[CRIU] [PATCH v3 0/2] mem: Address dropped guard pages from procfs report

[Adrian Reber]

On Thu, Jun 29, 2017 at 11:53:52PM +0300, Dmitry Safonov wrote:
> 2017-06-29 21:22 GMT+03:00 Adrian Reber <adrian at lisas.de>:
> > On Thu, Jun 29, 2017 at 08:31:20PM +0300, Dmitry Safonov wrote:
> >> 2017-06-29 19:48 GMT+03:00 Cyrill Gorcunov <gorcunov at gmail.com>:
> >> > On Thu, Jun 29, 2017 at 05:49:04PM +0200, Adrian Reber wrote:
> >> >> >
> >> >> > Hmm. Hard to say, maybe something with CONFIG_ symbols again?
> >> >> > Mind to try bisecting it?
> >> >>
> >> >> Bisecting does not really work as it does not work right now and if I
> >> >> remove the stack guard gap fix it also does not work. But I removed the
> >> >> following patchset (vdso rework):
> >> >> https://lists.openvz.org/pipermail/criu/2017-June/thread.html
> >> >>
> >> >> and I can run criu again on the RHEL kernel. The patch description talks
> >> >> about the vzkernel and
> >> >>
> >> >> """
> >> >>   I've tested the performance impact of the patches set on:
> >> >>   4.12.0-rc5+ kernel (needed arch_prctl() is available from v4.9 kernel)
> >> >>   With the current criu-dev af6399cc5 ("vma: Fix badly inherited FD
> >> >>   in filemap_open")
> >> >> """
> >> >>
> >> >> Does this mean that the RHEL kernel needs additional patches to work
> >> >> with criu-dev?
> >> >
> >> > Dima?
> >>
> >> Adrian, could you run it with the patch
> >> "criu/pie/parasite-vdso: Handle also systems without VVAR"
> >> it's on ml currently, and see if that helps?
> >
> > No, the error is still the same:
> >
> > (00.006730)pie: 24: Switched to the restorer 24
> > (00.006732)pie: 24: Error (criu/pie/parasite-vdso.c:49): vdso: BUG at criu/pi>
> > (00.006732)pie: 24: e/parasite-vdso.c:49
> > (00.008847) Error (criu/cr-restore.c:1472): 24 killed by signal 11: Segmentation fault
> > (00.008864) Error (criu/cr-restore.c:2363): Restoring FAILED.
> 
> 
> Hmm, could you try again, but after applying that patch,
> do `touch criu/pie/restorer.c` or `make clean` before
> building?
> 
> I suspect that patch should fix the issue, but the dependencies
> should be corrected for regeneration of restorer.

Yes, you are right. Rebuilding after 'make clean' fixes the errors I
have seen. ./zdtm.py run -f h,ns -a finishes now without any errors on
my RHEL kernel. Thanks!

		Adrian