[PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization v2
v2: BUG_ON not matching rings.
Signed-off-by: Christian K?nig
Reviewed-by: Jerome Glisse
---
drivers/gpu/drm/radeon/radeon.h | 25 ++-
drivers/gpu/drm/radeon/radeon_fence.c | 73 +
2 files changed, 87 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 5e259b4..ef132eb 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -257,8 +257,8 @@ struct radeon_fence_driver {
uint32_tscratch_reg;
uint64_tgpu_addr;
volatile uint32_t *cpu_addr;
- /* seq is protected by ring emission lock */
- uint64_tseq;
+ /* sync_seq is protected by ring emission lock */
+ uint64_tsync_seq[RADEON_NUM_RINGS];
atomic64_t last_seq;
unsigned long last_activity;
boolinitialized;
@@ -288,6 +288,27 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
void radeon_fence_unref(struct radeon_fence **fence);
unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
+bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
+void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
+static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
+ struct radeon_fence *b)
+{
+ if (!a) {
+ return b;
+ }
+
+ if (!b) {
+ return a;
+ }
+
+ BUG_ON(a->ring != b->ring);
+
+ if (a->seq > b->seq) {
+ return a;
+ } else {
+ return b;
+ }
+}
/*
* Tiling registers
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
b/drivers/gpu/drm/radeon/radeon_fence.c
index 401d346..7b55625 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
}
kref_init(&((*fence)->kref));
(*fence)->rdev = rdev;
- (*fence)->seq = ++rdev->fence_drv[ring].seq;
+ (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
(*fence)->ring = ring;
radeon_fence_ring_emit(rdev, ring, *fence);
trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
@@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
*rdev, int ring)
* wait.
*/
seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
- if (seq >= rdev->fence_drv[ring].seq) {
+ if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
/* nothing to wait for, last_seq is
already the last emited fence */
return -ENOENT;
@@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
*rdev, int ring)
* activity can be scheduled so there won't be concurrent access
* to seq value.
*/
- return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
+ return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].sync_seq[ring],
ring, false, false);
}
@@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
* but it's ok to report slightly wrong fence count here.
*/
radeon_fence_process(rdev, ring);
- emitted = rdev->fence_drv[ring].seq -
atomic64_read(&rdev->fence_drv[ring].last_seq);
+ emitted = rdev->fence_drv[ring].sync_seq[ring]
+ - atomic64_read(&rdev->fence_drv[ring].last_seq);
/* to avoid 32bits warp around */
if (emitted > 0x1000) {
emitted = 0x1000;
@@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
return (unsigned)emitted;
}
+bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *fdrv;
+
+ if (!fence) {
+ return false;
+ }
+
+ if (fence->ring == dst_ring) {
+ return false;
+ }
+
+ /* we are protected by the ring mutex */
+ fdrv = &fence->rdev->fence_drv[dst_ring];
+ if (fence->seq <= fdrv->sync_seq[fence->ring]) {
+ return false;
+ }
+
+ return true;
+}
+
+void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *dst, *src;
+ unsigned i;
+
+ if (!fence) {
+ return;
+ }
+
+ if (fence->ring == dst_ring) {
+ return;
+ }
+
+ /* we are protected by the ring mutex */
+ src = &fence->rdev->fence_drv[fence->ring];
+ dst = &fence->rdev->fence_drv[dst_ring];
+ for (i = 0; i
[PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization v2
v2: BUG_ON not matching rings.
Signed-off-by: Christian König
Reviewed-by: Jerome Glisse
---
drivers/gpu/drm/radeon/radeon.h | 25 ++-
drivers/gpu/drm/radeon/radeon_fence.c | 73 +
2 files changed, 87 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 5e259b4..ef132eb 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -257,8 +257,8 @@ struct radeon_fence_driver {
uint32_tscratch_reg;
uint64_tgpu_addr;
volatile uint32_t *cpu_addr;
- /* seq is protected by ring emission lock */
- uint64_tseq;
+ /* sync_seq is protected by ring emission lock */
+ uint64_tsync_seq[RADEON_NUM_RINGS];
atomic64_t last_seq;
unsigned long last_activity;
boolinitialized;
@@ -288,6 +288,27 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
void radeon_fence_unref(struct radeon_fence **fence);
unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
+bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
+void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
+static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
+ struct radeon_fence *b)
+{
+ if (!a) {
+ return b;
+ }
+
+ if (!b) {
+ return a;
+ }
+
+ BUG_ON(a->ring != b->ring);
+
+ if (a->seq > b->seq) {
+ return a;
+ } else {
+ return b;
+ }
+}
/*
* Tiling registers
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
b/drivers/gpu/drm/radeon/radeon_fence.c
index 401d346..7b55625 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
}
kref_init(&((*fence)->kref));
(*fence)->rdev = rdev;
- (*fence)->seq = ++rdev->fence_drv[ring].seq;
+ (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
(*fence)->ring = ring;
radeon_fence_ring_emit(rdev, ring, *fence);
trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
@@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
*rdev, int ring)
* wait.
