Skip to contentSkip to conversation

perf tools: rework of handling lack of sched_getcpu in old glibc

PATCH In review Petr Holasek 2 participants 3 messages May 19, 2015 perf events

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 perf tools: rework of handling lack of sched_getcpu in old glibc

Commit e1e455f4f4d3 ("perf tools: Work around lack of sched_getcpu in glibc < 2.6.") introduced weak function sched_getcpu() which should be overridden by sched_getcpu() provided by glibc >= 2.6. Unfortunately, the weak function was linked to sched_getcpu() used in perf bench numa and made the benchmark heavily relying on this function unusable.

This patch alternates weak function magic for preprocessor defines.

---
 tools/perf/util/cloexec.c | 13 +++++++++++--
 tools/perf/util/cloexec.h |  6 ------
 2 files changed, 11 insertions(+), 8 deletions(-)
tools/perf/util/cloexec.c +11 -2
index 85b5238..23b8aa9 100644
@@ -7,11 +7,20 @@
 
 static unsigned long flag = PERF_FLAG_FD_CLOEXEC;
 
-int __weak sched_getcpu(void)
+#ifdef __GLIBC_PREREQ
+#if !__GLIBC_PREREQ(2, 6)
+static int do_sched_getcpu(void)
 {
 	errno = ENOSYS;
 	return -1;
 }
+#else
+static int do_sched_getcpu(void)
+{
+	return sched_getcpu();
+}
+#endif
+#endif
 
 static int perf_flag_probe(void)
 {
@@ -27,7 +36,7 @@ static int perf_flag_probe(void)
 	pid_t pid = -1;
 	char sbuf[STRERR_BUFSIZE];
 
-	cpu = sched_getcpu();
+	cpu = do_sched_getcpu();
 	if (cpu < 0)
 		cpu = 0;
 
tools/perf/util/cloexec.h +0 -6
index 68888c2..94a5a7d 100644
@@ -3,10 +3,4 @@
 
 unsigned long perf_event_open_cloexec_flag(void);
 
-#ifdef __GLIBC_PREREQ
-#if !__GLIBC_PREREQ(2, 6)
-extern int sched_getcpu(void) __THROW;
-#endif
-#endif
-
 #endif /* __PERF_CLOEXEC_H */
Signature · 1 line
2.1.0
#1 lore raw

Em Tue, May 19, 2015 at 03:20:47PM +0200, Petr Holasek escreveu:

Quoted 7 lines

Commit e1e455f4f4d3 ("perf tools: Work around lack of sched_getcpu in glibc < 2.6.") introduced weak function sched_getcpu() which should be overridden by sched_getcpu() provided by glibc >= 2.6. Unfortunately, the weak function was linked to sched_getcpu() used in perf bench numa and made the benchmark heavily relying on this function unusable.

This patch alternates weak function magic for preprocessor defines.

With this you break the build again on RHEL5, right? That "fix" was a quickie, the good thing would be to have this sched_getcpu() thing somewhere else, far from its user, i.e. cloexec has no business doing this if there are other places (perf bench numa) that need this, right?

- Arnaldo

Quoted 58 lines
---
 tools/perf/util/cloexec.c | 13 +++++++++++--
 tools/perf/util/cloexec.h |  6 ------
 2 files changed, 11 insertions(+), 8 deletions(-)
tools/perf/util/cloexec.c +11 -2
index 85b5238..23b8aa9 100644
@@ -7,11 +7,20 @@
 
 static unsigned long flag = PERF_FLAG_FD_CLOEXEC;
 
-int __weak sched_getcpu(void)
+#ifdef __GLIBC_PREREQ
+#if !__GLIBC_PREREQ(2, 6)
+static int do_sched_getcpu(void)
 {
 	errno = ENOSYS;
 	return -1;
 }
+#else
+static int do_sched_getcpu(void)
+{
+	return sched_getcpu();
+}
+#endif
+#endif
 
 static int perf_flag_probe(void)
 {
@@ -27,7 +36,7 @@ static int perf_flag_probe(void)
 	pid_t pid = -1;
 	char sbuf[STRERR_BUFSIZE];
 
-	cpu = sched_getcpu();
+	cpu = do_sched_getcpu();
 	if (cpu < 0)
 		cpu = 0;
 
tools/perf/util/cloexec.h +0 -6
index 68888c2..94a5a7d 100644
@@ -3,10 +3,4 @@
 
 unsigned long perf_event_open_cloexec_flag(void);
 
-#ifdef __GLIBC_PREREQ
-#if !__GLIBC_PREREQ(2, 6)
-extern int sched_getcpu(void) __THROW;
-#endif
-#endif
-
 #endif /* __PERF_CLOEXEC_H */
Signature · 1 line
2.1.0
#2 lore raw
Quoted 7 lines · Arnaldo Carvalho de Melo

Em Tue, May 19, 2015 at 03:20:47PM +0200, Petr Holasek escreveu:

Quoted 6 lines

Commit e1e455f4f4d3 ("perf tools: Work around lack of sched_getcpu in glibc < 2.6.") introduced weak function sched_getcpu() which should be overridden by sched_getcpu() provided by glibc >= 2.6. Unfortunately, the weak function was linked to sched_getcpu() used in perf bench numa and made the benchmark heavily relying on this function unusable.

Quoted 9 lines
Quoted 1 line

This patch alternates weak function magic for preprocessor defines.

With this you break the build again on RHEL5, right? That "fix" was a quickie, the good thing would be to have this sched_getcpu() thing somewhere else, far from its user, i.e. cloexec has no business doing this if there are other places (perf bench numa) that need this, right?

- Arnaldo

This fix is also quickie, since perf bench numa was unusable from 4.0-rc1 and I needed to fix it somehow. But I agree, the right place for the fix should be *far* from its users.

Quoted 59 lines
Quoted 59 lines
---
 tools/perf/util/cloexec.c | 13 +++++++++++--
 tools/perf/util/cloexec.h |  6 ------
 2 files changed, 11 insertions(+), 8 deletions(-)
tools/perf/util/cloexec.c +11 -2
index 85b5238..23b8aa9 100644
@@ -7,11 +7,20 @@
 
 static unsigned long flag = PERF_FLAG_FD_CLOEXEC;
 
-int __weak sched_getcpu(void)
+#ifdef __GLIBC_PREREQ
+#if !__GLIBC_PREREQ(2, 6)
+static int do_sched_getcpu(void)
 {
 	errno = ENOSYS;
 	return -1;
 }
+#else
+static int do_sched_getcpu(void)
+{
+	return sched_getcpu();
+}
+#endif
+#endif
 
 static int perf_flag_probe(void)
 {
@@ -27,7 +36,7 @@ static int perf_flag_probe(void)
 	pid_t pid = -1;
 	char sbuf[STRERR_BUFSIZE];
 
-	cpu = sched_getcpu();
+	cpu = do_sched_getcpu();
 	if (cpu < 0)
 		cpu = 0;
 
tools/perf/util/cloexec.h +0 -6
index 68888c2..94a5a7d 100644
@@ -3,10 +3,4 @@
 
 unsigned long perf_event_open_cloexec_flag(void);
 
-#ifdef __GLIBC_PREREQ
-#if !__GLIBC_PREREQ(2, 6)
-extern int sched_getcpu(void) __THROW;
-#endif
-#endif
-
 #endif /* __PERF_CLOEXEC_H */
Signature · 1 line
2.1.0
Signature · 2 lines
Petr Holasek
[email protected]