From: Christian K?nig <christian.koe...@amd.com>
This allows us to run different engines concurrently and
still have the fences sync to all operations as currently
required by TTM.
Signed-off-by: Christian K?nig <christian.koenig at amd.com>
---
drivers/gpu/drm/radeon/radeon.h | 3 ++-
drivers/gpu/drm/radeon/radeon_cs.c | 18 ++++++++++++------
drivers/gpu/drm/radeon/radeon_ib.c | 30 ++++++++++++++++++++++--------
drivers/gpu/drm/radeon/radeon_vm.c | 8 ++++----
4 files changed, 40 insertions(+), 19 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index e715e0c..4579361 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -800,7 +800,8 @@ struct radeon_ib {
struct radeon_fence *fence;
struct radeon_vm *vm;
bool is_const_ib;
- struct radeon_semaphore *semaphore;
+ struct radeon_semaphore *presync;
+ struct radeon_semaphore *postsync;
};
struct radeon_ring {
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c
b/drivers/gpu/drm/radeon/radeon_cs.c
index ee712c1..2be4fc5 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -228,11 +228,15 @@ static void radeon_cs_sync_rings(struct radeon_cs_parser
*p)
int i;
for (i = 0; i < p->nrelocs; i++) {
- if (!p->relocs[i].robj)
+ struct radeon_cs_reloc *reloc = &p->relocs[i];
+ struct radeon_bo *bo = reloc->robj;
+ struct radeon_fence *fence;
+
+ if (!bo)
continue;
- radeon_semaphore_sync_to(p->ib.semaphore,
- p->relocs[i].robj->tbo.sync_obj);
+ fence = bo->tbo.sync_obj;
+ radeon_semaphore_sync_to(p->ib.presync, fence);
}
}
@@ -252,9 +256,11 @@ int radeon_cs_parser_init(struct radeon_cs_parser *p, void
*data)
INIT_LIST_HEAD(&p->validated);
p->idx = 0;
p->ib.sa_bo = NULL;
- p->ib.semaphore = NULL;
+ p->ib.presync = NULL;
+ p->ib.postsync = NULL;
p->const_ib.sa_bo = NULL;
- p->const_ib.semaphore = NULL;
+ p->const_ib.presync = NULL;
+ p->const_ib.postsync = NULL;
p->chunk_ib_idx = -1;
p->chunk_relocs_idx = -1;
p->chunk_flags_idx = -1;
@@ -537,7 +543,7 @@ static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
goto out;
}
radeon_cs_sync_rings(parser);
- radeon_semaphore_sync_to(parser->ib.semaphore, vm->fence);
+ radeon_semaphore_sync_to(parser->ib.presync, vm->fence);
if ((rdev->family >= CHIP_TAHITI) &&
(parser->chunk_const_ib_idx != -1)) {
diff --git a/drivers/gpu/drm/radeon/radeon_ib.c
b/drivers/gpu/drm/radeon/radeon_ib.c
index 65b0c21..be09d70 100644
--- a/drivers/gpu/drm/radeon/radeon_ib.c
+++ b/drivers/gpu/drm/radeon/radeon_ib.c
@@ -64,10 +64,13 @@ int radeon_ib_get(struct radeon_device *rdev, int ring,
return r;
}
- r = radeon_semaphore_create(rdev, &ib->semaphore);
- if (r) {
+ r = radeon_semaphore_create(rdev, &ib->presync);
+ if (r)
+ return r;
+
+ r = radeon_semaphore_create(rdev, &ib->postsync);
+ if (r)
return r;
- }
ib->ring = ring;
ib->fence = NULL;
@@ -96,7 +99,8 @@ int radeon_ib_get(struct radeon_device *rdev, int ring,
*/
void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib)
