* [PATCH v3 00/17] drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Previously, short term pinning in execbuf was required because i915_vma was
effectively independent from objects, and has its own refcount, locking,
lifetime rules and pinning.
This series removes the separate locking, by requiring vma->obj->resv to be
held when pinning and unbinding. This will also be required for VM_BIND work.
With pinning required for pinning and unbinding, the lock is enough to prevent
unbinding when trying to pin with the lock held, for example in execbuf.
This makes binding/unbinding similar to ttm_bo_validate()'s use, which just
cares that an object is in a certain place, without pinning it in place.
Having the VMA part of gem bo removes a lot of the vma refcounting, and makes
i915_vma more a part of the bo, instead of its own floating object that just
happens to be part of a bo. This is also required to make it more compatible
with TTM, and migration in general.
For future work, it makes things a lot simpler and clear. We want to end up
with i915_vma just being a specific mapping of the BO, just like is the
case in other drivers. i915_vma->active removal is the next step there,
and makes it when object is destroyed, the bindings are destroyed (after idle),
instead of object being destroyed when bindings are idle.
---
This is version 3, I changed the ordering of the patches slightly to be more
natural, and divided up some commits in different ways.
In particular, I added the ww ctx to trylock first, then converted
i915_gem_evict_vm(), the shrinker, and the other eviction functions all
in separate commits. Functionally the same, but easier to review and bisect.
In theory, it was already reviewed, but leaving those r-b's out as they're
technically different patches now.
The first patches prepare for vma->obj->resv requirement, by ensuring that we
unbind as needed, and take the ww lock in some places where they're required,
including when freeing the object.
After that we convert the shrinker, eviction functions, and then we can start
requiring the object lock for i915_vma_unbind.
Now that the conversion and requirement is added, we can remove ther special
case of assert_object_held_shared, which was added because we didn't always
require the object lock.
As last step we can remove the short term pins from execbuf. We still pin when
evicting the entire VM, but if it turns out to be an issue, it can be removed,
and replaced with unpinning the entire VM.
Maarten Lankhorst (17):
drm/i915: Remove unused bits of i915_vma/active api
drm/i915: Change shrink ordering to use locking around unbinding.
drm/i915: Remove pages_mutex and intel_gtt->vma_ops.set/clear_pages
members, v3.
drm/i915: Take object lock in i915_ggtt_pin if ww is not set
drm/i915: Force ww lock for i915_gem_object_ggtt_pin_ww, v2.
drm/i915: Ensure gem_contexts selftests work with unbind changes, v2.
drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking
changes.
drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC
errors
drm/i915: Trylock the object when shrinking
drm/i915: Require object lock when freeing pages during destruction
drm/i915: Add ww ctx to i915_gem_object_trylock
drm/i915: Add locking to i915_gem_evict_vm()
drm/i915: Add object locking to i915_gem_evict_for_node and
i915_gem_evict_something
drm/i915: Add i915_vma_unbind_unlocked, and take obj lock for
i915_vma_unbind, v2.
drm/i915: Remove assert_object_held_shared
drm/i915: Remove support for unlocked i915_vma unbind
drm/i915: Remove short-term pins from execbuf, v5.
drivers/gpu/drm/i915/display/intel_dpt.c | 2 -
drivers/gpu/drm/i915/display/intel_fb_pin.c | 2 +-
.../gpu/drm/i915/gem/i915_gem_execbuffer.c | 250 ++++----
drivers/gpu/drm/i915/gem/i915_gem_mman.c | 18 +-
drivers/gpu/drm/i915/gem/i915_gem_object.c | 9 +-
drivers/gpu/drm/i915/gem/i915_gem_object.h | 22 +-
drivers/gpu/drm/i915/gem/i915_gem_pages.c | 12 +-
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 44 +-
drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 2 +-
drivers/gpu/drm/i915/gem/i915_gem_userptr.c | 2 +-
.../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +-
.../i915/gem/selftests/i915_gem_client_blt.c | 2 +-
.../drm/i915/gem/selftests/i915_gem_context.c | 59 +-
.../drm/i915/gem/selftests/i915_gem_mman.c | 6 +
drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 15 -
drivers/gpu/drm/i915/gt/intel_engine_pm.c | 2 +-
drivers/gpu/drm/i915/gt/intel_ggtt.c | 454 ++------------
drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c | 1 -
drivers/gpu/drm/i915/gt/intel_gtt.c | 13 -
drivers/gpu/drm/i915/gt/intel_gtt.h | 7 -
drivers/gpu/drm/i915/gt/intel_ppgtt.c | 12 -
drivers/gpu/drm/i915/gt/mock_engine.c | 2 +-
drivers/gpu/drm/i915/gt/selftest_hangcheck.c | 2 +-
drivers/gpu/drm/i915/gt/selftest_migrate.c | 2 +-
drivers/gpu/drm/i915/gvt/aperture_gm.c | 2 +-
drivers/gpu/drm/i915/i915_active.c | 28 +-
drivers/gpu/drm/i915/i915_active.h | 17 +-
drivers/gpu/drm/i915/i915_drv.h | 12 +-
drivers/gpu/drm/i915/i915_gem.c | 32 +-
drivers/gpu/drm/i915/i915_gem_evict.c | 64 +-
drivers/gpu/drm/i915/i915_gem_gtt.c | 8 +-
drivers/gpu/drm/i915/i915_gem_gtt.h | 4 +
drivers/gpu/drm/i915/i915_vgpu.c | 2 +-
drivers/gpu/drm/i915/i915_vma.c | 579 +++++++++++++++---
drivers/gpu/drm/i915/i915_vma.h | 6 +-
drivers/gpu/drm/i915/i915_vma_types.h | 1 -
.../gpu/drm/i915/selftests/i915_gem_evict.c | 27 +-
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 50 +-
drivers/gpu/drm/i915/selftests/i915_vma.c | 19 +-
drivers/gpu/drm/i915/selftests/mock_gtt.c | 4 -
40 files changed, 951 insertions(+), 846 deletions(-)
--
2.34.1
^ permalink raw reply [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 00/17] drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Previously, short term pinning in execbuf was required because i915_vma was
effectively independent from objects, and has its own refcount, locking,
lifetime rules and pinning.
This series removes the separate locking, by requiring vma->obj->resv to be
held when pinning and unbinding. This will also be required for VM_BIND work.
With pinning required for pinning and unbinding, the lock is enough to prevent
unbinding when trying to pin with the lock held, for example in execbuf.
This makes binding/unbinding similar to ttm_bo_validate()'s use, which just
cares that an object is in a certain place, without pinning it in place.
Having the VMA part of gem bo removes a lot of the vma refcounting, and makes
i915_vma more a part of the bo, instead of its own floating object that just
happens to be part of a bo. This is also required to make it more compatible
with TTM, and migration in general.
For future work, it makes things a lot simpler and clear. We want to end up
with i915_vma just being a specific mapping of the BO, just like is the
case in other drivers. i915_vma->active removal is the next step there,
and makes it when object is destroyed, the bindings are destroyed (after idle),
instead of object being destroyed when bindings are idle.
---
This is version 3, I changed the ordering of the patches slightly to be more
natural, and divided up some commits in different ways.
In particular, I added the ww ctx to trylock first, then converted
i915_gem_evict_vm(), the shrinker, and the other eviction functions all
in separate commits. Functionally the same, but easier to review and bisect.
In theory, it was already reviewed, but leaving those r-b's out as they're
technically different patches now.
The first patches prepare for vma->obj->resv requirement, by ensuring that we
unbind as needed, and take the ww lock in some places where they're required,
including when freeing the object.
After that we convert the shrinker, eviction functions, and then we can start
requiring the object lock for i915_vma_unbind.
Now that the conversion and requirement is added, we can remove ther special
case of assert_object_held_shared, which was added because we didn't always
require the object lock.
As last step we can remove the short term pins from execbuf. We still pin when
evicting the entire VM, but if it turns out to be an issue, it can be removed,
and replaced with unpinning the entire VM.
Maarten Lankhorst (17):
drm/i915: Remove unused bits of i915_vma/active api
drm/i915: Change shrink ordering to use locking around unbinding.
drm/i915: Remove pages_mutex and intel_gtt->vma_ops.set/clear_pages
members, v3.
drm/i915: Take object lock in i915_ggtt_pin if ww is not set
drm/i915: Force ww lock for i915_gem_object_ggtt_pin_ww, v2.
drm/i915: Ensure gem_contexts selftests work with unbind changes, v2.
drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking
changes.
drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC
errors
drm/i915: Trylock the object when shrinking
drm/i915: Require object lock when freeing pages during destruction
drm/i915: Add ww ctx to i915_gem_object_trylock
drm/i915: Add locking to i915_gem_evict_vm()
drm/i915: Add object locking to i915_gem_evict_for_node and
i915_gem_evict_something
drm/i915: Add i915_vma_unbind_unlocked, and take obj lock for
i915_vma_unbind, v2.
drm/i915: Remove assert_object_held_shared
drm/i915: Remove support for unlocked i915_vma unbind
drm/i915: Remove short-term pins from execbuf, v5.
drivers/gpu/drm/i915/display/intel_dpt.c | 2 -
drivers/gpu/drm/i915/display/intel_fb_pin.c | 2 +-
.../gpu/drm/i915/gem/i915_gem_execbuffer.c | 250 ++++----
drivers/gpu/drm/i915/gem/i915_gem_mman.c | 18 +-
drivers/gpu/drm/i915/gem/i915_gem_object.c | 9 +-
drivers/gpu/drm/i915/gem/i915_gem_object.h | 22 +-
drivers/gpu/drm/i915/gem/i915_gem_pages.c | 12 +-
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 44 +-
drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 2 +-
drivers/gpu/drm/i915/gem/i915_gem_userptr.c | 2 +-
.../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +-
.../i915/gem/selftests/i915_gem_client_blt.c | 2 +-
.../drm/i915/gem/selftests/i915_gem_context.c | 59 +-
.../drm/i915/gem/selftests/i915_gem_mman.c | 6 +
drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 15 -
drivers/gpu/drm/i915/gt/intel_engine_pm.c | 2 +-
drivers/gpu/drm/i915/gt/intel_ggtt.c | 454 ++------------
drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c | 1 -
drivers/gpu/drm/i915/gt/intel_gtt.c | 13 -
drivers/gpu/drm/i915/gt/intel_gtt.h | 7 -
drivers/gpu/drm/i915/gt/intel_ppgtt.c | 12 -
drivers/gpu/drm/i915/gt/mock_engine.c | 2 +-
drivers/gpu/drm/i915/gt/selftest_hangcheck.c | 2 +-
drivers/gpu/drm/i915/gt/selftest_migrate.c | 2 +-
drivers/gpu/drm/i915/gvt/aperture_gm.c | 2 +-
drivers/gpu/drm/i915/i915_active.c | 28 +-
drivers/gpu/drm/i915/i915_active.h | 17 +-
drivers/gpu/drm/i915/i915_drv.h | 12 +-
drivers/gpu/drm/i915/i915_gem.c | 32 +-
drivers/gpu/drm/i915/i915_gem_evict.c | 64 +-
drivers/gpu/drm/i915/i915_gem_gtt.c | 8 +-
drivers/gpu/drm/i915/i915_gem_gtt.h | 4 +
drivers/gpu/drm/i915/i915_vgpu.c | 2 +-
drivers/gpu/drm/i915/i915_vma.c | 579 +++++++++++++++---
drivers/gpu/drm/i915/i915_vma.h | 6 +-
drivers/gpu/drm/i915/i915_vma_types.h | 1 -
.../gpu/drm/i915/selftests/i915_gem_evict.c | 27 +-
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 50 +-
drivers/gpu/drm/i915/selftests/i915_vma.c | 19 +-
drivers/gpu/drm/i915/selftests/mock_gtt.c | 4 -
40 files changed, 951 insertions(+), 846 deletions(-)
--
2.34.1
^ permalink raw reply [flat|nested] 57+ messages in thread
* [PATCH v3 01/17] drm/i915: Remove unused bits of i915_vma/active api
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Niranjana Vishwanathapura, dri-devel
When reworking the code to move the eviction fence to the object,
the best code is removed code.
Remove some functions that are unused, and change the function definition
if it's only used in 1 place.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
[mlankhorst: Remove new use of i915_active_has_exclusive]
---
drivers/gpu/drm/i915/i915_active.c | 28 +++-------------------------
drivers/gpu/drm/i915/i915_active.h | 17 +----------------
drivers/gpu/drm/i915/i915_vma.c | 24 ++++++++++--------------
drivers/gpu/drm/i915/i915_vma.h | 2 --
4 files changed, 14 insertions(+), 57 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
index 3103c1e1fd14..ee2b3a375362 100644
--- a/drivers/gpu/drm/i915/i915_active.c
+++ b/drivers/gpu/drm/i915/i915_active.c
@@ -426,8 +426,9 @@ replace_barrier(struct i915_active *ref, struct i915_active_fence *active)
return true;
}
-int i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence)
+int i915_active_add_request(struct i915_active *ref, struct i915_request *rq)
{
+ struct dma_fence *fence = &rq->fence;
struct i915_active_fence *active;
int err;
@@ -436,7 +437,7 @@ int i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence)
if (err)
return err;
- active = active_instance(ref, idx);
+ active = active_instance(ref, i915_request_timeline(rq)->fence_context);
if (!active) {
err = -ENOMEM;
goto out;
@@ -477,29 +478,6 @@ __i915_active_set_fence(struct i915_active *ref,
return prev;
}
-static struct i915_active_fence *
-__active_fence(struct i915_active *ref, u64 idx)
-{
- struct active_node *it;
-
- it = __active_lookup(ref, idx);
- if (unlikely(!it)) { /* Contention with parallel tree builders! */
- spin_lock_irq(&ref->tree_lock);
- it = __active_lookup(ref, idx);
- spin_unlock_irq(&ref->tree_lock);
- }
- GEM_BUG_ON(!it); /* slot must be preallocated */
-
- return &it->base;
-}
-
-struct dma_fence *
-__i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence)
-{
- /* Only valid while active, see i915_active_acquire_for_context() */
- return __i915_active_set_fence(ref, __active_fence(ref, idx), fence);
-}
-
struct dma_fence *
i915_active_set_exclusive(struct i915_active *ref, struct dma_fence *f)
{
diff --git a/drivers/gpu/drm/i915/i915_active.h b/drivers/gpu/drm/i915/i915_active.h
index 5fcdb0e2bc9e..7eb44132183a 100644
--- a/drivers/gpu/drm/i915/i915_active.h
+++ b/drivers/gpu/drm/i915/i915_active.h
@@ -164,26 +164,11 @@ void __i915_active_init(struct i915_active *ref,
__i915_active_init(ref, active, retire, flags, &__mkey, &__wkey); \
} while (0)
-struct dma_fence *
-__i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence);
-int i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence);
-
-static inline int
-i915_active_add_request(struct i915_active *ref, struct i915_request *rq)
-{
- return i915_active_ref(ref,
- i915_request_timeline(rq)->fence_context,
- &rq->fence);
-}
+int i915_active_add_request(struct i915_active *ref, struct i915_request *rq);
struct dma_fence *
i915_active_set_exclusive(struct i915_active *ref, struct dma_fence *f);
-static inline bool i915_active_has_exclusive(struct i915_active *ref)
-{
- return rcu_access_pointer(ref->excl.fence);
-}
-
int __i915_active_wait(struct i915_active *ref, int state);
static inline int i915_active_wait(struct i915_active *ref)
{
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 927f0d4f8e11..921d5b946c32 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -356,22 +356,18 @@ int i915_vma_wait_for_bind(struct i915_vma *vma)
#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)
static int i915_vma_verify_bind_complete(struct i915_vma *vma)
{
- int err = 0;
-
- if (i915_active_has_exclusive(&vma->active)) {
- struct dma_fence *fence =
- i915_active_fence_get(&vma->active.excl);
+ struct dma_fence *fence = i915_active_fence_get(&vma->active.excl);
+ int err;
- if (!fence)
- return 0;
+ if (!fence)
+ return 0;
- if (dma_fence_is_signaled(fence))
- err = fence->error;
- else
- err = -EBUSY;
+ if (dma_fence_is_signaled(fence))
+ err = fence->error;
+ else
+ err = -EBUSY;
- dma_fence_put(fence);
- }
+ dma_fence_put(fence);
return err;
}
@@ -1249,7 +1245,7 @@ __i915_request_await_bind(struct i915_request *rq, struct i915_vma *vma)
return __i915_request_await_exclusive(rq, &vma->active);
}
-int __i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *rq)
+static int __i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *rq)
{
int err;
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 4033aa08d5e4..9a931ecb09e5 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -55,8 +55,6 @@ static inline bool i915_vma_is_active(const struct i915_vma *vma)
/* do not reserve memory to prevent deadlocks */
#define __EXEC_OBJECT_NO_RESERVE BIT(31)
-int __must_check __i915_vma_move_to_active(struct i915_vma *vma,
- struct i915_request *rq);
int __must_check _i915_vma_move_to_active(struct i915_vma *vma,
struct i915_request *rq,
struct dma_fence *fence,
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 01/17] drm/i915: Remove unused bits of i915_vma/active api
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
When reworking the code to move the eviction fence to the object,
the best code is removed code.
Remove some functions that are unused, and change the function definition
if it's only used in 1 place.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
[mlankhorst: Remove new use of i915_active_has_exclusive]
---
drivers/gpu/drm/i915/i915_active.c | 28 +++-------------------------
drivers/gpu/drm/i915/i915_active.h | 17 +----------------
drivers/gpu/drm/i915/i915_vma.c | 24 ++++++++++--------------
drivers/gpu/drm/i915/i915_vma.h | 2 --
4 files changed, 14 insertions(+), 57 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
index 3103c1e1fd14..ee2b3a375362 100644
--- a/drivers/gpu/drm/i915/i915_active.c
+++ b/drivers/gpu/drm/i915/i915_active.c
@@ -426,8 +426,9 @@ replace_barrier(struct i915_active *ref, struct i915_active_fence *active)
return true;
}
-int i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence)
+int i915_active_add_request(struct i915_active *ref, struct i915_request *rq)
{
+ struct dma_fence *fence = &rq->fence;
struct i915_active_fence *active;
int err;
@@ -436,7 +437,7 @@ int i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence)
if (err)
return err;
- active = active_instance(ref, idx);
+ active = active_instance(ref, i915_request_timeline(rq)->fence_context);
if (!active) {
err = -ENOMEM;
goto out;
@@ -477,29 +478,6 @@ __i915_active_set_fence(struct i915_active *ref,
return prev;
}
-static struct i915_active_fence *
-__active_fence(struct i915_active *ref, u64 idx)
-{
- struct active_node *it;
-
- it = __active_lookup(ref, idx);
- if (unlikely(!it)) { /* Contention with parallel tree builders! */
- spin_lock_irq(&ref->tree_lock);
- it = __active_lookup(ref, idx);
- spin_unlock_irq(&ref->tree_lock);
- }
- GEM_BUG_ON(!it); /* slot must be preallocated */
-
- return &it->base;
-}
-
-struct dma_fence *
-__i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence)
-{
- /* Only valid while active, see i915_active_acquire_for_context() */
- return __i915_active_set_fence(ref, __active_fence(ref, idx), fence);
-}
-
struct dma_fence *
i915_active_set_exclusive(struct i915_active *ref, struct dma_fence *f)
{
diff --git a/drivers/gpu/drm/i915/i915_active.h b/drivers/gpu/drm/i915/i915_active.h
index 5fcdb0e2bc9e..7eb44132183a 100644
--- a/drivers/gpu/drm/i915/i915_active.h
+++ b/drivers/gpu/drm/i915/i915_active.h
@@ -164,26 +164,11 @@ void __i915_active_init(struct i915_active *ref,
__i915_active_init(ref, active, retire, flags, &__mkey, &__wkey); \
} while (0)
-struct dma_fence *
-__i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence);
-int i915_active_ref(struct i915_active *ref, u64 idx, struct dma_fence *fence);
-
-static inline int
-i915_active_add_request(struct i915_active *ref, struct i915_request *rq)
-{
- return i915_active_ref(ref,
- i915_request_timeline(rq)->fence_context,
- &rq->fence);
-}
+int i915_active_add_request(struct i915_active *ref, struct i915_request *rq);
struct dma_fence *
i915_active_set_exclusive(struct i915_active *ref, struct dma_fence *f);
-static inline bool i915_active_has_exclusive(struct i915_active *ref)
-{
- return rcu_access_pointer(ref->excl.fence);
-}
-
int __i915_active_wait(struct i915_active *ref, int state);
static inline int i915_active_wait(struct i915_active *ref)
{
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 927f0d4f8e11..921d5b946c32 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -356,22 +356,18 @@ int i915_vma_wait_for_bind(struct i915_vma *vma)
#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_GEM)
static int i915_vma_verify_bind_complete(struct i915_vma *vma)
{
- int err = 0;
-
- if (i915_active_has_exclusive(&vma->active)) {
- struct dma_fence *fence =
- i915_active_fence_get(&vma->active.excl);
+ struct dma_fence *fence = i915_active_fence_get(&vma->active.excl);
+ int err;
- if (!fence)
- return 0;
+ if (!fence)
+ return 0;
- if (dma_fence_is_signaled(fence))
- err = fence->error;
- else
- err = -EBUSY;
+ if (dma_fence_is_signaled(fence))
+ err = fence->error;
+ else
+ err = -EBUSY;
- dma_fence_put(fence);
- }
+ dma_fence_put(fence);
return err;
}
@@ -1249,7 +1245,7 @@ __i915_request_await_bind(struct i915_request *rq, struct i915_vma *vma)
return __i915_request_await_exclusive(rq, &vma->active);
}
-int __i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *rq)
+static int __i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *rq)
{
int err;
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 4033aa08d5e4..9a931ecb09e5 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -55,8 +55,6 @@ static inline bool i915_vma_is_active(const struct i915_vma *vma)
/* do not reserve memory to prevent deadlocks */
#define __EXEC_OBJECT_NO_RESERVE BIT(31)
-int __must_check __i915_vma_move_to_active(struct i915_vma *vma,
- struct i915_request *rq);
int __must_check _i915_vma_move_to_active(struct i915_vma *vma,
struct i915_request *rq,
struct dma_fence *fence,
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 02/17] drm/i915: Change shrink ordering to use locking around unbinding.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Niranjana Vishwanathapura, dri-devel
Call drop_pages with the gem object lock held, instead of the other
way around. This will allow us to drop the vma bindings with the
gem object lock held.
We plan to require the object lock for unpinning in the future,
and this is an easy target.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 38 ++++++++++----------
1 file changed, 18 insertions(+), 20 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
index 157a9765f483..eebff4735781 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
@@ -36,8 +36,8 @@ static bool can_release_pages(struct drm_i915_gem_object *obj)
return swap_available() || obj->mm.madv == I915_MADV_DONTNEED;
}
-static bool unsafe_drop_pages(struct drm_i915_gem_object *obj,
- unsigned long shrink, bool trylock_vm)
+static int drop_pages(struct drm_i915_gem_object *obj,
+ unsigned long shrink, bool trylock_vm)
{
unsigned long flags;
@@ -214,26 +214,24 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww,
spin_unlock_irqrestore(&i915->mm.obj_lock, flags);
- err = 0;
- if (unsafe_drop_pages(obj, shrink, trylock_vm)) {
- /* May arrive from get_pages on another bo */
- if (!ww) {
- if (!i915_gem_object_trylock(obj))
- goto skip;
- } else {
- err = i915_gem_object_lock(obj, ww);
- if (err)
- goto skip;
- }
-
- if (!__i915_gem_object_put_pages(obj)) {
- if (!try_to_writeback(obj, shrink))
- count += obj->base.size >> PAGE_SHIFT;
- }
- if (!ww)
- i915_gem_object_unlock(obj);
+ /* May arrive from get_pages on another bo */
+ if (!ww) {
+ if (!i915_gem_object_trylock(obj))
+ goto skip;
+ } else {
+ err = i915_gem_object_lock(obj, ww);
+ if (err)
+ goto skip;
}
+ if (drop_pages(obj, shrink, trylock_vm) &&
+ !__i915_gem_object_put_pages(obj) &&
+ !try_to_writeback(obj, shrink))
+ count += obj->base.size >> PAGE_SHIFT;
+
+ if (!ww)
+ i915_gem_object_unlock(obj);
+
scanned += obj->base.size >> PAGE_SHIFT;
skip:
i915_gem_object_put(obj);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 02/17] drm/i915: Change shrink ordering to use locking around unbinding.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Call drop_pages with the gem object lock held, instead of the other
way around. This will allow us to drop the vma bindings with the
gem object lock held.
We plan to require the object lock for unpinning in the future,
and this is an easy target.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 38 ++++++++++----------
1 file changed, 18 insertions(+), 20 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
index 157a9765f483..eebff4735781 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
@@ -36,8 +36,8 @@ static bool can_release_pages(struct drm_i915_gem_object *obj)
return swap_available() || obj->mm.madv == I915_MADV_DONTNEED;
}
-static bool unsafe_drop_pages(struct drm_i915_gem_object *obj,
- unsigned long shrink, bool trylock_vm)
+static int drop_pages(struct drm_i915_gem_object *obj,
+ unsigned long shrink, bool trylock_vm)
{
unsigned long flags;
@@ -214,26 +214,24 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww,
spin_unlock_irqrestore(&i915->mm.obj_lock, flags);
- err = 0;
- if (unsafe_drop_pages(obj, shrink, trylock_vm)) {
- /* May arrive from get_pages on another bo */
- if (!ww) {
- if (!i915_gem_object_trylock(obj))
- goto skip;
- } else {
- err = i915_gem_object_lock(obj, ww);
- if (err)
- goto skip;
- }
-
- if (!__i915_gem_object_put_pages(obj)) {
- if (!try_to_writeback(obj, shrink))
- count += obj->base.size >> PAGE_SHIFT;
- }
- if (!ww)
- i915_gem_object_unlock(obj);
+ /* May arrive from get_pages on another bo */
+ if (!ww) {
+ if (!i915_gem_object_trylock(obj))
+ goto skip;
+ } else {
+ err = i915_gem_object_lock(obj, ww);
+ if (err)
+ goto skip;
}
+ if (drop_pages(obj, shrink, trylock_vm) &&
+ !__i915_gem_object_put_pages(obj) &&
+ !try_to_writeback(obj, shrink))
+ count += obj->base.size >> PAGE_SHIFT;
+
+ if (!ww)
+ i915_gem_object_unlock(obj);
+
scanned += obj->base.size >> PAGE_SHIFT;
skip:
i915_gem_object_put(obj);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 03/17] drm/i915: Remove pages_mutex and intel_gtt->vma_ops.set/clear_pages members, v3.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Big delta, but boils down to moving set_pages to i915_vma.c, and removing
the special handling, all callers use the defaults anyway. We only remap
in ggtt, so default case will fall through.
Because we still don't require locking in i915_vma_unpin(), handle this by
using xchg in get_pages(), as it's locked with obj->mutex, and cmpxchg in
unpin, which only fails if we race a against a new pin.
Changes since v1:
- aliasing gtt sets ZERO_SIZE_PTR, not -ENODEV, remove special case
from __i915_vma_get_pages(). (Matt)
Changes since v2:
- Free correct old pages in __i915_vma_get_pages(). (Matt)
Remove race of clearing vma->pages accidentally from put,
free it but leave it set, as only get has the lock.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/display/intel_dpt.c | 2 -
drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 15 -
drivers/gpu/drm/i915/gt/intel_ggtt.c | 403 ----------------
drivers/gpu/drm/i915/gt/intel_gtt.c | 13 -
drivers/gpu/drm/i915/gt/intel_gtt.h | 7 -
drivers/gpu/drm/i915/gt/intel_ppgtt.c | 12 -
drivers/gpu/drm/i915/i915_vma.c | 449 ++++++++++++++++--
drivers/gpu/drm/i915/i915_vma.h | 3 +
drivers/gpu/drm/i915/i915_vma_types.h | 1 -
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 14 +-
drivers/gpu/drm/i915/selftests/mock_gtt.c | 4 -
11 files changed, 429 insertions(+), 494 deletions(-)
diff --git a/drivers/gpu/drm/i915/display/intel_dpt.c b/drivers/gpu/drm/i915/display/intel_dpt.c
index 963ca7155b06..cb1b956ca175 100644
--- a/drivers/gpu/drm/i915/display/intel_dpt.c
+++ b/drivers/gpu/drm/i915/display/intel_dpt.c
@@ -279,8 +279,6 @@ intel_dpt_create(struct intel_framebuffer *fb)
vm->vma_ops.bind_vma = dpt_bind_vma;
vm->vma_ops.unbind_vma = dpt_unbind_vma;
- vm->vma_ops.set_pages = ggtt_set_pages;
- vm->vma_ops.clear_pages = clear_pages;
vm->pte_encode = gen8_ggtt_pte_encode;
diff --git a/drivers/gpu/drm/i915/gt/gen6_ppgtt.c b/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
index c0d149f04949..6e9292918bfc 100644
--- a/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
+++ b/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
@@ -269,19 +269,6 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
free_pd(&ppgtt->base.vm, ppgtt->base.pd);
}
-static int pd_vma_set_pages(struct i915_vma *vma)
-{
- vma->pages = ERR_PTR(-ENODEV);
- return 0;
-}
-
-static void pd_vma_clear_pages(struct i915_vma *vma)
-{
- GEM_BUG_ON(!vma->pages);
-
- vma->pages = NULL;
-}
-
static void pd_vma_bind(struct i915_address_space *vm,
struct i915_vm_pt_stash *stash,
struct i915_vma *vma,
@@ -321,8 +308,6 @@ static void pd_vma_unbind(struct i915_address_space *vm, struct i915_vma *vma)
}
static const struct i915_vma_ops pd_vma_ops = {
- .set_pages = pd_vma_set_pages,
- .clear_pages = pd_vma_clear_pages,
.bind_vma = pd_vma_bind,
.unbind_vma = pd_vma_unbind,
};
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index d85a1050f4a8..a287c9186ec9 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -22,9 +22,6 @@
#include "intel_gtt.h"
#include "gen8_ppgtt.h"
-static int
-i915_get_ggtt_vma_pages(struct i915_vma *vma);
-
static void i915_ggtt_color_adjust(const struct drm_mm_node *node,
unsigned long color,
u64 *start,
@@ -892,21 +889,6 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size)
return 0;
}
-int ggtt_set_pages(struct i915_vma *vma)
-{
- int ret;
-
- GEM_BUG_ON(vma->pages);
-
- ret = i915_get_ggtt_vma_pages(vma);
- if (ret)
- return ret;
-
- vma->page_sizes = vma->obj->mm.page_sizes;
-
- return 0;
-}
-
static void gen6_gmch_remove(struct i915_address_space *vm)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
@@ -968,8 +950,6 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)
ggtt->vm.vma_ops.bind_vma = ggtt_bind_vma;
ggtt->vm.vma_ops.unbind_vma = ggtt_unbind_vma;
- ggtt->vm.vma_ops.set_pages = ggtt_set_pages;
- ggtt->vm.vma_ops.clear_pages = clear_pages;
ggtt->vm.pte_encode = gen8_ggtt_pte_encode;
@@ -1119,8 +1099,6 @@ static int gen6_gmch_probe(struct i915_ggtt *ggtt)
ggtt->vm.vma_ops.bind_vma = ggtt_bind_vma;
ggtt->vm.vma_ops.unbind_vma = ggtt_unbind_vma;
- ggtt->vm.vma_ops.set_pages = ggtt_set_pages;
- ggtt->vm.vma_ops.clear_pages = clear_pages;
return ggtt_probe_common(ggtt, size);
}
@@ -1165,8 +1143,6 @@ static int i915_gmch_probe(struct i915_ggtt *ggtt)
ggtt->vm.vma_ops.bind_vma = ggtt_bind_vma;
ggtt->vm.vma_ops.unbind_vma = ggtt_unbind_vma;
- ggtt->vm.vma_ops.set_pages = ggtt_set_pages;
- ggtt->vm.vma_ops.clear_pages = clear_pages;
if (unlikely(ggtt->do_idle_maps))
drm_notice(&i915->drm,
@@ -1336,382 +1312,3 @@ void i915_ggtt_resume(struct i915_ggtt *ggtt)
intel_ggtt_restore_fences(ggtt);
}
-
-static struct scatterlist *
-rotate_pages(struct drm_i915_gem_object *obj, unsigned int offset,
- unsigned int width, unsigned int height,
- unsigned int src_stride, unsigned int dst_stride,
- struct sg_table *st, struct scatterlist *sg)
-{
- unsigned int column, row;
- unsigned int src_idx;
-
- for (column = 0; column < width; column++) {
- unsigned int left;
-
- src_idx = src_stride * (height - 1) + column + offset;
- for (row = 0; row < height; row++) {
- st->nents++;
- /*
- * We don't need the pages, but need to initialize
- * the entries so the sg list can be happily traversed.
- * The only thing we need are DMA addresses.
- */
- sg_set_page(sg, NULL, I915_GTT_PAGE_SIZE, 0);
- sg_dma_address(sg) =
- i915_gem_object_get_dma_address(obj, src_idx);
- sg_dma_len(sg) = I915_GTT_PAGE_SIZE;
- sg = sg_next(sg);
- src_idx -= src_stride;
- }
-
- left = (dst_stride - height) * I915_GTT_PAGE_SIZE;
-
- if (!left)
- continue;
-
- st->nents++;
-
- /*
- * The DE ignores the PTEs for the padding tiles, the sg entry
- * here is just a conenience to indicate how many padding PTEs
- * to insert at this spot.
- */
- sg_set_page(sg, NULL, left, 0);
- sg_dma_address(sg) = 0;
- sg_dma_len(sg) = left;
- sg = sg_next(sg);
- }
-
- return sg;
-}
-
-static noinline struct sg_table *
-intel_rotate_pages(struct intel_rotation_info *rot_info,
- struct drm_i915_gem_object *obj)
-{
- unsigned int size = intel_rotation_info_size(rot_info);
- struct drm_i915_private *i915 = to_i915(obj->base.dev);
- struct sg_table *st;
- struct scatterlist *sg;
- int ret = -ENOMEM;
- int i;
-
- /* Allocate target SG list. */
- st = kmalloc(sizeof(*st), GFP_KERNEL);
- if (!st)
- goto err_st_alloc;
-
- ret = sg_alloc_table(st, size, GFP_KERNEL);
- if (ret)
- goto err_sg_alloc;
-
- st->nents = 0;
- sg = st->sgl;
-
- for (i = 0 ; i < ARRAY_SIZE(rot_info->plane); i++)
- sg = rotate_pages(obj, rot_info->plane[i].offset,
- rot_info->plane[i].width, rot_info->plane[i].height,
- rot_info->plane[i].src_stride,
- rot_info->plane[i].dst_stride,
- st, sg);
-
- return st;
-
-err_sg_alloc:
- kfree(st);
-err_st_alloc:
-
- drm_dbg(&i915->drm, "Failed to create rotated mapping for object size %zu! (%ux%u tiles, %u pages)\n",
- obj->base.size, rot_info->plane[0].width,
- rot_info->plane[0].height, size);
-
- return ERR_PTR(ret);
-}
-
-static struct scatterlist *
-add_padding_pages(unsigned int count,
- struct sg_table *st, struct scatterlist *sg)
-{
- st->nents++;
-
- /*
- * The DE ignores the PTEs for the padding tiles, the sg entry
- * here is just a convenience to indicate how many padding PTEs
- * to insert at this spot.
- */
- sg_set_page(sg, NULL, count * I915_GTT_PAGE_SIZE, 0);
- sg_dma_address(sg) = 0;
- sg_dma_len(sg) = count * I915_GTT_PAGE_SIZE;
- sg = sg_next(sg);
-
- return sg;
-}
-
-static struct scatterlist *
-remap_tiled_color_plane_pages(struct drm_i915_gem_object *obj,
- unsigned int offset, unsigned int alignment_pad,
- unsigned int width, unsigned int height,
- unsigned int src_stride, unsigned int dst_stride,
- struct sg_table *st, struct scatterlist *sg,
- unsigned int *gtt_offset)
-{
- unsigned int row;
-
- if (!width || !height)
- return sg;
-
- if (alignment_pad)
- sg = add_padding_pages(alignment_pad, st, sg);
-
- for (row = 0; row < height; row++) {
- unsigned int left = width * I915_GTT_PAGE_SIZE;
-
- while (left) {
- dma_addr_t addr;
- unsigned int length;
-
- /*
- * We don't need the pages, but need to initialize
- * the entries so the sg list can be happily traversed.
- * The only thing we need are DMA addresses.
- */
-
- addr = i915_gem_object_get_dma_address_len(obj, offset, &length);
-
- length = min(left, length);
-
- st->nents++;
-
- sg_set_page(sg, NULL, length, 0);
- sg_dma_address(sg) = addr;
- sg_dma_len(sg) = length;
- sg = sg_next(sg);
-
- offset += length / I915_GTT_PAGE_SIZE;
- left -= length;
- }
-
- offset += src_stride - width;
-
- left = (dst_stride - width) * I915_GTT_PAGE_SIZE;
-
- if (!left)
- continue;
-
- sg = add_padding_pages(left >> PAGE_SHIFT, st, sg);
- }
-
- *gtt_offset += alignment_pad + dst_stride * height;
-
- return sg;
-}
-
-static struct scatterlist *
-remap_contiguous_pages(struct drm_i915_gem_object *obj,
- unsigned int obj_offset,
- unsigned int count,
- struct sg_table *st, struct scatterlist *sg)
-{
- struct scatterlist *iter;
- unsigned int offset;
-
- iter = i915_gem_object_get_sg_dma(obj, obj_offset, &offset);
- GEM_BUG_ON(!iter);
-
- do {
- unsigned int len;
-
- len = min(sg_dma_len(iter) - (offset << PAGE_SHIFT),
- count << PAGE_SHIFT);
- sg_set_page(sg, NULL, len, 0);
- sg_dma_address(sg) =
- sg_dma_address(iter) + (offset << PAGE_SHIFT);
- sg_dma_len(sg) = len;
-
- st->nents++;
- count -= len >> PAGE_SHIFT;
- if (count == 0)
- return sg;
-
- sg = __sg_next(sg);
- iter = __sg_next(iter);
- offset = 0;
- } while (1);
-}
-
-static struct scatterlist *
-remap_linear_color_plane_pages(struct drm_i915_gem_object *obj,
- unsigned int obj_offset, unsigned int alignment_pad,
- unsigned int size,
- struct sg_table *st, struct scatterlist *sg,
- unsigned int *gtt_offset)
-{
- if (!size)
- return sg;
-
- if (alignment_pad)
- sg = add_padding_pages(alignment_pad, st, sg);
-
- sg = remap_contiguous_pages(obj, obj_offset, size, st, sg);
- sg = sg_next(sg);
-
- *gtt_offset += alignment_pad + size;
-
- return sg;
-}
-
-static struct scatterlist *
-remap_color_plane_pages(const struct intel_remapped_info *rem_info,
- struct drm_i915_gem_object *obj,
- int color_plane,
- struct sg_table *st, struct scatterlist *sg,
- unsigned int *gtt_offset)
-{
- unsigned int alignment_pad = 0;
-
- if (rem_info->plane_alignment)
- alignment_pad = ALIGN(*gtt_offset, rem_info->plane_alignment) - *gtt_offset;
-
- if (rem_info->plane[color_plane].linear)
- sg = remap_linear_color_plane_pages(obj,
- rem_info->plane[color_plane].offset,
- alignment_pad,
- rem_info->plane[color_plane].size,
- st, sg,
- gtt_offset);
-
- else
- sg = remap_tiled_color_plane_pages(obj,
- rem_info->plane[color_plane].offset,
- alignment_pad,
- rem_info->plane[color_plane].width,
- rem_info->plane[color_plane].height,
- rem_info->plane[color_plane].src_stride,
- rem_info->plane[color_plane].dst_stride,
- st, sg,
- gtt_offset);
-
- return sg;
-}
-
-static noinline struct sg_table *
-intel_remap_pages(struct intel_remapped_info *rem_info,
- struct drm_i915_gem_object *obj)
-{
- unsigned int size = intel_remapped_info_size(rem_info);
- struct drm_i915_private *i915 = to_i915(obj->base.dev);
- struct sg_table *st;
- struct scatterlist *sg;
- unsigned int gtt_offset = 0;
- int ret = -ENOMEM;
- int i;
-
- /* Allocate target SG list. */
- st = kmalloc(sizeof(*st), GFP_KERNEL);
- if (!st)
- goto err_st_alloc;
-
- ret = sg_alloc_table(st, size, GFP_KERNEL);
- if (ret)
- goto err_sg_alloc;
-
- st->nents = 0;
- sg = st->sgl;
-
- for (i = 0 ; i < ARRAY_SIZE(rem_info->plane); i++)
- sg = remap_color_plane_pages(rem_info, obj, i, st, sg, >t_offset);
-
- i915_sg_trim(st);
-
- return st;
-
-err_sg_alloc:
- kfree(st);
-err_st_alloc:
-
- drm_dbg(&i915->drm, "Failed to create remapped mapping for object size %zu! (%ux%u tiles, %u pages)\n",
- obj->base.size, rem_info->plane[0].width,
- rem_info->plane[0].height, size);
-
- return ERR_PTR(ret);
-}
-
-static noinline struct sg_table *
-intel_partial_pages(const struct i915_ggtt_view *view,
- struct drm_i915_gem_object *obj)
-{
- struct sg_table *st;
- struct scatterlist *sg;
- unsigned int count = view->partial.size;
- int ret = -ENOMEM;
-
- st = kmalloc(sizeof(*st), GFP_KERNEL);
- if (!st)
- goto err_st_alloc;
-
- ret = sg_alloc_table(st, count, GFP_KERNEL);
- if (ret)
- goto err_sg_alloc;
-
- st->nents = 0;
-
- sg = remap_contiguous_pages(obj, view->partial.offset, count, st, st->sgl);
-
- sg_mark_end(sg);
- i915_sg_trim(st); /* Drop any unused tail entries. */
-
- return st;
-
-err_sg_alloc:
- kfree(st);
-err_st_alloc:
- return ERR_PTR(ret);
-}
-
-static int
-i915_get_ggtt_vma_pages(struct i915_vma *vma)
-{
- int ret;
-
- /*
- * The vma->pages are only valid within the lifespan of the borrowed
- * obj->mm.pages. When the obj->mm.pages sg_table is regenerated, so
- * must be the vma->pages. A simple rule is that vma->pages must only
- * be accessed when the obj->mm.pages are pinned.
- */
- GEM_BUG_ON(!i915_gem_object_has_pinned_pages(vma->obj));
-
- switch (vma->ggtt_view.type) {
- default:
- GEM_BUG_ON(vma->ggtt_view.type);
- fallthrough;
- case I915_GGTT_VIEW_NORMAL:
- vma->pages = vma->obj->mm.pages;
- return 0;
-
- case I915_GGTT_VIEW_ROTATED:
- vma->pages =
- intel_rotate_pages(&vma->ggtt_view.rotated, vma->obj);
- break;
-
- case I915_GGTT_VIEW_REMAPPED:
- vma->pages =
- intel_remap_pages(&vma->ggtt_view.remapped, vma->obj);
- break;
-
- case I915_GGTT_VIEW_PARTIAL:
- vma->pages = intel_partial_pages(&vma->ggtt_view, vma->obj);
- break;
- }
-
- ret = 0;
- if (IS_ERR(vma->pages)) {
- ret = PTR_ERR(vma->pages);
- vma->pages = NULL;
- drm_err(&vma->vm->i915->drm,
- "Failed to get pages for VMA view type %u (%d)!\n",
- vma->ggtt_view.type, ret);
- }
- return ret;
-}
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c
index 30683c06b344..a94be0306464 100644
--- a/drivers/gpu/drm/i915/gt/intel_gtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_gtt.c
@@ -224,19 +224,6 @@ void i915_address_space_init(struct i915_address_space *vm, int subclass)
INIT_LIST_HEAD(&vm->bound_list);
}
-void clear_pages(struct i915_vma *vma)
-{
- GEM_BUG_ON(!vma->pages);
-
- if (vma->pages != vma->obj->mm.pages) {
- sg_free_table(vma->pages);
- kfree(vma->pages);
- }
- vma->pages = NULL;
-
- memset(&vma->page_sizes, 0, sizeof(vma->page_sizes));
-}
-
void *__px_vaddr(struct drm_i915_gem_object *p)
{
enum i915_map_type type;
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h b/drivers/gpu/drm/i915/gt/intel_gtt.h
index 15b98321e89a..177b42b935a1 100644
--- a/drivers/gpu/drm/i915/gt/intel_gtt.h
+++ b/drivers/gpu/drm/i915/gt/intel_gtt.h
@@ -209,9 +209,6 @@ struct i915_vma_ops {
*/
void (*unbind_vma)(struct i915_address_space *vm,
struct i915_vma *vma);
-
- int (*set_pages)(struct i915_vma *vma);
- void (*clear_pages)(struct i915_vma *vma);
};
struct i915_address_space {
@@ -601,10 +598,6 @@ release_pd_entry(struct i915_page_directory * const pd,
const struct drm_i915_gem_object * const scratch);
void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);
-int ggtt_set_pages(struct i915_vma *vma);
-int ppgtt_set_pages(struct i915_vma *vma);
-void clear_pages(struct i915_vma *vma);
-
void ppgtt_bind_vma(struct i915_address_space *vm,
struct i915_vm_pt_stash *stash,
struct i915_vma *vma,
diff --git a/drivers/gpu/drm/i915/gt/intel_ppgtt.c b/drivers/gpu/drm/i915/gt/intel_ppgtt.c
index 4396bfd630d8..083b3090c69c 100644
--- a/drivers/gpu/drm/i915/gt/intel_ppgtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ppgtt.c
@@ -289,16 +289,6 @@ void i915_vm_free_pt_stash(struct i915_address_space *vm,
}
}
-int ppgtt_set_pages(struct i915_vma *vma)
-{
- GEM_BUG_ON(vma->pages);
-
- vma->pages = vma->obj->mm.pages;
- vma->page_sizes = vma->obj->mm.page_sizes;
-
- return 0;
-}
-
void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,
unsigned long lmem_pt_obj_flags)
{
@@ -315,6 +305,4 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,
ppgtt->vm.vma_ops.bind_vma = ppgtt_bind_vma;
ppgtt->vm.vma_ops.unbind_vma = ppgtt_unbind_vma;
- ppgtt->vm.vma_ops.set_pages = ppgtt_set_pages;
- ppgtt->vm.vma_ops.clear_pages = clear_pages;
}
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 921d5b946c32..8eb531b04d3e 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -109,7 +109,6 @@ vma_create(struct drm_i915_gem_object *obj,
return ERR_PTR(-ENOMEM);
kref_init(&vma->ref);
- mutex_init(&vma->pages_mutex);
vma->vm = i915_vm_get(vm);
vma->ops = &vm->vma_ops;
vma->obj = obj;
@@ -415,7 +414,7 @@ int i915_vma_bind(struct i915_vma *vma,
if (bind_flags == 0)
return 0;
- GEM_BUG_ON(!vma->pages);
+ GEM_BUG_ON(!atomic_read(&vma->pages_count));
trace_i915_vma_bind(vma, bind_flags);
if (work && bind_flags & vma->vm->bind_async_flags) {
@@ -816,10 +815,392 @@ static bool try_qad_pin(struct i915_vma *vma, unsigned int flags)
return pinned;
}
-static int vma_get_pages(struct i915_vma *vma)
+static struct scatterlist *
+rotate_pages(struct drm_i915_gem_object *obj, unsigned int offset,
+ unsigned int width, unsigned int height,
+ unsigned int src_stride, unsigned int dst_stride,
+ struct sg_table *st, struct scatterlist *sg)
{
- int err = 0;
- bool pinned_pages = true;
+ unsigned int column, row;
+ unsigned int src_idx;
+
+ for (column = 0; column < width; column++) {
+ unsigned int left;
+
+ src_idx = src_stride * (height - 1) + column + offset;
+ for (row = 0; row < height; row++) {
+ st->nents++;
+ /*
+ * We don't need the pages, but need to initialize
+ * the entries so the sg list can be happily traversed.
