Skip to contentSkip to conversation

perf,p4 series for -tip

PATCH In review Cyrill Gorcunov 2 participants 10 messages May 8, 2010 – May 10, 2010

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

cover perf,p4 series for -tip

Hi,

please review and test if possible. Any complains
are quite welcome! Thanks!

Cyrill

#1 lore raw

PATCH 1/4 x86,perf: P4 PMU -- configurate predefined events

If an event is not RAW we should exit p4_hw_config early but call x86_setup_perfctr as well.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Frederic Weisbecker <[email protected]>
CC Lin Ming <[email protected]>
CC Robert Richter <[email protected]>

---

Robert, if only I'm not missing something this is
a side effect of commit 9d0fcba67e47ff398a6fa86476d4884d472dc98a, wonders why don't we hit it earlier. Am I wrong?

 arch/x86/kernel/cpu/perf_event_p4.c |   29 ++++++++++++++---------------
 1 file changed, 14 insertions(+), 15 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +14 -15
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
=====================================================================
@@ -439,21 +439,20 @@ static int p4_hw_config(struct perf_even
 	if (p4_ht_active() && p4_ht_thread(cpu))
 		event->hw.config = p4_set_ht_bit(event->hw.config);
 
-	if (event->attr.type != PERF_TYPE_RAW)
-		return 0;
-
-	/*
-	 * We don't control raw events so it's up to the caller
-	 * to pass sane values (and we don't count the thread number
-	 * on HT machine but allow HT-compatible specifics to be
-	 * passed on)
-	 *
-	 * XXX: HT wide things should check perf_paranoid_cpu() &&
-	 *      CAP_SYS_ADMIN
-	 */
-	event->hw.config |= event->attr.config &
-		(p4_config_pack_escr(P4_ESCR_MASK_HT) |
-		 p4_config_pack_cccr(P4_CCCR_MASK_HT));
+	if (event->attr.type == PERF_TYPE_RAW) {
+		/*
+		 * We don't control raw events so it's up to the caller
+		 * to pass sane values (and we don't count the thread number
+		 * on HT machine but allow HT-compatible specifics to be
+		 * passed on)
+		 *
+		 * XXX: HT wide things should check perf_paranoid_cpu() &&
+		 *      CAP_SYS_ADMIN
+		 */
+		event->hw.config |= event->attr.config &
+			(p4_config_pack_escr(P4_ESCR_MASK_HT) |
+			 p4_config_pack_cccr(P4_CCCR_MASK_HT));
+	}
 
 	return x86_setup_perfctr(event);
 }
#2 lore raw

Re: [patch 1/4] x86,perf: P4 PMU -- configurate predefined events

Quoted 3 lines · Cyrill Gorcunov

If an event is not RAW we should exit p4_hw_config early but call x86_setup_perfctr as well.

Actually i meant "should not"

Quoted 57 lines
CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Frederic Weisbecker <[email protected]>
CC Lin Ming <[email protected]>
CC Robert Richter <[email protected]>

---

Robert, if only I'm not missing something this is
a side effect of commit 9d0fcba67e47ff398a6fa86476d4884d472dc98a, wonders why don't we hit it earlier. Am I wrong?

 arch/x86/kernel/cpu/perf_event_p4.c |   29 ++++++++++++++---------------

 1 file changed, 14 insertions(+), 15 deletions(-)

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

arch/x86/kernel/cpu/perf_event_p4.c +14 -15
@@ -439,21 +439,20 @@ static int p4_hw_config(struct perf_even
        if (p4_ht_active() && p4_ht_thread(cpu))
                event->hw.config = p4_set_ht_bit(event->hw.config);
 
-       if (event->attr.type != PERF_TYPE_RAW)
-               return 0;
-
-       /*
-        * We don't control raw events so it's up to the caller
-        * to pass sane values (and we don't count the thread number
-        * on HT machine but allow HT-compatible specifics to be
-        * passed on)
-        *
-        * XXX: HT wide things should check perf_paranoid_cpu() &&
-        *      CAP_SYS_ADMIN
-        */
-       event->hw.config |= event->attr.config &
-               (p4_config_pack_escr(P4_ESCR_MASK_HT) |
-                p4_config_pack_cccr(P4_CCCR_MASK_HT));
+       if (event->attr.type == PERF_TYPE_RAW) {
+               /*
+                * We don't control raw events so it's up to the caller
+                * to pass sane values (and we don't count the thread number
+                * on HT machine but allow HT-compatible specifics to be
+                * passed on)
+                *
+                * XXX: HT wide things should check perf_paranoid_cpu() &&
+                *      CAP_SYS_ADMIN
+                */
+               event->hw.config |= event->attr.config &
+                       (p4_config_pack_escr(P4_ESCR_MASK_HT) |
+                        p4_config_pack_cccr(P4_CCCR_MASK_HT));
+       }
 
        return x86_setup_perfctr(event);
  }
#3 lore raw

Re: [patch 1/4] x86,perf: P4 PMU -- configurate predefined events

In reply to Cyrill Gorcunov
Quoted 9 lines · Cyrill Gorcunov

If an event is not RAW we should exit p4_hw_config early but call x86_setup_perfctr as well.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Frederic Weisbecker <[email protected]>
CC Lin Ming <[email protected]>
CC Robert Richter <[email protected]>
Quoted 5 lines

