2015-06-13 23:09:40

by Max Filippov

Subject: arc_pmu not initialized in the arc/kernel/perf_event.c?

Mischa, Vineet,

Looking at ARC PMU driver I've got an impression that its static arc_pmu
is left uninitialized, which possibly leads to Oops once registered
callbacks are invoked. I don't have neither ARC toolchain nor hardware to
test it, so I may be missing something. Below is the fix for it.

Thanks.
-- Max

---8<---
From: Max Filippov <[email protected]>
Subject: [PATCH] arc: fix use of uninitialized arc_pmu

static arc_pmu in the arch/arc/kernel/perf_event.c is not initialized as
it's shadowed by a local variable of the same name in the
arc_pmu_device_probe.

Signed-off-by: Max Filippov <[email protected]>
---
arch/arc/kernel/perf_event.c | 1 -
1 file changed, 1 deletion(-)

diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
index fd2ec50..57b58f5 100644
--- a/arch/arc/kernel/perf_event.c
+++ b/arch/arc/kernel/perf_event.c
@@ -266,7 +266,6 @@ static int arc_pmu_add(struct perf_event *event, int flags)

static int arc_pmu_device_probe(struct platform_device *pdev)
{
- struct arc_pmu *arc_pmu;
struct arc_reg_pct_build pct_bcr;
struct arc_reg_cc_build cc_bcr;
int i, j, ret;
--
1.8.1.4

2015-06-15 05:51:11

by Vineet Gupta

Subject: Re: arc_pmu not initialized in the arc/kernel/perf_event.c?

-CC Mischa (left Synopsys a while back)
+CC Alexey (does most fo perf work these days)

Hi Max,

On Sunday 14 June 2015 04:39 AM, Max Filippov wrote:
> Mischa, Vineet,
>
> Looking at ARC PMU driver I've got an impression that its static arc_pmu
> is left uninitialized, which possibly leads to Oops once registered
> callbacks are invoked. I don't have neither ARC toolchain nor hardware to
> test it, so I may be missing something. Below is the fix for it.
>
> Thanks.
> -- Max
>
> ---8<---
> From: Max Filippov <[email protected]>
> Subject: [PATCH] arc: fix use of uninitialized arc_pmu
>
> static arc_pmu in the arch/arc/kernel/perf_event.c is not initialized as
> it's shadowed by a local variable of the same name in the
> arc_pmu_device_probe.
>
> Signed-off-by: Max Filippov <[email protected]>

Thx for the fix. Indeed looks like a big SNAFU on my side. The reason we didn't
run into this despite pretty regular use of perf is that we have been using a
bunch of out of tree patches (recently floated for upstream) which removed the
offending code for other reasons. Neverthless I will add this to for-next and CC
stable as well.

Thx,
-Vineet

> ---
> arch/arc/kernel/perf_event.c | 1 -
> 1 file changed, 1 deletion(-)
>
> diff --git a/arch/arc/kernel/perf_event.c b/arch/arc/kernel/perf_event.c
> index fd2ec50..57b58f5 100644
> --- a/arch/arc/kernel/perf_event.c
> +++ b/arch/arc/kernel/perf_event.c
> @@ -266,7 +266,6 @@ static int arc_pmu_add(struct perf_event *event, int flags)
>
> static int arc_pmu_device_probe(struct platform_device *pdev)
> {
> - struct arc_pmu *arc_pmu;
> struct arc_reg_pct_build pct_bcr;
> struct arc_reg_cc_build cc_bcr;
> int i, j, ret;