Skip to contentSkip to conversation

perf, x86: add Intel IvyBridge event scheduling constraints

PATCH In review Stephane Eranian 4 participants 7 messages Feb 20, 2013 – Feb 26, 2013

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

PATCH 1/1 perf, x86: add Intel IvyBridge event scheduling constraints

Intel IvyBridge processor has different constraints compared to SandyBridge. Therefore it needs its own contraint table. This patch adds the constraint table. Without this patch, the events listed in the patch may not be scheduled correctly and bogus counts may be collected.

---
 arch/x86/kernel/cpu/perf_event_intel.c |   23 ++++++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/perf_event_intel.c +22 -1
index 4914e94..529c893 100644
@@ -107,6 +107,27 @@ static struct event_constraint intel_snb_event_constraints[] __read_mostly =
 	EVENT_CONSTRAINT_END
 };
 
+static struct event_constraint intel_ivb_event_constraints[] __read_mostly =
+{
+	FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
+	FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */
+	FIXED_EVENT_CONSTRAINT(0x0300, 2), /* CPU_CLK_UNHALTED.REF */
+	INTEL_UEVENT_CONSTRAINT(0x0148, 0x4), /* L1D_PEND_MISS.PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x0279, 0xf), /* IDQ.EMTPY */
+	INTEL_UEVENT_CONSTRAINT(0x019c, 0xf), /* IDQ_UOPS_NOT_DELIVERED.CORE */
+	INTEL_UEVENT_CONSTRAINT(0x04a3, 0xf), /* CYCLE_ACTIVITY.CYCLES_NO_EXECUTE */
+	INTEL_UEVENT_CONSTRAINT(0x05a3, 0xf), /* CYCLE_ACTIVITY.STALLS_L2_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x06a3, 0xf), /* CYCLE_ACTIVITY.STALLS_LDM_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x08a3, 0x4), /* CYCLE_ACTIVITY.CYCLES_L1D_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x0ca3, 0x4), /* CYCLE_ACTIVITY.STALLS_L1D_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x01c0, 0x2), /* INST_RETIRED.PREC_DIST */
+	INTEL_EVENT_CONSTRAINT(0xd0, 0xf), /* MEM_UOPS_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd1, 0xf), /* MEM_LOAD_UOPS_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd2, 0xf), /* MEM_LOAD_UOPS_LLC_HIT_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd3, 0xf), /*  MEM_LOAD_UOPS_LLC_MISS_RETIRED.* */
+	EVENT_CONSTRAINT_END
+};
+
 static struct extra_reg intel_westmere_extra_regs[] __read_mostly =
 {
 	INTEL_EVENT_EXTRA_REG(0xb7, MSR_OFFCORE_RSP_0, 0xffff, RSP_0),
@@ -2095,7 +2116,7 @@ __init int intel_pmu_init(void)
 
 		intel_pmu_lbr_init_snb();
 
-		x86_pmu.event_constraints = intel_snb_event_constraints;
+		x86_pmu.event_constraints = intel_ivb_event_constraints;
 		x86_pmu.pebs_constraints = intel_ivb_pebs_event_constraints;
 		x86_pmu.pebs_aliases = intel_pebs_aliases_snb;
 		x86_pmu.extra_regs = intel_snb_extra_regs;
Signature · 1 line
1.7.9.5
#1 lore raw

[tip:perf/urgent] perf/x86: Add Intel IvyBridge event scheduling constraints

perf/x86: Add Intel IvyBridge event scheduling constraints
Commit-ID:  69943182bb9e19e4b60ea5033f683ec1af1703a9
Gitweb:     http://git.kernel.org/tip/69943182bb9e19e4b60ea5033f683ec1af1703a9
Author:     Stephane Eranian <[email protected]>
AuthorDate: Wed, 20 Feb 2013 11:15:12 +0100
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 20 Feb 2013 11:22:46 +0100

perf/x86: Add Intel IvyBridge event scheduling constraints

Intel IvyBridge processor has different constraints compared to SandyBridge. Therefore it needs its own contraint table. This patch adds the constraint table.

Without this patch, the events listed in the patch may not be scheduled correctly and bogus counts may be collected.

