On Thu, Jul 02, 2015 at 08:35:57PM +0200, Ingo Molnar wrote:
>
> * Paul E. McKenney <paul...@linux.vnet.ibm.com> wrote:
>
> > > And it's not like it's that hard to stem the flow of algorithmic
> > > sloppiness at
> > > the source, right?
> >
> > OK, first let me make sure that I understand what you are asking for:
> >
> > 1. Completely eliminate synchronize_rcu_expedited() and
> > synchronize_sched_expedited(), replacing all uses with their
> > unexpedited counterparts. (Note that synchronize_srcu_expedited()
> > does not wake up CPUs, courtesy of its read-side memory barriers.)
> > The fast-boot guys are probably going to complain, along with
> > the networking guys.
> >
> > 2. Keep synchronize_rcu_expedited() and synchronize_sched_expedited(),
> > but push back hard on any new uses and question any existing uses.
> >
> > 3. Revert 74b51ee152b6 ("ACPI / osl: speedup grace period in
> > acpi_os_map_cleanup").
> >
> > 4. Something else?
>
> I'd love to have 1) but 2) would be a realistic second best option? ;-)
OK, how about the following checkpatch.pl patch?
And here are some other actions I have taken and will take to improve
the situation, both for OS jitter and for scalability:
o Reduce OS jitter by switching from try_stop_cpus() to
stop_one_cpu_nowait(), courtesy of Peter Zijlstra.
I expect to send this in v4.3 or v4.4, depending on how
testing and review goes.
o Eliminate expedited-grace-period-induced OS jitter on idle CPUs.
This went into v3.19. Note that this also avoids IPIing
nohz_full CPUs.
o I believe that I can reduce OS jitter by a further factor of two
(worst case) or factor of five (average case), but I am still
thinking about exactly how to do this. (This also has the
benefit of shutting up a lockdep false positive.)
o There is a global counter that synchronize_sched_expedited()
uses to determine when all the CPUs have passed through a
quiescent state. This is a scalability bottleneck on modest
systems under heavy load, so I will be switching this to
instead use the combining tree.
o Because both synchronize_sched_expedited() and
synchronize_rcu_expedited() can potentially wake up each and
every CPU, on sufficiently large systems, they are quite slow.
If this scalability problem ever becomes real, I intend to use
multiple kthreads to do the wakeups on large systems.
Seem reasonable?
Thanx, Paul
------------------------------------------------------------------------
scripts: Make checkpatch.pl warn on expedited RCU grace periods
The synchronize_rcu_expedited() and synchronize_sched_expedited()
expedited-grace-period primitives induce OS jitter, which can degrade
real-time response. This commit therefore adds a checkpatch.pl warning
on any patch adding them.
Note that this patch does not warn on synchronize_srcu_expedited()
because it does not induce OS jitter, courtesy of its otherwise
much-maligned read-side memory barriers.
Signed-off-by: Paul E. McKenney <paul...@linux.vnet.ibm.com>
Cc: Andy Whitcroft <a...@canonical.com>
Cc: Joe Perches <j...@perches.com>
diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
index 89b1df4e72ab..ddd82d743bba 100755
--- a/scripts/checkpatch.pl
+++ b/scripts/checkpatch.pl
@@ -4898,6 +4898,12 @@ sub process {
"memory barrier without comment\n" .
$herecurr);
}
}
+# check for expedited grace periods that interrupt CPUs.
+# note that synchronize_srcu_expedited() does -not- do this, so no complaints.
+ if ($line =~
/\b(synchronize_rcu_expedited|synchronize_sched_expedited)\(/) {
+ WARN("EXPEDITED_RCU_GRACE_PERIOD",
+ "expedited RCU grace periods should be avoided\n"
. $herecurr);
+ }
# check of hardware specific defines
if ($line =~
m@^.\s*\#\s*if.*\b(__i386__|__powerpc64__|__sun__|__s390x__)\b@ && $realfile !~
m@include/asm-@) {
CHK("ARCH_DEFINES",
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