+ * The only thing we need are DMA addresses.
+ */
+ sg_set_page(sg, NULL, I915_GTT_PAGE_SIZE, 0);
+ sg_dma_address(sg) =
+ i915_gem_object_get_dma_address(obj, src_idx);
+ sg_dma_len(sg) = I915_GTT_PAGE_SIZE;
+ sg = sg_next(sg);
+ src_idx -= src_stride;
+ }
+
+ left = (dst_stride - height) * I915_GTT_PAGE_SIZE;
+
+ if (!left)
+ continue;
+
+ st->nents++;
+
+ /*
+ * The DE ignores the PTEs for the padding tiles, the sg entry
+ * here is just a conenience to indicate how many padding PTEs
+ * to insert at this spot.
+ */
+ sg_set_page(sg, NULL, left, 0);
+ sg_dma_address(sg) = 0;
+ sg_dma_len(sg) = left;
+ sg = sg_next(sg);
+ }
+
+ return sg;
+}
+
+static noinline struct sg_table *
+intel_rotate_pages(struct intel_rotation_info *rot_info,
+ struct drm_i915_gem_object *obj)
+{
+ unsigned int size = intel_rotation_info_size(rot_info);
+ struct drm_i915_private *i915 = to_i915(obj->base.dev);
+ struct sg_table *st;
+ struct scatterlist *sg;
+ int ret = -ENOMEM;
+ int i;
+
+ /* Allocate target SG list. */
+ st = kmalloc(sizeof(*st), GFP_KERNEL);
+ if (!st)
+ goto err_st_alloc;
+
+ ret = sg_alloc_table(st, size, GFP_KERNEL);
+ if (ret)
+ goto err_sg_alloc;
+
+ st->nents = 0;
+ sg = st->sgl;
+
+ for (i = 0 ; i < ARRAY_SIZE(rot_info->plane); i++)
+ sg = rotate_pages(obj, rot_info->plane[i].offset,
+ rot_info->plane[i].width, rot_info->plane[i].height,
+ rot_info->plane[i].src_stride,
+ rot_info->plane[i].dst_stride,
+ st, sg);
+
+ return st;
+
+err_sg_alloc:
+ kfree(st);
+err_st_alloc:
+
+ drm_dbg(&i915->drm, "Failed to create rotated mapping for object size %zu! (%ux%u tiles, %u pages)\n",
+ obj->base.size, rot_info->plane[0].width,
+ rot_info->plane[0].height, size);
+
+ return ERR_PTR(ret);
+}
+
+static struct scatterlist *
+add_padding_pages(unsigned int count,
+ struct sg_table *st, struct scatterlist *sg)
+{
+ st->nents++;
+
+ /*
+ * The DE ignores the PTEs for the padding tiles, the sg entry
+ * here is just a convenience to indicate how many padding PTEs
+ * to insert at this spot.
+ */
+ sg_set_page(sg, NULL, count * I915_GTT_PAGE_SIZE, 0);
+ sg_dma_address(sg) = 0;
+ sg_dma_len(sg) = count * I915_GTT_PAGE_SIZE;
+ sg = sg_next(sg);
+
+ return sg;
+}
+
+static struct scatterlist *
+remap_tiled_color_plane_pages(struct drm_i915_gem_object *obj,
+ unsigned int offset, unsigned int alignment_pad,
+ unsigned int width, unsigned int height,
+ unsigned int src_stride, unsigned int dst_stride,
+ struct sg_table *st, struct scatterlist *sg,
+ unsigned int *gtt_offset)
+{
+ unsigned int row;
+
+ if (!width || !height)
+ return sg;
+
+ if (alignment_pad)
+ sg = add_padding_pages(alignment_pad, st, sg);
+
+ for (row = 0; row < height; row++) {
+ unsigned int left = width * I915_GTT_PAGE_SIZE;
+
+ while (left) {
+ dma_addr_t addr;
+ unsigned int length;
+
+ /*
+ * We don't need the pages, but need to initialize
+ * the entries so the sg list can be happily traversed.
+ * The only thing we need are DMA addresses.
+ */
+
+ addr = i915_gem_object_get_dma_address_len(obj, offset, &length);
+
+ length = min(left, length);
+
+ st->nents++;
+
+ sg_set_page(sg, NULL, length, 0);
+ sg_dma_address(sg) = addr;
+ sg_dma_len(sg) = length;
+ sg = sg_next(sg);
+
+ offset += length / I915_GTT_PAGE_SIZE;
+ left -= length;
+ }
+
+ offset += src_stride - width;
+
+ left = (dst_stride - width) * I915_GTT_PAGE_SIZE;
+
+ if (!left)
+ continue;
+
+ sg = add_padding_pages(left >> PAGE_SHIFT, st, sg);
+ }
+
+ *gtt_offset += alignment_pad + dst_stride * height;
+
+ return sg;
+}
+
+static struct scatterlist *
+remap_contiguous_pages(struct drm_i915_gem_object *obj,
+ unsigned int obj_offset,
+ unsigned int count,
+ struct sg_table *st, struct scatterlist *sg)
+{
+ struct scatterlist *iter;
+ unsigned int offset;
+
+ iter = i915_gem_object_get_sg_dma(obj, obj_offset, &offset);
+ GEM_BUG_ON(!iter);
+
+ do {
+ unsigned int len;
+
+ len = min(sg_dma_len(iter) - (offset << PAGE_SHIFT),
+ count << PAGE_SHIFT);
+ sg_set_page(sg, NULL, len, 0);
+ sg_dma_address(sg) =
+ sg_dma_address(iter) + (offset << PAGE_SHIFT);
+ sg_dma_len(sg) = len;
+
+ st->nents++;
+ count -= len >> PAGE_SHIFT;
+ if (count == 0)
+ return sg;
+
+ sg = __sg_next(sg);
+ iter = __sg_next(iter);
+ offset = 0;
+ } while (1);
+}
+
+static struct scatterlist *
+remap_linear_color_plane_pages(struct drm_i915_gem_object *obj,
+ unsigned int obj_offset, unsigned int alignment_pad,
+ unsigned int size,
+ struct sg_table *st, struct scatterlist *sg,
+ unsigned int *gtt_offset)
+{
+ if (!size)
+ return sg;
+
+ if (alignment_pad)
+ sg = add_padding_pages(alignment_pad, st, sg);
+
+ sg = remap_contiguous_pages(obj, obj_offset, size, st, sg);
+ sg = sg_next(sg);
+
+ *gtt_offset += alignment_pad + size;
+
+ return sg;
+}
+
+static struct scatterlist *
+remap_color_plane_pages(const struct intel_remapped_info *rem_info,
+ struct drm_i915_gem_object *obj,
+ int color_plane,
+ struct sg_table *st, struct scatterlist *sg,
+ unsigned int *gtt_offset)
+{
+ unsigned int alignment_pad = 0;
+
+ if (rem_info->plane_alignment)
+ alignment_pad = ALIGN(*gtt_offset, rem_info->plane_alignment) - *gtt_offset;
+
+ if (rem_info->plane[color_plane].linear)
+ sg = remap_linear_color_plane_pages(obj,
+ rem_info->plane[color_plane].offset,
+ alignment_pad,
+ rem_info->plane[color_plane].size,
+ st, sg,
+ gtt_offset);
+
+ else
+ sg = remap_tiled_color_plane_pages(obj,
+ rem_info->plane[color_plane].offset,
+ alignment_pad,
+ rem_info->plane[color_plane].width,
+ rem_info->plane[color_plane].height,
+ rem_info->plane[color_plane].src_stride,
+ rem_info->plane[color_plane].dst_stride,
+ st, sg,
+ gtt_offset);
+
+ return sg;
+}
+
+static noinline struct sg_table *
+intel_remap_pages(struct intel_remapped_info *rem_info,
+ struct drm_i915_gem_object *obj)
+{
+ unsigned int size = intel_remapped_info_size(rem_info);
+ struct drm_i915_private *i915 = to_i915(obj->base.dev);
+ struct sg_table *st;
+ struct scatterlist *sg;
+ unsigned int gtt_offset = 0;
+ int ret = -ENOMEM;
+ int i;
+
+ /* Allocate target SG list. */
+ st = kmalloc(sizeof(*st), GFP_KERNEL);
+ if (!st)
+ goto err_st_alloc;
+
+ ret = sg_alloc_table(st, size, GFP_KERNEL);
+ if (ret)
+ goto err_sg_alloc;
+
+ st->nents = 0;
+ sg = st->sgl;
+
+ for (i = 0 ; i < ARRAY_SIZE(rem_info->plane); i++)
+ sg = remap_color_plane_pages(rem_info, obj, i, st, sg, >t_offset);
+
+ i915_sg_trim(st);
+
+ return st;
+
+err_sg_alloc:
+ kfree(st);
+err_st_alloc:
+
+ drm_dbg(&i915->drm, "Failed to create remapped mapping for object size %zu! (%ux%u tiles, %u pages)\n",
+ obj->base.size, rem_info->plane[0].width,
+ rem_info->plane[0].height, size);
+
+ return ERR_PTR(ret);
+}
+
+static noinline struct sg_table *
+intel_partial_pages(const struct i915_ggtt_view *view,
+ struct drm_i915_gem_object *obj)
+{
+ struct sg_table *st;
+ struct scatterlist *sg;
+ unsigned int count = view->partial.size;
+ int ret = -ENOMEM;
+
+ st = kmalloc(sizeof(*st), GFP_KERNEL);
+ if (!st)
+ goto err_st_alloc;
+
+ ret = sg_alloc_table(st, count, GFP_KERNEL);
+ if (ret)
+ goto err_sg_alloc;
+
+ st->nents = 0;
+
+ sg = remap_contiguous_pages(obj, view->partial.offset, count, st, st->sgl);
+
+ sg_mark_end(sg);
+ i915_sg_trim(st); /* Drop any unused tail entries. */
+
+ return st;
+
+err_sg_alloc:
+ kfree(st);
+err_st_alloc:
+ return ERR_PTR(ret);
+}
+
+static int
+__i915_vma_get_pages(struct i915_vma *vma)
+{
+ struct sg_table *pages;
+ int ret;
+
+ /*
+ * The vma->pages are only valid within the lifespan of the borrowed
+ * obj->mm.pages. When the obj->mm.pages sg_table is regenerated, so
+ * must be the vma->pages. A simple rule is that vma->pages must only
+ * be accessed when the obj->mm.pages are pinned.
+ */
+ GEM_BUG_ON(!i915_gem_object_has_pinned_pages(vma->obj));
+
+ switch (vma->ggtt_view.type) {
+ default:
+ GEM_BUG_ON(vma->ggtt_view.type);
+ fallthrough;
+ case I915_GGTT_VIEW_NORMAL:
+ pages = vma->obj->mm.pages;
+ break;
+
+ case I915_GGTT_VIEW_ROTATED:
+ pages =
+ intel_rotate_pages(&vma->ggtt_view.rotated, vma->obj);
+ break;
+
+ case I915_GGTT_VIEW_REMAPPED:
+ pages =
+ intel_remap_pages(&vma->ggtt_view.remapped, vma->obj);
+ break;
+
+ case I915_GGTT_VIEW_PARTIAL:
+ pages = intel_partial_pages(&vma->ggtt_view, vma->obj);
+ break;
+ }
+
+ ret = 0;
+ if (IS_ERR(pages)) {
+ ret = PTR_ERR(pages);
+ pages = NULL;
+ drm_err(&vma->vm->i915->drm,
+ "Failed to get pages for VMA view type %u (%d)!\n",
+ vma->ggtt_view.type, ret);
+ }
+
+ vma->pages = pages;
+
+ return ret;
+}
+
+I915_SELFTEST_EXPORT int i915_vma_get_pages(struct i915_vma *vma)
+{
+ int err;
if (atomic_add_unless(&vma->pages_count, 1, 0))
return 0;
@@ -828,25 +1209,17 @@ static int vma_get_pages(struct i915_vma *vma)
if (err)
return err;
- /* Allocations ahoy! */
- if (mutex_lock_interruptible(&vma->pages_mutex)) {
- err = -EINTR;
- goto unpin;
- }
+ err = __i915_vma_get_pages(vma);
+ if (err)
+ goto err_unpin;
- if (!atomic_read(&vma->pages_count)) {
- err = vma->ops->set_pages(vma);
- if (err)
- goto unlock;
- pinned_pages = false;
- }
+ vma->page_sizes = vma->obj->mm.page_sizes;
atomic_inc(&vma->pages_count);
-unlock:
- mutex_unlock(&vma->pages_mutex);
-unpin:
- if (pinned_pages)
- __i915_gem_object_unpin_pages(vma->obj);
+ return 0;
+
+err_unpin:
+ __i915_gem_object_unpin_pages(vma->obj);
return err;
}
@@ -854,18 +1227,31 @@ static int vma_get_pages(struct i915_vma *vma)
static void __vma_put_pages(struct i915_vma *vma, unsigned int count)
{
/* We allocate under vma_get_pages, so beware the shrinker */
- mutex_lock_nested(&vma->pages_mutex, SINGLE_DEPTH_NESTING);
+ struct sg_table *pages = READ_ONCE(vma->pages);
+
GEM_BUG_ON(atomic_read(&vma->pages_count) < count);
+
if (atomic_sub_return(count, &vma->pages_count) == 0) {
- vma->ops->clear_pages(vma);
- GEM_BUG_ON(vma->pages);
+ /*
+ * The atomic_sub_return is a read barrier for the READ_ONCE of
+ * vma->pages above.
+ *
+ * READ_ONCE is safe because this is either called from the same
+ * function (i915_vma_pin_ww), or guarded by vma->vm->mutex.
+ *
+ * TODO: We're leaving vma->pages dangling, until vma->obj->resv
+ * lock is required.
+ */
+ if (pages != vma->obj->mm.pages) {
+ sg_free_table(pages);
+ kfree(pages);
+ }
i915_gem_object_unpin_pages(vma->obj);
}
- mutex_unlock(&vma->pages_mutex);
}
-static void vma_put_pages(struct i915_vma *vma)
+I915_SELFTEST_EXPORT void i915_vma_put_pages(struct i915_vma *vma)
{
if (atomic_add_unless(&vma->pages_count, -1, 1))
return;
@@ -896,10 +1282,8 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
unsigned int bound;
int err;
-#ifdef CONFIG_PROVE_LOCKING
- if (debug_locks && !WARN_ON(!ww))
- assert_vma_held(vma);
-#endif
+ assert_vma_held(vma);
+ GEM_BUG_ON(!ww);
BUILD_BUG_ON(PIN_GLOBAL != I915_VMA_GLOBAL_BIND);
BUILD_BUG_ON(PIN_USER != I915_VMA_LOCAL_BIND);
@@ -910,7 +1294,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
if (try_qad_pin(vma, flags & I915_VMA_BIND_MASK))
return 0;
- err = vma_get_pages(vma);
+ err = i915_vma_get_pages(vma);
if (err)
return err;
@@ -1038,10 +1422,11 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
err_rpm:
if (wakeref)
intel_runtime_pm_put(&vma->vm->i915->runtime_pm, wakeref);
+
if (moving)
dma_fence_put(moving);
- vma_put_pages(vma);
+ i915_vma_put_pages(vma);
return err;
}
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 9a931ecb09e5..32719431b3df 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -431,4 +431,7 @@ static inline int i915_vma_sync(struct i915_vma *vma)
void i915_vma_module_exit(void);
int i915_vma_module_init(void);
+I915_SELFTEST_DECLARE(int i915_vma_get_pages(struct i915_vma *vma));
+I915_SELFTEST_DECLARE(void i915_vma_put_pages(struct i915_vma *vma));
+
#endif
diff --git a/drivers/gpu/drm/i915/i915_vma_types.h b/drivers/gpu/drm/i915/i915_vma_types.h
index f03fa96a1701..ca575e129ced 100644
--- a/drivers/gpu/drm/i915/i915_vma_types.h
+++ b/drivers/gpu/drm/i915/i915_vma_types.h
@@ -270,7 +270,6 @@ struct i915_vma {
#define I915_VMA_PAGES_BIAS 24
#define I915_VMA_PAGES_ACTIVE (BIT(24) | 1)
atomic_t pages_count; /* number of active binds to the pages */
- struct mutex pages_mutex; /* protect acquire/release of backing pages */
/**
* Support different GGTT views into the same object.
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 46f4236039a9..88b3a4d9e27e 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -1275,7 +1275,7 @@ static void track_vma_bind(struct i915_vma *vma)
__i915_gem_object_pin_pages(obj);
- GEM_BUG_ON(vma->pages);
+ GEM_BUG_ON(atomic_read(&vma->pages_count));
atomic_set(&vma->pages_count, I915_VMA_PAGES_ACTIVE);
__i915_gem_object_pin_pages(obj);
vma->pages = obj->mm.pages;
@@ -1953,7 +1953,9 @@ static int igt_cs_tlb(void *arg)
goto end;
}
- err = vma->ops->set_pages(vma);
+ i915_gem_object_lock(bbe, NULL);
+ err = i915_vma_get_pages(vma);
+ i915_gem_object_unlock(bbe);
if (err)
goto end;
@@ -1994,7 +1996,7 @@ static int igt_cs_tlb(void *arg)
i915_request_put(rq);
}
- vma->ops->clear_pages(vma);
+ i915_vma_put_pages(vma);
err = context_sync(ce);
if (err) {
@@ -2009,7 +2011,9 @@ static int igt_cs_tlb(void *arg)
goto end;
}
- err = vma->ops->set_pages(vma);
+ i915_gem_object_lock(act, NULL);
+ err = i915_vma_get_pages(vma);
+ i915_gem_object_unlock(act);
if (err)
goto end;
@@ -2047,7 +2051,7 @@ static int igt_cs_tlb(void *arg)
}
end_spin(batch, count - 1);
- vma->ops->clear_pages(vma);
+ i915_vma_put_pages(vma);
err = context_sync(ce);
if (err) {
diff --git a/drivers/gpu/drm/i915/selftests/mock_gtt.c b/drivers/gpu/drm/i915/selftests/mock_gtt.c
index 32ca8962d0ab..19537191bf03 100644
--- a/drivers/gpu/drm/i915/selftests/mock_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/mock_gtt.c
@@ -87,8 +87,6 @@ struct i915_ppgtt *mock_ppgtt(struct drm_i915_private *i915, const char *name)
ppgtt->vm.vma_ops.bind_vma = mock_bind_ppgtt;
ppgtt->vm.vma_ops.unbind_vma = mock_unbind_ppgtt;
- ppgtt->vm.vma_ops.set_pages = ppgtt_set_pages;
- ppgtt->vm.vma_ops.clear_pages = clear_pages;
return ppgtt;
}
@@ -128,8 +126,6 @@ void mock_init_ggtt(struct drm_i915_private *i915, struct i915_ggtt *ggtt)
ggtt->vm.vma_ops.bind_vma = mock_bind_ggtt;
ggtt->vm.vma_ops.unbind_vma = mock_unbind_ggtt;
- ggtt->vm.vma_ops.set_pages = ggtt_set_pages;
- ggtt->vm.vma_ops.clear_pages = clear_pages;
i915_address_space_init(&ggtt->vm, VM_CLASS_GGTT);
i915->gt.ggtt = ggtt;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 03/17] drm/i915: Remove pages_mutex and intel_gtt->vma_ops.set/clear_pages members, v3.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Big delta, but boils down to moving set_pages to i915_vma.c, and removing
the special handling, all callers use the defaults anyway. We only remap
in ggtt, so default case will fall through.
Because we still don't require locking in i915_vma_unpin(), handle this by
using xchg in get_pages(), as it's locked with obj->mutex, and cmpxchg in
unpin, which only fails if we race a against a new pin.
Changes since v1:
- aliasing gtt sets ZERO_SIZE_PTR, not -ENODEV, remove special case
from __i915_vma_get_pages(). (Matt)
Changes since v2:
- Free correct old pages in __i915_vma_get_pages(). (Matt)
Remove race of clearing vma->pages accidentally from put,
free it but leave it set, as only get has the lock.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/display/intel_dpt.c | 2 -
drivers/gpu/drm/i915/gt/gen6_ppgtt.c | 15 -
drivers/gpu/drm/i915/gt/intel_ggtt.c | 403 ----------------
drivers/gpu/drm/i915/gt/intel_gtt.c | 13 -
drivers/gpu/drm/i915/gt/intel_gtt.h | 7 -
drivers/gpu/drm/i915/gt/intel_ppgtt.c | 12 -
drivers/gpu/drm/i915/i915_vma.c | 449 ++++++++++++++++--
drivers/gpu/drm/i915/i915_vma.h | 3 +
drivers/gpu/drm/i915/i915_vma_types.h | 1 -
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 14 +-
drivers/gpu/drm/i915/selftests/mock_gtt.c | 4 -
11 files changed, 429 insertions(+), 494 deletions(-)
diff --git a/drivers/gpu/drm/i915/display/intel_dpt.c b/drivers/gpu/drm/i915/display/intel_dpt.c
index 963ca7155b06..cb1b956ca175 100644
--- a/drivers/gpu/drm/i915/display/intel_dpt.c
+++ b/drivers/gpu/drm/i915/display/intel_dpt.c
@@ -279,8 +279,6 @@ intel_dpt_create(struct intel_framebuffer *fb)
vm->vma_ops.bind_vma = dpt_bind_vma;
vm->vma_ops.unbind_vma = dpt_unbind_vma;
- vm->vma_ops.set_pages = ggtt_set_pages;
- vm->vma_ops.clear_pages = clear_pages;
vm->pte_encode = gen8_ggtt_pte_encode;
diff --git a/drivers/gpu/drm/i915/gt/gen6_ppgtt.c b/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
index c0d149f04949..6e9292918bfc 100644
--- a/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
+++ b/drivers/gpu/drm/i915/gt/gen6_ppgtt.c
@@ -269,19 +269,6 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
free_pd(&ppgtt->base.vm, ppgtt->base.pd);
}
-static int pd_vma_set_pages(struct i915_vma *vma)
-{
- vma->pages = ERR_PTR(-ENODEV);
- return 0;
-}
-
-static void pd_vma_clear_pages(struct i915_vma *vma)
-{
- GEM_BUG_ON(!vma->pages);
-
- vma->pages = NULL;
-}
-
static void pd_vma_bind(struct i915_address_space *vm,
struct i915_vm_pt_stash *stash,
struct i915_vma *vma,
@@ -321,8 +308,6 @@ static void pd_vma_unbind(struct i915_address_space *vm, struct i915_vma *vma)
}
static const struct i915_vma_ops pd_vma_ops = {
- .set_pages = pd_vma_set_pages,
- .clear_pages = pd_vma_clear_pages,
.bind_vma = pd_vma_bind,
.unbind_vma = pd_vma_unbind,
};
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index d85a1050f4a8..a287c9186ec9 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -22,9 +22,6 @@
#include "intel_gtt.h"
#include "gen8_ppgtt.h"
-static int
-i915_get_ggtt_vma_pages(struct i915_vma *vma);
-
static void i915_ggtt_color_adjust(const struct drm_mm_node *node,
unsigned long color,
u64 *start,
@@ -892,21 +889,6 @@ static int ggtt_probe_common(struct i915_ggtt *ggtt, u64 size)
return 0;
}
-int ggtt_set_pages(struct i915_vma *vma)
-{
- int ret;
-
- GEM_BUG_ON(vma->pages);
-
- ret = i915_get_ggtt_vma_pages(vma);
- if (ret)
- return ret;
-
- vma->page_sizes = vma->obj->mm.page_sizes;
-
- return 0;
-}
-
static void gen6_gmch_remove(struct i915_address_space *vm)
{
struct i915_ggtt *ggtt = i915_vm_to_ggtt(vm);
@@ -968,8 +950,6 @@ static int gen8_gmch_probe(struct i915_ggtt *ggtt)
ggtt->vm.vma_ops.bind_vma = ggtt_bind_vma;
ggtt->vm.vma_ops.unbind_vma = ggtt_unbind_vma;
- ggtt->vm.vma_ops.set_pages = ggtt_set_pages;
- ggtt->vm.vma_ops.clear_pages = clear_pages;
ggtt->vm.pte_encode = gen8_ggtt_pte_encode;
@@ -1119,8 +1099,6 @@ static int gen6_gmch_probe(struct i915_ggtt *ggtt)
ggtt->vm.vma_ops.bind_vma = ggtt_bind_vma;
ggtt->vm.vma_ops.unbind_vma = ggtt_unbind_vma;
- ggtt->vm.vma_ops.set_pages = ggtt_set_pages;
- ggtt->vm.vma_ops.clear_pages = clear_pages;
return ggtt_probe_common(ggtt, size);
}
@@ -1165,8 +1143,6 @@ static int i915_gmch_probe(struct i915_ggtt *ggtt)
ggtt->vm.vma_ops.bind_vma = ggtt_bind_vma;
ggtt->vm.vma_ops.unbind_vma = ggtt_unbind_vma;
- ggtt->vm.vma_ops.set_pages = ggtt_set_pages;
- ggtt->vm.vma_ops.clear_pages = clear_pages;
if (unlikely(ggtt->do_idle_maps))
drm_notice(&i915->drm,
@@ -1336,382 +1312,3 @@ void i915_ggtt_resume(struct i915_ggtt *ggtt)
intel_ggtt_restore_fences(ggtt);
}
-
-static struct scatterlist *
-rotate_pages(struct drm_i915_gem_object *obj, unsigned int offset,
- unsigned int width, unsigned int height,
- unsigned int src_stride, unsigned int dst_stride,
- struct sg_table *st, struct scatterlist *sg)
-{
- unsigned int column, row;
- unsigned int src_idx;
-
- for (column = 0; column < width; column++) {
- unsigned int left;
-
- src_idx = src_stride * (height - 1) + column + offset;
- for (row = 0; row < height; row++) {
- st->nents++;
- /*
- * We don't need the pages, but need to initialize
- * the entries so the sg list can be happily traversed.
- * The only thing we need are DMA addresses.
- */
- sg_set_page(sg, NULL, I915_GTT_PAGE_SIZE, 0);
- sg_dma_address(sg) =
- i915_gem_object_get_dma_address(obj, src_idx);
- sg_dma_len(sg) = I915_GTT_PAGE_SIZE;
- sg = sg_next(sg);
- src_idx -= src_stride;
- }
-
- left = (dst_stride - height) * I915_GTT_PAGE_SIZE;
-
- if (!left)
- continue;
-
- st->nents++;
-
- /*
- * The DE ignores the PTEs for the padding tiles, the sg entry
- * here is just a conenience to indicate how many padding PTEs
- * to insert at this spot.
- */
- sg_set_page(sg, NULL, left, 0);
- sg_dma_address(sg) = 0;
- sg_dma_len(sg) = left;
- sg = sg_next(sg);
- }
-
- return sg;
-}
-
-static noinline struct sg_table *
-intel_rotate_pages(struct intel_rotation_info *rot_info,
- struct drm_i915_gem_object *obj)
-{
- unsigned int size = intel_rotation_info_size(rot_info);
- struct drm_i915_private *i915 = to_i915(obj->base.dev);
- struct sg_table *st;
- struct scatterlist *sg;
- int ret = -ENOMEM;
- int i;
-
- /* Allocate target SG list. */
- st = kmalloc(sizeof(*st), GFP_KERNEL);
- if (!st)
- goto err_st_alloc;
-
- ret = sg_alloc_table(st, size, GFP_KERNEL);
- if (ret)
- goto err_sg_alloc;
-
- st->nents = 0;
- sg = st->sgl;
-
- for (i = 0 ; i < ARRAY_SIZE(rot_info->plane); i++)
- sg = rotate_pages(obj, rot_info->plane[i].offset,
- rot_info->plane[i].width, rot_info->plane[i].height,
- rot_info->plane[i].src_stride,
- rot_info->plane[i].dst_stride,
- st, sg);
-
- return st;
-
-err_sg_alloc:
- kfree(st);
-err_st_alloc:
-
- drm_dbg(&i915->drm, "Failed to create rotated mapping for object size %zu! (%ux%u tiles, %u pages)\n",
- obj->base.size, rot_info->plane[0].width,
- rot_info->plane[0].height, size);
-
- return ERR_PTR(ret);
-}
-
-static struct scatterlist *
-add_padding_pages(unsigned int count,
- struct sg_table *st, struct scatterlist *sg)
-{
- st->nents++;
-
- /*
- * The DE ignores the PTEs for the padding tiles, the sg entry
- * here is just a convenience to indicate how many padding PTEs
- * to insert at this spot.
- */
- sg_set_page(sg, NULL, count * I915_GTT_PAGE_SIZE, 0);
- sg_dma_address(sg) = 0;
- sg_dma_len(sg) = count * I915_GTT_PAGE_SIZE;
- sg = sg_next(sg);
-
- return sg;
-}
-
-static struct scatterlist *
-remap_tiled_color_plane_pages(struct drm_i915_gem_object *obj,
- unsigned int offset, unsigned int alignment_pad,
- unsigned int width, unsigned int height,
- unsigned int src_stride, unsigned int dst_stride,
- struct sg_table *st, struct scatterlist *sg,
- unsigned int *gtt_offset)
-{
- unsigned int row;
-
- if (!width || !height)
- return sg;
-
- if (alignment_pad)
- sg = add_padding_pages(alignment_pad, st, sg);
-
- for (row = 0; row < height; row++) {
- unsigned int left = width * I915_GTT_PAGE_SIZE;
-
- while (left) {
- dma_addr_t addr;
- unsigned int length;
-
- /*
- * We don't need the pages, but need to initialize
- * the entries so the sg list can be happily traversed.
- * The only thing we need are DMA addresses.
- */
-
- addr = i915_gem_object_get_dma_address_len(obj, offset, &length);
-
- length = min(left, length);
-
- st->nents++;
-
- sg_set_page(sg, NULL, length, 0);
- sg_dma_address(sg) = addr;
- sg_dma_len(sg) = length;
- sg = sg_next(sg);
-
- offset += length / I915_GTT_PAGE_SIZE;
- left -= length;
- }
-
- offset += src_stride - width;
-
- left = (dst_stride - width) * I915_GTT_PAGE_SIZE;
-
- if (!left)
- continue;
-
- sg = add_padding_pages(left >> PAGE_SHIFT, st, sg);
- }
-
- *gtt_offset += alignment_pad + dst_stride * height;
-
- return sg;
-}
-
-static struct scatterlist *
-remap_contiguous_pages(struct drm_i915_gem_object *obj,
- unsigned int obj_offset,
- unsigned int count,
- struct sg_table *st, struct scatterlist *sg)
-{
- struct scatterlist *iter;
- unsigned int offset;
-
- iter = i915_gem_object_get_sg_dma(obj, obj_offset, &offset);
- GEM_BUG_ON(!iter);
-
- do {
- unsigned int len;
-
- len = min(sg_dma_len(iter) - (offset << PAGE_SHIFT),
- count << PAGE_SHIFT);
- sg_set_page(sg, NULL, len, 0);
- sg_dma_address(sg) =
- sg_dma_address(iter) + (offset << PAGE_SHIFT);
- sg_dma_len(sg) = len;
-
- st->nents++;
- count -= len >> PAGE_SHIFT;
- if (count == 0)
- return sg;
-
- sg = __sg_next(sg);
- iter = __sg_next(iter);
- offset = 0;
- } while (1);
-}
-
-static struct scatterlist *
-remap_linear_color_plane_pages(struct drm_i915_gem_object *obj,
- unsigned int obj_offset, unsigned int alignment_pad,
- unsigned int size,
- struct sg_table *st, struct scatterlist *sg,
- unsigned int *gtt_offset)
-{
- if (!size)
- return sg;
-
- if (alignment_pad)
- sg = add_padding_pages(alignment_pad, st, sg);
-
- sg = remap_contiguous_pages(obj, obj_offset, size, st, sg);
- sg = sg_next(sg);
-
- *gtt_offset += alignment_pad + size;
-
- return sg;
-}
-
-static struct scatterlist *
-remap_color_plane_pages(const struct intel_remapped_info *rem_info,
- struct drm_i915_gem_object *obj,
- int color_plane,
- struct sg_table *st, struct scatterlist *sg,
- unsigned int *gtt_offset)
-{
- unsigned int alignment_pad = 0;
-
- if (rem_info->plane_alignment)
- alignment_pad = ALIGN(*gtt_offset, rem_info->plane_alignment) - *gtt_offset;
-
- if (rem_info->plane[color_plane].linear)
- sg = remap_linear_color_plane_pages(obj,
- rem_info->plane[color_plane].offset,
- alignment_pad,
- rem_info->plane[color_plane].size,
- st, sg,
- gtt_offset);
-
- else
- sg = remap_tiled_color_plane_pages(obj,
- rem_info->plane[color_plane].offset,
- alignment_pad,
- rem_info->plane[color_plane].width,
- rem_info->plane[color_plane].height,
- rem_info->plane[color_plane].src_stride,
- rem_info->plane[color_plane].dst_stride,
- st, sg,
- gtt_offset);
-
- return sg;
-}
-
-static noinline struct sg_table *
-intel_remap_pages(struct intel_remapped_info *rem_info,
- struct drm_i915_gem_object *obj)
-{
- unsigned int size = intel_remapped_info_size(rem_info);
- struct drm_i915_private *i915 = to_i915(obj->base.dev);
- struct sg_table *st;
- struct scatterlist *sg;
- unsigned int gtt_offset = 0;
- int ret = -ENOMEM;
- int i;
-
- /* Allocate target SG list. */
- st = kmalloc(sizeof(*st), GFP_KERNEL);
- if (!st)
- goto err_st_alloc;
-
- ret = sg_alloc_table(st, size, GFP_KERNEL);
- if (ret)
- goto err_sg_alloc;
-
- st->nents = 0;
- sg = st->sgl;
-
- for (i = 0 ; i < ARRAY_SIZE(rem_info->plane); i++)
- sg = remap_color_plane_pages(rem_info, obj, i, st, sg, >t_offset);
-
- i915_sg_trim(st);
-
- return st;
-
-err_sg_alloc:
- kfree(st);
-err_st_alloc:
-
- drm_dbg(&i915->drm, "Failed to create remapped mapping for object size %zu! (%ux%u tiles, %u pages)\n",
- obj->base.size, rem_info->plane[0].width,
- rem_info->plane[0].height, size);
-
- return ERR_PTR(ret);
-}
-
-static noinline struct sg_table *
-intel_partial_pages(const struct i915_ggtt_view *view,
- struct drm_i915_gem_object *obj)
-{
- struct sg_table *st;
- struct scatterlist *sg;
- unsigned int count = view->partial.size;
- int ret = -ENOMEM;
-
- st = kmalloc(sizeof(*st), GFP_KERNEL);
- if (!st)
- goto err_st_alloc;
-
- ret = sg_alloc_table(st, count, GFP_KERNEL);
- if (ret)
- goto err_sg_alloc;
-
- st->nents = 0;
-
- sg = remap_contiguous_pages(obj, view->partial.offset, count, st, st->sgl);
-
- sg_mark_end(sg);
- i915_sg_trim(st); /* Drop any unused tail entries. */
-
- return st;
-
-err_sg_alloc:
- kfree(st);
-err_st_alloc:
- return ERR_PTR(ret);
-}
-
-static int
-i915_get_ggtt_vma_pages(struct i915_vma *vma)
-{
- int ret;
-
- /*
- * The vma->pages are only valid within the lifespan of the borrowed
- * obj->mm.pages. When the obj->mm.pages sg_table is regenerated, so
- * must be the vma->pages. A simple rule is that vma->pages must only
- * be accessed when the obj->mm.pages are pinned.
- */
- GEM_BUG_ON(!i915_gem_object_has_pinned_pages(vma->obj));
-
- switch (vma->ggtt_view.type) {
- default:
- GEM_BUG_ON(vma->ggtt_view.type);
- fallthrough;
- case I915_GGTT_VIEW_NORMAL:
- vma->pages = vma->obj->mm.pages;
- return 0;
-
- case I915_GGTT_VIEW_ROTATED:
- vma->pages =
- intel_rotate_pages(&vma->ggtt_view.rotated, vma->obj);
- break;
-
- case I915_GGTT_VIEW_REMAPPED:
- vma->pages =
- intel_remap_pages(&vma->ggtt_view.remapped, vma->obj);
- break;
-
- case I915_GGTT_VIEW_PARTIAL:
- vma->pages = intel_partial_pages(&vma->ggtt_view, vma->obj);
- break;
- }
-
- ret = 0;
- if (IS_ERR(vma->pages)) {
- ret = PTR_ERR(vma->pages);
- vma->pages = NULL;
- drm_err(&vma->vm->i915->drm,
- "Failed to get pages for VMA view type %u (%d)!\n",
- vma->ggtt_view.type, ret);
- }
- return ret;
-}
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.c b/drivers/gpu/drm/i915/gt/intel_gtt.c
index 30683c06b344..a94be0306464 100644
--- a/drivers/gpu/drm/i915/gt/intel_gtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_gtt.c
@@ -224,19 +224,6 @@ void i915_address_space_init(struct i915_address_space *vm, int subclass)
INIT_LIST_HEAD(&vm->bound_list);
}
-void clear_pages(struct i915_vma *vma)
-{
- GEM_BUG_ON(!vma->pages);
-
- if (vma->pages != vma->obj->mm.pages) {
- sg_free_table(vma->pages);
- kfree(vma->pages);
- }
- vma->pages = NULL;
-
- memset(&vma->page_sizes, 0, sizeof(vma->page_sizes));
-}
-
void *__px_vaddr(struct drm_i915_gem_object *p)
{
enum i915_map_type type;
diff --git a/drivers/gpu/drm/i915/gt/intel_gtt.h b/drivers/gpu/drm/i915/gt/intel_gtt.h
index 15b98321e89a..177b42b935a1 100644
--- a/drivers/gpu/drm/i915/gt/intel_gtt.h
+++ b/drivers/gpu/drm/i915/gt/intel_gtt.h
@@ -209,9 +209,6 @@ struct i915_vma_ops {
*/
void (*unbind_vma)(struct i915_address_space *vm,
struct i915_vma *vma);
-
- int (*set_pages)(struct i915_vma *vma);
- void (*clear_pages)(struct i915_vma *vma);
};
struct i915_address_space {
@@ -601,10 +598,6 @@ release_pd_entry(struct i915_page_directory * const pd,
const struct drm_i915_gem_object * const scratch);
void gen6_ggtt_invalidate(struct i915_ggtt *ggtt);
-int ggtt_set_pages(struct i915_vma *vma);
-int ppgtt_set_pages(struct i915_vma *vma);
-void clear_pages(struct i915_vma *vma);
-
void ppgtt_bind_vma(struct i915_address_space *vm,
struct i915_vm_pt_stash *stash,
struct i915_vma *vma,
diff --git a/drivers/gpu/drm/i915/gt/intel_ppgtt.c b/drivers/gpu/drm/i915/gt/intel_ppgtt.c
index 4396bfd630d8..083b3090c69c 100644
--- a/drivers/gpu/drm/i915/gt/intel_ppgtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ppgtt.c
@@ -289,16 +289,6 @@ void i915_vm_free_pt_stash(struct i915_address_space *vm,
}
}
-int ppgtt_set_pages(struct i915_vma *vma)
-{
- GEM_BUG_ON(vma->pages);
-
- vma->pages = vma->obj->mm.pages;
- vma->page_sizes = vma->obj->mm.page_sizes;
-
- return 0;
-}
-
void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,
unsigned long lmem_pt_obj_flags)
{
@@ -315,6 +305,4 @@ void ppgtt_init(struct i915_ppgtt *ppgtt, struct intel_gt *gt,
ppgtt->vm.vma_ops.bind_vma = ppgtt_bind_vma;
ppgtt->vm.vma_ops.unbind_vma = ppgtt_unbind_vma;
- ppgtt->vm.vma_ops.set_pages = ppgtt_set_pages;
- ppgtt->vm.vma_ops.clear_pages = clear_pages;
}
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 921d5b946c32..8eb531b04d3e 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -109,7 +109,6 @@ vma_create(struct drm_i915_gem_object *obj,
return ERR_PTR(-ENOMEM);
kref_init(&vma->ref);
- mutex_init(&vma->pages_mutex);
vma->vm = i915_vm_get(vm);
vma->ops = &vm->vma_ops;
vma->obj = obj;
@@ -415,7 +414,7 @@ int i915_vma_bind(struct i915_vma *vma,
if (bind_flags == 0)
return 0;
- GEM_BUG_ON(!vma->pages);
+ GEM_BUG_ON(!atomic_read(&vma->pages_count));
trace_i915_vma_bind(vma, bind_flags);
if (work && bind_flags & vma->vm->bind_async_flags) {
@@ -816,10 +815,392 @@ static bool try_qad_pin(struct i915_vma *vma, unsigned int flags)
return pinned;
}
-static int vma_get_pages(struct i915_vma *vma)
+static struct scatterlist *
+rotate_pages(struct drm_i915_gem_object *obj, unsigned int offset,
+ unsigned int width, unsigned int height,
+ unsigned int src_stride, unsigned int dst_stride,
+ struct sg_table *st, struct scatterlist *sg)
{
- int err = 0;
- bool pinned_pages = true;
+ unsigned int column, row;
+ unsigned int src_idx;
+
+ for (column = 0; column < width; column++) {
+ unsigned int left;
+
+ src_idx = src_stride * (height - 1) + column + offset;
+ for (row = 0; row < height; row++) {
+ st->nents++;
+ /*
+ * We don't need the pages, but need to initialize
+ * the entries so the sg list can be happily traversed.
+ * The only thing we need are DMA addresses.
+ */
+ sg_set_page(sg, NULL, I915_GTT_PAGE_SIZE, 0);
+ sg_dma_address(sg) =
+ i915_gem_object_get_dma_address(obj, src_idx);
+ sg_dma_len(sg) = I915_GTT_PAGE_SIZE;
+ sg = sg_next(sg);
+ src_idx -= src_stride;
+ }
+
+ left = (dst_stride - height) * I915_GTT_PAGE_SIZE;
+
+ if (!left)
+ continue;
+
+ st->nents++;
+
+ /*
+ * The DE ignores the PTEs for the padding tiles, the sg entry
+ * here is just a conenience to indicate how many padding PTEs
+ * to insert at this spot.
+ */
+ sg_set_page(sg, NULL, left, 0);
+ sg_dma_address(sg) = 0;
+ sg_dma_len(sg) = left;
+ sg = sg_next(sg);
+ }
+
+ return sg;
+}
+
+static noinline struct sg_table *
+intel_rotate_pages(struct intel_rotation_info *rot_info,
+ struct drm_i915_gem_object *obj)
+{
+ unsigned int size = intel_rotation_info_size(rot_info);
+ struct drm_i915_private *i915 = to_i915(obj->base.dev);
+ struct sg_table *st;
+ struct scatterlist *sg;
+ int ret = -ENOMEM;
+ int i;
+
+ /* Allocate target SG list. */
+ st = kmalloc(sizeof(*st), GFP_KERNEL);
+ if (!st)
+ goto err_st_alloc;
+
+ ret = sg_alloc_table(st, size, GFP_KERNEL);
+ if (ret)
+ goto err_sg_alloc;
+
+ st->nents = 0;
+ sg = st->sgl;
+
+ for (i = 0 ; i < ARRAY_SIZE(rot_info->plane); i++)
+ sg = rotate_pages(obj, rot_info->plane[i].offset,
+ rot_info->plane[i].width, rot_info->plane[i].height,
+ rot_info->plane[i].src_stride,
+ rot_info->plane[i].dst_stride,
+ st, sg);
+
+ return st;
+
+err_sg_alloc:
+ kfree(st);
+err_st_alloc:
+
+ drm_dbg(&i915->drm, "Failed to create rotated mapping for object size %zu! (%ux%u tiles, %u pages)\n",
+ obj->base.size, rot_info->plane[0].width,
+ rot_info->plane[0].height, size);
+
+ return ERR_PTR(ret);
+}
+
+static struct scatterlist *
+add_padding_pages(unsigned int count,
+ struct sg_table *st, struct scatterlist *sg)
+{
+ st->nents++;
+
+ /*
+ * The DE ignores the PTEs for the padding tiles, the sg entry
+ * here is just a convenience to indicate how many padding PTEs
+ * to insert at this spot.
+ */
+ sg_set_page(sg, NULL, count * I915_GTT_PAGE_SIZE, 0);
+ sg_dma_address(sg) = 0;
+ sg_dma_len(sg) = count * I915_GTT_PAGE_SIZE;
+ sg = sg_next(sg);
+
+ return sg;
+}
+
+static struct scatterlist *
+remap_tiled_color_plane_pages(struct drm_i915_gem_object *obj,
+ unsigned int offset, unsigned int alignment_pad,
+ unsigned int width, unsigned int height,
+ unsigned int src_stride, unsigned int dst_stride,
+ struct sg_table *st, struct scatterlist *sg,
+ unsigned int *gtt_offset)
+{
+ unsigned int row;
+
+ if (!width || !height)
+ return sg;
+
+ if (alignment_pad)
+ sg = add_padding_pages(alignment_pad, st, sg);
+
+ for (row = 0; row < height; row++) {
+ unsigned int left = width * I915_GTT_PAGE_SIZE;
+
+ while (left) {
+ dma_addr_t addr;
+ unsigned int length;
+
+ /*
+ * We don't need the pages, but need to initialize
+ * the entries so the sg list can be happily traversed.
+ * The only thing we need are DMA addresses.
