Re: [Intel-gfx] [PATCH] drm/i915: Don't disable interrupts independently of the lock
On 2019-10-10 21:30:35 [+0100], Chris Wilson wrote: > > | spin_lock_irq(>engine->active.lock); > > | list_del(>sched.link); > > | spin_unlock_irq(>engine->active.lock); > > | > > | spin_lock_irq(>lock); > > | i915_request_mark_complete(rq); > > … > > | spin_unlock_irq(>lock); > > > > has been avoided because an interrupt here could change something or if > > this is just an optimisation. > > Just avoiding the back-to-back enable/disable. as I assumed. Is the patch okay? > -Chris Sebastian ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [Intel-gfx] [PATCH] drm/i915: Don't disable interrupts independently of the lock
Quoting Sebastian Andrzej Siewior (2019-10-10 19:26:10) > On 2019-10-10 19:11:27 [+0100], Chris Wilson wrote: > > > --- a/drivers/gpu/drm/i915/i915_request.c > > > +++ b/drivers/gpu/drm/i915/i915_request.c > > > @@ -251,15 +251,13 @@ static bool i915_request_retire(struct i > > > active->retire(active, rq); > > > } > > > > > > - local_irq_disable(); > > > - > > > /* > > > * We only loosely track inflight requests across preemption, > > > * and so we may find ourselves attempting to retire a _completed_ > > > * request that we have removed from the HW and put back on a run > > > * queue. > > > */ > > > - spin_lock(>engine->active.lock); > > > + spin_lock_irq(>engine->active.lock); > > > list_del(>sched.link); > > > spin_unlock(>engine->active.lock); > > > > > > @@ -278,9 +276,7 @@ static bool i915_request_retire(struct i > > > __notify_execute_cb(rq); > > > } > > > GEM_BUG_ON(!list_empty(>execute_cb)); > > > - spin_unlock(>lock); > > > - > > > - local_irq_enable(); > > > + spin_unlock_irq(>lock); > > > > Nothing screams about the imbalance? irq off from one lock to the other? > > There is no imbalance, is there? Interrupts are disabled as part of > acquiring the first lock and enabled again as part of releasing the > second lock. > It may not look beautiful. Sure, it's at the same scope, I just expect at some point lockdep to complain :) > I'm just not sure if this > > | spin_lock_irq(>engine->active.lock); > | list_del(>sched.link); > | spin_unlock_irq(>engine->active.lock); > | > | spin_lock_irq(>lock); > | i915_request_mark_complete(rq); > … > | spin_unlock_irq(>lock); > > has been avoided because an interrupt here could change something or if > this is just an optimisation. Just avoiding the back-to-back enable/disable. -Chris ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [Intel-gfx] [PATCH] drm/i915: Don't disable interrupts independently of the lock
On 2019-10-10 19:11:27 [+0100], Chris Wilson wrote: > > --- a/drivers/gpu/drm/i915/i915_request.c > > +++ b/drivers/gpu/drm/i915/i915_request.c > > @@ -251,15 +251,13 @@ static bool i915_request_retire(struct i > > active->retire(active, rq); > > } > > > > - local_irq_disable(); > > - > > /* > > * We only loosely track inflight requests across preemption, > > * and so we may find ourselves attempting to retire a _completed_ > > * request that we have removed from the HW and put back on a run > > * queue. > > */ > > - spin_lock(>engine->active.lock); > > + spin_lock_irq(>engine->active.lock); > > list_del(>sched.link); > > spin_unlock(>engine->active.lock); > > > > @@ -278,9 +276,7 @@ static bool i915_request_retire(struct i > > __notify_execute_cb(rq); > > } > > GEM_BUG_ON(!list_empty(>execute_cb)); > > - spin_unlock(>lock); > > - > > - local_irq_enable(); > > + spin_unlock_irq(>lock); > > Nothing screams about the imbalance? irq off from one lock to the other? There is no imbalance, is there? Interrupts are disabled as part of acquiring the first lock and enabled again as part of releasing the second lock. It may not look beautiful. I'm just not sure if this | spin_lock_irq(>engine->active.lock); | list_del(>sched.link); | spin_unlock_irq(>engine->active.lock); | | spin_lock_irq(>lock); | i915_request_mark_complete(rq); … | spin_unlock_irq(>lock); has been avoided because an interrupt here could change something or if this is just an optimisation. > -Chris Sebastian ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [Intel-gfx] [PATCH] drm/i915: Don't disable interrupts independently of the lock
