2012-08-18 16:15:42

by Andi Kleen

Subject: [PATCH 1/2] pebs, x86: Make sure to really disable PEBS

From: Andi Kleen <[email protected]>

When disabling a PEBS event we need to check the old value
to determine if PEBS was running. Otherwise the last PEBS value
could be kept running.

Signed-off-by: Andi Kleen <[email protected]>
---
arch/x86/kernel/cpu/perf_event_intel_ds.c | 3 ++-
1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c b/arch/x86/kernel/cpu/perf_event_intel_ds.c
index e38d97b..db49ab3 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_ds.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c
@@ -438,9 +438,10 @@ void intel_pmu_pebs_disable(struct perf_event *event)
{
struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
struct hw_perf_event *hwc = &event->hw;
+ u64 old = cpuc->pebs_enabled;

cpuc->pebs_enabled &= ~(1ULL << hwc->idx);
- if (cpuc->enabled)
+ if (old)
wrmsrl(MSR_IA32_PEBS_ENABLE, cpuc->pebs_enabled);

hwc->config |= ARCH_PERFMON_EVENTSEL_INT;
--
1.7.7.6

2012-08-18 16:16:35

by Andi Kleen

Subject: [PATCH 2/2] perf: Enable function tracing for perf core

From: Andi Kleen <[email protected]>

According to Steven R. there is no reason left to not support
function tracing for the perf core. This makes it easier to debug
perf.

Don't remove -pg for the x86 and generic perf core.

Cc: [email protected]
Signed-off-by: Andi Kleen <[email protected]>
---
arch/x86/kernel/cpu/Makefile | 1 -
kernel/events/Makefile | 4 ----
2 files changed, 0 insertions(+), 5 deletions(-)

diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index d30a6a9..6a16725 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -5,7 +5,6 @@
# Don't trace early stages of a secondary CPU boot
ifdef CONFIG_FUNCTION_TRACER
CFLAGS_REMOVE_common.o = -pg
-CFLAGS_REMOVE_perf_event.o = -pg
endif

# Make sure load_percpu_segment has no stackprotector
diff --git a/kernel/events/Makefile b/kernel/events/Makefile
index 103f5d1..a630994 100644
--- a/kernel/events/Makefile
+++ b/kernel/events/Makefile
@@ -1,7 +1,3 @@
-ifdef CONFIG_FUNCTION_TRACER
-CFLAGS_REMOVE_core.o = -pg
-endif
-
obj-y := core.o ring_buffer.o callchain.o

obj-$(CONFIG_HAVE_HW_BREAKPOINT) += hw_breakpoint.o
--
1.7.7.6

2012-08-20 10:10:25

by Peter Zijlstra

Subject: Re: [PATCH 1/2] pebs, x86: Make sure to really disable PEBS

On Sat, 2012-08-18 at 09:15 -0700, Andi Kleen wrote:
> From: Andi Kleen <[email protected]>
>
> When disabling a PEBS event we need to check the old value
> to determine if PEBS was running. Otherwise the last PEBS value
> could be kept running.
>
> Signed-off-by: Andi Kleen <[email protected]>
> ---
> arch/x86/kernel/cpu/perf_event_intel_ds.c | 3 ++-
> 1 files changed, 2 insertions(+), 1 deletions(-)
>
> diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c b/arch/x86/kernel/cpu/perf_event_intel_ds.c
> index e38d97b..db49ab3 100644
> --- a/arch/x86/kernel/cpu/perf_event_intel_ds.c
> +++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c
> @@ -438,9 +438,10 @@ void intel_pmu_pebs_disable(struct perf_event *event)
> {
> struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
> struct hw_perf_event *hwc = &event->hw;
> + u64 old = cpuc->pebs_enabled;
>
> cpuc->pebs_enabled &= ~(1ULL << hwc->idx);
> - if (cpuc->enabled)
> + if (old)

cpuc->enabled != cpuc->pebs_enabled

> wrmsrl(MSR_IA32_PEBS_ENABLE, cpuc->pebs_enabled);
>
> hwc->config |= ARCH_PERFMON_EVENTSEL_INT;

2012-08-21 15:13:40

by Steven Rostedt

Subject: Re: [PATCH 2/2] perf: Enable function tracing for perf core

On Sat, 2012-08-18 at 09:15 -0700, Andi Kleen wrote:
> From: Andi Kleen <[email protected]>
>
> According to Steven R. there is no reason left to not support
> function tracing for the perf core. This makes it easier to debug
> perf.
>
> Don't remove -pg for the x86 and generic perf core.

There's some new patches queued for 3.7 that has better function tracing
recursion protection. This should go after those.

Thanks,

-- Steve