+ */
+
+ addr = i915_gem_object_get_dma_address_len(obj, offset, &length);
+
+ length = min(left, length);
+
+ st->nents++;
+
+ sg_set_page(sg, NULL, length, 0);
+ sg_dma_address(sg) = addr;
+ sg_dma_len(sg) = length;
+ sg = sg_next(sg);
+
+ offset += length / I915_GTT_PAGE_SIZE;
+ left -= length;
+ }
+
+ offset += src_stride - width;
+
+ left = (dst_stride - width) * I915_GTT_PAGE_SIZE;
+
+ if (!left)
+ continue;
+
+ sg = add_padding_pages(left >> PAGE_SHIFT, st, sg);
+ }
+
+ *gtt_offset += alignment_pad + dst_stride * height;
+
+ return sg;
+}
+
+static struct scatterlist *
+remap_contiguous_pages(struct drm_i915_gem_object *obj,
+ unsigned int obj_offset,
+ unsigned int count,
+ struct sg_table *st, struct scatterlist *sg)
+{
+ struct scatterlist *iter;
+ unsigned int offset;
+
+ iter = i915_gem_object_get_sg_dma(obj, obj_offset, &offset);
+ GEM_BUG_ON(!iter);
+
+ do {
+ unsigned int len;
+
+ len = min(sg_dma_len(iter) - (offset << PAGE_SHIFT),
+ count << PAGE_SHIFT);
+ sg_set_page(sg, NULL, len, 0);
+ sg_dma_address(sg) =
+ sg_dma_address(iter) + (offset << PAGE_SHIFT);
+ sg_dma_len(sg) = len;
+
+ st->nents++;
+ count -= len >> PAGE_SHIFT;
+ if (count == 0)
+ return sg;
+
+ sg = __sg_next(sg);
+ iter = __sg_next(iter);
+ offset = 0;
+ } while (1);
+}
+
+static struct scatterlist *
+remap_linear_color_plane_pages(struct drm_i915_gem_object *obj,
+ unsigned int obj_offset, unsigned int alignment_pad,
+ unsigned int size,
+ struct sg_table *st, struct scatterlist *sg,
+ unsigned int *gtt_offset)
+{
+ if (!size)
+ return sg;
+
+ if (alignment_pad)
+ sg = add_padding_pages(alignment_pad, st, sg);
+
+ sg = remap_contiguous_pages(obj, obj_offset, size, st, sg);
+ sg = sg_next(sg);
+
+ *gtt_offset += alignment_pad + size;
+
+ return sg;
+}
+
+static struct scatterlist *
+remap_color_plane_pages(const struct intel_remapped_info *rem_info,
+ struct drm_i915_gem_object *obj,
+ int color_plane,
+ struct sg_table *st, struct scatterlist *sg,
+ unsigned int *gtt_offset)
+{
+ unsigned int alignment_pad = 0;
+
+ if (rem_info->plane_alignment)
+ alignment_pad = ALIGN(*gtt_offset, rem_info->plane_alignment) - *gtt_offset;
+
+ if (rem_info->plane[color_plane].linear)
+ sg = remap_linear_color_plane_pages(obj,
+ rem_info->plane[color_plane].offset,
+ alignment_pad,
+ rem_info->plane[color_plane].size,
+ st, sg,
+ gtt_offset);
+
+ else
+ sg = remap_tiled_color_plane_pages(obj,
+ rem_info->plane[color_plane].offset,
+ alignment_pad,
+ rem_info->plane[color_plane].width,
+ rem_info->plane[color_plane].height,
+ rem_info->plane[color_plane].src_stride,
+ rem_info->plane[color_plane].dst_stride,
+ st, sg,
+ gtt_offset);
+
+ return sg;
+}
+
+static noinline struct sg_table *
+intel_remap_pages(struct intel_remapped_info *rem_info,
+ struct drm_i915_gem_object *obj)
+{
+ unsigned int size = intel_remapped_info_size(rem_info);
+ struct drm_i915_private *i915 = to_i915(obj->base.dev);
+ struct sg_table *st;
+ struct scatterlist *sg;
+ unsigned int gtt_offset = 0;
+ int ret = -ENOMEM;
+ int i;
+
+ /* Allocate target SG list. */
+ st = kmalloc(sizeof(*st), GFP_KERNEL);
+ if (!st)
+ goto err_st_alloc;
+
+ ret = sg_alloc_table(st, size, GFP_KERNEL);
+ if (ret)
+ goto err_sg_alloc;
+
+ st->nents = 0;
+ sg = st->sgl;
+
+ for (i = 0 ; i < ARRAY_SIZE(rem_info->plane); i++)
+ sg = remap_color_plane_pages(rem_info, obj, i, st, sg, >t_offset);
+
+ i915_sg_trim(st);
+
+ return st;
+
+err_sg_alloc:
+ kfree(st);
+err_st_alloc:
+
+ drm_dbg(&i915->drm, "Failed to create remapped mapping for object size %zu! (%ux%u tiles, %u pages)\n",
+ obj->base.size, rem_info->plane[0].width,
+ rem_info->plane[0].height, size);
+
+ return ERR_PTR(ret);
+}
+
+static noinline struct sg_table *
+intel_partial_pages(const struct i915_ggtt_view *view,
+ struct drm_i915_gem_object *obj)
+{
+ struct sg_table *st;
+ struct scatterlist *sg;
+ unsigned int count = view->partial.size;
+ int ret = -ENOMEM;
+
+ st = kmalloc(sizeof(*st), GFP_KERNEL);
+ if (!st)
+ goto err_st_alloc;
+
+ ret = sg_alloc_table(st, count, GFP_KERNEL);
+ if (ret)
+ goto err_sg_alloc;
+
+ st->nents = 0;
+
+ sg = remap_contiguous_pages(obj, view->partial.offset, count, st, st->sgl);
+
+ sg_mark_end(sg);
+ i915_sg_trim(st); /* Drop any unused tail entries. */
+
+ return st;
+
+err_sg_alloc:
+ kfree(st);
+err_st_alloc:
+ return ERR_PTR(ret);
+}
+
+static int
+__i915_vma_get_pages(struct i915_vma *vma)
+{
+ struct sg_table *pages;
+ int ret;
+
+ /*
+ * The vma->pages are only valid within the lifespan of the borrowed
+ * obj->mm.pages. When the obj->mm.pages sg_table is regenerated, so
+ * must be the vma->pages. A simple rule is that vma->pages must only
+ * be accessed when the obj->mm.pages are pinned.
+ */
+ GEM_BUG_ON(!i915_gem_object_has_pinned_pages(vma->obj));
+
+ switch (vma->ggtt_view.type) {
+ default:
+ GEM_BUG_ON(vma->ggtt_view.type);
+ fallthrough;
+ case I915_GGTT_VIEW_NORMAL:
+ pages = vma->obj->mm.pages;
+ break;
+
+ case I915_GGTT_VIEW_ROTATED:
+ pages =
+ intel_rotate_pages(&vma->ggtt_view.rotated, vma->obj);
+ break;
+
+ case I915_GGTT_VIEW_REMAPPED:
+ pages =
+ intel_remap_pages(&vma->ggtt_view.remapped, vma->obj);
+ break;
+
+ case I915_GGTT_VIEW_PARTIAL:
+ pages = intel_partial_pages(&vma->ggtt_view, vma->obj);
+ break;
+ }
+
+ ret = 0;
+ if (IS_ERR(pages)) {
+ ret = PTR_ERR(pages);
+ pages = NULL;
+ drm_err(&vma->vm->i915->drm,
+ "Failed to get pages for VMA view type %u (%d)!\n",
+ vma->ggtt_view.type, ret);
+ }
+
+ vma->pages = pages;
+
+ return ret;
+}
+
+I915_SELFTEST_EXPORT int i915_vma_get_pages(struct i915_vma *vma)
+{
+ int err;
if (atomic_add_unless(&vma->pages_count, 1, 0))
return 0;
@@ -828,25 +1209,17 @@ static int vma_get_pages(struct i915_vma *vma)
if (err)
return err;
- /* Allocations ahoy! */
- if (mutex_lock_interruptible(&vma->pages_mutex)) {
- err = -EINTR;
- goto unpin;
- }
+ err = __i915_vma_get_pages(vma);
+ if (err)
+ goto err_unpin;
- if (!atomic_read(&vma->pages_count)) {
- err = vma->ops->set_pages(vma);
- if (err)
- goto unlock;
- pinned_pages = false;
- }
+ vma->page_sizes = vma->obj->mm.page_sizes;
atomic_inc(&vma->pages_count);
-unlock:
- mutex_unlock(&vma->pages_mutex);
-unpin:
- if (pinned_pages)
- __i915_gem_object_unpin_pages(vma->obj);
+ return 0;
+
+err_unpin:
+ __i915_gem_object_unpin_pages(vma->obj);
return err;
}
@@ -854,18 +1227,31 @@ static int vma_get_pages(struct i915_vma *vma)
static void __vma_put_pages(struct i915_vma *vma, unsigned int count)
{
/* We allocate under vma_get_pages, so beware the shrinker */
- mutex_lock_nested(&vma->pages_mutex, SINGLE_DEPTH_NESTING);
+ struct sg_table *pages = READ_ONCE(vma->pages);
+
GEM_BUG_ON(atomic_read(&vma->pages_count) < count);
+
if (atomic_sub_return(count, &vma->pages_count) == 0) {
- vma->ops->clear_pages(vma);
- GEM_BUG_ON(vma->pages);
+ /*
+ * The atomic_sub_return is a read barrier for the READ_ONCE of
+ * vma->pages above.
+ *
+ * READ_ONCE is safe because this is either called from the same
+ * function (i915_vma_pin_ww), or guarded by vma->vm->mutex.
+ *
+ * TODO: We're leaving vma->pages dangling, until vma->obj->resv
+ * lock is required.
+ */
+ if (pages != vma->obj->mm.pages) {
+ sg_free_table(pages);
+ kfree(pages);
+ }
i915_gem_object_unpin_pages(vma->obj);
}
- mutex_unlock(&vma->pages_mutex);
}
-static void vma_put_pages(struct i915_vma *vma)
+I915_SELFTEST_EXPORT void i915_vma_put_pages(struct i915_vma *vma)
{
if (atomic_add_unless(&vma->pages_count, -1, 1))
return;
@@ -896,10 +1282,8 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
unsigned int bound;
int err;
-#ifdef CONFIG_PROVE_LOCKING
- if (debug_locks && !WARN_ON(!ww))
- assert_vma_held(vma);
-#endif
+ assert_vma_held(vma);
+ GEM_BUG_ON(!ww);
BUILD_BUG_ON(PIN_GLOBAL != I915_VMA_GLOBAL_BIND);
BUILD_BUG_ON(PIN_USER != I915_VMA_LOCAL_BIND);
@@ -910,7 +1294,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
if (try_qad_pin(vma, flags & I915_VMA_BIND_MASK))
return 0;
- err = vma_get_pages(vma);
+ err = i915_vma_get_pages(vma);
if (err)
return err;
@@ -1038,10 +1422,11 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
err_rpm:
if (wakeref)
intel_runtime_pm_put(&vma->vm->i915->runtime_pm, wakeref);
+
if (moving)
dma_fence_put(moving);
- vma_put_pages(vma);
+ i915_vma_put_pages(vma);
return err;
}
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 9a931ecb09e5..32719431b3df 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -431,4 +431,7 @@ static inline int i915_vma_sync(struct i915_vma *vma)
void i915_vma_module_exit(void);
int i915_vma_module_init(void);
+I915_SELFTEST_DECLARE(int i915_vma_get_pages(struct i915_vma *vma));
+I915_SELFTEST_DECLARE(void i915_vma_put_pages(struct i915_vma *vma));
+
#endif
diff --git a/drivers/gpu/drm/i915/i915_vma_types.h b/drivers/gpu/drm/i915/i915_vma_types.h
index f03fa96a1701..ca575e129ced 100644
--- a/drivers/gpu/drm/i915/i915_vma_types.h
+++ b/drivers/gpu/drm/i915/i915_vma_types.h
@@ -270,7 +270,6 @@ struct i915_vma {
#define I915_VMA_PAGES_BIAS 24
#define I915_VMA_PAGES_ACTIVE (BIT(24) | 1)
atomic_t pages_count; /* number of active binds to the pages */
- struct mutex pages_mutex; /* protect acquire/release of backing pages */
/**
* Support different GGTT views into the same object.
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 46f4236039a9..88b3a4d9e27e 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -1275,7 +1275,7 @@ static void track_vma_bind(struct i915_vma *vma)
__i915_gem_object_pin_pages(obj);
- GEM_BUG_ON(vma->pages);
+ GEM_BUG_ON(atomic_read(&vma->pages_count));
atomic_set(&vma->pages_count, I915_VMA_PAGES_ACTIVE);
__i915_gem_object_pin_pages(obj);
vma->pages = obj->mm.pages;
@@ -1953,7 +1953,9 @@ static int igt_cs_tlb(void *arg)
goto end;
}
- err = vma->ops->set_pages(vma);
+ i915_gem_object_lock(bbe, NULL);
+ err = i915_vma_get_pages(vma);
+ i915_gem_object_unlock(bbe);
if (err)
goto end;
@@ -1994,7 +1996,7 @@ static int igt_cs_tlb(void *arg)
i915_request_put(rq);
}
- vma->ops->clear_pages(vma);
+ i915_vma_put_pages(vma);
err = context_sync(ce);
if (err) {
@@ -2009,7 +2011,9 @@ static int igt_cs_tlb(void *arg)
goto end;
}
- err = vma->ops->set_pages(vma);
+ i915_gem_object_lock(act, NULL);
+ err = i915_vma_get_pages(vma);
+ i915_gem_object_unlock(act);
if (err)
goto end;
@@ -2047,7 +2051,7 @@ static int igt_cs_tlb(void *arg)
}
end_spin(batch, count - 1);
- vma->ops->clear_pages(vma);
+ i915_vma_put_pages(vma);
err = context_sync(ce);
if (err) {
diff --git a/drivers/gpu/drm/i915/selftests/mock_gtt.c b/drivers/gpu/drm/i915/selftests/mock_gtt.c
index 32ca8962d0ab..19537191bf03 100644
--- a/drivers/gpu/drm/i915/selftests/mock_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/mock_gtt.c
@@ -87,8 +87,6 @@ struct i915_ppgtt *mock_ppgtt(struct drm_i915_private *i915, const char *name)
ppgtt->vm.vma_ops.bind_vma = mock_bind_ppgtt;
ppgtt->vm.vma_ops.unbind_vma = mock_unbind_ppgtt;
- ppgtt->vm.vma_ops.set_pages = ppgtt_set_pages;
- ppgtt->vm.vma_ops.clear_pages = clear_pages;
return ppgtt;
}
@@ -128,8 +126,6 @@ void mock_init_ggtt(struct drm_i915_private *i915, struct i915_ggtt *ggtt)
ggtt->vm.vma_ops.bind_vma = mock_bind_ggtt;
ggtt->vm.vma_ops.unbind_vma = mock_unbind_ggtt;
- ggtt->vm.vma_ops.set_pages = ggtt_set_pages;
- ggtt->vm.vma_ops.clear_pages = clear_pages;
i915_address_space_init(&ggtt->vm, VM_CLASS_GGTT);
i915->gt.ggtt = ggtt;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 04/17] drm/i915: Take object lock in i915_ggtt_pin if ww is not set
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
i915_vma_wait_for_bind needs the vma lock held, fix the caller.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/i915_vma.c | 40 +++++++++++++++++++++++----------
1 file changed, 28 insertions(+), 12 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 8eb531b04d3e..de24e4b3b19b 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1441,23 +1441,15 @@ static void flush_idle_contexts(struct intel_gt *gt)
intel_gt_wait_for_idle(gt, MAX_SCHEDULE_TIMEOUT);
}
-int i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
- u32 align, unsigned int flags)
+static int __i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
+ u32 align, unsigned int flags)
{
struct i915_address_space *vm = vma->vm;
int err;
- GEM_BUG_ON(!i915_vma_is_ggtt(vma));
-
-#ifdef CONFIG_LOCKDEP
- WARN_ON(!ww && dma_resv_held(vma->obj->base.resv));
-#endif
-
do {
- if (ww)
- err = i915_vma_pin_ww(vma, ww, 0, align, flags | PIN_GLOBAL);
- else
- err = i915_vma_pin(vma, 0, align, flags | PIN_GLOBAL);
+ err = i915_vma_pin_ww(vma, ww, 0, align, flags | PIN_GLOBAL);
+
if (err != -ENOSPC) {
if (!err) {
err = i915_vma_wait_for_bind(vma);
@@ -1476,6 +1468,30 @@ int i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
} while (1);
}
+int i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
+ u32 align, unsigned int flags)
+{
+ struct i915_gem_ww_ctx _ww;
+ int err;
+
+ GEM_BUG_ON(!i915_vma_is_ggtt(vma));
+
+ if (ww)
+ return __i915_ggtt_pin(vma, ww, align, flags);
+
+#ifdef CONFIG_LOCKDEP
+ WARN_ON(dma_resv_held(vma->obj->base.resv));
+#endif
+
+ for_i915_gem_ww(&_ww, err, true) {
+ err = i915_gem_object_lock(vma->obj, &_ww);
+ if (!err)
+ err = __i915_ggtt_pin(vma, &_ww, align, flags);
+ }
+
+ return err;
+}
+
static void __vma_close(struct i915_vma *vma, struct intel_gt *gt)
{
/*
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 04/17] drm/i915: Take object lock in i915_ggtt_pin if ww is not set
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
i915_vma_wait_for_bind needs the vma lock held, fix the caller.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/i915_vma.c | 40 +++++++++++++++++++++++----------
1 file changed, 28 insertions(+), 12 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 8eb531b04d3e..de24e4b3b19b 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1441,23 +1441,15 @@ static void flush_idle_contexts(struct intel_gt *gt)
intel_gt_wait_for_idle(gt, MAX_SCHEDULE_TIMEOUT);
}
-int i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
- u32 align, unsigned int flags)
+static int __i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
+ u32 align, unsigned int flags)
{
struct i915_address_space *vm = vma->vm;
int err;
- GEM_BUG_ON(!i915_vma_is_ggtt(vma));
-
-#ifdef CONFIG_LOCKDEP
- WARN_ON(!ww && dma_resv_held(vma->obj->base.resv));
-#endif
-
do {
- if (ww)
- err = i915_vma_pin_ww(vma, ww, 0, align, flags | PIN_GLOBAL);
- else
- err = i915_vma_pin(vma, 0, align, flags | PIN_GLOBAL);
+ err = i915_vma_pin_ww(vma, ww, 0, align, flags | PIN_GLOBAL);
+
if (err != -ENOSPC) {
if (!err) {
err = i915_vma_wait_for_bind(vma);
@@ -1476,6 +1468,30 @@ int i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
} while (1);
}
+int i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
+ u32 align, unsigned int flags)
+{
+ struct i915_gem_ww_ctx _ww;
+ int err;
+
+ GEM_BUG_ON(!i915_vma_is_ggtt(vma));
+
+ if (ww)
+ return __i915_ggtt_pin(vma, ww, align, flags);
+
+#ifdef CONFIG_LOCKDEP
+ WARN_ON(dma_resv_held(vma->obj->base.resv));
+#endif
+
+ for_i915_gem_ww(&_ww, err, true) {
+ err = i915_gem_object_lock(vma->obj, &_ww);
+ if (!err)
+ err = __i915_ggtt_pin(vma, &_ww, align, flags);
+ }
+
+ return err;
+}
+
static void __vma_close(struct i915_vma *vma, struct intel_gt *gt)
{
/*
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 05/17] drm/i915: Force ww lock for i915_gem_object_ggtt_pin_ww, v2.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
We will need the lock to unbind the vma, and wait for bind to complete.
Remove the special casing for the !ww path, and force ww locking for all.
Changes since v1:
- Pass err to for_i915_gem_ww handling for -EDEADLK handling.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/i915_drv.h | 7 ++-----
drivers/gpu/drm/i915/i915_gem.c | 30 ++++++++++++++++++++++++++----
2 files changed, 28 insertions(+), 9 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index ed48179bacd8..d51628d10f9d 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1652,13 +1652,10 @@ i915_gem_object_ggtt_pin_ww(struct drm_i915_gem_object *obj,
const struct i915_ggtt_view *view,
u64 size, u64 alignment, u64 flags);
-static inline struct i915_vma * __must_check
+struct i915_vma * __must_check
i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
const struct i915_ggtt_view *view,
- u64 size, u64 alignment, u64 flags)
-{
- return i915_gem_object_ggtt_pin_ww(obj, NULL, view, size, alignment, flags);
-}
+ u64 size, u64 alignment, u64 flags);
int i915_gem_object_unbind(struct drm_i915_gem_object *obj,
unsigned long flags);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 527228d4da7e..6002045bd194 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -877,6 +877,8 @@ i915_gem_object_ggtt_pin_ww(struct drm_i915_gem_object *obj,
struct i915_vma *vma;
int ret;
+ GEM_WARN_ON(!ww);
+
if (flags & PIN_MAPPABLE &&
(!view || view->type == I915_GGTT_VIEW_NORMAL)) {
/*
@@ -936,10 +938,7 @@ i915_gem_object_ggtt_pin_ww(struct drm_i915_gem_object *obj,
return ERR_PTR(ret);
}
- if (ww)
- ret = i915_vma_pin_ww(vma, ww, size, alignment, flags | PIN_GLOBAL);
- else
- ret = i915_vma_pin(vma, size, alignment, flags | PIN_GLOBAL);
+ ret = i915_vma_pin_ww(vma, ww, size, alignment, flags | PIN_GLOBAL);
if (ret)
return ERR_PTR(ret);
@@ -959,6 +958,29 @@ i915_gem_object_ggtt_pin_ww(struct drm_i915_gem_object *obj,
return vma;
}
+struct i915_vma * __must_check
+i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
+ const struct i915_ggtt_view *view,
+ u64 size, u64 alignment, u64 flags)
+{
+ struct i915_gem_ww_ctx ww;
+ struct i915_vma *ret;
+ int err;
+
+ for_i915_gem_ww(&ww, err, true) {
+ err = i915_gem_object_lock(obj, &ww);
+ if (err)
+ continue;
+
+ ret = i915_gem_object_ggtt_pin_ww(obj, &ww, view, size,
+ alignment, flags);
+ if (IS_ERR(ret))
+ err = PTR_ERR(ret);
+ }
+
+ return err ? ERR_PTR(err) : ret;
+}
+
int
i915_gem_madvise_ioctl(struct drm_device *dev, void *data,
struct drm_file *file_priv)
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 05/17] drm/i915: Force ww lock for i915_gem_object_ggtt_pin_ww, v2.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
We will need the lock to unbind the vma, and wait for bind to complete.
Remove the special casing for the !ww path, and force ww locking for all.
Changes since v1:
- Pass err to for_i915_gem_ww handling for -EDEADLK handling.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/i915_drv.h | 7 ++-----
drivers/gpu/drm/i915/i915_gem.c | 30 ++++++++++++++++++++++++++----
2 files changed, 28 insertions(+), 9 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index ed48179bacd8..d51628d10f9d 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1652,13 +1652,10 @@ i915_gem_object_ggtt_pin_ww(struct drm_i915_gem_object *obj,
const struct i915_ggtt_view *view,
u64 size, u64 alignment, u64 flags);
-static inline struct i915_vma * __must_check
+struct i915_vma * __must_check
i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
const struct i915_ggtt_view *view,
- u64 size, u64 alignment, u64 flags)
-{
- return i915_gem_object_ggtt_pin_ww(obj, NULL, view, size, alignment, flags);
-}
+ u64 size, u64 alignment, u64 flags);
int i915_gem_object_unbind(struct drm_i915_gem_object *obj,
unsigned long flags);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 527228d4da7e..6002045bd194 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -877,6 +877,8 @@ i915_gem_object_ggtt_pin_ww(struct drm_i915_gem_object *obj,
struct i915_vma *vma;
int ret;
+ GEM_WARN_ON(!ww);
+
if (flags & PIN_MAPPABLE &&
(!view || view->type == I915_GGTT_VIEW_NORMAL)) {
/*
@@ -936,10 +938,7 @@ i915_gem_object_ggtt_pin_ww(struct drm_i915_gem_object *obj,
return ERR_PTR(ret);
}
- if (ww)
- ret = i915_vma_pin_ww(vma, ww, size, alignment, flags | PIN_GLOBAL);
- else
- ret = i915_vma_pin(vma, size, alignment, flags | PIN_GLOBAL);
+ ret = i915_vma_pin_ww(vma, ww, size, alignment, flags | PIN_GLOBAL);
if (ret)
return ERR_PTR(ret);
@@ -959,6 +958,29 @@ i915_gem_object_ggtt_pin_ww(struct drm_i915_gem_object *obj,
return vma;
}
+struct i915_vma * __must_check
+i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
+ const struct i915_ggtt_view *view,
+ u64 size, u64 alignment, u64 flags)
+{
+ struct i915_gem_ww_ctx ww;
+ struct i915_vma *ret;
+ int err;
+
+ for_i915_gem_ww(&ww, err, true) {
+ err = i915_gem_object_lock(obj, &ww);
+ if (err)
+ continue;
+
+ ret = i915_gem_object_ggtt_pin_ww(obj, &ww, view, size,
+ alignment, flags);
+ if (IS_ERR(ret))
+ err = PTR_ERR(ret);
+ }
+
+ return err ? ERR_PTR(err) : ret;
+}
+
int
i915_gem_madvise_ioctl(struct drm_device *dev, void *data,
struct drm_file *file_priv)
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 06/17] drm/i915: Ensure gem_contexts selftests work with unbind changes, v2.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
In the next commits, we may not evict when refcount = 0.
igt_vm_isolation() continuously tries to pin/unpin at same address,
but also calls put() on the object, which means the object may not
be unpinned in time.
Instead of this, re-use the same object over and over, so they can
be unbound as required.
Changes since v1:
- Fix cleaning up obj_b on failure. (Matt)
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
.../drm/i915/gem/selftests/i915_gem_context.c | 59 +++++++++++--------
1 file changed, 35 insertions(+), 24 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
index 21b71568cd5f..786c58412733 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
@@ -1481,10 +1481,10 @@ static int check_scratch(struct i915_address_space *vm, u64 offset)
static int write_to_scratch(struct i915_gem_context *ctx,
struct intel_engine_cs *engine,
+ struct drm_i915_gem_object *obj,
u64 offset, u32 value)
{
struct drm_i915_private *i915 = ctx->i915;
- struct drm_i915_gem_object *obj;
struct i915_address_space *vm;
struct i915_request *rq;
struct i915_vma *vma;
@@ -1497,15 +1497,9 @@ static int write_to_scratch(struct i915_gem_context *ctx,
if (err)
return err;
- obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
- if (IS_ERR(obj))
- return PTR_ERR(obj);
-
cmd = i915_gem_object_pin_map_unlocked(obj, I915_MAP_WB);
- if (IS_ERR(cmd)) {
- err = PTR_ERR(cmd);
- goto out;
- }
+ if (IS_ERR(cmd))
+ return PTR_ERR(cmd);
*cmd++ = MI_STORE_DWORD_IMM_GEN4;
if (GRAPHICS_VER(i915) >= 8) {
@@ -1569,17 +1563,19 @@ static int write_to_scratch(struct i915_gem_context *ctx,
i915_vma_unpin(vma);
out_vm:
i915_vm_put(vm);
-out:
- i915_gem_object_put(obj);
+
+ if (!err)
+ err = i915_gem_object_wait(obj, 0, MAX_SCHEDULE_TIMEOUT);
+
return err;
}
static int read_from_scratch(struct i915_gem_context *ctx,
struct intel_engine_cs *engine,
+ struct drm_i915_gem_object *obj,
u64 offset, u32 *value)
{
struct drm_i915_private *i915 = ctx->i915;
- struct drm_i915_gem_object *obj;
struct i915_address_space *vm;
const u32 result = 0x100;
struct i915_request *rq;
@@ -1594,10 +1590,6 @@ static int read_from_scratch(struct i915_gem_context *ctx,
if (err)
return err;
- obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
- if (IS_ERR(obj))
- return PTR_ERR(obj);
-
if (GRAPHICS_VER(i915) >= 8) {
const u32 GPR0 = engine->mmio_base + 0x600;
@@ -1615,7 +1607,7 @@ static int read_from_scratch(struct i915_gem_context *ctx,
cmd = i915_gem_object_pin_map_unlocked(obj, I915_MAP_WB);
if (IS_ERR(cmd)) {
err = PTR_ERR(cmd);
- goto out;
+ goto err_unpin;
}
memset(cmd, POISON_INUSE, PAGE_SIZE);
@@ -1651,7 +1643,7 @@ static int read_from_scratch(struct i915_gem_context *ctx,
cmd = i915_gem_object_pin_map_unlocked(obj, I915_MAP_WB);
if (IS_ERR(cmd)) {
err = PTR_ERR(cmd);
- goto out;
+ goto err_unpin;
}
memset(cmd, POISON_INUSE, PAGE_SIZE);
@@ -1722,8 +1714,10 @@ static int read_from_scratch(struct i915_gem_context *ctx,
i915_vma_unpin(vma);
out_vm:
i915_vm_put(vm);
-out:
- i915_gem_object_put(obj);
+
+ if (!err)
+ err = i915_gem_object_wait(obj, 0, MAX_SCHEDULE_TIMEOUT);
+
return err;
}
@@ -1757,6 +1751,7 @@ static int igt_vm_isolation(void *arg)
{
struct drm_i915_private *i915 = arg;
struct i915_gem_context *ctx_a, *ctx_b;
+ struct drm_i915_gem_object *obj_a, *obj_b;
unsigned long num_engines, count;
struct intel_engine_cs *engine;
struct igt_live_test t;
@@ -1810,6 +1805,18 @@ static int igt_vm_isolation(void *arg)
vm_total = ctx_a->vm->total;
GEM_BUG_ON(ctx_b->vm->total != vm_total);
+ obj_a = i915_gem_object_create_internal(i915, PAGE_SIZE);
+ if (IS_ERR(obj_a)) {
+ err = PTR_ERR(obj_a);
+ goto out_file;
+ }
+
+ obj_b = i915_gem_object_create_internal(i915, PAGE_SIZE);
+ if (IS_ERR(obj_b)) {
+ err = PTR_ERR(obj_b);
+ goto put_a;
+ }
+
count = 0;
num_engines = 0;
for_each_uabi_engine(engine, i915) {
@@ -1832,13 +1839,13 @@ static int igt_vm_isolation(void *arg)
I915_GTT_PAGE_SIZE, vm_total,
sizeof(u32), alignof_dword);
- err = write_to_scratch(ctx_a, engine,
+ err = write_to_scratch(ctx_a, engine, obj_a,
offset, 0xdeadbeef);
if (err == 0)
- err = read_from_scratch(ctx_b, engine,
+ err = read_from_scratch(ctx_b, engine, obj_b,
offset, &value);
if (err)
- goto out_file;
+ goto put_b;
if (value != expected) {
pr_err("%s: Read %08x from scratch (offset 0x%08x_%08x), after %lu reads!\n",
@@ -1847,7 +1854,7 @@ static int igt_vm_isolation(void *arg)
lower_32_bits(offset),
this);
err = -EINVAL;
- goto out_file;
+ goto put_b;
}
this++;
@@ -1858,6 +1865,10 @@ static int igt_vm_isolation(void *arg)
pr_info("Checked %lu scratch offsets across %lu engines\n",
count, num_engines);
+put_b:
+ i915_gem_object_put(obj_b);
+put_a:
+ i915_gem_object_put(obj_a);
out_file:
if (igt_live_test_end(&t))
err = -EIO;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 06/17] drm/i915: Ensure gem_contexts selftests work with unbind changes, v2.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
In the next commits, we may not evict when refcount = 0.
igt_vm_isolation() continuously tries to pin/unpin at same address,
but also calls put() on the object, which means the object may not
be unpinned in time.
Instead of this, re-use the same object over and over, so they can
be unbound as required.
Changes since v1:
- Fix cleaning up obj_b on failure. (Matt)
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
.../drm/i915/gem/selftests/i915_gem_context.c | 59 +++++++++++--------
1 file changed, 35 insertions(+), 24 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
index 21b71568cd5f..786c58412733 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_context.c
@@ -1481,10 +1481,10 @@ static int check_scratch(struct i915_address_space *vm, u64 offset)
static int write_to_scratch(struct i915_gem_context *ctx,
struct intel_engine_cs *engine,
+ struct drm_i915_gem_object *obj,
u64 offset, u32 value)
{
struct drm_i915_private *i915 = ctx->i915;
- struct drm_i915_gem_object *obj;
struct i915_address_space *vm;
struct i915_request *rq;
struct i915_vma *vma;
@@ -1497,15 +1497,9 @@ static int write_to_scratch(struct i915_gem_context *ctx,
if (err)
return err;
- obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
- if (IS_ERR(obj))
- return PTR_ERR(obj);
-
cmd = i915_gem_object_pin_map_unlocked(obj, I915_MAP_WB);
- if (IS_ERR(cmd)) {
- err = PTR_ERR(cmd);
- goto out;
- }
+ if (IS_ERR(cmd))
+ return PTR_ERR(cmd);
*cmd++ = MI_STORE_DWORD_IMM_GEN4;
if (GRAPHICS_VER(i915) >= 8) {
@@ -1569,17 +1563,19 @@ static int write_to_scratch(struct i915_gem_context *ctx,
i915_vma_unpin(vma);
out_vm:
i915_vm_put(vm);
-out:
- i915_gem_object_put(obj);
+
+ if (!err)
+ err = i915_gem_object_wait(obj, 0, MAX_SCHEDULE_TIMEOUT);
+
return err;
}
static int read_from_scratch(struct i915_gem_context *ctx,
struct intel_engine_cs *engine,
+ struct drm_i915_gem_object *obj,
u64 offset, u32 *value)
{
struct drm_i915_private *i915 = ctx->i915;
- struct drm_i915_gem_object *obj;
struct i915_address_space *vm;
const u32 result = 0x100;
struct i915_request *rq;
@@ -1594,10 +1590,6 @@ static int read_from_scratch(struct i915_gem_context *ctx,
if (err)
return err;
- obj = i915_gem_object_create_internal(i915, PAGE_SIZE);
- if (IS_ERR(obj))
- return PTR_ERR(obj);
-
if (GRAPHICS_VER(i915) >= 8) {
const u32 GPR0 = engine->mmio_base + 0x600;
@@ -1615,7 +1607,7 @@ static int read_from_scratch(struct i915_gem_context *ctx,
cmd = i915_gem_object_pin_map_unlocked(obj, I915_MAP_WB);
if (IS_ERR(cmd)) {
err = PTR_ERR(cmd);
- goto out;
+ goto err_unpin;
}
memset(cmd, POISON_INUSE, PAGE_SIZE);
@@ -1651,7 +1643,7 @@ static int read_from_scratch(struct i915_gem_context *ctx,
cmd = i915_gem_object_pin_map_unlocked(obj, I915_MAP_WB);
if (IS_ERR(cmd)) {
err = PTR_ERR(cmd);
- goto out;
+ goto err_unpin;
}
memset(cmd, POISON_INUSE, PAGE_SIZE);
@@ -1722,8 +1714,10 @@ static int read_from_scratch(struct i915_gem_context *ctx,
i915_vma_unpin(vma);
out_vm:
i915_vm_put(vm);
-out:
- i915_gem_object_put(obj);
+
+ if (!err)
+ err = i915_gem_object_wait(obj, 0, MAX_SCHEDULE_TIMEOUT);
+
return err;
}
@@ -1757,6 +1751,7 @@ static int igt_vm_isolation(void *arg)
{
struct drm_i915_private *i915 = arg;
struct i915_gem_context *ctx_a, *ctx_b;
+ struct drm_i915_gem_object *obj_a, *obj_b;
unsigned long num_engines, count;
struct intel_engine_cs *engine;
struct igt_live_test t;
@@ -1810,6 +1805,18 @@ static int igt_vm_isolation(void *arg)
vm_total = ctx_a->vm->total;
GEM_BUG_ON(ctx_b->vm->total != vm_total);
+ obj_a = i915_gem_object_create_internal(i915, PAGE_SIZE);
+ if (IS_ERR(obj_a)) {
+ err = PTR_ERR(obj_a);
+ goto out_file;
+ }
+
+ obj_b = i915_gem_object_create_internal(i915, PAGE_SIZE);
+ if (IS_ERR(obj_b)) {
+ err = PTR_ERR(obj_b);
+ goto put_a;
+ }
+
count = 0;
num_engines = 0;
for_each_uabi_engine(engine, i915) {
@@ -1832,13 +1839,13 @@ static int igt_vm_isolation(void *arg)
I915_GTT_PAGE_SIZE, vm_total,
sizeof(u32), alignof_dword);
- err = write_to_scratch(ctx_a, engine,
+ err = write_to_scratch(ctx_a, engine, obj_a,
offset, 0xdeadbeef);
if (err == 0)
- err = read_from_scratch(ctx_b, engine,
+ err = read_from_scratch(ctx_b, engine, obj_b,
offset, &value);
if (err)
- goto out_file;
+ goto put_b;
if (value != expected) {
pr_err("%s: Read %08x from scratch (offset 0x%08x_%08x), after %lu reads!\n",
@@ -1847,7 +1854,7 @@ static int igt_vm_isolation(void *arg)
lower_32_bits(offset),
this);
err = -EINVAL;
- goto out_file;
+ goto put_b;
}
this++;
@@ -1858,6 +1865,10 @@ static int igt_vm_isolation(void *arg)
pr_info("Checked %lu scratch offsets across %lu engines\n",
count, num_engines);
+put_b:
+ i915_gem_object_put(obj_b);
+put_a:
+ i915_gem_object_put(obj_a);
out_file:
if (igt_live_test_end(&t))
err = -EIO;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 07/17] drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking changes.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Now that we require locking to evict, multiple vmas from the same object
might not be evicted. This is expected and required, because execbuf will
move to short-term pinning by using the lock only. This will cause these
tests to fail, because they create a ton of vma's for the same object.
Unbind manually to prevent spurious -ENOSPC in those mock tests.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/selftests/i915_vma.c | 17 ++++++++++++++++-
1 file changed, 16 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/selftests/i915_vma.c b/drivers/gpu/drm/i915/selftests/i915_vma.c
index 1f10fe36619b..5c5809dfe9b2 100644
--- a/drivers/gpu/drm/i915/selftests/i915_vma.c
+++ b/drivers/gpu/drm/i915/selftests/i915_vma.c
@@ -691,7 +691,11 @@ static int igt_vma_rotate_remap(void *arg)
}
i915_vma_unpin(vma);
-
+ err = i915_vma_unbind(vma);
+ if (err) {
+ pr_err("Unbinding returned %i\n", err);
+ goto out_object;
+ }
cond_resched();
}
}
@@ -848,6 +852,11 @@ static int igt_vma_partial(void *arg)
i915_vma_unpin(vma);
nvma++;
+ err = i915_vma_unbind(vma);
+ if (err) {
+ pr_err("Unbinding returned %i\n", err);
+ goto out_object;
+ }
cond_resched();
}
@@ -882,6 +891,12 @@ static int igt_vma_partial(void *arg)
i915_vma_unpin(vma);
+ err = i915_vma_unbind(vma);
+ if (err) {
+ pr_err("Unbinding returned %i\n", err);
+ goto out_object;
+ }
+
count = 0;
list_for_each_entry(vma, &obj->vma.list, obj_link)
count++;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 07/17] drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking changes.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Now that we require locking to evict, multiple vmas from the same object
might not be evicted. This is expected and required, because execbuf will
move to short-term pinning by using the lock only. This will cause these
tests to fail, because they create a ton of vma's for the same object.
Unbind manually to prevent spurious -ENOSPC in those mock tests.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/selftests/i915_vma.c | 17 ++++++++++++++++-
1 file changed, 16 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/selftests/i915_vma.c b/drivers/gpu/drm/i915/selftests/i915_vma.c
index 1f10fe36619b..5c5809dfe9b2 100644
--- a/drivers/gpu/drm/i915/selftests/i915_vma.c
+++ b/drivers/gpu/drm/i915/selftests/i915_vma.c
@@ -691,7 +691,11 @@ static int igt_vma_rotate_remap(void *arg)
}
i915_vma_unpin(vma);
-
+ err = i915_vma_unbind(vma);
+ if (err) {
+ pr_err("Unbinding returned %i\n", err);
+ goto out_object;
+ }
cond_resched();
}
}
@@ -848,6 +852,11 @@ static int igt_vma_partial(void *arg)
i915_vma_unpin(vma);
nvma++;
+ err = i915_vma_unbind(vma);
+ if (err) {
+ pr_err("Unbinding returned %i\n", err);
+ goto out_object;
+ }
cond_resched();
}
@@ -882,6 +891,12 @@ static int igt_vma_partial(void *arg)
i915_vma_unpin(vma);
+ err = i915_vma_unbind(vma);
+ if (err) {
+ pr_err("Unbinding returned %i\n", err);
+ goto out_object;
+ }
+
count = 0;
list_for_each_entry(vma, &obj->vma.list, obj_link)
count++;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 08/17] drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Now that we cannot unbind kill the currently locked object directly
because we're removing short term pinning, we may have to unbind the
object from gtt manually, using a i915_gem_evict_vm() call.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_mman.c | 18 ++++++++++++++++--
1 file changed, 16 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
index af81d6c3332a..00cd9642669a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
@@ -358,8 +358,22 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
}
- /* The entire mappable GGTT is pinned? Unexpected! */
- GEM_BUG_ON(vma == ERR_PTR(-ENOSPC));
+ /*
+ * The entire mappable GGTT is pinned? Unexpected!
+ * Try to evict the object we locked too, as normally we skip it
+ * due to lack of short term pinning inside execbuf.
+ */
+ if (vma == ERR_PTR(-ENOSPC)) {
+ ret = mutex_lock_interruptible(&ggtt->vm.mutex);
+ if (!ret) {
+ ret = i915_gem_evict_vm(&ggtt->vm);
+ mutex_unlock(&ggtt->vm.mutex);
+ }
+ if (ret)
+ goto err_reset;
+ vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
+ }
+ GEM_WARN_ON(vma == ERR_PTR(-ENOSPC));
}
if (IS_ERR(vma)) {
ret = PTR_ERR(vma);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 08/17] drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Now that we cannot unbind kill the currently locked object directly
because we're removing short term pinning, we may have to unbind the
object from gtt manually, using a i915_gem_evict_vm() call.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_mman.c | 18 ++++++++++++++++--
1 file changed, 16 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
index af81d6c3332a..00cd9642669a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
@@ -358,8 +358,22 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
}
- /* The entire mappable GGTT is pinned? Unexpected! */
- GEM_BUG_ON(vma == ERR_PTR(-ENOSPC));
+ /*
+ * The entire mappable GGTT is pinned? Unexpected!
+ * Try to evict the object we locked too, as normally we skip it
+ * due to lack of short term pinning inside execbuf.
