| Message ID | 20211118165914.pckik75emivsilek@linutronix.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/i915: Don't disable interrupts and pretend a lock as been acquired in __timeline_mark_lock(). | expand |
On 2021-11-19 08:47:56 [-0000], Patchwork wrote: > == Series Details == > > Series: drm/i915: Don't disable interrupts and pretend a lock as been acquired in __timeline_mark_lock(). > URL : https://patchwork.freedesktop.org/series/97082/ > State : failure > > == Summary == > > CI Bug Log - changes from CI_DRM_10900_full -> Patchwork_21633_full > ==================================================== > > Summary > ------- > > **FAILURE** > > Serious unknown changes coming with Patchwork_21633_full absolutely need to be > verified manually. > > If you think the reported changes have nothing to do with the changes > introduced in Patchwork_21633_full, please notify your bug team to allow them > to document this new failure mode, which will reduce false positives in CI. My bug team? > #### Possible regressions #### > > * igt@kms_vblank@pipe-b-query-forked-hang: > - shard-tglb: [PASS][1] -> [INCOMPLETE][2] > [1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10900/shard-tglb8/igt@kms_vblank@pipe-b-query-forked-hang.html > [2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_21633/shard-tglb6/igt@kms_vblank@pipe-b-query-forked-hang.html By comparing the two dmesg outputs I see that the INCOMPLETE ones stops, while the PASS one continues with |<7> [82.256394] heartbeat rcs0 heartbeat {seqno:5:7, prio:2147483646} not ticking I cloned igt-gpu-tools and run 'kms_vblank' on my box. I was a little surprised by the suspend part but the test passed here. Sebastian
On Thu, Nov 18, 2021 at 05:59:14PM +0100, Sebastian Andrzej Siewior wrote: > This is a revert of commits > d67739268cf0e ("drm/i915/gt: Mark up the nested engine-pm timeline lock as irqsafe") > 6c69a45445af9 ("drm/i915/gt: Mark context->active_count as protected by timeline->mutex") > > The existing code leads to a different behaviour depending on whether > lockdep is enabled or not. Any following lock that is acquired without > disabling interrupts (but needs to) will not be noticed by lockdep. > > This it not just a lockdep annotation but is used but an actual mutex_t > that is properly used as a lock but in case of __timeline_mark_lock() > lockdep is only told that it is acquired but no lock has been acquired. > > It appears that its purpose is just satisfy the lockdep_assert_held() > check in intel_context_mark_active(). The other problem with disabling > interrupts is that on PREEMPT_RT interrupts are also disabled which > leads to problems for instance later during memory allocation. > > Add a CONTEXT_IS_PARKED bit to intel_engine_cs and set_bit/clear_bit it > instead of mutex_acquire/mutex_release. Use test_bit in the two > identified spots which relied on the lockdep annotation. > > Cc: Peter Zijlstra <peterz@infradead.org> > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Yeah if we can simplify this with reverts then I'm all for this. Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> I've asked drm/i915 maintainers to check&merge. -Daniel > --- > drivers/gpu/drm/i915/gt/intel_context.h | 3 +- > drivers/gpu/drm/i915/gt/intel_context_types.h | 1 > drivers/gpu/drm/i915/gt/intel_engine_pm.c | 38 +------------------------- > drivers/gpu/drm/i915/i915_request.h | 3 +- > 4 files changed, 7 insertions(+), 38 deletions(-) > > --- a/drivers/gpu/drm/i915/gt/intel_context.h > +++ b/drivers/gpu/drm/i915/gt/intel_context.h > @@ -211,7 +211,8 @@ static inline void intel_context_enter(s > > static inline void intel_context_mark_active(struct intel_context *ce) > { > - lockdep_assert_held(&ce->timeline->mutex); > + lockdep_assert(lockdep_is_held(&ce->timeline->mutex) || > + test_bit(CONTEXT_IS_PARKED, &ce->flags)); > ++ce->active_count; > } > > --- a/drivers/gpu/drm/i915/gt/intel_context_types.h > +++ b/drivers/gpu/drm/i915/gt/intel_context_types.h > @@ -118,6 +118,7 @@ struct intel_context { > #define CONTEXT_LRCA_DIRTY 9 > #define CONTEXT_GUC_INIT 10 > #define CONTEXT_PERMA_PIN 11 > +#define CONTEXT_IS_PARKED 12 > > struct { > u64 timeout_us; > --- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c > +++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.c > @@ -80,39 +80,6 @@ static int __engine_unpark(struct intel_ > return 0; > } > > -#if IS_ENABLED(CONFIG_LOCKDEP) > - > -static unsigned long __timeline_mark_lock(struct intel_context *ce) > -{ > - unsigned long flags; > - > - local_irq_save(flags); > - mutex_acquire(&ce->timeline->mutex.dep_map, 2, 0, _THIS_IP_); > - > - return flags; > -} > - > -static void __timeline_mark_unlock(struct intel_context *ce, > - unsigned long flags) > -{ > - mutex_release(&ce->timeline->mutex.dep_map, _THIS_IP_); > - local_irq_restore(flags); > -} > - > -#else > - > -static unsigned long __timeline_mark_lock(struct intel_context *ce) > -{ > - return 0; > -} > - > -static void __timeline_mark_unlock(struct intel_context *ce, > - unsigned long flags) > -{ > -} > - > -#endif /* !IS_ENABLED(CONFIG_LOCKDEP) */ > - > static void duration(struct dma_fence *fence, struct dma_fence_cb *cb) > { > struct i915_request *rq = to_request(fence); > @@ -159,7 +126,6 @@ static bool switch_to_kernel_context(str > { > struct intel_context *ce = engine->kernel_context; > struct i915_request *rq; > - unsigned long flags; > bool result = true; > > /* > @@ -214,7 +180,7 @@ static bool switch_to_kernel_context(str > * engine->wakeref.count, we may see the request completion and retire > * it causing an underflow of the engine->wakeref. > */ > - flags = __timeline_mark_lock(ce); > + set_bit(CONTEXT_IS_PARKED, &ce->flags); > GEM_BUG_ON(atomic_read(&ce->timeline->active_count) < 0); > > rq = __i915_request_create(ce, GFP_NOWAIT); > @@ -246,7 +212,7 @@ static bool switch_to_kernel_context(str > > result = false; > out_unlock: > - __timeline_mark_unlock(ce, flags); > + clear_bit(CONTEXT_IS_PARKED, &ce->flags); > return result; > } > > --- a/drivers/gpu/drm/i915/i915_request.h > +++ b/drivers/gpu/drm/i915/i915_request.h > @@ -642,7 +642,8 @@ i915_request_timeline(const struct i915_ > { > /* Valid only while the request is being constructed (or retired). */ > return rcu_dereference_protected(rq->timeline, > - lockdep_is_held(&rcu_access_pointer(rq->timeline)->mutex)); > + lockdep_is_held(&rcu_access_pointer(rq->timeline)->mutex) || > + test_bit(CONTEXT_IS_PARKED, &rq->context->flags)); > } > > static inline struct i915_gem_context *
On 2021-11-19 17:04:00 [+0100], Daniel Vetter wrote: > Yeah if we can simplify this with reverts then I'm all for this. > > Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> > > I've asked drm/i915 maintainers to check&merge. Thanks. Should I repost my queue (excluding this one) or should wait until this one has been taken care? > -Daniel Sebastian
On Tue, Nov 30, 2021 at 05:33:09PM +0100, Sebastian Andrzej Siewior wrote: > On 2021-11-19 17:04:00 [+0100], Daniel Vetter wrote: > > Yeah if we can simplify this with reverts then I'm all for this. > > > > Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch> > > > > I've asked drm/i915 maintainers to check&merge. > > Thanks. Should I repost my queue (excluding this one) or should wait > until this one has been taken care? No idea, Tvrtko (check latest MAINTAINERS) and Joonas need to take care of this. Holler again if it's falling through the cracks. -Daniel
On 18/11/2021 16:59, Sebastian Andrzej Siewior wrote: > This is a revert of commits > d67739268cf0e ("drm/i915/gt: Mark up the nested engine-pm timeline lock as irqsafe") > 6c69a45445af9 ("drm/i915/gt: Mark context->active_count as protected by timeline->mutex") 6dcb85a0ad99 ("drm/i915: Hold irq-off for the entire fake lock period") is very relevant as well. > The existing code leads to a different behaviour depending on whether > lockdep is enabled or not. Any following lock that is acquired without > disabling interrupts (but needs to) will not be noticed by lockdep. Agreed this is not good. > This it not just a lockdep annotation but is used but an actual mutex_t > that is properly used as a lock but in case of __timeline_mark_lock() > lockdep is only told that it is acquired but no lock has been acquired. > > It appears that its purpose is just satisfy the lockdep_assert_held() > check in intel_context_mark_active(). The other problem with disabling > interrupts is that on PREEMPT_RT interrupts are also disabled which > leads to problems for instance later during memory allocation. Hmm commit message of d67739268cf0 ("drm/i915/gt: Mark