From: Johannes Weiner <hannes@cmpxchg.org>
To: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>,
Andrew Morton <akpm@linux-foundation.org>,
Linus Torvalds <torvalds@linux-foundation.org>,
Tejun Heo <tj@kernel.org>, Suren Baghdasaryan <surenb@google.com>,
Daniel Drake <drake@endlessm.com>,
Vinayak Menon <vinmenon@codeaurora.org>,
Christopher Lameter <cl@linux.com>,
Mike Galbraith <efault@gmx.de>,
Shakeel Butt <shakeelb@google.com>,
Peter Enderborg <peter.enderborg@sony.com>,
linux-mm@kvack.org, cgroups@vger.kernel.org,
linux-kernel@vger.kernel.org, kernel-team@fb.com
Subject: Re: [PATCH 8/9] psi: pressure stall information for CPU, memory, and IO
Date: Mon, 6 Aug 2018 11:19:28 -0400 [thread overview]
Message-ID: <20180806151928.GB9888@cmpxchg.org> (raw)
In-Reply-To: <20180803165641.GA2476@hirez.programming.kicks-ass.net>
On Fri, Aug 03, 2018 at 06:56:41PM +0200, Peter Zijlstra wrote:
> On Wed, Aug 01, 2018 at 11:19:57AM -0400, Johannes Weiner wrote:
> > +static bool psi_update_stats(struct psi_group *group)
> > +{
> > + u64 deltas[NR_PSI_STATES - 1] = { 0, };
> > + unsigned long missed_periods = 0;
> > + unsigned long nonidle_total = 0;
> > + u64 now, expires, period;
> > + int cpu;
> > + int s;
> > +
> > + mutex_lock(&group->stat_lock);
> > +
> > + /*
> > + * Collect the per-cpu time buckets and average them into a
> > + * single time sample that is normalized to wallclock time.
> > + *
> > + * For averaging, each CPU is weighted by its non-idle time in
> > + * the sampling period. This eliminates artifacts from uneven
> > + * loading, or even entirely idle CPUs.
> > + *
> > + * We don't need to synchronize against CPU hotplugging. If we
> > + * see a CPU that's online and has samples, we incorporate it.
> > + */
> > + for_each_online_cpu(cpu) {
> > + struct psi_group_cpu *groupc = per_cpu_ptr(group->pcpu, cpu);
> > + u32 uninitialized_var(nonidle);
>
> urgh.. I can see why the compiler got confused. Dodgy :-)
:-) I think we can make this cleaner. Something like this (modulo the
READ_ONCE/WRITE_ONCE you pointed out in the other email)?
diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
index abccfddba5d5..ce6f02ada1cd 100644
--- a/kernel/sched/psi.c
+++ b/kernel/sched/psi.c
@@ -220,6 +220,49 @@ static bool test_state(unsigned int *tasks, enum psi_states state)
}
}
+static u32 read_update_delta(struct psi_group_cpu *groupc,
+ enum psi_states state, int cpu)
+{
+ u32 time, delta;
+
+ time = READ_ONCE(groupc->times[state]);
+ /*
+ * In addition to already concluded states, we also
+ * incorporate currently active states on the CPU, since
+ * states may last for many sampling periods.
+ *
+ * This way we keep our delta sampling buckets small (u32) and
+ * our reported pressure close to what's actually happening.
+ */
+ if (test_state(groupc->tasks, state)) {
+ /*
+ * We can race with a state change and need to make
+ * sure the state_start update is ordered against the
+ * updates to the live state and the time buckets
+ * (groupc->times).
+ *
+ * 1. If we observe task state that needs to be
+ * recorded, make sure we see state_start from when
+ * that state went into effect or we'll count time
+ * from the previous state.
+ *
+ * 2. If the time delta has already been added to the
+ * bucket, make sure we don't see it in state_start or
+ * we'll count it twice.
+ *
+ * If the time delta is out of state_start but not in
+ * the time bucket yet, we'll miss it entirely and
+ * handle it in the next period.
+ */
+ smp_rmb();
+ time += cpu_clock(cpu) - groupc->state_start;
+ }
+ delta = time - groupc->times_prev[state];
+ groupc->times_prev[state] = time;
+
+ return delta;
+}
+
static bool psi_update_stats(struct psi_group *group)
{
u64 deltas[NR_PSI_STATES - 1] = { 0, };
@@ -244,60 +287,17 @@ static bool psi_update_stats(struct psi_group *group)
*/
for_each_online_cpu(cpu) {
struct psi_group_cpu *groupc = per_cpu_ptr(group->pcpu, cpu);
- u32 uninitialized_var(nonidle);
-
- BUILD_BUG_ON(PSI_NONIDLE != NR_PSI_STATES - 1);
-
- for (s = PSI_NONIDLE; s >= 0; s--) {
- u32 time, delta;
-
- time = READ_ONCE(groupc->times[s]);
- /*
- * In addition to already concluded states, we
- * also incorporate currently active states on
- * the CPU, since states may last for many
- * sampling periods.
- *
- * This way we keep our delta sampling buckets
- * small (u32) and our reported pressure close
- * to what's actually happening.
- */
- if (test_state(groupc->tasks, cpu, s)) {
- /*
- * We can race with a state change and
- * need to make sure the state_start
- * update is ordered against the
- * updates to the live state and the
- * time buckets (groupc->times).
- *
- * 1. If we observe task state that
- * needs to be recorded, make sure we
- * see state_start from when that
- * state went into effect or we'll
- * count time from the previous state.