*/
seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
- if (seq >= rdev->fence_drv[ring].seq) {
+ if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
/* nothing to wait for, last_seq is
already the last emited fence */
return -ENOENT;
@@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
*rdev, int ring)
* activity can be scheduled so there won't be concurrent access
* to seq value.
*/
- return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
+ return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].sync_seq[ring],
ring, false, false);
}
@@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
* but it's ok to report slightly wrong fence count here.
*/
radeon_fence_process(rdev, ring);
- emitted = rdev->fence_drv[ring].seq -
atomic64_read(&rdev->fence_drv[ring].last_seq);
+ emitted = rdev->fence_drv[ring].sync_seq[ring]
+ - atomic64_read(&rdev->fence_drv[ring].last_seq);
/* to avoid 32bits warp around */
if (emitted > 0x1000) {
emitted = 0x1000;
@@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
return (unsigned)emitted;
}
+bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *fdrv;
+
+ if (!fence) {
+ return false;
+ }
+
+ if (fence->ring == dst_ring) {
+ return false;
+ }
+
+ /* we are protected by the ring mutex */
+ fdrv = &fence->rdev->fence_drv[dst_ring];
+ if (fence->seq <= fdrv->sync_seq[fence->ring]) {
+ return false;
+ }
+
+ return true;
+}
+
+void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *dst, *src;
+ unsigned i;
+
+ if (!fence) {
+ return;
+ }
+
+ if (fence->ring == dst_ring) {
+ return;
+ }
+
+ /* we are protected by the ring mutex */
+ src = &fence->rdev->fence_drv[fence->ring];
+ dst = &fence->rdev->fence_drv[dst_ring];
+ for (i = 0
[PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization
Signed-off-by: Christian K?nig
---
drivers/gpu/drm/radeon/radeon.h | 23 ++-
drivers/gpu/drm/radeon/radeon_fence.c | 73 +
2 files changed, 85 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 5e259b4..4e232c3 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -257,8 +257,8 @@ struct radeon_fence_driver {
uint32_tscratch_reg;
uint64_tgpu_addr;
volatile uint32_t *cpu_addr;
- /* seq is protected by ring emission lock */
- uint64_tseq;
+ /* sync_seq is protected by ring emission lock */
+ uint64_tsync_seq[RADEON_NUM_RINGS];
atomic64_t last_seq;
unsigned long last_activity;
boolinitialized;
@@ -288,6 +288,25 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
void radeon_fence_unref(struct radeon_fence **fence);
unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
+bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
+void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
+static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
+ struct radeon_fence *b)
+{
+ if (!a) {
+ return b;
+ }
+
+ if (!b) {
+ return a;
+ }
+
+ if (a->seq > b->seq) {
+ return a;
+ } else {
+ return b;
+ }
+}
/*
* Tiling registers
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
b/drivers/gpu/drm/radeon/radeon_fence.c
index 401d346..7b55625 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
}
kref_init(&((*fence)->kref));
(*fence)->rdev = rdev;
- (*fence)->seq = ++rdev->fence_drv[ring].seq;
+ (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
(*fence)->ring = ring;
radeon_fence_ring_emit(rdev, ring, *fence);
trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
@@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
*rdev, int ring)
* wait.
*/
seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
- if (seq >= rdev->fence_drv[ring].seq) {
+ if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
/* nothing to wait for, last_seq is
already the last emited fence */
return -ENOENT;
@@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
*rdev, int ring)
* activity can be scheduled so there won't be concurrent access
* to seq value.
*/
- return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
+ return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].sync_seq[ring],
ring, false, false);
}
@@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
* but it's ok to report slightly wrong fence count here.
