This is a note to let you know that I've just added the patch titled
powerpc/perf_event: Skip updating kernel counters if register value shrinks
to the 2.6.38-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
powerpc-perf_event-skip-updating-kernel-counters-if-register-value-shrinks.patch
and it can be found in the queue-2.6.38 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <[email protected]> know about it.
>From 86c74ab317c1ef4d37325e0d7ca8a01a796b0bd7 Mon Sep 17 00:00:00 2001
From: Eric B Munson <[email protected]>
Date: Fri, 15 Apr 2011 08:12:30 +0000
Subject: powerpc/perf_event: Skip updating kernel counters if register value
shrinks
From: Eric B Munson <[email protected]>
commit 86c74ab317c1ef4d37325e0d7ca8a01a796b0bd7 upstream.
Because of speculative event roll back, it is possible for some event coutners
to decrease between reads on POWER7. This causes a problem with the way that
counters are updated. Delta calues are calculated in a 64 bit value and the
top 32 bits are masked. If the register value has decreased, this leaves us
with a very large positive value added to the kernel counters. This patch
protects against this by skipping the update if the delta would be negative.
This can lead to a lack of precision in the coutner values, but from my testing
the value is typcially fewer than 10 samples at a time.
Signed-off-by: Eric B Munson <[email protected]>
Signed-off-by: Benjamin Herrenschmidt <[email protected]>
Signed-off-by: Greg Kroah-Hartman <[email protected]>
---
arch/powerpc/kernel/perf_event.c | 37 ++++++++++++++++++++++++++++++-------
1 file changed, 30 insertions(+), 7 deletions(-)
--- a/arch/powerpc/kernel/perf_event.c
+++ b/arch/powerpc/kernel/perf_event.c
@@ -398,6 +398,25 @@ static int check_excludes(struct perf_ev
return 0;
}
+static u64 check_and_compute_delta(u64 prev, u64 val)
+{
+ u64 delta = (val - prev) & 0xfffffffful;
+
+ /*
+ * POWER7 can roll back counter values, if the new value is smaller
+ * than the previous value it will cause the delta and the counter to
+ * have bogus values unless we rolled a counter over. If a coutner is
+ * rolled back, it will be smaller, but within 256, which is the maximum
+ * number of events to rollback at once. If we dectect a rollback
+ * return 0. This can lead to a small lack of precision in the
+ * counters.
+ */
+ if (prev > val && (prev - val) < 256)
+ delta = 0;
+
+ return delta;
+}
+
static void power_pmu_read(struct perf_event *event)
{
s64 val, delta, prev;
@@ -416,10 +435,11 @@ static void power_pmu_read(struct perf_e
prev = local64_read(&event->hw.prev_count);
barrier();
val = read_pmc(event->hw.idx);
+ delta = check_and_compute_delta(prev, val);
+ if (!delta)
+ return;
} while (local64_cmpxchg(&event->hw.prev_count, prev, val) != prev);
- /* The counters are only 32 bits wide */
- delta = (val - prev) & 0xfffffffful;
local64_add(delta, &event->count);
local64_sub(delta, &event->hw.period_left);
}
@@ -449,8 +469,9 @@ static void freeze_limited_counters(stru
val = (event->hw.idx == 5) ? pmc5 : pmc6;
prev = local64_read(&event->hw.prev_count);
event->hw.idx = 0;
- delta = (val - prev) & 0xfffffffful;
- local64_add(delta, &event->count);
+ delta = check_and_compute_delta(prev, val);
+ if (delta)
+ local64_add(delta, &event->count);
}
}
@@ -458,14 +479,16 @@ static void thaw_limited_counters(struct
unsigned long pmc5, unsigned long pmc6)
{
struct perf_event *event;
- u64 val;
+ u64 val, prev;
int i;
for (i = 0; i < cpuhw->n_limited; ++i) {
event = cpuhw->limited_counter[i];
event->hw.idx = cpuhw->limited_hwidx[i];
val = (event->hw.idx == 5) ? pmc5 : pmc6;
- local64_set(&event->hw.prev_count, val);
+ prev = local64_read(&event->hw.prev_count);
+ if (check_and_compute_delta(prev, val))
+ local64_set(&event->hw.prev_count, val);
perf_event_update_userpage(event);
}
}
@@ -1197,7 +1220,7 @@ static void record_and_restart(struct pe
/* we don't have to worry about interrupts here */
prev = local64_read(&event->hw.prev_count);
- delta = (val - prev) & 0xfffffffful;
+ delta = check_and_compute_delta(prev, val);
local64_add(delta, &event->count);
/*
Patches currently in stable-queue which might be from [email protected] are
queue-2.6.38/powerpc-perf_event-skip-updating-kernel-counters-if-register-value-shrinks.patch
_______________________________________________
stable mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/stable
