[CRIU] Fwd: [PATCH] zdtm: add test maps008 for dedup on unmap vma

[Pavel Tikhomirov]

Sorry missed mailing list.

Best Regards, Tikhomirov Pavel.


---------- Forwarded message ----------
From: Pavel Tikhomirov <snorcht at gmail.com>
Date: 2014-04-01 17:36 GMT+04:00
Subject: Re: [CRIU] [PATCH] zdtm: add test maps008 for dedup on unmap vma
To: Andrew Vagin <avagin at parallels.com>



2014-04-01 17:18 GMT+04:00 Andrew Vagin <avagin at parallels.com>:

On Tue, Apr 01, 2014 at 03:10:40PM +0400, Tikhomirov Pavel wrote:
> > this test mmaps 512M area of memory, than sleep to be dumped,
> > than it munmaps 128 odd areas of size 2M of all 256 such areas
> > and in second dump process will have only 256M of memory
>
> Why do we need so big slice of memory? What is a difference between 64
> Mb and 512 Mb for this test? Will it catch more bugs with 512Mb?


For this test it doesn't matter, yes, we can do even less than 64Mb.


>  >
> > if --auto-dedup option is on, all images in the end must be
> > of size 0
> >
> > use:
> > sudo bash test/zdtm.sh -P -i 2 --auto-dedup transition/maps008
> >
> > to work need "[PATCH] dedup: add dedup on unmap vma, into
> > auto-dedup and dedup"
> >
> > Signed-off-by: Tikhomirov Pavel <snorcht at gmail.com>
> > ---
> >  test/zdtm.sh                        |  1 +
> >  test/zdtm/live/transition/Makefile  |  1 +
> >  test/zdtm/live/transition/maps008.c | 65
> +++++++++++++++++++++++++++++++++++++
> >  3 files changed, 67 insertions(+)
> >  create mode 100644 test/zdtm/live/transition/maps008.c
> >
> > diff --git a/test/zdtm.sh b/test/zdtm.sh
> > index d7f8fca..e4ed770 100755
> > --- a/test/zdtm.sh
> > +++ b/test/zdtm.sh
> > @@ -103,6 +103,7 @@ static/chroot
> >  static/chroot-file
> >  static/rtc
> >  transition/maps007
> > +transition/maps008
> >  "
> >  # Duplicate list with ns/ prefix
> >  TEST_LIST=$TEST_LIST$(echo $TEST_LIST | tr ' ' '\n' | sed 's#^#ns/#')
> > diff --git a/test/zdtm/live/transition/Makefile
> b/test/zdtm/live/transition/Makefile
> > index a4d0127..c1f7a7a 100644
> > --- a/test/zdtm/live/transition/Makefile
> > +++ b/test/zdtm/live/transition/Makefile
> > @@ -13,6 +13,7 @@ TST_NOFILE  =       \
> >               fork2           \
> >               thread-bomb     \
> >               maps007         \
> > +             maps008         \
> >
> >  TST_FILE     =       \
> >               file_read       \
> > diff --git a/test/zdtm/live/transition/maps008.c
> b/test/zdtm/live/transition/maps008.c
> > new file mode 100644
> > index 0000000..04054be
> > --- /dev/null
> > +++ b/test/zdtm/live/transition/maps008.c
> > @@ -0,0 +1,65 @@
> > +#include <unistd.h>
> > +#include <sys/mman.h>
> > +
> > +#include "zdtmtst.h"
> > +
> > +#define MEM_SIZE (1 << 29)
> > +#define PAGE_SIZE 4096
> > +
> > +const char *test_doc = "create big mapping and unmap odd blocks of size
> 2M";
> > +
> > +int main(int argc, char **argv)
> > +{
> > +     void *start, *p;
> > +     unsigned long long count = 0;
> > +     unsigned long long i;
> > +
> > +     test_init(argc, argv);
> > +
> > +     /*
> > +      * allocate workspace
> > +      */
> > +     start = mmap(NULL, MEM_SIZE, PROT_READ|PROT_WRITE, MAP_PRIVATE |
> MAP_ANONYMOUS, -1, 0);
> > +     if (start == MAP_FAILED)
> > +             return -1;
> > +
> > +     test_msg("%p-%p\n", start, start + MEM_SIZE);
> > +
> > +     /*
> > +      * dirty pages to ensure allocation
> > +      */
> > +     for (i = 0; i < MEM_SIZE/PAGE_SIZE; i++) {
> > +             char *t = start + i * PAGE_SIZE;
> > +             t[0] = 'o';
> > +     }
> > +
> > +     test_daemon();
> > +
> > +     /*
> > +      * Sleep to make 1-st dump get all workspace
> > +      */
> > +     sleep(1);
>
> test_waitsig()


Oh, now I understand. Thanks!


>
>
 > +
> > +     while (test_go()) {
>
> I don't understand this loop. Pls explain.
>

OK. and may be i need to simplify it.


>
>
> +             int ret;
> > +             unsigned long size = 512 * PAGE_SIZE;
> > +
> > +             if (count * size >= MEM_SIZE)
> > +                     break;
>

count til the end of memory, than break.


>  > +             p = start + ((count) * size);
> > +             count+=2;
>

counts step is 2, because I want unmap odd blocks: 1-st,3-rd, etc.


>  > +
> > +             /*
> > +              * unmap every odd block of size 2M
> > +              */
> > +             ret = munmap(p, size);
>
> What is an idea to unmap one slice more than once?

I unmap one slice only one time resulting memory will look like:
    |maped|unmaped|maped|unmaped|....

>  > +             if (ret == -1) {
> > +                     err("%p-%p", p, p + size);
> > +                     return -1;
> > +             }
> > +     }
> > +
> > +     test_waitsig();
> > +     pass();
> > +     return 0;
> > +}
> > --
> > 1.8.3.2
> >
> > _______________________________________________
> > CRIU mailing list
> > CRIU at openvz.org
> > https://lists.openvz.org/mailman/listinfo/criu
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openvz.org/pipermail/criu/attachments/20140414/38f75e32/attachment.html>