Skip to contentSkip to conversation

x86,perf: P4 PMU -- protect sensible procedures from preemption

PATCH In review Cyrill Gorcunov 5 participants 13 messages May 5, 2010 – May 6, 2010

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86,perf: P4 PMU -- protect sensible procedures from preemption

Steven reported

|
| I'm getting:
|
| Pid: 3477, comm: perf Not tainted 2.6.34-rc6 #2727
| Call Trace:
|  [<ffffffff811c7565>] debug_smp_processor_id+0xd5/0xf0
|  [<ffffffff81019874>] p4_hw_config+0x2b/0x15c
|  [<ffffffff8107acbc>] ? trace_hardirqs_on_caller+0x12b/0x14f
|  [<ffffffff81019143>] hw_perf_event_init+0x468/0x7be
|  [<ffffffff810782fd>] ? debug_mutex_init+0x31/0x3c
|  [<ffffffff810c68b2>] T.850+0x273/0x42e
|  [<ffffffff810c6cab>] sys_perf_event_open+0x23e/0x3f1
|  [<ffffffff81009e6a>] ? sysret_check+0x2e/0x69
|  [<ffffffff81009e32>] system_call_fastpath+0x16/0x1b
|
| When running perf record in latest tip/perf/core
|

Due to the fact that p4 counters are shared between HT threads we synthetically divide the whole set of counters into two non-intersected subsets. And while we're borrowing counters from these subsets we should not be preempted. So use get_cpu/put_cpu pair.

---
 arch/x86/kernel/cpu/perf_event_p4.c |    9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +6 -3
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
=====================================================================
@@ -421,7 +421,7 @@ static u64 p4_pmu_event_map(int hw_event
 
 static int p4_hw_config(struct perf_event *event)
 {
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	u32 escr, cccr;
 
 	/*
@@ -440,7 +440,7 @@ static int p4_hw_config(struct perf_even
 		event->hw.config = p4_set_ht_bit(event->hw.config);
 
 	if (event->attr.type != PERF_TYPE_RAW)
-		return 0;
+		goto out;
 
 	/*
 	 * We don't control raw events so it's up to the caller
@@ -455,6 +455,8 @@ static int p4_hw_config(struct perf_even
 		(p4_config_pack_escr(P4_ESCR_MASK_HT) |
 		 p4_config_pack_cccr(P4_CCCR_MASK_HT));
 
+out:
+	put_cpu();
 	return 0;
 }
 
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 {
 	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 	unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	struct hw_perf_event *hwc;
 	struct p4_event_bind *bind;
 	unsigned int i, thread, num;
@@ -777,6 +779,7 @@ reserve:
 	}
 
 done:
+	put_cpu();
 	return num ? -ENOSPC : 0;
 }
 
#1 lore raw
Quoted 77 lines · Cyrill Gorcunov

Steven reported

|
| I'm getting:
|
| Pid: 3477, comm: perf Not tainted 2.6.34-rc6 #2727
| Call Trace:
|  [<ffffffff811c7565>] debug_smp_processor_id+0xd5/0xf0
|  [<ffffffff81019874>] p4_hw_config+0x2b/0x15c
|  [<ffffffff8107acbc>] ? trace_hardirqs_on_caller+0x12b/0x14f
|  [<ffffffff81019143>] hw_perf_event_init+0x468/0x7be
|  [<ffffffff810782fd>] ? debug_mutex_init+0x31/0x3c
|  [<ffffffff810c68b2>] T.850+0x273/0x42e
|  [<ffffffff810c6cab>] sys_perf_event_open+0x23e/0x3f1
|  [<ffffffff81009e6a>] ? sysret_check+0x2e/0x69
|  [<ffffffff81009e32>] system_call_fastpath+0x16/0x1b
|
| When running perf record in latest tip/perf/core
|

Due to the fact that p4 counters are shared between HT threads we synthetically divide the whole set of counters into two non-intersected subsets. And while we're borrowing counters from these subsets we should not be preempted. So use get_cpu/put_cpu pair.

