* [PATCH 0/5] drm/ttm, amdgpu: fix crashes due to shadow page table evictions
@ 2016-12-15 17:10 Nicolai Hähnle
[not found] ` <1481821815-9971-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
0 siblings, 1 reply; 15+ messages in thread
From: Nicolai Hähnle @ 2016-12-15 17:10 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Fix a bunch of related crashes in amdgpu that occur when shadow page tables
are kicked out of the GART.
One of the issues was that during command submission, we rely on a
device-global evictions counter to skip some of the work of page-table
validation. The driver was never informed of evictions from GART, so this
series adds an evict parameter to ttm_bo_driver::move_notify.
There's still the evict parameter on ttm_bo_driver::move which is used by
radeon and nouveau for the call to ttm_bo_move_accel_cleanup. The 'evict'
parameter there should probably be more accurately called 'wait', but unless
that wait can always be avoided, the evict parameter on ttm_bo_driver::move
needs to stay.
Please review!
Nicolai
---
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 20 ++++++++++++++------
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 5 +++++
drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 12 +++++++++++-
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 4 ----
drivers/gpu/drm/nouveau/nouveau_bo.c | 3 ++-
drivers/gpu/drm/qxl/qxl_ttm.c | 1 +
drivers/gpu/drm/radeon/radeon_object.c | 1 +
drivers/gpu/drm/radeon/radeon_object.h | 1 +
drivers/gpu/drm/ttm/ttm_bo.c | 8 ++++----
drivers/gpu/drm/virtio/virtgpu_ttm.c | 1 +
drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 1 +
include/drm/ttm/ttm_bo_driver.h | 10 ++++++++--
12 files changed, 49 insertions(+), 18 deletions(-)
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
* [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify
[not found] ` <1481821815-9971-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
@ 2016-12-15 17:10 ` Nicolai Hähnle
[not found] ` <1481821815-9971-2-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:10 ` [PATCH 2/5] drm/amd/amdgpu: move eviction counting to amdgpu_bo_move_notify Nicolai Hähnle
` (3 subsequent siblings)
4 siblings, 1 reply; 15+ messages in thread
From: Nicolai Hähnle @ 2016-12-15 17:10 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: Nicolai Hähnle, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Nicolai Hähnle <nicolai.haehnle@amd.com>
Ensure that the driver can listen to evictions even when they don't take the
path through ttm_bo_driver::move.
This is crucial for amdgpu, which relies on an eviction counter to skip
re-binding page tables when possible.
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 3 ++-
drivers/gpu/drm/nouveau/nouveau_bo.c | 3 ++-
drivers/gpu/drm/qxl/qxl_ttm.c | 1 +
drivers/gpu/drm/radeon/radeon_object.c | 1 +
drivers/gpu/drm/radeon/radeon_object.h | 1 +
drivers/gpu/drm/ttm/ttm_bo.c | 8 ++++----
drivers/gpu/drm/virtio/virtgpu_ttm.c | 1 +
drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 1 +
include/drm/ttm/ttm_bo_driver.h | 10 ++++++++--
10 files changed, 22 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index bf79b73..c29db99 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -842,20 +842,21 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
if (metadata_size)
*metadata_size = bo->metadata_size;
if (flags)
*flags = bo->metadata_flags;
return 0;
}
void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
+ bool evict,
struct ttm_mem_reg *new_mem)
{
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
struct amdgpu_bo *abo;
struct ttm_mem_reg *old_mem = &bo->mem;
if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
return;
abo = container_of(bo, struct amdgpu_bo, tbo);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 5cbf59e..4306b2f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -148,21 +148,22 @@ void amdgpu_bo_fini(struct amdgpu_device *adev);
int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
struct vm_area_struct *vma);
int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64 *tiling_flags);
int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
uint32_t metadata_size, uint64_t flags);
int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
size_t buffer_size, uint32_t *metadata_size,
uint64_t *flags);
void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
- struct ttm_mem_reg *new_mem);
+ bool evict,
+ struct ttm_mem_reg *new_mem);
int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
bool shared);
u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
struct amdgpu_ring *ring,
struct amdgpu_bo *bo,
struct reservation_object *resv,
struct dma_fence **fence, bool direct);
int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index e0c0007..6fa1521 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -1187,21 +1187,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
if (ret)
goto out;
out:
ttm_bo_mem_put(bo, &tmp_mem);
return ret;
}
static void
-nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem)
+nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
+ struct ttm_mem_reg *new_mem)
{
struct nouveau_bo *nvbo = nouveau_bo(bo);
struct nvkm_vma *vma;
/* ttm can now (stupidly) pass the driver bos it didn't create... */
if (bo->destroy != nouveau_bo_del_ttm)
return;
list_for_each_entry(vma, &nvbo->vma_list, head) {
if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
index 1176133..f3939a9 100644
--- a/drivers/gpu/drm/qxl/qxl_ttm.c
+++ b/drivers/gpu/drm/qxl/qxl_ttm.c
@@ -360,20 +360,21 @@ static int qxl_bo_move(struct ttm_buffer_object *bo,
if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
qxl_move_null(bo, new_mem);
return 0;
}
return ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
new_mem);
}
static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
+ bool evict,
struct ttm_mem_reg *new_mem)
{
struct qxl_bo *qbo;
struct qxl_device *qdev;
if (!qxl_ttm_bo_is_qxl_bo(bo))
return;
qbo = to_qxl_bo(bo);
qdev = qbo->gem_base.dev->dev_private;
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index 41b72ce..74b2760 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -758,20 +758,21 @@ int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
return 0;
}
if ((bo->surface_reg >= 0) && !has_moved)
return 0;
return radeon_bo_get_surface_reg(bo);
}
void radeon_bo_move_notify(struct ttm_buffer_object *bo,
+ bool evict,
struct ttm_mem_reg *new_mem)
{
struct radeon_bo *rbo;
if (!radeon_ttm_bo_is_radeon_bo(bo))
return;
rbo = container_of(bo, struct radeon_bo, tbo);
radeon_bo_check_tiling(rbo, 0, 1);
radeon_vm_bo_invalidate(rbo->rdev, rbo);
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index a10bb3d..9ffd821 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -143,20 +143,21 @@ extern void radeon_bo_fini(struct radeon_device *rdev);
extern int radeon_bo_list_validate(struct radeon_device *rdev,
struct ww_acquire_ctx *ticket,
struct list_head *head, int ring);
extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
u32 tiling_flags, u32 pitch);
extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
u32 *tiling_flags, u32 *pitch);
extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
bool force_drop);
extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
+ bool evict,
struct ttm_mem_reg *new_mem);
extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
extern void radeon_bo_fence(struct radeon_bo *bo, struct radeon_fence *fence,
bool shared);
/*
* sub allocation
*/
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index d506361..d4973e9c 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -335,45 +335,45 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
goto out_err;
if (mem->mem_type != TTM_PL_SYSTEM) {
ret = ttm_tt_bind(bo->ttm, mem);
if (ret)
goto out_err;
}
if (bo->mem.mem_type == TTM_PL_SYSTEM) {
if (bdev->driver->move_notify)
- bdev->driver->move_notify(bo, mem);
+ bdev->driver->move_notify(bo, evict, mem);
bo->mem = *mem;
mem->mm_node = NULL;
goto moved;
}
}
if (bdev->driver->move_notify)
- bdev->driver->move_notify(bo, mem);
+ bdev->driver->move_notify(bo, evict, mem);
if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
!(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
ret = ttm_bo_move_ttm(bo, interruptible, no_wait_gpu, mem);
else if (bdev->driver->move)
ret = bdev->driver->move(bo, evict, interruptible,
no_wait_gpu, mem);
else
ret = ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu, mem);
if (ret) {
if (bdev->driver->move_notify) {
struct ttm_mem_reg tmp_mem = *mem;
*mem = bo->mem;
bo->mem = tmp_mem;
- bdev->driver->move_notify(bo, mem);
+ bdev->driver->move_notify(bo, false, mem);
bo->mem = *mem;
*mem = tmp_mem;
}
goto out_err;
}
moved:
if (bo->evicted) {
if (bdev->driver->invalidate_caches) {
@@ -407,21 +407,21 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
* Call bo::reserved.
* Will release GPU memory type usage on destruction.
* This is the place to put in driver specific hooks to release
* driver private resources.
* Will release the bo::reserved lock.
*/
static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
{
if (bo->bdev->driver->move_notify)
- bo->bdev->driver->move_notify(bo, NULL);
+ bo->bdev->driver->move_notify(bo, false, NULL);
ttm_tt_destroy(bo->ttm);
bo->ttm = NULL;
ttm_bo_mem_put(bo, &bo->mem);
ww_mutex_unlock (&bo->resv->lock);
}
static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
{
diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
index 4a1de9f..10387d7 100644
--- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
+++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
@@ -379,20 +379,21 @@ static int virtio_gpu_bo_move(struct ttm_buffer_object *bo,
ret = ttm_bo_wait(bo, interruptible, no_wait_gpu);
if (ret)
return ret;
virtio_gpu_move_null(bo, new_mem);
return 0;
}
static void virtio_gpu_bo_move_notify(struct ttm_buffer_object *tbo,
+ bool evict,
struct ttm_mem_reg *new_mem)
{
struct virtio_gpu_object *bo;
struct virtio_gpu_device *vgdev;
bo = container_of(tbo, struct virtio_gpu_object, tbo);
vgdev = (struct virtio_gpu_device *)bo->gem_base.dev->dev_private;
if (!new_mem || (new_mem->placement & TTM_PL_FLAG_SYSTEM)) {
if (bo->hw_res_handle)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
index c894a48..caa279b 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
@@ -818,20 +818,21 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
* vmw_move_notify - TTM move_notify_callback
*
* @bo: The TTM buffer object about to move.
