[PATCH v3] lib/raid6: Reduce high latency by using migrate instead of preempt

[PATCH v3] lib/raid6: Reduce high latency by using migrate instead of preempt

[Yajun Deng]

We found an abnormally high latency when executing modprobe raid6_pq, the
latency is greater than 1.2s when CONFIG_PREEMPT_VOLUNTARY=y, greater than
67ms when CONFIG_PREEMPT=y, and greater than 16ms when CONFIG_PREEMPT_RT=y.

How to reproduce:
 - Install cyclictest
     sudo apt install rt-tests
 - Run cyclictest example in one terminal
     sudo cyclictest -S -p 95 -d 0 -i 1000 -D 24h -m
 - Modprobe raid6_pq in another terminal
     sudo modprobe raid6_pq

This is caused by ksoftirqd fail to scheduled due to disable preemption,
this time is too long and unreasonable.

Reduce high latency by using migrate_disabl()/emigrate_enable() instead of
preempt_disable()/preempt_enable(), the latency won't greater than 100us.

This patch beneficial for CONFIG_PREEMPT=y or CONFIG_PREEMPT_RT=y, but no
effect for CONFIG_PREEMPT_VOLUNTARY=y.

Cc: stable@vger.kernel.org
Fixes: fe5cbc6e06c7 ("md/raid6 algorithms: delta syndrome functions")
Fixes: cc4589ebfae6 ("Rename raid6 files now they're in a 'raid6' directory.")
Link: https://lore.kernel.org/linux-raid/b06c5e3ef3413f12a2c2b2a241005af9@linux.dev/T/#t # v1
Signed-off-by: Yajun Deng <yajun.deng@linux.dev>
---
 lib/raid6/algos.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/lib/raid6/algos.c b/lib/raid6/algos.c
index 6d5e5000fdd7..21611d05c34c 100644
--- a/lib/raid6/algos.c
+++ b/lib/raid6/algos.c
@@ -162,7 +162,7 @@ static inline const struct raid6_calls *raid6_choose_gen(
 
 			perf = 0;
 
-			preempt_disable();
+			migrate_disable();
 			j0 = jiffies;
 			while ((j1 = jiffies) == j0)
 				cpu_relax();
@@ -171,7 +171,7 @@ static inline const struct raid6_calls *raid6_choose_gen(
 				(*algo)->gen_syndrome(disks, PAGE_SIZE, *dptrs);
 				perf++;
 			}
-			preempt_enable();
+			migrate_enable();
 
 			if (perf > bestgenperf) {
 				bestgenperf = perf;
@@ -186,7 +186,7 @@ static inline const struct raid6_calls *raid6_choose_gen(
 
 			perf = 0;
 
-			preempt_disable();
+			migrate_disable();
 			j0 = jiffies;
 			while ((j1 = jiffies) == j0)
 				cpu_relax();
@@ -196,7 +196,7 @@ static inline const struct raid6_calls *raid6_choose_gen(
 						      PAGE_SIZE, *dptrs);
 				perf++;
 			}
-			preempt_enable();
+			migrate_enable();
 
 			if (best == *algo)
 				bestxorperf = perf;
-- 
2.32.0

Re: [PATCH v3] lib/raid6: Reduce high latency by using migrate instead of preempt

[Sebastian Andrzej Siewior]

On 2021-12-17 10:16:10 [+0800], Yajun Deng wrote:
> We found an abnormally high latency when executing modprobe raid6_pq, the
> latency is greater than 1.2s when CONFIG_PREEMPT_VOLUNTARY=y, greater than
> 67ms when CONFIG_PREEMPT=y, and greater than 16ms when CONFIG_PREEMPT_RT=y.
> 
> How to reproduce:
>  - Install cyclictest
>      sudo apt install rt-tests
>  - Run cyclictest example in one terminal
>      sudo cyclictest -S -p 95 -d 0 -i 1000 -D 24h -m
>  - Modprobe raid6_pq in another terminal
>      sudo modprobe raid6_pq
> 
> This is caused by ksoftirqd fail to scheduled due to disable preemption,
> this time is too long and unreasonable.
> 
> Reduce high latency by using migrate_disabl()/emigrate_enable() instead of
> preempt_disable()/preempt_enable(), the latency won't greater than 100us.
> 
> This patch beneficial for CONFIG_PREEMPT=y or CONFIG_PREEMPT_RT=y, but no
> effect for CONFIG_PREEMPT_VOLUNTARY=y.