*/
radeon_fence_process(rdev, ring);
- emitted = rdev->fence_drv[ring].seq -
atomic64_read(&rdev->fence_drv[ring].last_seq);
+ emitted = rdev->fence_drv[ring].sync_seq[ring]
+ - atomic64_read(&rdev->fence_drv[ring].last_seq);
/* to avoid 32bits warp around */
if (emitted > 0x1000) {
emitted = 0x1000;
@@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
return (unsigned)emitted;
}
+bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *fdrv;
+
+ if (!fence) {
+ return false;
+ }
+
+ if (fence->ring == dst_ring) {
+ return false;
+ }
+
+ /* we are protected by the ring mutex */
+ fdrv = &fence->rdev->fence_drv[dst_ring];
+ if (fence->seq <= fdrv->sync_seq[fence->ring]) {
+ return false;
+ }
+
+ return true;
+}
+
+void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *dst, *src;
+ unsigned i;
+
+ if (!fence) {
+ return;
+ }
+
+ if (fence->ring == dst_ring) {
+ return;
+ }
+
+ /* we are protected by the ring mutex */
+ src = &fence->rdev->fence_drv[fence->ring];
+ dst = &fence->rdev->fence_drv[dst_ring];
+ for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+ if (i == dst_ring) {
+ continue;
[PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization
On Thu, May 31, 2012 at 4:15 PM, Christian K?nig
wrote:
> Signed-off-by: Christian K?nig
> ---
> ?drivers/gpu/drm/radeon/radeon.h ? ? ? | ? 23 ++-
> ?drivers/gpu/drm/radeon/radeon_fence.c | ? 73
> +
> ?2 files changed, 85 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
> index 5e259b4..4e232c3 100644
> --- a/drivers/gpu/drm/radeon/radeon.h
> +++ b/drivers/gpu/drm/radeon/radeon.h
> @@ -257,8 +257,8 @@ struct radeon_fence_driver {
> ? ? ? ?uint32_t ? ? ? ? ? ? ? ? ? ? ? ?scratch_reg;
> ? ? ? ?uint64_t ? ? ? ? ? ? ? ? ? ? ? ?gpu_addr;
> ? ? ? ?volatile uint32_t ? ? ? ? ? ? ? *cpu_addr;
> - ? ? ? /* seq is protected by ring emission lock */
> - ? ? ? uint64_t ? ? ? ? ? ? ? ? ? ? ? ?seq;
> + ? ? ? /* sync_seq is protected by ring emission lock */
> + ? ? ? uint64_t ? ? ? ? ? ? ? ? ? ? ? ?sync_seq[RADEON_NUM_RINGS];
> ? ? ? ?atomic64_t ? ? ? ? ? ? ? ? ? ? ?last_seq;
> ? ? ? ?unsigned long ? ? ? ? ? ? ? ? ? last_activity;
> ? ? ? ?bool ? ? ? ? ? ? ? ? ? ? ? ? ? ?initialized;
> @@ -288,6 +288,25 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
> ?struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
> ?void radeon_fence_unref(struct radeon_fence **fence);
> ?unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
> +bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
> +void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
> +static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? struct radeon_fence *b)
> +{
> + ? ? ? if (!a) {
> + ? ? ? ? ? ? ? return b;
> + ? ? ? }
> +
> + ? ? ? if (!b) {
> + ? ? ? ? ? ? ? return a;
> + ? ? ? }
Please add :
BUG_ON(a->ring != b->ring);
So we can catch if someone badly use this function.
> +
> + ? ? ? if (a->seq > b->seq) {
> + ? ? ? ? ? ? ? return a;
> + ? ? ? } else {
> + ? ? ? ? ? ? ? return b;
> + ? ? ? }
> +}
>
> ?/*
> ?* Tiling registers
> diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
> b/drivers/gpu/drm/radeon/radeon_fence.c
> index 401d346..7b55625 100644
> --- a/drivers/gpu/drm/radeon/radeon_fence.c
> +++ b/drivers/gpu/drm/radeon/radeon_fence.c
> @@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
> ? ? ? ?}
> ? ? ? ?kref_init(&((*fence)->kref));
> ? ? ? ?(*fence)->rdev = rdev;
> - ? ? ? (*fence)->seq = ++rdev->fence_drv[ring].seq;
> + ? ? ? (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
> ? ? ? ?(*fence)->ring = ring;
> ? ? ? ?radeon_fence_ring_emit(rdev, ring, *fence);
> ? ? ? ?trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
> @@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
> *rdev, int ring)
> ? ? ? ? * wait.
> ? ? ? ? */
> ? ? ? ?seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
> - ? ? ? if (seq >= rdev->fence_drv[ring].seq) {
> + ? ? ? if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
> ? ? ? ? ? ? ? ?/* nothing to wait for, last_seq is
> ? ? ? ? ? ? ? ? ? already the last emited fence */
> ? ? ? ? ? ? ? ?return -ENOENT;
> @@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
> *rdev, int ring)
> ? ? ? ? * activity can be scheduled so there won't be concurrent access
> ? ? ? ? * to seq value.
> ? ? ? ? */
> - ? ? ? return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
> + ? ? ? return radeon_fence_wait_seq(rdev,
> rdev->fence_drv[ring].sync_seq[ring],
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ring, false, false);
> ?}
>
> @@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
> *rdev, int ring)
> ? ? ? ? * but it's ok to report slightly wrong fence count here.
