| Message ID | 20140512152208.GA12309@potion.brq.redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Mon, May 12, 2014 at 05:22:08PM +0200, Radim Kr?má? wrote: > 2014-05-07 11:01-0400, Waiman Long: > > From: Peter Zijlstra <peterz@infradead.org> > > > > Because the qspinlock needs to touch a second cacheline; add a pending > > bit and allow a single in-word spinner before we punt to the second > > cacheline. > > I think there is an unwanted scenario on virtual machines: > 1) VCPU sets the pending bit and start spinning. > 2) Pending VCPU gets descheduled. > - we have PLE and lock holder isn't running [1] > - the hypervisor randomly preempts us > 3) Lock holder unlocks while pending VCPU is waiting in queue. > 4) Subsequent lockers will see free lock with set pending bit and will > loop in trylock's 'for (;;)' > - the worst-case is lock starving [2] > - PLE can save us from wasting whole timeslice > > Retry threshold is the easiest solution, regardless of its ugliness [4]. > > Another minor design flaw is that formerly first VCPU gets appended to > the tail when it decides to queue; > is the performance gain worth it? This is all for real hardware, I've not yet stared at the (para)virt crap. My primary concern is that native hardware runs good and that the (para)virt support does rape the code -- so far its failing hard on the second. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 05/12/2014 11:22 AM, Radim Kr?má? wrote: > 2014-05-07 11:01-0400, Waiman Long: >> From: Peter Zijlstra<peterz@infradead.org> >> >> Because the qspinlock needs to touch a second cacheline; add a pending >> bit and allow a single in-word spinner before we punt to the second >> cacheline. > I think there is an unwanted scenario on virtual machines: > 1) VCPU sets the pending bit and start spinning. > 2) Pending VCPU gets descheduled. > - we have PLE and lock holder isn't running [1] > - the hypervisor randomly preempts us > 3) Lock holder unlocks while pending VCPU is waiting in queue. > 4) Subsequent lockers will see free lock with set pending bit and will > loop in trylock's 'for (;;)' > - the worst-case is lock starving [2] > - PLE can save us from wasting whole timeslice > > Retry threshold is the easiest solution, regardless of its ugliness [4]. Yes, that can be a real issue. Some sort of retry threshold, as you said, should be able to handle it. BTW, the relevant patch should be 16/19 where the PV spinlock stuff should be discussed. This patch is perfectly fine. > Another minor design flaw is that formerly first VCPU gets appended to > the tail when it decides to queue; > is the performance gain worth it? > > Thanks. Yes, the performance gain is worth it. The primary goal is to be not worse than ticket spinlock in light load situation which is the most common case. This feature is need to achieve that. -Longman -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
2014-05-13 15:47-0400, Waiman Long: > On 05/12/2014 11:22 AM, Radim Kr?má? wrote: > >I think there is an unwanted scenario on virtual machines: > >1) VCPU sets the pending bit and start spinning. > >2) Pending VCPU gets descheduled. > > - we have PLE and lock holder isn't running [1] > > - the hypervisor randomly preempts us > >3) Lock holder unlocks while pending VCPU is waiting in queue. > >4) Subsequent lockers will see free lock with set pending bit and will > > loop in trylock's 'for (;;)' > > - the worst-case is lock starving [2] > > - PLE can save us from wasting whole timeslice > > > >Retry threshold is the easiest solution, regardless of its ugliness [4]. > > Yes, that can be a real issue. Some sort of retry threshold, as you said, > should be able to handle it. > > BTW, the relevant patch should be 16/19 where the PV spinlock stuff should > be discussed. This patch is perfectly fine. Ouch, my apology to Peter didn't make it ... Agreed, I should have split the comment under patches [06/19] (part quoted above; does not depend on PV), [16/19] (part quoted below) and [17/19] (general doubts). > >Another minor design flaw is that formerly first VCPU gets appended to > >the tail when it decides to queue; > >is the performance gain worth it? > > > >Thanks. > > Yes, the performance gain is worth it. The primary goal is to be not worse > than ticket spinlock in light load situation which is the most common case. > This feature is need to achieve that. Ok. I've seen merit in pvqspinlock even with slightly slower first-waiter, so I would have happily sacrificed those horrible branches. (I prefer elegant to optimized code, but I can see why we want to be strictly better than ticketlock.) Peter mentioned that we are focusing on bare-metal patches, so I'll withold my other paravirt rants until they are polished. And to forcefully bring this thread a little bit on-topic: Pending-bit is effectively a lock in a lock, so I was wondering why don't we use more pending bits; advantages are the same, just diminished by the probability of having an ideally contended lock: - waiter won't be blocked on RAM access if critical section (or more) ends sooner - some unlucky cacheline is not forgotten - faster unlock (no need for tail operations) (- ?) disadvantages are magnified: - increased complexity - intense cacheline sharing (I thought that this is the main disadvantage of ticketlock.) (- ?) One bit still improved performance, is it the best we got? Thanks. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, May 14, 2014 at 06:51:24PM +0200, Radim Kr?má? wrote: > Ok. > I've seen merit in pvqspinlock even with slightly slower first-waiter, > so I would have happily sacrificed those horrible branches. > (I prefer elegant to optimized code, but I can see why we want to be > strictly better than ticketlock.) > Peter mentioned that we are focusing on bare-metal patches, so I'll > withold my other paravirt rants until they are polished. Well, paravirt must happen too, but comes later in this series, patch 3 which we're replying to is still very much in the bare metal part of the series. I've not had time yet to decode all that Waiman has done to make paravirt work. But as a general rule I like patches that start with something simple and working and then optimize it, this series doesn't seem to quite grasp that. > And to forcefully bring this thread a little bit on-topic: > > Pending-bit is effectively a lock in a lock, so I was wondering why > don't we use more pending bits; advantages are the same, just diminished > by the probability of having an ideally contended lock: > - waiter won't be blocked on RAM access if critical section (or more) > ends sooner > - some unlucky cacheline is not forgotten > - faster unlock (no need for tail operations) > (- ?) > disadvantages are magnified: > - increased complexity > - intense cacheline sharing > (I thought that this is the main disadvantage of ticketlock.) > (- ?) > > One bit still improved performance, is it the best we got? So, the advantage of one bit is that if we use a whole byte for 1 bit we can avoid some atomic ops. The entire reason for this in-word spinner is to amortize the cost of hitting the external node cacheline. So traditional locks like test-and-test and the ticket lock only ever access the spinlock word itsef, this MCS style queueing lock has a second (and, see my other rants in this thread, when done wrong more than 2) cacheline to touch. That said, all our benchmarking is pretty much for the cache-hot case, so I'm not entirely convinced yet that the one pending bit makes up for it, it does in the cache-hot case. But... writing cache-cold benchmarks is _hard_ :/
2014-05-14 19:00+0200, Peter Zijlstra: > On Wed, May 14, 2014 at 06:51:24PM +0200, Radim Kr?má? wrote: > > Ok. > > I've seen merit in pvqspinlock even with slightly slower first-waiter, > > so I would have happily sacrificed those horrible branches. > > (I prefer elegant to optimized code, but I can see why we want to be > > strictly better than ticketlock.) > > Peter mentioned that we are focusing on bare-metal patches, so I'll > > withold my other paravirt rants until they are polished. (It was an ambiguous sentence, I have comments for later patches.) > Well, paravirt must happen too, but comes later in this series, patch 3 > which we're replying to is still very much in the bare metal part of the > series. (I think that bare metal spans the first 7 patches.) > I've not had time yet to decode all that Waiman has done to make > paravirt work. > > But as a general rule I like patches that start with something simple > and working and then optimize it, this series doesn't seem to quite > grasp that. > > > And to forcefully bring this thread a little bit on-topic: > > > > Pending-bit is effectively a lock in a lock, so I was wondering why > > don't we use more pending bits; advantages are the same, just diminished > > by the probability of having an ideally contended lock: > > - waiter won't be blocked on RAM access if critical section (or more) > > ends sooner > > - some unlucky cacheline is not forgotten > > - faster unlock (no need for tail operations) > > (- ?) > > disadvantages are magnified: > > - increased complexity > > - intense cacheline sharing > > (I thought that this is the main disadvantage of ticketlock.) > > (- ?) > > > > One bit still improved performance, is it the best we got? > > So, the advantage of one bit is that if we use a whole byte for 1 bit we > can avoid some atomic ops. > > The entire reason for this in-word spinner is to amortize the cost of > hitting the external node cacheline. Every pending CPU removes one length of the critical section from the delay caused by cacheline propagation and really cold cache is hundreds(?) of cycles, so we could burn some to ensure correctness and still be waiting when the first pending CPU unlocks. > So traditional locks like test-and-test and the ticket lock only ever > access the spinlock word itsef, this MCS style queueing lock has a > second (and, see my other rants in this thread, when done wrong more > than 2) cacheline to touch. > > That said, all our benchmarking is pretty much for the cache-hot case, > so I'm not entirely convinced yet that the one pending bit makes up for > it, it does in the cache-hot case. Yeah, we probably use the faster pre-lock quite a lot. Cover letter states that queue depth 1-3 is a bit slower than ticket spinlock, so we might not be losing if we implemented a faster in-word-lock of this capacity. (Not that I'm a fan of the hybrid lock.) > But... writing cache-cold benchmarks is _hard_ :/ Wouldn't