{
- radeon_semaphore_free(rdev, &ib->semaphore, ib->fence);
+ radeon_semaphore_free(rdev, &ib->presync, ib->fence);
+ radeon_semaphore_free(rdev, &ib->postsync, ib->fence);
radeon_sa_bo_free(rdev, &ib->sa_bo, ib->fence);
radeon_fence_unref(&ib->fence);
}
@@ -144,11 +148,11 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct
radeon_ib *ib,
if (ib->vm) {
struct radeon_fence *vm_id_fence;
vm_id_fence = radeon_vm_grab_id(rdev, ib->vm, ib->ring);
- radeon_semaphore_sync_to(ib->semaphore, vm_id_fence);
+ radeon_semaphore_sync_to(ib->presync, vm_id_fence);
}
- /* sync with other rings */
- r = radeon_semaphore_sync_rings(rdev, ib->semaphore, ib->ring);
+ /* sync with other rings before IB execution */
+ r = radeon_semaphore_sync_rings(rdev, ib->presync, ib->ring);
if (r) {
dev_err(rdev->dev, "failed to sync rings (%d)\n", r);
radeon_ring_unlock_undo(rdev, ring);
@@ -160,9 +164,19 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct
radeon_ib *ib,
if (const_ib) {
radeon_ring_ib_execute(rdev, const_ib->ring, const_ib);
- radeon_semaphore_free(rdev, &const_ib->semaphore, NULL);
+ radeon_semaphore_free(rdev, &const_ib->presync, NULL);
+ radeon_semaphore_free(rdev, &const_ib->postsync, NULL);
}
radeon_ring_ib_execute(rdev, ib->ring, ib);
+
+ /* sync with other rings after IB execution */
+ r = radeon_semaphore_sync_rings(rdev, ib->postsync, ib->ring);
+ if (r) {
+ dev_err(rdev->dev, "failed to sync rings (%d)\n", r);
+ radeon_ring_unlock_undo(rdev, ring);
+ return r;
+ }
+
r = radeon_fence_emit(rdev, &ib->fence, ib->ring);
if (r) {
dev_err(rdev->dev, "failed to emit fence for new IB (%d)\n", r);
diff --git a/drivers/gpu/drm/radeon/radeon_vm.c
b/drivers/gpu/drm/radeon/radeon_vm.c
index b2c336a..61f2e6a 100644
--- a/drivers/gpu/drm/radeon/radeon_vm.c
+++ b/drivers/gpu/drm/radeon/radeon_vm.c
@@ -696,8 +696,8 @@ int radeon_vm_update_page_directory(struct radeon_device
*rdev,
if (ib.length_dw != 0) {
radeon_asic_vm_pad_ib(rdev, &ib);
- radeon_semaphore_sync_to(ib.semaphore, pd->tbo.sync_obj);
- radeon_semaphore_sync_to(ib.semaphore, vm->last_id_use);
+ radeon_semaphore_sync_to(ib.presync, pd->tbo.sync_obj);
+ radeon_semaphore_sync_to(ib.presync, vm->last_id_use);
WARN_ON(ib.length_dw > ndw);
r = radeon_ib_schedule(rdev, &ib, NULL);
if (r) {
@@ -823,7 +823,7 @@ static void radeon_vm_update_ptes(struct radeon_device
*rdev,
unsigned nptes;
uint64_t pte;
- radeon_semaphore_sync_to(ib->semaphore, pt->tbo.sync_obj);
+ radeon_semaphore_sync_to(ib->presync, pt->tbo.sync_obj);
if ((addr & ~mask) == (end & ~mask))
nptes = end - addr;
@@ -965,7 +965,7 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
radeon_asic_vm_pad_ib(rdev, &ib);
WARN_ON(ib.length_dw > ndw);
- radeon_semaphore_sync_to(ib.semaphore, vm->fence);
+ radeon_semaphore_sync_to(ib.presync, vm->fence);
r = radeon_ib_schedule(rdev, &ib, NULL);
if (r) {
radeon_ib_free(rdev, &ib);
--
1.9.1