+ */
+ if (vma == ERR_PTR(-ENOSPC)) {
+ ret = mutex_lock_interruptible(&ggtt->vm.mutex);
+ if (!ret) {
+ ret = i915_gem_evict_vm(&ggtt->vm);
+ mutex_unlock(&ggtt->vm.mutex);
+ }
+ if (ret)
+ goto err_reset;
+ vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
+ }
+ GEM_WARN_ON(vma == ERR_PTR(-ENOSPC));
}
if (IS_ERR(vma)) {
ret = PTR_ERR(vma);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 09/17] drm/i915: Trylock the object when shrinking
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
We're working on requiring the obj->resv lock during unbind, fix
the shrinker to take the objectl ock.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
index eebff4735781..ad2123369e0d 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
@@ -405,12 +405,18 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr
list_for_each_entry_safe(vma, next,
&i915->ggtt.vm.bound_list, vm_link) {
unsigned long count = vma->node.size >> PAGE_SHIFT;
+ struct drm_i915_gem_object *obj = vma->obj;
if (!vma->iomap || i915_vma_is_active(vma))
continue;
+ if (!i915_gem_object_trylock(obj))
+ continue;
+
if (__i915_vma_unbind(vma) == 0)
freed_pages += count;
+
+ i915_gem_object_unlock(obj);
}
mutex_unlock(&i915->ggtt.vm.mutex);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 09/17] drm/i915: Trylock the object when shrinking
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
We're working on requiring the obj->resv lock during unbind, fix
the shrinker to take the objectl ock.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
index eebff4735781..ad2123369e0d 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
@@ -405,12 +405,18 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr
list_for_each_entry_safe(vma, next,
&i915->ggtt.vm.bound_list, vm_link) {
unsigned long count = vma->node.size >> PAGE_SHIFT;
+ struct drm_i915_gem_object *obj = vma->obj;
if (!vma->iomap || i915_vma_is_active(vma))
continue;
+ if (!i915_gem_object_trylock(obj))
+ continue;
+
if (__i915_vma_unbind(vma) == 0)
freed_pages += count;
+
+ i915_gem_object_unlock(obj);
}
mutex_unlock(&i915->ggtt.vm.mutex);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 10/17] drm/i915: Require object lock when freeing pages during destruction
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
TTM already requires this, and we require it for delayed destroy.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index 5fac9b560b73..39cd563544a5 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -262,6 +262,8 @@ static void __i915_gem_object_free_mmaps(struct drm_i915_gem_object *obj)
*/
void __i915_gem_object_pages_fini(struct drm_i915_gem_object *obj)
{
+ assert_object_held(obj);
+
if (!list_empty(&obj->vma.list)) {
struct i915_vma *vma;
@@ -328,7 +330,10 @@ static void __i915_gem_free_objects(struct drm_i915_private *i915,
obj->ops->delayed_free(obj);
continue;
}
+
+ i915_gem_object_lock(obj, NULL);
__i915_gem_object_pages_fini(obj);
+ i915_gem_object_unlock(obj);
__i915_gem_free_object(obj);
/* But keep the pointer alive for RCU-protected lookups */
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 10/17] drm/i915: Require object lock when freeing pages during destruction
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
TTM already requires this, and we require it for delayed destroy.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index 5fac9b560b73..39cd563544a5 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -262,6 +262,8 @@ static void __i915_gem_object_free_mmaps(struct drm_i915_gem_object *obj)
*/
void __i915_gem_object_pages_fini(struct drm_i915_gem_object *obj)
{
+ assert_object_held(obj);
+
if (!list_empty(&obj->vma.list)) {
struct i915_vma *vma;
@@ -328,7 +330,10 @@ static void __i915_gem_free_objects(struct drm_i915_private *i915,
obj->ops->delayed_free(obj);
continue;
}
+
+ i915_gem_object_lock(obj, NULL);
__i915_gem_object_pages_fini(obj);
+ i915_gem_object_unlock(obj);
__i915_gem_free_object(obj);
/* But keep the pointer alive for RCU-protected lookups */
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 11/17] drm/i915: Add ww ctx to i915_gem_object_trylock
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
This is required for i915_gem_evict_vm, to be able to evict the entire VM,
including objects that are already locked to the current ww ctx.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_object.h | 8 ++++++--
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 4 ++--
drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 2 +-
drivers/gpu/drm/i915/gt/intel_engine_pm.c | 2 +-
drivers/gpu/drm/i915/gt/mock_engine.c | 2 +-
drivers/gpu/drm/i915/gt/selftest_migrate.c | 2 +-
6 files changed, 12 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h
index 66f20b803b01..f66d46882ea7 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
@@ -210,9 +210,13 @@ static inline int i915_gem_object_lock_interruptible(struct drm_i915_gem_object
return __i915_gem_object_lock(obj, ww, true);
}
-static inline bool i915_gem_object_trylock(struct drm_i915_gem_object *obj)
+static inline bool i915_gem_object_trylock(struct drm_i915_gem_object *obj,
+ struct i915_gem_ww_ctx *ww)
{
- return dma_resv_trylock(obj->base.resv);
+ if (!ww)
+ return dma_resv_trylock(obj->base.resv);
+ else
+ return ww_mutex_trylock(&obj->base.resv->lock, &ww->ctx);
}
static inline void i915_gem_object_unlock(struct drm_i915_gem_object *obj)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
index ad2123369e0d..85ff182ddbc2 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
@@ -216,7 +216,7 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww,
/* May arrive from get_pages on another bo */
if (!ww) {
- if (!i915_gem_object_trylock(obj))
+ if (!i915_gem_object_trylock(obj, NULL))
goto skip;
} else {
err = i915_gem_object_lock(obj, ww);
@@ -410,7 +410,7 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr
if (!vma->iomap || i915_vma_is_active(vma))
continue;
- if (!i915_gem_object_trylock(obj))
+ if (!i915_gem_object_trylock(obj, NULL))
continue;
if (__i915_vma_unbind(vma) == 0)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
index ad7a8e9e13e7..7df50fd6cc7b 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
@@ -656,7 +656,7 @@ static int __i915_gem_object_create_stolen(struct intel_memory_region *mem,
cache_level = HAS_LLC(mem->i915) ? I915_CACHE_LLC : I915_CACHE_NONE;
i915_gem_object_set_cache_coherency(obj, cache_level);
- if (WARN_ON(!i915_gem_object_trylock(obj)))
+ if (WARN_ON(!i915_gem_object_trylock(obj, NULL)))
return -EBUSY;
i915_gem_object_init_memory_region(obj, mem);
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pm.c b/drivers/gpu/drm/i915/gt/intel_engine_pm.c
index a8a2ad44b7e3..b0a4a2dbe3ee 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.c
@@ -26,7 +26,7 @@ static void dbg_poison_ce(struct intel_context *ce)
int type = i915_coherent_map_type(ce->engine->i915, obj, true);
void *map;
- if (!i915_gem_object_trylock(obj))
+ if (!i915_gem_object_trylock(obj, NULL))
return;
map = i915_gem_object_pin_map(obj, type);
diff --git a/drivers/gpu/drm/i915/gt/mock_engine.c b/drivers/gpu/drm/i915/gt/mock_engine.c
index bb99fc03f503..240265354274 100644
--- a/drivers/gpu/drm/i915/gt/mock_engine.c
+++ b/drivers/gpu/drm/i915/gt/mock_engine.c
@@ -17,7 +17,7 @@ static int mock_timeline_pin(struct intel_timeline *tl)
{
int err;
- if (WARN_ON(!i915_gem_object_trylock(tl->hwsp_ggtt->obj)))
+ if (WARN_ON(!i915_gem_object_trylock(tl->hwsp_ggtt->obj, NULL)))
return -EBUSY;
err = intel_timeline_pin_map(tl);
diff --git a/drivers/gpu/drm/i915/gt/selftest_migrate.c b/drivers/gpu/drm/i915/gt/selftest_migrate.c
index e21787301bbd..cb574ea6b29d 100644
--- a/drivers/gpu/drm/i915/gt/selftest_migrate.c
+++ b/drivers/gpu/drm/i915/gt/selftest_migrate.c
@@ -465,7 +465,7 @@ create_init_lmem_internal(struct intel_gt *gt, size_t sz, bool try_lmem)
return obj;
}
- i915_gem_object_trylock(obj);
+ i915_gem_object_trylock(obj, NULL);
err = i915_gem_object_pin_pages(obj);
if (err) {
i915_gem_object_unlock(obj);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 11/17] drm/i915: Add ww ctx to i915_gem_object_trylock
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
This is required for i915_gem_evict_vm, to be able to evict the entire VM,
including objects that are already locked to the current ww ctx.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_object.h | 8 ++++++--
drivers/gpu/drm/i915/gem/i915_gem_shrinker.c | 4 ++--
drivers/gpu/drm/i915/gem/i915_gem_stolen.c | 2 +-
drivers/gpu/drm/i915/gt/intel_engine_pm.c | 2 +-
drivers/gpu/drm/i915/gt/mock_engine.c | 2 +-
drivers/gpu/drm/i915/gt/selftest_migrate.c | 2 +-
6 files changed, 12 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h
index 66f20b803b01..f66d46882ea7 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
@@ -210,9 +210,13 @@ static inline int i915_gem_object_lock_interruptible(struct drm_i915_gem_object
return __i915_gem_object_lock(obj, ww, true);
}
-static inline bool i915_gem_object_trylock(struct drm_i915_gem_object *obj)
+static inline bool i915_gem_object_trylock(struct drm_i915_gem_object *obj,
+ struct i915_gem_ww_ctx *ww)
{
- return dma_resv_trylock(obj->base.resv);
+ if (!ww)
+ return dma_resv_trylock(obj->base.resv);
+ else
+ return ww_mutex_trylock(&obj->base.resv->lock, &ww->ctx);
}
static inline void i915_gem_object_unlock(struct drm_i915_gem_object *obj)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
index ad2123369e0d..85ff182ddbc2 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
@@ -216,7 +216,7 @@ i915_gem_shrink(struct i915_gem_ww_ctx *ww,
/* May arrive from get_pages on another bo */
if (!ww) {
- if (!i915_gem_object_trylock(obj))
+ if (!i915_gem_object_trylock(obj, NULL))
goto skip;
} else {
err = i915_gem_object_lock(obj, ww);
@@ -410,7 +410,7 @@ i915_gem_shrinker_vmap(struct notifier_block *nb, unsigned long event, void *ptr
if (!vma->iomap || i915_vma_is_active(vma))
continue;
- if (!i915_gem_object_trylock(obj))
+ if (!i915_gem_object_trylock(obj, NULL))
continue;
if (__i915_vma_unbind(vma) == 0)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
index ad7a8e9e13e7..7df50fd6cc7b 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_stolen.c
@@ -656,7 +656,7 @@ static int __i915_gem_object_create_stolen(struct intel_memory_region *mem,
cache_level = HAS_LLC(mem->i915) ? I915_CACHE_LLC : I915_CACHE_NONE;
i915_gem_object_set_cache_coherency(obj, cache_level);
- if (WARN_ON(!i915_gem_object_trylock(obj)))
+ if (WARN_ON(!i915_gem_object_trylock(obj, NULL)))
return -EBUSY;
i915_gem_object_init_memory_region(obj, mem);
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pm.c b/drivers/gpu/drm/i915/gt/intel_engine_pm.c
index a8a2ad44b7e3..b0a4a2dbe3ee 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.c
@@ -26,7 +26,7 @@ static void dbg_poison_ce(struct intel_context *ce)
int type = i915_coherent_map_type(ce->engine->i915, obj, true);
void *map;
- if (!i915_gem_object_trylock(obj))
+ if (!i915_gem_object_trylock(obj, NULL))
return;
map = i915_gem_object_pin_map(obj, type);
diff --git a/drivers/gpu/drm/i915/gt/mock_engine.c b/drivers/gpu/drm/i915/gt/mock_engine.c
index bb99fc03f503..240265354274 100644
--- a/drivers/gpu/drm/i915/gt/mock_engine.c
+++ b/drivers/gpu/drm/i915/gt/mock_engine.c
@@ -17,7 +17,7 @@ static int mock_timeline_pin(struct intel_timeline *tl)
{
int err;
- if (WARN_ON(!i915_gem_object_trylock(tl->hwsp_ggtt->obj)))
+ if (WARN_ON(!i915_gem_object_trylock(tl->hwsp_ggtt->obj, NULL)))
return -EBUSY;
err = intel_timeline_pin_map(tl);
diff --git a/drivers/gpu/drm/i915/gt/selftest_migrate.c b/drivers/gpu/drm/i915/gt/selftest_migrate.c
index e21787301bbd..cb574ea6b29d 100644
--- a/drivers/gpu/drm/i915/gt/selftest_migrate.c
+++ b/drivers/gpu/drm/i915/gt/selftest_migrate.c
@@ -465,7 +465,7 @@ create_init_lmem_internal(struct intel_gt *gt, size_t sz, bool try_lmem)
return obj;
}
- i915_gem_object_trylock(obj);
+ i915_gem_object_trylock(obj, NULL);
err = i915_gem_object_pin_pages(obj);
if (err) {
i915_gem_object_unlock(obj);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 12/17] drm/i915: Add locking to i915_gem_evict_vm()
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
i915_gem_evict_vm will need to be able to evict objects that are
locked by the current ctx. By testing if the current context already
locked the object, we can do this correctly. This allows us to
evict the entire vm even if we already hold some objects' locks.
Previously, this was spread over several commits, but it makes
more sense to commit the changes to i915_gem_evict_vm separately
from the changes to i915_gem_evict_something() and
i915_gem_evict_for_node().
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
.../gpu/drm/i915/gem/i915_gem_execbuffer.c | 2 +-
drivers/gpu/drm/i915/gem/i915_gem_mman.c | 2 +-
drivers/gpu/drm/i915/i915_drv.h | 3 +-
drivers/gpu/drm/i915/i915_gem_evict.c | 30 +++++++++++++++++--
drivers/gpu/drm/i915/i915_vma.c | 7 ++++-
.../gpu/drm/i915/selftests/i915_gem_evict.c | 10 +++++--
6 files changed, 46 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 2213f7b613da..eb3649e844ff 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -766,7 +766,7 @@ static int eb_reserve(struct i915_execbuffer *eb)
case 1:
/* Too fragmented, unbind everything and retry */
mutex_lock(&eb->context->vm->mutex);
- err = i915_gem_evict_vm(eb->context->vm);
+ err = i915_gem_evict_vm(eb->context->vm, &eb->ww);
mutex_unlock(&eb->context->vm->mutex);
if (err)
return err;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
index 00cd9642669a..2856098cb449 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
@@ -366,7 +366,7 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
if (vma == ERR_PTR(-ENOSPC)) {
ret = mutex_lock_interruptible(&ggtt->vm.mutex);
if (!ret) {
- ret = i915_gem_evict_vm(&ggtt->vm);
+ ret = i915_gem_evict_vm(&ggtt->vm, &ww);
mutex_unlock(&ggtt->vm.mutex);
}
if (ret)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index d51628d10f9d..c180019c607f 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1725,7 +1725,8 @@ int __must_check i915_gem_evict_something(struct i915_address_space *vm,
int __must_check i915_gem_evict_for_node(struct i915_address_space *vm,
struct drm_mm_node *node,
unsigned int flags);
-int i915_gem_evict_vm(struct i915_address_space *vm);
+int i915_gem_evict_vm(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww);
/* i915_gem_internal.c */
struct drm_i915_gem_object *
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 2b73ddb11c66..bfd66f539fc1 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -367,7 +367,7 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
* To clarify: This is for freeing up virtual address space, not for freeing
* memory in e.g. the shrinker.
*/
-int i915_gem_evict_vm(struct i915_address_space *vm)
+int i915_gem_evict_vm(struct i915_address_space *vm, struct i915_gem_ww_ctx *ww)
{
int ret = 0;
@@ -388,24 +388,50 @@ int i915_gem_evict_vm(struct i915_address_space *vm)
do {
struct i915_vma *vma, *vn;
LIST_HEAD(eviction_list);
+ LIST_HEAD(locked_eviction_list);
list_for_each_entry(vma, &vm->bound_list, vm_link) {
if (i915_vma_is_pinned(vma))
continue;
+ /*
+ * If we already own the lock, trylock fails. In case the resv
+ * is shared among multiple objects, we still need the object ref.
+ */
+ if (ww && (dma_resv_locking_ctx(vma->obj->base.resv) == &ww->ctx)) {
+ __i915_vma_pin(vma);
+ list_add(&vma->evict_link, &locked_eviction_list);
+ continue;
+ }
+
+ if (!i915_gem_object_trylock(vma->obj, ww))
+ continue;
+
__i915_vma_pin(vma);
list_add(&vma->evict_link, &eviction_list);
}
- if (list_empty(&eviction_list))
+ if (list_empty(&eviction_list) && list_empty(&locked_eviction_list))
break;
ret = 0;
+ /* Unbind locked objects first, before unlocking the eviction_list */
+ list_for_each_entry_safe(vma, vn, &locked_eviction_list, evict_link) {
+ __i915_vma_unpin(vma);
+
+ if (ret == 0)
+ ret = __i915_vma_unbind(vma);
+ if (ret != -EINTR) /* "Get me out of here!" */
+ ret = 0;
+ }
+
list_for_each_entry_safe(vma, vn, &eviction_list, evict_link) {
__i915_vma_unpin(vma);
if (ret == 0)
ret = __i915_vma_unbind(vma);
if (ret != -EINTR) /* "Get me out of here!" */
ret = 0;
+
+ i915_gem_object_unlock(vma->obj);
}
} while (ret == 0);
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index de24e4b3b19b..d24e90eac948 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1462,7 +1462,12 @@ static int __i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
/* Unlike i915_vma_pin, we don't take no for an answer! */
flush_idle_contexts(vm->gt);
if (mutex_lock_interruptible(&vm->mutex) == 0) {
- i915_gem_evict_vm(vm);
+ /*
+ * We pass NULL ww here, as we don't want to unbind
+ * locked objects when called from execbuf when pinning
+ * is removed. This would probably regress badly.
+ */
+ i915_gem_evict_vm(vm, NULL);
mutex_unlock(&vm->mutex);
}
} while (1);
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
index 7e0658a77659..7178811366af 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
@@ -331,6 +331,7 @@ static int igt_evict_vm(void *arg)
{
struct intel_gt *gt = arg;
struct i915_ggtt *ggtt = gt->ggtt;
+ struct i915_gem_ww_ctx ww;
LIST_HEAD(objects);
int err;
@@ -342,7 +343,7 @@ static int igt_evict_vm(void *arg)
/* Everything is pinned, nothing should happen */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_vm(&ggtt->vm);
+ err = i915_gem_evict_vm(&ggtt->vm, NULL);
mutex_unlock(&ggtt->vm.mutex);
if (err) {
pr_err("i915_gem_evict_vm on a full GGTT returned err=%d]\n",
@@ -352,9 +353,14 @@ static int igt_evict_vm(void *arg)
unpin_ggtt(ggtt);
+ i915_gem_ww_ctx_init(&ww, false);
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_vm(&ggtt->vm);
+ err = i915_gem_evict_vm(&ggtt->vm, &ww);
mutex_unlock(&ggtt->vm.mutex);
+
+ /* no -EDEADLK handling; can't happen with vm.mutex in place */
+ i915_gem_ww_ctx_fini(&ww);
+
if (err) {
pr_err("i915_gem_evict_vm on a full GGTT returned err=%d]\n",
err);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 12/17] drm/i915: Add locking to i915_gem_evict_vm()
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
i915_gem_evict_vm will need to be able to evict objects that are
locked by the current ctx. By testing if the current context already
locked the object, we can do this correctly. This allows us to
evict the entire vm even if we already hold some objects' locks.
Previously, this was spread over several commits, but it makes
more sense to commit the changes to i915_gem_evict_vm separately
from the changes to i915_gem_evict_something() and
i915_gem_evict_for_node().
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
.../gpu/drm/i915/gem/i915_gem_execbuffer.c | 2 +-
drivers/gpu/drm/i915/gem/i915_gem_mman.c | 2 +-
drivers/gpu/drm/i915/i915_drv.h | 3 +-
drivers/gpu/drm/i915/i915_gem_evict.c | 30 +++++++++++++++++--
drivers/gpu/drm/i915/i915_vma.c | 7 ++++-
.../gpu/drm/i915/selftests/i915_gem_evict.c | 10 +++++--
6 files changed, 46 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 2213f7b613da..eb3649e844ff 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -766,7 +766,7 @@ static int eb_reserve(struct i915_execbuffer *eb)
case 1:
/* Too fragmented, unbind everything and retry */
mutex_lock(&eb->context->vm->mutex);
- err = i915_gem_evict_vm(eb->context->vm);
+ err = i915_gem_evict_vm(eb->context->vm, &eb->ww);
mutex_unlock(&eb->context->vm->mutex);
if (err)
return err;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
index 00cd9642669a..2856098cb449 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
@@ -366,7 +366,7 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
if (vma == ERR_PTR(-ENOSPC)) {
ret = mutex_lock_interruptible(&ggtt->vm.mutex);
if (!ret) {
- ret = i915_gem_evict_vm(&ggtt->vm);
+ ret = i915_gem_evict_vm(&ggtt->vm, &ww);
mutex_unlock(&ggtt->vm.mutex);
}
if (ret)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index d51628d10f9d..c180019c607f 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1725,7 +1725,8 @@ int __must_check i915_gem_evict_something(struct i915_address_space *vm,
int __must_check i915_gem_evict_for_node(struct i915_address_space *vm,
struct drm_mm_node *node,
unsigned int flags);
-int i915_gem_evict_vm(struct i915_address_space *vm);
+int i915_gem_evict_vm(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww);
/* i915_gem_internal.c */
struct drm_i915_gem_object *
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index 2b73ddb11c66..bfd66f539fc1 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -367,7 +367,7 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
* To clarify: This is for freeing up virtual address space, not for freeing
* memory in e.g. the shrinker.
*/
-int i915_gem_evict_vm(struct i915_address_space *vm)
+int i915_gem_evict_vm(struct i915_address_space *vm, struct i915_gem_ww_ctx *ww)
{
int ret = 0;
@@ -388,24 +388,50 @@ int i915_gem_evict_vm(struct i915_address_space *vm)
do {
struct i915_vma *vma, *vn;
LIST_HEAD(eviction_list);
+ LIST_HEAD(locked_eviction_list);
list_for_each_entry(vma, &vm->bound_list, vm_link) {
if (i915_vma_is_pinned(vma))
continue;
+ /*
+ * If we already own the lock, trylock fails. In case the resv
+ * is shared among multiple objects, we still need the object ref.
+ */
+ if (ww && (dma_resv_locking_ctx(vma->obj->base.resv) == &ww->ctx)) {
+ __i915_vma_pin(vma);
+ list_add(&vma->evict_link, &locked_eviction_list);
+ continue;
+ }
+
+ if (!i915_gem_object_trylock(vma->obj, ww))
+ continue;
+
__i915_vma_pin(vma);
list_add(&vma->evict_link, &eviction_list);
}
- if (list_empty(&eviction_list))
+ if (list_empty(&eviction_list) && list_empty(&locked_eviction_list))
break;
ret = 0;
+ /* Unbind locked objects first, before unlocking the eviction_list */
+ list_for_each_entry_safe(vma, vn, &locked_eviction_list, evict_link) {
+ __i915_vma_unpin(vma);
+
+ if (ret == 0)
+ ret = __i915_vma_unbind(vma);
+ if (ret != -EINTR) /* "Get me out of here!" */
+ ret = 0;
+ }
+
list_for_each_entry_safe(vma, vn, &eviction_list, evict_link) {
__i915_vma_unpin(vma);
if (ret == 0)
ret = __i915_vma_unbind(vma);
if (ret != -EINTR) /* "Get me out of here!" */
ret = 0;
+
+ i915_gem_object_unlock(vma->obj);
}
} while (ret == 0);
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index de24e4b3b19b..d24e90eac948 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1462,7 +1462,12 @@ static int __i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
/* Unlike i915_vma_pin, we don't take no for an answer! */
flush_idle_contexts(vm->gt);
if (mutex_lock_interruptible(&vm->mutex) == 0) {
- i915_gem_evict_vm(vm);
+ /*
+ * We pass NULL ww here, as we don't want to unbind
+ * locked objects when called from execbuf when pinning
+ * is removed. This would probably regress badly.
+ */
+ i915_gem_evict_vm(vm, NULL);
mutex_unlock(&vm->mutex);
}
} while (1);
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
index 7e0658a77659..7178811366af 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
@@ -331,6 +331,7 @@ static int igt_evict_vm(void *arg)
{
struct intel_gt *gt = arg;
struct i915_ggtt *ggtt = gt->ggtt;
+ struct i915_gem_ww_ctx ww;
LIST_HEAD(objects);
int err;
@@ -342,7 +343,7 @@ static int igt_evict_vm(void *arg)
/* Everything is pinned, nothing should happen */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_vm(&ggtt->vm);
+ err = i915_gem_evict_vm(&ggtt->vm, NULL);
mutex_unlock(&ggtt->vm.mutex);
if (err) {
pr_err("i915_gem_evict_vm on a full GGTT returned err=%d]\n",
@@ -352,9 +353,14 @@ static int igt_evict_vm(void *arg)
unpin_ggtt(ggtt);
+ i915_gem_ww_ctx_init(&ww, false);
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_vm(&ggtt->vm);
+ err = i915_gem_evict_vm(&ggtt->vm, &ww);
mutex_unlock(&ggtt->vm.mutex);
+
+ /* no -EDEADLK handling; can't happen with vm.mutex in place */
+ i915_gem_ww_ctx_fini(&ww);
+
if (err) {
pr_err("i915_gem_evict_vm on a full GGTT returned err=%d]\n",
err);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 13/17] drm/i915: Add object locking to i915_gem_evict_for_node and i915_gem_evict_something
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Because we will start to require the obj->resv lock for unbinding,
ensure these shrinker functions also take the lock.
This requires some function signature changes, to ensure that the
ww context is passed around, but is mostly straightforward.
Previously this was split up into several patches, but reworking
should allow for easier bisection.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/gt/intel_ggtt.c | 2 +-
drivers/gpu/drm/i915/gt/selftest_hangcheck.c | 2 +-
drivers/gpu/drm/i915/gvt/aperture_gm.c | 2 +-
drivers/gpu/drm/i915/i915_drv.h | 2 ++
drivers/gpu/drm/i915/i915_gem_evict.c | 34 +++++++++++++++----
drivers/gpu/drm/i915/i915_gem_gtt.c | 8 +++--
drivers/gpu/drm/i915/i915_gem_gtt.h | 3 ++
drivers/gpu/drm/i915/i915_vgpu.c | 2 +-
drivers/gpu/drm/i915/i915_vma.c | 9 ++---
.../gpu/drm/i915/selftests/i915_gem_evict.c | 17 +++++-----
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 14 ++++----
11 files changed, 63 insertions(+), 32 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index a287c9186ec9..ed43e9c80aaa 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -504,7 +504,7 @@ static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt)
GEM_BUG_ON(ggtt->vm.total <= GUC_GGTT_TOP);
size = ggtt->vm.total - GUC_GGTT_TOP;
- ret = i915_gem_gtt_reserve(&ggtt->vm, &ggtt->uc_fw, size,
+ ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, &ggtt->uc_fw, size,
GUC_GGTT_TOP, I915_COLOR_UNEVICTABLE,
PIN_NOEVICT);
if (ret)
diff --git a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
index e5ad4d5a91c0..a3597a6bb805 100644
--- a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
+++ b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
@@ -1382,7 +1382,7 @@ static int evict_vma(void *data)
complete(&arg->completion);
mutex_lock(&vm->mutex);
- err = i915_gem_evict_for_node(vm, &evict, 0);
+ err = i915_gem_evict_for_node(vm, NULL, &evict, 0);
mutex_unlock(&vm->mutex);
return err;
diff --git a/drivers/gpu/drm/i915/gvt/aperture_gm.c b/drivers/gpu/drm/i915/gvt/aperture_gm.c
index 0d6d59871308..c08098a167e9 100644
--- a/drivers/gpu/drm/i915/gvt/aperture_gm.c
+++ b/drivers/gpu/drm/i915/gvt/aperture_gm.c
@@ -63,7 +63,7 @@ static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm)
mutex_lock(>->ggtt->vm.mutex);
mmio_hw_access_pre(gt);
- ret = i915_gem_gtt_insert(>->ggtt->vm, node,
+ ret = i915_gem_gtt_insert(>->ggtt->vm, NULL, node,
size, I915_GTT_PAGE_SIZE,
I915_COLOR_UNEVICTABLE,
start, end, flags);
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index c180019c607f..2a98192a89c1 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1718,11 +1718,13 @@ i915_gem_vm_lookup(struct drm_i915_file_private *file_priv, u32 id)
/* i915_gem_evict.c */
int __must_check i915_gem_evict_something(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
u64 min_size, u64 alignment,
unsigned long color,
u64 start, u64 end,
unsigned flags);
int __must_check i915_gem_evict_for_node(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
unsigned int flags);
int i915_gem_evict_vm(struct i915_address_space *vm,
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index bfd66f539fc1..f502a617b35c 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -51,6 +51,7 @@ static int ggtt_flush(struct intel_gt *gt)
static bool
mark_free(struct drm_mm_scan *scan,
+ struct i915_gem_ww_ctx *ww,
struct i915_vma *vma,
unsigned int flags,
struct list_head *unwind)
@@ -58,6 +59,9 @@ mark_free(struct drm_mm_scan *scan,
if (i915_vma_is_pinned(vma))
return false;
+ if (!i915_gem_object_trylock(vma->obj, ww))
+ return false;
+
list_add(&vma->evict_link, unwind);
return drm_mm_scan_add_block(scan, &vma->node);
}
@@ -98,6 +102,7 @@ static bool defer_evict(struct i915_vma *vma)
*/
int
i915_gem_evict_something(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
u64 min_size, u64 alignment,
unsigned long color,
u64 start, u64 end,
@@ -170,7 +175,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
continue;
}
- if (mark_free(&scan, vma, flags, &eviction_list))
+ if (mark_free(&scan, ww, vma, flags, &eviction_list))
goto found;
}
@@ -178,6 +183,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
ret = drm_mm_scan_remove_block(&scan, &vma->node);
BUG_ON(ret);
+ i915_gem_object_unlock(vma->obj);
}
/*
@@ -222,10 +228,12 @@ i915_gem_evict_something(struct i915_address_space *vm,
* of any of our objects, thus corrupting the list).
*/
list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
- if (drm_mm_scan_remove_block(&scan, &vma->node))
+ if (drm_mm_scan_remove_block(&scan, &vma->node)) {
__i915_vma_pin(vma);
- else
+ } else {
list_del(&vma->evict_link);
+ i915_gem_object_unlock(vma->obj);
+ }
}
/* Unbinding will emit any required flushes */
@@ -234,16 +242,22 @@ i915_gem_evict_something(struct i915_address_space *vm,
__i915_vma_unpin(vma);
if (ret == 0)
ret = __i915_vma_unbind(vma);
+
+ i915_gem_object_unlock(vma->obj);
}
while (ret == 0 && (node = drm_mm_scan_color_evict(&scan))) {
vma = container_of(node, struct i915_vma, node);
+
/* If we find any non-objects (!vma), we cannot evict them */
- if (vma->node.color != I915_COLOR_UNEVICTABLE)
+ if (vma->node.color != I915_COLOR_UNEVICTABLE &&
+ i915_gem_object_trylock(vma->obj, ww)) {
ret = __i915_vma_unbind(vma);
- else
- ret = -ENOSPC; /* XXX search failed, try again? */
+ i915_gem_object_unlock(vma->obj);
+ } else {
+ ret = -ENOSPC;
+ }
}
return ret;
@@ -261,6 +275,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
* memory in e.g. the shrinker.
*/
int i915_gem_evict_for_node(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *target,
unsigned int flags)
{
@@ -333,6 +348,11 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
break;
}
+ if (!i915_gem_object_trylock(vma->obj, ww)) {
+ ret = -ENOSPC;
+ break;
+ }
+
/*
* Never show fear in the face of dragons!
*
@@ -350,6 +370,8 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
__i915_vma_unpin(vma);
if (ret == 0)
ret = __i915_vma_unbind(vma);
+
+ i915_gem_object_unlock(vma->obj);
}
return ret;
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index cd5f2348a187..c810da476c2b 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -93,6 +93,7 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
* asked to wait for eviction and interrupted.
*/
int i915_gem_gtt_reserve(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
u64 size, u64 offset, unsigned long color,
unsigned int flags)
@@ -117,7 +118,7 @@ int i915_gem_gtt_reserve(struct i915_address_space *vm,
if (flags & PIN_NOEVICT)
return -ENOSPC;
- err = i915_gem_evict_for_node(vm, node, flags);
+ err = i915_gem_evict_for_node(vm, ww, node, flags);
if (err == 0)
err = drm_mm_reserve_node(&vm->mm, node);
@@ -184,6 +185,7 @@ static u64 random_offset(u64 start, u64 end, u64 len, u64 align)
* asked to wait for eviction and interrupted.
*/
int i915_gem_gtt_insert(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
u64 size, u64 alignment, unsigned long color,
u64 start, u64 end, unsigned int flags)
@@ -269,7 +271,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
*/
offset = random_offset(start, end,
size, alignment ?: I915_GTT_MIN_ALIGNMENT);
- err = i915_gem_gtt_reserve(vm, node, size, offset, color, flags);
+ err = i915_gem_gtt_reserve(vm, ww, node, size, offset, color, flags);
if (err != -ENOSPC)
return err;
@@ -277,7 +279,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
return -ENOSPC;
/* Randomly selected placement is pinned, do a search */
- err = i915_gem_evict_something(vm, size, alignment, color,
+ err = i915_gem_evict_something(vm, ww, size, alignment, color,
start, end, flags);
if (err)
return err;
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index c9b0ee5e1d23..e4938aba3fe9 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -16,6 +16,7 @@
struct drm_i915_gem_object;
struct i915_address_space;
+struct i915_gem_ww_ctx;
int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
struct sg_table *pages);
@@ -23,11 +24,13 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
struct sg_table *pages);
int i915_gem_gtt_reserve(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
u64 size, u64 offset, unsigned long color,
unsigned int flags);
int i915_gem_gtt_insert(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
u64 size, u64 alignment, unsigned long color,
u64 start, u64 end, unsigned int flags);
diff --git a/drivers/gpu/drm/i915/i915_vgpu.c b/drivers/gpu/drm/i915/i915_vgpu.c
index 31a105bc1792..c97323973f9b 100644
--- a/drivers/gpu/drm/i915/i915_vgpu.c
+++ b/drivers/gpu/drm/i915/i915_vgpu.c
@@ -197,7 +197,7 @@ static int vgt_balloon_space(struct i915_ggtt *ggtt,
drm_info(&dev_priv->drm,
"balloon space: range [ 0x%lx - 0x%lx ] %lu KiB.\n",
start, end, size / 1024);
- ret = i915_gem_gtt_reserve(&ggtt->vm, node,
+ ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, node,
size, start, I915_COLOR_UNEVICTABLE,
0);
if (!ret)
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index d24e90eac948..83df1290df5e 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -650,7 +650,8 @@ bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long color)
* 0 on success, negative error code otherwise.
*/
static int
-i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
+i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
+ u64 size, u64 alignment, u64 flags)
{
unsigned long color;
u64 start, end;
@@ -702,7 +703,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
range_overflows(offset, size, end))
return -EINVAL;
- ret = i915_gem_gtt_reserve(vma->vm, &vma->node,
+ ret = i915_gem_gtt_reserve(vma->vm, ww, &vma->node,
size, offset, color,
flags);
if (ret)
@@ -741,7 +742,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
size = round_up(size, I915_GTT_PAGE_SIZE_2M);
}
- ret = i915_gem_gtt_insert(vma->vm, &vma->node,
+ ret = i915_gem_gtt_insert(vma->vm, ww, &vma->node,
size, alignment, color,
start, end, flags);
if (ret)
@@ -1382,7 +1383,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
goto err_unlock;
if (!(bound & I915_VMA_BIND_MASK)) {
- err = i915_vma_insert(vma, size, alignment, flags);
+ err = i915_vma_insert(vma, ww, size, alignment, flags);
if (err)
goto err_active;
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
index 7178811366af..c80e1483d603 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
@@ -117,7 +117,7 @@ static int igt_evict_something(void *arg)
/* Everything is pinned, nothing should happen */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_something(&ggtt->vm,
+ err = i915_gem_evict_something(&ggtt->vm, NULL,
I915_GTT_PAGE_SIZE, 0, 0,
0, U64_MAX,
0);
@@ -132,11 +132,12 @@ static int igt_evict_something(void *arg)
/* Everything is unpinned, we should be able to evict something */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_something(&ggtt->vm,
+ err = i915_gem_evict_something(&ggtt->vm, NULL,
I915_GTT_PAGE_SIZE, 0, 0,
0, U64_MAX,
0);
mutex_unlock(&ggtt->vm.mutex);
+
if (err) {
pr_err("i915_gem_evict_something failed on a full GGTT with err=%d\n",
err);
@@ -204,7 +205,7 @@ static int igt_evict_for_vma(void *arg)
/* Everything is pinned, nothing should happen */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
+ err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
mutex_unlock(&ggtt->vm.mutex);
if (err != -ENOSPC) {
pr_err("i915_gem_evict_for_node on a full GGTT returned err=%d\n",
@@ -216,7 +217,7 @@ static int igt_evict_for_vma(void *arg)
/* Everything is unpinned, we should be able to evict the node */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
+ err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
mutex_unlock(&ggtt->vm.mutex);
if (err) {
pr_err("i915_gem_evict_for_node returned err=%d\n",
@@ -297,7 +298,7 @@ static int igt_evict_for_cache_color(void *arg)
/* Remove just the second vma */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
+ err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
mutex_unlock(&ggtt->vm.mutex);
if (err) {
pr_err("[0]i915_gem_evict_for_node returned err=%d\n", err);
@@ -310,7 +311,7 @@ static int igt_evict_for_cache_color(void *arg)
target.color = I915_CACHE_L3_LLC;
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
+ err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
mutex_unlock(&ggtt->vm.mutex);
if (!err) {
pr_err("[1]i915_gem_evict_for_node returned err=%d\n", err);
@@ -408,7 +409,7 @@ static int igt_evict_contexts(void *arg)
/* Reserve a block so that we know we have enough to fit a few rq */
memset(&hole, 0, sizeof(hole));
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &hole,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &hole,
PRETEND_GGTT_SIZE, 0, I915_COLOR_UNEVICTABLE,
0, ggtt->vm.total,
PIN_NOEVICT);
@@ -428,7 +429,7 @@ static int igt_evict_contexts(void *arg)
goto out_locked;
}
- if (i915_gem_gtt_insert(&ggtt->vm, &r->node,
+ if (i915_gem_gtt_insert(&ggtt->vm, NULL, &r->node,
1ul << 20, 0, I915_COLOR_UNEVICTABLE,
0, ggtt->vm.total,
PIN_NOEVICT)) {
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 88b3a4d9e27e..169e2330b386 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -1378,7 +1378,7 @@ static int igt_gtt_reserve(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
obj->base.size,
total,
obj->cache_level,
@@ -1430,7 +1430,7 @@ static int igt_gtt_reserve(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
obj->base.size,
total,
obj->cache_level,
@@ -1477,7 +1477,7 @@ static int igt_gtt_reserve(void *arg)
I915_GTT_MIN_ALIGNMENT);
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
obj->base.size,
offset,
obj->cache_level,
@@ -1552,7 +1552,7 @@ static int igt_gtt_insert(void *arg)
/* Check a couple of obviously invalid requests */
for (ii = invalid_insert; ii->size; ii++) {
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &tmp,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &tmp,
ii->size, ii->alignment,
I915_COLOR_UNEVICTABLE,
ii->start, ii->end,
@@ -1594,7 +1594,7 @@ static int igt_gtt_insert(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
obj->base.size, 0, obj->cache_level,
0, ggtt->vm.total,
0);
@@ -1654,7 +1654,7 @@ static int igt_gtt_insert(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
obj->base.size, 0, obj->cache_level,
0, ggtt->vm.total,
0);
@@ -1703,7 +1703,7 @@ static int igt_gtt_insert(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
obj->base.size, 0, obj->cache_level,
0, ggtt->vm.total,
0);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 13/17] drm/i915: Add object locking to i915_gem_evict_for_node and i915_gem_evict_something
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Because we will start to require the obj->resv lock for unbinding,
ensure these shrinker functions also take the lock.
This requires some function signature changes, to ensure that the
ww context is passed around, but is mostly straightforward.
Previously this was split up into several patches, but reworking
should allow for easier bisection.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
---
drivers/gpu/drm/i915/gt/intel_ggtt.c | 2 +-
drivers/gpu/drm/i915/gt/selftest_hangcheck.c | 2 +-
drivers/gpu/drm/i915/gvt/aperture_gm.c | 2 +-
drivers/gpu/drm/i915/i915_drv.h | 2 ++
drivers/gpu/drm/i915/i915_gem_evict.c | 34 +++++++++++++++----
drivers/gpu/drm/i915/i915_gem_gtt.c | 8 +++--
drivers/gpu/drm/i915/i915_gem_gtt.h | 3 ++
drivers/gpu/drm/i915/i915_vgpu.c | 2 +-
drivers/gpu/drm/i915/i915_vma.c | 9 ++---
.../gpu/drm/i915/selftests/i915_gem_evict.c | 17 +++++-----
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 14 ++++----
11 files changed, 63 insertions(+), 32 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index a287c9186ec9..ed43e9c80aaa 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -504,7 +504,7 @@ static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt)
GEM_BUG_ON(ggtt->vm.total <= GUC_GGTT_TOP);
size = ggtt->vm.total - GUC_GGTT_TOP;
- ret = i915_gem_gtt_reserve(&ggtt->vm, &ggtt->uc_fw, size,
+ ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, &ggtt->uc_fw, size,
GUC_GGTT_TOP, I915_COLOR_UNEVICTABLE,
PIN_NOEVICT);
if (ret)
diff --git a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
index e5ad4d5a91c0..a3597a6bb805 100644
--- a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
+++ b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
@@ -1382,7 +1382,7 @@ static int evict_vma(void *data)
complete(&arg->completion);
mutex_lock(&vm->mutex);
- err = i915_gem_evict_for_node(vm, &evict, 0);
+ err = i915_gem_evict_for_node(vm, NULL, &evict, 0);
mutex_unlock(&vm->mutex);
return err;
diff --git a/drivers/gpu/drm/i915/gvt/aperture_gm.c b/drivers/gpu/drm/i915/gvt/aperture_gm.c
index 0d6d59871308..c08098a167e9 100644
--- a/drivers/gpu/drm/i915/gvt/aperture_gm.c
+++ b/drivers/gpu/drm/i915/gvt/aperture_gm.c
@@ -63,7 +63,7 @@ static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm)
mutex_lock(>->ggtt->vm.mutex);
mmio_hw_access_pre(gt);
- ret = i915_gem_gtt_insert(>->ggtt->vm, node,
+ ret = i915_gem_gtt_insert(>->ggtt->vm, NULL, node,
size, I915_GTT_PAGE_SIZE,
I915_COLOR_UNEVICTABLE,
start, end, flags);
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index c180019c607f..2a98192a89c1 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1718,11 +1718,13 @@ i915_gem_vm_lookup(struct drm_i915_file_private *file_priv, u32 id)
/* i915_gem_evict.c */
int __must_check i915_gem_evict_something(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
u64 min_size, u64 alignment,
unsigned long color,
u64 start, u64 end,
unsigned flags);
int __must_check i915_gem_evict_for_node(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
unsigned int flags);
int i915_gem_evict_vm(struct i915_address_space *vm,
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index bfd66f539fc1..f502a617b35c 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -51,6 +51,7 @@ static int ggtt_flush(struct intel_gt *gt)
static bool
mark_free(struct drm_mm_scan *scan,
+ struct i915_gem_ww_ctx *ww,
struct i915_vma *vma,
unsigned int flags,
struct list_head *unwind)
@@ -58,6 +59,9 @@ mark_free(struct drm_mm_scan *scan,
if (i915_vma_is_pinned(vma))
return false;
+ if (!i915_gem_object_trylock(vma->obj, ww))
+ return false;
+
list_add(&vma->evict_link, unwind);
return drm_mm_scan_add_block(scan, &vma->node);
}
@@ -98,6 +102,7 @@ static bool defer_evict(struct i915_vma *vma)
*/
int
i915_gem_evict_something(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
u64 min_size, u64 alignment,
unsigned long color,
u64 start, u64 end,
@@ -170,7 +175,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
continue;
}
- if (mark_free(&scan, vma, flags, &eviction_list))
+ if (mark_free(&scan, ww, vma, flags, &eviction_list))
goto found;
}
@@ -178,6 +183,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
ret = drm_mm_scan_remove_block(&scan, &vma->node);
BUG_ON(ret);
+ i915_gem_object_unlock(vma->obj);
}
/*
@@ -222,10 +228,12 @@ i915_gem_evict_something(struct i915_address_space *vm,
* of any of our objects, thus corrupting the list).
*/
list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
- if (drm_mm_scan_remove_block(&scan, &vma->node))
+ if (drm_mm_scan_remove_block(&scan, &vma->node)) {
__i915_vma_pin(vma);
- else
+ } else {
list_del(&vma->evict_link);
+ i915_gem_object_unlock(vma->obj);
+ }
}
/* Unbinding will emit any required flushes */
@@ -234,16 +242,22 @@ i915_gem_evict_something(struct i915_address_space *vm,
__i915_vma_unpin(vma);
if (ret == 0)
ret = __i915_vma_unbind(vma);
+
+ i915_gem_object_unlock(vma->obj);
}
while (ret == 0 && (node = drm_mm_scan_color_evict(&scan))) {
vma = container_of(node, struct i915_vma, node);
+
/* If we find any non-objects (!vma), we cannot evict them */
- if (vma->node.color != I915_COLOR_UNEVICTABLE)
+ if (vma->node.color != I915_COLOR_UNEVICTABLE &&
+ i915_gem_object_trylock(vma->obj, ww)) {
ret = __i915_vma_unbind(vma);
- else
- ret = -ENOSPC; /* XXX search failed, try again? */
+ i915_gem_object_unlock(vma->obj);
+ } else {
+ ret = -ENOSPC;
+ }
}
return ret;
@@ -261,6 +275,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
* memory in e.g. the shrinker.
*/
int i915_gem_evict_for_node(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *target,
unsigned int flags)
{
@@ -333,6 +348,11 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
break;
}
+ if (!i915_gem_object_trylock(vma->obj, ww)) {
+ ret = -ENOSPC;
+ break;
+ }
+
/*
* Never show fear in the face of dragons!
*
@@ -350,6 +370,8 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
__i915_vma_unpin(vma);
if (ret == 0)
ret = __i915_vma_unbind(vma);
+
+ i915_gem_object_unlock(vma->obj);
}
return ret;
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index cd5f2348a187..c810da476c2b 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -93,6 +93,7 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
* asked to wait for eviction and interrupted.
*/
int i915_gem_gtt_reserve(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
u64 size, u64 offset, unsigned long color,
unsigned int flags)
@@ -117,7 +118,7 @@ int i915_gem_gtt_reserve(struct i915_address_space *vm,
if (flags & PIN_NOEVICT)
return -ENOSPC;
- err = i915_gem_evict_for_node(vm, node, flags);
+ err = i915_gem_evict_for_node(vm, ww, node, flags);
if (err == 0)
err = drm_mm_reserve_node(&vm->mm, node);
@@ -184,6 +185,7 @@ static u64 random_offset(u64 start, u64 end, u64 len, u64 align)
* asked to wait for eviction and interrupted.
*/
int i915_gem_gtt_insert(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
u64 size, u64 alignment, unsigned long color,
u64 start, u64 end, unsigned int flags)
@@ -269,7 +271,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
*/
offset = random_offset(start, end,
size, alignment ?: I915_GTT_MIN_ALIGNMENT);
- err = i915_gem_gtt_reserve(vm, node, size, offset, color, flags);
+ err = i915_gem_gtt_reserve(vm, ww, node, size, offset, color, flags);
if (err != -ENOSPC)
return err;
@@ -277,7 +279,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
return -ENOSPC;
/* Randomly selected placement is pinned, do a search */
- err = i915_gem_evict_something(vm, size, alignment, color,
+ err = i915_gem_evict_something(vm, ww, size, alignment, color,
start, end, flags);
if (err)
return err;
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index c9b0ee5e1d23..e4938aba3fe9 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -16,6 +16,7 @@
struct drm_i915_gem_object;
struct i915_address_space;
+struct i915_gem_ww_ctx;
int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
struct sg_table *pages);
@@ -23,11 +24,13 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
struct sg_table *pages);
int i915_gem_gtt_reserve(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
u64 size, u64 offset, unsigned long color,
unsigned int flags);
int i915_gem_gtt_insert(struct i915_address_space *vm,
+ struct i915_gem_ww_ctx *ww,
struct drm_mm_node *node,
u64 size, u64 alignment, unsigned long color,
u64 start, u64 end, unsigned int flags);
diff --git a/drivers/gpu/drm/i915/i915_vgpu.c b/drivers/gpu/drm/i915/i915_vgpu.c
index 31a105bc1792..c97323973f9b 100644
--- a/drivers/gpu/drm/i915/i915_vgpu.c
+++ b/drivers/gpu/drm/i915/i915_vgpu.c
@@ -197,7 +197,7 @@ static int vgt_balloon_space(struct i915_ggtt *ggtt,
drm_info(&dev_priv->drm,
"balloon space: range [ 0x%lx - 0x%lx ] %lu KiB.\n",
start, end, size / 1024);
- ret = i915_gem_gtt_reserve(&ggtt->vm, node,
+ ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, node,
size, start, I915_COLOR_UNEVICTABLE,
0);
if (!ret)
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index d24e90eac948..83df1290df5e 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -650,7 +650,8 @@ bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long color)
* 0 on success, negative error code otherwise.