---

Robert, if only I'm not missing something this is
a side effect of commit 9d0fcba67e47ff398a6fa86476d4884d472dc98a, wonders why don't we hit it earlier. Am I wrong?

Yes, that's true, the function had have two exit points. Thanks for the fix.

-Robert

Signature · 3 lines
Advanced Micro Devices, Inc.
Operating System Research Center
email: [email protected]
#4 lore raw

PATCH 2/4 x86,perf: P4 PMU -- protect sensible procedures from preemption

In reply to 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 (well, strictly speaking in p4_hw_config we just pre-set reference to the subset which allow to save some cycles in schedule routine if it happens on the same cpu). So use get_cpu/put_cpu pair.

Also p4_pmu_schedule_events should use smp_processor_id rather than raw_ version. This allow us to catch up preemption issue (if there will ever be).

CC Steven Rostedt <[email protected]>
CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Frederic Weisbecker <[email protected]>
CC Lin Ming <[email protected]>

---
Note: I've removed Steven's Tested-by tag, since due to commit 9d0fcba67e47ff398a6fa86476d4884d472dc98a the former patch Steven were testing is not applicable anymore.

 arch/x86/kernel/cpu/perf_event_p4.c |    8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +6 -2
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
=====================================================================
@@ -421,7 +421,8 @@ 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();
+	int rc = 0;
 	u32 escr, cccr;
 
 	/*
@@ -454,7 +455,10 @@ static int p4_hw_config(struct perf_even
 			 p4_config_pack_cccr(P4_CCCR_MASK_HT));
 	}
 
-	return x86_setup_perfctr(event);
+	rc = x86_setup_perfctr(event);
+	put_cpu();
+
+	return rc;
 }
 
 static inline void p4_pmu_clear_cccr_ovf(struct hw_perf_event *hwc)
#5 lore raw

[tip:perf/core] x86, perf: P4 PMU -- protect sensible procedures from preemption

Commit-ID:  137351e0feeb9f25d99488ee1afc1c79f5499a9a
Gitweb:     http://git.kernel.org/tip/137351e0feeb9f25d99488ee1afc1c79f5499a9a
Author:     Cyrill Gorcunov <[email protected]>
AuthorDate: Sat, 8 May 2010 15:25:52 +0400
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sat, 8 May 2010 14:17:53 +0200

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 (well, strictly speaking in p4_hw_config we just pre-set reference to the subset which allow to save some cycles in schedule routine if it happens on the same cpu). So use get_cpu/put_cpu pair.

Also p4_pmu_schedule_events should use smp_processor_id rather than raw_ version. This allow us to catch up preemption issue (if there will ever be).

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/cpu/perf_event_p4.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +6 -2
index b1f532d..ca40180 100644
@@ -421,7 +421,8 @@ 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();
+	int rc = 0;
 	u32 escr, cccr;
 
 	/*
@@ -454,7 +455,10 @@ static int p4_hw_config(struct perf_event *event)
 			 p4_config_pack_cccr(P4_CCCR_MASK_HT));
 	}
 
-	return x86_setup_perfctr(event);
+	rc = x86_setup_perfctr(event);
+	put_cpu();
+
+	return rc;
 }
 
 static inline void p4_pmu_clear_cccr_ovf(struct hw_perf_event *hwc)
#6 lore raw

PATCH 3/4 x86,perf: P4 PMU -- Get rid of redundant check for array index

In reply to Cyrill Gorcunov

The caller already has done such a check.
And it was wrong anyway, it had to be '>=' rather then '>'

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Frederic Weisbecker <[email protected]>
CC Lin Ming <[email protected]>
---
 arch/x86/kernel/cpu/perf_event_p4.c |    5 -----
 1 file changed, 5 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +0 -5
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
=====================================================================
@@ -406,11 +406,6 @@ static u64 p4_pmu_event_map(int hw_event
 	unsigned int esel;
 	u64 config;
 
-	if (hw_event > ARRAY_SIZE(p4_general_events)) {
-		printk_once(KERN_ERR "P4 PMU: Bad index: %i\n", hw_event);
-		return 0;
-	}
-
 	config = p4_general_events[hw_event];
 	bind = p4_config_get_bind(config);
 	esel = P4_OPCODE_ESEL(bind->opcode);
#7 lore raw

