perf, x86: Consolidate some code repetition
authorPeter Zijlstra <peterz@infradead.org>
Thu, 8 Apr 2010 21:03:20 +0000 (23:03 +0200)
committerIngo Molnar <mingo@elte.hu>
Fri, 7 May 2010 09:31:02 +0000 (11:31 +0200)
Remove some duplicated logic.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/kernel/cpu/perf_event_intel_ds.c

index 080b9b0..35056f7 100644 (file)
@@ -452,14 +452,54 @@ static int intel_pmu_pebs_fixup_ip(struct pt_regs *regs)
 
 static int intel_pmu_save_and_restart(struct perf_event *event);
 
+static void __intel_pmu_pebs_event(struct perf_event *event,
+                                  struct pt_regs *iregs, void *__pebs)
+{
+       /*
+        * We cast to pebs_record_core since that is a subset of
+        * both formats and we don't use the other fields in this
+        * routine.
+        */
+       struct pebs_record_core *pebs = __pebs;
+       struct perf_sample_data data;
+       struct pt_regs regs;
+
+       if (!intel_pmu_save_and_restart(event))
+               return;
+
+       perf_sample_data_init(&data, 0);
+       data.period = event->hw.last_period;
+
+       /*
+        * We use the interrupt regs as a base because the PEBS record
+        * does not contain a full regs set, specifically it seems to
+        * lack segment descriptors, which get used by things like
+        * user_mode().
+        *
+        * In the simple case fix up only the IP and BP,SP regs, for
+        * PERF_SAMPLE_IP and PERF_SAMPLE_CALLCHAIN to function properly.
+        * A possible PERF_SAMPLE_REGS will have to transfer all regs.
+        */
+       regs = *iregs;
+       regs.ip = pebs->ip;
+       regs.bp = pebs->bp;
+       regs.sp = pebs->sp;
+
+       if (intel_pmu_pebs_fixup_ip(regs))
+               regs.flags |= PERF_EFLAGS_EXACT;
+       else
+               regs.flags &= ~PERF_EFLAGS_EXACT;
+
+       if (perf_event_overflow(event, 1, &data, &regs))
+               x86_pmu_stop(event);
+}
+
 static void intel_pmu_drain_pebs_core(struct pt_regs *iregs)
 {
        struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
        struct debug_store *ds = cpuc->ds;
        struct perf_event *event = cpuc->events[0]; /* PMC0 only */
        struct pebs_record_core *at, *top;
-       struct perf_sample_data data;
-       struct pt_regs regs;
        int n;
 
        if (!ds || !x86_pmu.pebs)
@@ -485,9 +525,6 @@ static void intel_pmu_drain_pebs_core(struct pt_regs *iregs)
        if (n <= 0)
                return;
 
-       if (!intel_pmu_save_and_restart(event))
-               return;
-
        /*
         * Should not happen, we program the threshold at 1 and do not
         * set a reset value.
@@ -495,31 +532,7 @@ static void intel_pmu_drain_pebs_core(struct pt_regs *iregs)
        WARN_ON_ONCE(n > 1);
        at += n - 1;
 
-       perf_sample_data_init(&data, 0);
-       data.period = event->hw.last_period;
-
-       /*
-        * We use the interrupt regs as a base because the PEBS record
-        * does not contain a full regs set, specifically it seems to
-        * lack segment descriptors, which get used by things like
-        * user_mode().
-        *
-        * In the simple case fix up only the IP and BP,SP regs, for
-        * PERF_SAMPLE_IP and PERF_SAMPLE_CALLCHAIN to function properly.
-        * A possible PERF_SAMPLE_REGS will have to transfer all regs.
-        */
-       regs = *iregs;
-       regs.ip = at->ip;
-       regs.bp = at->bp;
-       regs.sp = at->sp;
-
-       if (intel_pmu_pebs_fixup_ip(&regs))
-               regs.flags |= PERF_EFLAGS_EXACT;
-       else
-               regs.flags &= ~PERF_EFLAGS_EXACT;
-
-       if (perf_event_overflow(event, 1, &data, &regs))
-               x86_pmu_stop(event);
+       __intel_pmu_pebs_event(event, iregs, at);
 }
 
 static void intel_pmu_drain_pebs_nhm(struct pt_regs *iregs)
@@ -527,9 +540,7 @@ static void intel_pmu_drain_pebs_nhm(struct pt_regs *iregs)
        struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
        struct debug_store *ds = cpuc->ds;
        struct pebs_record_nhm *at, *top;
-       struct perf_sample_data data;
        struct perf_event *event = NULL;
-       struct pt_regs regs;
        u64 status = 0;
        int bit, n;
 
@@ -571,27 +582,7 @@ static void intel_pmu_drain_pebs_nhm(struct pt_regs *iregs)
                if (!event || bit >= MAX_PEBS_EVENTS)
                        continue;
 
-               if (!intel_pmu_save_and_restart(event))
-                       continue;
-
-               perf_sample_data_init(&data, 0);
-               data.period = event->hw.last_period;
-
-               /*
-                * See the comment in intel_pmu_drain_pebs_core()
-                */
-               regs = *iregs;
-               regs.ip = at->ip;
-               regs.bp = at->bp;
-               regs.sp = at->sp;
-
-               if (intel_pmu_pebs_fixup_ip(&regs))
-                       regs.flags |= PERF_EFLAGS_EXACT;
-               else
-                       regs.flags &= ~PERF_EFLAGS_EXACT;
-
-               if (perf_event_overflow(event, 1, &data, &regs))
-                       x86_pmu_stop(event);
+               __intel_pmu_pebs_event(event, iregs, at);
        }
 }