perf tool: improve error handling in perf_flag_probe()

PATCH In review Yann Droneaud 2 participants 6 messages Oct 1, 2015 – Oct 2, 2015 2 messages not in the archive perf events

lore

Share & tools

lore

Yann Droneaud

author

2015-10-01 15:17 UTC #0 lore raw

v2 cover perf tool: improve error handling in perf_flag_probe()

Hi,

Following the EBUSY errors reported by Jiri Olsa [1], I've tryed to improve a bit the way perf_flag_probe() handle errors.

In case EBUSY is returned by perf_event_open(), testing the function again without PERF_FLAG_FD_CLOEXEC is meaningless: EBUSY is not related to close-on-exec flag, so there's nothing to confirm.

For other errors, not yet explicitly handled by perf_flag_probe(), it's pointless to report a second error message for the same error
code: the second check should not print an error if the error is the same as the one returned for the first check.

Changes from v1 [2]:
- resend

Changes from v0 [3]:
- rebased on top of commit 48536c9195ae ("perf tools: Fix probing for PERF_FLAG_FD_CLOEXEC flag");
- update a bit commit message;
- the previous patchset was acked by Jiri[4], but I haven't
reported Acked-by: in this updated patchset.

[1] http://lkml.kernel.org/r/[email protected]
[2] http://lkml.kernel.org/r/[email protected]
[3] http://lkml.kernel.org/r/[email protected]
[4] http://lkml.kernel.org/r/[email protected]

Yann Droneaud (2):
perf tools: shortcut PERF_FLAG_FD_CLOEXEC probing in case of EBUSY
error
perf tools: report PERF_FLAG_FD_CLOEXEC probing error once

 tools/perf/util/cloexec.c | 27 ++++++++++++++++-----------
 1 file changed, 16 insertions(+), 11 deletions(-)

Signature · 1 line

2.4.3

Yann Droneaud

author

2015-10-01 15:19 UTC #1 lore raw

PATCH v2 1/2 perf tools: shortcut PERF_FLAG_FD_CLOEXEC probing in case of EBUSY error

This patch is a simplification of the logic introduced as part of
commit 63914aca8f7e ('perf tools: Show better error message in case we fail to open counters due to EBUSY error'):
if EBUSY is reported by perf_event_open(), it will not be possible
to probe PERF_FLAG_FD_CLOEXEC, so it's safe to leave early.

It should be noted that -EBUSY errors are not really expected here
since commit 038fa0b9739d ('perf tools: Fix PERF_FLAG_FD_CLOEXEC flag probing event type open counters due to EBUSY error'):
the perf event type used now should be safe to use regarding -EBUSY
error.

Cc Adrian Hunter <[email protected]>

Cc David Ahern <[email protected]>

Cc Frederic Weisbecker <[email protected]>

Cc Jiri Olsa <[email protected]>

Cc Namhyung Kim <[email protected]>

Cc Paul Mackerras <[email protected]>

Cc Peter Zijlstra <[email protected]>

Cc Stephane Eranian <[email protected]>

Cc William Cohen <[email protected]>

Link http://lkml.kernel.org/r/[email protected] open here

Signed-off-by Yann Droneaud <[email protected]>

---
 tools/perf/util/cloexec.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

tools/perf/util/cloexec.c +6 -2

index 2babddaa2481..be226d293c91 100644
@@ -56,7 +56,11 @@ static int perf_flag_probe(void)
 		return 1;
 	}
 
-	WARN_ONCE(err != EINVAL && err != EBUSY,
+	/* ignore busy errors */
+	if (err == EBUSY)
+		return -1;
+
+	WARN_ONCE(err != EINVAL,
 		  "perf_event_open(..., PERF_FLAG_FD_CLOEXEC) failed with unexpected error %d (%s)\n",
 		  err, strerror_r(err, sbuf, sizeof(sbuf)));

@@ -74,7 +78,7 @@ static int perf_flag_probe(void)
 	if (fd >= 0)
 		close(fd);
 
-	if (WARN_ONCE(fd < 0 && err != EBUSY,
+	if (WARN_ONCE(fd < 0,
 		      "perf_event_open(..., 0) failed unexpectedly with error %d (%s)\n",
 		      err, strerror_r(err, sbuf, sizeof(sbuf))))
 		return -1;

Signature · 1 line

2.4.3

A message that is not in the archive (sent off-list or not yet received): [email protected]