* @mem: The struct ttm_mem_reg indicating to what memory
* region the move is taking place.
*
* Calls move_notify for all subsystems needing it.
* (currently only resources).
*/
static void vmw_move_notify(struct ttm_buffer_object *bo,
+ bool evict,
struct ttm_mem_reg *mem)
{
vmw_resource_move_notify(bo, mem);
vmw_query_move_notify(bo, mem);
}
/**
* vmw_swap_notify - TTM move_notify_callback
*
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index cdbdb40..8b4bec7 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -424,23 +424,29 @@ struct ttm_bo_driver {
*
* Called from the map / write / read methods to verify that the
* caller is permitted to access the buffer object.
* This member may be set to NULL, which will refuse this kind of
* access for all buffer objects.
* This function should return 0 if access is granted, -EPERM otherwise.
*/
int (*verify_access)(struct ttm_buffer_object *bo,
struct file *filp);
- /* hook to notify driver about a driver move so it
- * can do tiling things */
+ /**
+ * Hook to notify driver about a driver move so it
+ * can do tiling things and book-keeping.
+ *
+ * @evict: whether this move is evicting the buffer from the graphics
+ * address space
+ */
void (*move_notify)(struct ttm_buffer_object *bo,
+ bool evict,
struct ttm_mem_reg *new_mem);
/* notify the driver we are taking a fault on this BO
* and have reserved it */
int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
/**
* notify the driver that we're about to swap out this bo
*/
void (*swap_notify)(struct ttm_buffer_object *bo);
--
2.7.4
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 2/5] drm/amd/amdgpu: move eviction counting to amdgpu_bo_move_notify
[not found] ` <1481821815-9971-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:10 ` [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify Nicolai Hähnle
@ 2016-12-15 17:10 ` Nicolai Hähnle
[not found] ` <1481821815-9971-3-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:10 ` [PATCH 3/5] drm/amd/amdgpu: add amdgpu_bo_gpu_accessible helper function Nicolai Hähnle
` (2 subsequent siblings)
4 siblings, 1 reply; 15+ messages in thread
From: Nicolai Hähnle @ 2016-12-15 17:10 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: Nicolai Hähnle, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Nicolai Hähnle <nicolai.haehnle@amd.com>
This catches evictions of shadow page tables from the GART. Since shadow
page tables are always stored in system memory, amdgpu_bo_move is never
called for them.
This fixes a crash during command submission that occurs when only a shadow
page table and no other BOs were evicted since the last submission.
Fixes: 1baa439fb2f4e586 ("drm/amdgpu: allocate shadow for pd/pt bo V2")
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 ++++
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 4 ----
2 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index c29db99..d94cdef 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -855,20 +855,24 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
struct amdgpu_bo *abo;
struct ttm_mem_reg *old_mem = &bo->mem;
if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
return;
abo = container_of(bo, struct amdgpu_bo, tbo);
amdgpu_vm_bo_invalidate(adev, abo);
+ /* remember the eviction */
+ if (evict)
+ atomic64_inc(&adev->num_evictions);
+
/* update statistics */
if (!new_mem)
return;
/* move_notify is called before move happens */
amdgpu_update_memory_usage(adev, &bo->mem, new_mem);
trace_amdgpu_ttm_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 8f18b8e..80924c2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -460,24 +460,20 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo,
struct ttm_mem_reg *old_mem = &bo->mem;
int r;
/* Can't move a pinned BO */
abo = container_of(bo, struct amdgpu_bo, tbo);
if (WARN_ON_ONCE(abo->pin_count > 0))
return -EINVAL;
adev = amdgpu_ttm_adev(bo->bdev);
- /* remember the eviction */
- if (evict)
- atomic64_inc(&adev->num_evictions);
-
if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
amdgpu_move_null(bo, new_mem);
return 0;
}
if ((old_mem->mem_type == TTM_PL_TT &&
new_mem->mem_type == TTM_PL_SYSTEM) ||
(old_mem->mem_type == TTM_PL_SYSTEM &&
new_mem->mem_type == TTM_PL_TT)) {
/* bind is enough */
amdgpu_move_null(bo, new_mem);
--
2.7.4
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 3/5] drm/amd/amdgpu: add amdgpu_bo_gpu_accessible helper function
[not found] ` <1481821815-9971-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:10 ` [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify Nicolai Hähnle
2016-12-15 17:10 ` [PATCH 2/5] drm/amd/amdgpu: move eviction counting to amdgpu_bo_move_notify Nicolai Hähnle
@ 2016-12-15 17:10 ` Nicolai Hähnle
[not found] ` <1481821815-9971-4-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:10 ` [PATCH 4/5] drm/amd/amdgpu: add check that shadow page directory is GPU-accessible Nicolai Hähnle
2016-12-15 17:10 ` [PATCH 5/5] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible Nicolai Hähnle
4 siblings, 1 reply; 15+ messages in thread
From: Nicolai Hähnle @ 2016-12-15 17:10 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: Nicolai Hähnle, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Nicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index 4306b2f..15a723a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -107,20 +107,29 @@ static inline unsigned amdgpu_bo_gpu_page_alignment(struct amdgpu_bo *bo)
* amdgpu_bo_mmap_offset - return mmap offset of bo
* @bo: amdgpu object for which we query the offset
*
* Returns mmap offset of the object.
*/
static inline u64 amdgpu_bo_mmap_offset(struct amdgpu_bo *bo)
{
return drm_vma_node_offset_addr(&bo->tbo.vma_node);
}
+/**
+ * amdgpu_bo_gpu_accessible - return whether the bo is currently in memory that
+ * is accessible to the GPU.
+ */
+static inline bool amdgpu_bo_gpu_accessible(struct amdgpu_bo *bo)
+{
+ return bo->tbo.mem.mem_type != TTM_PL_SYSTEM;
+}
+
int amdgpu_bo_create(struct amdgpu_device *adev,
unsigned long size, int byte_align,
bool kernel, u32 domain, u64 flags,
struct sg_table *sg,
struct reservation_object *resv,
struct amdgpu_bo **bo_ptr);
int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
unsigned long size, int byte_align,
bool kernel, u32 domain, u64 flags,
struct sg_table *sg,
--
2.7.4
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 4/5] drm/amd/amdgpu: add check that shadow page directory is GPU-accessible
[not found] ` <1481821815-9971-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
` (2 preceding siblings ...)
2016-12-15 17:10 ` [PATCH 3/5] drm/amd/amdgpu: add amdgpu_bo_gpu_accessible helper function Nicolai Hähnle
@ 2016-12-15 17:10 ` Nicolai Hähnle
[not found] ` <1481821815-9971-5-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:10 ` [PATCH 5/5] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible Nicolai Hähnle
4 siblings, 1 reply; 15+ messages in thread
From: Nicolai Hähnle @ 2016-12-15 17:10 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: Nicolai Hähnle, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Nicolai Hähnle <nicolai.haehnle@amd.com>
Skip amdgpu_gem_va_update_vm when shadow the page directory is swapped out.
Clean up the check for non-shadow BOs as well using the new helper function.
This fixes a crash with the stack trace:
amdgpu_gem_va_update_vm
-> amdgpu_vm_update_page_directory
-> amdgpu_ttm_bind
-> amdgpu_gtt_mgr_alloc
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 11 ++++++++---
1 file changed, 8 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index cd62f6f..4e1eb05 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -489,44 +489,49 @@ static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
* vital here, so they are not reported back to userspace.
*/
static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
struct amdgpu_bo_va *bo_va,
uint32_t operation)
{
struct ttm_validate_buffer tv, *entry;
struct amdgpu_bo_list_entry vm_pd;
struct ww_acquire_ctx ticket;
struct list_head list, duplicates;
- unsigned domain;
int r;
INIT_LIST_HEAD(&list);
INIT_LIST_HEAD(&duplicates);
tv.bo = &bo_va->bo->tbo;
tv.shared = true;
list_add(&tv.head, &list);
amdgpu_vm_get_pd_bo(bo_va->vm, &list, &vm_pd);
/* Provide duplicates to avoid -EALREADY */
r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
if (r)
goto error_print;
list_for_each_entry(entry, &list, head) {
- domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type);
+ struct amdgpu_bo *bo =
+ container_of(entry->bo, struct amdgpu_bo, tbo);
+
/* if anything is swapped out don't swap it in here,
just abort and wait for the next CS */
- if (domain == AMDGPU_GEM_DOMAIN_CPU)
+ if (!amdgpu_bo_gpu_accessible(bo))
+ goto error_unreserve;
+
+ if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
goto error_unreserve;
}
+
r = amdgpu_vm_validate_pt_bos(adev, bo_va->vm, amdgpu_gem_va_check,
NULL);
if (r)
goto error_unreserve;
r = amdgpu_vm_update_page_directory(adev, bo_va->vm);
if (r)
goto error_unreserve;
r = amdgpu_vm_clear_freed(adev, bo_va->vm);
--
2.7.4
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 5/5] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible
[not found] ` <1481821815-9971-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
` (3 preceding siblings ...)
