[Intel-gfx] [PATCH 01/22] drm/i915: Combine loops within i915_gem_evict_something
Chris Wilson
chris at chris-wilson.co.uk
Wed Jul 27 11:14:39 UTC 2016
Slight micro-optimise to produce combine loops so that gcc is able to
optimise the inner-loops concisely. Since we are reviewing the loops, we
can update the comments to describe the current state of affairs, in
particular the distinction between evicting from the global GTT (which
may contain untracked items and transient global pins) and the
per-process GTT.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
drivers/gpu/drm/i915/i915_gem_evict.c | 143 +++++++++++++++++-----------------
1 file changed, 70 insertions(+), 73 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 3437ced76cb6..016be7316676 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -34,6 +34,19 @@
#include "i915_trace.h"
static bool
+gpu_is_idle(struct drm_i915_private *dev_priv)
+{
+ struct intel_engine_cs *engine;
+
+ for_each_engine(engine, dev_priv) {
+ if (!list_empty(&engine->request_list))
+ return false;
+ }
+
+ return true;
+}
+
+static bool
mark_free(struct i915_vma *vma, struct list_head *unwind)
{
if (vma->pin_count)
@@ -76,37 +89,31 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,
unsigned long start, unsigned long end,
unsigned flags)
{
- struct list_head eviction_list, unwind_list;
- struct i915_vma *vma;
- int ret = 0;
- int pass = 0;
+ struct drm_i915_private *dev_priv = to_i915(dev);
+ struct list_head eviction_list;
+ struct list_head *phases[] = {
+ &vm->inactive_list,
+ &vm->active_list,
+ NULL,
+ }, **phase;
+ struct i915_vma *vma, *next;
+ int ret;
trace_i915_gem_evict(dev, min_size, alignment, flags);
/*
* The goal is to evict objects and amalgamate space in LRU order.
* The oldest idle objects reside on the inactive list, which is in
- * retirement order. The next objects to retire are those on the (per
- * ring) active list that do not have an outstanding flush. Once the
- * hardware reports completion (the seqno is updated after the
- * batchbuffer has been finished) the clean buffer objects would
- * be retired to the inactive list. Any dirty objects would be added
- * to the tail of the flushing list. So after processing the clean
- * active objects we need to emit a MI_FLUSH to retire the flushing
- * list, hence the retirement order of the flushing list is in
- * advance of the dirty objects on the active lists.
+ * retirement order. The next objects to retire are those in flight,
+ * on the active list, again in retirement order.
*
* The retirement sequence is thus:
* 1. Inactive objects (already retired)
- * 2. Clean active objects
- * 3. Flushing list
- * 4. Dirty active objects.
+ * 2. Active objects (will stall on unbinding)
*
* On each list, the oldest objects lie at the HEAD with the freshest
* object on the TAIL.
*/
-
- INIT_LIST_HEAD(&unwind_list);
if (start != 0 || end != vm->total) {
drm_mm_init_scan_with_range(&vm->mm, min_size,
alignment, cache_level,
@@ -114,79 +121,71 @@ i915_gem_evict_something(struct drm_device *dev, struct i915_address_space *vm,
} else
drm_mm_init_scan(&vm->mm, min_size, alignment, cache_level);
-search_again:
- /* First see if there is a large enough contiguous idle region... */
- list_for_each_entry(vma, &vm->inactive_list, vm_link) {
- if (mark_free(vma, &unwind_list))
- goto found;
- }
-
if (flags & PIN_NONBLOCK)
- goto none;
+ phases[1] = NULL;
- /* Now merge in the soon-to-be-expired objects... */
- list_for_each_entry(vma, &vm->active_list, vm_link) {
- if (mark_free(vma, &unwind_list))
- goto found;
- }
+search_again:
+ INIT_LIST_HEAD(&eviction_list);
+ phase = phases;
+ do {
+ list_for_each_entry(vma, *phase, vm_link)
+ if (mark_free(vma, &eviction_list))
+ goto found;
+ } while (*++phase);
-none:
/* Nothing found, clean up and bail out! */
- while (!list_empty(&unwind_list)) {
- vma = list_first_entry(&unwind_list,
- struct i915_vma,
- exec_list);
+ list_for_each_entry_safe(vma, next, &eviction_list, exec_list) {
ret = drm_mm_scan_remove_block(&vma->node);
BUG_ON(ret);
- list_del_init(&vma->exec_list);
+ INIT_LIST_HEAD(&vma->exec_list);
}
/* Can we unpin some objects such as idle hw contents,
- * or pending flips?
