[PATCH] powerpc, perf: Clear out branch entries to avoid any previous stale values

[PATCH] powerpc, perf: Clear out branch entries to avoid any previous stale values

[Anshuman Khandual]

The 'to' field inside branch entries might contain stale values from previous
PMU interrupt instances which had indirect branches. So clear all the values
before reading a fresh set of BHRB entries after a PMU interrupt.

Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
---
 arch/powerpc/perf/core-book3s.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
index c627843..09db68d 100644
--- a/arch/powerpc/perf/core-book3s.c
+++ b/arch/powerpc/perf/core-book3s.c
@@ -1590,6 +1590,8 @@ static void record_and_restart(struct perf_event *event, unsigned long val,
 		if (event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK) {
 			struct cpu_hw_events *cpuhw;
 			cpuhw = &__get_cpu_var(cpu_hw_events);
+			memset(cpuhw->bhrb_entries, 0,
+				sizeof(struct perf_branch_entry) * BHRB_MAX_ENTRIES);
 			power_pmu_bhrb_read(cpuhw);
 			data.br_stack = &cpuhw->bhrb_stack;
 		}
-- 
1.7.11.7

Re: [PATCH] powerpc, perf: Clear out branch entries to avoid any previous stale values

[Michael Ellerman]

On Mon, 2013-05-06 at 13:34 +0530, Anshuman Khandual wrote:
> The 'to' field inside branch entries might contain stale values from previous
> PMU interrupt instances which had indirect branches. So clear all the values
> before reading a fresh set of BHRB entries after a PMU interrupt.
> 
> Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
> ---
>  arch/powerpc/perf/core-book3s.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
> index c627843..09db68d 100644
> --- a/arch/powerpc/perf/core-book3s.c
> +++ b/arch/powerpc/perf/core-book3s.c
> @@ -1590,6 +1590,8 @@ static void record_and_restart(struct perf_event *event, unsigned long val,
>  		if (event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK) {
>  			struct cpu_hw_events *cpuhw;
>  			cpuhw = &__get_cpu_var(cpu_hw_events);
> +			memset(cpuhw->bhrb_entries, 0,
> +				sizeof(struct perf_branch_entry) * BHRB_MAX_ENTRIES);
>  			power_pmu_bhrb_read(cpuhw);
>  			data.br_stack = &cpuhw->bhrb_stack;
>  		}

Wouldn't it be just as effective, and less overhead, to set .to = 0; in
the else branch in power_pmu_bhrb_read() ?

eg:

diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
index c627843..30af11a4 100644
--- a/arch/powerpc/perf/core-book3s.c
+++ b/arch/powerpc/perf/core-book3s.c
@@ -1516,6 +1516,7 @@ void power_pmu_bhrb_read(struct cpu_hw_events *cpuhw)
                        } else {
                                /* Update address, flags for current entry */
                                cpuhw->bhrb_entries[u_index].from = addr;
+                               cpuhw->bhrb_entries[u_index].to = 0;
                                cpuhw->bhrb_entries[u_index].mispred = pred;
                                cpuhw->bhrb_entries[u_index].predicted = ~pred;
 

cheers

[PATCH] powerpc, perf: Clear out branch entries to avoid any previous stale values

[Anshuman Khandual]

The 'to' field inside branch entries might contain stale values from previous
PMU interrupt instances which had indirect branches. So clear all the values
before reading a fresh set of BHRB entries after a PMU interrupt.

Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
---
 arch/powerpc/perf/core-book3s.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
index c627843..09db68d 100644
--- a/arch/powerpc/perf/core-book3s.c
+++ b/arch/powerpc/perf/core-book3s.c
@@ -1590,6 +1590,8 @@ static void record_and_restart(struct perf_event *event, unsigned long val,
 		if (event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK) {
 			struct cpu_hw_events *cpuhw;
 			cpuhw = &__get_cpu_var(cpu_hw_events);
+			memset(cpuhw->bhrb_entries, 0,
+				sizeof(struct perf_branch_entry) * BHRB_MAX_ENTRIES);
 			power_pmu_bhrb_read(cpuhw);
 			data.br_stack = &cpuhw->bhrb_stack;
 		}
-- 
1.7.11.7