*/
static int
-i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
+i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
+ u64 size, u64 alignment, u64 flags)
{
unsigned long color;
u64 start, end;
@@ -702,7 +703,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
range_overflows(offset, size, end))
return -EINVAL;
- ret = i915_gem_gtt_reserve(vma->vm, &vma->node,
+ ret = i915_gem_gtt_reserve(vma->vm, ww, &vma->node,
size, offset, color,
flags);
if (ret)
@@ -741,7 +742,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
size = round_up(size, I915_GTT_PAGE_SIZE_2M);
}
- ret = i915_gem_gtt_insert(vma->vm, &vma->node,
+ ret = i915_gem_gtt_insert(vma->vm, ww, &vma->node,
size, alignment, color,
start, end, flags);
if (ret)
@@ -1382,7 +1383,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
goto err_unlock;
if (!(bound & I915_VMA_BIND_MASK)) {
- err = i915_vma_insert(vma, size, alignment, flags);
+ err = i915_vma_insert(vma, ww, size, alignment, flags);
if (err)
goto err_active;
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
index 7178811366af..c80e1483d603 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
@@ -117,7 +117,7 @@ static int igt_evict_something(void *arg)
/* Everything is pinned, nothing should happen */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_something(&ggtt->vm,
+ err = i915_gem_evict_something(&ggtt->vm, NULL,
I915_GTT_PAGE_SIZE, 0, 0,
0, U64_MAX,
0);
@@ -132,11 +132,12 @@ static int igt_evict_something(void *arg)
/* Everything is unpinned, we should be able to evict something */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_something(&ggtt->vm,
+ err = i915_gem_evict_something(&ggtt->vm, NULL,
I915_GTT_PAGE_SIZE, 0, 0,
0, U64_MAX,
0);
mutex_unlock(&ggtt->vm.mutex);
+
if (err) {
pr_err("i915_gem_evict_something failed on a full GGTT with err=%d\n",
err);
@@ -204,7 +205,7 @@ static int igt_evict_for_vma(void *arg)
/* Everything is pinned, nothing should happen */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
+ err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
mutex_unlock(&ggtt->vm.mutex);
if (err != -ENOSPC) {
pr_err("i915_gem_evict_for_node on a full GGTT returned err=%d\n",
@@ -216,7 +217,7 @@ static int igt_evict_for_vma(void *arg)
/* Everything is unpinned, we should be able to evict the node */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
+ err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
mutex_unlock(&ggtt->vm.mutex);
if (err) {
pr_err("i915_gem_evict_for_node returned err=%d\n",
@@ -297,7 +298,7 @@ static int igt_evict_for_cache_color(void *arg)
/* Remove just the second vma */
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
+ err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
mutex_unlock(&ggtt->vm.mutex);
if (err) {
pr_err("[0]i915_gem_evict_for_node returned err=%d\n", err);
@@ -310,7 +311,7 @@ static int igt_evict_for_cache_color(void *arg)
target.color = I915_CACHE_L3_LLC;
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
+ err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
mutex_unlock(&ggtt->vm.mutex);
if (!err) {
pr_err("[1]i915_gem_evict_for_node returned err=%d\n", err);
@@ -408,7 +409,7 @@ static int igt_evict_contexts(void *arg)
/* Reserve a block so that we know we have enough to fit a few rq */
memset(&hole, 0, sizeof(hole));
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &hole,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &hole,
PRETEND_GGTT_SIZE, 0, I915_COLOR_UNEVICTABLE,
0, ggtt->vm.total,
PIN_NOEVICT);
@@ -428,7 +429,7 @@ static int igt_evict_contexts(void *arg)
goto out_locked;
}
- if (i915_gem_gtt_insert(&ggtt->vm, &r->node,
+ if (i915_gem_gtt_insert(&ggtt->vm, NULL, &r->node,
1ul << 20, 0, I915_COLOR_UNEVICTABLE,
0, ggtt->vm.total,
PIN_NOEVICT)) {
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 88b3a4d9e27e..169e2330b386 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -1378,7 +1378,7 @@ static int igt_gtt_reserve(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
obj->base.size,
total,
obj->cache_level,
@@ -1430,7 +1430,7 @@ static int igt_gtt_reserve(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
obj->base.size,
total,
obj->cache_level,
@@ -1477,7 +1477,7 @@ static int igt_gtt_reserve(void *arg)
I915_GTT_MIN_ALIGNMENT);
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
obj->base.size,
offset,
obj->cache_level,
@@ -1552,7 +1552,7 @@ static int igt_gtt_insert(void *arg)
/* Check a couple of obviously invalid requests */
for (ii = invalid_insert; ii->size; ii++) {
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &tmp,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &tmp,
ii->size, ii->alignment,
I915_COLOR_UNEVICTABLE,
ii->start, ii->end,
@@ -1594,7 +1594,7 @@ static int igt_gtt_insert(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
obj->base.size, 0, obj->cache_level,
0, ggtt->vm.total,
0);
@@ -1654,7 +1654,7 @@ static int igt_gtt_insert(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
obj->base.size, 0, obj->cache_level,
0, ggtt->vm.total,
0);
@@ -1703,7 +1703,7 @@ static int igt_gtt_insert(void *arg)
}
mutex_lock(&ggtt->vm.mutex);
- err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
+ err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
obj->base.size, 0, obj->cache_level,
0, ggtt->vm.total,
0);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 14/17] drm/i915: Add i915_vma_unbind_unlocked, and take obj lock for i915_vma_unbind, v2.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
We want to remove more members of i915_vma, which requires the locking to be
held more often.
Start requiring gem object lock for i915_vma_unbind, as it's one of the
callers that may unpin pages.
Some special care is needed when evicting, because the last reference to the
object may be held by the VMA, so after __i915_vma_unbind, vma may be garbage,
and we need to cache vma->obj before unlocking.
Changes since v1:
- Make trylock failing a WARN. (Matt)
- Remove double i915_vma_wait_for_bind() (Matt)
- Move atomic_set to right before mutex_unlock(), to make it more clear
they belong together. (Matt)
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.william.auld@gmail.com>
---
drivers/gpu/drm/i915/display/intel_fb_pin.c | 2 +-
.../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +-
.../i915/gem/selftests/i915_gem_client_blt.c | 2 +-
.../drm/i915/gem/selftests/i915_gem_mman.c | 6 +++
drivers/gpu/drm/i915/gt/intel_ggtt.c | 49 ++++++++++++++++---
drivers/gpu/drm/i915/i915_gem.c | 2 +
drivers/gpu/drm/i915/i915_vma.c | 27 +++++++++-
drivers/gpu/drm/i915/i915_vma.h | 1 +
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 22 ++++-----
drivers/gpu/drm/i915/selftests/i915_vma.c | 8 +--
10 files changed, 95 insertions(+), 26 deletions(-)
diff --git a/drivers/gpu/drm/i915/display/intel_fb_pin.c b/drivers/gpu/drm/i915/display/intel_fb_pin.c
index 31c15e5fca95..9c555f6d1958 100644
--- a/drivers/gpu/drm/i915/display/intel_fb_pin.c
+++ b/drivers/gpu/drm/i915/display/intel_fb_pin.c
@@ -47,7 +47,7 @@ intel_pin_fb_obj_dpt(struct drm_framebuffer *fb,
goto err;
if (i915_vma_misplaced(vma, 0, alignment, 0)) {
- ret = i915_vma_unbind(vma);
+ ret = i915_vma_unbind_unlocked(vma);
if (ret) {
vma = ERR_PTR(ret);
goto err;
diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
index c69c7d45aabc..0db62652e3ba 100644
--- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
+++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
@@ -647,7 +647,7 @@ static int igt_mock_ppgtt_misaligned_dma(void *arg)
* pages.
*/
for (offset = 4096; offset < page_size; offset += 4096) {
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err)
goto out_unpin;
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
index 8402ed925a69..8fb5be799b3c 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
@@ -318,7 +318,7 @@ static int pin_buffer(struct i915_vma *vma, u64 addr)
int err;
if (drm_mm_node_allocated(&vma->node) && vma->node.start != addr) {
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err)
return err;
}
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
index 6d30cdfa80f3..e69e8861352d 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
@@ -165,7 +165,9 @@ static int check_partial_mapping(struct drm_i915_gem_object *obj,
kunmap(p);
out:
+ i915_gem_object_lock(obj, NULL);
__i915_vma_put(vma);
+ i915_gem_object_unlock(obj);
return err;
}
@@ -259,7 +261,9 @@ static int check_partial_mappings(struct drm_i915_gem_object *obj,
if (err)
return err;
+ i915_gem_object_lock(obj, NULL);
__i915_vma_put(vma);
+ i915_gem_object_unlock(obj);
if (igt_timeout(end_time,
"%s: timed out after tiling=%d stride=%d\n",
@@ -1349,7 +1353,9 @@ static int __igt_mmap_revoke(struct drm_i915_private *i915,
* for other objects. Ergo we have to revoke the previous mmap PTE
* access as it no longer points to the same object.
*/
+ i915_gem_object_lock(obj, NULL);
err = i915_gem_object_unbind(obj, I915_GEM_OBJECT_UNBIND_ACTIVE);
+ i915_gem_object_unlock(obj);
if (err) {
pr_err("Failed to unbind object!\n");
goto out_unmap;
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index ed43e9c80aaa..8bba5b9ceeb0 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -129,22 +129,49 @@ void i915_ggtt_suspend_vm(struct i915_address_space *vm)
drm_WARN_ON(&vm->i915->drm, !vm->is_ggtt && !vm->is_dpt);
+retry:
+ i915_gem_drain_freed_objects(vm->i915);
+
mutex_lock(&vm->mutex);
/* Skip rewriting PTE on VMA unbind. */
open = atomic_xchg(&vm->open, 0);
list_for_each_entry_safe(vma, vn, &vm->bound_list, vm_link) {
+ struct drm_i915_gem_object *obj = vma->obj;
+
GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
- i915_vma_wait_for_bind(vma);
- if (i915_vma_is_pinned(vma))
+ if (i915_vma_is_pinned(vma) || !i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND))
continue;
- if (!i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND)) {
- __i915_vma_evict(vma);
- drm_mm_remove_node(&vma->node);
+ /* unlikely to race when GPU is idle, so no worry about slowpath.. */
+ if (WARN_ON(!i915_gem_object_trylock(obj, NULL))) {
+ /*
+ * No dead objects should appear here, GPU should be
+ * completely idle, and userspace suspended
+ */
+ i915_gem_object_get(obj);
+
+ atomic_set(&vm->open, open);
+ mutex_unlock(&vm->mutex);
+
+ i915_gem_object_lock(obj, NULL);
+ open = i915_vma_unbind(vma);
+ i915_gem_object_unlock(obj);
+
+ GEM_WARN_ON(open);
+
+ i915_gem_object_put(obj);
+ goto retry;
}
+
+ i915_vma_wait_for_bind(vma);
+
+ __i915_vma_evict(vma);
+ drm_mm_remove_node(&vma->node);
+
+ i915_gem_object_unlock(obj);
}
vm->clear_range(vm, 0, vm->total);
@@ -742,11 +769,21 @@ static void ggtt_cleanup_hw(struct i915_ggtt *ggtt)
atomic_set(&ggtt->vm.open, 0);
flush_workqueue(ggtt->vm.i915->wq);
+ i915_gem_drain_freed_objects(ggtt->vm.i915);
mutex_lock(&ggtt->vm.mutex);
- list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link)
+ list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link) {
+ struct drm_i915_gem_object *obj = vma->obj;
+ bool trylock;
+
+ trylock = i915_gem_object_trylock(obj, NULL);
+ WARN_ON(!trylock);
+
WARN_ON(__i915_vma_unbind(vma));
+ if (trylock)
+ i915_gem_object_unlock(obj);
+ }
if (drm_mm_node_allocated(&ggtt->error_capture))
drm_mm_remove_node(&ggtt->error_capture);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 6002045bd194..c31ef87397f1 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -118,6 +118,8 @@ int i915_gem_object_unbind(struct drm_i915_gem_object *obj,
struct i915_vma *vma;
int ret;
+ assert_object_held(obj);
+
if (list_empty(&obj->vma.list))
return 0;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 83df1290df5e..565819c8bf49 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1604,8 +1604,16 @@ void i915_vma_parked(struct intel_gt *gt)
struct drm_i915_gem_object *obj = vma->obj;
struct i915_address_space *vm = vma->vm;
- INIT_LIST_HEAD(&vma->closed_link);
- __i915_vma_put(vma);
+ if (i915_gem_object_trylock(obj, NULL)) {
+ INIT_LIST_HEAD(&vma->closed_link);
+ __i915_vma_put(vma);
+ i915_gem_object_unlock(obj);
+ } else {
+ /* back you go.. */
+ spin_lock_irq(>->closed_lock);
+ list_add(&vma->closed_link, >->closed_vma);
+ spin_unlock_irq(>->closed_lock);
+ }
i915_gem_object_put(obj);
i915_vm_close(vm);
@@ -1721,6 +1729,7 @@ int _i915_vma_move_to_active(struct i915_vma *vma,
void __i915_vma_evict(struct i915_vma *vma)
{
GEM_BUG_ON(i915_vma_is_pinned(vma));
+ assert_object_held_shared(vma->obj);
if (i915_vma_is_map_and_fenceable(vma)) {
/* Force a pagefault for domain tracking on next user access */
@@ -1766,6 +1775,7 @@ int __i915_vma_unbind(struct i915_vma *vma)
int ret;
lockdep_assert_held(&vma->vm->mutex);
+ assert_object_held_shared(vma->obj);
if (!drm_mm_node_allocated(&vma->node))
return 0;
@@ -1797,6 +1807,8 @@ int i915_vma_unbind(struct i915_vma *vma)
intel_wakeref_t wakeref = 0;
int err;
+ assert_object_held_shared(vma->obj);
+
/* Optimistic wait before taking the mutex */
err = i915_vma_sync(vma);
if (err)
@@ -1827,6 +1839,17 @@ int i915_vma_unbind(struct i915_vma *vma)
return err;
}
+int i915_vma_unbind_unlocked(struct i915_vma *vma)
+{
+ int err;
+
+ i915_gem_object_lock(vma->obj, NULL);
+ err = i915_vma_unbind(vma);
+ i915_gem_object_unlock(vma->obj);
+
+ return err;
+}
+
struct i915_vma *i915_vma_make_unshrinkable(struct i915_vma *vma)
{
i915_gem_object_make_unshrinkable(vma->obj);
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 32719431b3df..da69ecb1b860 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -214,6 +214,7 @@ void i915_vma_revoke_mmap(struct i915_vma *vma);
void __i915_vma_evict(struct i915_vma *vma);
int __i915_vma_unbind(struct i915_vma *vma);
int __must_check i915_vma_unbind(struct i915_vma *vma);
+int __must_check i915_vma_unbind_unlocked(struct i915_vma *vma);
void i915_vma_unlink_ctx(struct i915_vma *vma);
void i915_vma_close(struct i915_vma *vma);
void i915_vma_reopen(struct i915_vma *vma);
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 169e2330b386..cafe7936fb6f 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -385,7 +385,7 @@ static void close_object_list(struct list_head *objects,
vma = i915_vma_instance(obj, vm, NULL);
if (!IS_ERR(vma))
- ignored = i915_vma_unbind(vma);
+ ignored = i915_vma_unbind_unlocked(vma);
list_del(&obj->st_link);
i915_gem_object_put(obj);
@@ -496,7 +496,7 @@ static int fill_hole(struct i915_address_space *vm,
goto err;
}
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("%s(%s) (forward) unbind of vma.node=%llx + %llx failed with err=%d\n",
__func__, p->name, vma->node.start, vma->node.size,
@@ -569,7 +569,7 @@ static int fill_hole(struct i915_address_space *vm,
goto err;
}
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("%s(%s) (backward) unbind of vma.node=%llx + %llx failed with err=%d\n",
__func__, p->name, vma->node.start, vma->node.size,
@@ -655,7 +655,7 @@ static int walk_hole(struct i915_address_space *vm,
goto err_put;
}
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("%s unbind failed at %llx + %llx with err=%d\n",
__func__, addr, vma->size, err);
@@ -732,13 +732,13 @@ static int pot_hole(struct i915_address_space *vm,
pr_err("%s incorrect at %llx + %llx\n",
__func__, addr, vma->size);
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
err = -EINVAL;
goto err_obj;
}
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
GEM_BUG_ON(err);
}
@@ -832,13 +832,13 @@ static int drunk_hole(struct i915_address_space *vm,
pr_err("%s incorrect at %llx + %llx\n",
__func__, addr, BIT_ULL(size));
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
err = -EINVAL;
goto err_obj;
}
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
GEM_BUG_ON(err);
if (igt_timeout(end_time,
@@ -906,7 +906,7 @@ static int __shrink_hole(struct i915_address_space *vm,
pr_err("%s incorrect at %llx + %llx\n",
__func__, addr, size);
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
err = -EINVAL;
break;
}
@@ -1465,7 +1465,7 @@ static int igt_gtt_reserve(void *arg)
goto out;
}
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("i915_vma_unbind failed with err=%d!\n", err);
goto out;
@@ -1647,7 +1647,7 @@ static int igt_gtt_insert(void *arg)
GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
offset = vma->node.start;
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("i915_vma_unbind failed with err=%d!\n", err);
goto out;
diff --git a/drivers/gpu/drm/i915/selftests/i915_vma.c b/drivers/gpu/drm/i915/selftests/i915_vma.c
index 5c5809dfe9b2..2c73f7448df7 100644
--- a/drivers/gpu/drm/i915/selftests/i915_vma.c
+++ b/drivers/gpu/drm/i915/selftests/i915_vma.c
@@ -340,7 +340,7 @@ static int igt_vma_pin1(void *arg)
if (!err) {
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("Failed to unbind single page from GGTT, err=%d\n", err);
goto out;
@@ -691,7 +691,7 @@ static int igt_vma_rotate_remap(void *arg)
}
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("Unbinding returned %i\n", err);
goto out_object;
@@ -852,7 +852,7 @@ static int igt_vma_partial(void *arg)
i915_vma_unpin(vma);
nvma++;
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("Unbinding returned %i\n", err);
goto out_object;
@@ -891,7 +891,7 @@ static int igt_vma_partial(void *arg)
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("Unbinding returned %i\n", err);
goto out_object;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 14/17] drm/i915: Add i915_vma_unbind_unlocked, and take obj lock for i915_vma_unbind, v2.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
We want to remove more members of i915_vma, which requires the locking to be
held more often.
Start requiring gem object lock for i915_vma_unbind, as it's one of the
callers that may unpin pages.
Some special care is needed when evicting, because the last reference to the
object may be held by the VMA, so after __i915_vma_unbind, vma may be garbage,
and we need to cache vma->obj before unlocking.
Changes since v1:
- Make trylock failing a WARN. (Matt)
- Remove double i915_vma_wait_for_bind() (Matt)
- Move atomic_set to right before mutex_unlock(), to make it more clear
they belong together. (Matt)
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.william.auld@gmail.com>
---
drivers/gpu/drm/i915/display/intel_fb_pin.c | 2 +-
.../gpu/drm/i915/gem/selftests/huge_pages.c | 2 +-
.../i915/gem/selftests/i915_gem_client_blt.c | 2 +-
.../drm/i915/gem/selftests/i915_gem_mman.c | 6 +++
drivers/gpu/drm/i915/gt/intel_ggtt.c | 49 ++++++++++++++++---
drivers/gpu/drm/i915/i915_gem.c | 2 +
drivers/gpu/drm/i915/i915_vma.c | 27 +++++++++-
drivers/gpu/drm/i915/i915_vma.h | 1 +
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 22 ++++-----
drivers/gpu/drm/i915/selftests/i915_vma.c | 8 +--
10 files changed, 95 insertions(+), 26 deletions(-)
diff --git a/drivers/gpu/drm/i915/display/intel_fb_pin.c b/drivers/gpu/drm/i915/display/intel_fb_pin.c
index 31c15e5fca95..9c555f6d1958 100644
--- a/drivers/gpu/drm/i915/display/intel_fb_pin.c
+++ b/drivers/gpu/drm/i915/display/intel_fb_pin.c
@@ -47,7 +47,7 @@ intel_pin_fb_obj_dpt(struct drm_framebuffer *fb,
goto err;
if (i915_vma_misplaced(vma, 0, alignment, 0)) {
- ret = i915_vma_unbind(vma);
+ ret = i915_vma_unbind_unlocked(vma);
if (ret) {
vma = ERR_PTR(ret);
goto err;
diff --git a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
index c69c7d45aabc..0db62652e3ba 100644
--- a/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
+++ b/drivers/gpu/drm/i915/gem/selftests/huge_pages.c
@@ -647,7 +647,7 @@ static int igt_mock_ppgtt_misaligned_dma(void *arg)
* pages.
*/
for (offset = 4096; offset < page_size; offset += 4096) {
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err)
goto out_unpin;
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
index 8402ed925a69..8fb5be799b3c 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_client_blt.c
@@ -318,7 +318,7 @@ static int pin_buffer(struct i915_vma *vma, u64 addr)
int err;
if (drm_mm_node_allocated(&vma->node) && vma->node.start != addr) {
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err)
return err;
}
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
index 6d30cdfa80f3..e69e8861352d 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
@@ -165,7 +165,9 @@ static int check_partial_mapping(struct drm_i915_gem_object *obj,
kunmap(p);
out:
+ i915_gem_object_lock(obj, NULL);
__i915_vma_put(vma);
+ i915_gem_object_unlock(obj);
return err;
}
@@ -259,7 +261,9 @@ static int check_partial_mappings(struct drm_i915_gem_object *obj,
if (err)
return err;
+ i915_gem_object_lock(obj, NULL);
__i915_vma_put(vma);
+ i915_gem_object_unlock(obj);
if (igt_timeout(end_time,
"%s: timed out after tiling=%d stride=%d\n",
@@ -1349,7 +1353,9 @@ static int __igt_mmap_revoke(struct drm_i915_private *i915,
* for other objects. Ergo we have to revoke the previous mmap PTE
* access as it no longer points to the same object.
*/
+ i915_gem_object_lock(obj, NULL);
err = i915_gem_object_unbind(obj, I915_GEM_OBJECT_UNBIND_ACTIVE);
+ i915_gem_object_unlock(obj);
if (err) {
pr_err("Failed to unbind object!\n");
goto out_unmap;
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
index ed43e9c80aaa..8bba5b9ceeb0 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
@@ -129,22 +129,49 @@ void i915_ggtt_suspend_vm(struct i915_address_space *vm)
drm_WARN_ON(&vm->i915->drm, !vm->is_ggtt && !vm->is_dpt);
+retry:
+ i915_gem_drain_freed_objects(vm->i915);
+
mutex_lock(&vm->mutex);
/* Skip rewriting PTE on VMA unbind. */
open = atomic_xchg(&vm->open, 0);
list_for_each_entry_safe(vma, vn, &vm->bound_list, vm_link) {
+ struct drm_i915_gem_object *obj = vma->obj;
+
GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
- i915_vma_wait_for_bind(vma);
- if (i915_vma_is_pinned(vma))
+ if (i915_vma_is_pinned(vma) || !i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND))
continue;
- if (!i915_vma_is_bound(vma, I915_VMA_GLOBAL_BIND)) {
- __i915_vma_evict(vma);
- drm_mm_remove_node(&vma->node);
+ /* unlikely to race when GPU is idle, so no worry about slowpath.. */
+ if (WARN_ON(!i915_gem_object_trylock(obj, NULL))) {
+ /*
+ * No dead objects should appear here, GPU should be
+ * completely idle, and userspace suspended
+ */
+ i915_gem_object_get(obj);
+
+ atomic_set(&vm->open, open);
+ mutex_unlock(&vm->mutex);
+
+ i915_gem_object_lock(obj, NULL);
+ open = i915_vma_unbind(vma);
+ i915_gem_object_unlock(obj);
+
+ GEM_WARN_ON(open);
+
+ i915_gem_object_put(obj);
+ goto retry;
}
+
+ i915_vma_wait_for_bind(vma);
+
+ __i915_vma_evict(vma);
+ drm_mm_remove_node(&vma->node);
+
+ i915_gem_object_unlock(obj);
}
vm->clear_range(vm, 0, vm->total);
@@ -742,11 +769,21 @@ static void ggtt_cleanup_hw(struct i915_ggtt *ggtt)
atomic_set(&ggtt->vm.open, 0);
flush_workqueue(ggtt->vm.i915->wq);
+ i915_gem_drain_freed_objects(ggtt->vm.i915);
mutex_lock(&ggtt->vm.mutex);
- list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link)
+ list_for_each_entry_safe(vma, vn, &ggtt->vm.bound_list, vm_link) {
+ struct drm_i915_gem_object *obj = vma->obj;
+ bool trylock;
+
+ trylock = i915_gem_object_trylock(obj, NULL);
+ WARN_ON(!trylock);
+
WARN_ON(__i915_vma_unbind(vma));
+ if (trylock)
+ i915_gem_object_unlock(obj);
+ }
if (drm_mm_node_allocated(&ggtt->error_capture))
drm_mm_remove_node(&ggtt->error_capture);
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 6002045bd194..c31ef87397f1 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -118,6 +118,8 @@ int i915_gem_object_unbind(struct drm_i915_gem_object *obj,
struct i915_vma *vma;
int ret;
+ assert_object_held(obj);
+
if (list_empty(&obj->vma.list))
return 0;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 83df1290df5e..565819c8bf49 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1604,8 +1604,16 @@ void i915_vma_parked(struct intel_gt *gt)
struct drm_i915_gem_object *obj = vma->obj;
struct i915_address_space *vm = vma->vm;
- INIT_LIST_HEAD(&vma->closed_link);
- __i915_vma_put(vma);
+ if (i915_gem_object_trylock(obj, NULL)) {
+ INIT_LIST_HEAD(&vma->closed_link);
+ __i915_vma_put(vma);
+ i915_gem_object_unlock(obj);
+ } else {
+ /* back you go.. */
+ spin_lock_irq(>->closed_lock);
+ list_add(&vma->closed_link, >->closed_vma);
+ spin_unlock_irq(>->closed_lock);
+ }
i915_gem_object_put(obj);
i915_vm_close(vm);
@@ -1721,6 +1729,7 @@ int _i915_vma_move_to_active(struct i915_vma *vma,
void __i915_vma_evict(struct i915_vma *vma)
{
GEM_BUG_ON(i915_vma_is_pinned(vma));
+ assert_object_held_shared(vma->obj);
if (i915_vma_is_map_and_fenceable(vma)) {
/* Force a pagefault for domain tracking on next user access */
@@ -1766,6 +1775,7 @@ int __i915_vma_unbind(struct i915_vma *vma)
int ret;
lockdep_assert_held(&vma->vm->mutex);
+ assert_object_held_shared(vma->obj);
if (!drm_mm_node_allocated(&vma->node))
return 0;
@@ -1797,6 +1807,8 @@ int i915_vma_unbind(struct i915_vma *vma)
intel_wakeref_t wakeref = 0;
int err;
+ assert_object_held_shared(vma->obj);
+
/* Optimistic wait before taking the mutex */
err = i915_vma_sync(vma);
if (err)
@@ -1827,6 +1839,17 @@ int i915_vma_unbind(struct i915_vma *vma)
return err;
}
+int i915_vma_unbind_unlocked(struct i915_vma *vma)
+{
+ int err;
+
+ i915_gem_object_lock(vma->obj, NULL);
+ err = i915_vma_unbind(vma);
+ i915_gem_object_unlock(vma->obj);
+
+ return err;
+}
+
struct i915_vma *i915_vma_make_unshrinkable(struct i915_vma *vma)
{
i915_gem_object_make_unshrinkable(vma->obj);
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 32719431b3df..da69ecb1b860 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -214,6 +214,7 @@ void i915_vma_revoke_mmap(struct i915_vma *vma);
void __i915_vma_evict(struct i915_vma *vma);
int __i915_vma_unbind(struct i915_vma *vma);
int __must_check i915_vma_unbind(struct i915_vma *vma);
+int __must_check i915_vma_unbind_unlocked(struct i915_vma *vma);
void i915_vma_unlink_ctx(struct i915_vma *vma);
void i915_vma_close(struct i915_vma *vma);
void i915_vma_reopen(struct i915_vma *vma);
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 169e2330b386..cafe7936fb6f 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -385,7 +385,7 @@ static void close_object_list(struct list_head *objects,
vma = i915_vma_instance(obj, vm, NULL);
if (!IS_ERR(vma))
- ignored = i915_vma_unbind(vma);
+ ignored = i915_vma_unbind_unlocked(vma);
list_del(&obj->st_link);
i915_gem_object_put(obj);
@@ -496,7 +496,7 @@ static int fill_hole(struct i915_address_space *vm,
goto err;
}
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("%s(%s) (forward) unbind of vma.node=%llx + %llx failed with err=%d\n",
__func__, p->name, vma->node.start, vma->node.size,
@@ -569,7 +569,7 @@ static int fill_hole(struct i915_address_space *vm,
goto err;
}
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("%s(%s) (backward) unbind of vma.node=%llx + %llx failed with err=%d\n",
__func__, p->name, vma->node.start, vma->node.size,
@@ -655,7 +655,7 @@ static int walk_hole(struct i915_address_space *vm,
goto err_put;
}
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("%s unbind failed at %llx + %llx with err=%d\n",
__func__, addr, vma->size, err);
@@ -732,13 +732,13 @@ static int pot_hole(struct i915_address_space *vm,
pr_err("%s incorrect at %llx + %llx\n",
__func__, addr, vma->size);
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
err = -EINVAL;
goto err_obj;
}
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
GEM_BUG_ON(err);
}
@@ -832,13 +832,13 @@ static int drunk_hole(struct i915_address_space *vm,
pr_err("%s incorrect at %llx + %llx\n",
__func__, addr, BIT_ULL(size));
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
err = -EINVAL;
goto err_obj;
}
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
GEM_BUG_ON(err);
if (igt_timeout(end_time,
@@ -906,7 +906,7 @@ static int __shrink_hole(struct i915_address_space *vm,
pr_err("%s incorrect at %llx + %llx\n",
__func__, addr, size);
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
err = -EINVAL;
break;
}
@@ -1465,7 +1465,7 @@ static int igt_gtt_reserve(void *arg)
goto out;
}
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("i915_vma_unbind failed with err=%d!\n", err);
goto out;
@@ -1647,7 +1647,7 @@ static int igt_gtt_insert(void *arg)
GEM_BUG_ON(!drm_mm_node_allocated(&vma->node));
offset = vma->node.start;
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("i915_vma_unbind failed with err=%d!\n", err);
goto out;
diff --git a/drivers/gpu/drm/i915/selftests/i915_vma.c b/drivers/gpu/drm/i915/selftests/i915_vma.c
index 5c5809dfe9b2..2c73f7448df7 100644
--- a/drivers/gpu/drm/i915/selftests/i915_vma.c
+++ b/drivers/gpu/drm/i915/selftests/i915_vma.c
@@ -340,7 +340,7 @@ static int igt_vma_pin1(void *arg)
if (!err) {
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("Failed to unbind single page from GGTT, err=%d\n", err);
goto out;
@@ -691,7 +691,7 @@ static int igt_vma_rotate_remap(void *arg)
}
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("Unbinding returned %i\n", err);
goto out_object;
@@ -852,7 +852,7 @@ static int igt_vma_partial(void *arg)
i915_vma_unpin(vma);
nvma++;
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("Unbinding returned %i\n", err);
goto out_object;
@@ -891,7 +891,7 @@ static int igt_vma_partial(void *arg)
i915_vma_unpin(vma);
- err = i915_vma_unbind(vma);
+ err = i915_vma_unbind_unlocked(vma);
if (err) {
pr_err("Unbinding returned %i\n", err);
goto out_object;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 15/17] drm/i915: Remove assert_object_held_shared
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
This duck tape workaround is no longer required, unbind and destroy are
fixed to take the obj->resv mutex before destroying and obj->mm.lock has
been removed, always requiring obj->resv as well.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 4 ++--
drivers/gpu/drm/i915/gem/i915_gem_object.h | 14 --------------
drivers/gpu/drm/i915/gem/i915_gem_pages.c | 12 ++++++------
drivers/gpu/drm/i915/gem/i915_gem_userptr.c | 2 +-
drivers/gpu/drm/i915/i915_vma.c | 6 +++---
5 files changed, 12 insertions(+), 26 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index 39cd563544a5..0ae86812ed66 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -544,7 +544,7 @@ bool i915_gem_object_has_struct_page(const struct drm_i915_gem_object *obj)
#ifdef CONFIG_LOCKDEP
if (IS_DGFX(to_i915(obj->base.dev)) &&
i915_gem_object_evictable((void __force *)obj))
- assert_object_held_shared(obj);
+ assert_object_held(obj);
#endif
return obj->mem_flags & I915_BO_FLAG_STRUCT_PAGE;
}
@@ -563,7 +563,7 @@ bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)
#ifdef CONFIG_LOCKDEP
if (IS_DGFX(to_i915(obj->base.dev)) &&
i915_gem_object_evictable((void __force *)obj))
- assert_object_held_shared(obj);
+ assert_object_held(obj);
#endif
return obj->mem_flags & I915_BO_FLAG_IOMEM;
}
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h
index f66d46882ea7..09cf1c92373a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
@@ -157,20 +157,6 @@ i915_gem_object_put(struct drm_i915_gem_object *obj)
#define assert_object_held(obj) dma_resv_assert_held((obj)->base.resv)
-/*
- * If more than one potential simultaneous locker, assert held.
- */
-static inline void assert_object_held_shared(const struct drm_i915_gem_object *obj)
-{
- /*
- * Note mm list lookup is protected by
- * kref_get_unless_zero().
- */
- if (IS_ENABLED(CONFIG_LOCKDEP) &&
- kref_read(&obj->base.refcount) > 0)
- assert_object_held(obj);
-}
-
static inline int __i915_gem_object_lock(struct drm_i915_gem_object *obj,
struct i915_gem_ww_ctx *ww,
bool intr)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
index 89b70f5cde7a..95f2744e504c 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
@@ -19,7 +19,7 @@ void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
bool shrinkable;
int i;
- assert_object_held_shared(obj);
+ assert_object_held(obj);
if (i915_gem_object_is_volatile(obj))
obj->mm.madv = I915_MADV_DONTNEED;
@@ -95,7 +95,7 @@ int ____i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
struct drm_i915_private *i915 = to_i915(obj->base.dev);
int err;
- assert_object_held_shared(obj);
+ assert_object_held(obj);
if (unlikely(obj->mm.madv != I915_MADV_WILLNEED)) {
drm_dbg(&i915->drm,
@@ -122,7 +122,7 @@ int __i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
assert_object_held(obj);
- assert_object_held_shared(obj);
+ assert_object_held(obj);
if (unlikely(!i915_gem_object_has_pages(obj))) {
GEM_BUG_ON(i915_gem_object_has_pinned_pages(obj));
@@ -171,7 +171,7 @@ int i915_gem_object_truncate(struct drm_i915_gem_object *obj)
/* Try to discard unwanted pages */
void i915_gem_object_writeback(struct drm_i915_gem_object *obj)
{
- assert_object_held_shared(obj);
+ assert_object_held(obj);
GEM_BUG_ON(i915_gem_object_has_pages(obj));
if (obj->ops->writeback)
@@ -202,7 +202,7 @@ __i915_gem_object_unset_pages(struct drm_i915_gem_object *obj)
{
struct sg_table *pages;
- assert_object_held_shared(obj);
+ assert_object_held(obj);
pages = fetch_and_zero(&obj->mm.pages);
if (IS_ERR_OR_NULL(pages))
@@ -233,7 +233,7 @@ int __i915_gem_object_put_pages(struct drm_i915_gem_object *obj)
return -EBUSY;
/* May be called by shrinker from within get_pages() (on another bo) */
- assert_object_held_shared(obj);
+ assert_object_held(obj);
i915_gem_object_release_mmap_offset(obj);
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
index 3173c9f9a040..a315c010f635 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
@@ -109,7 +109,7 @@ static void i915_gem_object_userptr_drop_ref(struct drm_i915_gem_object *obj)
{
struct page **pvec = NULL;
- assert_object_held_shared(obj);
+ assert_object_held(obj);
if (!--obj->userptr.page_ref) {
pvec = obj->userptr.pvec;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 565819c8bf49..97093ecc3f97 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1729,7 +1729,7 @@ int _i915_vma_move_to_active(struct i915_vma *vma,
void __i915_vma_evict(struct i915_vma *vma)
{
GEM_BUG_ON(i915_vma_is_pinned(vma));
- assert_object_held_shared(vma->obj);
+ assert_object_held(vma->obj);
if (i915_vma_is_map_and_fenceable(vma)) {
/* Force a pagefault for domain tracking on next user access */
@@ -1775,7 +1775,7 @@ int __i915_vma_unbind(struct i915_vma *vma)
int ret;
lockdep_assert_held(&vma->vm->mutex);
- assert_object_held_shared(vma->obj);
+ assert_object_held(vma->obj);
if (!drm_mm_node_allocated(&vma->node))
return 0;
@@ -1807,7 +1807,7 @@ int i915_vma_unbind(struct i915_vma *vma)
intel_wakeref_t wakeref = 0;
int err;
- assert_object_held_shared(vma->obj);
+ assert_object_held(vma->obj);
/* Optimistic wait before taking the mutex */
err = i915_vma_sync(vma);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 15/17] drm/i915: Remove assert_object_held_shared
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
This duck tape workaround is no longer required, unbind and destroy are
fixed to take the obj->resv mutex before destroying and obj->mm.lock has
been removed, always requiring obj->resv as well.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
drivers/gpu/drm/i915/gem/i915_gem_object.c | 4 ++--
drivers/gpu/drm/i915/gem/i915_gem_object.h | 14 --------------
drivers/gpu/drm/i915/gem/i915_gem_pages.c | 12 ++++++------
drivers/gpu/drm/i915/gem/i915_gem_userptr.c | 2 +-
drivers/gpu/drm/i915/i915_vma.c | 6 +++---
5 files changed, 12 insertions(+), 26 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index 39cd563544a5..0ae86812ed66 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -544,7 +544,7 @@ bool i915_gem_object_has_struct_page(const struct drm_i915_gem_object *obj)
#ifdef CONFIG_LOCKDEP
if (IS_DGFX(to_i915(obj->base.dev)) &&
i915_gem_object_evictable((void __force *)obj))
- assert_object_held_shared(obj);
+ assert_object_held(obj);
#endif
return obj->mem_flags & I915_BO_FLAG_STRUCT_PAGE;
}
@@ -563,7 +563,7 @@ bool i915_gem_object_has_iomem(const struct drm_i915_gem_object *obj)
#ifdef CONFIG_LOCKDEP
if (IS_DGFX(to_i915(obj->base.dev)) &&
i915_gem_object_evictable((void __force *)obj))
- assert_object_held_shared(obj);
+ assert_object_held(obj);
#endif
return obj->mem_flags & I915_BO_FLAG_IOMEM;
}
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.h b/drivers/gpu/drm/i915/gem/i915_gem_object.h
index f66d46882ea7..09cf1c92373a 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.h
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.h
@@ -157,20 +157,6 @@ i915_gem_object_put(struct drm_i915_gem_object *obj)
#define assert_object_held(obj) dma_resv_assert_held((obj)->base.resv)
-/*
- * If more than one potential simultaneous locker, assert held.
- */
-static inline void assert_object_held_shared(const struct drm_i915_gem_object *obj)
-{
- /*
- * Note mm list lookup is protected by
- * kref_get_unless_zero().
- */
- if (IS_ENABLED(CONFIG_LOCKDEP) &&
- kref_read(&obj->base.refcount) > 0)
- assert_object_held(obj);
-}
-
static inline int __i915_gem_object_lock(struct drm_i915_gem_object *obj,
struct i915_gem_ww_ctx *ww,
bool intr)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pages.c b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
index 89b70f5cde7a..95f2744e504c 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_pages.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_pages.c
@@ -19,7 +19,7 @@ void __i915_gem_object_set_pages(struct drm_i915_gem_object *obj,
bool shrinkable;
int i;
- assert_object_held_shared(obj);
+ assert_object_held(obj);
if (i915_gem_object_is_volatile(obj))
obj->mm.madv = I915_MADV_DONTNEED;
@@ -95,7 +95,7 @@ int ____i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
struct drm_i915_private *i915 = to_i915(obj->base.dev);
int err;
- assert_object_held_shared(obj);
+ assert_object_held(obj);
if (unlikely(obj->mm.madv != I915_MADV_WILLNEED)) {
drm_dbg(&i915->drm,
@@ -122,7 +122,7 @@ int __i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
assert_object_held(obj);
- assert_object_held_shared(obj);
+ assert_object_held(obj);
if (unlikely(!i915_gem_object_has_pages(obj))) {
GEM_BUG_ON(i915_gem_object_has_pinned_pages(obj));
@@ -171,7 +171,7 @@ int i915_gem_object_truncate(struct drm_i915_gem_object *obj)
/* Try to discard unwanted pages */
void i915_gem_object_writeback(struct drm_i915_gem_object *obj)
{
- assert_object_held_shared(obj);
+ assert_object_held(obj);
GEM_BUG_ON(i915_gem_object_has_pages(obj));
if (obj->ops->writeback)
@@ -202,7 +202,7 @@ __i915_gem_object_unset_pages(struct drm_i915_gem_object *obj)
{
struct sg_table *pages;
- assert_object_held_shared(obj);
+ assert_object_held(obj);
pages = fetch_and_zero(&obj->mm.pages);
if (IS_ERR_OR_NULL(pages))
@@ -233,7 +233,7 @@ int __i915_gem_object_put_pages(struct drm_i915_gem_object *obj)
return -EBUSY;
/* May be called by shrinker from within get_pages() (on another bo) */
- assert_object_held_shared(obj);
+ assert_object_held(obj);
i915_gem_object_release_mmap_offset(obj);
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
index 3173c9f9a040..a315c010f635 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_userptr.c
@@ -109,7 +109,7 @@ static void i915_gem_object_userptr_drop_ref(struct drm_i915_gem_object *obj)
{
struct page **pvec = NULL;
- assert_object_held_shared(obj);
+ assert_object_held(obj);
if (!--obj->userptr.page_ref) {
pvec = obj->userptr.pvec;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 565819c8bf49..97093ecc3f97 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1729,7 +1729,7 @@ int _i915_vma_move_to_active(struct i915_vma *vma,
void __i915_vma_evict(struct i915_vma *vma)
{
GEM_BUG_ON(i915_vma_is_pinned(vma));
- assert_object_held_shared(vma->obj);
+ assert_object_held(vma->obj);
if (i915_vma_is_map_and_fenceable(vma)) {
/* Force a pagefault for domain tracking on next user access */
@@ -1775,7 +1775,7 @@ int __i915_vma_unbind(struct i915_vma *vma)
int ret;
lockdep_assert_held(&vma->vm->mutex);
- assert_object_held_shared(vma->obj);
+ assert_object_held(vma->obj);
if (!drm_mm_node_allocated(&vma->node))
return 0;
@@ -1807,7 +1807,7 @@ int i915_vma_unbind(struct i915_vma *vma)
intel_wakeref_t wakeref = 0;
int err;
- assert_object_held_shared(vma->obj);
+ assert_object_held(vma->obj);
/* Optimistic wait before taking the mutex */
err = i915_vma_sync(vma);
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 16/17] drm/i915: Remove support for unlocked i915_vma unbind
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: dri-devel
Now that we require the object lock for all ops, some code handling
race conditions can be removed.
This is required to not take short-term pins inside execbuf.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Acked-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
---
drivers/gpu/drm/i915/i915_vma.c | 55 +++++----------------------------
1 file changed, 8 insertions(+), 47 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 97093ecc3f97..bdd348fc755c 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -776,7 +776,6 @@ i915_vma_detach(struct i915_vma *vma)
static bool try_qad_pin(struct i915_vma *vma, unsigned int flags)
{
unsigned int bound;
- bool pinned = true;
bound = atomic_read(&vma->flags);
do {
@@ -786,34 +785,10 @@ static bool try_qad_pin(struct i915_vma *vma, unsigned int flags)
if (unlikely(bound & (I915_VMA_OVERFLOW | I915_VMA_ERROR)))
return false;
- if (!(bound & I915_VMA_PIN_MASK))
- goto unpinned;
-
GEM_BUG_ON(((bound + 1) & I915_VMA_PIN_MASK) == 0);
} while (!atomic_try_cmpxchg(&vma->flags, &bound, bound + 1));
return true;
-
-unpinned:
- /*
- * If pin_count==0, but we are bound, check under the lock to avoid
- * racing with a concurrent i915_vma_unbind().
- */
- mutex_lock(&vma->vm->mutex);
- do {
- if (unlikely(bound & (I915_VMA_OVERFLOW | I915_VMA_ERROR))) {
- pinned = false;
- break;
- }
-
- if (unlikely(flags & ~bound)) {
- pinned = false;
- break;
- }
- } while (!atomic_try_cmpxchg(&vma->flags, &bound, bound + 1));
- mutex_unlock(&vma->vm->mutex);
-
- return pinned;
}
static struct scatterlist *
@@ -1152,7 +1127,6 @@ static int
__i915_vma_get_pages(struct i915_vma *vma)
{
struct sg_table *pages;
- int ret;
/*
* The vma->pages are only valid within the lifespan of the borrowed
@@ -1185,18 +1159,16 @@ __i915_vma_get_pages(struct i915_vma *vma)
break;
}
- ret = 0;
if (IS_ERR(pages)) {
- ret = PTR_ERR(pages);
- pages = NULL;
drm_err(&vma->vm->i915->drm,
- "Failed to get pages for VMA view type %u (%d)!\n",
- vma->ggtt_view.type, ret);
+ "Failed to get pages for VMA view type %u (%ld)!\n",
+ vma->ggtt_view.type, PTR_ERR(pages));
+ return PTR_ERR(pages);
}
vma->pages = pages;
- return ret;
+ return 0;
}
I915_SELFTEST_EXPORT int i915_vma_get_pages(struct i915_vma *vma)
@@ -1228,25 +1200,14 @@ I915_SELFTEST_EXPORT int i915_vma_get_pages(struct i915_vma *vma)
static void __vma_put_pages(struct i915_vma *vma, unsigned int count)
{
/* We allocate under vma_get_pages, so beware the shrinker */
- struct sg_table *pages = READ_ONCE(vma->pages);
-
GEM_BUG_ON(atomic_read(&vma->pages_count) < count);
if (atomic_sub_return(count, &vma->pages_count) == 0) {
- /*
- * The atomic_sub_return is a read barrier for the READ_ONCE of
- * vma->pages above.
- *
- * READ_ONCE is safe because this is either called from the same
- * function (i915_vma_pin_ww), or guarded by vma->vm->mutex.
- *
- * TODO: We're leaving vma->pages dangling, until vma->obj->resv
- * lock is required.
- */
- if (pages != vma->obj->mm.pages) {
- sg_free_table(pages);
- kfree(pages);
+ if (vma->pages != vma->obj->mm.pages) {
+ sg_free_table(vma->pages);
+ kfree(vma->pages);
}
+ vma->pages = NULL;
i915_gem_object_unpin_pages(vma->obj);
}
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 16/17] drm/i915: Remove support for unlocked i915_vma unbind
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Niranjana Vishwanathapura, dri-devel
Now that we require the object lock for all ops, some code handling
race conditions can be removed.
This is required to not take short-term pins inside execbuf.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Acked-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
---
drivers/gpu/drm/i915/i915_vma.c | 55 +++++----------------------------
1 file changed, 8 insertions(+), 47 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 97093ecc3f97..bdd348fc755c 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -776,7 +776,6 @@ i915_vma_detach(struct i915_vma *vma)
static bool try_qad_pin(struct i915_vma *vma, unsigned int flags)
{
unsigned int bound;
- bool pinned = true;
bound = atomic_read(&vma->flags);
do {
@@ -786,34 +785,10 @@ static bool try_qad_pin(struct i915_vma *vma, unsigned int flags)
if (unlikely(bound & (I915_VMA_OVERFLOW | I915_VMA_ERROR)))
return false;
- if (!(bound & I915_VMA_PIN_MASK))
- goto unpinned;
-
GEM_BUG_ON(((bound + 1) & I915_VMA_PIN_MASK) == 0);
} while (!atomic_try_cmpxchg(&vma->flags, &bound, bound + 1));
return true;
-
-unpinned:
- /*
- * If pin_count==0, but we are bound, check under the lock to avoid
- * racing with a concurrent i915_vma_unbind().
- */
- mutex_lock(&vma->vm->mutex);
- do {
- if (unlikely(bound & (I915_VMA_OVERFLOW | I915_VMA_ERROR))) {
- pinned = false;
- break;
- }
-
- if (unlikely(flags & ~bound)) {
- pinned = false;
- break;
- }
- } while (!atomic_try_cmpxchg(&vma->flags, &bound, bound + 1));
- mutex_unlock(&vma->vm->mutex);
-
- return pinned;
}
static struct scatterlist *
@@ -1152,7 +1127,6 @@ static int
__i915_vma_get_pages(struct i915_vma *vma)
{
struct sg_table *pages;
- int ret;
/*
* The vma->pages are only valid within the lifespan of the borrowed
@@ -1185,18 +1159,16 @@ __i915_vma_get_pages(struct i915_vma *vma)
break;
}
- ret = 0;
if (IS_ERR(pages)) {
- ret = PTR_ERR(pages);
- pages = NULL;
drm_err(&vma->vm->i915->drm,
- "Failed to get pages for VMA view type %u (%d)!\n",
- vma->ggtt_view.type, ret);
+ "Failed to get pages for VMA view type %u (%ld)!\n",
+ vma->ggtt_view.type, PTR_ERR(pages));
+ return PTR_ERR(pages);
}
vma->pages = pages;
- return ret;
+ return 0;
}
I915_SELFTEST_EXPORT int i915_vma_get_pages(struct i915_vma *vma)
@@ -1228,25 +1200,14 @@ I915_SELFTEST_EXPORT int i915_vma_get_pages(struct i915_vma *vma)
static void __vma_put_pages(struct i915_vma *vma, unsigned int count)
{
/* We allocate under vma_get_pages, so beware the shrinker */
- struct sg_table *pages = READ_ONCE(vma->pages);
-
GEM_BUG_ON(atomic_read(&vma->pages_count) < count);
if (atomic_sub_return(count, &vma->pages_count) == 0) {
- /*
- * The atomic_sub_return is a read barrier for the READ_ONCE of
- * vma->pages above.