+ * or pending flips? But since only the GGTT has global entries
+ * such as scanouts, rinbuffers and contexts, we can skip the
+ * purge when inspecting per-process local address spaces.
*/
- if (flags & PIN_NONBLOCK)
+ if (!i915_is_ggtt(vm) || flags & PIN_NONBLOCK)
return -ENOSPC;
- /* Only idle the GPU and repeat the search once */
- if (pass++ == 0) {
- struct drm_i915_private *dev_priv = to_i915(dev);
-
- if (i915_is_ggtt(vm)) {
- ret = i915_gem_switch_to_kernel_context(dev_priv);
- if (ret)
- return ret;
- }
-
- ret = i915_gem_wait_for_idle(dev_priv);
- if (ret)
- return ret;
-
- i915_gem_retire_requests(dev_priv);
- goto search_again;
+ if (gpu_is_idle(dev_priv)) {
+ /* If we still have pending pageflip completions, drop
+ * back to userspace to give our workqueues time to
+ * acquire our locks and unpin the old scanouts.
+ */
+ return intel_has_pending_fb_unpin(dev) ? -EAGAIN : -ENOSPC;
}
- /* If we still have pending pageflip completions, drop
- * back to userspace to give our workqueues time to
- * acquire our locks and unpin the old scanouts.
+ /* Not everything in the GGTT is tracked via vma (otherwise we
+ * could evict as required with minimal stalling) so we are forced
+ * to idle the GPU and explicitly retire outstanding requests in
+ * the hopes that we can then remove contexts and the like only
+ * bound by their active reference.
*/
- return intel_has_pending_fb_unpin(dev) ? -EAGAIN : -ENOSPC;
+ ret = i915_gem_switch_to_kernel_context(dev_priv);
+ if (ret)
+ return ret;
+
+ ret = i915_gem_wait_for_idle(dev_priv);
+ if (ret)
+ return ret;
+
+ i915_gem_retire_requests(dev_priv);
+ goto search_again;
found:
/* drm_mm doesn't allow any other other operations while
- * scanning, therefore store to be evicted objects on a
- * temporary list. */
- INIT_LIST_HEAD(&eviction_list);
- while (!list_empty(&unwind_list)) {
- vma = list_first_entry(&unwind_list,
- struct i915_vma,
- exec_list);
- if (drm_mm_scan_remove_block(&vma->node)) {
+ * scanning, therefore store to-be-evicted objects on a
+ * temporary list and take a reference for all before
+ * calling unbind (which may remove the active reference
+ * of any of our objects, thus corrupting the list).
+ */
+ list_for_each_entry_safe(vma, next, &eviction_list, exec_list) {
+ if (drm_mm_scan_remove_block(&vma->node))
vma->pin_count++;
- list_move(&vma->exec_list, &eviction_list);
- continue;
- }
- list_del_init(&vma->exec_list);
+ else
+ list_del_init(&vma->exec_list);
}
/* Unbinding will emit any required flushes */
@@ -200,7 +199,6 @@ found:
if (ret == 0)
ret = i915_vma_unbind(vma);
}
-
return ret;
}
@@ -279,7 +277,6 @@ int i915_gem_evict_vm(struct i915_address_space *vm, bool do_idle)
return ret;
i915_gem_retire_requests(dev_priv);
-
WARN_ON(!list_empty(&vm->active_list));
}
--
2.8.1
More information about the Intel-gfx
mailing list