---
 arch/x86/kernel/cpu/perf_event_p4.c |    9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
=====================================================================

arch/x86/kernel/cpu/perf_event_p4.c +5 -3
@@ -421,7 +421,7 @@ static u64 p4_pmu_event_map(int hw_event
 
 static int p4_hw_config(struct perf_event *event)
 {
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	u32 escr, cccr;
 
 	/*
@@ -440,7 +440,7 @@ static int p4_hw_config(struct perf_even
 		event->hw.config = p4_set_ht_bit(event->hw.config);
 
 	if (event->attr.type != PERF_TYPE_RAW)
-		return 0;
+		goto out;
 
 	/*
 	 * We don't control raw events so it's up to the caller
@@ -455,6 +455,8 @@ static int p4_hw_config(struct perf_even
 		(p4_config_pack_escr(P4_ESCR_MASK_HT) |
 		 p4_config_pack_cccr(P4_CCCR_MASK_HT));
 
+out:
+	put_cpu();
 	return 0;
 }
 
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 {
 	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 	unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	struct hw_perf_event *hwc;
 	struct p4_event_bind *bind;
 	unsigned int i, thread, num;

arch/x86/kernel/cpu/perf_event_p4.c reserve: · line 784

Quoted 6 lines
arch/x86/kernel/cpu/perf_event_p4.c +1 -0
@@ -777,6 +779,7 @@ reserve:
 	}
 
 done:
+	put_cpu();
 	return num ? -ENOSPC : 0;
 }

That's no big deal. But I think the schedule_events() is called on pmu::enable() time, when preemption is already disabled.

#2 lore raw

On Wed, May 05, 2010 at 06:57:34PM +0200, Frederic Weisbecker wrote:
...

Quoted 12 lines
Quoted 12 lines
arch/x86/kernel/cpu/perf_event_p4.c +1 -1
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 {
 	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 	unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	struct hw_perf_event *hwc;
 	struct p4_event_bind *bind;
 	unsigned int i, thread, num;
@@ -777,6 +779,7 @@ reserve:
 	}
 
Quoted 8 lines

arch/x86/kernel/cpu/perf_event_p4.c reserve: · line 784

Quoted 4 lines
arch/x86/kernel/cpu/perf_event_p4.c +1 -0
 done:
+	put_cpu();
 	return num ? -ENOSPC : 0;
 }

That's no big deal. But I think the schedule_events() is called on pmu::enable() time, when preemption is already disabled.

We'll be on a safe side using get/put_cpu here (ie in case if something get changed one day).

-- Cyrill

#3 lore raw
Quoted 23 lines · Cyrill Gorcunov

On Wed, May 05, 2010 at 06:57:34PM +0200, Frederic Weisbecker wrote:
...

Quoted 12 lines
Quoted 12 lines
arch/x86/kernel/cpu/perf_event_p4.c +1 -1
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 {
 	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 	unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	struct hw_perf_event *hwc;
 	struct p4_event_bind *bind;
 	unsigned int i, thread, num;
@@ -777,6 +779,7 @@ reserve:
 	}
 
Quoted 8 lines

arch/x86/kernel/cpu/perf_event_p4.c reserve: · line 784

Quoted 4 lines
arch/x86/kernel/cpu/perf_event_p4.c +1 -0
 done:
+	put_cpu();
 	return num ? -ENOSPC : 0;
 }

That's no big deal. But I think the schedule_events() is called on pmu::enable() time, when preemption is already disabled.

Quoted 4 lines

We'll be on a safe side using get/put_cpu here (ie in case if something get changed one day).

-- Cyrill

Sure, no problem with that.

#4 lore raw
In reply to Cyrill Gorcunov
Quoted 23 lines · Cyrill Gorcunov

On Wed, May 05, 2010 at 06:57:34PM +0200, Frederic Weisbecker wrote:
...

Quoted 12 lines
Quoted 12 lines
arch/x86/kernel/cpu/perf_event_p4.c +1 -1
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 {
 	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 	unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	struct hw_perf_event *hwc;
 	struct p4_event_bind *bind;
 	unsigned int i, thread, num;
@@ -777,6 +779,7 @@ reserve:
 	}
 
Quoted 8 lines

arch/x86/kernel/cpu/perf_event_p4.c reserve: · line 784

Quoted 4 lines
arch/x86/kernel/cpu/perf_event_p4.c +1 -0
 done:
+	put_cpu();
 	return num ? -ENOSPC : 0;
 }

That's no big deal. But I think the schedule_events() is called on pmu::enable() time, when preemption is already disabled.

Quoted 2 lines

We'll be on a safe side using get/put_cpu here (ie in case if something get changed one day).

hm, when 'something gets changed one day' we'll see a warning when using unsafe primitives.

So if preemption is always off here we really should not add extra runtime overhead via get_cpu()/put_cpu().

So wouldnt it be better (and faster) to disable preemption in hw_perf_event_init(), which seems to be the bit missing?

