Re: [LTP] [PATCH] patch for mincore01

[Bruce D. <bru...@gm...>]

Doug Chapman wrote:
> > On Tue, 2007-04-24 at 13:50 -0500, Bruce Dubbs wrote:
> >
>> >> In any case, I downloaded ltp-full-20070331.tgz and tried to apply
your
>> >> patch and got:
>> >>
>> >> $ patch -Np1 -i ../mincore01.patch
>> >> patching file testcases/kernel/syscalls/mincore/mincore01.c
>> >> Hunk #1 FAILED at 27.
>> >> Hunk #3 FAILED at 95.
>> >> Hunk #4 FAILED at 194.
>> >> 3 out of 4 hunks FAILED -- saving rejects to file
>> >> testcases/kernel/syscalls/mincore/mincore01.c.rej
> >
> > odd, it applies cleanly for me:
> >
> > [root@kona1 ~]# rm -rf ltp-full-20070331
> > [root@kona1 ~]# tar zxf ltp-full-20070331.tgz
> > [root@kona1 ~]# cd ltp-full-20070331
> > [root@kona1 ltp-full-20070331]# patch -p1 < ../mincore01.patch
> > patching file testcases/kernel/syscalls/mincore/mincore01.c
> > [root@kona1 ltp-full-20070331]#
> >
> >
> > The patch likely got mangled by your email client.  I took the mail and
> > saved it as a text file and tried it that way as well and it still
> > applied cleanly.

I did use your method and it applied cleanly.  I'm sure it was me, but
I'm not sure why it failed.  In any case, its OK.

> > I saw your email on that when I went digging through the archives.  I
> > agree that in theory this case is possible but since it maps the memory
> > itself and then runs the test the only way it could fail is if somehow
> > it allocated another page and it just happened to end up at the same
> > virtual address and the page we just unmapped.
> >
> > Have you actually seen the test fail (i.e. mincore pass unexpectedly) in
> > this way?

Yes, but it doesn't always fail.  However, I have had it fail and I got
others to confirm it.

See the thread at starting at:

http://linuxfromscratch.org/pipermail/lfs-dev/2007-March/059128.html

As I dug through the kernel source to see what was happening, there is
no tie between memory being in core and a specific process instance that
I could see.  The 'unexpected pass' seemed to be specific to IA32 builds.

I sent in a patch to LKML to change the functionality of the mincore
system call (tie the call to a single mmapped file), but it was
rejected.  The -21 kernel will return success to an anonymous mmapped
region.  That will certainly increase the chances that the test will
unexpectedly pass.

http://lkml.org/lkml/2007/3/22/385
http://lkml.org/lkml/2007/3/25/90

  -- Bruce