[PATCH] powerpc: perf: Do not reset a counter value on stop/start

From: Heechul Yun
Date: Mon Aug 13 2012 - 08:01:41 EST


A performance counter of e500mc processor core generates an overflow
interrupt when the counter value reaches to 0x80000000. In order to
set up a sampling counter, the kernel sets the counter value to
0x80000000 - <samples> to cause an interrupt right after <samples>
are incremented.

In the current code, the problem arises when the kernel stops and
re-starts the sampling counter (fsl_emb_pmu_stop() and
fsl_emb_pmu_start() are called repsectively). When fsl_emb_pmu_stop()
is called, called when a user adjusts the sampling period, it currently
re-writes the counter value to zero, which breaks working of the
sampling counter. Similarily fsl_emb_pmu_start() re-write the counter
value.

This patch fixes the bug by not re-writing the counter unnecessarily.
Also, it sets the counter freeze bit in fsl_emb_pmu_stop(), so that
the counter does not increase until fsl_emb_pmu_start() is called
again.

Signed-off-by: Heechul Yun <heechul@xxxxxxxxxxxx>
---
arch/powerpc/perf/core-fsl-emb.c | 11 ++++++-----
1 files changed, 6 insertions(+), 5 deletions(-)

diff --git a/arch/powerpc/perf/core-fsl-emb.c b/arch/powerpc/perf/core-fsl-emb.c
index 106c533..4cd4f12 100644
--- a/arch/powerpc/perf/core-fsl-emb.c
+++ b/arch/powerpc/perf/core-fsl-emb.c
@@ -365,7 +365,6 @@ static void fsl_emb_pmu_del(struct perf_event *event, int flags)
static void fsl_emb_pmu_start(struct perf_event *event, int ef_flags)
{
unsigned long flags;
- s64 left;

if (event->hw.idx < 0 || !event->hw.sample_period)
return;
@@ -380,8 +379,9 @@ static void fsl_emb_pmu_start(struct perf_event *event, int ef_flags)
perf_pmu_disable(event->pmu);

event->hw.state = 0;
- left = local64_read(&event->hw.period_left);
- write_pmc(event->hw.idx, left);
+
+ /* start the counter. */
+ write_pmlca(event->hw.idx, event->hw.config_base);

perf_event_update_userpage(event);
perf_pmu_enable(event->pmu);
@@ -403,7 +403,9 @@ static void fsl_emb_pmu_stop(struct perf_event *event, int ef_flags)

fsl_emb_pmu_read(event);
event->hw.state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
- write_pmc(event->hw.idx, 0);
+
+ /* freeze the counter. */
+ write_pmlca(event->hw.idx, event->hw.config_base | PMLCA_FC);

perf_event_update_userpage(event);
perf_pmu_enable(event->pmu);
@@ -575,7 +577,6 @@ static void record_and_restart(struct perf_event *event, unsigned long val,
int record = 0;

if (event->hw.state & PERF_HES_STOPPED) {
- write_pmc(event->hw.idx, 0);
return;
}

--
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/