add(&obj_priv->evict_list, &unwind_list);
+ if (drm_mm_scan_add_block(obj_priv->gtt_space))
+ goto found;
+ }
+ /* Now merge in the soon-to-be-expired objects... */
+ render_iter = dev_priv->render_ring.active_list.next;
+ bsd_iter = dev_priv->bsd_ring.active_list.next;
+ while ((obj_priv = i915_gem_next_active_object(dev, &render_iter, &bsd_iter)) != NULL) {
+ /* Does the object require an outstanding flush? */
+ if (obj_priv->base.write_domain || obj_priv->pin_count)
continue;
- }
-
- if (HAS_BSD(dev) && !list_empty(&bsd_ring->request_list)) {
- struct drm_i915_gem_request *request;
- request = list_first_entry(&bsd_ring->request_list,
- struct drm_i915_gem_request,
- list);
+ list_add(&obj_priv->evict_list, &unwind_list);
+ if (drm_mm_scan_add_block(obj_priv->gtt_space))
+ goto found;
+ }
- ret = i915_do_wait_request(dev, request->seqno, true, request->ring);
- if (ret)
- return ret;
+ /* Finally add anything with a pending flush (in order of retirement). */
+ list_for_each_entry(obj_priv, &dev_priv->mm.flushing_list, list) {
+ if (obj_priv->pin_count)
+ continue;
+ list_add(&obj_priv->evict_list, &unwind_list);
+ if (drm_mm_scan_add_block(obj_priv->gtt_space))
+ goto found;
+ }
+ render_iter = dev_priv->render_ring.active_list.next;
+ bsd_iter = dev_priv->bsd_ring.active_list.next;
+ while ((obj_priv = i915_gem_next_active_object(dev, &render_iter, &bsd_iter)) != NULL) {
+ if (! obj_priv->base.write_domain || obj_priv->pin_count)
continue;
- }
- /* If we didn't have anything on the request list but there
- * are buffers awaiting a flush, emit one and try again.
- * When we wait on it, those buffers waiting for that flush
- * will get moved to inactive.
- */
- if (!list_empty(&dev_priv->mm.flushing_list)) {
- struct drm_gem_object *obj = NULL;
- struct drm_i915_gem_object *obj_priv;
-
- /* Find an object that we can immediately reuse */
- list_for_each_entry(obj_priv, &dev_priv->mm.flushing_list, list) {
- obj = &obj_priv->base;
- if (obj->size >= min_size)
- break;
-
- obj = NULL;
- }
-
- if (obj != NULL) {
- uint32_t seqno;
-
- i915_gem_flush_ring(dev,
- obj->write_domain,
- obj->write_domain,
- obj_priv->ring);
- seqno = i915_add_request(dev, NULL,
- obj->write_domain,
- obj_priv->ring);
- if (seqno == 0)
- return -ENOMEM;
- continue;
- }
+ list_add(&obj_priv->evict_list, &unwind_list);
+ if (drm_mm_scan_add_block(obj_priv->gtt_space))
+ goto found;
+ }
+
+ /* Nothing found, clean up and bail out! */
+ list_for_each_entry(obj_priv, &unwind_list, evict_list) {
+ ret = drm_mm_scan_remove_block(obj_priv->gtt_space);
+ BUG_ON(ret);
+ }
+
+ return i915_gem_evict_everything(dev);
+
+found:
+ INIT_LIST_HEAD(&eviction_list);
+ list_for_each_entry_safe(obj_priv, tmp_obj_priv, &unwind_list, evict_list) {
+ if (drm_mm_scan_remove_block(obj_priv->gtt_space)) {
+ /* drm_mm doesn't allow any other other operations while
+ * scanning, therefore store to be evicted objects on a
+ * temporary list. */
+ list_move(&obj_priv->evict_list, &eviction_list);
}
+ }
- /* If we didn't do any of the above, there's no single buffer
- * large enough to swap out for the new one, so just evict
- * everything and start again. (This should be rare.)
- */
- if (!list_empty(&dev_priv->mm.inactive_list))
- return i915_gem_evict_inactive(dev);
- else
- return i915_gem_evict_everything(dev);
+ /* Unbinding will emit any required flushes */
+ list_for_each_entry_safe(obj_priv, tmp_obj_priv,
+ &eviction_list, evict_list) {
+#if WATCH_LRU
+ DRM_INFO("%s: evicting %p\n", __func__, obj);
+#endif
+ ret = i915_gem_object_unbind(&obj_priv->base);
+ if (ret)
+ return ret;
}
+
+ /* The just created free hole should be on the top of the free stack
+ * maintained by drm_mm, so this BUG_ON actually executes in O(1).
+ * Furthermore all accessed data has just recently been used, so it
+ * should be really fast, too. */
+ BUG_ON(!drm_mm_search_free(&dev_priv->mm.gtt_space, min_size,
+ alignment, 0));
+
+ return 0;
}
int
--
1.7.1
More information about the Intel-gfx
mailing list