2016-12-15 17:10 ` [PATCH 4/5] drm/amd/amdgpu: add check that shadow page directory is GPU-accessible Nicolai Hähnle
@ 2016-12-15 17:10 ` Nicolai Hähnle
[not found] ` <1481821815-9971-6-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
4 siblings, 1 reply; 15+ messages in thread
From: Nicolai Hähnle @ 2016-12-15 17:10 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: Nicolai Hähnle, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Nicolai Hähnle <nicolai.haehnle@amd.com>
Skip amdgpu_gem_va_update_vm otherwise. Also clean up the check for the
non-shadow page tables using the new helper function.
This fixes a crash with the stack trace:
amdgpu_gem_va_update_vm
-> amdgpu_vm_update_page_directory
-> amdgpu_ttm_bind
-> amdgpu_gtt_mgr_alloc
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 4e1eb05..d91c80b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -464,26 +464,29 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
unreserve:
amdgpu_bo_unreserve(robj);
out:
drm_gem_object_unreference_unlocked(gobj);
return r;
}
static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
{
- unsigned domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
-
/* if anything is swapped out don't swap it in here,
just abort and wait for the next CS */
+ if (!amdgpu_bo_gpu_accessible(bo))
+ return -ERESTARTSYS;
+
+ if (bo->shadow && !amdgpu_bo_gpu_accessible(bo))
+ return -ERESTARTSYS;
- return domain == AMDGPU_GEM_DOMAIN_CPU ? -ERESTARTSYS : 0;
+ return 0;
}
/**
* amdgpu_gem_va_update_vm -update the bo_va in its VM
*
* @adev: amdgpu_device pointer
* @bo_va: bo_va to update
*
* Update the bo_va directly after setting it's address. Errors are not
* vital here, so they are not reported back to userspace.
--
2.7.4
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH v2] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible
[not found] ` <1481821815-9971-6-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
@ 2016-12-15 17:59 ` Nicolai Hähnle
[not found] ` <1481824743-11139-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-16 2:07 ` [PATCH 5/5] " zhoucm1
1 sibling, 1 reply; 15+ messages in thread
From: Nicolai Hähnle @ 2016-12-15 17:59 UTC (permalink / raw)
To: amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: Nicolai Hähnle, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
From: Nicolai Hähnle <nicolai.haehnle@amd.com>
Skip amdgpu_gem_va_update_vm otherwise. Also clean up the check for the
non-shadow page tables using the new helper function.
This fixes a crash with the stack trace:
amdgpu_gem_va_update_vm
-> amdgpu_vm_update_page_directory
-> amdgpu_ttm_bind
-> amdgpu_gtt_mgr_alloc
v2: actually check bo->shadow instead of just checking bo twice
Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 4e1eb05..9bd1b4e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -464,26 +464,29 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
unreserve:
amdgpu_bo_unreserve(robj);
out:
drm_gem_object_unreference_unlocked(gobj);
return r;
}
static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
{
- unsigned domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
-
/* if anything is swapped out don't swap it in here,
just abort and wait for the next CS */
+ if (!amdgpu_bo_gpu_accessible(bo))
+ return -ERESTARTSYS;
+
+ if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
+ return -ERESTARTSYS;
- return domain == AMDGPU_GEM_DOMAIN_CPU ? -ERESTARTSYS : 0;
+ return 0;
}
/**
* amdgpu_gem_va_update_vm -update the bo_va in its VM
*
* @adev: amdgpu_device pointer
* @bo_va: bo_va to update
*
* Update the bo_va directly after setting it's address. Errors are not
* vital here, so they are not reported back to userspace.
--
2.7.4
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply related [flat|nested] 15+ messages in thread
* Re: [PATCH 5/5] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible
[not found] ` <1481821815-9971-6-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:59 ` [PATCH v2] " Nicolai Hähnle
@ 2016-12-16 2:07 ` zhoucm1
1 sibling, 0 replies; 15+ messages in thread
From: zhoucm1 @ 2016-12-16 2:07 UTC (permalink / raw)
To: Nicolai Hähnle, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Nicolai Hähnle
On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Skip amdgpu_gem_va_update_vm otherwise. Also clean up the check for the
> non-shadow page tables using the new helper function.
>
> This fixes a crash with the stack trace:
>
> amdgpu_gem_va_update_vm
> -> amdgpu_vm_update_page_directory
> -> amdgpu_ttm_bind
> -> amdgpu_gtt_mgr_alloc
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> index 4e1eb05..d91c80b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> @@ -464,26 +464,29 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
>
> unreserve:
> amdgpu_bo_unreserve(robj);
> out:
> drm_gem_object_unreference_unlocked(gobj);
> return r;
> }
>
> static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
> {
> - unsigned domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
> -
> /* if anything is swapped out don't swap it in here,
> just abort and wait for the next CS */
> + if (!amdgpu_bo_gpu_accessible(bo))
> + return -ERESTARTSYS;
> +
> + if (bo->shadow && !amdgpu_bo_gpu_accessible(bo))
this should be 'if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))'.
Regards,
David Zhou
> + return -ERESTARTSYS;
>
> - return domain == AMDGPU_GEM_DOMAIN_CPU ? -ERESTARTSYS : 0;
> + return 0;
> }
>
> /**
> * amdgpu_gem_va_update_vm -update the bo_va in its VM
> *
> * @adev: amdgpu_device pointer
> * @bo_va: bo_va to update
> *
> * Update the bo_va directly after setting it's address. Errors are not
> * vital here, so they are not reported back to userspace.
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH v2] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible
[not found] ` <1481824743-11139-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
@ 2016-12-16 2:48 ` zhoucm1
0 siblings, 0 replies; 15+ messages in thread
From: zhoucm1 @ 2016-12-16 2:48 UTC (permalink / raw)
To: Nicolai Hähnle, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Nicolai Hähnle
On 2016年12月16日 01:59, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Skip amdgpu_gem_va_update_vm otherwise. Also clean up the check for the
> non-shadow page tables using the new helper function.
>
> This fixes a crash with the stack trace:
>
> amdgpu_gem_va_update_vm
> -> amdgpu_vm_update_page_directory
> -> amdgpu_ttm_bind
> -> amdgpu_gtt_mgr_alloc
>
> v2: actually check bo->shadow instead of just checking bo twice
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 9 ++++++---
> 1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> index 4e1eb05..9bd1b4e 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> @@ -464,26 +464,29 @@ int amdgpu_gem_metadata_ioctl(struct drm_device *dev, void *data,
>
> unreserve:
> amdgpu_bo_unreserve(robj);
> out:
> drm_gem_object_unreference_unlocked(gobj);
> return r;
> }
>
> static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
> {
> - unsigned domain = amdgpu_mem_type_to_domain(bo->tbo.mem.mem_type);
> -
> /* if anything is swapped out don't swap it in here,
> just abort and wait for the next CS */
> + if (!amdgpu_bo_gpu_accessible(bo))
> + return -ERESTARTSYS;
> +
> + if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
> + return -ERESTARTSYS;
>
> - return domain == AMDGPU_GEM_DOMAIN_CPU ? -ERESTARTSYS : 0;
> + return 0;
> }
>
> /**
> * amdgpu_gem_va_update_vm -update the bo_va in its VM
> *
> * @adev: amdgpu_device pointer
> * @bo_va: bo_va to update
> *
> * Update the bo_va directly after setting it's address. Errors are not
> * vital here, so they are not reported back to userspace.
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 3/5] drm/amd/amdgpu: add amdgpu_bo_gpu_accessible helper function
[not found] ` <1481821815-9971-4-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
@ 2016-12-16 2:48 ` zhoucm1
0 siblings, 0 replies; 15+ messages in thread
From: zhoucm1 @ 2016-12-16 2:48 UTC (permalink / raw)
To: Nicolai Hähnle, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Nicolai Hähnle
On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 9 +++++++++
> 1 file changed, 9 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> index 4306b2f..15a723a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> @@ -107,20 +107,29 @@ static inline unsigned amdgpu_bo_gpu_page_alignment(struct amdgpu_bo *bo)
> * amdgpu_bo_mmap_offset - return mmap offset of bo
> * @bo: amdgpu object for which we query the offset
> *
> * Returns mmap offset of the object.
> */
> static inline u64 amdgpu_bo_mmap_offset(struct amdgpu_bo *bo)
> {
> return drm_vma_node_offset_addr(&bo->tbo.vma_node);
> }
>
> +/**
> + * amdgpu_bo_gpu_accessible - return whether the bo is currently in memory that
> + * is accessible to the GPU.
> + */
> +static inline bool amdgpu_bo_gpu_accessible(struct amdgpu_bo *bo)
> +{
> + return bo->tbo.mem.mem_type != TTM_PL_SYSTEM;
> +}
> +
> int amdgpu_bo_create(struct amdgpu_device *adev,
> unsigned long size, int byte_align,
> bool kernel, u32 domain, u64 flags,
> struct sg_table *sg,
> struct reservation_object *resv,
> struct amdgpu_bo **bo_ptr);
> int amdgpu_bo_create_restricted(struct amdgpu_device *adev,
> unsigned long size, int byte_align,
> bool kernel, u32 domain, u64 flags,
> struct sg_table *sg,
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 4/5] drm/amd/amdgpu: add check that shadow page directory is GPU-accessible
[not found] ` <1481821815-9971-5-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
@ 2016-12-16 2:49 ` zhoucm1
0 siblings, 0 replies; 15+ messages in thread
From: zhoucm1 @ 2016-12-16 2:49 UTC (permalink / raw)
To: Nicolai Hähnle, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Nicolai Hähnle
On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Skip amdgpu_gem_va_update_vm when shadow the page directory is swapped out.
> Clean up the check for non-shadow BOs as well using the new helper function.
