The check needs to be for > 1, because ctx->acquired is already incremented. This will prevent ww_mutex_lock_slow from returning -EDEADLK and not locking the mutex. It caused a lot of false gpu lockups on radeon with CONFIG_DEBUG_WW_MUTEX_SLOWPATH=y because a function that shouldn't be able to return -EDEADLK did.
Cc: Alex Deucher alexander.deucher@amd.com Signed-off-by: Maarten Lankhorst maarten.lankhorst@canonical.com --- diff --git a/kernel/mutex.c b/kernel/mutex.c index ff05f4b..a52ee7bb 100644 --- a/kernel/mutex.c +++ b/kernel/mutex.c @@ -686,7 +686,7 @@ __ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) might_sleep(); ret = __mutex_lock_common(&lock->base, TASK_UNINTERRUPTIBLE, 0, &ctx->dep_map, _RET_IP_, ctx); - if (!ret && ctx->acquired > 0) + if (!ret && ctx->acquired > 1) return ww_mutex_deadlock_injection(lock, ctx);
return ret; @@ -702,7 +702,7 @@ __ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) ret = __mutex_lock_common(&lock->base, TASK_INTERRUPTIBLE, 0, &ctx->dep_map, _RET_IP_, ctx);
- if (!ret && ctx->acquired > 0) + if (!ret && ctx->acquired > 1) return ww_mutex_deadlock_injection(lock, ctx);
return ret;
On Tue, Jul 30, 2013 at 10:13:41AM +0200, Maarten Lankhorst wrote:
The check needs to be for > 1, because ctx->acquired is already incremented. This will prevent ww_mutex_lock_slow from returning -EDEADLK and not locking the mutex. It caused a lot of false gpu lockups on radeon with CONFIG_DEBUG_WW_MUTEX_SLOWPATH=y because a function that shouldn't be able to return -EDEADLK did.
Cc: Alex Deucher alexander.deucher@amd.com Signed-off-by: Maarten Lankhorst maarten.lankhorst@canonical.com
Thanks!
On Tue, Jul 30, 2013 at 6:41 PM, Peter Zijlstra peterz@infradead.org wrote:
On Tue, Jul 30, 2013 at 10:13:41AM +0200, Maarten Lankhorst wrote:
The check needs to be for > 1, because ctx->acquired is already incremented. This will prevent ww_mutex_lock_slow from returning -EDEADLK and not locking the mutex. It caused a lot of false gpu lockups on radeon with CONFIG_DEBUG_WW_MUTEX_SLOWPATH=y because a function that shouldn't be able to return -EDEADLK did.
Cc: Alex Deucher alexander.deucher@amd.com Signed-off-by: Maarten Lankhorst maarten.lankhorst@canonical.com
Should this be merged via Ingo? or will I queue it in my -fixes?
Dave.
Op 07-08-13 02:05, Dave Airlie schreef:
On Tue, Jul 30, 2013 at 6:41 PM, Peter Zijlstra peterz@infradead.org wrote:
On Tue, Jul 30, 2013 at 10:13:41AM +0200, Maarten Lankhorst wrote:
The check needs to be for > 1, because ctx->acquired is already incremented. This will prevent ww_mutex_lock_slow from returning -EDEADLK and not locking the mutex. It caused a lot of false gpu lockups on radeon with CONFIG_DEBUG_WW_MUTEX_SLOWPATH=y because a function that shouldn't be able to return -EDEADLK did.
Cc: Alex Deucher alexander.deucher@amd.com Signed-off-by: Maarten Lankhorst maarten.lankhorst@canonical.com
Should this be merged via Ingo? or will I queue it in my -fixes?
Dave.
It's in tip:core/urgent, so I imagine you don't need to queue it.
On Tue, Jul 30, 2013 at 4:13 AM, Maarten Lankhorst maarten.lankhorst@canonical.com wrote:
The check needs to be for > 1, because ctx->acquired is already incremented. This will prevent ww_mutex_lock_slow from returning -EDEADLK and not locking the mutex. It caused a lot of false gpu lockups on radeon with CONFIG_DEBUG_WW_MUTEX_SLOWPATH=y because a function that shouldn't be able to return -EDEADLK did.
I haven't followed the new reservation stuff too closely, but seems plausible.
Acked-by: Alex Deucher alexander.deucher@amd.com
Cc: Alex Deucher alexander.deucher@amd.com Signed-off-by: Maarten Lankhorst maarten.lankhorst@canonical.com
diff --git a/kernel/mutex.c b/kernel/mutex.c index ff05f4b..a52ee7bb 100644 --- a/kernel/mutex.c +++ b/kernel/mutex.c @@ -686,7 +686,7 @@ __ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) might_sleep(); ret = __mutex_lock_common(&lock->base, TASK_UNINTERRUPTIBLE, 0, &ctx->dep_map, _RET_IP_, ctx);
if (!ret && ctx->acquired > 0)
if (!ret && ctx->acquired > 1) return ww_mutex_deadlock_injection(lock, ctx); return ret;@@ -702,7 +702,7 @@ __ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) ret = __mutex_lock_common(&lock->base, TASK_INTERRUPTIBLE, 0, &ctx->dep_map, _RET_IP_, ctx);
if (!ret && ctx->acquired > 0)
if (!ret && ctx->acquired > 1) return ww_mutex_deadlock_injection(lock, ctx); return ret;
dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel
On Tue, Jul 30, 2013 at 10:13:41AM +0200, Maarten Lankhorst wrote:
The check needs to be for > 1, because ctx->acquired is already incremented. This will prevent ww_mutex_lock_slow from returning -EDEADLK and not locking the mutex. It caused a lot of false gpu lockups on radeon with CONFIG_DEBUG_WW_MUTEX_SLOWPATH=y because a function that shouldn't be able to return -EDEADLK did.
Cc: Alex Deucher alexander.deucher@amd.com Signed-off-by: Maarten Lankhorst maarten.lankhorst@canonical.com
Oops, thanks for catching this.
Reviewed-by: Daniel Vetter daniel.vetter@ffwll.ch
diff --git a/kernel/mutex.c b/kernel/mutex.c index ff05f4b..a52ee7bb 100644 --- a/kernel/mutex.c +++ b/kernel/mutex.c @@ -686,7 +686,7 @@ __ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) might_sleep(); ret = __mutex_lock_common(&lock->base, TASK_UNINTERRUPTIBLE, 0, &ctx->dep_map, _RET_IP_, ctx);
- if (!ret && ctx->acquired > 0)
if (!ret && ctx->acquired > 1) return ww_mutex_deadlock_injection(lock, ctx);
return ret;
@@ -702,7 +702,7 @@ __ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) ret = __mutex_lock_common(&lock->base, TASK_INTERRUPTIBLE, 0, &ctx->dep_map, _RET_IP_, ctx);
- if (!ret && ctx->acquired > 0)
if (!ret && ctx->acquired > 1) return ww_mutex_deadlock_injection(lock, ctx);
return ret;
dri-devel mailing list dri-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/dri-devel
dri-devel@lists.freedesktop.org
-
Alex Deucher -
Daniel Vetter -
Dave Airlie -
Maarten Lankhorst -
Peter Zijlstra