clflush of the second cacheline before trying for the lock give us a rough estimate? -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 05/14/2014 03:13 PM, Radim Kr?má? wrote: > 2014-05-14 19:00+0200, Peter Zijlstra: >> On Wed, May 14, 2014 at 06:51:24PM +0200, Radim Kr?má? wrote: >>> Ok. >>> I've seen merit in pvqspinlock even with slightly slower first-waiter, >>> so I would have happily sacrificed those horrible branches. >>> (I prefer elegant to optimized code, but I can see why we want to be >>> strictly better than ticketlock.) >>> Peter mentioned that we are focusing on bare-metal patches, so I'll >>> withold my other paravirt rants until they are polished. > (It was an ambiguous sentence, I have comments for later patches.) > >> Well, paravirt must happen too, but comes later in this series, patch 3 >> which we're replying to is still very much in the bare metal part of the >> series. > (I think that bare metal spans the first 7 patches.) > >> I've not had time yet to decode all that Waiman has done to make >> paravirt work. >> >> But as a general rule I like patches that start with something simple >> and working and then optimize it, this series doesn't seem to quite >> grasp that. >> >>> And to forcefully bring this thread a little bit on-topic: >>> >>> Pending-bit is effectively a lock in a lock, so I was wondering why >>> don't we use more pending bits; advantages are the same, just diminished >>> by the probability of having an ideally contended lock: >>> - waiter won't be blocked on RAM access if critical section (or more) >>> ends sooner >>> - some unlucky cacheline is not forgotten >>> - faster unlock (no need for tail operations) >>> (- ?) >>> disadvantages are magnified: >>> - increased complexity >>> - intense cacheline sharing >>> (I thought that this is the main disadvantage of ticketlock.) >>> (- ?) >>> >>> One bit still improved performance, is it the best we got? >> So, the advantage of one bit is that if we use a whole byte for 1 bit we >> can avoid some atomic ops. >> >> The entire reason for this in-word spinner is to amortize the cost of >> hitting the external node cacheline. > Every pending CPU removes one length of the critical section from the > delay caused by cacheline propagation and really cold cache is > hundreds(?) of cycles, so we could burn some to ensure correctness and > still be waiting when the first pending CPU unlocks. Assuming that taking a spinlock is fairly frequent in the kernel, the node structure cacheline won't be so cold after all. >> So traditional locks like test-and-test and the ticket lock only ever >> access the spinlock word itsef, this MCS style queueing lock has a >> second (and, see my other rants in this thread, when done wrong more >> than 2) cacheline to touch. >> >> That said, all our benchmarking is pretty much for the cache-hot case, >> so I'm not entirely convinced yet that the one pending bit makes up for >> it, it does in the cache-hot case. > Yeah, we probably use the faster pre-lock quite a lot. > Cover letter states that queue depth 1-3 is a bit slower than ticket > spinlock, so we might not be losing if we implemented a faster > in-word-lock of this capacity. (Not that I'm a fan of the hybrid lock.) I had tried an experimental patch with 2 pending bits. However, the benchmark test that I used show the performance is even worse than without any pending bit. I probably need to revisit that later as to why this is the case. As for now, I will focus on just having one pending bit. If we could find a way to get better performance out of more than 1 pending bit later on, we could always submit another patch to do that. >> But... writing cache-cold benchmarks is _hard_ :/ > Wouldn't clflush of the second cacheline before trying for the lock give > us a rough estimate? clflush is a very expensive operation and I doubt that it will be indicative of real life performance at all. BTW, there is no way to write a cache-cold benchmark for that 2nd cacheline as any call to spin_lock will likely to access it if there is enough contention. -Longman -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c index 37b5c7f..cd45c27 100644 --- a/kernel/locking/qspinlock.c +++ b/kernel/locking/qspinlock.c @@ -573,7 +573,7 @@ static __always_inline int get_qlock(struct qspinlock *lock) static inline int trylock_pending(struct qspinlock *lock, u32 *pval) { u32 old, new, val = *pval; - int retry = 1; + int retry = 0; /* * trylock || pending @@ -595,9 +595,9 @@ static inline int trylock_pending(struct qspinlock *lock, u32 *pval) * a while to see if that either bit will be cleared. * If that is no change, we return and be queued. */ - if (!retry) + if (retry) return 0; - retry--; + retry++; cpu_relax(); cpu_relax(); *pval = val = atomic_read(&lock->val); @@ -608,7 +608,11 @@ static inline int trylock_pending(struct qspinlock *lock, u32 *pval) * Assuming that the pending bit holder is going to * set the lock bit and clear the pending bit soon, * it is better to wait than to exit at this point. + * Our assumption does not hold on hypervisors, where + * the pending bit holder doesn't have to be running. */ + if (cpu_has_hypervisor && ++retry > MAGIC) + return 0; cpu_relax(); *pval = val = atomic_read(&lock->val); continue;