Yann Droneaud

author

2015-10-01 21:32 UTC #3 lore raw

Re: [PATCH v2 1/2] perf tools: shortcut PERF_FLAG_FD_CLOEXEC probing in case of EBUSY error

Hi,

Quoted 37 lines · Jiri Olsa

Quoted 36 lines · Yann Droneaud
This patch is a simplification of the logic introduced as part of commit 63914aca8f7e ('perf tools: Show better error message in case we fail to open counters due to EBUSY error'):
if EBUSY is reported by perf_event_open(), it will not be possible to probe PERF_FLAG_FD_CLOEXEC, so it's safe to leave early.
It should be noted that -EBUSY errors are not really expected here since commit 038fa0b9739d ('perf tools: Fix PERF_FLAG_FD_CLOEXEC flag probing event type open counters due to EBUSY error'):
the perf event type used now should be safe to use regarding -EBUSY error.
Cc: Adrian Hunter <[email protected]>
Cc: David Ahern <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Stephane Eranian <[email protected]>
Cc: William Cohen <[email protected]>
Link:
http://lkml.kernel.org/r/[email protected]
Signed-off-by Yann Droneaud <[email protected]>
---
 tools/perf/util/cloexec.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
tools/perf/util/cloexec.c +0 -0
index 2babddaa2481..be226d293c91 100644
@@ -56,7 +56,11 @@ static int perf_flag_probe(void)
 		return 1;
 	}
 

Quoted 7 lines

tools/perf/util/cloexec.c static int perf_flag_probe(void) · line 61
Quoted 4 lines
tools/perf/util/cloexec.c +3 -1
-	WARN_ONCE(err != EINVAL && err != EBUSY,
+	/* ignore busy errors */
+	if (err == EBUSY)
+		return -1;
so what's now the error message in case we get EBUSY ?

None, just as it does currently.

perf_event_open_cloexec_flag() will still will return 0 just like PERF_FLAG_FD_CLOEXEC was not available.

Handling -EBUSY is left to the function which will try to make use of perf_event_open().

Regards.

Signature · 2 lines

Yann Droneaud
OPTEYA

Jiri Olsa

Acked-by

2015-10-02 13:19 UTC #4 lore raw

Re: [PATCH v2 1/2] perf tools: shortcut PERF_FLAG_FD_CLOEXEC probing in case of EBUSY error

Quoted 48 lines · Yann Droneaud

Hi,
Quoted 37 lines · Jiri Olsa
Quoted 36 lines · Yann Droneaud
This patch is a simplification of the logic introduced as part of commit 63914aca8f7e ('perf tools: Show better error message in case we fail to open counters due to EBUSY error'):
if EBUSY is reported by perf_event_open(), it will not be possible to probe PERF_FLAG_FD_CLOEXEC, so it's safe to leave early.
It should be noted that -EBUSY errors are not really expected here since commit 038fa0b9739d ('perf tools: Fix PERF_FLAG_FD_CLOEXEC flag probing event type open counters due to EBUSY error'):
the perf event type used now should be safe to use regarding -EBUSY error.
Cc: Adrian Hunter <[email protected]>
Cc: David Ahern <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Stephane Eranian <[email protected]>
Cc: William Cohen <[email protected]>
Link:
http://lkml.kernel.org/r/[email protected]
Signed-off-by Yann Droneaud <[email protected]>
---
 tools/perf/util/cloexec.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
tools/perf/util/cloexec.c +0 -0
index 2babddaa2481..be226d293c91 100644
@@ -56,7 +56,11 @@ static int perf_flag_probe(void)
 		return 1;
 	}
 
Quoted 7 lines
tools/perf/util/cloexec.c static int perf_flag_probe(void) · line 61
Quoted 4 lines
tools/perf/util/cloexec.c +3 -1
-	WARN_ONCE(err != EINVAL && err != EBUSY,
+	/* ignore busy errors */
+	if (err == EBUSY)
+		return -1;
so what's now the error message in case we get EBUSY ?

Quoted 7 lines

None, just as it does currently.
perf_event_open_cloexec_flag() will still will return 0 just like PERF_FLAG_FD_CLOEXEC was not available.
Handling -EBUSY is left to the function which will try to make use of perf_event_open().

