Skip to contentSkip to conversation

perf stat: handle Ctrl-C

tip:perfcounters/core tip-bot for Ingo Molnar 3 participants 4 messages May 15, 2009 – May 29, 2009 Documentation

lore
Share & tools
lore
#0 lore raw
Commit-ID:  58d7e993b16b62d30f8ef27757614056fe4def11
Gitweb:     http://git.kernel.org/tip/58d7e993b16b62d30f8ef27757614056fe4def11
Author:     Ingo Molnar <[email protected]>
AuthorDate: Fri, 15 May 2009 11:03:23 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Fri, 15 May 2009 12:09:54 +0200

perf stat: handle Ctrl-C

Before this change, if a long-running perf stat workload was Ctrl-C-ed, the utility exited without displaying statistics.

After the change, the Ctrl-C gets propagated into the workload (and causes its early exit there), but perf stat itself will still continue to run and will display counter results.

This is useful to run open-ended workloads, let them run for a while, then Ctrl-C them to get the stats.

[ Impact: extend perf stat with new functionality ]

Cc: Peter Zijlstra <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Corey Ashford <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
LKML-Reference: <new-submission>

---
 Documentation/perf_counter/builtin-stat.c |   16 ++++++++++++++++
 1 files changed, 16 insertions(+), 0 deletions(-)
Documentation/perf_counter/builtin-stat.c +16 -0
index cf575c3..03518d7 100644
@@ -538,8 +538,14 @@ static void process_options(int argc, char **argv)
 	}
 }
 
+static void skip_signal(int signo)
+{
+}
+
 int cmd_stat(int argc, char **argv, const char *prefix)
 {
+	sigset_t blocked;
+
 	page_size = sysconf(_SC_PAGE_SIZE);
 
 	process_options(argc, argv);
@@ -548,5 +554,15 @@ int cmd_stat(int argc, char **argv, const char *prefix)
 	assert(nr_cpus <= MAX_NR_CPUS);
 	assert(nr_cpus >= 0);
 
+	/*
+	 * We dont want to block the signals - that would cause
+	 * child tasks to inherit that and Ctrl-C would not work.
+	 * What we want is for Ctrl-C to work in the exec()-ed
+	 * task, but being ignored by perf stat itself:
+	 */
+	signal(SIGINT,  skip_signal);
+	signal(SIGALRM, skip_signal);
+	signal(SIGABRT, skip_signal);
+
 	return do_perfstat(argc, argv);
 }
#1 lore raw
Quoted 11 lines · tip-bot for Ingo Molnar

perf stat: handle Ctrl-C

Before this change, if a long-running perf stat workload was Ctrl-C-ed, the utility exited without displaying statistics.

After the change, the Ctrl-C gets propagated into the workload (and causes its early exit there), but perf stat itself will still continue to run and will display counter results.

This is useful to run open-ended workloads, let them run for a while, then Ctrl-C them to get the stats.

Unfortunately it means that if you do e.g.

$ while true; do perf stat something; done

it's impossible to kill the loop with ctrl-C. To fix this we need to make perf stat kill itself with the signal after printing the results, so bash sees the died-due-to-signal exit status and stops the loop.

Paul.

#2 lore raw
Quoted 14 lines · Paul Mackerras
Quoted 11 lines · tip-bot for Ingo Molnar

perf stat: handle Ctrl-C

Before this change, if a long-running perf stat workload was Ctrl-C-ed, the utility exited without displaying statistics.

After the change, the Ctrl-C gets propagated into the workload (and causes its early exit there), but perf stat itself will still continue to run and will display counter results.

This is useful to run open-ended workloads, let them run for a while, then Ctrl-C them to get the stats.

Quoted 7 lines

Unfortunately it means that if you do e.g.

$ while true; do perf stat something; done

it's impossible to kill the loop with ctrl-C. To fix this we need to make perf stat kill itself with the signal after printing the results, so bash sees the died-due-to-signal exit status and stops the loop.

Yep, just ran into the same..

^Z kill $! worked though, but that's not ideal.

#3 lore raw
Quoted 19 lines · Peter Zijlstra
Quoted 18 lines · Paul Mackerras
Quoted 11 lines · tip-bot for Ingo Molnar

perf stat: handle Ctrl-C

Before this change, if a long-running perf stat workload was Ctrl-C-ed, the utility exited without displaying statistics.

After the change, the Ctrl-C gets propagated into the workload (and causes its early exit there), but perf stat itself will still continue to run and will display counter results.

This is useful to run open-ended workloads, let them run for a while, then Ctrl-C them to get the stats.

Unfortunately it means that if you do e.g.

$ while true; do perf stat something; done
Quoted 7 lines
Quoted 3 lines

it's impossible to kill the loop with ctrl-C. To fix this we need to make perf stat kill itself with the signal after printing the results, so bash sees the died-due-to-signal exit status and stops the loop.

Yep, just ran into the same..

^Z kill $! worked though, but that's not ideal.

would be nice to have a fix for this - i suspect people will run into this frequently.

Ingo