It is illegal to wait on an another vma while holding the vm->mutex, as
that easily leads to ABBA deadlocks (we wait on a second vma that waits
on us to release the vm->mutex). So while the vm->mutex exists, move the
waiting outside of the lock into the async binding pipeline.
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
.../gpu/drm/i915/gem/i915_gem_execbuffer.c | 21 +++--
drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c | 87 ++++++++++++++++++-
drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h | 5 ++
3 files changed, 103 insertions(+), 10 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 7fb47ff185a3..db5ad6a0df28 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -472,13 +472,19 @@ eb_pin_vma(struct i915_execbuffer *eb,
return false;
if (unlikely(ev->flags & EXEC_OBJECT_NEEDS_FENCE)) {
- if (unlikely(i915_vma_pin_fence(vma))) {
- i915_vma_unpin(vma);
- return false;
- }
+ struct i915_fence_reg *reg = vma->fence;
- if (vma->fence)
+ /* Avoid waiting to change the fence; defer to async worker */
+ if (reg) {
+ if (READ_ONCE(reg->dirty))
+ return false;
+
+ atomic_inc(®->pin_count);
ev->flags |= __EXEC_OBJECT_HAS_FENCE;
+ } else {
+ if (i915_gem_object_is_tiled(vma->obj))
+ return false;
+ }
}
ev->flags |= __EXEC_OBJECT_HAS_PIN;
@@ -955,7 +961,7 @@ static int eb_reserve_vma(struct eb_vm_work *work, struct eb_vma *ev)
pin:
if (unlikely(exec_flags & EXEC_OBJECT_NEEDS_FENCE)) {
- err = __i915_vma_pin_fence(vma); /* XXX no waiting */
+ err = __i915_vma_pin_fence_async(vma, &work->base);
if (unlikely(err))
return err;
@@ -1030,6 +1036,9 @@ static int eb_bind_vma(struct dma_fence_work *base)
GEM_BUG_ON(vma->vm != vm);
+ if (ev->flags & __EXEC_OBJECT_HAS_FENCE)
+ __i915_vma_apply_fence_async(vma);
+
if (!ev->bind_flags)
goto put;
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
index 97659a1249fd..d99d57be3505 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
@@ -21,10 +21,13 @@
* IN THE SOFTWARE.
*/
+#include "i915_active.h"
#include "i915_drv.h"
#include "i915_scatterlist.h"
+#include "i915_sw_fence_work.h"
#include "i915_pvinfo.h"
#include "i915_vgpu.h"
+#include "i915_vma.h"
/**
* DOC: fence register handling
@@ -336,16 +339,14 @@ static struct i915_fence_reg *fence_find(struct i915_ggtt *ggtt)
int __i915_vma_pin_fence(struct i915_vma *vma)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vma->vm);
- struct i915_fence_reg *fence;
+ struct i915_fence_reg *fence = vma->fence;
struct i915_vma *set = i915_gem_object_is_tiled(vma->obj) ? vma : NULL;
int err;
lockdep_assert_held(&vma->vm->mutex);
/* Just update our place in the LRU if our fence is getting reused. */
- if (vma->fence) {
- fence = vma->fence;
- GEM_BUG_ON(fence->vma != vma);
+ if (fence && fence->vma == vma) {
atomic_inc(&fence->pin_count);
if (!fence->dirty) {
list_move_tail(&fence->link, &ggtt->fence_list);
@@ -377,6 +378,84 @@ int __i915_vma_pin_fence(struct i915_vma *vma)
return err;
}
+int __i915_vma_pin_fence_async(struct i915_vma *vma,
+ struct dma_fence_work *work)
+{
+ struct i915_ggtt *ggtt = i915_vm_to_ggtt(vma->vm);
+ struct i915_vma *set = i915_gem_object_is_tiled(vma->obj) ? vma : NULL;
+ struct i915_fence_reg *fence = vma->fence;
+ int err;
+
+ lockdep_assert_held(&vma->vm->mutex);
+
+ /* Just update our place in the LRU if our fence is getting reused. */
+ if (fence && fence->vma == vma) {
+ } else if (set) {
+ fence = fence_find(ggtt);
+ if (IS_ERR(fence))
+ return PTR_ERR(fence);
+
+ GEM_BUG_ON(atomic_read(&fence->pin_count));
+ fence->dirty = true;
+ } else {
+ return 0;
+ }
+
+ atomic_inc(&fence->pin_count);
+ if (!fence->dirty) {
+ list_move_tail(&fence->link, &ggtt->fence_list);
+ return 0;
+ }
+
+ fence->tiling = 0;
+ if (set) {
+ if (INTEL_GEN(fence_to_i915(fence)) < 4) {
+ /* implicit 'unfenced' GPU blits */
+ err = i915_sw_fence_await_active(&work->chain,
+ &vma->active,
+ I915_ACTIVE_AWAIT_ALL);
+ if (err) {
+ atomic_dec(&fence->pin_count);
+ return err;
+ }
+ }
+
+ fence->start = vma->node.start;
+ fence->size = vma->fence_size;
+ fence->stride = i915_gem_object_get_stride(vma->obj);
+ fence->tiling = i915_gem_object_get_tiling(vma->obj);
+ }
+
+ set = xchg(&fence->vma, vma);
+ if (set) {
+ err = i915_sw_fence_await_active(&work->chain,
+ &fence->active,
+ I915_ACTIVE_AWAIT_ALL);
+ if (err) {
+ fence->vma = set;
+ atomic_dec(&fence->pin_count);
+ return err;
+ }
+
+ if (set != vma) {
+ GEM_BUG_ON(set->fence != fence);
+ i915_vma_revoke_mmap(set);
+ set->fence = NULL;
+ }
+ }
+
+ vma->fence = fence;
+ return 0;
+}
+
+void __i915_vma_apply_fence_async(struct i915_vma *vma)
+{
+ struct i915_fence_reg *fence = vma->fence;
+
+ if (fence->dirty)
+ fence_write(fence);
+}
+
/**
* i915_vma_pin_fence - set up fencing for a vma
* @vma: vma to map through a fence reg
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h
index 9eef679e1311..d306ac14d47e 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h
@@ -30,6 +30,7 @@
#include "i915_active.h"
+struct dma_fence_work;
struct drm_i915_gem_object;
struct i915_ggtt;
struct i915_vma;
@@ -70,6 +71,10 @@ void i915_gem_object_do_bit_17_swizzle(struct drm_i915_gem_object *obj,
void i915_gem_object_save_bit_17_swizzle(struct drm_i915_gem_object *obj,
struct sg_table *pages);
+int __i915_vma_pin_fence_async(struct i915_vma *vma,
+ struct dma_fence_work *work);
+void __i915_vma_apply_fence_async(struct i915_vma *vma);
+
void intel_ggtt_init_fences(struct i915_ggtt *ggtt);
void intel_ggtt_fini_fences(struct i915_ggtt *ggtt);
--
2.20.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