- *
- * 2. If the time delta has already
- * been added to the bucket, make sure
- * we don't see it in state_start or
- * we'll count it twice.
- *
- * If the time delta is out of
- * state_start but not in the time
- * bucket yet, we'll miss it entirely
- * and handle it in the next period.
- */
- smp_rmb();
- time += cpu_clock(cpu) - groupc->state_start;
- }
- delta = time - groupc->times_prev[s];
- groupc->times_prev[s] = time;
-
- if (s == PSI_NONIDLE) {
- nonidle = nsecs_to_jiffies(delta);
- nonidle_total += nonidle;
- } else {
- deltas[s] += (u64)delta * nonidle;
- }
+ u32 nonidle;
+
+ nonidle = read_update_delta(groupc, PSI_NONIDLE, cpu);
+ nonidle = nsecs_to_jiffies(nonidle);
+ nonidle_total += nonidle;
+
+ for (s = 0; s < PSI_NONIDLE; s++) {
+ u32 delta;
+
+ delta = read_update_delta(groupc, s, cpu);
+ deltas[s] += (u64)delta * nonidle;
}
}
next prev parent reply other threads:[~2018-08-06 15:16 UTC|newest]
Thread overview: 41+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-08-01 15:19 [PATCH 0/9] psi: pressure stall information for CPU, memory, and IO v3 Johannes Weiner
2018-08-01 15:19 ` [PATCH 1/9] mm: workingset: don't drop refault information prematurely Johannes Weiner
2018-08-01 15:19 ` [PATCH 2/9] mm: workingset: tell cache transitions from workingset thrashing Johannes Weiner
2018-08-01 21:56 ` Suren Baghdasaryan
2018-08-02 12:28 ` Johannes Weiner
2018-08-01 15:19 ` [PATCH 3/9] delayacct: track delays from thrashing cache pages Johannes Weiner
2018-08-01 15:19 ` [PATCH 4/9] sched: loadavg: consolidate LOAD_INT, LOAD_FRAC, CALC_LOAD Johannes Weiner
2018-08-01 15:19 ` [PATCH 5/9] sched: loadavg: make calc_load_n() public Johannes Weiner
2018-08-01 15:19 ` [PATCH 6/9] sched: sched.h: make rq locking and clock functions available in stats.h Johannes Weiner
2018-08-01 15:19 ` [PATCH 7/9] sched: introduce this_rq_lock_irq() Johannes Weiner
2018-08-01 15:19 ` [PATCH 8/9] psi: pressure stall information for CPU, memory, and IO Johannes Weiner
2018-08-03 16:56 ` Peter Zijlstra
2018-08-06 15:05 ` Johannes Weiner
2018-08-06 15:25 ` Peter Zijlstra
2018-08-06 15:40 ` Johannes Weiner
2018-08-06 15:19 ` Johannes Weiner [this message]
2018-08-06 16:03 ` Peter Zijlstra
2018-08-21 19:44 ` Johannes Weiner
2018-08-22 9:16 ` Peter Zijlstra
2018-08-03 17:07 ` Peter Zijlstra
2018-08-06 15:23 ` Johannes Weiner
2018-08-03 17:15 ` Peter Zijlstra
2018-08-03 17:21 ` Peter Zijlstra
2018-08-21 20:11 ` Johannes Weiner
2018-08-22 9:10 ` Peter Zijlstra
2018-08-22 17:28 ` Johannes Weiner
2018-08-01 15:19 ` [PATCH 9/9] psi: cgroup support Johannes Weiner
2018-08-07 11:50 ` [PATCH 0/9] psi: pressure stall information for CPU, memory, and IO v3 peter enderborg
2018-08-07 17:51 ` Johannes Weiner
-- strict thread matches above, loose matches on Subject: below --
2018-08-28 17:22 [PATCH 0/9] psi: pressure stall information for CPU, memory, and IO v4 Johannes Weiner
2018-08-28 17:22 ` [PATCH 8/9] psi: pressure stall information for CPU, memory, and IO Johannes Weiner
2018-08-28 20:11 ` Randy Dunlap
2018-08-28 20:56 ` Johannes Weiner
2018-08-28 21:30 ` Randy Dunlap
2018-09-07 10:16 ` Peter Zijlstra
2018-09-07 10:21 ` Peter Zijlstra
2018-09-07 14:44 ` Johannes Weiner
2018-09-07 14:58 ` Peter Zijlstra
2018-09-07 17:50 ` Johannes Weiner
2018-09-07 10:24 ` Peter Zijlstra
2018-09-07 14:54 ` Johannes Weiner
2018-08-01 15:12 Johannes Weiner
2018-08-01 15:13 ` [PATCH 8/9] psi: pressure stall information for CPU, memory, and IO Johannes Weiner
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20180806151928.GB9888@cmpxchg.org \
--to=hannes@cmpxchg.org \
--cc=akpm@linux-foundation.org \
--cc=cgroups@vger.kernel.org \
--cc=cl@linux.com \
--cc=drake@endlessm.com \
--cc=efault@gmx.de \
--cc=kernel-team@fb.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mingo@redhat.com \
--cc=peter.enderborg@sony.com \
--cc=peterz@infradead.org \
--cc=shakeelb@google.com \
--cc=surenb@google.com \
--cc=tj@kernel.org \
--cc=torvalds@linux-foundation.org \
--cc=vinmenon@codeaurora.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).