Ingo

#5 lore raw
Quoted 32 lines · Ingo Molnar
Quoted 23 lines · Cyrill Gorcunov

On Wed, May 05, 2010 at 06:57:34PM +0200, Frederic Weisbecker wrote:
...

Quoted 12 lines
Quoted 12 lines
arch/x86/kernel/cpu/perf_event_p4.c +1 -1
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
  {
   unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
   unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
- int cpu = raw_smp_processor_id();
+ int cpu = get_cpu();
   struct hw_perf_event *hwc;
   struct p4_event_bind *bind;
   unsigned int i, thread, num;
@@ -777,6 +779,7 @@ reserve:
   }
 
Quoted 8 lines
Quoted 4 lines

 done:
+ put_cpu();
  return num ? -ENOSPC : 0;
 }

That's no big deal. But I think the schedule_events() is called on pmu::enable() time, when preemption is already disabled.

Quoted 2 lines

We'll be on a safe side using get/put_cpu here (ie in case if something get changed one day).

hm, when 'something gets changed one day' we'll see a warning when using unsafe primitives.

Quoted 8 lines

So if preemption is always off here we really should not add extra runtime overhead via get_cpu()/put_cpu().

So wouldnt it be better (and faster) to disable preemption in hw_perf_event_init(), which seems to be the bit missing?

       Ingo

the thing are that p4 is only snippet here which is sensible to preemtion, and hw_perf_event_init is executing with preemtion off (but i could miss the details here, dont have code under my hands at moment, so PeterZ help is needed ;) but more important reason why i've saved get/put here is that otherwise i would not have rights to put tested-by tag, since it would not be the patch Steven has tested. We could make a patch on top of this one, or we could drop this one, make new with explicit preemt off in caller and use smp_processor_id in p4 schedule routine. What is preferred?

#6 lore raw
Quoted 42 lines · Cyrill Gorcunov
Quoted 32 lines · Ingo Molnar
Quoted 23 lines · Cyrill Gorcunov

On Wed, May 05, 2010 at 06:57:34PM +0200, Frederic Weisbecker wrote:
...

Quoted 12 lines
Quoted 12 lines
(quoted diff) +1 -1
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 ?{
 ? unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 ? unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
- int cpu = raw_smp_processor_id();
+ int cpu = get_cpu();
 ? struct hw_perf_event *hwc;
 ? struct p4_event_bind *bind;
 ? unsigned int i, thread, num;
@@ -777,6 +779,7 @@ reserve:
 ? }
 
Quoted 8 lines
Quoted 4 lines

?done:
+ put_cpu();
? return num ? -ENOSPC : 0;
?}

That's no big deal. But I think the schedule_events() is called on pmu::enable() time, when preemption is already disabled.

Quoted 2 lines

We'll be on a safe side using get/put_cpu here (ie in case if something get changed one day).

hm, when 'something gets changed one day' we'll see a warning when using unsafe primitives.

Quoted 8 lines

So if preemption is always off here we really should not add extra runtime overhead via get_cpu()/put_cpu().

So wouldnt it be better (and faster) to disable preemption in hw_perf_event_init(), which seems to be the bit missing?

? ? ? ?Ingo

Quoted 8 lines

the thing are that p4 is only snippet here which is sensible to preemtion, and hw_perf_event_init is executing with preemtion off (but i could miss the details here, dont have code under my hands at moment, so PeterZ help is needed ;) but more important reason why i've saved get/put here is that otherwise i would not have rights to put tested-by tag, since it would not be the patch Steven has tested. We could make a patch on top of this one, or we could drop this one, make new with explicit preemt off in caller and use smp_processor_id in p4 schedule routine. What is preferred?

We want the one with the least runtime overhead. These are instrumentation routines, so we want to optimize them as much as possible.

Thanks,

Ingo

#7 lore raw
Quoted 54 lines · Ingo Molnar
Quoted 42 lines · Cyrill Gorcunov
Quoted 32 lines · Ingo Molnar
Quoted 23 lines · Cyrill Gorcunov

On Wed, May 05, 2010 at 06:57:34PM +0200, Frederic Weisbecker wrote:
...

Quoted 12 lines
Quoted 12 lines
(quoted diff) +1 -1
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 ?{
 ? unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 ? unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
- int cpu = raw_smp_processor_id();
+ int cpu = get_cpu();
 ? struct hw_perf_event *hwc;
 ? struct p4_event_bind *bind;
 ? unsigned int i, thread, num;
@@ -777,6 +779,7 @@ reserve:
 ? }
 
Quoted 8 lines
Quoted 4 lines

?done:
+ put_cpu();
? return num ? -ENOSPC : 0;
?}

That's no big deal. But I think the schedule_events() is called on pmu::enable() time, when preemption is already disabled.