- *
- * READ_ONCE is safe because this is either called from the same
- * function (i915_vma_pin_ww), or guarded by vma->vm->mutex.
- *
- * TODO: We're leaving vma->pages dangling, until vma->obj->resv
- * lock is required.
- */
- if (pages != vma->obj->mm.pages) {
- sg_free_table(pages);
- kfree(pages);
+ if (vma->pages != vma->obj->mm.pages) {
+ sg_free_table(vma->pages);
+ kfree(vma->pages);
}
+ vma->pages = NULL;
i915_gem_object_unpin_pages(vma->obj);
}
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [PATCH v3 17/17] drm/i915: Remove short-term pins from execbuf, v5.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-16 14:27 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
Add a flag PIN_VALIDATE, to indicate we don't need to pin and only
protected by the object lock.
This removes the need to unpin, which is done by just releasing the
lock.
eb_reserve is slightly reworked for readability, but the same steps
are still done:
- First pass pins with NONBLOCK.
- Second pass unbinds all objects first, then pins.
- Third pass is only called when not all objects are softpinned, and
unbinds all objects, then calls i915_gem_evict_vm(), then pins.
When evicting the entire vm in eb_reserve() we do temporarily pin objects
that are marked with EXEC_OBJECT_PINNED. This is because they are already
at their destination, and i915_gem_evict_vm() would otherwise unbind them.
However, we reduce the visibility of those pins by limiting the pin
to our call to i915_gem_evict_vm() only, and pin with vm->mutex held,
instead of the entire duration of the execbuf.
Not sure the latter matters, one can hope..
In theory we could kill the pinning by adding an extra flag to the vma
to temporarily prevent unbinding for gtt for i915_gem_evict_vm only, but
I think that might be overkill. We're still holding the object lock, and
we don't have blocking eviction yet. It's likely sufficient to simply
enforce EXEC_OBJECT_PINNED for all objects on >= gen12.
Changes since v1:
- Split out eb_reserve() into separate functions for readability.
Changes since v2:
- Make batch buffer mappable on platforms where only GGTT is available,
to prevent moving the batch buffer during relocations.
Changes since v3:
- Preserve current behavior for batch buffer, instead be cautious when
calling i915_gem_object_ggtt_pin_ww, and re-use the current batch vma
if it's inside ggtt and map-and-fenceable.
- Remove impossible condition check from eb_reserve. (Matt)
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
.../gpu/drm/i915/gem/i915_gem_execbuffer.c | 250 ++++++++++--------
drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c | 1 -
drivers/gpu/drm/i915/i915_gem_gtt.h | 1 +
drivers/gpu/drm/i915/i915_vma.c | 24 +-
4 files changed, 158 insertions(+), 118 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index eb3649e844ff..3fe5d5811cdb 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -441,7 +441,7 @@ eb_pin_vma(struct i915_execbuffer *eb,
else
pin_flags = entry->offset & PIN_OFFSET_MASK;
- pin_flags |= PIN_USER | PIN_NOEVICT | PIN_OFFSET_FIXED;
+ pin_flags |= PIN_USER | PIN_NOEVICT | PIN_OFFSET_FIXED | PIN_VALIDATE;
if (unlikely(ev->flags & EXEC_OBJECT_NEEDS_GTT))
pin_flags |= PIN_GLOBAL;
@@ -459,17 +459,15 @@ eb_pin_vma(struct i915_execbuffer *eb,
entry->pad_to_size,
entry->alignment,
eb_pin_flags(entry, ev->flags) |
- PIN_USER | PIN_NOEVICT);
+ PIN_USER | PIN_NOEVICT | PIN_VALIDATE);
if (unlikely(err))
return err;
}
if (unlikely(ev->flags & EXEC_OBJECT_NEEDS_FENCE)) {
err = i915_vma_pin_fence(vma);
- if (unlikely(err)) {
- i915_vma_unpin(vma);
+ if (unlikely(err))
return err;
- }
if (vma->fence)
ev->flags |= __EXEC_OBJECT_HAS_FENCE;
@@ -485,13 +483,9 @@ eb_pin_vma(struct i915_execbuffer *eb,
static inline void
eb_unreserve_vma(struct eb_vma *ev)
{
- if (!(ev->flags & __EXEC_OBJECT_HAS_PIN))
- return;
-
if (unlikely(ev->flags & __EXEC_OBJECT_HAS_FENCE))
__i915_vma_unpin_fence(ev->vma);
- __i915_vma_unpin(ev->vma);
ev->flags &= ~__EXEC_OBJECT_RESERVED;
}
@@ -684,10 +678,8 @@ static int eb_reserve_vma(struct i915_execbuffer *eb,
if (unlikely(ev->flags & EXEC_OBJECT_NEEDS_FENCE)) {
err = i915_vma_pin_fence(vma);
- if (unlikely(err)) {
- i915_vma_unpin(vma);
+ if (unlikely(err))
return err;
- }
if (vma->fence)
ev->flags |= __EXEC_OBJECT_HAS_FENCE;
@@ -699,85 +691,125 @@ static int eb_reserve_vma(struct i915_execbuffer *eb,
return 0;
}
-static int eb_reserve(struct i915_execbuffer *eb)
+static int eb_evict_vm(struct i915_execbuffer *eb)
{
const unsigned int count = eb->buffer_count;
- unsigned int pin_flags = PIN_USER | PIN_NONBLOCK;
+ unsigned int i;
+ int err;
+
+ err = mutex_lock_interruptible(&eb->context->vm->mutex);
+ if (err)
+ return err;
+
+ /* pin to protect against i915_gem_evict_vm evicting below */
+ for (i = 0; i < count; i++) {
+ struct eb_vma *ev = &eb->vma[i];
+
+ if (ev->flags & __EXEC_OBJECT_HAS_PIN)
+ __i915_vma_pin(ev->vma);
+ }
+
+ /* Too fragmented, unbind everything and retry */
+ err = i915_gem_evict_vm(eb->context->vm, &eb->ww);
+
+ /* unpin objects.. */
+ for (i = 0; i < count; i++) {
+ struct eb_vma *ev = &eb->vma[i];
+
+ if (ev->flags & __EXEC_OBJECT_HAS_PIN)
+ i915_vma_unpin(ev->vma);
+ }
+
+ mutex_unlock(&eb->context->vm->mutex);
+
+ return err;
+}
+
+static bool eb_unbind(struct i915_execbuffer *eb)
+{
+ const unsigned int count = eb->buffer_count;
+ unsigned int i;
struct list_head last;
+ bool unpinned = false;
+
+ /* Resort *all* the objects into priority order */
+ INIT_LIST_HEAD(&eb->unbound);
+ INIT_LIST_HEAD(&last);
+
+ for (i = 0; i < count; i++) {
+ struct eb_vma *ev = &eb->vma[i];
+ unsigned int flags = ev->flags;
+
+ if (flags & EXEC_OBJECT_PINNED &&
+ flags & __EXEC_OBJECT_HAS_PIN)
+ continue;
+
+ unpinned = true;
+ eb_unreserve_vma(ev);
+
+ if (flags & EXEC_OBJECT_PINNED)
+ /* Pinned must have their slot */
+ list_add(&ev->bind_link, &eb->unbound);
+ else if (flags & __EXEC_OBJECT_NEEDS_MAP)
+ /* Map require the lowest 256MiB (aperture) */
+ list_add_tail(&ev->bind_link, &eb->unbound);
+ else if (!(flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS))
+ /* Prioritise 4GiB region for restricted bo */
+ list_add(&ev->bind_link, &last);
+ else
+ list_add_tail(&ev->bind_link, &last);
+ }
+
+ list_splice_tail(&last, &eb->unbound);
+ return unpinned;
+}
+
+static int eb_reserve(struct i915_execbuffer *eb)
+{
struct eb_vma *ev;
- unsigned int i, pass;
+ unsigned int pass;
int err = 0;
+ bool unpinned;
/*
* Attempt to pin all of the buffers into the GTT.
- * This is done in 3 phases:
+ * This is done in 2 phases:
*
- * 1a. Unbind all objects that do not match the GTT constraints for
- * the execbuffer (fenceable, mappable, alignment etc).
- * 1b. Increment pin count for already bound objects.
- * 2. Bind new objects.
- * 3. Decrement pin count.
+ * 1. Unbind all objects that do not match the GTT constraints for
+ * the execbuffer (fenceable, mappable, alignment etc).
+ * 2. Bind new objects.
*
* This avoid unnecessary unbinding of later objects in order to make
* room for the earlier objects *unless* we need to defragment.
+ *
+ * Defragmenting is skipped if all objects are pinned at a fixed location.
*/
- pass = 0;
- do {
- list_for_each_entry(ev, &eb->unbound, bind_link) {
- err = eb_reserve_vma(eb, ev, pin_flags);
- if (err)
- break;
- }
- if (err != -ENOSPC)
- return err;
+ for (pass = 0; pass <= 2; pass++) {
+ int pin_flags = PIN_USER | PIN_VALIDATE;
- /* Resort *all* the objects into priority order */
- INIT_LIST_HEAD(&eb->unbound);
- INIT_LIST_HEAD(&last);
- for (i = 0; i < count; i++) {
- unsigned int flags;
+ if (pass == 0)
+ pin_flags |= PIN_NONBLOCK;
- ev = &eb->vma[i];
- flags = ev->flags;
- if (flags & EXEC_OBJECT_PINNED &&
- flags & __EXEC_OBJECT_HAS_PIN)
- continue;
+ if (pass >= 1)
+ unpinned = eb_unbind(eb);
- eb_unreserve_vma(ev);
-
- if (flags & EXEC_OBJECT_PINNED)
- /* Pinned must have their slot */
- list_add(&ev->bind_link, &eb->unbound);
- else if (flags & __EXEC_OBJECT_NEEDS_MAP)
- /* Map require the lowest 256MiB (aperture) */
- list_add_tail(&ev->bind_link, &eb->unbound);
- else if (!(flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS))
- /* Prioritise 4GiB region for restricted bo */
- list_add(&ev->bind_link, &last);
- else
- list_add_tail(&ev->bind_link, &last);
- }
- list_splice_tail(&last, &eb->unbound);
-
- switch (pass++) {
- case 0:
- break;
-
- case 1:
- /* Too fragmented, unbind everything and retry */
- mutex_lock(&eb->context->vm->mutex);
- err = i915_gem_evict_vm(eb->context->vm, &eb->ww);
- mutex_unlock(&eb->context->vm->mutex);
+ if (pass == 2) {
+ err = eb_evict_vm(eb);
if (err)
return err;
- break;
+ }
- default:
- return -ENOSPC;
+ list_for_each_entry(ev, &eb->unbound, bind_link) {
+ err = eb_reserve_vma(eb, ev, pin_flags);
+ if (err)
+ break;
}
- pin_flags = PIN_USER;
- } while (1);
+ if (err != -ENOSPC)
+ break;
+ }
+
+ return err;
}
static int eb_select_context(struct i915_execbuffer *eb)
@@ -1184,10 +1216,11 @@ static void *reloc_kmap(struct drm_i915_gem_object *obj,
return vaddr;
}
-static void *reloc_iomap(struct drm_i915_gem_object *obj,
+static void *reloc_iomap(struct i915_vma *batch,
struct i915_execbuffer *eb,
unsigned long page)
{
+ struct drm_i915_gem_object *obj = batch->obj;
struct reloc_cache *cache = &eb->reloc_cache;
struct i915_ggtt *ggtt = cache_to_ggtt(cache);
unsigned long offset;
@@ -1197,7 +1230,7 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
intel_gt_flush_ggtt_writes(ggtt->vm.gt);
io_mapping_unmap_atomic((void __force __iomem *) unmask_page(cache->vaddr));
} else {
- struct i915_vma *vma;
+ struct i915_vma *vma = ERR_PTR(-ENODEV);
int err;
if (i915_gem_object_is_tiled(obj))
@@ -1210,10 +1243,23 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
if (err)
return ERR_PTR(err);
- vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0,
- PIN_MAPPABLE |
- PIN_NONBLOCK /* NOWARN */ |
- PIN_NOEVICT);
+ /*
+ * i915_gem_object_ggtt_pin_ww may attempt to remove the batch
+ * VMA from the object list because we no longer pin.
+ *
+ * Only attempt to pin the batch buffer to ggtt if the current batch
+ * is not inside ggtt, or the batch buffer is not misplaced.
+ */
+ if (!i915_is_ggtt(batch->vm)) {
+ vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0,
+ PIN_MAPPABLE |
+ PIN_NONBLOCK /* NOWARN */ |
+ PIN_NOEVICT);
+ } else if (i915_vma_is_map_and_fenceable(batch)) {
+ __i915_vma_pin(batch);
+ vma = batch;
+ }
+
if (vma == ERR_PTR(-EDEADLK))
return vma;
@@ -1251,7 +1297,7 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
return vaddr;
}
-static void *reloc_vaddr(struct drm_i915_gem_object *obj,
+static void *reloc_vaddr(struct i915_vma *vma,
struct i915_execbuffer *eb,
unsigned long page)
{
@@ -1263,9 +1309,9 @@ static void *reloc_vaddr(struct drm_i915_gem_object *obj,
} else {
vaddr = NULL;
if ((cache->vaddr & KMAP) == 0)
- vaddr = reloc_iomap(obj, eb, page);
+ vaddr = reloc_iomap(vma, eb, page);
if (!vaddr)
- vaddr = reloc_kmap(obj, cache, page);
+ vaddr = reloc_kmap(vma->obj, cache, page);
}
return vaddr;
@@ -1306,7 +1352,7 @@ relocate_entry(struct i915_vma *vma,
void *vaddr;
repeat:
- vaddr = reloc_vaddr(vma->obj, eb,
+ vaddr = reloc_vaddr(vma, eb,
offset >> PAGE_SHIFT);
if (IS_ERR(vaddr))
return PTR_ERR(vaddr);
@@ -2161,7 +2207,7 @@ shadow_batch_pin(struct i915_execbuffer *eb,
if (IS_ERR(vma))
return vma;
- err = i915_vma_pin_ww(vma, &eb->ww, 0, 0, flags);
+ err = i915_vma_pin_ww(vma, &eb->ww, 0, 0, flags | PIN_VALIDATE);
if (err)
return ERR_PTR(err);
@@ -2175,7 +2221,7 @@ static struct i915_vma *eb_dispatch_secure(struct i915_execbuffer *eb, struct i9
* batch" bit. Hence we need to pin secure batches into the global gtt.
* hsw should have this fixed, but bdw mucks it up again. */
if (eb->batch_flags & I915_DISPATCH_SECURE)
- return i915_gem_object_ggtt_pin_ww(vma->obj, &eb->ww, NULL, 0, 0, 0);
+ return i915_gem_object_ggtt_pin_ww(vma->obj, &eb->ww, NULL, 0, 0, PIN_VALIDATE);
return NULL;
}
@@ -2226,13 +2272,12 @@ static int eb_parse(struct i915_execbuffer *eb)
err = i915_gem_object_lock(pool->obj, &eb->ww);
if (err)
- goto err;
+ return err;
shadow = shadow_batch_pin(eb, pool->obj, eb->context->vm, PIN_USER);
- if (IS_ERR(shadow)) {
- err = PTR_ERR(shadow);
- goto err;
- }
+ if (IS_ERR(shadow))
+ return PTR_ERR(shadow);
+
intel_gt_buffer_pool_mark_used(pool);
i915_gem_object_set_readonly(shadow->obj);
shadow->private = pool;
@@ -2244,25 +2289,21 @@ static int eb_parse(struct i915_execbuffer *eb)
shadow = shadow_batch_pin(eb, pool->obj,
&eb->gt->ggtt->vm,
PIN_GLOBAL);
- if (IS_ERR(shadow)) {
- err = PTR_ERR(shadow);
- shadow = trampoline;
- goto err_shadow;
- }
+ if (IS_ERR(shadow))
+ return PTR_ERR(shadow);
+
shadow->private = pool;
eb->batch_flags |= I915_DISPATCH_SECURE;
}
batch = eb_dispatch_secure(eb, shadow);
- if (IS_ERR(batch)) {
- err = PTR_ERR(batch);
- goto err_trampoline;
- }
+ if (IS_ERR(batch))
+ return PTR_ERR(batch);
err = dma_resv_reserve_shared(shadow->obj->base.resv, 1);
if (err)
- goto err_trampoline;
+ return err;
err = intel_engine_cmd_parser(eb->context->engine,
eb->batches[0]->vma,
@@ -2270,7 +2311,7 @@ static int eb_parse(struct i915_execbuffer *eb)
eb->batch_len[0],
shadow, trampoline);
if (err)
- goto err_unpin_batch;
+ return err;
eb->batches[0] = &eb->vma[eb->buffer_count++];
eb->batches[0]->vma = i915_vma_get(shadow);
@@ -2289,17 +2330,6 @@ static int eb_parse(struct i915_execbuffer *eb)
eb->batches[0]->vma = i915_vma_get(batch);
}
return 0;
-
-err_unpin_batch:
- if (batch)
- i915_vma_unpin(batch);
-err_trampoline:
- if (trampoline)
- i915_vma_unpin(trampoline);
-err_shadow:
- i915_vma_unpin(shadow);
-err:
- return err;
}
static int eb_request_submit(struct i915_execbuffer *eb,
@@ -3420,8 +3450,6 @@ i915_gem_do_execbuffer(struct drm_device *dev,
err_vma:
eb_release_vmas(&eb, true);
- if (eb.trampoline)
- i915_vma_unpin(eb.trampoline);
WARN_ON(err == -EDEADLK);
i915_gem_ww_ctx_fini(&eb.ww);
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
index f8948de72036..bbcd0522f68e 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
@@ -425,7 +425,6 @@ int i915_vma_pin_fence(struct i915_vma *vma)
* must keep the device awake whilst using the fence.
*/
assert_rpm_wakelock_held(vma->vm->gt->uncore->rpm);
- GEM_BUG_ON(!i915_vma_is_pinned(vma));
GEM_BUG_ON(!i915_vma_is_ggtt(vma));
err = mutex_lock_interruptible(&vma->vm->mutex);
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index e4938aba3fe9..8c2f57eb5dda 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -44,6 +44,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
#define PIN_HIGH BIT_ULL(5)
#define PIN_OFFSET_BIAS BIT_ULL(6)
#define PIN_OFFSET_FIXED BIT_ULL(7)
+#define PIN_VALIDATE BIT_ULL(8) /* validate placement only, no need to call unpin() */
#define PIN_GLOBAL BIT_ULL(10) /* I915_VMA_GLOBAL_BIND */
#define PIN_USER BIT_ULL(11) /* I915_VMA_LOCAL_BIND */
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index bdd348fc755c..aa951a246802 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -778,6 +778,15 @@ static bool try_qad_pin(struct i915_vma *vma, unsigned int flags)
unsigned int bound;
bound = atomic_read(&vma->flags);
+
+ if (flags & PIN_VALIDATE) {
+ flags &= I915_VMA_BIND_MASK;
+
+ return (flags & bound) == flags;
+ }
+
+ /* with the lock mandatory for unbind, we don't race here */
+ flags &= I915_VMA_BIND_MASK;
do {
if (unlikely(flags & ~bound))
return false;
@@ -1253,7 +1262,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
GEM_BUG_ON(!(flags & (PIN_USER | PIN_GLOBAL)));
/* First try and grab the pin without rebinding the vma */
- if (try_qad_pin(vma, flags & I915_VMA_BIND_MASK))
+ if (try_qad_pin(vma, flags))
return 0;
err = i915_vma_get_pages(vma);
@@ -1335,7 +1344,8 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
}
if (unlikely(!(flags & ~bound & I915_VMA_BIND_MASK))) {
- __i915_vma_pin(vma);
+ if (!(flags & PIN_VALIDATE))
+ __i915_vma_pin(vma);
goto err_unlock;
}
@@ -1364,8 +1374,10 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
atomic_add(I915_VMA_PAGES_ACTIVE, &vma->pages_count);
list_move_tail(&vma->vm_link, &vma->vm->bound_list);
- __i915_vma_pin(vma);
- GEM_BUG_ON(!i915_vma_is_pinned(vma));
+ if (!(flags & PIN_VALIDATE)) {
+ __i915_vma_pin(vma);
+ GEM_BUG_ON(!i915_vma_is_pinned(vma));
+ }
GEM_BUG_ON(!i915_vma_is_bound(vma, flags));
GEM_BUG_ON(i915_vma_misplaced(vma, size, alignment, flags));
@@ -1625,8 +1637,6 @@ static int __i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *
{
int err;
- GEM_BUG_ON(!i915_vma_is_pinned(vma));
-
/* Wait for the vma to be bound before we start! */
err = __i915_request_await_bind(rq, vma);
if (err)
@@ -1645,6 +1655,8 @@ int _i915_vma_move_to_active(struct i915_vma *vma,
assert_object_held(obj);
+ GEM_BUG_ON(!vma->pages);
+
err = __i915_vma_move_to_active(vma, rq);
if (unlikely(err))
return err;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] [PATCH v3 17/17] drm/i915: Remove short-term pins from execbuf, v5.
@ 2021-12-16 14:27 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-16 14:27 UTC (permalink / raw)
To: intel-gfx; +Cc: Matthew Auld, dri-devel
Add a flag PIN_VALIDATE, to indicate we don't need to pin and only
protected by the object lock.
This removes the need to unpin, which is done by just releasing the
lock.
eb_reserve is slightly reworked for readability, but the same steps
are still done:
- First pass pins with NONBLOCK.
- Second pass unbinds all objects first, then pins.
- Third pass is only called when not all objects are softpinned, and
unbinds all objects, then calls i915_gem_evict_vm(), then pins.
When evicting the entire vm in eb_reserve() we do temporarily pin objects
that are marked with EXEC_OBJECT_PINNED. This is because they are already
at their destination, and i915_gem_evict_vm() would otherwise unbind them.
However, we reduce the visibility of those pins by limiting the pin
to our call to i915_gem_evict_vm() only, and pin with vm->mutex held,
instead of the entire duration of the execbuf.
Not sure the latter matters, one can hope..
In theory we could kill the pinning by adding an extra flag to the vma
to temporarily prevent unbinding for gtt for i915_gem_evict_vm only, but
I think that might be overkill. We're still holding the object lock, and
we don't have blocking eviction yet. It's likely sufficient to simply
enforce EXEC_OBJECT_PINNED for all objects on >= gen12.
Changes since v1:
- Split out eb_reserve() into separate functions for readability.
Changes since v2:
- Make batch buffer mappable on platforms where only GGTT is available,
to prevent moving the batch buffer during relocations.
Changes since v3:
- Preserve current behavior for batch buffer, instead be cautious when
calling i915_gem_object_ggtt_pin_ww, and re-use the current batch vma
if it's inside ggtt and map-and-fenceable.
- Remove impossible condition check from eb_reserve. (Matt)
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
---
.../gpu/drm/i915/gem/i915_gem_execbuffer.c | 250 ++++++++++--------
drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c | 1 -
drivers/gpu/drm/i915/i915_gem_gtt.h | 1 +
drivers/gpu/drm/i915/i915_vma.c | 24 +-
4 files changed, 158 insertions(+), 118 deletions(-)
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index eb3649e844ff..3fe5d5811cdb 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -441,7 +441,7 @@ eb_pin_vma(struct i915_execbuffer *eb,
else
pin_flags = entry->offset & PIN_OFFSET_MASK;
- pin_flags |= PIN_USER | PIN_NOEVICT | PIN_OFFSET_FIXED;
+ pin_flags |= PIN_USER | PIN_NOEVICT | PIN_OFFSET_FIXED | PIN_VALIDATE;
if (unlikely(ev->flags & EXEC_OBJECT_NEEDS_GTT))
pin_flags |= PIN_GLOBAL;
@@ -459,17 +459,15 @@ eb_pin_vma(struct i915_execbuffer *eb,
entry->pad_to_size,
entry->alignment,
eb_pin_flags(entry, ev->flags) |
- PIN_USER | PIN_NOEVICT);
+ PIN_USER | PIN_NOEVICT | PIN_VALIDATE);
if (unlikely(err))
return err;
}
if (unlikely(ev->flags & EXEC_OBJECT_NEEDS_FENCE)) {
err = i915_vma_pin_fence(vma);
- if (unlikely(err)) {
- i915_vma_unpin(vma);
+ if (unlikely(err))
return err;
- }
if (vma->fence)
ev->flags |= __EXEC_OBJECT_HAS_FENCE;
@@ -485,13 +483,9 @@ eb_pin_vma(struct i915_execbuffer *eb,
static inline void
eb_unreserve_vma(struct eb_vma *ev)
{
- if (!(ev->flags & __EXEC_OBJECT_HAS_PIN))
- return;
-
if (unlikely(ev->flags & __EXEC_OBJECT_HAS_FENCE))
__i915_vma_unpin_fence(ev->vma);
- __i915_vma_unpin(ev->vma);
ev->flags &= ~__EXEC_OBJECT_RESERVED;
}
@@ -684,10 +678,8 @@ static int eb_reserve_vma(struct i915_execbuffer *eb,
if (unlikely(ev->flags & EXEC_OBJECT_NEEDS_FENCE)) {
err = i915_vma_pin_fence(vma);
- if (unlikely(err)) {
- i915_vma_unpin(vma);
+ if (unlikely(err))
return err;
- }
if (vma->fence)
ev->flags |= __EXEC_OBJECT_HAS_FENCE;
@@ -699,85 +691,125 @@ static int eb_reserve_vma(struct i915_execbuffer *eb,
return 0;
}
-static int eb_reserve(struct i915_execbuffer *eb)
+static int eb_evict_vm(struct i915_execbuffer *eb)
{
const unsigned int count = eb->buffer_count;
- unsigned int pin_flags = PIN_USER | PIN_NONBLOCK;
+ unsigned int i;
+ int err;
+
+ err = mutex_lock_interruptible(&eb->context->vm->mutex);
+ if (err)
+ return err;
+
+ /* pin to protect against i915_gem_evict_vm evicting below */
+ for (i = 0; i < count; i++) {
+ struct eb_vma *ev = &eb->vma[i];
+
+ if (ev->flags & __EXEC_OBJECT_HAS_PIN)
+ __i915_vma_pin(ev->vma);
+ }
+
+ /* Too fragmented, unbind everything and retry */
+ err = i915_gem_evict_vm(eb->context->vm, &eb->ww);
+
+ /* unpin objects.. */
+ for (i = 0; i < count; i++) {
+ struct eb_vma *ev = &eb->vma[i];
+
+ if (ev->flags & __EXEC_OBJECT_HAS_PIN)
+ i915_vma_unpin(ev->vma);
+ }
+
+ mutex_unlock(&eb->context->vm->mutex);
+
+ return err;
+}
+
+static bool eb_unbind(struct i915_execbuffer *eb)
+{
+ const unsigned int count = eb->buffer_count;
+ unsigned int i;
struct list_head last;
+ bool unpinned = false;
+
+ /* Resort *all* the objects into priority order */
+ INIT_LIST_HEAD(&eb->unbound);
+ INIT_LIST_HEAD(&last);
+
+ for (i = 0; i < count; i++) {
+ struct eb_vma *ev = &eb->vma[i];
+ unsigned int flags = ev->flags;
+
+ if (flags & EXEC_OBJECT_PINNED &&
+ flags & __EXEC_OBJECT_HAS_PIN)
+ continue;
+
+ unpinned = true;
+ eb_unreserve_vma(ev);
+
+ if (flags & EXEC_OBJECT_PINNED)
+ /* Pinned must have their slot */
+ list_add(&ev->bind_link, &eb->unbound);
+ else if (flags & __EXEC_OBJECT_NEEDS_MAP)
+ /* Map require the lowest 256MiB (aperture) */
+ list_add_tail(&ev->bind_link, &eb->unbound);
+ else if (!(flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS))
+ /* Prioritise 4GiB region for restricted bo */
+ list_add(&ev->bind_link, &last);
+ else
+ list_add_tail(&ev->bind_link, &last);
+ }
+
+ list_splice_tail(&last, &eb->unbound);
+ return unpinned;
+}
+
+static int eb_reserve(struct i915_execbuffer *eb)
+{
struct eb_vma *ev;
- unsigned int i, pass;
+ unsigned int pass;
int err = 0;
+ bool unpinned;
/*
* Attempt to pin all of the buffers into the GTT.
- * This is done in 3 phases:
+ * This is done in 2 phases:
*
- * 1a. Unbind all objects that do not match the GTT constraints for
- * the execbuffer (fenceable, mappable, alignment etc).
- * 1b. Increment pin count for already bound objects.
- * 2. Bind new objects.
- * 3. Decrement pin count.
+ * 1. Unbind all objects that do not match the GTT constraints for
+ * the execbuffer (fenceable, mappable, alignment etc).
+ * 2. Bind new objects.
*
* This avoid unnecessary unbinding of later objects in order to make
* room for the earlier objects *unless* we need to defragment.
+ *
+ * Defragmenting is skipped if all objects are pinned at a fixed location.
*/
- pass = 0;
- do {
- list_for_each_entry(ev, &eb->unbound, bind_link) {
- err = eb_reserve_vma(eb, ev, pin_flags);
- if (err)
- break;
- }
- if (err != -ENOSPC)
- return err;
+ for (pass = 0; pass <= 2; pass++) {
+ int pin_flags = PIN_USER | PIN_VALIDATE;
- /* Resort *all* the objects into priority order */
- INIT_LIST_HEAD(&eb->unbound);
- INIT_LIST_HEAD(&last);
- for (i = 0; i < count; i++) {
- unsigned int flags;
+ if (pass == 0)
+ pin_flags |= PIN_NONBLOCK;
- ev = &eb->vma[i];
- flags = ev->flags;
- if (flags & EXEC_OBJECT_PINNED &&
- flags & __EXEC_OBJECT_HAS_PIN)
- continue;
+ if (pass >= 1)
+ unpinned = eb_unbind(eb);
- eb_unreserve_vma(ev);
-
- if (flags & EXEC_OBJECT_PINNED)
- /* Pinned must have their slot */
- list_add(&ev->bind_link, &eb->unbound);
- else if (flags & __EXEC_OBJECT_NEEDS_MAP)
- /* Map require the lowest 256MiB (aperture) */
- list_add_tail(&ev->bind_link, &eb->unbound);
- else if (!(flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS))
- /* Prioritise 4GiB region for restricted bo */
- list_add(&ev->bind_link, &last);
- else
- list_add_tail(&ev->bind_link, &last);
- }
- list_splice_tail(&last, &eb->unbound);
-
- switch (pass++) {
- case 0:
- break;
-
- case 1:
- /* Too fragmented, unbind everything and retry */
- mutex_lock(&eb->context->vm->mutex);
- err = i915_gem_evict_vm(eb->context->vm, &eb->ww);
- mutex_unlock(&eb->context->vm->mutex);
+ if (pass == 2) {
+ err = eb_evict_vm(eb);
if (err)
return err;
- break;
+ }
- default:
- return -ENOSPC;
+ list_for_each_entry(ev, &eb->unbound, bind_link) {
+ err = eb_reserve_vma(eb, ev, pin_flags);
+ if (err)
+ break;
}
- pin_flags = PIN_USER;
- } while (1);
+ if (err != -ENOSPC)
+ break;
+ }
+
+ return err;
}
static int eb_select_context(struct i915_execbuffer *eb)
@@ -1184,10 +1216,11 @@ static void *reloc_kmap(struct drm_i915_gem_object *obj,
return vaddr;
}
-static void *reloc_iomap(struct drm_i915_gem_object *obj,
+static void *reloc_iomap(struct i915_vma *batch,
struct i915_execbuffer *eb,
unsigned long page)
{
+ struct drm_i915_gem_object *obj = batch->obj;
struct reloc_cache *cache = &eb->reloc_cache;
struct i915_ggtt *ggtt = cache_to_ggtt(cache);
unsigned long offset;
@@ -1197,7 +1230,7 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
intel_gt_flush_ggtt_writes(ggtt->vm.gt);
io_mapping_unmap_atomic((void __force __iomem *) unmask_page(cache->vaddr));
} else {
- struct i915_vma *vma;
+ struct i915_vma *vma = ERR_PTR(-ENODEV);
int err;
if (i915_gem_object_is_tiled(obj))
@@ -1210,10 +1243,23 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
if (err)
return ERR_PTR(err);
- vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0,
- PIN_MAPPABLE |
- PIN_NONBLOCK /* NOWARN */ |
- PIN_NOEVICT);
+ /*
+ * i915_gem_object_ggtt_pin_ww may attempt to remove the batch
+ * VMA from the object list because we no longer pin.
+ *
+ * Only attempt to pin the batch buffer to ggtt if the current batch
+ * is not inside ggtt, or the batch buffer is not misplaced.
+ */
+ if (!i915_is_ggtt(batch->vm)) {
+ vma = i915_gem_object_ggtt_pin_ww(obj, &eb->ww, NULL, 0, 0,
+ PIN_MAPPABLE |
+ PIN_NONBLOCK /* NOWARN */ |
+ PIN_NOEVICT);
+ } else if (i915_vma_is_map_and_fenceable(batch)) {
+ __i915_vma_pin(batch);
+ vma = batch;
+ }
+
if (vma == ERR_PTR(-EDEADLK))
return vma;
@@ -1251,7 +1297,7 @@ static void *reloc_iomap(struct drm_i915_gem_object *obj,
return vaddr;
}
-static void *reloc_vaddr(struct drm_i915_gem_object *obj,
+static void *reloc_vaddr(struct i915_vma *vma,
struct i915_execbuffer *eb,
unsigned long page)
{
@@ -1263,9 +1309,9 @@ static void *reloc_vaddr(struct drm_i915_gem_object *obj,
} else {
vaddr = NULL;
if ((cache->vaddr & KMAP) == 0)
- vaddr = reloc_iomap(obj, eb, page);
+ vaddr = reloc_iomap(vma, eb, page);
if (!vaddr)
- vaddr = reloc_kmap(obj, cache, page);
+ vaddr = reloc_kmap(vma->obj, cache, page);
}
return vaddr;
@@ -1306,7 +1352,7 @@ relocate_entry(struct i915_vma *vma,
void *vaddr;
repeat:
- vaddr = reloc_vaddr(vma->obj, eb,
+ vaddr = reloc_vaddr(vma, eb,
offset >> PAGE_SHIFT);
if (IS_ERR(vaddr))
return PTR_ERR(vaddr);
@@ -2161,7 +2207,7 @@ shadow_batch_pin(struct i915_execbuffer *eb,
if (IS_ERR(vma))
return vma;
- err = i915_vma_pin_ww(vma, &eb->ww, 0, 0, flags);
+ err = i915_vma_pin_ww(vma, &eb->ww, 0, 0, flags | PIN_VALIDATE);
if (err)
return ERR_PTR(err);
@@ -2175,7 +2221,7 @@ static struct i915_vma *eb_dispatch_secure(struct i915_execbuffer *eb, struct i9
* batch" bit. Hence we need to pin secure batches into the global gtt.
* hsw should have this fixed, but bdw mucks it up again. */
if (eb->batch_flags & I915_DISPATCH_SECURE)
- return i915_gem_object_ggtt_pin_ww(vma->obj, &eb->ww, NULL, 0, 0, 0);
+ return i915_gem_object_ggtt_pin_ww(vma->obj, &eb->ww, NULL, 0, 0, PIN_VALIDATE);
return NULL;
}
@@ -2226,13 +2272,12 @@ static int eb_parse(struct i915_execbuffer *eb)
err = i915_gem_object_lock(pool->obj, &eb->ww);
if (err)
- goto err;
+ return err;
shadow = shadow_batch_pin(eb, pool->obj, eb->context->vm, PIN_USER);
- if (IS_ERR(shadow)) {
- err = PTR_ERR(shadow);
- goto err;
- }
+ if (IS_ERR(shadow))
+ return PTR_ERR(shadow);
+
intel_gt_buffer_pool_mark_used(pool);
i915_gem_object_set_readonly(shadow->obj);
shadow->private = pool;
@@ -2244,25 +2289,21 @@ static int eb_parse(struct i915_execbuffer *eb)
shadow = shadow_batch_pin(eb, pool->obj,
&eb->gt->ggtt->vm,
PIN_GLOBAL);
- if (IS_ERR(shadow)) {
- err = PTR_ERR(shadow);
- shadow = trampoline;
- goto err_shadow;
- }
+ if (IS_ERR(shadow))
+ return PTR_ERR(shadow);
+
shadow->private = pool;
eb->batch_flags |= I915_DISPATCH_SECURE;
}
batch = eb_dispatch_secure(eb, shadow);
- if (IS_ERR(batch)) {
- err = PTR_ERR(batch);
- goto err_trampoline;
- }
+ if (IS_ERR(batch))
+ return PTR_ERR(batch);
err = dma_resv_reserve_shared(shadow->obj->base.resv, 1);
if (err)
- goto err_trampoline;
+ return err;
err = intel_engine_cmd_parser(eb->context->engine,
eb->batches[0]->vma,
@@ -2270,7 +2311,7 @@ static int eb_parse(struct i915_execbuffer *eb)
eb->batch_len[0],
shadow, trampoline);
if (err)
- goto err_unpin_batch;
+ return err;
eb->batches[0] = &eb->vma[eb->buffer_count++];
eb->batches[0]->vma = i915_vma_get(shadow);
@@ -2289,17 +2330,6 @@ static int eb_parse(struct i915_execbuffer *eb)
eb->batches[0]->vma = i915_vma_get(batch);
}
return 0;
-
-err_unpin_batch:
- if (batch)
- i915_vma_unpin(batch);
-err_trampoline:
- if (trampoline)
- i915_vma_unpin(trampoline);
-err_shadow:
- i915_vma_unpin(shadow);
-err:
- return err;
}
static int eb_request_submit(struct i915_execbuffer *eb,
@@ -3420,8 +3450,6 @@ i915_gem_do_execbuffer(struct drm_device *dev,
err_vma:
eb_release_vmas(&eb, true);
- if (eb.trampoline)
- i915_vma_unpin(eb.trampoline);
WARN_ON(err == -EDEADLK);
i915_gem_ww_ctx_fini(&eb.ww);
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
index f8948de72036..bbcd0522f68e 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
@@ -425,7 +425,6 @@ int i915_vma_pin_fence(struct i915_vma *vma)
* must keep the device awake whilst using the fence.
*/
assert_rpm_wakelock_held(vma->vm->gt->uncore->rpm);
- GEM_BUG_ON(!i915_vma_is_pinned(vma));
GEM_BUG_ON(!i915_vma_is_ggtt(vma));
err = mutex_lock_interruptible(&vma->vm->mutex);
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index e4938aba3fe9..8c2f57eb5dda 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -44,6 +44,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
#define PIN_HIGH BIT_ULL(5)
#define PIN_OFFSET_BIAS BIT_ULL(6)
#define PIN_OFFSET_FIXED BIT_ULL(7)
+#define PIN_VALIDATE BIT_ULL(8) /* validate placement only, no need to call unpin() */
#define PIN_GLOBAL BIT_ULL(10) /* I915_VMA_GLOBAL_BIND */
#define PIN_USER BIT_ULL(11) /* I915_VMA_LOCAL_BIND */
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index bdd348fc755c..aa951a246802 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -778,6 +778,15 @@ static bool try_qad_pin(struct i915_vma *vma, unsigned int flags)
unsigned int bound;
bound = atomic_read(&vma->flags);
+
+ if (flags & PIN_VALIDATE) {
+ flags &= I915_VMA_BIND_MASK;
+
+ return (flags & bound) == flags;
+ }
+
+ /* with the lock mandatory for unbind, we don't race here */
+ flags &= I915_VMA_BIND_MASK;
do {
if (unlikely(flags & ~bound))
return false;
@@ -1253,7 +1262,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
GEM_BUG_ON(!(flags & (PIN_USER | PIN_GLOBAL)));
/* First try and grab the pin without rebinding the vma */
- if (try_qad_pin(vma, flags & I915_VMA_BIND_MASK))
+ if (try_qad_pin(vma, flags))
return 0;
err = i915_vma_get_pages(vma);
@@ -1335,7 +1344,8 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
}
if (unlikely(!(flags & ~bound & I915_VMA_BIND_MASK))) {
- __i915_vma_pin(vma);
+ if (!(flags & PIN_VALIDATE))
+ __i915_vma_pin(vma);
goto err_unlock;
}
@@ -1364,8 +1374,10 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
atomic_add(I915_VMA_PAGES_ACTIVE, &vma->pages_count);
list_move_tail(&vma->vm_link, &vma->vm->bound_list);
- __i915_vma_pin(vma);
- GEM_BUG_ON(!i915_vma_is_pinned(vma));
+ if (!(flags & PIN_VALIDATE)) {
+ __i915_vma_pin(vma);
+ GEM_BUG_ON(!i915_vma_is_pinned(vma));
+ }
GEM_BUG_ON(!i915_vma_is_bound(vma, flags));
GEM_BUG_ON(i915_vma_misplaced(vma, size, alignment, flags));
@@ -1625,8 +1637,6 @@ static int __i915_vma_move_to_active(struct i915_vma *vma, struct i915_request *
{
int err;
- GEM_BUG_ON(!i915_vma_is_pinned(vma));
-
/* Wait for the vma to be bound before we start! */
err = __i915_request_await_bind(rq, vma);
if (err)
@@ -1645,6 +1655,8 @@ int _i915_vma_move_to_active(struct i915_vma *vma,
assert_object_held(obj);
+ GEM_BUG_ON(!vma->pages);
+
err = __i915_vma_move_to_active(vma, rq);
if (unlikely(err))
return err;
--
2.34.1
^ permalink raw reply related [flat|nested] 57+ messages in thread
* [Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
` (17 preceding siblings ...)
(?)
@ 2021-12-17 8:46 ` Patchwork
-1 siblings, 0 replies; 57+ messages in thread
From: Patchwork @ 2021-12-17 8:46 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: intel-gfx
== Series Details ==
Series: drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
URL : https://patchwork.freedesktop.org/series/98137/
State : warning
== Summary ==
$ dim checkpatch origin/drm-tip
bc08f58eeb3f drm/i915: Remove unused bits of i915_vma/active api
91bcecf270f0 drm/i915: Change shrink ordering to use locking around unbinding.
-:28: CHECK:PARENTHESIS_ALIGNMENT: Alignment should match open parenthesis
#28: FILE: drivers/gpu/drm/i915/gem/i915_gem_shrinker.c:40:
+static int drop_pages(struct drm_i915_gem_object *obj,
+ unsigned long shrink, bool trylock_vm)
total: 0 errors, 0 warnings, 1 checks, 52 lines checked
316d336c98ea drm/i915: Remove pages_mutex and intel_gtt->vma_ops.set/clear_pages members, v3.
9d385c135ad3 drm/i915: Take object lock in i915_ggtt_pin if ww is not set
c293606bcd36 drm/i915: Force ww lock for i915_gem_object_ggtt_pin_ww, v2.
e7fbd118c9b2 drm/i915: Ensure gem_contexts selftests work with unbind changes, v2.
c9892c726d3c drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking changes.
cc5dd7eac42f drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors
1631c2059017 drm/i915: Trylock the object when shrinking
d1406f612444 drm/i915: Require object lock when freeing pages during destruction
97cc19bfa2a0 drm/i915: Add ww ctx to i915_gem_object_trylock
44fc65c4b062 drm/i915: Add locking to i915_gem_evict_vm()
8b8d9bb4ffaa drm/i915: Add object locking to i915_gem_evict_for_node and i915_gem_evict_something
-:148: CHECK:LINE_SPACING: Please don't use multiple blank lines
#148: FILE: drivers/gpu/drm/i915/i915_gem_evict.c:252:
+
total: 0 errors, 0 warnings, 1 checks, 364 lines checked
d2242714f33e drm/i915: Add i915_vma_unbind_unlocked, and take obj lock for i915_vma_unbind, v2.
-:7: WARNING:COMMIT_LOG_LONG_LINE: Possible unwrapped commit description (prefer a maximum 75 chars per line)
#7:
We want to remove more members of i915_vma, which requires the locking to be
total: 0 errors, 1 warnings, 0 checks, 317 lines checked
ece7fb471984 drm/i915: Remove assert_object_held_shared
d5c98a7587ba drm/i915: Remove support for unlocked i915_vma unbind
c54bb29dd0de drm/i915: Remove short-term pins from execbuf, v5.
^ permalink raw reply [flat|nested] 57+ messages in thread
* [Intel-gfx] ✗ Fi.CI.SPARSE: warning for drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
` (18 preceding siblings ...)
(?)
@ 2021-12-17 8:47 ` Patchwork
-1 siblings, 0 replies; 57+ messages in thread
From: Patchwork @ 2021-12-17 8:47 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: intel-gfx
== Series Details ==
Series: drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
URL : https://patchwork.freedesktop.org/series/98137/
State : warning
== Summary ==
$ dim sparse --fast origin/drm-tip
Sparse version: v0.6.2
Fast mode used, each commit won't be checked separately.
^ permalink raw reply [flat|nested] 57+ messages in thread
* [Intel-gfx] ✗ Fi.CI.DOCS: warning for drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
` (19 preceding siblings ...)
(?)
@ 2021-12-17 8:51 ` Patchwork
-1 siblings, 0 replies; 57+ messages in thread
From: Patchwork @ 2021-12-17 8:51 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: intel-gfx
== Series Details ==
Series: drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
URL : https://patchwork.freedesktop.org/series/98137/
State : warning
== Summary ==
$ make htmldocs 2>&1 > /dev/null | grep i915
./drivers/gpu/drm/i915/i915_gem_evict.c:110: warning: Function parameter or member 'ww' not described in 'i915_gem_evict_something'
./drivers/gpu/drm/i915/i915_gem_evict.c:281: warning: Function parameter or member 'ww' not described in 'i915_gem_evict_for_node'
./drivers/gpu/drm/i915/i915_gem_evict.c:393: warning: Function parameter or member 'ww' not described in 'i915_gem_evict_vm'
./drivers/gpu/drm/i915/i915_gem_gtt.c:100: warning: Function parameter or member 'ww' not described in 'i915_gem_gtt_reserve'
./drivers/gpu/drm/i915/i915_gem_gtt.c:192: warning: Function parameter or member 'ww' not described in 'i915_gem_gtt_insert'
^ permalink raw reply [flat|nested] 57+ messages in thread
* [Intel-gfx] ✓ Fi.CI.BAT: success for drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
` (20 preceding siblings ...)
(?)