Why does it matter? This is only during boot-up/ module loading or do I
miss something? 
The delay is a jiffy so it depends on CONFIG_HZ. You do benchmark for
the best algorithm and if you get preempted during that period then your
results may be wrong and you make a bad selection.

You can either enable one algorithm and or disable
CONFIG_RAID6_PQ_BENCHMARK. I don't see the need for this patch not to
mention the stable tree.

Sebastian

Re: [PATCH v3] lib/raid6: Reduce high latency by using migrate instead of preempt

[Song Liu]

On Fri, Dec 17, 2021 at 5:42 AM Sebastian Andrzej Siewior
<bigeasy@linutronix.de> wrote:
>
> On 2021-12-17 10:16:10 [+0800], Yajun Deng wrote:
> > We found an abnormally high latency when executing modprobe raid6_pq, the
> > latency is greater than 1.2s when CONFIG_PREEMPT_VOLUNTARY=y, greater than
> > 67ms when CONFIG_PREEMPT=y, and greater than 16ms when CONFIG_PREEMPT_RT=y.
> >
> > How to reproduce:
> >  - Install cyclictest
> >      sudo apt install rt-tests
> >  - Run cyclictest example in one terminal
> >      sudo cyclictest -S -p 95 -d 0 -i 1000 -D 24h -m
> >  - Modprobe raid6_pq in another terminal
> >      sudo modprobe raid6_pq
> >
> > This is caused by ksoftirqd fail to scheduled due to disable preemption,
> > this time is too long and unreasonable.
> >
> > Reduce high latency by using migrate_disabl()/emigrate_enable() instead of
> > preempt_disable()/preempt_enable(), the latency won't greater than 100us.
> >
> > This patch beneficial for CONFIG_PREEMPT=y or CONFIG_PREEMPT_RT=y, but no
> > effect for CONFIG_PREEMPT_VOLUNTARY=y.
>
> Why does it matter? This is only during boot-up/ module loading or do I
> miss something?

Yes this only happens on boot-up and module loading.I don't know RT well
enough to tell whether latency during module loading is an issue.

> The delay is a jiffy so it depends on CONFIG_HZ. You do benchmark for
> the best algorithm and if you get preempted during that period then your
> results may be wrong and you make a bad selection.

With current code, the delay _should be_ 16 jiffies. However, the experiment
hits way longer latencies. I agree this may cause inaccurate benchmark results
and thus suboptimal RAID algorithm.

I guess the key question is whether long latency at module loading time matters.
If that doesn't matter, we should just drop this.

Thanks,
Song

Re: [PATCH v3] lib/raid6: Reduce high latency by using migrate instead of preempt

[Sebastian Andrzej Siewior]

On 2021-12-17 09:25:25 [-0800], Song Liu wrote:
> > The delay is a jiffy so it depends on CONFIG_HZ. You do benchmark for
> > the best algorithm and if you get preempted during that period then your
> > results may be wrong and you make a bad selection.
> 
> With current code, the delay _should be_ 16 jiffies. However, the experiment
> hits way longer latencies. I agree this may cause inaccurate benchmark results
> and thus suboptimal RAID algorithm.

Everything less than CONFIG_PREEMPT does not have an explicit
requirement for preemption so higher latencies are not unusual. *If*
this is a problem on <= PREEMPT_VOLUNTARY then a cond_resched() between
loops would be the usual thing to do. But only *if* it is a real problem
which I doubt. It is not a preemtible kernel after all…

> I guess the key question is whether long latency at module loading time matters.
> If that doesn't matter, we should just drop this.

Correct. And should this be problematic on PREEMPT_RT then I would
restrict CONFIG_RAID6_PQ_BENCHMARK to !PREEMPT_RT.