>
> This fixes a crash with the stack trace:
>
> amdgpu_gem_va_update_vm
> -> amdgpu_vm_update_page_directory
> -> amdgpu_ttm_bind
> -> amdgpu_gtt_mgr_alloc
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 11 ++++++++---
> 1 file changed, 8 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> index cd62f6f..4e1eb05 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
> @@ -489,44 +489,49 @@ static int amdgpu_gem_va_check(void *param, struct amdgpu_bo *bo)
> * vital here, so they are not reported back to userspace.
> */
> static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
> struct amdgpu_bo_va *bo_va,
> uint32_t operation)
> {
> struct ttm_validate_buffer tv, *entry;
> struct amdgpu_bo_list_entry vm_pd;
> struct ww_acquire_ctx ticket;
> struct list_head list, duplicates;
> - unsigned domain;
> int r;
>
> INIT_LIST_HEAD(&list);
> INIT_LIST_HEAD(&duplicates);
>
> tv.bo = &bo_va->bo->tbo;
> tv.shared = true;
> list_add(&tv.head, &list);
>
> amdgpu_vm_get_pd_bo(bo_va->vm, &list, &vm_pd);
>
> /* Provide duplicates to avoid -EALREADY */
> r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
> if (r)
> goto error_print;
>
> list_for_each_entry(entry, &list, head) {
> - domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type);
> + struct amdgpu_bo *bo =
> + container_of(entry->bo, struct amdgpu_bo, tbo);
> +
> /* if anything is swapped out don't swap it in here,
> just abort and wait for the next CS */
> - if (domain == AMDGPU_GEM_DOMAIN_CPU)
> + if (!amdgpu_bo_gpu_accessible(bo))
> + goto error_unreserve;
> +
> + if (bo->shadow && !amdgpu_bo_gpu_accessible(bo->shadow))
> goto error_unreserve;
> }
> +
> r = amdgpu_vm_validate_pt_bos(adev, bo_va->vm, amdgpu_gem_va_check,
> NULL);
> if (r)
> goto error_unreserve;
>
> r = amdgpu_vm_update_page_directory(adev, bo_va->vm);
> if (r)
> goto error_unreserve;
>
> r = amdgpu_vm_clear_freed(adev, bo_va->vm);
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify
[not found] ` <1481821815-9971-2-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
@ 2016-12-16 2:49 ` zhoucm1
[not found] ` <5853564E.5020005-5C7GfCeVMHo@public.gmane.org>
0 siblings, 1 reply; 15+ messages in thread
From: zhoucm1 @ 2016-12-16 2:49 UTC (permalink / raw)
To: Nicolai Hähnle, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Nicolai Hähnle
On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> Ensure that the driver can listen to evictions even when they don't take the
> path through ttm_bo_driver::move.
>
> This is crucial for amdgpu, which relies on an eviction counter to skip
> re-binding page tables when possible.
>
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Acked-by: Chunming Zhou <david1.zhou@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 3 ++-
> drivers/gpu/drm/nouveau/nouveau_bo.c | 3 ++-
> drivers/gpu/drm/qxl/qxl_ttm.c | 1 +
> drivers/gpu/drm/radeon/radeon_object.c | 1 +
> drivers/gpu/drm/radeon/radeon_object.h | 1 +
> drivers/gpu/drm/ttm/ttm_bo.c | 8 ++++----
> drivers/gpu/drm/virtio/virtgpu_ttm.c | 1 +
> drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 1 +
> include/drm/ttm/ttm_bo_driver.h | 10 ++++++++--
> 10 files changed, 22 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index bf79b73..c29db99 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -842,20 +842,21 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
>
> if (metadata_size)
> *metadata_size = bo->metadata_size;
> if (flags)
> *flags = bo->metadata_flags;
>
> return 0;
> }
>
> void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
> + bool evict,
> struct ttm_mem_reg *new_mem)
> {
> struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
> struct amdgpu_bo *abo;
> struct ttm_mem_reg *old_mem = &bo->mem;
>
> if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
> return;
>
> abo = container_of(bo, struct amdgpu_bo, tbo);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> index 5cbf59e..4306b2f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
> @@ -148,21 +148,22 @@ void amdgpu_bo_fini(struct amdgpu_device *adev);
> int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
> struct vm_area_struct *vma);
> int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
> void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64 *tiling_flags);
> int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
> uint32_t metadata_size, uint64_t flags);
> int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
> size_t buffer_size, uint32_t *metadata_size,
> uint64_t *flags);
> void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
> - struct ttm_mem_reg *new_mem);
> + bool evict,
> + struct ttm_mem_reg *new_mem);
> int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
> void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
> bool shared);
> u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
> int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
> struct amdgpu_ring *ring,
> struct amdgpu_bo *bo,
> struct reservation_object *resv,
> struct dma_fence **fence, bool direct);
> int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
> index e0c0007..6fa1521 100644
> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
> @@ -1187,21 +1187,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object *bo, bool evict, bool intr,
> ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
> if (ret)
> goto out;
>
> out:
> ttm_bo_mem_put(bo, &tmp_mem);
> return ret;
> }
>
> static void
> -nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem)
> +nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
> + struct ttm_mem_reg *new_mem)
> {
> struct nouveau_bo *nvbo = nouveau_bo(bo);
> struct nvkm_vma *vma;
>
> /* ttm can now (stupidly) pass the driver bos it didn't create... */
> if (bo->destroy != nouveau_bo_del_ttm)
> return;
>
> list_for_each_entry(vma, &nvbo->vma_list, head) {
> if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
> diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c b/drivers/gpu/drm/qxl/qxl_ttm.c
> index 1176133..f3939a9 100644
> --- a/drivers/gpu/drm/qxl/qxl_ttm.c
> +++ b/drivers/gpu/drm/qxl/qxl_ttm.c
> @@ -360,20 +360,21 @@ static int qxl_bo_move(struct ttm_buffer_object *bo,
>
> if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
> qxl_move_null(bo, new_mem);
> return 0;
> }
> return ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
> new_mem);
> }
>
> static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
> + bool evict,
> struct ttm_mem_reg *new_mem)
> {
> struct qxl_bo *qbo;
> struct qxl_device *qdev;
>
> if (!qxl_ttm_bo_is_qxl_bo(bo))
> return;
> qbo = to_qxl_bo(bo);
> qdev = qbo->gem_base.dev->dev_private;
>
> diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
> index 41b72ce..74b2760 100644
> --- a/drivers/gpu/drm/radeon/radeon_object.c
> +++ b/drivers/gpu/drm/radeon/radeon_object.c
> @@ -758,20 +758,21 @@ int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
> return 0;
> }
>
> if ((bo->surface_reg >= 0) && !has_moved)
> return 0;
>
> return radeon_bo_get_surface_reg(bo);
> }
>
> void radeon_bo_move_notify(struct ttm_buffer_object *bo,
> + bool evict,
> struct ttm_mem_reg *new_mem)
> {
> struct radeon_bo *rbo;
>
> if (!radeon_ttm_bo_is_radeon_bo(bo))
> return;
>
> rbo = container_of(bo, struct radeon_bo, tbo);
> radeon_bo_check_tiling(rbo, 0, 1);
> radeon_vm_bo_invalidate(rbo->rdev, rbo);
> diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
> index a10bb3d..9ffd821 100644
> --- a/drivers/gpu/drm/radeon/radeon_object.h
> +++ b/drivers/gpu/drm/radeon/radeon_object.h
> @@ -143,20 +143,21 @@ extern void radeon_bo_fini(struct radeon_device *rdev);
> extern int radeon_bo_list_validate(struct radeon_device *rdev,
> struct ww_acquire_ctx *ticket,
> struct list_head *head, int ring);
> extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
> u32 tiling_flags, u32 pitch);
> extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
> u32 *tiling_flags, u32 *pitch);
> extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
> bool force_drop);
> extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
> + bool evict,
> struct ttm_mem_reg *new_mem);
> extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
> extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
> extern void radeon_bo_fence(struct radeon_bo *bo, struct radeon_fence *fence,
> bool shared);
>
> /*
> * sub allocation
> */
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index d506361..d4973e9c 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -335,45 +335,45 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
> goto out_err;
>
> if (mem->mem_type != TTM_PL_SYSTEM) {
> ret = ttm_tt_bind(bo->ttm, mem);
> if (ret)
> goto out_err;
> }
>
> if (bo->mem.mem_type == TTM_PL_SYSTEM) {
> if (bdev->driver->move_notify)
> - bdev->driver->move_notify(bo, mem);
> + bdev->driver->move_notify(bo, evict, mem);
> bo->mem = *mem;
> mem->mm_node = NULL;
> goto moved;
> }
> }
>
> if (bdev->driver->move_notify)
> - bdev->driver->move_notify(bo, mem);
> + bdev->driver->move_notify(bo, evict, mem);
>
> if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
> !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
> ret = ttm_bo_move_ttm(bo, interruptible, no_wait_gpu, mem);
> else if (bdev->driver->move)
> ret = bdev->driver->move(bo, evict, interruptible,
> no_wait_gpu, mem);
> else
> ret = ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu, mem);
>
> if (ret) {
> if (bdev->driver->move_notify) {
> struct ttm_mem_reg tmp_mem = *mem;
> *mem = bo->mem;
> bo->mem = tmp_mem;
> - bdev->driver->move_notify(bo, mem);
> + bdev->driver->move_notify(bo, false, mem);
> bo->mem = *mem;
> *mem = tmp_mem;
> }
>
> goto out_err;
> }
>
> moved:
> if (bo->evicted) {
> if (bdev->driver->invalidate_caches) {
> @@ -407,21 +407,21 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
> * Call bo::reserved.