@ 2021-12-17 9:10 ` Patchwork
-1 siblings, 0 replies; 57+ messages in thread
From: Patchwork @ 2021-12-17 9:10 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: intel-gfx
[-- Attachment #1: Type: text/plain, Size: 4087 bytes --]
== Series Details ==
Series: drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
URL : https://patchwork.freedesktop.org/series/98137/
State : success
== Summary ==
CI Bug Log - changes from CI_DRM_11009 -> Patchwork_21861
====================================================
Summary
-------
**SUCCESS**
No regressions found.
External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/index.html
Participating hosts (43 -> 34)
------------------------------
Additional (1): fi-pnv-d510
Missing (10): fi-kbl-soraka bat-dg1-6 bat-dg1-5 fi-icl-u2 fi-bsw-cyan bat-adlp-6 bat-adlp-4 fi-bdw-samus bat-jsl-2 bat-jsl-1
Known issues
------------
Here are the changes found in Patchwork_21861 that come from known issues:
### IGT changes ###
#### Issues hit ####
* igt@gem_exec_suspend@basic-s0:
- fi-tgl-1115g4: [PASS][1] -> [FAIL][2] ([i915#1888])
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/fi-tgl-1115g4/igt@gem_exec_suspend@basic-s0.html
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/fi-tgl-1115g4/igt@gem_exec_suspend@basic-s0.html
* igt@gem_exec_suspend@basic-s3:
- fi-skl-6600u: [PASS][3] -> [INCOMPLETE][4] ([i915#4547])
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/fi-skl-6600u/igt@gem_exec_suspend@basic-s3.html
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/fi-skl-6600u/igt@gem_exec_suspend@basic-s3.html
* igt@prime_vgem@basic-userptr:
- fi-pnv-d510: NOTRUN -> [SKIP][5] ([fdo#109271]) +57 similar issues
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/fi-pnv-d510/igt@prime_vgem@basic-userptr.html
#### Possible fixes ####
* igt@gem_exec_suspend@basic-s3:
- fi-tgl-1115g4: [FAIL][6] ([i915#1888]) -> [PASS][7]
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/fi-tgl-1115g4/igt@gem_exec_suspend@basic-s3.html
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/fi-tgl-1115g4/igt@gem_exec_suspend@basic-s3.html
[fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271
[i915#1888]: https://gitlab.freedesktop.org/drm/intel/issues/1888
[i915#4547]: https://gitlab.freedesktop.org/drm/intel/issues/4547
Build changes
-------------
* Linux: CI_DRM_11009 -> Patchwork_21861
CI-20190529: 20190529
CI_DRM_11009: 9efbfd937cc876674559ddf8fb1897a00c10019b @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_6313: 1793ed798cc09966c27bf478781e0c1d6bb23bad @ https://gitlab.freedesktop.org/drm/igt-gpu-tools.git
Patchwork_21861: c54bb29dd0de9657c0becea36437f5bcca1b36fd @ git://anongit.freedesktop.org/gfx-ci/linux
== Linux commits ==
c54bb29dd0de drm/i915: Remove short-term pins from execbuf, v5.
d5c98a7587ba drm/i915: Remove support for unlocked i915_vma unbind
ece7fb471984 drm/i915: Remove assert_object_held_shared
d2242714f33e drm/i915: Add i915_vma_unbind_unlocked, and take obj lock for i915_vma_unbind, v2.
8b8d9bb4ffaa drm/i915: Add object locking to i915_gem_evict_for_node and i915_gem_evict_something
44fc65c4b062 drm/i915: Add locking to i915_gem_evict_vm()
97cc19bfa2a0 drm/i915: Add ww ctx to i915_gem_object_trylock
d1406f612444 drm/i915: Require object lock when freeing pages during destruction
1631c2059017 drm/i915: Trylock the object when shrinking
cc5dd7eac42f drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors
c9892c726d3c drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking changes.
e7fbd118c9b2 drm/i915: Ensure gem_contexts selftests work with unbind changes, v2.
c293606bcd36 drm/i915: Force ww lock for i915_gem_object_ggtt_pin_ww, v2.
9d385c135ad3 drm/i915: Take object lock in i915_ggtt_pin if ww is not set
316d336c98ea drm/i915: Remove pages_mutex and intel_gtt->vma_ops.set/clear_pages members, v3.
91bcecf270f0 drm/i915: Change shrink ordering to use locking around unbinding.
bc08f58eeb3f drm/i915: Remove unused bits of i915_vma/active api
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/index.html
[-- Attachment #2: Type: text/html, Size: 4944 bytes --]
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [PATCH v3 04/17] drm/i915: Take object lock in i915_ggtt_pin if ww is not set
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-17 10:44 ` Matthew Auld
-1 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 10:44 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> i915_vma_wait_for_bind needs the vma lock held, fix the caller.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 04/17] drm/i915: Take object lock in i915_ggtt_pin if ww is not set
@ 2021-12-17 10:44 ` Matthew Auld
0 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 10:44 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> i915_vma_wait_for_bind needs the vma lock held, fix the caller.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 05/17] drm/i915: Force ww lock for i915_gem_object_ggtt_pin_ww, v2.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
(?)
@ 2021-12-17 10:46 ` Matthew Auld
-1 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 10:46 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> We will need the lock to unbind the vma, and wait for bind to complete.
> Remove the special casing for the !ww path, and force ww locking for all.
>
> Changes since v1:
> - Pass err to for_i915_gem_ww handling for -EDEADLK handling.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Deja-vu,
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* [Intel-gfx] ✗ Fi.CI.IGT: failure for drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
` (21 preceding siblings ...)
(?)
@ 2021-12-17 10:48 ` Patchwork
-1 siblings, 0 replies; 57+ messages in thread
From: Patchwork @ 2021-12-17 10:48 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: intel-gfx
[-- Attachment #1: Type: text/plain, Size: 30297 bytes --]
== Series Details ==
Series: drm/i915: Remove short term pins from execbuf by requiring lock to unbind.
URL : https://patchwork.freedesktop.org/series/98137/
State : failure
== Summary ==
CI Bug Log - changes from CI_DRM_11009_full -> Patchwork_21861_full
====================================================
Summary
-------
**FAILURE**
Serious unknown changes coming with Patchwork_21861_full absolutely need to be
verified manually.
If you think the reported changes have nothing to do with the changes
introduced in Patchwork_21861_full, please notify your bug team to allow them
to document this new failure mode, which will reduce false positives in CI.
Participating hosts (10 -> 10)
------------------------------
No changes in participating hosts
Possible new issues
-------------------
Here are the unknown changes that may have been introduced in Patchwork_21861_full:
### IGT changes ###
#### Possible regressions ####
* igt@gem_exec_suspend@basic-s3:
- shard-skl: [PASS][1] -> [INCOMPLETE][2]
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-skl9/igt@gem_exec_suspend@basic-s3.html
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl6/igt@gem_exec_suspend@basic-s3.html
* igt@gem_mmap_gtt@cpuset-big-copy-xy:
- shard-iclb: [PASS][3] -> [INCOMPLETE][4]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-iclb1/igt@gem_mmap_gtt@cpuset-big-copy-xy.html
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb3/igt@gem_mmap_gtt@cpuset-big-copy-xy.html
- shard-glk: [PASS][5] -> [INCOMPLETE][6]
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk5/igt@gem_mmap_gtt@cpuset-big-copy-xy.html
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk8/igt@gem_mmap_gtt@cpuset-big-copy-xy.html
- shard-tglb: [PASS][7] -> [INCOMPLETE][8]
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-tglb7/igt@gem_mmap_gtt@cpuset-big-copy-xy.html
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-tglb3/igt@gem_mmap_gtt@cpuset-big-copy-xy.html
* igt@gem_ppgtt@blt-vs-render-ctx0:
- shard-tglb: [PASS][9] -> [FAIL][10]
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-tglb5/igt@gem_ppgtt@blt-vs-render-ctx0.html
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-tglb3/igt@gem_ppgtt@blt-vs-render-ctx0.html
* igt@i915_pm_dc@dc6-psr:
- shard-skl: NOTRUN -> [CRASH][11]
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl3/igt@i915_pm_dc@dc6-psr.html
Known issues
------------
Here are the changes found in Patchwork_21861_full that come from known issues:
### CI changes ###
#### Possible fixes ####
* boot:
- shard-glk: ([PASS][12], [PASS][13], [PASS][14], [FAIL][15], [PASS][16], [PASS][17], [PASS][18], [PASS][19], [PASS][20], [PASS][21], [PASS][22], [PASS][23], [PASS][24], [PASS][25], [PASS][26], [PASS][27], [PASS][28], [PASS][29], [PASS][30], [PASS][31], [PASS][32], [PASS][33], [PASS][34], [PASS][35], [PASS][36]) ([i915#4392]) -> ([PASS][37], [PASS][38], [PASS][39], [PASS][40], [PASS][41], [PASS][42], [PASS][43], [PASS][44], [PASS][45], [PASS][46], [PASS][47], [PASS][48], [PASS][49], [PASS][50], [PASS][51], [PASS][52], [PASS][53], [PASS][54], [PASS][55], [PASS][56], [PASS][57], [PASS][58], [PASS][59], [PASS][60], [PASS][61])
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk2/boot.html
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk9/boot.html
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk9/boot.html
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk8/boot.html
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk8/boot.html
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk8/boot.html
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk8/boot.html
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk1/boot.html
[20]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk7/boot.html
[21]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk7/boot.html
[22]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk6/boot.html
[23]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk6/boot.html
[24]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk5/boot.html
[25]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk5/boot.html
[26]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk5/boot.html
[27]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk4/boot.html
[28]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk1/boot.html
[29]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk4/boot.html
[30]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk4/boot.html
[31]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk3/boot.html
[32]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk1/boot.html
[33]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk3/boot.html
[34]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk3/boot.html
[35]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk2/boot.html
[36]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk2/boot.html
[37]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk9/boot.html
[38]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk9/boot.html
[39]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk8/boot.html
[40]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk8/boot.html
[41]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk8/boot.html
[42]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk7/boot.html
[43]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk7/boot.html
[44]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk7/boot.html
[45]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/boot.html
[46]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/boot.html
[47]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk5/boot.html
[48]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk5/boot.html
[49]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk5/boot.html
[50]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk5/boot.html
[51]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk4/boot.html
[52]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk4/boot.html
[53]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk3/boot.html
[54]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk3/boot.html
[55]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk3/boot.html
[56]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk2/boot.html
[57]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk2/boot.html
[58]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk2/boot.html
[59]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk1/boot.html
[60]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk1/boot.html
[61]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk1/boot.html
### IGT changes ###
#### Issues hit ####
* igt@feature_discovery@display-4x:
- shard-iclb: NOTRUN -> [SKIP][62] ([i915#1839])
[62]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@feature_discovery@display-4x.html
* igt@feature_discovery@psr2:
- shard-iclb: NOTRUN -> [SKIP][63] ([i915#658])
[63]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb6/igt@feature_discovery@psr2.html
* igt@gem_ctx_isolation@preservation-s3@vcs1:
- shard-tglb: [PASS][64] -> [INCOMPLETE][65] ([i915#456])
[64]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-tglb5/igt@gem_ctx_isolation@preservation-s3@vcs1.html
[65]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-tglb8/igt@gem_ctx_isolation@preservation-s3@vcs1.html
* igt@gem_ctx_persistence@many-contexts:
- shard-tglb: [PASS][66] -> [FAIL][67] ([i915#2410])
[66]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-tglb2/igt@gem_ctx_persistence@many-contexts.html
[67]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-tglb7/igt@gem_ctx_persistence@many-contexts.html
* igt@gem_eio@unwedge-stress:
- shard-skl: NOTRUN -> [TIMEOUT][68] ([i915#3063])
[68]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl7/igt@gem_eio@unwedge-stress.html
* igt@gem_exec_capture@pi@bcs0:
- shard-skl: [PASS][69] -> [INCOMPLETE][70] ([i915#4547])
[69]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-skl5/igt@gem_exec_capture@pi@bcs0.html
[70]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl8/igt@gem_exec_capture@pi@bcs0.html
* igt@gem_exec_capture@pi@rcs0:
- shard-tglb: [PASS][71] -> [INCOMPLETE][72] ([i915#3371])
[71]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-tglb6/igt@gem_exec_capture@pi@rcs0.html
[72]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-tglb1/igt@gem_exec_capture@pi@rcs0.html
* igt@gem_exec_fair@basic-deadline:
- shard-glk: [PASS][73] -> [FAIL][74] ([i915#2846])
[73]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk3/igt@gem_exec_fair@basic-deadline.html
[74]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk3/igt@gem_exec_fair@basic-deadline.html
* igt@gem_exec_fair@basic-pace-share@rcs0:
- shard-apl: [PASS][75] -> [FAIL][76] ([i915#2842])
[75]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-apl2/igt@gem_exec_fair@basic-pace-share@rcs0.html
[76]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl7/igt@gem_exec_fair@basic-pace-share@rcs0.html
* igt@gem_exec_fair@basic-pace@vcs1:
- shard-iclb: NOTRUN -> [FAIL][77] ([i915#2842]) +1 similar issue
[77]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb2/igt@gem_exec_fair@basic-pace@vcs1.html
* igt@gem_exec_fair@basic-pace@vecs0:
- shard-kbl: [PASS][78] -> [FAIL][79] ([i915#2842]) +2 similar issues
[78]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-kbl3/igt@gem_exec_fair@basic-pace@vecs0.html
[79]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl2/igt@gem_exec_fair@basic-pace@vecs0.html
* igt@gem_exec_fair@basic-throttle@rcs0:
- shard-iclb: [PASS][80] -> [FAIL][81] ([i915#2849])
[80]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-iclb3/igt@gem_exec_fair@basic-throttle@rcs0.html
[81]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb7/igt@gem_exec_fair@basic-throttle@rcs0.html
* igt@gem_exec_flush@basic-batch-kernel-default-uc:
- shard-glk: [PASS][82] -> [DMESG-WARN][83] ([i915#118]) +1 similar issue
[82]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk6/igt@gem_exec_flush@basic-batch-kernel-default-uc.html
[83]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk9/igt@gem_exec_flush@basic-batch-kernel-default-uc.html
* igt@gem_lmem_swapping@parallel-random:
- shard-apl: NOTRUN -> [SKIP][84] ([fdo#109271] / [i915#4613])
[84]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl7/igt@gem_lmem_swapping@parallel-random.html
- shard-kbl: NOTRUN -> [SKIP][85] ([fdo#109271] / [i915#4613]) +1 similar issue
[85]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl2/igt@gem_lmem_swapping@parallel-random.html
* igt@gem_lmem_swapping@verify-random:
- shard-skl: NOTRUN -> [SKIP][86] ([fdo#109271] / [i915#4613]) +4 similar issues
[86]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl1/igt@gem_lmem_swapping@verify-random.html
* igt@gem_ppgtt@flink-and-close-vma-leak:
- shard-apl: [PASS][87] -> [FAIL][88] ([i915#644])
[87]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-apl4/igt@gem_ppgtt@flink-and-close-vma-leak.html
[88]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl2/igt@gem_ppgtt@flink-and-close-vma-leak.html
* igt@gem_pread@exhaustion:
- shard-skl: NOTRUN -> [WARN][89] ([i915#2658])
[89]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl1/igt@gem_pread@exhaustion.html
* igt@gem_render_copy@yf-tiled-to-vebox-linear:
- shard-iclb: NOTRUN -> [SKIP][90] ([i915#768]) +1 similar issue
[90]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@gem_render_copy@yf-tiled-to-vebox-linear.html
* igt@gen7_exec_parse@basic-allocation:
- shard-glk: NOTRUN -> [SKIP][91] ([fdo#109271]) +34 similar issues
[91]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/igt@gen7_exec_parse@basic-allocation.html
* igt@i915_pm_dc@dc6-dpms:
- shard-skl: NOTRUN -> [FAIL][92] ([i915#454])
[92]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl7/igt@i915_pm_dc@dc6-dpms.html
* igt@i915_pm_dc@dc9-dpms:
- shard-iclb: [PASS][93] -> [SKIP][94] ([i915#4281])
[93]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-iclb1/igt@i915_pm_dc@dc9-dpms.html
[94]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb3/igt@i915_pm_dc@dc9-dpms.html
* igt@i915_pm_rpm@gem-execbuf-stress-pc8:
- shard-iclb: NOTRUN -> [SKIP][95] ([fdo#109293] / [fdo#109506])
[95]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@i915_pm_rpm@gem-execbuf-stress-pc8.html
* igt@i915_selftest@live@gt_pm:
- shard-skl: NOTRUN -> [DMESG-FAIL][96] ([i915#1886] / [i915#2291])
[96]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl3/igt@i915_selftest@live@gt_pm.html
* igt@kms_big_fb@x-tiled-max-hw-stride-32bpp-rotate-180-async-flip:
- shard-skl: NOTRUN -> [FAIL][97] ([i915#3743]) +1 similar issue
[97]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl5/igt@kms_big_fb@x-tiled-max-hw-stride-32bpp-rotate-180-async-flip.html
* igt@kms_big_fb@x-tiled-max-hw-stride-64bpp-rotate-180-hflip:
- shard-glk: NOTRUN -> [SKIP][98] ([fdo#109271] / [i915#3777])
[98]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/igt@kms_big_fb@x-tiled-max-hw-stride-64bpp-rotate-180-hflip.html
* igt@kms_big_fb@y-tiled-64bpp-rotate-90:
- shard-iclb: NOTRUN -> [SKIP][99] ([fdo#110725] / [fdo#111614]) +2 similar issues
[99]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_big_fb@y-tiled-64bpp-rotate-90.html
* igt@kms_big_fb@y-tiled-max-hw-stride-64bpp-rotate-0-async-flip:
- shard-tglb: [PASS][100] -> [FAIL][101] ([i915#3743])
[100]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-tglb3/igt@kms_big_fb@y-tiled-max-hw-stride-64bpp-rotate-0-async-flip.html
[101]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-tglb1/igt@kms_big_fb@y-tiled-max-hw-stride-64bpp-rotate-0-async-flip.html
- shard-skl: NOTRUN -> [FAIL][102] ([i915#3763])
[102]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl7/igt@kms_big_fb@y-tiled-max-hw-stride-64bpp-rotate-0-async-flip.html
* igt@kms_big_fb@y-tiled-max-hw-stride-64bpp-rotate-0-hflip:
- shard-skl: NOTRUN -> [SKIP][103] ([fdo#109271] / [i915#3777])
[103]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl10/igt@kms_big_fb@y-tiled-max-hw-stride-64bpp-rotate-0-hflip.html
* igt@kms_ccs@pipe-a-crc-sprite-planes-basic-y_tiled_gen12_mc_ccs:
- shard-apl: NOTRUN -> [SKIP][104] ([fdo#109271] / [i915#3886]) +1 similar issue
[104]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl7/igt@kms_ccs@pipe-a-crc-sprite-planes-basic-y_tiled_gen12_mc_ccs.html
* igt@kms_ccs@pipe-b-bad-rotation-90-y_tiled_gen12_mc_ccs:
- shard-kbl: NOTRUN -> [SKIP][105] ([fdo#109271] / [i915#3886]) +3 similar issues
[105]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl2/igt@kms_ccs@pipe-b-bad-rotation-90-y_tiled_gen12_mc_ccs.html
* igt@kms_ccs@pipe-b-crc-primary-basic-y_tiled_gen12_rc_ccs_cc:
- shard-iclb: NOTRUN -> [SKIP][106] ([fdo#109278] / [i915#3886])
[106]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_ccs@pipe-b-crc-primary-basic-y_tiled_gen12_rc_ccs_cc.html
* igt@kms_ccs@pipe-b-crc-primary-rotation-180-y_tiled_gen12_rc_ccs_cc:
- shard-skl: NOTRUN -> [SKIP][107] ([fdo#109271] / [i915#3886]) +12 similar issues
[107]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl9/igt@kms_ccs@pipe-b-crc-primary-rotation-180-y_tiled_gen12_rc_ccs_cc.html
* igt@kms_ccs@pipe-c-crc-sprite-planes-basic-y_tiled_gen12_rc_ccs_cc:
- shard-glk: NOTRUN -> [SKIP][108] ([fdo#109271] / [i915#3886]) +1 similar issue
[108]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/igt@kms_ccs@pipe-c-crc-sprite-planes-basic-y_tiled_gen12_rc_ccs_cc.html
* igt@kms_ccs@pipe-d-crc-sprite-planes-basic-y_tiled_gen12_rc_ccs_cc:
- shard-skl: NOTRUN -> [SKIP][109] ([fdo#109271]) +282 similar issues
[109]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl9/igt@kms_ccs@pipe-d-crc-sprite-planes-basic-y_tiled_gen12_rc_ccs_cc.html
* igt@kms_cdclk@mode-transition:
- shard-iclb: NOTRUN -> [SKIP][110] ([i915#3742])
[110]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_cdclk@mode-transition.html
* igt@kms_chamelium@dp-hpd:
- shard-iclb: NOTRUN -> [SKIP][111] ([fdo#109284] / [fdo#111827])
[111]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_chamelium@dp-hpd.html
* igt@kms_chamelium@hdmi-crc-single:
- shard-glk: NOTRUN -> [SKIP][112] ([fdo#109271] / [fdo#111827]) +3 similar issues
[112]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/igt@kms_chamelium@hdmi-crc-single.html
* igt@kms_chamelium@vga-hpd-after-suspend:
- shard-skl: NOTRUN -> [SKIP][113] ([fdo#109271] / [fdo#111827]) +21 similar issues
[113]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl7/igt@kms_chamelium@vga-hpd-after-suspend.html
* igt@kms_color@pipe-b-ctm-0-25:
- shard-skl: [PASS][114] -> [DMESG-WARN][115] ([i915#1982])
[114]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-skl7/igt@kms_color@pipe-b-ctm-0-25.html
[115]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl8/igt@kms_color@pipe-b-ctm-0-25.html
* igt@kms_color@pipe-d-ctm-0-25:
- shard-iclb: NOTRUN -> [SKIP][116] ([fdo#109278] / [i915#1149])
[116]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_color@pipe-d-ctm-0-25.html
* igt@kms_color_chamelium@pipe-a-ctm-0-5:
- shard-apl: NOTRUN -> [SKIP][117] ([fdo#109271] / [fdo#111827]) +3 similar issues
[117]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl6/igt@kms_color_chamelium@pipe-a-ctm-0-5.html
* igt@kms_color_chamelium@pipe-a-ctm-blue-to-red:
- shard-kbl: NOTRUN -> [SKIP][118] ([fdo#109271] / [fdo#111827]) +9 similar issues
[118]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl4/igt@kms_color_chamelium@pipe-a-ctm-blue-to-red.html
* igt@kms_cursor_crc@pipe-a-cursor-512x512-random:
- shard-iclb: NOTRUN -> [SKIP][119] ([fdo#109278] / [fdo#109279])
[119]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_cursor_crc@pipe-a-cursor-512x512-random.html
* igt@kms_cursor_crc@pipe-d-cursor-128x128-onscreen:
- shard-apl: NOTRUN -> [SKIP][120] ([fdo#109271]) +47 similar issues
[120]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl7/igt@kms_cursor_crc@pipe-d-cursor-128x128-onscreen.html
* igt@kms_cursor_crc@pipe-d-cursor-suspend:
- shard-tglb: [PASS][121] -> [INCOMPLETE][122] ([i915#2411] / [i915#4211])
[121]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-tglb5/igt@kms_cursor_crc@pipe-d-cursor-suspend.html
[122]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-tglb7/igt@kms_cursor_crc@pipe-d-cursor-suspend.html
* igt@kms_cursor_legacy@2x-cursor-vs-flip-atomic:
- shard-iclb: NOTRUN -> [SKIP][123] ([fdo#109274] / [fdo#109278]) +1 similar issue
[123]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_cursor_legacy@2x-cursor-vs-flip-atomic.html
* igt@kms_cursor_legacy@flip-vs-cursor-varying-size:
- shard-iclb: [PASS][124] -> [FAIL][125] ([i915#2346])
[124]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-iclb3/igt@kms_cursor_legacy@flip-vs-cursor-varying-size.html
[125]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb7/igt@kms_cursor_legacy@flip-vs-cursor-varying-size.html
* igt@kms_flip@2x-absolute-wf_vblank-interruptible:
- shard-iclb: NOTRUN -> [SKIP][126] ([fdo#109274]) +2 similar issues
[126]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb6/igt@kms_flip@2x-absolute-wf_vblank-interruptible.html
* igt@kms_flip@plain-flip-fb-recreate@c-edp1:
- shard-skl: [PASS][127] -> [FAIL][128] ([i915#2122]) +1 similar issue
[127]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-skl5/igt@kms_flip@plain-flip-fb-recreate@c-edp1.html
[128]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl3/igt@kms_flip@plain-flip-fb-recreate@c-edp1.html
* igt@kms_flip_scaled_crc@flip-32bpp-ytile-to-32bpp-ytileccs:
- shard-skl: NOTRUN -> [INCOMPLETE][129] ([i915#3699])
[129]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl10/igt@kms_flip_scaled_crc@flip-32bpp-ytile-to-32bpp-ytileccs.html
* igt@kms_flip_scaled_crc@flip-64bpp-ytile-to-32bpp-ytilercccs:
- shard-skl: NOTRUN -> [SKIP][130] ([fdo#109271] / [i915#2672])
[130]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl7/igt@kms_flip_scaled_crc@flip-64bpp-ytile-to-32bpp-ytilercccs.html
- shard-kbl: NOTRUN -> [SKIP][131] ([fdo#109271] / [i915#2672])
[131]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl2/igt@kms_flip_scaled_crc@flip-64bpp-ytile-to-32bpp-ytilercccs.html
- shard-apl: NOTRUN -> [SKIP][132] ([fdo#109271] / [i915#2672])
[132]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl7/igt@kms_flip_scaled_crc@flip-64bpp-ytile-to-32bpp-ytilercccs.html
* igt@kms_frontbuffer_tracking@fbc-2p-primscrn-spr-indfb-draw-blt:
- shard-kbl: NOTRUN -> [SKIP][133] ([fdo#109271]) +119 similar issues
[133]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl4/igt@kms_frontbuffer_tracking@fbc-2p-primscrn-spr-indfb-draw-blt.html
* igt@kms_frontbuffer_tracking@fbc-suspend:
- shard-kbl: [PASS][134] -> [DMESG-WARN][135] ([i915#180]) +5 similar issues
[134]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-kbl3/igt@kms_frontbuffer_tracking@fbc-suspend.html
[135]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl7/igt@kms_frontbuffer_tracking@fbc-suspend.html
- shard-apl: [PASS][136] -> [DMESG-WARN][137] ([i915#180]) +3 similar issues
[136]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-apl4/igt@kms_frontbuffer_tracking@fbc-suspend.html
[137]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl2/igt@kms_frontbuffer_tracking@fbc-suspend.html
* igt@kms_frontbuffer_tracking@psr-2p-primscrn-pri-indfb-draw-mmap-cpu:
- shard-iclb: NOTRUN -> [SKIP][138] ([fdo#109280]) +6 similar issues
[138]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_frontbuffer_tracking@psr-2p-primscrn-pri-indfb-draw-mmap-cpu.html
* igt@kms_pipe_crc_basic@compare-crc-sanitycheck-pipe-d:
- shard-glk: NOTRUN -> [SKIP][139] ([fdo#109271] / [i915#533])
[139]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/igt@kms_pipe_crc_basic@compare-crc-sanitycheck-pipe-d.html
* igt@kms_pipe_crc_basic@read-crc-pipe-d:
- shard-skl: NOTRUN -> [SKIP][140] ([fdo#109271] / [i915#533]) +1 similar issue
[140]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl7/igt@kms_pipe_crc_basic@read-crc-pipe-d.html
* igt@kms_pipe_crc_basic@read-crc-pipe-d-frame-sequence:
- shard-iclb: NOTRUN -> [SKIP][141] ([fdo#109278]) +9 similar issues
[141]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_pipe_crc_basic@read-crc-pipe-d-frame-sequence.html
* igt@kms_plane_alpha_blend@pipe-a-alpha-7efc:
- shard-skl: NOTRUN -> [FAIL][142] ([fdo#108145] / [i915#265]) +2 similar issues
[142]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl10/igt@kms_plane_alpha_blend@pipe-a-alpha-7efc.html
* igt@kms_plane_alpha_blend@pipe-a-alpha-transparent-fb:
- shard-apl: NOTRUN -> [FAIL][143] ([i915#265])
[143]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl6/igt@kms_plane_alpha_blend@pipe-a-alpha-transparent-fb.html
* igt@kms_plane_alpha_blend@pipe-a-coverage-7efc:
- shard-skl: [PASS][144] -> [FAIL][145] ([fdo#108145] / [i915#265])
[144]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-skl8/igt@kms_plane_alpha_blend@pipe-a-coverage-7efc.html
[145]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl10/igt@kms_plane_alpha_blend@pipe-a-coverage-7efc.html
* igt@kms_plane_alpha_blend@pipe-b-alpha-opaque-fb:
- shard-glk: NOTRUN -> [FAIL][146] ([fdo#108145] / [i915#265]) +1 similar issue
[146]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/igt@kms_plane_alpha_blend@pipe-b-alpha-opaque-fb.html
* igt@kms_plane_lowres@pipe-c-tiling-x:
- shard-iclb: NOTRUN -> [SKIP][147] ([i915#3536])
[147]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb6/igt@kms_plane_lowres@pipe-c-tiling-x.html
* igt@kms_psr2_sf@overlay-primary-update-sf-dmg-area:
- shard-kbl: NOTRUN -> [SKIP][148] ([fdo#109271] / [i915#658])
[148]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl4/igt@kms_psr2_sf@overlay-primary-update-sf-dmg-area.html
* igt@kms_psr@psr2_primary_mmap_gtt:
- shard-iclb: NOTRUN -> [SKIP][149] ([fdo#109441])
[149]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb8/igt@kms_psr@psr2_primary_mmap_gtt.html
* igt@kms_psr@psr2_sprite_plane_move:
- shard-iclb: [PASS][150] -> [SKIP][151] ([fdo#109441]) +2 similar issues
[150]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-iclb2/igt@kms_psr@psr2_sprite_plane_move.html
[151]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb4/igt@kms_psr@psr2_sprite_plane_move.html
* igt@kms_sysfs_edid_timing:
- shard-apl: NOTRUN -> [FAIL][152] ([IGT#2])
[152]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-apl7/igt@kms_sysfs_edid_timing.html
- shard-kbl: NOTRUN -> [FAIL][153] ([IGT#2])
[153]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl2/igt@kms_sysfs_edid_timing.html
- shard-skl: NOTRUN -> [FAIL][154] ([IGT#2])
[154]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl7/igt@kms_sysfs_edid_timing.html
* igt@kms_writeback@writeback-fb-id:
- shard-kbl: NOTRUN -> [SKIP][155] ([fdo#109271] / [i915#2437])
[155]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl4/igt@kms_writeback@writeback-fb-id.html
- shard-skl: NOTRUN -> [SKIP][156] ([fdo#109271] / [i915#2437])
[156]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl9/igt@kms_writeback@writeback-fb-id.html
* igt@perf@polling-parameterized:
- shard-glk: [PASS][157] -> [FAIL][158] ([i915#1542])
[157]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-glk3/igt@perf@polling-parameterized.html
[158]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk4/igt@perf@polling-parameterized.html
* igt@sysfs_clients@fair-7:
- shard-iclb: NOTRUN -> [SKIP][159] ([i915#2994])
[159]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-iclb6/igt@sysfs_clients@fair-7.html
* igt@sysfs_clients@recycle-many:
- shard-glk: NOTRUN -> [SKIP][160] ([fdo#109271] / [i915#2994])
[160]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-glk6/igt@sysfs_clients@recycle-many.html
- shard-skl: NOTRUN -> [SKIP][161] ([fdo#109271] / [i915#2994]) +2 similar issues
[161]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-skl3/igt@sysfs_clients@recycle-many.html
#### Possible fixes ####
* igt@gem_ctx_isolation@preservation-s3@vecs0:
- shard-kbl: [DMESG-WARN][162] ([i915#180]) -> [PASS][163] +4 similar issues
[162]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_11009/shard-kbl4/igt@gem_ctx_isolation@preservation-s3@vecs0.html
[163]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/shard-kbl7/igt@gem_ctx_isolation@preserv
== Logs ==
For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21861/index.html
[-- Attachment #2: Type: text/html, Size: 33634 bytes --]
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [PATCH v3 07/17] drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking changes.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-17 11:25 ` Matthew Auld
-1 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 11:25 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> Now that we require locking to evict, multiple vmas from the same object
> might not be evicted. This is expected and required, because execbuf will
> move to short-term pinning by using the lock only. This will cause these
> tests to fail, because they create a ton of vma's for the same object.
>
> Unbind manually to prevent spurious -ENOSPC in those mock tests.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 07/17] drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking changes.
@ 2021-12-17 11:25 ` Matthew Auld
0 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 11:25 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> Now that we require locking to evict, multiple vmas from the same object
> might not be evicted. This is expected and required, because execbuf will
> move to short-term pinning by using the lock only. This will cause these
> tests to fail, because they create a ton of vma's for the same object.
>
> Unbind manually to prevent spurious -ENOSPC in those mock tests.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [PATCH v3 08/17] drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-17 11:58 ` Matthew Auld
-1 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 11:58 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> Now that we cannot unbind kill the currently locked object directly
"unbind kill"
> because we're removing short term pinning, we may have to unbind the
> object from gtt manually, using a i915_gem_evict_vm() call.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Maybe mention that this only in preparation for some future patches,
once the actual eviction is trylock and evict_for_vm can also handle
shared dma-resv? At this point in the series we shouldn't expect to
hit -ENOSPC, right?
> ---
> drivers/gpu/drm/i915/gem/i915_gem_mman.c | 18 ++++++++++++++++--
> 1 file changed, 16 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> index af81d6c3332a..00cd9642669a 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> @@ -358,8 +358,22 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
> vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
> }
>
> - /* The entire mappable GGTT is pinned? Unexpected! */
> - GEM_BUG_ON(vma == ERR_PTR(-ENOSPC));
> + /*
> + * The entire mappable GGTT is pinned? Unexpected!
> + * Try to evict the object we locked too, as normally we skip it
> + * due to lack of short term pinning inside execbuf.
> + */
> + if (vma == ERR_PTR(-ENOSPC)) {
> + ret = mutex_lock_interruptible(&ggtt->vm.mutex);
> + if (!ret) {
> + ret = i915_gem_evict_vm(&ggtt->vm);
> + mutex_unlock(&ggtt->vm.mutex);
> + }
> + if (ret)
> + goto err_reset;
> + vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
> + }
> + GEM_WARN_ON(vma == ERR_PTR(-ENOSPC));
Looks like this is being triggered in CI, I assume because the trylock
could easily fail, due to contention? Is this expected for now? Do we
keep the WARN and track it as a known issue?
> }
> if (IS_ERR(vma)) {
> ret = PTR_ERR(vma);
> --
> 2.34.1
>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 08/17] drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors
@ 2021-12-17 11:58 ` Matthew Auld
0 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 11:58 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> Now that we cannot unbind kill the currently locked object directly
"unbind kill"
> because we're removing short term pinning, we may have to unbind the
> object from gtt manually, using a i915_gem_evict_vm() call.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Maybe mention that this only in preparation for some future patches,
once the actual eviction is trylock and evict_for_vm can also handle
shared dma-resv? At this point in the series we shouldn't expect to
hit -ENOSPC, right?
> ---
> drivers/gpu/drm/i915/gem/i915_gem_mman.c | 18 ++++++++++++++++--
> 1 file changed, 16 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> index af81d6c3332a..00cd9642669a 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> @@ -358,8 +358,22 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
> vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
> }
>
> - /* The entire mappable GGTT is pinned? Unexpected! */
> - GEM_BUG_ON(vma == ERR_PTR(-ENOSPC));
> + /*
> + * The entire mappable GGTT is pinned? Unexpected!
> + * Try to evict the object we locked too, as normally we skip it
> + * due to lack of short term pinning inside execbuf.
> + */
> + if (vma == ERR_PTR(-ENOSPC)) {
> + ret = mutex_lock_interruptible(&ggtt->vm.mutex);
> + if (!ret) {
> + ret = i915_gem_evict_vm(&ggtt->vm);
> + mutex_unlock(&ggtt->vm.mutex);
> + }
> + if (ret)
> + goto err_reset;
> + vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
> + }
> + GEM_WARN_ON(vma == ERR_PTR(-ENOSPC));
Looks like this is being triggered in CI, I assume because the trylock
could easily fail, due to contention? Is this expected for now? Do we
keep the WARN and track it as a known issue?
> }
> if (IS_ERR(vma)) {
> ret = PTR_ERR(vma);
> --
> 2.34.1
>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [PATCH v3 09/17] drm/i915: Trylock the object when shrinking
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
@ 2021-12-17 12:53 ` Matthew Auld
-1 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 12:53 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> We're working on requiring the obj->resv lock during unbind, fix
> the shrinker to take the objectl ock.
lock
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 09/17] drm/i915: Trylock the object when shrinking
@ 2021-12-17 12:53 ` Matthew Auld
0 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 12:53 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> We're working on requiring the obj->resv lock during unbind, fix
> the shrinker to take the objectl ock.
lock
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 11/17] drm/i915: Add ww ctx to i915_gem_object_trylock
2021-12-16 14:27 ` Maarten Lankhorst
(?)
@ 2021-12-17 12:55 ` Matthew Auld
-1 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 12:55 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> This is required for i915_gem_evict_vm, to be able to evict the entire VM,
> including objects that are already locked to the current ww ctx.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 12/17] drm/i915: Add locking to i915_gem_evict_vm()
2021-12-16 14:27 ` Maarten Lankhorst
(?)
@ 2021-12-17 13:19 ` Matthew Auld
-1 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 13:19 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> i915_gem_evict_vm will need to be able to evict objects that are
> locked by the current ctx. By testing if the current context already
> locked the object, we can do this correctly. This allows us to
> evict the entire vm even if we already hold some objects' locks.
>
> Previously, this was spread over several commits, but it makes
> more sense to commit the changes to i915_gem_evict_vm separately
> from the changes to i915_gem_evict_something() and
> i915_gem_evict_for_node().
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> ---
> .../gpu/drm/i915/gem/i915_gem_execbuffer.c | 2 +-
> drivers/gpu/drm/i915/gem/i915_gem_mman.c | 2 +-
> drivers/gpu/drm/i915/i915_drv.h | 3 +-
> drivers/gpu/drm/i915/i915_gem_evict.c | 30 +++++++++++++++++--
> drivers/gpu/drm/i915/i915_vma.c | 7 ++++-
> .../gpu/drm/i915/selftests/i915_gem_evict.c | 10 +++++--
> 6 files changed, 46 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> index 2213f7b613da..eb3649e844ff 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
> @@ -766,7 +766,7 @@ static int eb_reserve(struct i915_execbuffer *eb)
> case 1:
> /* Too fragmented, unbind everything and retry */
> mutex_lock(&eb->context->vm->mutex);
> - err = i915_gem_evict_vm(eb->context->vm);
> + err = i915_gem_evict_vm(eb->context->vm, &eb->ww);
> mutex_unlock(&eb->context->vm->mutex);
> if (err)
> return err;
> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> index 00cd9642669a..2856098cb449 100644
> --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
> @@ -366,7 +366,7 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
> if (vma == ERR_PTR(-ENOSPC)) {
> ret = mutex_lock_interruptible(&ggtt->vm.mutex);
> if (!ret) {
> - ret = i915_gem_evict_vm(&ggtt->vm);
> + ret = i915_gem_evict_vm(&ggtt->vm, &ww);
> mutex_unlock(&ggtt->vm.mutex);
> }
> if (ret)
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index d51628d10f9d..c180019c607f 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1725,7 +1725,8 @@ int __must_check i915_gem_evict_something(struct i915_address_space *vm,
> int __must_check i915_gem_evict_for_node(struct i915_address_space *vm,
> struct drm_mm_node *node,
> unsigned int flags);
> -int i915_gem_evict_vm(struct i915_address_space *vm);
> +int i915_gem_evict_vm(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww);
>
> /* i915_gem_internal.c */
> struct drm_i915_gem_object *
> diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
> index 2b73ddb11c66..bfd66f539fc1 100644
> --- a/drivers/gpu/drm/i915/i915_gem_evict.c
> +++ b/drivers/gpu/drm/i915/i915_gem_evict.c
> @@ -367,7 +367,7 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
> * To clarify: This is for freeing up virtual address space, not for freeing
> * memory in e.g. the shrinker.
> */
> -int i915_gem_evict_vm(struct i915_address_space *vm)
> +int i915_gem_evict_vm(struct i915_address_space *vm, struct i915_gem_ww_ctx *ww)
> {
> int ret = 0;
>
> @@ -388,24 +388,50 @@ int i915_gem_evict_vm(struct i915_address_space *vm)
> do {
> struct i915_vma *vma, *vn;
> LIST_HEAD(eviction_list);
> + LIST_HEAD(locked_eviction_list);
>
> list_for_each_entry(vma, &vm->bound_list, vm_link) {
> if (i915_vma_is_pinned(vma))
> continue;
>
> + /*
> + * If we already own the lock, trylock fails. In case the resv
> + * is shared among multiple objects, we still need the object ref.
What is "object ref" here? I assume it's just leftovers...
Otherwise,
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> + */
> + if (ww && (dma_resv_locking_ctx(vma->obj->base.resv) == &ww->ctx)) {
> + __i915_vma_pin(vma);
> + list_add(&vma->evict_link, &locked_eviction_list);
> + continue;
> + }
> +
> + if (!i915_gem_object_trylock(vma->obj, ww))
> + continue;
> +
> __i915_vma_pin(vma);
> list_add(&vma->evict_link, &eviction_list);
> }
> - if (list_empty(&eviction_list))
> + if (list_empty(&eviction_list) && list_empty(&locked_eviction_list))
> break;
>
> ret = 0;
> + /* Unbind locked objects first, before unlocking the eviction_list */
> + list_for_each_entry_safe(vma, vn, &locked_eviction_list, evict_link) {
> + __i915_vma_unpin(vma);
> +
> + if (ret == 0)
> + ret = __i915_vma_unbind(vma);
> + if (ret != -EINTR) /* "Get me out of here!" */
> + ret = 0;
> + }
> +
> list_for_each_entry_safe(vma, vn, &eviction_list, evict_link) {
> __i915_vma_unpin(vma);
> if (ret == 0)
> ret = __i915_vma_unbind(vma);
> if (ret != -EINTR) /* "Get me out of here!" */
> ret = 0;
> +
> + i915_gem_object_unlock(vma->obj);
> }
> } while (ret == 0);
>
> diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
> index de24e4b3b19b..d24e90eac948 100644
> --- a/drivers/gpu/drm/i915/i915_vma.c
> +++ b/drivers/gpu/drm/i915/i915_vma.c
> @@ -1462,7 +1462,12 @@ static int __i915_ggtt_pin(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
> /* Unlike i915_vma_pin, we don't take no for an answer! */
> flush_idle_contexts(vm->gt);
> if (mutex_lock_interruptible(&vm->mutex) == 0) {
> - i915_gem_evict_vm(vm);
> + /*
> + * We pass NULL ww here, as we don't want to unbind
> + * locked objects when called from execbuf when pinning
> + * is removed. This would probably regress badly.
> + */
> + i915_gem_evict_vm(vm, NULL);
> mutex_unlock(&vm->mutex);
> }
> } while (1);
> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> index 7e0658a77659..7178811366af 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> +++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> @@ -331,6 +331,7 @@ static int igt_evict_vm(void *arg)
> {
> struct intel_gt *gt = arg;
> struct i915_ggtt *ggtt = gt->ggtt;
> + struct i915_gem_ww_ctx ww;
> LIST_HEAD(objects);
> int err;
>
> @@ -342,7 +343,7 @@ static int igt_evict_vm(void *arg)
>
> /* Everything is pinned, nothing should happen */
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_evict_vm(&ggtt->vm);
> + err = i915_gem_evict_vm(&ggtt->vm, NULL);
> mutex_unlock(&ggtt->vm.mutex);
> if (err) {
> pr_err("i915_gem_evict_vm on a full GGTT returned err=%d]\n",
> @@ -352,9 +353,14 @@ static int igt_evict_vm(void *arg)
>
> unpin_ggtt(ggtt);
>
> + i915_gem_ww_ctx_init(&ww, false);
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_evict_vm(&ggtt->vm);
> + err = i915_gem_evict_vm(&ggtt->vm, &ww);
> mutex_unlock(&ggtt->vm.mutex);
> +
> + /* no -EDEADLK handling; can't happen with vm.mutex in place */
> + i915_gem_ww_ctx_fini(&ww);
> +
> if (err) {
> pr_err("i915_gem_evict_vm on a full GGTT returned err=%d]\n",
> err);
> --
> 2.34.1
>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 13/17] drm/i915: Add object locking to i915_gem_evict_for_node and i915_gem_evict_something
2021-12-16 14:27 ` Maarten Lankhorst
(?)
@ 2021-12-17 13:55 ` Matthew Auld
2021-12-17 15:04 ` Maarten Lankhorst
-1 siblings, 1 reply; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 13:55 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> Because we will start to require the obj->resv lock for unbinding,
> ensure these shrinker functions also take the lock.
>
> This requires some function signature changes, to ensure that the
> ww context is passed around, but is mostly straightforward.
Do we even need this? We aren't handling the shared dma-resv for these
ones, right? Or more just to have a consistent-ish interface?
