Messages in this thread | | | Date | Tue, 16 Nov 2010 23:08:51 +0100 | From | Ingo Molnar <> | Subject | Re: [ANNOUNCE] New utility: 'trace' |
| |
* Peter Zijlstra <peterz@infradead.org> wrote:
> On Tue, 2010-11-16 at 22:59 +0100, Ingo Molnar wrote: > > > > > > Would you be interested in helping out with (and testing) such a more generic > > approach? > > how about having perf-record open a named pipe and rewriting everything that comes > in through that as an event and storing it in the buffer?
Yeah, that would work too in most cases - except if there's no VFS connection between tracer and tracee. (as it might be in the case of say chroot environments, etc.)
I think it's better to have kernel solutions that dont modify the VFS beyond the trace.data session info. A task could thus generate a user-space event without having to negotiate with the tracer.
Thanks,
Ingo
| |