> * Will release GPU memory type usage on destruction.
> * This is the place to put in driver specific hooks to release
> * driver private resources.
> * Will release the bo::reserved lock.
> */
>
> static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
> {
> if (bo->bdev->driver->move_notify)
> - bo->bdev->driver->move_notify(bo, NULL);
> + bo->bdev->driver->move_notify(bo, false, NULL);
>
> ttm_tt_destroy(bo->ttm);
> bo->ttm = NULL;
> ttm_bo_mem_put(bo, &bo->mem);
>
> ww_mutex_unlock (&bo->resv->lock);
> }
>
> static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
> {
> diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c b/drivers/gpu/drm/virtio/virtgpu_ttm.c
> index 4a1de9f..10387d7 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
> @@ -379,20 +379,21 @@ static int virtio_gpu_bo_move(struct ttm_buffer_object *bo,
>
> ret = ttm_bo_wait(bo, interruptible, no_wait_gpu);
> if (ret)
> return ret;
>
> virtio_gpu_move_null(bo, new_mem);
> return 0;
> }
>
> static void virtio_gpu_bo_move_notify(struct ttm_buffer_object *tbo,
> + bool evict,
> struct ttm_mem_reg *new_mem)
> {
> struct virtio_gpu_object *bo;
> struct virtio_gpu_device *vgdev;
>
> bo = container_of(tbo, struct virtio_gpu_object, tbo);
> vgdev = (struct virtio_gpu_device *)bo->gem_base.dev->dev_private;
>
> if (!new_mem || (new_mem->placement & TTM_PL_FLAG_SYSTEM)) {
> if (bo->hw_res_handle)
> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> index c894a48..caa279b 100644
> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
> @@ -818,20 +818,21 @@ static int vmw_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
> * vmw_move_notify - TTM move_notify_callback
> *
> * @bo: The TTM buffer object about to move.
> * @mem: The struct ttm_mem_reg indicating to what memory
> * region the move is taking place.
> *
> * Calls move_notify for all subsystems needing it.
> * (currently only resources).
> */
> static void vmw_move_notify(struct ttm_buffer_object *bo,
> + bool evict,
> struct ttm_mem_reg *mem)
> {
> vmw_resource_move_notify(bo, mem);
> vmw_query_move_notify(bo, mem);
> }
>
>
> /**
> * vmw_swap_notify - TTM move_notify_callback
> *
> diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
> index cdbdb40..8b4bec7 100644
> --- a/include/drm/ttm/ttm_bo_driver.h
> +++ b/include/drm/ttm/ttm_bo_driver.h
> @@ -424,23 +424,29 @@ struct ttm_bo_driver {
> *
> * Called from the map / write / read methods to verify that the
> * caller is permitted to access the buffer object.
> * This member may be set to NULL, which will refuse this kind of
> * access for all buffer objects.
> * This function should return 0 if access is granted, -EPERM otherwise.
> */
> int (*verify_access)(struct ttm_buffer_object *bo,
> struct file *filp);
>
> - /* hook to notify driver about a driver move so it
> - * can do tiling things */
> + /**
> + * Hook to notify driver about a driver move so it
> + * can do tiling things and book-keeping.
> + *
> + * @evict: whether this move is evicting the buffer from the graphics
> + * address space
> + */
> void (*move_notify)(struct ttm_buffer_object *bo,
> + bool evict,
> struct ttm_mem_reg *new_mem);
> /* notify the driver we are taking a fault on this BO
> * and have reserved it */
> int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
>
> /**
> * notify the driver that we're about to swap out this bo
> */
> void (*swap_notify)(struct ttm_buffer_object *bo);
>
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 2/5] drm/amd/amdgpu: move eviction counting to amdgpu_bo_move_notify
[not found] ` <1481821815-9971-3-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
@ 2016-12-16 2:50 ` zhoucm1
0 siblings, 0 replies; 15+ messages in thread
From: zhoucm1 @ 2016-12-16 2:50 UTC (permalink / raw)
To: Nicolai Hähnle, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Nicolai Hähnle
On 2016年12月16日 01:10, Nicolai Hähnle wrote:
> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>
> This catches evictions of shadow page tables from the GART. Since shadow
> page tables are always stored in system memory, amdgpu_bo_move is never
> called for them.
>
> This fixes a crash during command submission that occurs when only a shadow
> page table and no other BOs were evicted since the last submission.
>
> Fixes: 1baa439fb2f4e586 ("drm/amdgpu: allocate shadow for pd/pt bo V2")
> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Acked-by: Chunming Zhou <david1.zhou@amd.com>
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 ++++
> drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 4 ----
> 2 files changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> index c29db99..d94cdef 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
> @@ -855,20 +855,24 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
> struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
> struct amdgpu_bo *abo;
> struct ttm_mem_reg *old_mem = &bo->mem;
>
> if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
> return;
>
> abo = container_of(bo, struct amdgpu_bo, tbo);
> amdgpu_vm_bo_invalidate(adev, abo);
>
> + /* remember the eviction */
> + if (evict)
> + atomic64_inc(&adev->num_evictions);
> +
> /* update statistics */
> if (!new_mem)
> return;
>
> /* move_notify is called before move happens */
> amdgpu_update_memory_usage(adev, &bo->mem, new_mem);
>
> trace_amdgpu_ttm_bo_move(abo, new_mem->mem_type, old_mem->mem_type);
> }
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> index 8f18b8e..80924c2 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
> @@ -460,24 +460,20 @@ static int amdgpu_bo_move(struct ttm_buffer_object *bo,
> struct ttm_mem_reg *old_mem = &bo->mem;
> int r;
>
> /* Can't move a pinned BO */
> abo = container_of(bo, struct amdgpu_bo, tbo);
> if (WARN_ON_ONCE(abo->pin_count > 0))
> return -EINVAL;
>
> adev = amdgpu_ttm_adev(bo->bdev);
>
> - /* remember the eviction */
> - if (evict)
> - atomic64_inc(&adev->num_evictions);
> -
> if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
> amdgpu_move_null(bo, new_mem);
> return 0;
> }
> if ((old_mem->mem_type == TTM_PL_TT &&
> new_mem->mem_type == TTM_PL_SYSTEM) ||
> (old_mem->mem_type == TTM_PL_SYSTEM &&
> new_mem->mem_type == TTM_PL_TT)) {
> /* bind is enough */
> amdgpu_move_null(bo, new_mem);
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify
[not found] ` <5853564E.5020005-5C7GfCeVMHo@public.gmane.org>
@ 2016-12-21 15:12 ` Nicolai Hähnle
[not found] ` <8f68adf1-52bc-b2ef-2fa7-79a8f9695164-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
0 siblings, 1 reply; 15+ messages in thread
From: Nicolai Hähnle @ 2016-12-21 15:12 UTC (permalink / raw)
To: zhoucm1, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW, Nicolai Hähnle
On 16.12.2016 03:49, zhoucm1 wrote:
> On 2016年12月16日 01:10, Nicolai Hähnle wrote:
>> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>>
>> Ensure that the driver can listen to evictions even when they don't
>> take the
>> path through ttm_bo_driver::move.
>>
>> This is crucial for amdgpu, which relies on an eviction counter to skip
>> re-binding page tables when possible.
>>
>> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
> Acked-by: Chunming Zhou <david1.zhou@amd.com>
Thanks. Ping for feedback from non-AMD people?
