On Wednesday 15 July 2015 11:53:09 Arnaldo Carvalho de Melo wrote:
> Executive summary:
>
> . Check the latest sources, i.e. 4.2-rc
I could also try to compile perf from git, if you tell me what branch and
remote I should use.
> . Some of the stuff you report was implemented or fixed already
>
> . Details about some of the limitations and advantages of 'perf trace' below
>
> . Including an auto-tracing that will fix one of the problems you reported
Nice, thanks for that!
<snip>
> [acme@zoo ~]$ trace -e file cat file
> Error: Invalid syscall file
> Hint: try 'perf list syscalls:sys_enter_*'
> Hint: and: 'man syscalls'
Just to be clear: Supporting the strace groups as a collection of syscalls
will be implemented? Or is that not possible, as it goes straight down to the
perf event subsystem which does not know this group?
<snip>
> Ok, that was good, using trace to trace trace and fix it, after lunch I'll
> give it a try.
Cool, so you'll add that internally such that the hoops are not necessary on
the user side to get the filename? Excellent! I also notice that
close/read/write and others /sometimes/ write out the filename:
10.404 ( 0.001 ms): kmimetypefinde/8217 read(fd: 6</etc/localtime>, buf:
0x7ffe2b6d1319, count: 15671 ) = 0
10.529 ( 0.004 ms): kmimetypefinde/8494 read(fd: 6, buf: 0x7ffc61c45b40,
count: 8192 ) = 4
This is also pretty odd.
<snip>
> > O_RDONLY flag is not properly handled (compare to the strace output
> > above).
>
> It is zero, right? yeap:
>
> #define O_RDONLY 00000000
>
> syscall arguments with a zero payload are removed trying to make the output
> more compact, and also for cases where some of the args are optional, to
> reduce the logic needed to know when things are optional, just don't print
> zeroes.
>
> This shortcut has issues, as in this case, where an exception seems to be
> needed, i.e. we should print O_RDONLY, d'accord?
Yes, that would be good to have here, I guess. In general, the closer the
output gets to `strace`, the better.
> > I'd love to see someone working on this gem, as it really promises to be a
> > good replacement for strace.
>
> Well, you're doing some work, getting some discussion going, I can code some
> more if someone like you tests it and tries to use it in your use cases.
>
> Bonus points go to whoever follows these discussions and writes some
> documentation or blogs about it ;-)
>
> Ok, now to juggle this with the eBPF enablement of perf... I can see
> some synergies, like: please, trace network traffic to host foo.bar,
> please... :-)
Sounds cool :)
--
Milian Wolff
m...@milianw.de
http://milianw.de
--
To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
