Skip to contentSkip to conversation

powerpc/perf/hv-24x7: Check support before registering

PATCH In review Sukadev Bhattiprolu 2 participants 2 messages May 20, 2015 – Jun 4, 2015 PowerPC

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 powerpc/perf/hv-24x7: Check support before registering

Quoted 1 line

From 955102eacf035b19080dc659a15d9b8fbd8fae7f Mon Sep 17 00:00:00 2001

From: Sukadev Bhattiprolu <[email protected]>
Date: Tue, 28 Apr 2015 18:47:58 -0400
Subject: [PATCH 1/1] powerpc/perf/hv-24x7: Check support before registering
PMU

We currently try to register the 24x7 PMU unconditionally. Not all Power systems support 24x7 counters (eg: Power7). On these systems we get a backtrace during boot when trying to register the 24x7 PMU.

Check if the hypervisor supports 24x7 counters before attempting to register the 24x7 PMU.

---

Changelog[v2]
- [Michael Ellerman] Simplify check with bogus parameters.

---
 arch/powerpc/perf/hv-24x7.c |   21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
arch/powerpc/perf/hv-24x7.c +21 -0
index ec2eb20..c04a332 100644
@@ -1268,12 +1268,33 @@ static struct pmu h_24x7_pmu = {
 	.read        = h_24x7_event_read,
 };
 
+/*
+ * Return 1 if we can access the 24x7 counter catalog from the hypervisor.
+ * Return 0 otherwise.
+ */
+static bool hv_has_24x7(void)
+{
+	unsigned long hret;
+
+	hret = h_get_24x7_catalog_page(0, 0, 0);
+
+	if (hret != H_FUNCTION)
+		pr_err("Error %ld reading catalog, disabling 24x7 PMU\n", hret);
+
+	return hret == 0;
+
+}
+
+
 static int hv_24x7_init(void)
 {
 	int r;
 	unsigned long hret;
 	struct hv_perf_caps caps;
 
+	if (!hv_has_24x7())
+		return -ENODEV;
+
 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
 		pr_debug("not a virtualized system, not enabling\n");
 		return -ENODEV;
Signature · 1 line
1.7.9.5
#1 lore raw
Quoted 28 lines · Sukadev Bhattiprolu
Quoted 1 line

From 955102eacf035b19080dc659a15d9b8fbd8fae7f Mon Sep 17 00:00:00 2001

From: Sukadev Bhattiprolu <[email protected]>
Date: Tue, 28 Apr 2015 18:47:58 -0400
Subject: [PATCH 1/1] powerpc/perf/hv-24x7: Check support before registering
PMU

We currently try to register the 24x7 PMU unconditionally. Not all Power systems support 24x7 counters (eg: Power7). On these systems we get a backtrace during boot when trying to register the 24x7 PMU.

Check if the hypervisor supports 24x7 counters before attempting to register the 24x7 PMU.

---

Changelog[v2]
- [Michael Ellerman] Simplify check with bogus parameters.

---
 arch/powerpc/perf/hv-24x7.c |   21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
arch/powerpc/perf/hv-24x7.c +0 -0
index ec2eb20..c04a332 100644

arch/powerpc/perf/hv-24x7.c static struct pmu h_24x7_pmu = { · line 1273

Quoted 6 lines
arch/powerpc/perf/hv-24x7.c +3 -0
@@ -1268,12 +1268,33 @@ static struct pmu h_24x7_pmu = {
 	.read        = h_24x7_event_read,
 };
 
+/*
+ * Return 1 if we can access the 24x7 counter catalog from the hypervisor.
+ * Return 0 otherwise.

Comment is wrong.

arch/powerpc/perf/hv-24x7.c static struct pmu h_24x7_pmu = { · line 1277

Quoted 4 lines
arch/powerpc/perf/hv-24x7.c +4 -0
+ */
+static bool hv_has_24x7(void)
+{
+	unsigned long hret;

ret would be fine.

arch/powerpc/perf/hv-24x7.c static struct pmu h_24x7_pmu = { · line 1284

Quoted 7 lines
arch/powerpc/perf/hv-24x7.c +7 -0
+
+	hret = h_get_24x7_catalog_page(0, 0, 0);
+
+	if (hret != H_FUNCTION)
+		pr_err("Error %ld reading catalog, disabling 24x7 PMU\n", hret);
+
+	return hret == 0;

I don't get what you're doing here.

You check for something other than H_FUNCTION, and print, and then you just compare against 0. But I wouldn't ever expect that to return 0, because you passed it bogus args.

The logic should be:

static bool is_24x7_supported(void)

{
	if (h_get_24x7_catalog_page(0, 0, 0) == H_FUNCTION)
		return false;

	return true;
}

arch/powerpc/perf/hv-24x7.c static struct pmu h_24x7_pmu = { · line 1297

Quoted 9 lines
arch/powerpc/perf/hv-24x7.c +3 -0
 static int hv_24x7_init(void)
 {
 	int r;
 	unsigned long hret;
 	struct hv_perf_caps caps;
 
+	if (!hv_has_24x7())
+		return -ENODEV;
+

This is no good. You're doing the check, which involves a hcall, before you even check if you're running with a hypervisor (below).

arch/powerpc/perf/hv-24x7.c static struct pmu h_24x7_pmu = { · line 1300

Quoted 3 lines
arch/powerpc/perf/hv-24x7.c +0 -0
 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
 		pr_debug("not a virtualized system, not enabling\n");
 		return -ENODEV;

cheers