---
 arch/x86/kernel/cpu/perf_event_intel.c | 23 ++++++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)
arch/x86/kernel/cpu/perf_event_intel.c +22 -1
index 4914e94..529c893 100644
@@ -107,6 +107,27 @@ static struct event_constraint intel_snb_event_constraints[] __read_mostly =
 	EVENT_CONSTRAINT_END
 };
 
+static struct event_constraint intel_ivb_event_constraints[] __read_mostly =
+{
+	FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
+	FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */
+	FIXED_EVENT_CONSTRAINT(0x0300, 2), /* CPU_CLK_UNHALTED.REF */
+	INTEL_UEVENT_CONSTRAINT(0x0148, 0x4), /* L1D_PEND_MISS.PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x0279, 0xf), /* IDQ.EMTPY */
+	INTEL_UEVENT_CONSTRAINT(0x019c, 0xf), /* IDQ_UOPS_NOT_DELIVERED.CORE */
+	INTEL_UEVENT_CONSTRAINT(0x04a3, 0xf), /* CYCLE_ACTIVITY.CYCLES_NO_EXECUTE */
+	INTEL_UEVENT_CONSTRAINT(0x05a3, 0xf), /* CYCLE_ACTIVITY.STALLS_L2_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x06a3, 0xf), /* CYCLE_ACTIVITY.STALLS_LDM_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x08a3, 0x4), /* CYCLE_ACTIVITY.CYCLES_L1D_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x0ca3, 0x4), /* CYCLE_ACTIVITY.STALLS_L1D_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x01c0, 0x2), /* INST_RETIRED.PREC_DIST */
+	INTEL_EVENT_CONSTRAINT(0xd0, 0xf), /* MEM_UOPS_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd1, 0xf), /* MEM_LOAD_UOPS_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd2, 0xf), /* MEM_LOAD_UOPS_LLC_HIT_RETIRED.* */
+	INTEL_EVENT_CONSTRAINT(0xd3, 0xf), /*  MEM_LOAD_UOPS_LLC_MISS_RETIRED.* */
+	EVENT_CONSTRAINT_END
+};
+
 static struct extra_reg intel_westmere_extra_regs[] __read_mostly =
 {
 	INTEL_EVENT_EXTRA_REG(0xb7, MSR_OFFCORE_RSP_0, 0xffff, RSP_0),
@@ -2095,7 +2116,7 @@ __init int intel_pmu_init(void)
 
 		intel_pmu_lbr_init_snb();
 
-		x86_pmu.event_constraints = intel_snb_event_constraints;
+		x86_pmu.event_constraints = intel_ivb_event_constraints;
 		x86_pmu.pebs_constraints = intel_ivb_pebs_event_constraints;
 		x86_pmu.pebs_aliases = intel_pebs_aliases_snb;
 		x86_pmu.extra_regs = intel_snb_extra_regs;
#2 lore raw
In reply to Stephane Eranian
Quoted 5 lines · Stephane Eranian

Intel IvyBridge processor has different constraints compared to SandyBridge. Therefore it needs its own contraint table. This patch adds the constraint table. Without this patch, the events listed in the patch may not be scheduled correctly and bogus counts may be collected.

Thanks. I ran into this problem too and was about to write a similar patch.

arch/x86/kernel/cpu/perf_event_intel.c static struct event_constraint intel_snb_event_constraints[] __read_mostly = · line 122

Quoted 5 lines
arch/x86/kernel/cpu/perf_event_intel.c +5 -0
+	INTEL_UEVENT_CONSTRAINT(0x04a3, 0xf), /* CYCLE_ACTIVITY.CYCLES_NO_EXECUTE */
+	INTEL_UEVENT_CONSTRAINT(0x05a3, 0xf), /* CYCLE_ACTIVITY.STALLS_L2_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x06a3, 0xf), /* CYCLE_ACTIVITY.STALLS_LDM_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x08a3, 0x4), /* CYCLE_ACTIVITY.CYCLES_L1D_PENDING */
+	INTEL_UEVENT_CONSTRAINT(0x0ca3, 0x4), /* CYCLE_ACTIVITY.STALLS_L1D_PENDING */

Except for LDM_PENDING the CYCLE_ACTIVITY events have been also added to Sandy Bridge. So it should be also added there.

In fact I think you can still share the table because it would just add some non existent events to Sandy Bridge, which is a noop.

-Andi

#3 lore raw
Quoted 17 lines · Andi Kleen
Quoted 5 lines · Stephane Eranian

Intel IvyBridge processor has different constraints compared to SandyBridge. Therefore it needs its own contraint table. This patch adds the constraint table. Without this patch, the events listed in the patch may not be scheduled correctly and bogus counts may be collected.