Quoted 2 lines

We'll be on a safe side using get/put_cpu here (ie in case if something get changed one day).

hm, when 'something gets changed one day' we'll see a warning when using unsafe primitives.

Quoted 8 lines

So if preemption is always off here we really should not add extra runtime overhead via get_cpu()/put_cpu().

So wouldnt it be better (and faster) to disable preemption in hw_perf_event_init(), which seems to be the bit missing?

 ? ? ? ?Ingo

Quoted 8 lines

the thing are that p4 is only snippet here which is sensible to preemtion, and hw_perf_event_init is executing with preemtion off (but i could miss the details here, dont have code under my hands at moment, so PeterZ help is needed ;) but more important reason why i've saved get/put here is that otherwise i would not have rights to put tested-by tag, since it would not be the patch Steven has tested. We could make a patch on top of this one, or we could drop this one, make new with explicit preemt off in caller and use smp_processor_id in p4 schedule routine. What is preferred?

Quoted 7 lines

We want the one with the least runtime overhead. These are instrumentation routines, so we want to optimize them as much as possible.

Thanks,

       Ingo

ok, Ingo, dont apply this patch then for a while.

#8 lore raw
Quoted 11 lines · Cyrill Gorcunov
Quoted 9 lines · Ingo Molnar
Quoted 8 lines

the thing are that p4 is only snippet here which is sensible to preemtion, and hw_perf_event_init is executing with preemtion off (but i could miss the details here, dont have code under my hands at moment, so PeterZ help is needed ;) but more important reason why i've saved get/put here is that otherwise i would not have rights to put tested-by tag, since it would not be the patch Steven has tested. We could make a patch on top of this one, or we could drop this one, make new with explicit preemt off in caller and use smp_processor_id in p4 schedule routine. What is preferred?

Quoted 2 lines
Quoted 2 lines

We want the one with the least runtime overhead. These are instrumentation routines, so we want to optimize them as much as possible.

Yeah, my point was either disable preemption or keep the checks. In other words, if you don't disable preemption, do not use raw_smp_procesor_id(), because then we will not catch it if it changes in the future.

Quoted 1 line

ok, Ingo, dont apply this patch then for a while.

Send another patch, I'll test it again ;-)

-- Steve

#9 lore raw

On Thu, May 06, 2010 at 09:45:24AM -0400, Steven Rostedt wrote:
...

Quoted 9 lines
Quoted 2 lines
Quoted 2 lines

We want the one with the least runtime overhead. These are instrumentation routines, so we want to optimize them as much as possible.

Yeah, my point was either disable preemption or keep the checks. In other words, if you don't disable preemption, do not use raw_smp_procesor_id(), because then we will not catch it if it changes in the future.

Quoted 7 lines
Quoted 1 line

ok, Ingo, dont apply this patch then for a while.

Send another patch, I'll test it again ;-)

-- Steve

Ingo, Steven, it seems we have potential preemtion available in perf_event.c:validate_group:x86_pmu.schedule_events() which is reached via syscall from userspace perf_event_open() call, so get_cpu is still needed. But I'm a bit messed with call graph at the moment :(

-- Cyrill

#10 lore raw
Quoted 19 lines · Cyrill Gorcunov

On Thu, May 06, 2010 at 09:45:24AM -0400, Steven Rostedt wrote:
...

Quoted 9 lines
Quoted 2 lines
Quoted 2 lines

We want the one with the least runtime overhead. These are instrumentation routines, so we want to optimize them as much as possible.

Yeah, my point was either disable preemption or keep the checks. In other words, if you don't disable preemption, do not use raw_smp_procesor_id(), because then we will not catch it if it changes in the future.

Quoted 7 lines
Quoted 1 line

ok, Ingo, dont apply this patch then for a while.

Send another patch, I'll test it again ;-)

-- Steve

Quoted 7 lines

Ingo, Steven, it seems we have potential preemtion available in perf_event.c:validate_group:x86_pmu.schedule_events() which is reached via syscall from userspace perf_event_open() call, so get_cpu is still needed. But I'm a bit messed with call graph at the moment :(

-- Cyrill

Steve, while I'm diving through call graph could you give this patch a shot? If preemtion happens -- it'll trigger it fast.