Nicolai
>> ---
>> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
>> drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 3 ++-
>> drivers/gpu/drm/nouveau/nouveau_bo.c | 3 ++-
>> drivers/gpu/drm/qxl/qxl_ttm.c | 1 +
>> drivers/gpu/drm/radeon/radeon_object.c | 1 +
>> drivers/gpu/drm/radeon/radeon_object.h | 1 +
>> drivers/gpu/drm/ttm/ttm_bo.c | 8 ++++----
>> drivers/gpu/drm/virtio/virtgpu_ttm.c | 1 +
>> drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 1 +
>> include/drm/ttm/ttm_bo_driver.h | 10 ++++++++--
>> 10 files changed, 22 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> index bf79b73..c29db99 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>> @@ -842,20 +842,21 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo,
>> void *buffer,
>> if (metadata_size)
>> *metadata_size = bo->metadata_size;
>> if (flags)
>> *flags = bo->metadata_flags;
>> return 0;
>> }
>> void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>> + bool evict,
>> struct ttm_mem_reg *new_mem)
>> {
>> struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
>> struct amdgpu_bo *abo;
>> struct ttm_mem_reg *old_mem = &bo->mem;
>> if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
>> return;
>> abo = container_of(bo, struct amdgpu_bo, tbo);
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> index 5cbf59e..4306b2f 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>> @@ -148,21 +148,22 @@ void amdgpu_bo_fini(struct amdgpu_device *adev);
>> int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
>> struct vm_area_struct *vma);
>> int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64 tiling_flags);
>> void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64
>> *tiling_flags);
>> int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
>> uint32_t metadata_size, uint64_t flags);
>> int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
>> size_t buffer_size, uint32_t *metadata_size,
>> uint64_t *flags);
>> void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>> - struct ttm_mem_reg *new_mem);
>> + bool evict,
>> + struct ttm_mem_reg *new_mem);
>> int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
>> void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
>> bool shared);
>> u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
>> int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
>> struct amdgpu_ring *ring,
>> struct amdgpu_bo *bo,
>> struct reservation_object *resv,
>> struct dma_fence **fence, bool direct);
>> int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> index e0c0007..6fa1521 100644
>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>> @@ -1187,21 +1187,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object
>> *bo, bool evict, bool intr,
>> ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
>> if (ret)
>> goto out;
>> out:
>> ttm_bo_mem_put(bo, &tmp_mem);
>> return ret;
>> }
>> static void
>> -nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg
>> *new_mem)
>> +nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
>> + struct ttm_mem_reg *new_mem)
>> {
>> struct nouveau_bo *nvbo = nouveau_bo(bo);
>> struct nvkm_vma *vma;
>> /* ttm can now (stupidly) pass the driver bos it didn't
>> create... */
>> if (bo->destroy != nouveau_bo_del_ttm)
>> return;
>> list_for_each_entry(vma, &nvbo->vma_list, head) {
>> if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
>> diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c
>> b/drivers/gpu/drm/qxl/qxl_ttm.c
>> index 1176133..f3939a9 100644
>> --- a/drivers/gpu/drm/qxl/qxl_ttm.c
>> +++ b/drivers/gpu/drm/qxl/qxl_ttm.c
>> @@ -360,20 +360,21 @@ static int qxl_bo_move(struct ttm_buffer_object
>> *bo,
>> if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
>> qxl_move_null(bo, new_mem);
>> return 0;
>> }
>> return ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
>> new_mem);
>> }
>> static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
>> + bool evict,
>> struct ttm_mem_reg *new_mem)
>> {
>> struct qxl_bo *qbo;
>> struct qxl_device *qdev;
>> if (!qxl_ttm_bo_is_qxl_bo(bo))
>> return;
>> qbo = to_qxl_bo(bo);
>> qdev = qbo->gem_base.dev->dev_private;
>> diff --git a/drivers/gpu/drm/radeon/radeon_object.c
>> b/drivers/gpu/drm/radeon/radeon_object.c
>> index 41b72ce..74b2760 100644
>> --- a/drivers/gpu/drm/radeon/radeon_object.c
>> +++ b/drivers/gpu/drm/radeon/radeon_object.c
>> @@ -758,20 +758,21 @@ int radeon_bo_check_tiling(struct radeon_bo *bo,
>> bool has_moved,
>> return 0;
>> }
>> if ((bo->surface_reg >= 0) && !has_moved)
>> return 0;
>> return radeon_bo_get_surface_reg(bo);
>> }
>> void radeon_bo_move_notify(struct ttm_buffer_object *bo,
>> + bool evict,
>> struct ttm_mem_reg *new_mem)
>> {
>> struct radeon_bo *rbo;
>> if (!radeon_ttm_bo_is_radeon_bo(bo))
>> return;
>> rbo = container_of(bo, struct radeon_bo, tbo);
>> radeon_bo_check_tiling(rbo, 0, 1);
>> radeon_vm_bo_invalidate(rbo->rdev, rbo);
>> diff --git a/drivers/gpu/drm/radeon/radeon_object.h
>> b/drivers/gpu/drm/radeon/radeon_object.h
>> index a10bb3d..9ffd821 100644
>> --- a/drivers/gpu/drm/radeon/radeon_object.h
>> +++ b/drivers/gpu/drm/radeon/radeon_object.h
>> @@ -143,20 +143,21 @@ extern void radeon_bo_fini(struct radeon_device
>> *rdev);
>> extern int radeon_bo_list_validate(struct radeon_device *rdev,
>> struct ww_acquire_ctx *ticket,
>> struct list_head *head, int ring);
>> extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
>> u32 tiling_flags, u32 pitch);
>> extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
>> u32 *tiling_flags, u32 *pitch);
>> extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool has_moved,
>> bool force_drop);
>> extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
>> + bool evict,
>> struct ttm_mem_reg *new_mem);
>> extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object
>> *bo);
>> extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
>> extern void radeon_bo_fence(struct radeon_bo *bo, struct
>> radeon_fence *fence,
>> bool shared);
>> /*
>> * sub allocation
>> */
>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>> b/drivers/gpu/drm/ttm/ttm_bo.c
>> index d506361..d4973e9c 100644
>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>> @@ -335,45 +335,45 @@ static int ttm_bo_handle_move_mem(struct
>> ttm_buffer_object *bo,
>> goto out_err;
>> if (mem->mem_type != TTM_PL_SYSTEM) {
>> ret = ttm_tt_bind(bo->ttm, mem);
>> if (ret)
>> goto out_err;
>> }
>> if (bo->mem.mem_type == TTM_PL_SYSTEM) {
>> if (bdev->driver->move_notify)
>> - bdev->driver->move_notify(bo, mem);
>> + bdev->driver->move_notify(bo, evict, mem);
>> bo->mem = *mem;
>> mem->mm_node = NULL;
>> goto moved;
>> }
>> }
>> if (bdev->driver->move_notify)
>> - bdev->driver->move_notify(bo, mem);
>> + bdev->driver->move_notify(bo, evict, mem);
>> if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
>> !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
>> ret = ttm_bo_move_ttm(bo, interruptible, no_wait_gpu, mem);
>> else if (bdev->driver->move)
>> ret = bdev->driver->move(bo, evict, interruptible,
>> no_wait_gpu, mem);
>> else
>> ret = ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu, mem);
>> if (ret) {
>> if (bdev->driver->move_notify) {
>> struct ttm_mem_reg tmp_mem = *mem;
>> *mem = bo->mem;
>> bo->mem = tmp_mem;
>> - bdev->driver->move_notify(bo, mem);
>> + bdev->driver->move_notify(bo, false, mem);
>> bo->mem = *mem;
>> *mem = tmp_mem;
>> }
>> goto out_err;
>> }
>> moved:
>> if (bo->evicted) {
>> if (bdev->driver->invalidate_caches) {
>> @@ -407,21 +407,21 @@ static int ttm_bo_handle_move_mem(struct
>> ttm_buffer_object *bo,
>> * Call bo::reserved.
>> * Will release GPU memory type usage on destruction.
>> * This is the place to put in driver specific hooks to release
>> * driver private resources.
>> * Will release the bo::reserved lock.
>> */
>> static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
>> {
>> if (bo->bdev->driver->move_notify)
>> - bo->bdev->driver->move_notify(bo, NULL);
>> + bo->bdev->driver->move_notify(bo, false, NULL);
>> ttm_tt_destroy(bo->ttm);
>> bo->ttm = NULL;
>> ttm_bo_mem_put(bo, &bo->mem);
>> ww_mutex_unlock (&bo->resv->lock);
>> }
>> static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
>> {
>> diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c
>> b/drivers/gpu/drm/virtio/virtgpu_ttm.c
>> index 4a1de9f..10387d7 100644
>> --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
>> +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
>> @@ -379,20 +379,21 @@ static int virtio_gpu_bo_move(struct
>> ttm_buffer_object *bo,
>> ret = ttm_bo_wait(bo, interruptible, no_wait_gpu);
>> if (ret)
>> return ret;
>> virtio_gpu_move_null(bo, new_mem);
>> return 0;
>> }
>> static void virtio_gpu_bo_move_notify(struct ttm_buffer_object *tbo,
>> + bool evict,
>> struct ttm_mem_reg *new_mem)
>> {
>> struct virtio_gpu_object *bo;
>> struct virtio_gpu_device *vgdev;
>> bo = container_of(tbo, struct virtio_gpu_object, tbo);
>> vgdev = (struct virtio_gpu_device *)bo->gem_base.dev->dev_private;
>> if (!new_mem || (new_mem->placement & TTM_PL_FLAG_SYSTEM)) {
>> if (bo->hw_res_handle)
>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>> b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>> index c894a48..caa279b 100644
>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>> @@ -818,20 +818,21 @@ static int vmw_ttm_fault_reserve_notify(struct
>> ttm_buffer_object *bo)
>> * vmw_move_notify - TTM move_notify_callback
>> *
>> * @bo: The TTM buffer object about to move.
>> * @mem: The struct ttm_mem_reg indicating to what memory
>> * region the move is taking place.
>> *
>> * Calls move_notify for all subsystems needing it.
>> * (currently only resources).
>> */
>> static void vmw_move_notify(struct ttm_buffer_object *bo,
>> + bool evict,
>> struct ttm_mem_reg *mem)
>> {
>> vmw_resource_move_notify(bo, mem);
>> vmw_query_move_notify(bo, mem);
>> }
>> /**
>> * vmw_swap_notify - TTM move_notify_callback
>> *
>> diff --git a/include/drm/ttm/ttm_bo_driver.h
>> b/include/drm/ttm/ttm_bo_driver.h
>> index cdbdb40..8b4bec7 100644
>> --- a/include/drm/ttm/ttm_bo_driver.h
>> +++ b/include/drm/ttm/ttm_bo_driver.h
>> @@ -424,23 +424,29 @@ struct ttm_bo_driver {
>> *
>> * Called from the map / write / read methods to verify that the
>> * caller is permitted to access the buffer object.
>> * This member may be set to NULL, which will refuse this kind of
>> * access for all buffer objects.
>> * This function should return 0 if access is granted, -EPERM
>> otherwise.
>> */
>> int (*verify_access)(struct ttm_buffer_object *bo,
>> struct file *filp);
>> - /* hook to notify driver about a driver move so it
>> - * can do tiling things */
>> + /**
>> + * Hook to notify driver about a driver move so it
>> + * can do tiling things and book-keeping.
