Skip to contentSkip to conversation

oprofile: Allow setting EDGE/INV/CMASK for Intel counter events

PATCH In review Andi Kleen 3 participants 6 messages Mar 11, 2011 – Mar 16, 2011 x86

lore
Share & tools
lore

Andi Kleen

author
#0 lore raw

PATCH 1/1 oprofile: Allow setting EDGE/INV/CMASK for Intel counter events

From: Andi Kleen <[email protected]>

For some performance events it's useful to set the EDGE and INV bits and the CMASK mask in the counter control register. The list of predefined events Intel releases for each CPU has some events which require these settings to get more "natural" to use higher level events.

oprofile currently doesn't allow this.

This patch adds new extra configuration fields for them, so that they can be specified in oprofilefs.

An updated oprofile daemon can then make use of this to set them.

---
 arch/x86/oprofile/nmi_int.c    |    4 ++++
 arch/x86/oprofile/op_counter.h |    1 +
 2 files changed, 5 insertions(+), 0 deletions(-)
arch/x86/oprofile/nmi_int.c +4 -0
index e2b7b0c..bace252 100644
@@ -49,6 +49,9 @@ u64 op_x86_get_ctrl(struct op_x86_model_spec const *model,
 	val |= counter_config->user ? ARCH_PERFMON_EVENTSEL_USR : 0;
 	val |= counter_config->kernel ? ARCH_PERFMON_EVENTSEL_OS : 0;
 	val |= (counter_config->unit_mask & 0xFF) << 8;
+	val |= counter_config->extra & (ARCH_PERFMON_EVENTSEL_INV|
+					ARCH_PERFMON_EVENTSEL_EDGE|
+					ARCH_PERFMON_EVENTSEL_CMASK);
 	event &= model->event_mask ? model->event_mask : 0xFF;
 	val |= event & 0xFF;
 	val |= (event & 0x0F00) << 24;
@@ -440,6 +443,7 @@ static int nmi_create_files(struct super_block *sb, struct dentry *root)
 		oprofilefs_create_ulong(sb, dir, "unit_mask", &counter_config[i].unit_mask);
 		oprofilefs_create_ulong(sb, dir, "kernel", &counter_config[i].kernel);
 		oprofilefs_create_ulong(sb, dir, "user", &counter_config[i].user);
+		oprofilefs_create_ulong(sb, dir, "extra", &counter_config[i].extra);
 	}
 
 	return 0;
arch/x86/oprofile/op_counter.h +1 -0
index e28398d..0b7b7b1 100644
@@ -22,6 +22,7 @@ struct op_counter_config {
 	unsigned long kernel;
 	unsigned long user;
 	unsigned long unit_mask;
+	unsigned long extra;
 };
 
 extern struct op_counter_config counter_config[];
Signature · 1 line
1.7.4
#1 lore raw
Quoted 25 lines · Andi Kleen

From: Andi Kleen <[email protected]>

For some performance events it's useful to set the EDGE and INV bits and the CMASK mask in the counter control register. The list of predefined events Intel releases for each CPU has some events which require these settings to get more "natural" to use higher level events.

oprofile currently doesn't allow this.

This patch adds new extra configuration fields for them, so that they can be specified in oprofilefs.

An updated oprofile daemon can then make use of this to set them.

---
 arch/x86/oprofile/nmi_int.c    |    4 ++++
 arch/x86/oprofile/op_counter.h |    1 +
 2 files changed, 5 insertions(+), 0 deletions(-)
arch/x86/oprofile/nmi_int.c +0 -0
index e2b7b0c..bace252 100644

arch/x86/oprofile/nmi_int.c u64 op_x86_get_ctrl(struct op_x86_model_spec const *model, · line 54

Quoted 6 lines
arch/x86/oprofile/nmi_int.c +3 -0
@@ -49,6 +49,9 @@ u64 op_x86_get_ctrl(struct op_x86_model_spec const *model,
 	val |= counter_config->user ? ARCH_PERFMON_EVENTSEL_USR : 0;
 	val |= counter_config->kernel ? ARCH_PERFMON_EVENTSEL_OS : 0;
 	val |= (counter_config->unit_mask & 0xFF) << 8;
+	val |= counter_config->extra & (ARCH_PERFMON_EVENTSEL_INV|
+					ARCH_PERFMON_EVENTSEL_EDGE|
+					ARCH_PERFMON_EVENTSEL_CMASK);

I would like to write the actual value back so that userland may read it:

	counter_config->extra &= (ARCH_PERFMON_EVENTSEL_INV|
                         ARCH_PERFMON_EVENTSEL_EDGE|
                         ARCH_PERFMON_EVENTSEL_CMASK);
	val |= counter_config->extra;

The value will not be updated emmediately but at least after profiling was started.

As an alternative, maybe we better put this extra bits in the unit_mask, e.g. in bits [31:16] of the unit mask for bits [31:16] of PerfEvtSel? Then, we simply could use the current userland to set it up.

-Robert

Signature · 2 lines
Advanced Micro Devices, Inc.
Operating System Research Center
#2 lore raw
Quoted 10 lines

I would like to write the actual value back so that userland may read it:

	counter_config->extra &= (ARCH_PERFMON_EVENTSEL_INV|
                         ARCH_PERFMON_EVENTSEL_EDGE|
                         ARCH_PERFMON_EVENTSEL_CMASK);
	val |= counter_config->extra;

The value will not be updated emmediately but at least after profiling was started.

Okay.

Quoted 5 lines

As an alternative, maybe we better put this extra bits in the unit_mask, e.g. in bits [31:16] of the unit mask for bits [31:16] of PerfEvtSel? Then, we simply could use the current userland to set it up.

We can't anyways, it doesn't support it.

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only
#3 lore raw
Quoted 14 lines · Andi Kleen
Quoted 10 lines

I would like to write the actual value back so that userland may read it:

	counter_config->extra &= (ARCH_PERFMON_EVENTSEL_INV|
                         ARCH_PERFMON_EVENTSEL_EDGE|
                         ARCH_PERFMON_EVENTSEL_CMASK);
	val |= counter_config->extra;

The value will not be updated emmediately but at least after profiling was started.

Okay.

Quoted 1 line
Quoted 6 lines
Quoted 4 lines

As an alternative, maybe we better put this extra bits in the unit_mask, e.g. in bits [31:16] of the unit mask for bits [31:16] of PerfEvtSel? Then, we simply could use the current userland to set it up.

We can't anyways, it doesn't support it.

Shouldn't something like the follwing work (assuming the umask table is modified)?

opcontrol --event SOME_EVENT:100000:0x00800000 ...

(setting the INV flag)

-Robert

Signature · 2 lines
Advanced Micro Devices, Inc.
Operating System Research Center
author
#4 lore raw
Quoted 4 lines

Shouldn't something like the follwing work (assuming the umask table is modified)?

opcontrol --event SOME_EVENT:100000:0x00800000 ...

I think there was some problem with this when I tried it.

Besides I'm more interested in using named events for this; the raw numbers are quite evil. So this needs new user space in any case.

-Andi

#5 lore raw
Quoted 10 lines · Andi Kleen
Quoted 4 lines

Shouldn't something like the follwing work (assuming the umask table is modified)?

opcontrol --event SOME_EVENT:100000:0x00800000 ...

I think there was some problem with this when I tried it.

Besides I'm more interested in using named events for this; the raw numbers are quite evil. So this needs new user space in any case.

Ok, in this case we can use also /extra. So I am fine with this too.

Thanks,

-Robert

Signature · 2 lines
Advanced Micro Devices, Inc.
Operating System Research Center