Quoting Sebastian Andrzej Siewior (2019-10-10 17:06:40) > The locks (active.lock and rq->lock) need to be taken with disabled > interrupts. This is done in i915_request_retire() by disabling the > interrupts independently of the locks itself. > While local_irq_disable()+spin_lock() equals spin_lock_irq() on vanilla > it does not on PREEMPT_RT. Also, it is not obvious if there is a special > reason > to why the interrupts are disabled independently of the lock. > > Enable/disable interrupts as part of the locking instruction. > > Signed-off-by: Sebastian Andrzej Siewior > --- > drivers/gpu/drm/i915/i915_request.c |8 ++-- > 1 file changed, 2 insertions(+), 6 deletions(-) > > --- a/drivers/gpu/drm/i915/i915_request.c > +++ b/drivers/gpu/drm/i915/i915_request.c > @@ -251,15 +251,13 @@ static bool i915_request_retire(struct i > active->retire(active, rq); > } > > - local_irq_disable(); > - > /* > * We only loosely track inflight requests across preemption, > * and so we may find ourselves attempting to retire a _completed_ > * request that we have removed from the HW and put back on a run > * queue. > */ > - spin_lock(>engine->active.lock); > + spin_lock_irq(>engine->active.lock); > list_del(>sched.link); > spin_unlock(>engine->active.lock); > > @@ -278,9 +276,7 @@ static bool i915_request_retire(struct i > __notify_execute_cb(rq); > } > GEM_BUG_ON(!list_empty(>execute_cb)); > - spin_unlock(>lock); > - > - local_irq_enable(); > + spin_unlock_irq(>lock); Nothing screams about the imbalance? irq off from one lock to the other? -Chris ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
[Intel-gfx] [PATCH] drm/i915: Don't disable interrupts independently of the lock
The locks (active.lock and rq->lock) need to be taken with disabled interrupts. This is done in i915_request_retire() by disabling the interrupts independently of the locks itself. While local_irq_disable()+spin_lock() equals spin_lock_irq() on vanilla it does not on PREEMPT_RT. Also, it is not obvious if there is a special reason to why the interrupts are disabled independently of the lock. Enable/disable interrupts as part of the locking instruction. Signed-off-by: Sebastian Andrzej Siewior --- drivers/gpu/drm/i915/i915_request.c |8 ++-- 1 file changed, 2 insertions(+), 6 deletions(-) --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -251,15 +251,13 @@ static bool i915_request_retire(struct i active->retire(active, rq); } - local_irq_disable(); - /* * We only loosely track inflight requests across preemption, * and so we may find ourselves attempting to retire a _completed_ * request that we have removed from the HW and put back on a run * queue. */ - spin_lock(>engine->active.lock); + spin_lock_irq(>engine->active.lock); list_del(>sched.link); spin_unlock(>engine->active.lock); @@ -278,9 +276,7 @@ static bool i915_request_retire(struct i __notify_execute_cb(rq); } GEM_BUG_ON(!list_empty(>execute_cb)); - spin_unlock(>lock); - - local_irq_enable(); + spin_unlock_irq(>lock); remove_from_client(rq); list_del(>link); ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
[Intel-gfx] [PATCH] drm/i915: Don't disable interrupts independently of the lock
The locks (timeline->lock and rq->lock) need to be taken with disabled interrupts. This is done in __retire_engine_request() by disabling the interrupts independently of the locks itself. While local_irq_disable()+spin_lock() equals spin_lock_irq() on vanilla it does not on RT. Also, it is not obvious if there is a special reason to why the interrupts are disabled independently of the lock. Enable/disable interrupts as part of the locking instruction. Signed-off-by: Sebastian Andrzej Siewior --- drivers/gpu/drm/i915/i915_request.c | 8 ++-- 1 file changed, 2 insertions(+), 6 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c index ca95ab2f4cfa3..8744d20ac1681 100644 --- a/drivers/gpu/drm/i915/i915_request.c +++ b/drivers/gpu/drm/i915/i915_request.c @@ -278,9 +278,7 @@ static void __retire_engine_request(struct intel_engine_cs *engine, GEM_BUG_ON(!i915_request_completed(rq)); - local_irq_disable(); - - spin_lock(>timeline.lock); + spin_lock_irq(>timeline.lock); GEM_BUG_ON(!list_is_first(>link, >timeline.requests)); list_del_init(>link); spin_unlock(>timeline.lock); @@ -294,9 +292,7 @@ static void __retire_engine_request(struct intel_engine_cs *engine, GEM_BUG_ON(!atomic_read(>i915->gt_pm.rps.num_waiters)); atomic_dec(>i915->gt_pm.rps.num_waiters); } - spin_unlock(>lock); - - local_irq_enable(); + spin_unlock_irq(>lock); /* * The backing object for the context is done after switching to the -- 2.20.1 ___ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