>> + *
>> + * @evict: whether this move is evicting the buffer from the
>> graphics
>> + * address space
>> + */
>> void (*move_notify)(struct ttm_buffer_object *bo,
>> + bool evict,
>> struct ttm_mem_reg *new_mem);
>> /* notify the driver we are taking a fault on this BO
>> * and have reserved it */
>> int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
>> /**
>> * notify the driver that we're about to swap out this bo
>> */
>> void (*swap_notify)(struct ttm_buffer_object *bo);
>>
>
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify
[not found] ` <8f68adf1-52bc-b2ef-2fa7-79a8f9695164-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
@ 2017-01-03 11:42 ` Christian König
0 siblings, 0 replies; 15+ messages in thread
From: Christian König @ 2017-01-03 11:42 UTC (permalink / raw)
To: Nicolai Hähnle, zhoucm1, amd-gfx-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Cc: Nicolai Hähnle, dri-devel-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW
Am 21.12.2016 um 16:12 schrieb Nicolai Hähnle:
> On 16.12.2016 03:49, zhoucm1 wrote:
>> On 2016年12月16日 01:10, Nicolai Hähnle wrote:
>>> From: Nicolai Hähnle <nicolai.haehnle@amd.com>
>>>
>>> Ensure that the driver can listen to evictions even when they don't
>>> take the
>>> path through ttm_bo_driver::move.
>>>
>>> This is crucial for amdgpu, which relies on an eviction counter to skip
>>> re-binding page tables when possible.
>>>
>>> Signed-off-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
>> Acked-by: Chunming Zhou <david1.zhou@amd.com>
Feel free to add my Reviewed-by: Christian König
<christian.koenig@amd.com> to patch #1-#4 and V2 of patch #5 as well.
>
> Thanks. Ping for feedback from non-AMD people?
Ping once more. Would be nice if somebody else can take a look as well.
Christian.
>
> Nicolai
>
>
>>> ---
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 1 +
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 3 ++-
>>> drivers/gpu/drm/nouveau/nouveau_bo.c | 3 ++-
>>> drivers/gpu/drm/qxl/qxl_ttm.c | 1 +
>>> drivers/gpu/drm/radeon/radeon_object.c | 1 +
>>> drivers/gpu/drm/radeon/radeon_object.h | 1 +
>>> drivers/gpu/drm/ttm/ttm_bo.c | 8 ++++----
>>> drivers/gpu/drm/virtio/virtgpu_ttm.c | 1 +
>>> drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c | 1 +
>>> include/drm/ttm/ttm_bo_driver.h | 10 ++++++++--
>>> 10 files changed, 22 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> index bf79b73..c29db99 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
>>> @@ -842,20 +842,21 @@ int amdgpu_bo_get_metadata(struct amdgpu_bo *bo,
>>> void *buffer,
>>> if (metadata_size)
>>> *metadata_size = bo->metadata_size;
>>> if (flags)
>>> *flags = bo->metadata_flags;
>>> return 0;
>>> }
>>> void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>>> + bool evict,
>>> struct ttm_mem_reg *new_mem)
>>> {
>>> struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
>>> struct amdgpu_bo *abo;
>>> struct ttm_mem_reg *old_mem = &bo->mem;
>>> if (!amdgpu_ttm_bo_is_amdgpu_bo(bo))
>>> return;
>>> abo = container_of(bo, struct amdgpu_bo, tbo);
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>>> index 5cbf59e..4306b2f 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
>>> @@ -148,21 +148,22 @@ void amdgpu_bo_fini(struct amdgpu_device *adev);
>>> int amdgpu_bo_fbdev_mmap(struct amdgpu_bo *bo,
>>> struct vm_area_struct *vma);
>>> int amdgpu_bo_set_tiling_flags(struct amdgpu_bo *bo, u64
>>> tiling_flags);
>>> void amdgpu_bo_get_tiling_flags(struct amdgpu_bo *bo, u64
>>> *tiling_flags);
>>> int amdgpu_bo_set_metadata (struct amdgpu_bo *bo, void *metadata,
>>> uint32_t metadata_size, uint64_t flags);
>>> int amdgpu_bo_get_metadata(struct amdgpu_bo *bo, void *buffer,
>>> size_t buffer_size, uint32_t *metadata_size,
>>> uint64_t *flags);
>>> void amdgpu_bo_move_notify(struct ttm_buffer_object *bo,
>>> - struct ttm_mem_reg *new_mem);
>>> + bool evict,
>>> + struct ttm_mem_reg *new_mem);
>>> int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo);
>>> void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence,
>>> bool shared);
>>> u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo);
>>> int amdgpu_bo_backup_to_shadow(struct amdgpu_device *adev,
>>> struct amdgpu_ring *ring,
>>> struct amdgpu_bo *bo,
>>> struct reservation_object *resv,
>>> struct dma_fence **fence, bool direct);
>>> int amdgpu_bo_restore_from_shadow(struct amdgpu_device *adev,
>>> diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> index e0c0007..6fa1521 100644
>>> --- a/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> +++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
>>> @@ -1187,21 +1187,22 @@ nouveau_bo_move_flips(struct ttm_buffer_object
>>> *bo, bool evict, bool intr,
>>> ret = nouveau_bo_move_m2mf(bo, true, intr, no_wait_gpu, new_mem);
>>> if (ret)
>>> goto out;
>>> out:
>>> ttm_bo_mem_put(bo, &tmp_mem);
>>> return ret;
>>> }
>>> static void
>>> -nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, struct ttm_mem_reg
>>> *new_mem)
>>> +nouveau_bo_move_ntfy(struct ttm_buffer_object *bo, bool evict,
>>> + struct ttm_mem_reg *new_mem)
>>> {
>>> struct nouveau_bo *nvbo = nouveau_bo(bo);
>>> struct nvkm_vma *vma;
>>> /* ttm can now (stupidly) pass the driver bos it didn't
>>> create... */
>>> if (bo->destroy != nouveau_bo_del_ttm)
>>> return;
>>> list_for_each_entry(vma, &nvbo->vma_list, head) {
>>> if (new_mem && new_mem->mem_type != TTM_PL_SYSTEM &&
>>> diff --git a/drivers/gpu/drm/qxl/qxl_ttm.c
>>> b/drivers/gpu/drm/qxl/qxl_ttm.c
>>> index 1176133..f3939a9 100644
>>> --- a/drivers/gpu/drm/qxl/qxl_ttm.c
>>> +++ b/drivers/gpu/drm/qxl/qxl_ttm.c
>>> @@ -360,20 +360,21 @@ static int qxl_bo_move(struct ttm_buffer_object
>>> *bo,
>>> if (old_mem->mem_type == TTM_PL_SYSTEM && bo->ttm == NULL) {
>>> qxl_move_null(bo, new_mem);
>>> return 0;
>>> }
>>> return ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
>>> new_mem);
>>> }
>>> static void qxl_bo_move_notify(struct ttm_buffer_object *bo,
>>> + bool evict,
>>> struct ttm_mem_reg *new_mem)
>>> {
>>> struct qxl_bo *qbo;
>>> struct qxl_device *qdev;
>>> if (!qxl_ttm_bo_is_qxl_bo(bo))
>>> return;
>>> qbo = to_qxl_bo(bo);
>>> qdev = qbo->gem_base.dev->dev_private;
>>> diff --git a/drivers/gpu/drm/radeon/radeon_object.c
>>> b/drivers/gpu/drm/radeon/radeon_object.c
>>> index 41b72ce..74b2760 100644
>>> --- a/drivers/gpu/drm/radeon/radeon_object.c
>>> +++ b/drivers/gpu/drm/radeon/radeon_object.c
>>> @@ -758,20 +758,21 @@ int radeon_bo_check_tiling(struct radeon_bo *bo,
>>> bool has_moved,
>>> return 0;
>>> }
>>> if ((bo->surface_reg >= 0) && !has_moved)
>>> return 0;
>>> return radeon_bo_get_surface_reg(bo);
>>> }
>>> void radeon_bo_move_notify(struct ttm_buffer_object *bo,
>>> + bool evict,
>>> struct ttm_mem_reg *new_mem)
>>> {
>>> struct radeon_bo *rbo;
>>> if (!radeon_ttm_bo_is_radeon_bo(bo))
>>> return;
>>> rbo = container_of(bo, struct radeon_bo, tbo);
>>> radeon_bo_check_tiling(rbo, 0, 1);
>>> radeon_vm_bo_invalidate(rbo->rdev, rbo);
>>> diff --git a/drivers/gpu/drm/radeon/radeon_object.h
>>> b/drivers/gpu/drm/radeon/radeon_object.h
>>> index a10bb3d..9ffd821 100644
>>> --- a/drivers/gpu/drm/radeon/radeon_object.h
>>> +++ b/drivers/gpu/drm/radeon/radeon_object.h
>>> @@ -143,20 +143,21 @@ extern void radeon_bo_fini(struct radeon_device
>>> *rdev);
>>> extern int radeon_bo_list_validate(struct radeon_device *rdev,
>>> struct ww_acquire_ctx *ticket,
>>> struct list_head *head, int ring);
>>> extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
>>> u32 tiling_flags, u32 pitch);
>>> extern void radeon_bo_get_tiling_flags(struct radeon_bo *bo,
>>> u32 *tiling_flags, u32 *pitch);
>>> extern int radeon_bo_check_tiling(struct radeon_bo *bo, bool
>>> has_moved,
>>> bool force_drop);
>>> extern void radeon_bo_move_notify(struct ttm_buffer_object *bo,
>>> + bool evict,
>>> struct ttm_mem_reg *new_mem);
>>> extern int radeon_bo_fault_reserve_notify(struct ttm_buffer_object
>>> *bo);
>>> extern int radeon_bo_get_surface_reg(struct radeon_bo *bo);
>>> extern void radeon_bo_fence(struct radeon_bo *bo, struct
>>> radeon_fence *fence,
>>> bool shared);
>>> /*
>>> * sub allocation
>>> */
>>> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c
>>> b/drivers/gpu/drm/ttm/ttm_bo.c
>>> index d506361..d4973e9c 100644
>>> --- a/drivers/gpu/drm/ttm/ttm_bo.c
>>> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
>>> @@ -335,45 +335,45 @@ static int ttm_bo_handle_move_mem(struct
>>> ttm_buffer_object *bo,
>>> goto out_err;
>>> if (mem->mem_type != TTM_PL_SYSTEM) {
>>> ret = ttm_tt_bind(bo->ttm, mem);
>>> if (ret)
>>> goto out_err;
>>> }
>>> if (bo->mem.mem_type == TTM_PL_SYSTEM) {
>>> if (bdev->driver->move_notify)
>>> - bdev->driver->move_notify(bo, mem);
>>> + bdev->driver->move_notify(bo, evict, mem);
>>> bo->mem = *mem;
>>> mem->mm_node = NULL;
>>> goto moved;
>>> }
>>> }
>>> if (bdev->driver->move_notify)
>>> - bdev->driver->move_notify(bo, mem);
>>> + bdev->driver->move_notify(bo, evict, mem);
>>> if (!(old_man->flags & TTM_MEMTYPE_FLAG_FIXED) &&
>>> !(new_man->flags & TTM_MEMTYPE_FLAG_FIXED))
>>> ret = ttm_bo_move_ttm(bo, interruptible, no_wait_gpu, mem);
>>> else if (bdev->driver->move)
>>> ret = bdev->driver->move(bo, evict, interruptible,
>>> no_wait_gpu, mem);
>>> else
>>> ret = ttm_bo_move_memcpy(bo, interruptible, no_wait_gpu,
>>> mem);
>>> if (ret) {
>>> if (bdev->driver->move_notify) {
>>> struct ttm_mem_reg tmp_mem = *mem;
>>> *mem = bo->mem;
>>> bo->mem = tmp_mem;
>>> - bdev->driver->move_notify(bo, mem);
>>> + bdev->driver->move_notify(bo, false, mem);
>>> bo->mem = *mem;
>>> *mem = tmp_mem;
>>> }
>>> goto out_err;
>>> }
>>> moved:
>>> if (bo->evicted) {
>>> if (bdev->driver->invalidate_caches) {
>>> @@ -407,21 +407,21 @@ static int ttm_bo_handle_move_mem(struct
>>> ttm_buffer_object *bo,
>>> * Call bo::reserved.
