Message ID | 20190925110823.1242-3-tz.stoyanov@gmail.com (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
Series | Reset CPU mask | expand |
On Wed, 25 Sep 2019 14:08:23 +0300 "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote: > "trace-cmd reset" command should put all ftrace config to its default > state, but trace cpumask was not reseted. The patch sets cpumask to > its default value - all CPUs are enabled for tracing. > > Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> > --- > tracecmd/trace-record.c | 19 +++++++++++++++++++ > 1 file changed, 19 insertions(+) > > diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c > index 69de82a..c41f55f 100644 > --- a/tracecmd/trace-record.c > +++ b/tracecmd/trace-record.c > @@ -4096,6 +4096,24 @@ static void reset_clock(void) > write_instance_file(instance, "trace_clock", "local", "clock"); > } > > +static void reset_cpu_mask(void) > +{ > + char str[24]; > + int cpumask = 0; > + int cpus = count_cpus(); > + struct buffer_instance *instance; > + > + while (cpus--) { > + cpumask <<= 1; > + cpumask |= 1; > + } First, you can accomplish the same with: (1 << cpus) - 1; But then this only works if we have less than 32 CPUs. What we would want is: int fullwords; char *buf; int bits; int cpus; int len; fullwords = cpus / 32; bits = cpus % 32; len = (fullwords + 1) * 8; buf = malloc(len + 1); buf[0] = '\0'; if (bits) sprintf(buf, "%x", (1 << bits) - 1); while (fullwords-- > 0) strcat(buf, "ffffffff"); Because we may run this on machines with 1000s of CPUs! (BTW, I tested the above under valgrind with the following values: 1 31 32 33 126 127 128 129 with no memory leaks and the results looked good) -- Steve > + if (snprintf(str, 24, "%x", cpumask) <= 0) > + return; > + > + for_all_instances(instance) > + write_instance_file(instance, "tracing_cpumask", str, "cpumask"); > +} > + > static void reset_event_pid(void) > { > add_event_pid(""); > @@ -4808,6 +4826,7 @@ void trace_reset(int argc, char **argv) > reset_clock(); > reset_event_pid(); > reset_max_latency_instance(); > + reset_cpu_mask(); > tracecmd_remove_instances(); > clear_func_filters(); > /* restore tracing_on to 1 */
[ Added back the list, as this is good to document ] On Tue, 1 Oct 2019 14:17:08 +0200 Eugene Syromyatnikov <evgsyr@gmail.com> wrote: > On Mon, Sep 30, 2019 at 11:13 PM Steven Rostedt <rostedt@goodmis.org> wrote: > > if (bits) > > sprintf(buf, "%x", (1 << bits) - 1); > > while (fullwords-- > 0) > > strcat(buf, "ffffffff"); > Additional commas that split 32-bit chunks are needed here, if I read > lib/bitmap.c:__bitmap_parse correctly. > You're right. I just tested it: # echo fffffffff > /debug/tracing/tracing_cpumask -bash: echo: write error: Value too large for defined data type # echo f,ffffffff > /debug/tracing/tracing_cpumask # This should work then: int fullwords; char *buf; int bits; int cpus; int len; fullwords = (cpus - 1) / 32; bits = (cpus - 1) % 32 + 1; len = (fullwords + 1) * 9; buf = malloc(len + 1); buf[0] = '\0'; if (bits) sprintf(buf, "%x", (unsigned int)((1ULL << bits) - 1)); while (fullwords-- > 0) strcat(buf, ",ffffffff"); -- Steve
diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c index 69de82a..c41f55f 100644 --- a/tracecmd/trace-record.c +++ b/tracecmd/trace-record.c @@ -4096,6 +4096,24 @@ static void reset_clock(void) write_instance_file(instance, "trace_clock", "local", "clock"); } +static void reset_cpu_mask(void) +{ + char str[24]; + int cpumask = 0; + int cpus = count_cpus(); + struct buffer_instance *instance; + + while (cpus--) { + cpumask <<= 1; + cpumask |= 1; + } + if (snprintf(str, 24, "%x", cpumask) <= 0) + return; + + for_all_instances(instance) + write_instance_file(instance, "tracing_cpumask", str, "cpumask"); +} + static void reset_event_pid(void) { add_event_pid(""); @@ -4808,6 +4826,7 @@ void trace_reset(int argc, char **argv) reset_clock(); reset_event_pid(); reset_max_latency_instance(); + reset_cpu_mask(); tracecmd_remove_instances(); clear_func_filters(); /* restore tracing_on to 1 */
"trace-cmd reset" command should put all ftrace config to its default state, but trace cpumask was not reseted. The patch sets cpumask to its default value - all CPUs are enabled for tracing. Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com> --- tracecmd/trace-record.c | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)