> ? ? ? ? */
> ? ? ? ?radeon_fence_process(rdev, ring);
> - ? ? ? emitted = rdev->fence_drv[ring].seq -
> atomic64_read(&rdev->fence_drv[ring].last_seq);
> + ? ? ? emitted = rdev->fence_drv[ring].sync_seq[ring]
> + ? ? ? ? ? ? ? - atomic64_read(&rdev->fence_drv[ring].last_seq);
> ? ? ? ?/* to avoid 32bits warp around */
> ? ? ? ?if (emitted > 0x1000) {
> ? ? ? ? ? ? ? ?emitted = 0x1000;
> @@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
> *rdev, int ring)
> ? ? ? ?return (unsigned)emitted;
> ?}
>
> +bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
> +{
> + ? ? ? struct radeon_fence_driver *fdrv;
> +
> + ? ? ? if (!fence) {
> + ? ? ? ? ? ? ? return false;
> + ? ? ? }
> +
> + ? ? ? if (fence->ring == dst_ring) {
> + ? ? ? ? ? ? ? return false;
> + ? ? ? }
> +
> + ? ? ? /* we are protected by the ring mutex */
> + ? ? ? fdrv = &fence->rdev->fence_drv[dst_ring];
> + ? ? ? if (fence->seq <= fdrv->sync_seq[fence->ring]) {
> + ? ? ? ? ? ? ? return false;
> + ? ? ? }
> +
> + ? ? ? return true;
> +}
> +
> +void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
> +{
> + ? ? ? struct radeon_fence_driver *dst, *src;
> + ? ? ? unsigned i;
> +
> + ? ? ? if (!fence) {
>
Re: [PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization
On Thu, May 31, 2012 at 4:15 PM, Christian König
wrote:
> Signed-off-by: Christian König
> ---
> drivers/gpu/drm/radeon/radeon.h | 23 ++-
> drivers/gpu/drm/radeon/radeon_fence.c | 73
> +
> 2 files changed, 85 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
> index 5e259b4..4e232c3 100644
> --- a/drivers/gpu/drm/radeon/radeon.h
> +++ b/drivers/gpu/drm/radeon/radeon.h
> @@ -257,8 +257,8 @@ struct radeon_fence_driver {
> uint32_t scratch_reg;
> uint64_t gpu_addr;
> volatile uint32_t *cpu_addr;
> - /* seq is protected by ring emission lock */
> - uint64_t seq;
> + /* sync_seq is protected by ring emission lock */
> + uint64_t sync_seq[RADEON_NUM_RINGS];
> atomic64_t last_seq;
> unsigned long last_activity;
> bool initialized;
> @@ -288,6 +288,25 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
> struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
> void radeon_fence_unref(struct radeon_fence **fence);
> unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
> +bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
> +void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
> +static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
> + struct radeon_fence *b)
> +{
> + if (!a) {
> + return b;
> + }
> +
> + if (!b) {
> + return a;
> + }
Please add :
BUG_ON(a->ring != b->ring);
So we can catch if someone badly use this function.
> +
> + if (a->seq > b->seq) {
> + return a;
> + } else {
> + return b;
> + }
> +}
>
> /*
> * Tiling registers
> diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
> b/drivers/gpu/drm/radeon/radeon_fence.c
> index 401d346..7b55625 100644
> --- a/drivers/gpu/drm/radeon/radeon_fence.c
> +++ b/drivers/gpu/drm/radeon/radeon_fence.c
> @@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
> }
> kref_init(&((*fence)->kref));
> (*fence)->rdev = rdev;
> - (*fence)->seq = ++rdev->fence_drv[ring].seq;
> + (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
> (*fence)->ring = ring;
> radeon_fence_ring_emit(rdev, ring, *fence);
> trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
> @@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
> *rdev, int ring)
> * wait.
> */
> seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
> - if (seq >= rdev->fence_drv[ring].seq) {
> + if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
> /* nothing to wait for, last_seq is
> already the last emited fence */
> return -ENOENT;
> @@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
> *rdev, int ring)
> * activity can be scheduled so there won't be concurrent access
> * to seq value.
> */
> - return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
> + return radeon_fence_wait_seq(rdev,
> rdev->fence_drv[ring].sync_seq[ring],
> ring, false, false);
> }
>
> @@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
> *rdev, int ring)
> * but it's ok to report slightly wrong fence count here.