>
> Previously this was split up into several patches, but reworking
> should allow for easier bisection.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> ---
> drivers/gpu/drm/i915/gt/intel_ggtt.c | 2 +-
> drivers/gpu/drm/i915/gt/selftest_hangcheck.c | 2 +-
> drivers/gpu/drm/i915/gvt/aperture_gm.c | 2 +-
> drivers/gpu/drm/i915/i915_drv.h | 2 ++
> drivers/gpu/drm/i915/i915_gem_evict.c | 34 +++++++++++++++----
> drivers/gpu/drm/i915/i915_gem_gtt.c | 8 +++--
> drivers/gpu/drm/i915/i915_gem_gtt.h | 3 ++
> drivers/gpu/drm/i915/i915_vgpu.c | 2 +-
> drivers/gpu/drm/i915/i915_vma.c | 9 ++---
> .../gpu/drm/i915/selftests/i915_gem_evict.c | 17 +++++-----
> drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 14 ++++----
> 11 files changed, 63 insertions(+), 32 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
> index a287c9186ec9..ed43e9c80aaa 100644
> --- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
> +++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
> @@ -504,7 +504,7 @@ static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt)
> GEM_BUG_ON(ggtt->vm.total <= GUC_GGTT_TOP);
> size = ggtt->vm.total - GUC_GGTT_TOP;
>
> - ret = i915_gem_gtt_reserve(&ggtt->vm, &ggtt->uc_fw, size,
> + ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, &ggtt->uc_fw, size,
> GUC_GGTT_TOP, I915_COLOR_UNEVICTABLE,
> PIN_NOEVICT);
> if (ret)
> diff --git a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
> index e5ad4d5a91c0..a3597a6bb805 100644
> --- a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
> +++ b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
> @@ -1382,7 +1382,7 @@ static int evict_vma(void *data)
> complete(&arg->completion);
>
> mutex_lock(&vm->mutex);
> - err = i915_gem_evict_for_node(vm, &evict, 0);
> + err = i915_gem_evict_for_node(vm, NULL, &evict, 0);
> mutex_unlock(&vm->mutex);
>
> return err;
> diff --git a/drivers/gpu/drm/i915/gvt/aperture_gm.c b/drivers/gpu/drm/i915/gvt/aperture_gm.c
> index 0d6d59871308..c08098a167e9 100644
> --- a/drivers/gpu/drm/i915/gvt/aperture_gm.c
> +++ b/drivers/gpu/drm/i915/gvt/aperture_gm.c
> @@ -63,7 +63,7 @@ static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm)
>
> mutex_lock(>->ggtt->vm.mutex);
> mmio_hw_access_pre(gt);
> - ret = i915_gem_gtt_insert(>->ggtt->vm, node,
> + ret = i915_gem_gtt_insert(>->ggtt->vm, NULL, node,
> size, I915_GTT_PAGE_SIZE,
> I915_COLOR_UNEVICTABLE,
> start, end, flags);
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index c180019c607f..2a98192a89c1 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1718,11 +1718,13 @@ i915_gem_vm_lookup(struct drm_i915_file_private *file_priv, u32 id)
>
> /* i915_gem_evict.c */
> int __must_check i915_gem_evict_something(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww,
> u64 min_size, u64 alignment,
> unsigned long color,
> u64 start, u64 end,
> unsigned flags);
> int __must_check i915_gem_evict_for_node(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww,
> struct drm_mm_node *node,
> unsigned int flags);
> int i915_gem_evict_vm(struct i915_address_space *vm,
> diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
> index bfd66f539fc1..f502a617b35c 100644
> --- a/drivers/gpu/drm/i915/i915_gem_evict.c
> +++ b/drivers/gpu/drm/i915/i915_gem_evict.c
> @@ -51,6 +51,7 @@ static int ggtt_flush(struct intel_gt *gt)
>
> static bool
> mark_free(struct drm_mm_scan *scan,
> + struct i915_gem_ww_ctx *ww,
> struct i915_vma *vma,
> unsigned int flags,
> struct list_head *unwind)
> @@ -58,6 +59,9 @@ mark_free(struct drm_mm_scan *scan,
> if (i915_vma_is_pinned(vma))
> return false;
>
> + if (!i915_gem_object_trylock(vma->obj, ww))
> + return false;
> +
> list_add(&vma->evict_link, unwind);
> return drm_mm_scan_add_block(scan, &vma->node);
> }
> @@ -98,6 +102,7 @@ static bool defer_evict(struct i915_vma *vma)
> */
> int
> i915_gem_evict_something(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww,
> u64 min_size, u64 alignment,
> unsigned long color,
> u64 start, u64 end,
> @@ -170,7 +175,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
> continue;
> }
>
> - if (mark_free(&scan, vma, flags, &eviction_list))
> + if (mark_free(&scan, ww, vma, flags, &eviction_list))
> goto found;
> }
>
> @@ -178,6 +183,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
> list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
> ret = drm_mm_scan_remove_block(&scan, &vma->node);
> BUG_ON(ret);
> + i915_gem_object_unlock(vma->obj);
> }
>
> /*
> @@ -222,10 +228,12 @@ i915_gem_evict_something(struct i915_address_space *vm,
> * of any of our objects, thus corrupting the list).
> */
> list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
> - if (drm_mm_scan_remove_block(&scan, &vma->node))
> + if (drm_mm_scan_remove_block(&scan, &vma->node)) {
> __i915_vma_pin(vma);
> - else
> + } else {
> list_del(&vma->evict_link);
> + i915_gem_object_unlock(vma->obj);
> + }
> }
>
> /* Unbinding will emit any required flushes */
> @@ -234,16 +242,22 @@ i915_gem_evict_something(struct i915_address_space *vm,
> __i915_vma_unpin(vma);
> if (ret == 0)
> ret = __i915_vma_unbind(vma);
> +
> + i915_gem_object_unlock(vma->obj);
> }
>
> while (ret == 0 && (node = drm_mm_scan_color_evict(&scan))) {
> vma = container_of(node, struct i915_vma, node);
>
> +
Spurious new line.
Otherwise,
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
> /* If we find any non-objects (!vma), we cannot evict them */
> - if (vma->node.color != I915_COLOR_UNEVICTABLE)
> + if (vma->node.color != I915_COLOR_UNEVICTABLE &&
> + i915_gem_object_trylock(vma->obj, ww)) {
> ret = __i915_vma_unbind(vma);
> - else
> - ret = -ENOSPC; /* XXX search failed, try again? */
> + i915_gem_object_unlock(vma->obj);
> + } else {
> + ret = -ENOSPC;
> + }
> }
>
> return ret;
> @@ -261,6 +275,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
> * memory in e.g. the shrinker.
> */
> int i915_gem_evict_for_node(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww,
> struct drm_mm_node *target,
> unsigned int flags)
> {
> @@ -333,6 +348,11 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
> break;
> }
>
> + if (!i915_gem_object_trylock(vma->obj, ww)) {
> + ret = -ENOSPC;
> + break;
> + }
> +
> /*
> * Never show fear in the face of dragons!
> *
> @@ -350,6 +370,8 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
> __i915_vma_unpin(vma);
> if (ret == 0)
> ret = __i915_vma_unbind(vma);
> +
> + i915_gem_object_unlock(vma->obj);
> }
>
> return ret;
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
> index cd5f2348a187..c810da476c2b 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
> @@ -93,6 +93,7 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
> * asked to wait for eviction and interrupted.
> */
> int i915_gem_gtt_reserve(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww,
> struct drm_mm_node *node,
> u64 size, u64 offset, unsigned long color,
> unsigned int flags)
> @@ -117,7 +118,7 @@ int i915_gem_gtt_reserve(struct i915_address_space *vm,
> if (flags & PIN_NOEVICT)
> return -ENOSPC;
>
> - err = i915_gem_evict_for_node(vm, node, flags);
> + err = i915_gem_evict_for_node(vm, ww, node, flags);
> if (err == 0)
> err = drm_mm_reserve_node(&vm->mm, node);
>
> @@ -184,6 +185,7 @@ static u64 random_offset(u64 start, u64 end, u64 len, u64 align)
> * asked to wait for eviction and interrupted.
> */
> int i915_gem_gtt_insert(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww,
> struct drm_mm_node *node,
> u64 size, u64 alignment, unsigned long color,
> u64 start, u64 end, unsigned int flags)
> @@ -269,7 +271,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
> */
> offset = random_offset(start, end,
> size, alignment ?: I915_GTT_MIN_ALIGNMENT);
> - err = i915_gem_gtt_reserve(vm, node, size, offset, color, flags);
> + err = i915_gem_gtt_reserve(vm, ww, node, size, offset, color, flags);
> if (err != -ENOSPC)
> return err;
>
> @@ -277,7 +279,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
> return -ENOSPC;
>
> /* Randomly selected placement is pinned, do a search */
> - err = i915_gem_evict_something(vm, size, alignment, color,
> + err = i915_gem_evict_something(vm, ww, size, alignment, color,
> start, end, flags);
> if (err)
> return err;
> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
> index c9b0ee5e1d23..e4938aba3fe9 100644
> --- a/drivers/gpu/drm/i915/i915_gem_gtt.h
> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
> @@ -16,6 +16,7 @@
>
> struct drm_i915_gem_object;
> struct i915_address_space;
> +struct i915_gem_ww_ctx;
>
> int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
> struct sg_table *pages);
> @@ -23,11 +24,13 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
> struct sg_table *pages);
>
> int i915_gem_gtt_reserve(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww,
> struct drm_mm_node *node,
> u64 size, u64 offset, unsigned long color,
> unsigned int flags);
>
> int i915_gem_gtt_insert(struct i915_address_space *vm,
> + struct i915_gem_ww_ctx *ww,
> struct drm_mm_node *node,
> u64 size, u64 alignment, unsigned long color,
> u64 start, u64 end, unsigned int flags);
> diff --git a/drivers/gpu/drm/i915/i915_vgpu.c b/drivers/gpu/drm/i915/i915_vgpu.c
> index 31a105bc1792..c97323973f9b 100644
> --- a/drivers/gpu/drm/i915/i915_vgpu.c
> +++ b/drivers/gpu/drm/i915/i915_vgpu.c
> @@ -197,7 +197,7 @@ static int vgt_balloon_space(struct i915_ggtt *ggtt,
> drm_info(&dev_priv->drm,
> "balloon space: range [ 0x%lx - 0x%lx ] %lu KiB.\n",
> start, end, size / 1024);
> - ret = i915_gem_gtt_reserve(&ggtt->vm, node,
> + ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, node,
> size, start, I915_COLOR_UNEVICTABLE,
> 0);
> if (!ret)
> diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
> index d24e90eac948..83df1290df5e 100644
> --- a/drivers/gpu/drm/i915/i915_vma.c
> +++ b/drivers/gpu/drm/i915/i915_vma.c
> @@ -650,7 +650,8 @@ bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long color)
> * 0 on success, negative error code otherwise.
> */
> static int
> -i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
> +i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
> + u64 size, u64 alignment, u64 flags)
> {
> unsigned long color;
> u64 start, end;
> @@ -702,7 +703,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
> range_overflows(offset, size, end))
> return -EINVAL;
>
> - ret = i915_gem_gtt_reserve(vma->vm, &vma->node,
> + ret = i915_gem_gtt_reserve(vma->vm, ww, &vma->node,
> size, offset, color,
> flags);
> if (ret)
> @@ -741,7 +742,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
> size = round_up(size, I915_GTT_PAGE_SIZE_2M);
> }
>
> - ret = i915_gem_gtt_insert(vma->vm, &vma->node,
> + ret = i915_gem_gtt_insert(vma->vm, ww, &vma->node,
> size, alignment, color,
> start, end, flags);
> if (ret)
> @@ -1382,7 +1383,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
> goto err_unlock;
>
> if (!(bound & I915_VMA_BIND_MASK)) {
> - err = i915_vma_insert(vma, size, alignment, flags);
> + err = i915_vma_insert(vma, ww, size, alignment, flags);
> if (err)
> goto err_active;
>
> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> index 7178811366af..c80e1483d603 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> +++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
> @@ -117,7 +117,7 @@ static int igt_evict_something(void *arg)
>
> /* Everything is pinned, nothing should happen */
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_evict_something(&ggtt->vm,
> + err = i915_gem_evict_something(&ggtt->vm, NULL,
> I915_GTT_PAGE_SIZE, 0, 0,
> 0, U64_MAX,
> 0);
> @@ -132,11 +132,12 @@ static int igt_evict_something(void *arg)
>
> /* Everything is unpinned, we should be able to evict something */
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_evict_something(&ggtt->vm,
> + err = i915_gem_evict_something(&ggtt->vm, NULL,
> I915_GTT_PAGE_SIZE, 0, 0,
> 0, U64_MAX,
> 0);
> mutex_unlock(&ggtt->vm.mutex);
> +
> if (err) {
> pr_err("i915_gem_evict_something failed on a full GGTT with err=%d\n",
> err);
> @@ -204,7 +205,7 @@ static int igt_evict_for_vma(void *arg)
>
> /* Everything is pinned, nothing should happen */
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
> + err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
> mutex_unlock(&ggtt->vm.mutex);
> if (err != -ENOSPC) {
> pr_err("i915_gem_evict_for_node on a full GGTT returned err=%d\n",
> @@ -216,7 +217,7 @@ static int igt_evict_for_vma(void *arg)
>
> /* Everything is unpinned, we should be able to evict the node */
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
> + err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
> mutex_unlock(&ggtt->vm.mutex);
> if (err) {
> pr_err("i915_gem_evict_for_node returned err=%d\n",
> @@ -297,7 +298,7 @@ static int igt_evict_for_cache_color(void *arg)
>
> /* Remove just the second vma */
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
> + err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
> mutex_unlock(&ggtt->vm.mutex);
> if (err) {
> pr_err("[0]i915_gem_evict_for_node returned err=%d\n", err);
> @@ -310,7 +311,7 @@ static int igt_evict_for_cache_color(void *arg)
> target.color = I915_CACHE_L3_LLC;
>
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
> + err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
> mutex_unlock(&ggtt->vm.mutex);
> if (!err) {
> pr_err("[1]i915_gem_evict_for_node returned err=%d\n", err);
> @@ -408,7 +409,7 @@ static int igt_evict_contexts(void *arg)
> /* Reserve a block so that we know we have enough to fit a few rq */
> memset(&hole, 0, sizeof(hole));
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_gtt_insert(&ggtt->vm, &hole,
> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &hole,
> PRETEND_GGTT_SIZE, 0, I915_COLOR_UNEVICTABLE,
> 0, ggtt->vm.total,
> PIN_NOEVICT);
> @@ -428,7 +429,7 @@ static int igt_evict_contexts(void *arg)
> goto out_locked;
> }
>
> - if (i915_gem_gtt_insert(&ggtt->vm, &r->node,
> + if (i915_gem_gtt_insert(&ggtt->vm, NULL, &r->node,
> 1ul << 20, 0, I915_COLOR_UNEVICTABLE,
> 0, ggtt->vm.total,
> PIN_NOEVICT)) {
> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
> index 88b3a4d9e27e..169e2330b386 100644
> --- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
> +++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
> @@ -1378,7 +1378,7 @@ static int igt_gtt_reserve(void *arg)
> }
>
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
> + err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
> obj->base.size,
> total,
> obj->cache_level,
> @@ -1430,7 +1430,7 @@ static int igt_gtt_reserve(void *arg)
> }
>
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
> + err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
> obj->base.size,
> total,
> obj->cache_level,
> @@ -1477,7 +1477,7 @@ static int igt_gtt_reserve(void *arg)
> I915_GTT_MIN_ALIGNMENT);
>
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
> + err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
> obj->base.size,
> offset,
> obj->cache_level,
> @@ -1552,7 +1552,7 @@ static int igt_gtt_insert(void *arg)
> /* Check a couple of obviously invalid requests */
> for (ii = invalid_insert; ii->size; ii++) {
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_gtt_insert(&ggtt->vm, &tmp,
> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &tmp,
> ii->size, ii->alignment,
> I915_COLOR_UNEVICTABLE,
> ii->start, ii->end,
> @@ -1594,7 +1594,7 @@ static int igt_gtt_insert(void *arg)
> }
>
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
> obj->base.size, 0, obj->cache_level,
> 0, ggtt->vm.total,
> 0);
> @@ -1654,7 +1654,7 @@ static int igt_gtt_insert(void *arg)
> }
>
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
> obj->base.size, 0, obj->cache_level,
> 0, ggtt->vm.total,
> 0);
> @@ -1703,7 +1703,7 @@ static int igt_gtt_insert(void *arg)
> }
>
> mutex_lock(&ggtt->vm.mutex);
> - err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
> obj->base.size, 0, obj->cache_level,
> 0, ggtt->vm.total,
> 0);
> --
> 2.34.1
>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 13/17] drm/i915: Add object locking to i915_gem_evict_for_node and i915_gem_evict_something
2021-12-17 13:55 ` [Intel-gfx] " Matthew Auld
@ 2021-12-17 15:04 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-17 15:04 UTC (permalink / raw)
To: Matthew Auld; +Cc: Intel Graphics Development, ML dri-devel
On 17-12-2021 14:55, Matthew Auld wrote:
> On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
> <maarten.lankhorst@linux.intel.com> wrote:
>> Because we will start to require the obj->resv lock for unbinding,
>> ensure these shrinker functions also take the lock.
>>
>> This requires some function signature changes, to ensure that the
>> ww context is passed around, but is mostly straightforward.
> Do we even need this? We aren't handling the shared dma-resv for these
> ones, right? Or more just to have a consistent-ish interface?
Yeah, we don't currently use it, but it will allow for easier eviction in the future, with -EDEADLK handling
when required. It will currently only annotate the objects as being held by the ww context, which can cause -EDEADLK when other threads try to lock.
Ideally we'd take a ref to the object and perform a blocking lock, but our locking and eviction code can't handle that right now.
>> Previously this was split up into several patches, but reworking
>> should allow for easier bisection.
>>
>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
>> ---
>> drivers/gpu/drm/i915/gt/intel_ggtt.c | 2 +-
>> drivers/gpu/drm/i915/gt/selftest_hangcheck.c | 2 +-
>> drivers/gpu/drm/i915/gvt/aperture_gm.c | 2 +-
>> drivers/gpu/drm/i915/i915_drv.h | 2 ++
>> drivers/gpu/drm/i915/i915_gem_evict.c | 34 +++++++++++++++----
>> drivers/gpu/drm/i915/i915_gem_gtt.c | 8 +++--
>> drivers/gpu/drm/i915/i915_gem_gtt.h | 3 ++
>> drivers/gpu/drm/i915/i915_vgpu.c | 2 +-
>> drivers/gpu/drm/i915/i915_vma.c | 9 ++---
>> .../gpu/drm/i915/selftests/i915_gem_evict.c | 17 +++++-----
>> drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 14 ++++----
>> 11 files changed, 63 insertions(+), 32 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt.c b/drivers/gpu/drm/i915/gt/intel_ggtt.c
>> index a287c9186ec9..ed43e9c80aaa 100644
>> --- a/drivers/gpu/drm/i915/gt/intel_ggtt.c
>> +++ b/drivers/gpu/drm/i915/gt/intel_ggtt.c
>> @@ -504,7 +504,7 @@ static int ggtt_reserve_guc_top(struct i915_ggtt *ggtt)
>> GEM_BUG_ON(ggtt->vm.total <= GUC_GGTT_TOP);
>> size = ggtt->vm.total - GUC_GGTT_TOP;
>>
>> - ret = i915_gem_gtt_reserve(&ggtt->vm, &ggtt->uc_fw, size,
>> + ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, &ggtt->uc_fw, size,
>> GUC_GGTT_TOP, I915_COLOR_UNEVICTABLE,
>> PIN_NOEVICT);
>> if (ret)
>> diff --git a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
>> index e5ad4d5a91c0..a3597a6bb805 100644
>> --- a/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
>> +++ b/drivers/gpu/drm/i915/gt/selftest_hangcheck.c
>> @@ -1382,7 +1382,7 @@ static int evict_vma(void *data)
>> complete(&arg->completion);
>>
>> mutex_lock(&vm->mutex);
>> - err = i915_gem_evict_for_node(vm, &evict, 0);
>> + err = i915_gem_evict_for_node(vm, NULL, &evict, 0);
>> mutex_unlock(&vm->mutex);
>>
>> return err;
>> diff --git a/drivers/gpu/drm/i915/gvt/aperture_gm.c b/drivers/gpu/drm/i915/gvt/aperture_gm.c
>> index 0d6d59871308..c08098a167e9 100644
>> --- a/drivers/gpu/drm/i915/gvt/aperture_gm.c
>> +++ b/drivers/gpu/drm/i915/gvt/aperture_gm.c
>> @@ -63,7 +63,7 @@ static int alloc_gm(struct intel_vgpu *vgpu, bool high_gm)
>>
>> mutex_lock(>->ggtt->vm.mutex);
>> mmio_hw_access_pre(gt);
>> - ret = i915_gem_gtt_insert(>->ggtt->vm, node,
>> + ret = i915_gem_gtt_insert(>->ggtt->vm, NULL, node,
>> size, I915_GTT_PAGE_SIZE,
>> I915_COLOR_UNEVICTABLE,
>> start, end, flags);
>> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
>> index c180019c607f..2a98192a89c1 100644
>> --- a/drivers/gpu/drm/i915/i915_drv.h
>> +++ b/drivers/gpu/drm/i915/i915_drv.h
>> @@ -1718,11 +1718,13 @@ i915_gem_vm_lookup(struct drm_i915_file_private *file_priv, u32 id)
>>
>> /* i915_gem_evict.c */
>> int __must_check i915_gem_evict_something(struct i915_address_space *vm,
>> + struct i915_gem_ww_ctx *ww,
>> u64 min_size, u64 alignment,
>> unsigned long color,
>> u64 start, u64 end,
>> unsigned flags);
>> int __must_check i915_gem_evict_for_node(struct i915_address_space *vm,
>> + struct i915_gem_ww_ctx *ww,
>> struct drm_mm_node *node,
>> unsigned int flags);
>> int i915_gem_evict_vm(struct i915_address_space *vm,
>> diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
>> index bfd66f539fc1..f502a617b35c 100644
>> --- a/drivers/gpu/drm/i915/i915_gem_evict.c
>> +++ b/drivers/gpu/drm/i915/i915_gem_evict.c
>> @@ -51,6 +51,7 @@ static int ggtt_flush(struct intel_gt *gt)
>>
>> static bool
>> mark_free(struct drm_mm_scan *scan,
>> + struct i915_gem_ww_ctx *ww,
>> struct i915_vma *vma,
>> unsigned int flags,
>> struct list_head *unwind)
>> @@ -58,6 +59,9 @@ mark_free(struct drm_mm_scan *scan,
>> if (i915_vma_is_pinned(vma))
>> return false;
>>
>> + if (!i915_gem_object_trylock(vma->obj, ww))
>> + return false;
>> +
>> list_add(&vma->evict_link, unwind);
>> return drm_mm_scan_add_block(scan, &vma->node);
>> }
>> @@ -98,6 +102,7 @@ static bool defer_evict(struct i915_vma *vma)
>> */
>> int
>> i915_gem_evict_something(struct i915_address_space *vm,
>> + struct i915_gem_ww_ctx *ww,
>> u64 min_size, u64 alignment,
>> unsigned long color,
>> u64 start, u64 end,
>> @@ -170,7 +175,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
>> continue;
>> }
>>
>> - if (mark_free(&scan, vma, flags, &eviction_list))
>> + if (mark_free(&scan, ww, vma, flags, &eviction_list))
>> goto found;
>> }
>>
>> @@ -178,6 +183,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
>> list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
>> ret = drm_mm_scan_remove_block(&scan, &vma->node);
>> BUG_ON(ret);
>> + i915_gem_object_unlock(vma->obj);
>> }
>>
>> /*
>> @@ -222,10 +228,12 @@ i915_gem_evict_something(struct i915_address_space *vm,
>> * of any of our objects, thus corrupting the list).
>> */
>> list_for_each_entry_safe(vma, next, &eviction_list, evict_link) {
>> - if (drm_mm_scan_remove_block(&scan, &vma->node))
>> + if (drm_mm_scan_remove_block(&scan, &vma->node)) {
>> __i915_vma_pin(vma);
>> - else
>> + } else {
>> list_del(&vma->evict_link);
>> + i915_gem_object_unlock(vma->obj);
>> + }
>> }
>>
>> /* Unbinding will emit any required flushes */
>> @@ -234,16 +242,22 @@ i915_gem_evict_something(struct i915_address_space *vm,
>> __i915_vma_unpin(vma);
>> if (ret == 0)
>> ret = __i915_vma_unbind(vma);
>> +
>> + i915_gem_object_unlock(vma->obj);
>> }
>>
>> while (ret == 0 && (node = drm_mm_scan_color_evict(&scan))) {
>> vma = container_of(node, struct i915_vma, node);
>>
>> +
> Spurious new line.
>
> Otherwise,
> Reviewed-by: Matthew Auld <matthew.auld@intel.com>
>
>
>> /* If we find any non-objects (!vma), we cannot evict them */
>> - if (vma->node.color != I915_COLOR_UNEVICTABLE)
>> + if (vma->node.color != I915_COLOR_UNEVICTABLE &&
>> + i915_gem_object_trylock(vma->obj, ww)) {
>> ret = __i915_vma_unbind(vma);
>> - else
>> - ret = -ENOSPC; /* XXX search failed, try again? */
>> + i915_gem_object_unlock(vma->obj);
>> + } else {
>> + ret = -ENOSPC;
>> + }
>> }
>>
>> return ret;
>> @@ -261,6 +275,7 @@ i915_gem_evict_something(struct i915_address_space *vm,
>> * memory in e.g. the shrinker.
>> */
>> int i915_gem_evict_for_node(struct i915_address_space *vm,
>> + struct i915_gem_ww_ctx *ww,
>> struct drm_mm_node *target,
>> unsigned int flags)
>> {
>> @@ -333,6 +348,11 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
>> break;
>> }
>>
>> + if (!i915_gem_object_trylock(vma->obj, ww)) {
>> + ret = -ENOSPC;
>> + break;
>> + }
>> +
>> /*
>> * Never show fear in the face of dragons!
>> *
>> @@ -350,6 +370,8 @@ int i915_gem_evict_for_node(struct i915_address_space *vm,
>> __i915_vma_unpin(vma);
>> if (ret == 0)
>> ret = __i915_vma_unbind(vma);
>> +
>> + i915_gem_object_unlock(vma->obj);
>> }
>>
>> return ret;
>> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
>> index cd5f2348a187..c810da476c2b 100644
>> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
>> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
>> @@ -93,6 +93,7 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
>> * asked to wait for eviction and interrupted.
>> */
>> int i915_gem_gtt_reserve(struct i915_address_space *vm,
>> + struct i915_gem_ww_ctx *ww,
>> struct drm_mm_node *node,
>> u64 size, u64 offset, unsigned long color,
>> unsigned int flags)
>> @@ -117,7 +118,7 @@ int i915_gem_gtt_reserve(struct i915_address_space *vm,
>> if (flags & PIN_NOEVICT)
>> return -ENOSPC;
>>
>> - err = i915_gem_evict_for_node(vm, node, flags);
>> + err = i915_gem_evict_for_node(vm, ww, node, flags);
>> if (err == 0)
>> err = drm_mm_reserve_node(&vm->mm, node);
>>
>> @@ -184,6 +185,7 @@ static u64 random_offset(u64 start, u64 end, u64 len, u64 align)
>> * asked to wait for eviction and interrupted.
>> */
>> int i915_gem_gtt_insert(struct i915_address_space *vm,
>> + struct i915_gem_ww_ctx *ww,
>> struct drm_mm_node *node,
>> u64 size, u64 alignment, unsigned long color,
>> u64 start, u64 end, unsigned int flags)
>> @@ -269,7 +271,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
>> */
>> offset = random_offset(start, end,
>> size, alignment ?: I915_GTT_MIN_ALIGNMENT);
>> - err = i915_gem_gtt_reserve(vm, node, size, offset, color, flags);
>> + err = i915_gem_gtt_reserve(vm, ww, node, size, offset, color, flags);
>> if (err != -ENOSPC)
>> return err;
>>
>> @@ -277,7 +279,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
>> return -ENOSPC;
>>
>> /* Randomly selected placement is pinned, do a search */
>> - err = i915_gem_evict_something(vm, size, alignment, color,
>> + err = i915_gem_evict_something(vm, ww, size, alignment, color,
>> start, end, flags);
>> if (err)
>> return err;
>> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
>> index c9b0ee5e1d23..e4938aba3fe9 100644
>> --- a/drivers/gpu/drm/i915/i915_gem_gtt.h
>> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
>> @@ -16,6 +16,7 @@
>>
>> struct drm_i915_gem_object;
>> struct i915_address_space;
>> +struct i915_gem_ww_ctx;
>>
>> int __must_check i915_gem_gtt_prepare_pages(struct drm_i915_gem_object *obj,
>> struct sg_table *pages);
>> @@ -23,11 +24,13 @@ void i915_gem_gtt_finish_pages(struct drm_i915_gem_object *obj,
>> struct sg_table *pages);
>>
>> int i915_gem_gtt_reserve(struct i915_address_space *vm,
>> + struct i915_gem_ww_ctx *ww,
>> struct drm_mm_node *node,
>> u64 size, u64 offset, unsigned long color,
>> unsigned int flags);
>>
>> int i915_gem_gtt_insert(struct i915_address_space *vm,
>> + struct i915_gem_ww_ctx *ww,
>> struct drm_mm_node *node,
>> u64 size, u64 alignment, unsigned long color,
>> u64 start, u64 end, unsigned int flags);
>> diff --git a/drivers/gpu/drm/i915/i915_vgpu.c b/drivers/gpu/drm/i915/i915_vgpu.c
>> index 31a105bc1792..c97323973f9b 100644
>> --- a/drivers/gpu/drm/i915/i915_vgpu.c
>> +++ b/drivers/gpu/drm/i915/i915_vgpu.c
>> @@ -197,7 +197,7 @@ static int vgt_balloon_space(struct i915_ggtt *ggtt,
>> drm_info(&dev_priv->drm,
>> "balloon space: range [ 0x%lx - 0x%lx ] %lu KiB.\n",
>> start, end, size / 1024);
>> - ret = i915_gem_gtt_reserve(&ggtt->vm, node,
>> + ret = i915_gem_gtt_reserve(&ggtt->vm, NULL, node,
>> size, start, I915_COLOR_UNEVICTABLE,
>> 0);
>> if (!ret)
>> diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
>> index d24e90eac948..83df1290df5e 100644
>> --- a/drivers/gpu/drm/i915/i915_vma.c
>> +++ b/drivers/gpu/drm/i915/i915_vma.c
>> @@ -650,7 +650,8 @@ bool i915_gem_valid_gtt_space(struct i915_vma *vma, unsigned long color)
>> * 0 on success, negative error code otherwise.
>> */
>> static int
>> -i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
>> +i915_vma_insert(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
>> + u64 size, u64 alignment, u64 flags)
>> {
>> unsigned long color;
>> u64 start, end;
>> @@ -702,7 +703,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
>> range_overflows(offset, size, end))
>> return -EINVAL;
>>
>> - ret = i915_gem_gtt_reserve(vma->vm, &vma->node,
>> + ret = i915_gem_gtt_reserve(vma->vm, ww, &vma->node,
>> size, offset, color,
>> flags);
>> if (ret)
>> @@ -741,7 +742,7 @@ i915_vma_insert(struct i915_vma *vma, u64 size, u64 alignment, u64 flags)
>> size = round_up(size, I915_GTT_PAGE_SIZE_2M);
>> }
>>
>> - ret = i915_gem_gtt_insert(vma->vm, &vma->node,
>> + ret = i915_gem_gtt_insert(vma->vm, ww, &vma->node,
>> size, alignment, color,
>> start, end, flags);
>> if (ret)
>> @@ -1382,7 +1383,7 @@ int i915_vma_pin_ww(struct i915_vma *vma, struct i915_gem_ww_ctx *ww,
>> goto err_unlock;
>>
>> if (!(bound & I915_VMA_BIND_MASK)) {
>> - err = i915_vma_insert(vma, size, alignment, flags);
>> + err = i915_vma_insert(vma, ww, size, alignment, flags);
>> if (err)
>> goto err_active;
>>
>> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
>> index 7178811366af..c80e1483d603 100644
>> --- a/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
>> +++ b/drivers/gpu/drm/i915/selftests/i915_gem_evict.c
>> @@ -117,7 +117,7 @@ static int igt_evict_something(void *arg)
>>
>> /* Everything is pinned, nothing should happen */
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_evict_something(&ggtt->vm,
>> + err = i915_gem_evict_something(&ggtt->vm, NULL,
>> I915_GTT_PAGE_SIZE, 0, 0,
>> 0, U64_MAX,
>> 0);
>> @@ -132,11 +132,12 @@ static int igt_evict_something(void *arg)
>>
>> /* Everything is unpinned, we should be able to evict something */
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_evict_something(&ggtt->vm,
>> + err = i915_gem_evict_something(&ggtt->vm, NULL,
>> I915_GTT_PAGE_SIZE, 0, 0,
>> 0, U64_MAX,
>> 0);
>> mutex_unlock(&ggtt->vm.mutex);
>> +
>> if (err) {
>> pr_err("i915_gem_evict_something failed on a full GGTT with err=%d\n",
>> err);
>> @@ -204,7 +205,7 @@ static int igt_evict_for_vma(void *arg)
>>
>> /* Everything is pinned, nothing should happen */
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
>> + err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
>> mutex_unlock(&ggtt->vm.mutex);
>> if (err != -ENOSPC) {
>> pr_err("i915_gem_evict_for_node on a full GGTT returned err=%d\n",
>> @@ -216,7 +217,7 @@ static int igt_evict_for_vma(void *arg)
>>
>> /* Everything is unpinned, we should be able to evict the node */
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
>> + err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
>> mutex_unlock(&ggtt->vm.mutex);
>> if (err) {
>> pr_err("i915_gem_evict_for_node returned err=%d\n",
>> @@ -297,7 +298,7 @@ static int igt_evict_for_cache_color(void *arg)
>>
>> /* Remove just the second vma */
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
>> + err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
>> mutex_unlock(&ggtt->vm.mutex);
>> if (err) {
>> pr_err("[0]i915_gem_evict_for_node returned err=%d\n", err);
>> @@ -310,7 +311,7 @@ static int igt_evict_for_cache_color(void *arg)
>> target.color = I915_CACHE_L3_LLC;
>>
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_evict_for_node(&ggtt->vm, &target, 0);
>> + err = i915_gem_evict_for_node(&ggtt->vm, NULL, &target, 0);
>> mutex_unlock(&ggtt->vm.mutex);
>> if (!err) {
>> pr_err("[1]i915_gem_evict_for_node returned err=%d\n", err);
>> @@ -408,7 +409,7 @@ static int igt_evict_contexts(void *arg)
>> /* Reserve a block so that we know we have enough to fit a few rq */
>> memset(&hole, 0, sizeof(hole));
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_gtt_insert(&ggtt->vm, &hole,
>> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &hole,
>> PRETEND_GGTT_SIZE, 0, I915_COLOR_UNEVICTABLE,
>> 0, ggtt->vm.total,
>> PIN_NOEVICT);
>> @@ -428,7 +429,7 @@ static int igt_evict_contexts(void *arg)
>> goto out_locked;
>> }
>>
>> - if (i915_gem_gtt_insert(&ggtt->vm, &r->node,
>> + if (i915_gem_gtt_insert(&ggtt->vm, NULL, &r->node,
>> 1ul << 20, 0, I915_COLOR_UNEVICTABLE,
>> 0, ggtt->vm.total,
>> PIN_NOEVICT)) {
>> diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
>> index 88b3a4d9e27e..169e2330b386 100644
>> --- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
>> +++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
>> @@ -1378,7 +1378,7 @@ static int igt_gtt_reserve(void *arg)
>> }
>>
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
>> + err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
>> obj->base.size,
>> total,
>> obj->cache_level,
>> @@ -1430,7 +1430,7 @@ static int igt_gtt_reserve(void *arg)
>> }
>>
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
>> + err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
>> obj->base.size,
>> total,
>> obj->cache_level,
>> @@ -1477,7 +1477,7 @@ static int igt_gtt_reserve(void *arg)
>> I915_GTT_MIN_ALIGNMENT);
>>
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_gtt_reserve(&ggtt->vm, &vma->node,
>> + err = i915_gem_gtt_reserve(&ggtt->vm, NULL, &vma->node,
>> obj->base.size,
>> offset,
>> obj->cache_level,
>> @@ -1552,7 +1552,7 @@ static int igt_gtt_insert(void *arg)
>> /* Check a couple of obviously invalid requests */
>> for (ii = invalid_insert; ii->size; ii++) {
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_gtt_insert(&ggtt->vm, &tmp,
>> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &tmp,
>> ii->size, ii->alignment,
>> I915_COLOR_UNEVICTABLE,
>> ii->start, ii->end,
>> @@ -1594,7 +1594,7 @@ static int igt_gtt_insert(void *arg)
>> }
>>
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
>> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
>> obj->base.size, 0, obj->cache_level,
>> 0, ggtt->vm.total,
>> 0);
>> @@ -1654,7 +1654,7 @@ static int igt_gtt_insert(void *arg)
>> }
>>
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
>> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
>> obj->base.size, 0, obj->cache_level,
>> 0, ggtt->vm.total,
>> 0);
>> @@ -1703,7 +1703,7 @@ static int igt_gtt_insert(void *arg)
>> }
>>
>> mutex_lock(&ggtt->vm.mutex);
>> - err = i915_gem_gtt_insert(&ggtt->vm, &vma->node,
>> + err = i915_gem_gtt_insert(&ggtt->vm, NULL, &vma->node,
>> obj->base.size, 0, obj->cache_level,
>> 0, ggtt->vm.total,
>> 0);
>> --
>> 2.34.1
>>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 03/17] drm/i915: Remove pages_mutex and intel_gtt->vma_ops.set/clear_pages members, v3.
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
(?)
@ 2021-12-17 15:21 ` Matthew Auld
-1 siblings, 0 replies; 57+ messages in thread
From: Matthew Auld @ 2021-12-17 15:21 UTC (permalink / raw)
To: Maarten Lankhorst; +Cc: Intel Graphics Development, ML dri-devel
On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
<maarten.lankhorst@linux.intel.com> wrote:
>
> Big delta, but boils down to moving set_pages to i915_vma.c, and removing
> the special handling, all callers use the defaults anyway. We only remap
> in ggtt, so default case will fall through.
>
> Because we still don't require locking in i915_vma_unpin(), handle this by
> using xchg in get_pages(), as it's locked with obj->mutex, and cmpxchg in
> unpin, which only fails if we race a against a new pin.
>
> Changes since v1:
> - aliasing gtt sets ZERO_SIZE_PTR, not -ENODEV, remove special case
> from __i915_vma_get_pages(). (Matt)
> Changes since v2:
> - Free correct old pages in __i915_vma_get_pages(). (Matt)
> Remove race of clearing vma->pages accidentally from put,
> free it but leave it set, as only get has the lock.
>
> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [PATCH v3 08/17] drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors
2021-12-17 11:58 ` [Intel-gfx] " Matthew Auld
@ 2021-12-17 15:29 ` Maarten Lankhorst
-1 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-17 15:29 UTC (permalink / raw)
To: Matthew Auld; +Cc: Intel Graphics Development, ML dri-devel
On 17-12-2021 12:58, Matthew Auld wrote:
> On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
> <maarten.lankhorst@linux.intel.com> wrote:
>> Now that we cannot unbind kill the currently locked object directly
> "unbind kill"
>
>> because we're removing short term pinning, we may have to unbind the
>> object from gtt manually, using a i915_gem_evict_vm() call.
>>
>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> Maybe mention that this only in preparation for some future patches,
> once the actual eviction is trylock and evict_for_vm can also handle
> shared dma-resv? At this point in the series we shouldn't expect to
> hit -ENOSPC, right?
>
>> ---
>> drivers/gpu/drm/i915/gem/i915_gem_mman.c | 18 ++++++++++++++++--
>> 1 file changed, 16 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
>> index af81d6c3332a..00cd9642669a 100644
>> --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
>> @@ -358,8 +358,22 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
>> vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
>> }
>>
>> - /* The entire mappable GGTT is pinned? Unexpected! */
>> - GEM_BUG_ON(vma == ERR_PTR(-ENOSPC));
>> + /*
>> + * The entire mappable GGTT is pinned? Unexpected!
>> + * Try to evict the object we locked too, as normally we skip it
>> + * due to lack of short term pinning inside execbuf.
>> + */
>> + if (vma == ERR_PTR(-ENOSPC)) {
>> + ret = mutex_lock_interruptible(&ggtt->vm.mutex);
>> + if (!ret) {
>> + ret = i915_gem_evict_vm(&ggtt->vm);
>> + mutex_unlock(&ggtt->vm.mutex);
>> + }
>> + if (ret)
>> + goto err_reset;
>> + vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
>> + }
>> + GEM_WARN_ON(vma == ERR_PTR(-ENOSPC));
> Looks like this is being triggered in CI, I assume because the trylock
> could easily fail, due to contention? Is this expected for now? Do we
> keep the WARN and track it as a known issue?
I think it makes sense. I can probably fix i915_gem_evict_vm to attempt to take all objects in a blocking way.
I had some primitives that could lock for eviction, and keep a refcount on the object. i915_gem_evict_vm could probably be changed to use it.
>> }
>> if (IS_ERR(vma)) {
>> ret = PTR_ERR(vma);
>> --
>> 2.34.1
>>
^ permalink raw reply [flat|nested] 57+ messages in thread
* Re: [Intel-gfx] [PATCH v3 08/17] drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors
@ 2021-12-17 15:29 ` Maarten Lankhorst
0 siblings, 0 replies; 57+ messages in thread
From: Maarten Lankhorst @ 2021-12-17 15:29 UTC (permalink / raw)
To: Matthew Auld; +Cc: Intel Graphics Development, ML dri-devel
On 17-12-2021 12:58, Matthew Auld wrote:
> On Thu, 16 Dec 2021 at 14:28, Maarten Lankhorst
> <maarten.lankhorst@linux.intel.com> wrote:
>> Now that we cannot unbind kill the currently locked object directly
> "unbind kill"
>
>> because we're removing short term pinning, we may have to unbind the
>> object from gtt manually, using a i915_gem_evict_vm() call.
>>
>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> Maybe mention that this only in preparation for some future patches,
> once the actual eviction is trylock and evict_for_vm can also handle
> shared dma-resv? At this point in the series we shouldn't expect to
> hit -ENOSPC, right?
>
>> ---
>> drivers/gpu/drm/i915/gem/i915_gem_mman.c | 18 ++++++++++++++++--
>> 1 file changed, 16 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
>> index af81d6c3332a..00cd9642669a 100644
>> --- a/drivers/gpu/drm/i915/gem/i915_gem_mman.c
>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_mman.c
>> @@ -358,8 +358,22 @@ static vm_fault_t vm_fault_gtt(struct vm_fault *vmf)
>> vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
>> }
>>
>> - /* The entire mappable GGTT is pinned? Unexpected! */
>> - GEM_BUG_ON(vma == ERR_PTR(-ENOSPC));
>> + /*
>> + * The entire mappable GGTT is pinned? Unexpected!
>> + * Try to evict the object we locked too, as normally we skip it
>> + * due to lack of short term pinning inside execbuf.
>> + */
>> + if (vma == ERR_PTR(-ENOSPC)) {
>> + ret = mutex_lock_interruptible(&ggtt->vm.mutex);
>> + if (!ret) {
>> + ret = i915_gem_evict_vm(&ggtt->vm);
>> + mutex_unlock(&ggtt->vm.mutex);
>> + }
>> + if (ret)
>> + goto err_reset;
>> + vma = i915_gem_object_ggtt_pin_ww(obj, &ww, &view, 0, 0, flags);
>> + }
>> + GEM_WARN_ON(vma == ERR_PTR(-ENOSPC));
> Looks like this is being triggered in CI, I assume because the trylock
> could easily fail, due to contention? Is this expected for now? Do we
> keep the WARN and track it as a known issue?
I think it makes sense. I can probably fix i915_gem_evict_vm to attempt to take all objects in a blocking way.
I had some primitives that could lock for eviction, and keep a refcount on the object. i915_gem_evict_vm could probably be changed to use it.
>> }
>> if (IS_ERR(vma)) {
>> ret = PTR_ERR(vma);
>> --
>> 2.34.1
>>
^ permalink raw reply [flat|nested] 57+ messages in thread
end of thread, other threads:[~2021-12-17 15:29 UTC | newest]
Thread overview: 57+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-12-16 14:27 [PATCH v3 00/17] drm/i915: Remove short term pins from execbuf by requiring lock to unbind Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-16 14:27 ` [PATCH v3 01/17] drm/i915: Remove unused bits of i915_vma/active api Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-16 14:27 ` [PATCH v3 02/17] drm/i915: Change shrink ordering to use locking around unbinding Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-16 14:27 ` [PATCH v3 03/17] drm/i915: Remove pages_mutex and intel_gtt->vma_ops.set/clear_pages members, v3 Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-17 15:21 ` Matthew Auld
2021-12-16 14:27 ` [PATCH v3 04/17] drm/i915: Take object lock in i915_ggtt_pin if ww is not set Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-17 10:44 ` Matthew Auld
2021-12-17 10:44 ` [Intel-gfx] " Matthew Auld
2021-12-16 14:27 ` [PATCH v3 05/17] drm/i915: Force ww lock for i915_gem_object_ggtt_pin_ww, v2 Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-17 10:46 ` Matthew Auld
2021-12-16 14:27 ` [PATCH v3 06/17] drm/i915: Ensure gem_contexts selftests work with unbind changes, v2 Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-16 14:27 ` [PATCH v3 07/17] drm/i915: Ensure i915_vma tests do not get -ENOSPC with the locking changes Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-17 11:25 ` Matthew Auld
2021-12-17 11:25 ` [Intel-gfx] " Matthew Auld
2021-12-16 14:27 ` [PATCH v3 08/17] drm/i915: Call i915_gem_evict_vm in vm_fault_gtt to prevent new ENOSPC errors Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-17 11:58 ` Matthew Auld
2021-12-17 11:58 ` [Intel-gfx] " Matthew Auld
2021-12-17 15:29 ` Maarten Lankhorst
2021-12-17 15:29 ` [Intel-gfx] " Maarten Lankhorst
2021-12-16 14:27 ` [PATCH v3 09/17] drm/i915: Trylock the object when shrinking Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-17 12:53 ` Matthew Auld
2021-12-17 12:53 ` [Intel-gfx] " Matthew Auld
2021-12-16 14:27 ` [Intel-gfx] [PATCH v3 10/17] drm/i915: Require object lock when freeing pages during destruction Maarten Lankhorst
2021-12-16 14:27 ` Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] [PATCH v3 11/17] drm/i915: Add ww ctx to i915_gem_object_trylock Maarten Lankhorst
2021-12-16 14:27 ` Maarten Lankhorst
2021-12-17 12:55 ` [Intel-gfx] " Matthew Auld
2021-12-16 14:27 ` [Intel-gfx] [PATCH v3 12/17] drm/i915: Add locking to i915_gem_evict_vm() Maarten Lankhorst
2021-12-16 14:27 ` Maarten Lankhorst
2021-12-17 13:19 ` [Intel-gfx] " Matthew Auld
2021-12-16 14:27 ` [Intel-gfx] [PATCH v3 13/17] drm/i915: Add object locking to i915_gem_evict_for_node and i915_gem_evict_something Maarten Lankhorst
2021-12-16 14:27 ` Maarten Lankhorst
2021-12-17 13:55 ` [Intel-gfx] " Matthew Auld
2021-12-17 15:04 ` Maarten Lankhorst
2021-12-16 14:27 ` [PATCH v3 14/17] drm/i915: Add i915_vma_unbind_unlocked, and take obj lock for i915_vma_unbind, v2 Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-16 14:27 ` [PATCH v3 15/17] drm/i915: Remove assert_object_held_shared Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] [PATCH v3 16/17] drm/i915: Remove support for unlocked i915_vma unbind Maarten Lankhorst
2021-12-16 14:27 ` Maarten Lankhorst
2021-12-16 14:27 ` [PATCH v3 17/17] drm/i915: Remove short-term pins from execbuf, v5 Maarten Lankhorst
2021-12-16 14:27 ` [Intel-gfx] " Maarten Lankhorst
2021-12-17 8:46 ` [Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for drm/i915: Remove short term pins from execbuf by requiring lock to unbind Patchwork
2021-12-17 8:47 ` [Intel-gfx] ✗ Fi.CI.SPARSE: " Patchwork
2021-12-17 8:51 ` [Intel-gfx] ✗ Fi.CI.DOCS: " Patchwork
2021-12-17 9:10 ` [Intel-gfx] ✓ Fi.CI.BAT: success " Patchwork
2021-12-17 10:48 ` [Intel-gfx] ✗ Fi.CI.IGT: failure " Patchwork
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.