up the nested engine-pm timeline lock as irqsafe") makes it sound like parking can run with interrupts disabled already. I mean outside of the irq disable hack in here. I don't see it possible though, both due might_lock in intel_wakeref_put and GFP_NOWAIT in switch_to_kernel_context. So I will assume that commit message refers to an earlier state of the code base. So the approach in this patch looks good to me. > Add a CONTEXT_IS_PARKED bit to intel_engine_cs and set_bit/clear_bit it > instead of mutex_acquire/mutex_release. Use test_bit in the two > identified spots which relied on the lockdep annotation. > > Cc: Peter Zijlstra <peterz@infradead.org> > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> > --- > drivers/gpu/drm/i915/gt/intel_context.h | 3 +- > drivers/gpu/drm/i915/gt/intel_context_types.h | 1 > drivers/gpu/drm/i915/gt/intel_engine_pm.c | 38 +------------------------- > drivers/gpu/drm/i915/i915_request.h | 3 +- > 4 files changed, 7 insertions(+), 38 deletions(-) > > --- a/drivers/gpu/drm/i915/gt/intel_context.h > +++ b/drivers/gpu/drm/i915/gt/intel_context.h > @@ -211,7 +211,8 @@ static inline void intel_context_enter(s > > static inline void intel_context_mark_active(struct intel_context *ce) > { > - lockdep_assert_held(&ce->timeline->mutex); > + lockdep_assert(lockdep_is_held(&ce->timeline->mutex) || > + test_bit(CONTEXT_IS_PARKED, &ce->flags)); > ++ce->active_count; > } > > --- a/drivers/gpu/drm/i915/gt/intel_context_types.h > +++ b/drivers/gpu/drm/i915/gt/intel_context_types.h > @@ -118,6 +118,7 @@ struct intel_context { > #define CONTEXT_LRCA_DIRTY 9 > #define CONTEXT_GUC_INIT 10 > #define CONTEXT_PERMA_PIN 11 > +#define CONTEXT_IS_PARKED 12 Pedantic comment is that semantics of the bit flag suggest it should be name CONTEXT_IS_PARKING, or along those lines. Since the flag gets clear as soon as the parking completes. I'd suggest doing that change for the self-documenting benefit. Regards, Tvrtko > > struct { > u64 timeout_us; > --- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c > +++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.c > @@ -80,39 +80,6 @@ static int __engine_unpark(struct intel_ > return 0; > } > > -#if IS_ENABLED(CONFIG_LOCKDEP) > - > -static unsigned long __timeline_mark_lock(struct intel_context *ce) > -{ > - unsigned long flags; > - > - local_irq_save(flags); > - mutex_acquire(&ce->timeline->mutex.dep_map, 2, 0, _THIS_IP_); > - > - return flags; > -} > - > -static void __timeline_mark_unlock(struct intel_context *ce, > - unsigned long flags) > -{ > - mutex_release(&ce->timeline->mutex.dep_map, _THIS_IP_); > - local_irq_restore(flags); > -} > - > -#else > - > -static unsigned long __timeline_mark_lock(struct intel_context *ce) > -{ > - return 0; > -} > - > -static void __timeline_mark_unlock(struct intel_context *ce, > - unsigned long flags) > -{ > -} > - > -#endif /* !IS_ENABLED(CONFIG_LOCKDEP) */ > - > static void duration(struct dma_fence *fence, struct dma_fence_cb *cb) > { > struct i915_request *rq = to_request(fence); > @@ -159,7 +126,6 @@ static bool switch_to_kernel_context(str > { > struct intel_context *ce = engine->kernel_context; > struct i915_request *rq; > - unsigned long flags; > bool result = true; > > /* > @@ -214,7 +180,7 @@ static bool switch_to_kernel_context(str > * engine->wakeref.count, we may see the request completion and retire > * it causing an underflow of the engine->wakeref. > */ > - flags = __timeline_mark_lock(ce); > + set_bit(CONTEXT_IS_PARKED, &ce->flags); > GEM_BUG_ON(atomic_read(&ce->timeline->active_count) < 0); > > rq = __i915_request_create(ce, GFP_NOWAIT); > @@ -246,7 +212,7 @@ static bool switch_to_kernel_context(str > > result = false; > out_unlock: > - __timeline_mark_unlock(ce, flags); > + clear_bit(CONTEXT_IS_PARKED, &ce->flags); > return result; > } > > --- a/drivers/gpu/drm/i915/i915_request.h > +++ b/drivers/gpu/drm/i915/i915_request.h > @@ -642,7 +642,8 @@ i915_request_timeline(const struct i915_ > { > /* Valid only while the request is being constructed (or retired). */ > return rcu_dereference_protected(rq->timeline, > - lockdep_is_held(&rcu_access_pointer(rq->timeline)->mutex)); > + lockdep_is_held(&rcu_access_pointer(rq->timeline)->mutex) || > + test_bit(CONTEXT_IS_PARKED, &rq->context->flags)); > } > > static inline struct i915_gem_context * >