> */
> radeon_fence_process(rdev, ring);
> - emitted = rdev->fence_drv[ring].seq -
> atomic64_read(&rdev->fence_drv[ring].last_seq);
> + emitted = rdev->fence_drv[ring].sync_seq[ring]
> + - atomic64_read(&rdev->fence_drv[ring].last_seq);
> /* to avoid 32bits warp around */
> if (emitted > 0x1000) {
> emitted = 0x1000;
> @@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
> *rdev, int ring)
> return (unsigned)emitted;
> }
>
> +bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
> +{
> + struct radeon_fence_driver *fdrv;
> +
> + if (!fence) {
> + return false;
> + }
> +
> + if (fence->ring == dst_ring) {
> + return false;
> + }
> +
> + /* we are protected by the ring mutex */
> + fdrv = &fence->rdev->fence_drv[dst_ring];
> + if (fence->seq <= fdrv->sync_seq[fence->ring]) {
> + return false;
> + }
> +
> + return true;
> +}
> +
> +void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
> +{
> + struct radeon_fence_driver *dst, *src;
> + unsigned i;
> +
> + if (!fence) {
>
[PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization
Signed-off-by: Christian König
---
drivers/gpu/drm/radeon/radeon.h | 23 ++-
drivers/gpu/drm/radeon/radeon_fence.c | 73 +
2 files changed, 85 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 5e259b4..4e232c3 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -257,8 +257,8 @@ struct radeon_fence_driver {
uint32_tscratch_reg;
uint64_tgpu_addr;
volatile uint32_t *cpu_addr;
- /* seq is protected by ring emission lock */
- uint64_tseq;
+ /* sync_seq is protected by ring emission lock */
+ uint64_tsync_seq[RADEON_NUM_RINGS];
atomic64_t last_seq;
unsigned long last_activity;
boolinitialized;
@@ -288,6 +288,25 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
void radeon_fence_unref(struct radeon_fence **fence);
unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
+bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
+void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
+static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
+ struct radeon_fence *b)
+{
+ if (!a) {
+ return b;
+ }
+
+ if (!b) {
+ return a;
+ }
+
+ if (a->seq > b->seq) {
+ return a;
+ } else {
+ return b;
+ }
+}
/*
* Tiling registers
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
b/drivers/gpu/drm/radeon/radeon_fence.c
index 401d346..7b55625 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
}
kref_init(&((*fence)->kref));
(*fence)->rdev = rdev;
- (*fence)->seq = ++rdev->fence_drv[ring].seq;
+ (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
(*fence)->ring = ring;
radeon_fence_ring_emit(rdev, ring, *fence);
trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
@@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
*rdev, int ring)
* wait.
*/
seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
- if (seq >= rdev->fence_drv[ring].seq) {
+ if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
/* nothing to wait for, last_seq is
already the last emited fence */
return -ENOENT;
@@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
*rdev, int ring)
* activity can be scheduled so there won't be concurrent access
* to seq value.
*/
- return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
+ return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].sync_seq[ring],
ring, false, false);
}
@@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
* but it's ok to report slightly wrong fence count here.
*/
radeon_fence_process(rdev, ring);
- emitted = rdev->fence_drv[ring].seq -
atomic64_read(&rdev->fence_drv[ring].last_seq);
+ emitted = rdev->fence_drv[ring].sync_seq[ring]
+ - atomic64_read(&rdev->fence_drv[ring].last_seq);
/* to avoid 32bits warp around */
if (emitted > 0x1000) {
emitted = 0x1000;
@@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
return (unsigned)emitted;
}
+bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *fdrv;
+
+ if (!fence) {
+ return false;
+ }
+
+ if (fence->ring == dst_ring) {
+ return false;
+ }
+
+ /* we are protected by the ring mutex */
+ fdrv = &fence->rdev->fence_drv[dst_ring];
+ if (fence->seq <= fdrv->sync_seq[fence->ring]) {
+ return false;
+ }
+
+ return true;
+}
+
+void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *dst, *src;
+ unsigned i;
+
+ if (!fence) {
+ return;
+ }
+
+ if (fence->ring == dst_ring) {
+ return;
+ }
+
+ /* we are protected by the ring mutex */
+ src = &fence->rdev->fence_drv[fence->ring];
+ dst = &fence->rdev->fence_drv[dst_ring];
+ for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+ if (i == dst_ring) {
+ contin
[PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization
On Thu, May 24, 2012 at 09:49:06AM +0200, Christian K?nig wrote:
> Signed-off-by: Christian K?nig
Need a small improvement see below, otherwise
Reviewed-by: Jerome Glisse
> ---
> drivers/gpu/drm/radeon/radeon.h | 23 ++-
> drivers/gpu/drm/radeon/radeon_fence.c | 73
> +
> 2 files changed, 85 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
> index 5e259b4..4e232c3 100644
> --- a/drivers/gpu/drm/radeon/radeon.h
> +++ b/drivers/gpu/drm/radeon/radeon.h
> @@ -257,8 +257,8 @@ struct radeon_fence_driver {
> uint32_tscratch_reg;
> uint64_tgpu_addr;
> volatile uint32_t *cpu_addr;
> - /* seq is protected by ring emission lock */
> - uint64_tseq;
> + /* sync_seq is protected by ring emission lock */
> + uint64_tsync_seq[RADEON_NUM_RINGS];
> atomic64_t last_seq;
> unsigned long last_activity;
> boolinitialized;
> @@ -288,6 +288,25 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
> struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
> void radeon_fence_unref(struct radeon_fence **fence);
> unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
> +bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
> +void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
> +static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
> + struct radeon_fence *b)
> +{
> + if (!a) {
> + return b;
> + }
> +
> + if (!b) {
> + return a;
> + }
Please add :
BUG_ON(a->ring != b->ring);
So we can catch if someone badly use this function.