I see, ook

Acked-by Jiri Olsa <[email protected]>

thanks,
jirka

Yann Droneaud

author

2015-10-01 15:19 UTC #5 lore raw

PATCH v2 2/2 perf tools: report PERF_FLAG_FD_CLOEXEC probing error once

In reply to Yann Droneaud

In case of failure, unrelated to PERF_FLAG_FD_CLOEXEC, perf_flag_probe() reports the error twice. For example:

  $ perf record ls
  Error:
  perf_event_open(..., PERF_FLAG_FD_CLOEXEC) failed with unexpected error 16 (Device or resource busy)
  perf_event_open(..., 0) failed unexpectedly with error 16 (Device or resource busy)

There's no need for the second error message, so this patch changes the function to only report a second error message when the two calls to perf_even_open(2) fail with different error codes.

Cc Adrian Hunter <[email protected]>

Cc David Ahern <[email protected]>

Cc Frederic Weisbecker <[email protected]>

Cc Jiri Olsa <[email protected]>

Cc Namhyung Kim <[email protected]>

Cc Paul Mackerras <[email protected]>

Cc Peter Zijlstra <[email protected]>

Cc Stephane Eranian <[email protected]>

Cc William Cohen <[email protected]>

Link http://lkml.kernel.org/r/[email protected] open here

Reported-by Jiri Olsa <[email protected]>

Signed-off-by Yann Droneaud <[email protected]>

---
 tools/perf/util/cloexec.c | 25 +++++++++++++------------
 1 file changed, 13 insertions(+), 12 deletions(-)

tools/perf/util/cloexec.c +13 ~~-12~~

index be226d293c91..552528698566 100644
@@ -26,7 +26,7 @@ static int perf_flag_probe(void)
 		.exclude_kernel = 1,
 	};
 	int fd;
-	int err;
+	int err0, err1;
 	int cpu;
 	pid_t pid = -1;
 	char sbuf[STRERR_BUFSIZE];

@@ -49,7 +49,7 @@ static int perf_flag_probe(void)
 		}
 		break;
 	}
-	err = errno;
+	err0 = errno;
 
 	if (fd >= 0) {
 		close(fd);

@@ -57,12 +57,12 @@ static int perf_flag_probe(void)
 	}
 
 	/* ignore busy errors */
-	if (err == EBUSY)
+	if (err0 == EBUSY)
 		return -1;
 
-	WARN_ONCE(err != EINVAL,
+	WARN_ONCE(err0 != EINVAL,
 		  "perf_event_open(..., PERF_FLAG_FD_CLOEXEC) failed with unexpected error %d (%s)\n",
-		  err, strerror_r(err, sbuf, sizeof(sbuf)));
+		  err0, strerror_r(err0, sbuf, sizeof(sbuf)));
 
 	/* not supported, confirm error related to PERF_FLAG_FD_CLOEXEC */
 	while (1) {

@@ -73,17 +73,18 @@ static int perf_flag_probe(void)
 		}
 		break;
 	}
-	err = errno;
+	err1 = errno;
 
-	if (fd >= 0)
+	if (fd >= 0) {
 		close(fd);
+		return 0;
+	}
 
-	if (WARN_ONCE(fd < 0,
-		      "perf_event_open(..., 0) failed unexpectedly with error %d (%s)\n",
-		      err, strerror_r(err, sbuf, sizeof(sbuf))))
-		return -1;
+	WARN_ONCE(err0 != err1,
+		  "perf_event_open(..., 0) failed unexpectedly with error %d (%s)\n",
+		  err1, strerror_r(err1, sbuf, sizeof(sbuf)));
 
-	return 0;
+	return -1;
 }
 
 unsigned long perf_event_open_cloexec_flag(void)

Signature · 1 line

2.4.3

A message that is not in the archive (sent off-list or not yet received): [email protected]

Yann Droneaud

author

2015-10-02 13:43 UTC #7 lore raw

Re: [PATCH v2 2/2] perf tools: report PERF_FLAG_FD_CLOEXEC probing error once

Quoted 11 lines · Jiri Olsa

Quoted 9 lines · Yann Droneaud
In case of failure, unrelated to PERF_FLAG_FD_CLOEXEC, perf_flag_probe() reports the error twice. For example:
  $ perf record ls
  Error:
  perf_event_open(..., PERF_FLAG_FD_CLOEXEC) failed with unexpected
error 16 (Device or resource busy)
  perf_event_open(..., 0) failed unexpectedly with error 16 (Device
or resource busy)

Quoted 3 lines

so this error message is no longer possible due to your earlier patch right?

My bad, the error used to illustrate the behavior is not well choosed.

Quoted 3 lines

could you please provide current 'before&after' output into changelog otherwise it looks ok

OK. I'm updating the message.

Regards.

Signature · 2 lines

Yann Droneaud
OPTEYA