>>> * Will release GPU memory type usage on destruction.
>>> * This is the place to put in driver specific hooks to release
>>> * driver private resources.
>>> * Will release the bo::reserved lock.
>>> */
>>> static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object
>>> *bo)
>>> {
>>> if (bo->bdev->driver->move_notify)
>>> - bo->bdev->driver->move_notify(bo, NULL);
>>> + bo->bdev->driver->move_notify(bo, false, NULL);
>>> ttm_tt_destroy(bo->ttm);
>>> bo->ttm = NULL;
>>> ttm_bo_mem_put(bo, &bo->mem);
>>> ww_mutex_unlock (&bo->resv->lock);
>>> }
>>> static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
>>> {
>>> diff --git a/drivers/gpu/drm/virtio/virtgpu_ttm.c
>>> b/drivers/gpu/drm/virtio/virtgpu_ttm.c
>>> index 4a1de9f..10387d7 100644
>>> --- a/drivers/gpu/drm/virtio/virtgpu_ttm.c
>>> +++ b/drivers/gpu/drm/virtio/virtgpu_ttm.c
>>> @@ -379,20 +379,21 @@ static int virtio_gpu_bo_move(struct
>>> ttm_buffer_object *bo,
>>> ret = ttm_bo_wait(bo, interruptible, no_wait_gpu);
>>> if (ret)
>>> return ret;
>>> virtio_gpu_move_null(bo, new_mem);
>>> return 0;
>>> }
>>> static void virtio_gpu_bo_move_notify(struct ttm_buffer_object
>>> *tbo,
>>> + bool evict,
>>> struct ttm_mem_reg *new_mem)
>>> {
>>> struct virtio_gpu_object *bo;
>>> struct virtio_gpu_device *vgdev;
>>> bo = container_of(tbo, struct virtio_gpu_object, tbo);
>>> vgdev = (struct virtio_gpu_device
>>> *)bo->gem_base.dev->dev_private;
>>> if (!new_mem || (new_mem->placement & TTM_PL_FLAG_SYSTEM)) {
>>> if (bo->hw_res_handle)
>>> diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> index c894a48..caa279b 100644
>>> --- a/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> +++ b/drivers/gpu/drm/vmwgfx/vmwgfx_buffer.c
>>> @@ -818,20 +818,21 @@ static int vmw_ttm_fault_reserve_notify(struct
>>> ttm_buffer_object *bo)
>>> * vmw_move_notify - TTM move_notify_callback
>>> *
>>> * @bo: The TTM buffer object about to move.
>>> * @mem: The struct ttm_mem_reg indicating to what memory
>>> * region the move is taking place.
>>> *
>>> * Calls move_notify for all subsystems needing it.
>>> * (currently only resources).
>>> */
>>> static void vmw_move_notify(struct ttm_buffer_object *bo,
>>> + bool evict,
>>> struct ttm_mem_reg *mem)
>>> {
>>> vmw_resource_move_notify(bo, mem);
>>> vmw_query_move_notify(bo, mem);
>>> }
>>> /**
>>> * vmw_swap_notify - TTM move_notify_callback
>>> *
>>> diff --git a/include/drm/ttm/ttm_bo_driver.h
>>> b/include/drm/ttm/ttm_bo_driver.h
>>> index cdbdb40..8b4bec7 100644
>>> --- a/include/drm/ttm/ttm_bo_driver.h
>>> +++ b/include/drm/ttm/ttm_bo_driver.h
>>> @@ -424,23 +424,29 @@ struct ttm_bo_driver {
>>> *
>>> * Called from the map / write / read methods to verify that the
>>> * caller is permitted to access the buffer object.
>>> * This member may be set to NULL, which will refuse this kind of
>>> * access for all buffer objects.
>>> * This function should return 0 if access is granted, -EPERM
>>> otherwise.
>>> */
>>> int (*verify_access)(struct ttm_buffer_object *bo,
>>> struct file *filp);
>>> - /* hook to notify driver about a driver move so it
>>> - * can do tiling things */
>>> + /**
>>> + * Hook to notify driver about a driver move so it
>>> + * can do tiling things and book-keeping.
>>> + *
>>> + * @evict: whether this move is evicting the buffer from the
>>> graphics
>>> + * address space
>>> + */
>>> void (*move_notify)(struct ttm_buffer_object *bo,
>>> + bool evict,
>>> struct ttm_mem_reg *new_mem);
>>> /* notify the driver we are taking a fault on this BO
>>> * and have reserved it */
>>> int (*fault_reserve_notify)(struct ttm_buffer_object *bo);
>>> /**
>>> * notify the driver that we're about to swap out this bo
>>> */
>>> void (*swap_notify)(struct ttm_buffer_object *bo);
>>>
>>
> _______________________________________________
> amd-gfx mailing list
> amd-gfx@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/amd-gfx
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
^ permalink raw reply [flat|nested] 15+ messages in thread
end of thread, other threads:[~2017-01-03 11:42 UTC | newest]
Thread overview: 15+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-12-15 17:10 [PATCH 0/5] drm/ttm, amdgpu: fix crashes due to shadow page table evictions Nicolai Hähnle
[not found] ` <1481821815-9971-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:10 ` [PATCH 1/5] drm/ttm: add evict parameter to ttm_bo_driver::move_notify Nicolai Hähnle
[not found] ` <1481821815-9971-2-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-16 2:49 ` zhoucm1
[not found] ` <5853564E.5020005-5C7GfCeVMHo@public.gmane.org>
2016-12-21 15:12 ` Nicolai Hähnle
[not found] ` <8f68adf1-52bc-b2ef-2fa7-79a8f9695164-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2017-01-03 11:42 ` Christian König
2016-12-15 17:10 ` [PATCH 2/5] drm/amd/amdgpu: move eviction counting to amdgpu_bo_move_notify Nicolai Hähnle
[not found] ` <1481821815-9971-3-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-16 2:50 ` zhoucm1
2016-12-15 17:10 ` [PATCH 3/5] drm/amd/amdgpu: add amdgpu_bo_gpu_accessible helper function Nicolai Hähnle
[not found] ` <1481821815-9971-4-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-16 2:48 ` zhoucm1
2016-12-15 17:10 ` [PATCH 4/5] drm/amd/amdgpu: add check that shadow page directory is GPU-accessible Nicolai Hähnle
[not found] ` <1481821815-9971-5-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-16 2:49 ` zhoucm1
2016-12-15 17:10 ` [PATCH 5/5] drm/amd/amdgpu: add check that shadow page tables are GPU-accessible Nicolai Hähnle
[not found] ` <1481821815-9971-6-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-15 17:59 ` [PATCH v2] " Nicolai Hähnle
[not found] ` <1481824743-11139-1-git-send-email-nhaehnle-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
2016-12-16 2:48 ` zhoucm1
2016-12-16 2:07 ` [PATCH 5/5] " zhoucm1
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.