> +
> + if (a->seq > b->seq) {
> + return a;
> + } else {
> + return b;
> + }
> +}
>
> /*
> * Tiling registers
> diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
> b/drivers/gpu/drm/radeon/radeon_fence.c
> index 401d346..7b55625 100644
> --- a/drivers/gpu/drm/radeon/radeon_fence.c
> +++ b/drivers/gpu/drm/radeon/radeon_fence.c
> @@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
> }
> kref_init(&((*fence)->kref));
> (*fence)->rdev = rdev;
> - (*fence)->seq = ++rdev->fence_drv[ring].seq;
> + (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
> (*fence)->ring = ring;
> radeon_fence_ring_emit(rdev, ring, *fence);
> trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
> @@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
> *rdev, int ring)
>* wait.
>*/
> seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
> - if (seq >= rdev->fence_drv[ring].seq) {
> + if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
> /* nothing to wait for, last_seq is
> already the last emited fence */
> return -ENOENT;
> @@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
> *rdev, int ring)
>* activity can be scheduled so there won't be concurrent access
>* to seq value.
>*/
> - return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
> + return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].sync_seq[ring],
>ring, false, false);
> }
>
> @@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
> *rdev, int ring)
>* but it's ok to report slightly wrong fence count here.
>*/
> radeon_fence_process(rdev, ring);
> - emitted = rdev->fence_drv[ring].seq -
> atomic64_read(&rdev->fence_drv[ring].last_seq);
> + emitted = rdev->fence_drv[ring].sync_seq[ring]
> + - atomic64_read(&rdev->fence_drv[ring].last_seq);
> /* to avoid 32bits warp around */
> if (emitted > 0x1000) {
> emitted = 0x1000;
> @@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
> *rdev, int ring)
> return (unsigned)emitted;
> }
>
> +bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
> +{
> + struct radeon_fence_driver *fdrv;
> +
> + if (!fence) {
> + return false;
> + }
> +
> + if (fence->ring == dst_ring) {
> + return false;
> + }
> +
> + /* we are protected by the ring mutex */
> + fdrv = &fence->rdev->fence_drv[dst_ring];
> + if (fence->seq <= fdrv->sync_seq[fence->ring]) {
> + return false;
> + }
> +
> + return true;
> +}
> +
> +void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
> +{
> + struct radeon_fence_driver *dst, *src;
> + unsigned i;
> +
> + if (!fence) {
> + return;
>
[PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization
Signed-off-by: Christian K?nig
---
drivers/gpu/drm/radeon/radeon.h | 23 ++-
drivers/gpu/drm/radeon/radeon_fence.c | 73 +
2 files changed, 85 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 5e259b4..4e232c3 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -257,8 +257,8 @@ struct radeon_fence_driver {
uint32_tscratch_reg;
uint64_tgpu_addr;
volatile uint32_t *cpu_addr;
- /* seq is protected by ring emission lock */
- uint64_tseq;
+ /* sync_seq is protected by ring emission lock */
+ uint64_tsync_seq[RADEON_NUM_RINGS];
atomic64_t last_seq;
unsigned long last_activity;
boolinitialized;
@@ -288,6 +288,25 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
void radeon_fence_unref(struct radeon_fence **fence);
unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
+bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
+void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
+static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
+ struct radeon_fence *b)
+{
+ if (!a) {
+ return b;
+ }
+
+ if (!b) {
+ return a;
+ }
+
+ if (a->seq > b->seq) {
+ return a;
+ } else {
+ return b;
+ }
+}
/*
* Tiling registers
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
b/drivers/gpu/drm/radeon/radeon_fence.c
index 401d346..7b55625 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
}
kref_init(&((*fence)->kref));
(*fence)->rdev = rdev;
- (*fence)->seq = ++rdev->fence_drv[ring].seq;
+ (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
(*fence)->ring = ring;
radeon_fence_ring_emit(rdev, ring, *fence);
trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
@@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
*rdev, int ring)
* wait.
*/
seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
- if (seq >= rdev->fence_drv[ring].seq) {
+ if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
/* nothing to wait for, last_seq is
already the last emited fence */
return -ENOENT;
@@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
*rdev, int ring)
* activity can be scheduled so there won't be concurrent access
* to seq value.
*/
- return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
+ return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].sync_seq[ring],
ring, false, false);
}
@@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
* but it's ok to report slightly wrong fence count here.
