On Tue, 28 Mar 2023 18:03:21 +0300
"Tzvetomir Stoyanov (VMware)" <tz.stoyanov@xxxxxxxxx> wrote:
> @@ -45,9 +46,11 @@ as a field for both events to calculate the delta in nanoseconds, or use
> *TRACEFS_TIMESTAMP_USECS* as the compare fields for both events to calculate the
> delta in microseconds. This is used as the example below.
>
> -*tracefs_synth_create*() creates the synthetic event in the system. The synthetic events apply
> -across all instances. A synthetic event must be created with *tracefs_synth_alloc*(3) before
> -it can be created.
> +*tracefs_synth_create*() creates the synthetic event in the system. By default, the synthetic
> +events are created in the top trace instance and apply across all instances.
> +The *tracefs_synth_set_instance()* API can be used to set a custom instance, where the synthetic
> +event will be created. In that case the event operates only in that instance. A synthetic event
The above is somewhat incorrect. That makes it sound like you can not
use the event elsewhere, where you most certainly can.
># cd /sys/kernel/tracing
># mkdir instances/foo
># echo 'wakeup pid_t pid; u64 lat' > synthetic_events
># echo 'hist:keys=pid:ts=common_timestamp.usecs' > instances/foo/events/sched/sched_waking/trigger
># echo 'hist:keys=next_pid:lat=common_timestamp.usecs-$ts:onmatch(sched.sched_waking).trace(wakeup,next_pid,$lat)' > instances/foo/events/sched/sched_switch/trigger
># mkdir instances/bar
># echo 1 > instances/bar/events/synthetic/wakeup/enable
># cat instances/bar/trace
# tracer: nop
#
# entries-in-buffer/entries-written: 73761/73761 #P:8
#
# _-----=> irqs-off/BH-disabled
# / _----=> need-resched
# | / _---=> hardirq/softirq
# || / _--=> preempt-depth
# ||| / _-=> migrate-disable
# |||| / delay
# TASK-PID CPU# ||||| TIMESTAMP FUNCTION
# | | | ||||| | |
<idle>-0 [001] d..4. 56208.972686: wakeup: pid=1352 lat=25
bash-843 [002] d..4. 56208.972784: wakeup: pid=1353 lat=20
<idle>-0 [004] d..4. 56208.972894: wakeup: pid=3408 lat=25
<idle>-0 [006] d..4. 56208.972937: wakeup: pid=820 lat=33
<idle>-0 [004] d..4. 56208.972951: wakeup: pid=3408 lat=23
<idle>-0 [004] d..4. 56208.972993: wakeup: pid=1355 lat=12
<idle>-0 [003] d..4. 56208.973030: wakeup: pid=1354 lat=26
Now, I am going to apply these patches, but I also will update them to
reflect the actual benefit of them. The issue I see is that you have
triggers in the toplevel (histograms). And perhaps you don't want those
histograms there, and want to keep the histograms used to create a
synthetic event in a particular instance. That makes more sense,
although it's not that much of a interference, as you can add more than
one histogram to an event.
># cat instances/foo/events/sched/sched_waking/hist
# event histogram
#
# trigger info: hist:keys=pid:vals=hitcount:ts=common_timestamp.usecs:sort=hitcount:size=2048:clock=global [active]
#
{ pid: 2 } hitcount: 1
{ pid: 41 } hitcount: 1
{ pid: 3609 } hitcount: 1
{ pid: 26 } hitcount: 1
{ pid: 3612 } hitcount: 1
{ pid: 15 } hitcount: 1
[..]
{ pid: 289 } hitcount: 1920
{ pid: 454 } hitcount: 3747
{ pid: 84 } hitcount: 4714
{ pid: 3257 } hitcount: 83897
{ pid: 1351 } hitcount: 681823
{ pid: 1357 } hitcount: 681830
{ pid: 1353 } hitcount: 681831
{ pid: 1352 } hitcount: 681834
{ pid: 1358 } hitcount: 681837
{ pid: 1356 } hitcount: 681869
{ pid: 1355 } hitcount: 681872
{ pid: 1354 } hitcount: 681878
Totals:
Hits: 5564963
Entries: 89
Dropped: 0
># cat events/sched/sched_waking/hist
>#
-- Steve
> +must be created with *tracefs_synth_alloc*(3) before it can be created.
>
|