-- Cyrill

---
 arch/x86/kernel/cpu/perf_event_p4.c |    8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +5 -3
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
=====================================================================
@@ -421,7 +421,7 @@ static u64 p4_pmu_event_map(int hw_event
 
 static int p4_hw_config(struct perf_event *event)
 {
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	u32 escr, cccr;
 
 	/*
@@ -440,7 +440,7 @@ static int p4_hw_config(struct perf_even
 		event->hw.config = p4_set_ht_bit(event->hw.config);
 
 	if (event->attr.type != PERF_TYPE_RAW)
-		return 0;
+		goto out;
 
 	/*
 	 * We don't control raw events so it's up to the caller
@@ -455,6 +455,8 @@ static int p4_hw_config(struct perf_even
 		(p4_config_pack_escr(P4_ESCR_MASK_HT) |
 		 p4_config_pack_cccr(P4_CCCR_MASK_HT));
 
+out:
+	put_cpu();
 	return 0;
 }
 
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 {
 	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 	unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
-	int cpu = raw_smp_processor_id();
+	int cpu = smp_processor_id();
 	struct hw_perf_event *hwc;
 	struct p4_event_bind *bind;
 	unsigned int i, thread, num;
Tested-by
#11 lore raw
Quoted 26 lines · Cyrill Gorcunov
Quoted 25 lines · Cyrill Gorcunov

On Thu, May 06, 2010 at 09:45:24AM -0400, Steven Rostedt wrote:
...

Quoted 9 lines
Quoted 2 lines
Quoted 2 lines

We want the one with the least runtime overhead. These are instrumentation routines, so we want to optimize them as much as possible.

Yeah, my point was either disable preemption or keep the checks. In other words, if you don't disable preemption, do not use raw_smp_procesor_id(), because then we will not catch it if it changes in the future.

Quoted 7 lines
Quoted 1 line

ok, Ingo, dont apply this patch then for a while.

Send another patch, I'll test it again ;-)

-- Steve

Ingo, Steven, it seems we have potential preemtion available in perf_event.c:validate_group:x86_pmu.schedule_events() which is reached via syscall from userspace perf_event_open() call, so get_cpu is still needed. But I'm a bit messed with call graph at the moment :(

Quoted 4 lines
Quoted 1 line

-- Cyrill

Steve, while I'm diving through call graph could you give this patch a shot? If preemtion happens -- it'll trigger it fast.

This one works too.

-- Steve

Quoted 46 lines

-- Cyrill

---
 arch/x86/kernel/cpu/perf_event_p4.c |    8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
=====================================================================

arch/x86/kernel/cpu/perf_event_p4.c +5 -3
@@ -421,7 +421,7 @@ static u64 p4_pmu_event_map(int hw_event
 
 static int p4_hw_config(struct perf_event *event)
 {
-	int cpu = raw_smp_processor_id();
+	int cpu = get_cpu();
 	u32 escr, cccr;
 
 	/*
@@ -440,7 +440,7 @@ static int p4_hw_config(struct perf_even
 		event->hw.config = p4_set_ht_bit(event->hw.config);
 
 	if (event->attr.type != PERF_TYPE_RAW)
-		return 0;
+		goto out;
 
 	/*
 	 * We don't control raw events so it's up to the caller
@@ -455,6 +455,8 @@ static int p4_hw_config(struct perf_even
 		(p4_config_pack_escr(P4_ESCR_MASK_HT) |
 		 p4_config_pack_cccr(P4_CCCR_MASK_HT));
 
+out:
+	put_cpu();
 	return 0;
 }
 
@@ -741,7 +743,7 @@ static int p4_pmu_schedule_events(struct
 {
 	unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
 	unsigned long escr_mask[BITS_TO_LONGS(ARCH_P4_TOTAL_ESCR)];
-	int cpu = raw_smp_processor_id();
+	int cpu = smp_processor_id();
 	struct hw_perf_event *hwc;
 	struct p4_event_bind *bind;
 	unsigned int i, thread, num;
#12 lore raw

On Thu, May 06, 2010 at 02:32:33PM -0400, Steven Rostedt wrote:
[...]

Quoted 10 lines
Quoted 3 lines

Steve, while I'm diving through call graph could you give this patch a shot? If preemtion happens -- it'll trigger it fast.

This one works too.

-- Steve

Thanks a huge, Steven! I'll tinkle a bit more with p4 pmu and will send a series of patches with this one included.

-- Cyrill