*/
radeon_fence_process(rdev, ring);
- emitted = rdev->fence_drv[ring].seq -
atomic64_read(&rdev->fence_drv[ring].last_seq);
+ emitted = rdev->fence_drv[ring].sync_seq[ring]
+ - atomic64_read(&rdev->fence_drv[ring].last_seq);
/* to avoid 32bits warp around */
if (emitted > 0x1000) {
emitted = 0x1000;
@@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
return (unsigned)emitted;
}
+bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *fdrv;
+
+ if (!fence) {
+ return false;
+ }
+
+ if (fence->ring == dst_ring) {
+ return false;
+ }
+
+ /* we are protected by the ring mutex */
+ fdrv = &fence->rdev->fence_drv[dst_ring];
+ if (fence->seq <= fdrv->sync_seq[fence->ring]) {
+ return false;
+ }
+
+ return true;
+}
+
+void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *dst, *src;
+ unsigned i;
+
+ if (!fence) {
+ return;
+ }
+
+ if (fence->ring == dst_ring) {
+ return;
+ }
+
+ /* we are protected by the ring mutex */
+ src = &fence->rdev->fence_drv[fence->ring];
+ dst = &fence->rdev->fence_drv[dst_ring];
+ for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+ if (i == dst_ring) {
+ continue;
Re: [PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization
On Thu, May 24, 2012 at 09:49:06AM +0200, Christian König wrote:
> Signed-off-by: Christian König
Need a small improvement see below, otherwise
Reviewed-by: Jerome Glisse
> ---
> drivers/gpu/drm/radeon/radeon.h | 23 ++-
> drivers/gpu/drm/radeon/radeon_fence.c | 73
> +
> 2 files changed, 85 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
> index 5e259b4..4e232c3 100644
> --- a/drivers/gpu/drm/radeon/radeon.h
> +++ b/drivers/gpu/drm/radeon/radeon.h
> @@ -257,8 +257,8 @@ struct radeon_fence_driver {
> uint32_tscratch_reg;
> uint64_tgpu_addr;
> volatile uint32_t *cpu_addr;
> - /* seq is protected by ring emission lock */
> - uint64_tseq;
> + /* sync_seq is protected by ring emission lock */
> + uint64_tsync_seq[RADEON_NUM_RINGS];
> atomic64_t last_seq;
> unsigned long last_activity;
> boolinitialized;
> @@ -288,6 +288,25 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
> struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
> void radeon_fence_unref(struct radeon_fence **fence);
> unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
> +bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
> +void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
> +static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
> + struct radeon_fence *b)
> +{
> + if (!a) {
> + return b;
> + }
> +
> + if (!b) {
> + return a;
> + }
Please add :
BUG_ON(a->ring != b->ring);
So we can catch if someone badly use this function.
> +
> + if (a->seq > b->seq) {
> + return a;
> + } else {
> + return b;
> + }
> +}
>
> /*
> * Tiling registers
> diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
> b/drivers/gpu/drm/radeon/radeon_fence.c
> index 401d346..7b55625 100644
> --- a/drivers/gpu/drm/radeon/radeon_fence.c
> +++ b/drivers/gpu/drm/radeon/radeon_fence.c
> @@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
> }
> kref_init(&((*fence)->kref));
> (*fence)->rdev = rdev;
> - (*fence)->seq = ++rdev->fence_drv[ring].seq;
> + (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
> (*fence)->ring = ring;
> radeon_fence_ring_emit(rdev, ring, *fence);
> trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
> @@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
> *rdev, int ring)
>* wait.
>*/
> seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
> - if (seq >= rdev->fence_drv[ring].seq) {
> + if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
> /* nothing to wait for, last_seq is
> already the last emited fence */
> return -ENOENT;
> @@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
> *rdev, int ring)
>* activity can be scheduled so there won't be concurrent access
>* to seq value.
>*/
> - return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
> + return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].sync_seq[ring],
>ring, false, false);
> }
>
> @@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
> *rdev, int ring)
>* but it's ok to report slightly wrong fence count here.
