lkml.org 
[lkml]   [2015]   [Oct]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH 5/5] x86, perf: Avoid context switching LBR_INFO when not needed
Date
From: Andi Kleen <ak@linux.intel.com>

We context switch LBRs in call stack mode. Currently LBR_INFO
was also context switched, but we normally don't need that
in call stack mode.

Make the context switch check the NO_CYCLES|NO_FLAGS event
flags that were earlier added, and if set avoid writing
the LBR_INFO MSRs unnecessarily.

The same is done for the LBR reset code.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
---
arch/x86/kernel/cpu/perf_event.h | 3 ++-
arch/x86/kernel/cpu/perf_event_intel.c | 2 +-
arch/x86/kernel/cpu/perf_event_intel_lbr.c | 25 ++++++++++++++++---------
3 files changed, 19 insertions(+), 11 deletions(-)

diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h
index 1b47164..4ae66e3 100644
--- a/arch/x86/kernel/cpu/perf_event.h
+++ b/arch/x86/kernel/cpu/perf_event.h
@@ -634,6 +634,7 @@ struct x86_perf_task_context {
int tos;
int lbr_callstack_users;
int lbr_stack_state;
+ int need_info;
};

#define x86_add_quirk(func_) \
@@ -887,7 +888,7 @@ void intel_ds_init(void);

void intel_pmu_lbr_sched_task(struct perf_event_context *ctx, bool sched_in);

-void intel_pmu_lbr_reset(void);
+void intel_pmu_lbr_reset(bool need_info);

void intel_pmu_lbr_enable(struct perf_event *event);

diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
index f17772a..42f21f0 100644
--- a/arch/x86/kernel/cpu/perf_event_intel.c
+++ b/arch/x86/kernel/cpu/perf_event_intel.c
@@ -2844,7 +2844,7 @@ static void intel_pmu_cpu_starting(int cpu)
/*
* Deal with CPUs that don't clear their LBRs on power-up.
*/
- intel_pmu_lbr_reset();
+ intel_pmu_lbr_reset(1);

cpuc->lbr_sel = NULL;

diff --git a/arch/x86/kernel/cpu/perf_event_intel_lbr.c b/arch/x86/kernel/cpu/perf_event_intel_lbr.c
index 60e71b7..7c21efb 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_lbr.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_lbr.c
@@ -195,27 +195,30 @@ static void intel_pmu_lbr_reset_32(void)
wrmsrl(x86_pmu.lbr_from + i, 0);
}

-static void intel_pmu_lbr_reset_64(void)
+static void intel_pmu_lbr_reset_64(bool need_info)
{
int i;

for (i = 0; i < x86_pmu.lbr_nr; i++) {
wrmsrl(x86_pmu.lbr_from + i, 0);
wrmsrl(x86_pmu.lbr_to + i, 0);
- if (x86_pmu.intel_cap.lbr_format == LBR_FORMAT_INFO)
+ if (need_info)
wrmsrl(MSR_LBR_INFO_0 + i, 0);
}
}

-void intel_pmu_lbr_reset(void)
+void intel_pmu_lbr_reset(bool need_info)
{
if (!x86_pmu.lbr_nr)
return;

+ if (x86_pmu.intel_cap.lbr_format != LBR_FORMAT_INFO)
+ need_info = false;
+
if (x86_pmu.intel_cap.lbr_format == LBR_FORMAT_32)
intel_pmu_lbr_reset_32();
else
- intel_pmu_lbr_reset_64();
+ intel_pmu_lbr_reset_64(need_info);
}

/*
@@ -242,7 +245,7 @@ static void __intel_pmu_lbr_restore(struct x86_perf_task_context *task_ctx)

if (task_ctx->lbr_callstack_users == 0 ||
task_ctx->lbr_stack_state == LBR_NONE) {
- intel_pmu_lbr_reset();
+ intel_pmu_lbr_reset(task_ctx->need_info > 0);
return;
}

@@ -252,7 +255,7 @@ static void __intel_pmu_lbr_restore(struct x86_perf_task_context *task_ctx)
lbr_idx = (tos - i) & mask;
wrmsrl(x86_pmu.lbr_from + lbr_idx, task_ctx->lbr_from[i]);
wrmsrl(x86_pmu.lbr_to + lbr_idx, task_ctx->lbr_to[i]);
- if (x86_pmu.intel_cap.lbr_format == LBR_FORMAT_INFO)
+ if (task_ctx->need_info)
wrmsrl(MSR_LBR_INFO_0 + lbr_idx, task_ctx->lbr_info[i]);
}
wrmsrl(x86_pmu.lbr_tos, tos);
@@ -276,7 +279,7 @@ static void __intel_pmu_lbr_save(struct x86_perf_task_context *task_ctx)
lbr_idx = (tos - i) & mask;
rdmsrl(x86_pmu.lbr_from + lbr_idx, task_ctx->lbr_from[i]);
rdmsrl(x86_pmu.lbr_to + lbr_idx, task_ctx->lbr_to[i]);
- if (x86_pmu.intel_cap.lbr_format == LBR_FORMAT_INFO)
+ if (task_ctx->need_info)
rdmsrl(MSR_LBR_INFO_0 + lbr_idx, task_ctx->lbr_info[i]);
}
task_ctx->tos = tos;
@@ -317,7 +320,7 @@ void intel_pmu_lbr_sched_task(struct perf_event_context *ctx, bool sched_in)
* stack with branch from multiple tasks.
*/
if (sched_in) {
- intel_pmu_lbr_reset();
+ intel_pmu_lbr_reset(!task_ctx || task_ctx->need_info > 0);
cpuc->lbr_context = ctx;
}
}
@@ -340,7 +343,7 @@ void intel_pmu_lbr_enable(struct perf_event *event)
* avoid data leaks.
*/
if (event->ctx->task && cpuc->lbr_context != event->ctx) {
- intel_pmu_lbr_reset();
+ intel_pmu_lbr_reset(!(event->hw.branch_reg.reg & LBR_NO_INFO));
cpuc->lbr_context = event->ctx;
}
cpuc->br_sel = event->hw.branch_reg.reg;
@@ -349,6 +352,8 @@ void intel_pmu_lbr_enable(struct perf_event *event)
event->ctx->task_ctx_data) {
task_ctx = event->ctx->task_ctx_data;
task_ctx->lbr_callstack_users++;
+ if (!(cpuc->br_sel & LBR_NO_INFO))
+ task_ctx->need_info++;
}

cpuc->lbr_users++;
@@ -367,6 +372,8 @@ void intel_pmu_lbr_disable(struct perf_event *event)
event->ctx->task_ctx_data) {
task_ctx = event->ctx->task_ctx_data;
task_ctx->lbr_callstack_users--;
+ if (!(cpuc->br_sel & LBR_NO_INFO))
+ task_ctx->need_info--;
}

cpuc->lbr_users--;
--
2.4.3


\
 
 \ /
  Last update: 2015-10-20 21:01    [W:0.321 / U:0.016 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site