When under severe stress for GTT mappable space, the LRU eviction model
falls off a cliff. We spend all our time scanning the much larger
non-mappable area searching for something within the mappable zone we can
evict. Turn this on its head by only using the full vma for the object if
it is already pinned in the mappable zone or there is sufficient *free*
space to accommodate it (prioritizing speedy reuse). If there is not,
immediately fall back to using small chunks (tilerow for GTT mmap, single
pages for pwrite/relocation) and using random eviction before doing a full
search.
Testcase: igt/gem_concurrent_blt
References: https://bugs.freedesktop.org/show_bug.cgi?id=110848
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190821123234.19194-1-chris@chris-wilson.co.uk
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0,
PIN_MAPPABLE |
- PIN_NONBLOCK |
- PIN_NONFAULT);
+ PIN_NONBLOCK /* NOWARN */ |
+ PIN_NOEVICT);
if (IS_ERR(vma)) {
memset(&cache->node, 0, sizeof(cache->node));
err = drm_mm_insert_node_in_range
/* Now pin it into the GTT as needed */
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0,
PIN_MAPPABLE |
- PIN_NONBLOCK |
- PIN_NONFAULT);
+ PIN_NONBLOCK /* NOWARN */ |
+ PIN_NOSEARCH);
if (IS_ERR(vma)) {
/* Use a partial view if it is bigger than available space */
struct i915_ggtt_view view =
compute_partial_view(obj, page_offset, MIN_CHUNK_PAGES);
unsigned int flags;
- flags = PIN_MAPPABLE;
+ flags = PIN_MAPPABLE | PIN_NOSEARCH;
if (view.type == I915_GGTT_VIEW_NORMAL)
flags |= PIN_NONBLOCK; /* avoid warnings for pinned */
*/
vma = i915_gem_object_ggtt_pin(obj, &view, 0, 0, flags);
- if (IS_ERR(vma) && !view.type) {
+ if (IS_ERR(vma)) {
flags = PIN_MAPPABLE;
view.type = I915_GGTT_VIEW_PARTIAL;
vma = i915_gem_object_ggtt_pin(obj, &view, 0, 0, flags);
wakeref = intel_runtime_pm_get(&i915->runtime_pm);
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0,
PIN_MAPPABLE |
- PIN_NONFAULT |
- PIN_NONBLOCK);
+ PIN_NONBLOCK /* NOWARN */ |
+ PIN_NOEVICT);
if (!IS_ERR(vma)) {
node.start = i915_ggtt_offset(vma);
node.allocated = false;
vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0,
PIN_MAPPABLE |
- PIN_NONFAULT |
- PIN_NONBLOCK);
+ PIN_NONBLOCK /* NOWARN */ |
+ PIN_NOEVICT);
if (!IS_ERR(vma)) {
node.start = i915_ggtt_offset(vma);
node.allocated = false;
if (i915_vma_is_pinned(vma))
return false;
- if (flags & PIN_NONFAULT && i915_vma_has_userfault(vma))
- return false;
-
list_add(&vma->evict_link, unwind);
return drm_mm_scan_add_block(scan, &vma->node);
}
break;
}
- if (flags & PIN_NONFAULT && i915_vma_has_userfault(vma)) {
- ret = -ENOSPC;
- break;
- }
-
/* Overlap of objects in the same batch? */
if (i915_vma_is_pinned(vma)) {
ret = -ENOSPC;
if (flags & PIN_NOEVICT)
return -ENOSPC;
- /* No free space, pick a slot at random.
+ /*
+ * No free space, pick a slot at random.
*
* There is a pathological case here using a GTT shared between
* mmap and GPU (i.e. ggtt/aliasing_ppgtt but not full-ppgtt):
if (err != -ENOSPC)
return err;
+ if (flags & PIN_NOSEARCH)
+ return -ENOSPC;
+
/* Randomly selected placement is pinned, do a search */
err = i915_gem_evict_something(vm, size, alignment, color,
start, end, flags);
u64 start, u64 end, unsigned int flags);
/* Flags used by pin/bind&friends. */
-#define PIN_NONBLOCK BIT_ULL(0)
-#define PIN_NONFAULT BIT_ULL(1)
-#define PIN_NOEVICT BIT_ULL(2)
+#define PIN_NOEVICT BIT_ULL(0)
+#define PIN_NOSEARCH BIT_ULL(1)
+#define PIN_NONBLOCK BIT_ULL(2)
#define PIN_MAPPABLE BIT_ULL(3)
#define PIN_ZONE_4G BIT_ULL(4)
#define PIN_HIGH BIT_ULL(5)