[tip:perf/core] x86, perf: P4 PMU -- Get rid of redundant check for array index

Commit-ID:  3f51b7119d052827dbb0e40c966acdf2bdc6f47f
Gitweb:     http://git.kernel.org/tip/3f51b7119d052827dbb0e40c966acdf2bdc6f47f
Author:     Cyrill Gorcunov <[email protected]>
AuthorDate: Sat, 8 May 2010 15:25:53 +0400
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sat, 8 May 2010 14:17:53 +0200

x86, perf: P4 PMU -- Get rid of redundant check for array index

The caller already has done such a check.
And it was wrong anyway, it had to be '>=' rather than '>'

Cc Peter Zijlstra <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Lin Ming <[email protected]>

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/cpu/perf_event_p4.c |    5 -----
 1 files changed, 0 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +0 -5
index ca40180..b8c2d37 100644
@@ -406,11 +406,6 @@ static u64 p4_pmu_event_map(int hw_event)
 	unsigned int esel;
 	u64 config;
 
-	if (hw_event > ARRAY_SIZE(p4_general_events)) {
-		printk_once(KERN_ERR "P4 PMU: Bad index: %i\n", hw_event);
-		return 0;
-	}
-
 	config = p4_general_events[hw_event];
 	bind = p4_config_get_bind(config);
 	esel = P4_OPCODE_ESEL(bind->opcode);
#8 lore raw

PATCH 4/4 x86,perf: P4 PMU -- check for proper event index in RAW events

In reply to Cyrill Gorcunov

RAW events are special and we should be ready for user passing insane event index.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Frederic Weisbecker <[email protected]>
CC Lin Ming <[email protected]>
---
 arch/x86/kernel/cpu/perf_event_p4.c |   11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/perf_event_p4.c +10 -1
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
=====================================================================
@@ -418,6 +418,7 @@ static int p4_hw_config(struct perf_even
 {
 	int cpu = get_cpu();
 	int rc = 0;
+	unsigned int evnt;
 	u32 escr, cccr;
 
 	/*
@@ -436,6 +437,14 @@ 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) {
+
+		/* user data may have out-of-bound event index */
+		evnt = p4_config_unpack_event(event->attr.config);
+		if (evnt >= ARRAY_SIZE(p4_event_bind_map)) {
+			rc = -EINVAL;
+			goto out;
+		}
+
 		/*
 		 * We don't control raw events so it's up to the caller
 		 * to pass sane values (and we don't count the thread number
@@ -451,8 +460,8 @@ static int p4_hw_config(struct perf_even
 	}
 
 	rc = x86_setup_perfctr(event);
+out:
 	put_cpu();
-
 	return rc;
 }
 
#9 lore raw

[tip:perf/core] x86, perf: P4 PMU -- check for proper event index in RAW events

Commit-ID:  c7993165ef0c1d636ca05f4787739f8414584e6d
Gitweb:     http://git.kernel.org/tip/c7993165ef0c1d636ca05f4787739f8414584e6d
Author:     Cyrill Gorcunov <[email protected]>
AuthorDate: Sat, 8 May 2010 15:25:54 +0400
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sat, 8 May 2010 14:17:53 +0200

x86, perf: P4 PMU -- check for proper event index in RAW events

RAW events are special and we should be ready for user passing in insane event index values.

Cc Peter Zijlstra <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Lin Ming <[email protected]>

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/cpu/perf_event_p4.c |   11 ++++++++++-
 1 files changed, 10 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event_p4.c +10 -1
index b8c2d37..a603930 100644
@@ -418,6 +418,7 @@ static int p4_hw_config(struct perf_event *event)
 {
 	int cpu = get_cpu();
 	int rc = 0;
+	unsigned int evnt;
 	u32 escr, cccr;
 
 	/*
@@ -436,6 +437,14 @@ static int p4_hw_config(struct perf_event *event)
 		event->hw.config = p4_set_ht_bit(event->hw.config);
 
 	if (event->attr.type == PERF_TYPE_RAW) {
+
+		/* user data may have out-of-bound event index */
+		evnt = p4_config_unpack_event(event->attr.config);
+		if (evnt >= ARRAY_SIZE(p4_event_bind_map)) {
+			rc = -EINVAL;
+			goto out;
+		}
+
 		/*
 		 * We don't control raw events so it's up to the caller
 		 * to pass sane values (and we don't count the thread number
@@ -451,8 +460,8 @@ static int p4_hw_config(struct perf_event *event)
 	}
 
 	rc = x86_setup_perfctr(event);
+out:
 	put_cpu();
-
 	return rc;
 }