Thanks. I ran into this problem too and was about to write a similar patch.

arch/x86/kernel/cpu/perf_event_intel.c static struct event_constraint intel_snb_event_constraints[] __read_mostly = · line 122

Quoted 5 lines
arch/x86/kernel/cpu/perf_event_intel.c +5 -0
+     INTEL_UEVENT_CONSTRAINT(0x04a3, 0xf), /* CYCLE_ACTIVITY.CYCLES_NO_EXECUTE */
+     INTEL_UEVENT_CONSTRAINT(0x05a3, 0xf), /* CYCLE_ACTIVITY.STALLS_L2_PENDING */
+     INTEL_UEVENT_CONSTRAINT(0x06a3, 0xf), /* CYCLE_ACTIVITY.STALLS_LDM_PENDING */
+     INTEL_UEVENT_CONSTRAINT(0x08a3, 0x4), /* CYCLE_ACTIVITY.CYCLES_L1D_PENDING */
+     INTEL_UEVENT_CONSTRAINT(0x0ca3, 0x4), /* CYCLE_ACTIVITY.STALLS_L1D_PENDING */
Quoted 3 lines

Except for LDM_PENDING the CYCLE_ACTIVITY events have been also added to Sandy Bridge. So it should be also added there.

As far as I know and I double-checked the documentation I have, there is no CYCLE_ACTIVITY
event on SNB or SNB-EP.

Quoted 3 lines

In fact I think you can still share the table because it would just add some non existent events to Sandy Bridge, which is a noop.

I don't see the point of this, except saving a few bytes. Isn't it better to keep each PMU separate?

Quoted 2 lines

-Andi

#4 lore raw
Quoted 6 lines
Quoted 3 lines

Except for LDM_PENDING the CYCLE_ACTIVITY events have been also added to Sandy Bridge. So it should be also added there.

As far as I know and I double-checked the documentation I have, there is no CYCLE_ACTIVITY
event on SNB or SNB-EP.

page 19-28 in the Jan 2013 SDM. It was only recently added. It's only a subset of what you have on Ivy Bridge.

Quoted 6 lines
Quoted 4 lines

In fact I think you can still share the table because it would just add some non existent events to Sandy Bridge, which is a noop.

I don't see the point of this, except saving a few bytes. Isn't it better to keep each PMU separate?

Ok, as long as you update the SNB table too.

-Andi

#5 lore raw
Quoted 10 lines · Andi Kleen
Quoted 6 lines
Quoted 3 lines

Except for LDM_PENDING the CYCLE_ACTIVITY events have been also added to Sandy Bridge. So it should be also added there.

As far as I know and I double-checked the documentation I have, there is no CYCLE_ACTIVITY
event on SNB or SNB-EP.

page 19-28 in the Jan 2013 SDM. It was only recently added. It's only a subset of what you have on Ivy Bridge.

I have confirmed that this is a typo in the SDM. This event is not available on SandyBridge.

Quoted 10 lines
Quoted 6 lines
Quoted 4 lines

In fact I think you can still share the table because it would just add some non existent events to Sandy Bridge, which is a noop.

I don't see the point of this, except saving a few bytes. Isn't it better to keep each PMU separate?

Ok, as long as you update the SNB table too.

-Andi

#6 lore raw
In reply to Stephane Eranian

Hi Stephane,

I was wondering what the differences are between CYCLE_ACTIVITY.CYCLES_**_PENDING and CYCLE_ACTIVITY.STALLS_**_PENDING, because I could only find following events in the SDM, which seem to be different from the ones provided here. Correct me if I'm wrong.

A3H 01H CYCLE_ACTIVITY.CYCLES_L2_PENDING
- Cycles with pending L2 miss loads. Set Cmask=2 tocount cycle. Use only
when HTT is off
A3H 02H CYCLE_ACTIVITY.CYCLES_LDM_PENDING
- Cycles with pending memory loads. Set Cmask=2 to count cycle.
A3H 05H CYCLE_ACTIVITY.STALLS_L2_PENDING
- Number of loads missed L2. Use only when HTT is off
A3H 08H CYCLE_ACTIVITY.CYCLES_L1D_PENDING
- Cycles with pending L1 cache miss loads. SetCmask=8 to count cycle.PMC2
only

Thanks a lot!

--
View this message in context: http://linux-kernel.2935.n7.nabble.com/PATCH-perf-x86-add-Intel-IvyBridge-event-scheduling-constraints-tp604032p606450.html
Sent from the Linux Kernel mailing list archive at Nabble.com.