From: Matthew Brost <matthew.br...@intel.com>
Introduce a fences argument to xe_vm_range_tilemask_tlb_invalidation,
allowing callers to provide fences and defer waiting to a later point.
Signed-off-by: Matthew Brost <matthew.br...@intel.com>
---
drivers/gpu/drm/xe/xe_svm.c | 3 ++-
drivers/gpu/drm/xe/xe_vm.c | 26 +++++++++++++++++---------
drivers/gpu/drm/xe/xe_vm.h | 6 ++++--
3 files changed, 23 insertions(+), 12 deletions(-)
diff --git a/drivers/gpu/drm/xe/xe_svm.c b/drivers/gpu/drm/xe/xe_svm.c
index 23c5b363261c..82a598c8d56e 100644
--- a/drivers/gpu/drm/xe/xe_svm.c
+++ b/drivers/gpu/drm/xe/xe_svm.c
@@ -226,7 +226,8 @@ static void xe_svm_invalidate_twopass(struct drm_gpusvm
*gpusvm,
xe_device_wmb(xe);
- err = xe_vm_range_tilemask_tlb_invalidation(vm, adj_start, adj_end,
tile_mask);
+ err = xe_vm_range_tilemask_tlb_invalidation(vm, NULL, adj_start,
+ adj_end, tile_mask);
WARN_ON_ONCE(err);
range_notifier_event_end:
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 148a2425006f..52242fac6969 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -3846,6 +3846,7 @@ void xe_vm_unlock(struct xe_vm *vm)
* xe_vm_range_tilemask_tlb_invalidation - Issue a TLB invalidation on this
tilemask for an
* address range
* @vm: The VM
+ * @fences: Caller provided fences, caller owns waiting if non-NULL
* @start: start address
* @end: end address
* @tile_mask: mask for which gt's issue tlb invalidation
@@ -3854,10 +3855,12 @@ void xe_vm_unlock(struct xe_vm *vm)
*
* Returns 0 for success, negative error code otherwise.
*/
-int xe_vm_range_tilemask_tlb_invalidation(struct xe_vm *vm, u64 start,
- u64 end, u8 tile_mask)
+int xe_vm_range_tilemask_tlb_invalidation(struct xe_vm *vm,
+ struct xe_gt_tlb_invalidation_fence
*fences,
+ u64 start, u64 end, u8 tile_mask)
{
struct xe_gt_tlb_invalidation_fence fence[XE_MAX_TILES_PER_DEVICE *
XE_MAX_GT_PER_TILE];
+ struct xe_gt_tlb_invalidation_fence *__fence = fences ?: fence;
struct xe_tile *tile;
u32 fence_id = 0;
u8 id;
@@ -3869,37 +3872,41 @@ int xe_vm_range_tilemask_tlb_invalidation(struct xe_vm
*vm, u64 start,
for_each_tile(tile, vm->xe, id) {
if (tile_mask & BIT(id)) {
xe_gt_tlb_invalidation_fence_init(tile->primary_gt,
- &fence[fence_id],
true);
+ __fence, true);
err = xe_gt_tlb_invalidation_range(tile->primary_gt,
- &fence[fence_id],
+ __fence,
start,
end,
vm->usm.asid);
if (err)
goto wait;
++fence_id;
+ ++__fence;
if (!tile->media_gt)
continue;
xe_gt_tlb_invalidation_fence_init(tile->media_gt,
- &fence[fence_id],
true);
+ __fence, true);
err = xe_gt_tlb_invalidation_range(tile->media_gt,
- &fence[fence_id],
+ __fence,
start,
end,
vm->usm.asid);
if (err)
goto wait;
++fence_id;
+ ++__fence;
}
}
wait:
- for (id = 0; id < fence_id; ++id)
- xe_gt_tlb_invalidation_fence_wait(&fence[id]);
+ if (!fences) {
+ for (id = 0; id < fence_id; ++id)
+ xe_gt_tlb_invalidation_fence_wait(&fence[id]);
+ }
return err;
}
@@ -3958,7 +3965,8 @@ int xe_vm_invalidate_vma(struct xe_vma *vma)
xe_device_wmb(xe);
- ret = xe_vm_range_tilemask_tlb_invalidation(xe_vma_vm(vma),
xe_vma_start(vma),
+ ret = xe_vm_range_tilemask_tlb_invalidation(xe_vma_vm(vma), NULL,
+ xe_vma_start(vma),
xe_vma_end(vma), tile_mask);
/* WRITE_ONCE pairs with READ_ONCE in xe_vm_has_valid_gpu_mapping() */
diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
index 3475a118f666..d1c3c9aa8d03 100644
--- a/drivers/gpu/drm/xe/xe_vm.h
+++ b/drivers/gpu/drm/xe/xe_vm.h
@@ -22,6 +22,7 @@ struct dma_fence;
struct xe_exec_queue;
struct xe_file;
+struct xe_gt_tlb_invalidation_fence;
struct xe_sync_entry;
struct xe_svm_range;
struct drm_exec;
@@ -228,8 +229,9 @@ struct dma_fence *xe_vm_range_rebind(struct xe_vm *vm,
struct dma_fence *xe_vm_range_unbind(struct xe_vm *vm,
struct xe_svm_range *range);
-int xe_vm_range_tilemask_tlb_invalidation(struct xe_vm *vm, u64 start,
- u64 end, u8 tile_mask);
+int xe_vm_range_tilemask_tlb_invalidation(struct xe_vm *vm,
+ struct xe_gt_tlb_invalidation_fence
*fences,
+ u64 start, u64 end, u8 tile_mask);
int xe_vm_invalidate_vma(struct xe_vma *vma);
--
2.50.1
