This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: Measure the Accept Queueing Time
- From: fche at redhat dot com (Frank Ch. Eigler)
- To: Andi Kleen <ak at suse dot de>
- Cc: systemtap at sources dot redhat dot com
- Date: 17 Feb 2006 11:59:24 -0500
- Subject: Re: Measure the Accept Queueing Time
- References: <43E28A07.1040604@tamu.edu> <200602171601.36205.ak@suse.de> <20060217161538.GF3170@redhat.com> <200602171722.54795.ak@suse.de>
ak wrote:
> interrupt 1
> calls function to be measured
> startprobe
> ... function runs....
> interrupt 2 from a different device
> calls function to be measured again
> startprobe nested
> ... function runs nested ...
> endprobe nested
> interrupt 2 ends
> ... function still runs ...
> endprobe
> [...]
If this is happening on the same CPU, then it really shouldn't be
happening at all. On one hand, at probe start time, systemtap checks
for reentrancy and aborts any nested probes during context allocation,
so the nested probe handler should never get to even start. On the
other hand, the probe entry boilerplate should explicitly disable
interrupts/preemption while a probe handler is running, to prevent
"interrupt 2" from actually dispatching. We may be missing some code
for that (bug #2293).
- FChE