>*/
> radeon_fence_process(rdev, ring);
> - emitted = rdev->fence_drv[ring].seq -
> atomic64_read(&rdev->fence_drv[ring].last_seq);
> + emitted = rdev->fence_drv[ring].sync_seq[ring]
> + - atomic64_read(&rdev->fence_drv[ring].last_seq);
> /* to avoid 32bits warp around */
> if (emitted > 0x1000) {
> emitted = 0x1000;
> @@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
> *rdev, int ring)
> return (unsigned)emitted;
> }
>
> +bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
> +{
> + struct radeon_fence_driver *fdrv;
> +
> + if (!fence) {
> + return false;
> + }
> +
> + if (fence->ring == dst_ring) {
> + return false;
> + }
> +
> + /* we are protected by the ring mutex */
> + fdrv = &fence->rdev->fence_drv[dst_ring];
> + if (fence->seq <= fdrv->sync_seq[fence->ring]) {
> + return false;
> + }
> +
> + return true;
> +}
> +
> +void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
> +{
> + struct radeon_fence_driver *dst, *src;
> + unsigned i;
> +
> + if (!fence) {
> + return;
>
[PATCH 02/10] drm/radeon: add infrastructure for advanced ring synchronization
Signed-off-by: Christian König
---
drivers/gpu/drm/radeon/radeon.h | 23 ++-
drivers/gpu/drm/radeon/radeon_fence.c | 73 +
2 files changed, 85 insertions(+), 11 deletions(-)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 5e259b4..4e232c3 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -257,8 +257,8 @@ struct radeon_fence_driver {
uint32_tscratch_reg;
uint64_tgpu_addr;
volatile uint32_t *cpu_addr;
- /* seq is protected by ring emission lock */
- uint64_tseq;
+ /* sync_seq is protected by ring emission lock */
+ uint64_tsync_seq[RADEON_NUM_RINGS];
atomic64_t last_seq;
unsigned long last_activity;
boolinitialized;
@@ -288,6 +288,25 @@ int radeon_fence_wait_any(struct radeon_device *rdev,
struct radeon_fence *radeon_fence_ref(struct radeon_fence *fence);
void radeon_fence_unref(struct radeon_fence **fence);
unsigned radeon_fence_count_emitted(struct radeon_device *rdev, int ring);
+bool radeon_fence_need_sync(struct radeon_fence *fence, int ring);
+void radeon_fence_note_sync(struct radeon_fence *fence, int ring);
+static inline struct radeon_fence *radeon_fence_later(struct radeon_fence *a,
+ struct radeon_fence *b)
+{
+ if (!a) {
+ return b;
+ }
+
+ if (!b) {
+ return a;
+ }
+
+ if (a->seq > b->seq) {
+ return a;
+ } else {
+ return b;
+ }
+}
/*
* Tiling registers
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c
b/drivers/gpu/drm/radeon/radeon_fence.c
index 401d346..7b55625 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -72,7 +72,7 @@ int radeon_fence_emit(struct radeon_device *rdev,
}
kref_init(&((*fence)->kref));
(*fence)->rdev = rdev;
- (*fence)->seq = ++rdev->fence_drv[ring].seq;
+ (*fence)->seq = ++rdev->fence_drv[ring].sync_seq[ring];
(*fence)->ring = ring;
radeon_fence_ring_emit(rdev, ring, *fence);
trace_radeon_fence_emit(rdev->ddev, (*fence)->seq);
@@ -449,7 +449,7 @@ int radeon_fence_wait_next_locked(struct radeon_device
*rdev, int ring)
* wait.
*/
seq = atomic64_read(&rdev->fence_drv[ring].last_seq) + 1ULL;
- if (seq >= rdev->fence_drv[ring].seq) {
+ if (seq >= rdev->fence_drv[ring].sync_seq[ring]) {
/* nothing to wait for, last_seq is
already the last emited fence */
return -ENOENT;
@@ -464,7 +464,7 @@ int radeon_fence_wait_empty_locked(struct radeon_device
*rdev, int ring)
* activity can be scheduled so there won't be concurrent access
* to seq value.
*/
- return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].seq,
+ return radeon_fence_wait_seq(rdev, rdev->fence_drv[ring].sync_seq[ring],
ring, false, false);
}
@@ -492,7 +492,8 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
* but it's ok to report slightly wrong fence count here.
*/
radeon_fence_process(rdev, ring);
- emitted = rdev->fence_drv[ring].seq -
atomic64_read(&rdev->fence_drv[ring].last_seq);
+ emitted = rdev->fence_drv[ring].sync_seq[ring]
+ - atomic64_read(&rdev->fence_drv[ring].last_seq);
/* to avoid 32bits warp around */
if (emitted > 0x1000) {
emitted = 0x1000;
@@ -500,6 +501,51 @@ unsigned radeon_fence_count_emitted(struct radeon_device
*rdev, int ring)
return (unsigned)emitted;
}
+bool radeon_fence_need_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *fdrv;
+
+ if (!fence) {
+ return false;
+ }
+
+ if (fence->ring == dst_ring) {
+ return false;
+ }
+
+ /* we are protected by the ring mutex */
+ fdrv = &fence->rdev->fence_drv[dst_ring];
+ if (fence->seq <= fdrv->sync_seq[fence->ring]) {
+ return false;
+ }
+
+ return true;
+}
+
+void radeon_fence_note_sync(struct radeon_fence *fence, int dst_ring)
+{
+ struct radeon_fence_driver *dst, *src;
+ unsigned i;
+
+ if (!fence) {
+ return;
+ }
+
+ if (fence->ring == dst_ring) {
+ return;
+ }
+
+ /* we are protected by the ring mutex */
+ src = &fence->rdev->fence_drv[fence->ring];
+ dst = &fence->rdev->fence_drv[dst_ring];
+ for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+ if (i == dst_ring) {
+ contin