> Thanks,
> Song

Sebastian

Re: [PATCH v3] lib/raid6: Reduce high latency by using migrate instead of preempt

[Daniel Vacek]

On Fri, Dec 17, 2021 at 8:09 PM Sebastian Andrzej Siewior
<bigeasy@linutronix.de> wrote:
>
> On 2021-12-17 10:16:10 [+0800], Yajun Deng wrote:
> > We found an abnormally high latency when executing modprobe raid6_pq, the
> > latency is greater than 1.2s when CONFIG_PREEMPT_VOLUNTARY=y, greater than
> > 67ms when CONFIG_PREEMPT=y, and greater than 16ms when CONFIG_PREEMPT_RT=y.
> >
> > How to reproduce:
> >  - Install cyclictest
> >      sudo apt install rt-tests
> >  - Run cyclictest example in one terminal
> >      sudo cyclictest -S -p 95 -d 0 -i 1000 -D 24h -m
> >  - Modprobe raid6_pq in another terminal
> >      sudo modprobe raid6_pq
> >
> > This is caused by ksoftirqd fail to scheduled due to disable preemption,
> > this time is too long and unreasonable.
> >
> > Reduce high latency by using migrate_disabl()/emigrate_enable() instead of
> > preempt_disable()/preempt_enable(), the latency won't greater than 100us.
> >
> > This patch beneficial for CONFIG_PREEMPT=y or CONFIG_PREEMPT_RT=y, but no
> > effect for CONFIG_PREEMPT_VOLUNTARY=y.
>
> Why does it matter? This is only during boot-up/ module loading or do I
> miss something?
> The delay is a jiffy so it depends on CONFIG_HZ. You do benchmark for
> the best algorithm and if you get preempted during that period then your
> results may be wrong and you make a bad selection.
>
> You can either enable one algorithm and or disable
> CONFIG_RAID6_PQ_BENCHMARK. I don't see the need for this patch not to
> mention the stable tree.

Exactly. We should not touch this. I've just sent a verbose
explanation in the original report thread.

--nX

> Sebastian

Re: [PATCH v3] lib/raid6: Reduce high latency by using migrate instead of preempt

[Daniel Vacek]

On Fri, Dec 17, 2021 at 10:57 PM Song Liu <song@kernel.org> wrote:
>
> On Fri, Dec 17, 2021 at 5:42 AM Sebastian Andrzej Siewior
> <bigeasy@linutronix.de> wrote:
> >
> > On 2021-12-17 10:16:10 [+0800], Yajun Deng wrote:
> > > We found an abnormally high latency when executing modprobe raid6_pq, the
> > > latency is greater than 1.2s when CONFIG_PREEMPT_VOLUNTARY=y, greater than
> > > 67ms when CONFIG_PREEMPT=y, and greater than 16ms when CONFIG_PREEMPT_RT=y.
> > >
> > > How to reproduce:
> > >  - Install cyclictest
> > >      sudo apt install rt-tests
> > >  - Run cyclictest example in one terminal
> > >      sudo cyclictest -S -p 95 -d 0 -i 1000 -D 24h -m
> > >  - Modprobe raid6_pq in another terminal
> > >      sudo modprobe raid6_pq
> > >
> > > This is caused by ksoftirqd fail to scheduled due to disable preemption,
> > > this time is too long and unreasonable.
> > >
> > > Reduce high latency by using migrate_disabl()/emigrate_enable() instead of
> > > preempt_disable()/preempt_enable(), the latency won't greater than 100us.
> > >
> > > This patch beneficial for CONFIG_PREEMPT=y or CONFIG_PREEMPT_RT=y, but no
> > > effect for CONFIG_PREEMPT_VOLUNTARY=y.
> >
> > Why does it matter? This is only during boot-up/ module loading or do I
> > miss something?
>
> Yes this only happens on boot-up and module loading.I don't know RT well
> enough to tell whether latency during module loading is an issue.

Nope. It is not.

> > The delay is a jiffy so it depends on CONFIG_HZ. You do benchmark for
> > the best algorithm and if you get preempted during that period then your
> > results may be wrong and you make a bad selection.
>
> With current code, the delay _should be_ 16 jiffies. However, the experiment
> hits way longer latencies. I agree this may cause inaccurate benchmark results
> and thus suboptimal RAID algorithm.

I explained this in the original thread. All the observed latencies
are really expected.

> I guess the key question is whether long latency at module loading time matters.
> If that doesn't matter, we should just drop this.

Again, it does not matter at all and here it is rather desired by design.

Drop this, please.

--nX

> Thanks,
> Song