--- a/drivers/gpu/drm/i915/gt/intel_context.h +++ b/drivers/gpu/drm/i915/gt/intel_context.h @@ -211,7 +211,8 @@ static inline void intel_context_enter(s static inline void intel_context_mark_active(struct intel_context *ce) { - lockdep_assert_held(&ce->timeline->mutex); + lockdep_assert(lockdep_is_held(&ce->timeline->mutex) || + test_bit(CONTEXT_IS_PARKED, &ce->flags)); ++ce->active_count; } --- a/drivers/gpu/drm/i915/gt/intel_context_types.h +++ b/drivers/gpu/drm/i915/gt/intel_context_types.h @@ -118,6 +118,7 @@ struct intel_context { #define CONTEXT_LRCA_DIRTY 9 #define CONTEXT_GUC_INIT 10 #define CONTEXT_PERMA_PIN 11 +#define CONTEXT_IS_PARKED 12 struct { u64 timeout_us; --- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c +++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.c @@ -80,39 +80,6 @@ static int __engine_unpark(struct intel_ return 0; } -#if IS_ENABLED(CONFIG_LOCKDEP) - -static unsigned long __timeline_mark_lock(struct intel_context *ce) -{ - unsigned long flags; - - local_irq_save(flags); - mutex_acquire(&ce->timeline->mutex.dep_map, 2, 0, _THIS_IP_); - - return flags; -} - -static void __timeline_mark_unlock(struct intel_context *ce, - unsigned long flags) -{ - mutex_release(&ce->timeline->mutex.dep_map, _THIS_IP_); - local_irq_restore(flags); -} - -#else - -static unsigned long __timeline_mark_lock(struct intel_context *ce) -{ - return 0; -} - -static void __timeline_mark_unlock(struct intel_context *ce, - unsigned long flags) -{ -} - -#endif /* !IS_ENABLED(CONFIG_LOCKDEP) */ - static void duration(struct dma_fence *fence, struct dma_fence_cb *cb) { struct i915_request *rq = to_request(fence); @@ -159,7 +126,6 @@ static bool switch_to_kernel_context(str { struct intel_context *ce = engine->kernel_context; struct i915_request *rq; - unsigned long flags; bool result = true; /* @@ -214,7 +180,7 @@ static bool switch_to_kernel_context(str * engine->wakeref.count, we may see the request completion and retire * it causing an underflow of the engine->wakeref. */ - flags = __timeline_mark_lock(ce); + set_bit(CONTEXT_IS_PARKED, &ce->flags); GEM_BUG_ON(atomic_read(&ce->timeline->active_count) < 0); rq = __i915_request_create(ce, GFP_NOWAIT); @@ -246,7 +212,7 @@ static bool switch_to_kernel_context(str result = false; out_unlock: - __timeline_mark_unlock(ce, flags); + clear_bit(CONTEXT_IS_PARKED, &ce->flags); return result; } --- a/drivers/gpu/drm/i915/i915_request.h +++ b/drivers/gpu/drm/i915/i915_request.h @@ -642,7 +642,8 @@ i915_request_timeline(const struct i915_ { /* Valid only while the request is being constructed (or retired). */ return rcu_dereference_protected(rq->timeline, - lockdep_is_held(&rcu_access_pointer(rq->timeline)->mutex)); + lockdep_is_held(&rcu_access_pointer(rq->timeline)->mutex) || + test_bit(CONTEXT_IS_PARKED, &rq->context->flags)); } static inline struct i915_gem_context *
This is a revert of commits d67739268cf0e ("drm/i915/gt: Mark up the nested engine-pm timeline lock as irqsafe") 6c69a45445af9 ("drm/i915/gt: Mark context->active_count as protected by timeline->mutex") The existing code leads to a different behaviour depending on whether lockdep is enabled or not. Any following lock that is acquired without disabling interrupts (but needs to) will not be noticed by lockdep. This it not just a lockdep annotation but is used but an actual mutex_t that is properly used as a lock but in case of __timeline_mark_lock() lockdep is only told that it is acquired but no lock has been acquired. It appears that its purpose is just satisfy the lockdep_assert_held() check in intel_context_mark_active(). The other problem with disabling interrupts is that on PREEMPT_RT interrupts are also disabled which leads to problems for instance later during memory allocation. Add a CONTEXT_IS_PARKED bit to intel_engine_cs and set_bit/clear_bit it instead of mutex_acquire/mutex_release. Use test_bit in the two identified spots which relied on the lockdep annotation. Cc: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> --- drivers/gpu/drm/i915/gt/intel_context.h | 3 +- drivers/gpu/drm/i915/gt/intel_context_types.h | 1 drivers/gpu/drm/i915/gt/intel_engine_pm.c | 38 +------------------------- drivers/gpu/drm/i915/i915_request.h | 3 +- 4 files changed, 7 insertions(+), 38 deletions(-)