Comment # 23
on bug 13049
from Gerald Combs
(In reply to Pascal Quantin from comment #21)
> (In reply to Gerald Combs from comment #16)
> > (In reply to Pascal Quantin from comment #15)
> > > (In reply to Pascal Quantin from comment #3)
> > > > No problem running this capture on my machine. Given the size of the pcap
> > > > (28.1 MB) I guess the buildbot ran out of memory without an infinite loop.
> > >
> > > Gerald, is there a way to confirm this?
> >
> > That appears to be the case. With the default (unlimited) virtual memory
> > limit "valgrind --tool=memcheck tshark" takes a while but completes. With
> > the virtual memory limit set to 750000 (which is what valgrind-wireshark.sh
> > uses) it crashes with an allocation error.
>
> So should we remove the ulimit when fuzzing with valgrind memchek, as we did
> for ASAN?
That would mean that we don't fuzz with a memory limit in master since both the
fuzz and randpkt tests run under ASAN. If possible I'd like to keep the limit
in place in order to limit any collateral damage we might otherwise encounter
from running out of system memory (which also suggests that the Valgrind tests
should run first).
Running valgrind+tshark without a limit on fuzz-2016-10-24-11612.pcap consumes
about 792 MB. I'll check in a change to increase MAX_VMEM accordingly.
You are receiving this mail because:
- You are watching all bug changes.