[Crash-utility] Can't open i386 vmcore with crash on 2.6.20-rc2kernels

[Dave Anderson]

Vivek Goyal wrote:

> On Wed, Jan 03, 2007 at 09:29:52AM -0500, Dave Anderson wrote:
> > >
> > > WARNING: invalid linux_banner pointer: 756e694c
> > > crash: vmlinux and vmcore do not match!
> > >
> >
> > Hi Vivek,
> >
> > It has to do with how the linux_banner symbol is declared.  The crash code
> > in question does this:
> >
> >
> >         if (!(sp = symbol_search("linux_banner")))
> >                 error(FATAL, "linux_banner symbol does not exist?\n");
> >         else if (sp->type == 'R')
> >                 linux_banner = symbol_value("linux_banner");
> >         else
> >                 get_symbol_data("linux_banner", sizeof(ulong), &linux_banner);
> >
> >         if (!IS_KVADDR(linux_banner))
> >                 error(WARNING, "invalid linux_banner pointer: %lx\n",
> >                         linux_banner);
> >
> > Up until 2.6.20-rc2 (apparently), if the linux_banner symbol
> > was in the readonly section ('R'), the string data was located
> > at the address of the linux_banner symbol.  Otherwise, the
> > linux_banner symbol contained the address of the string data.
> >
> > In your kernel, what does "nm -Bn vmlinux | grep linux_banner" show?
> > Maybe 'r' instead of 'R'?
> >
>
> Hi Dave,
>
> You are right. Following is "nm -Bn vmlinux | grep linux_banner" output.
>
> c0411000 r linux_banner
>
> I checked that in 2.6.19, following was definition of linux_banner.
>
> const char linux_banner[] =
>         "Linux version " UTS_RELEASE " (" LINUX_COMPILE_BY "@"
>         LINUX_COMPILE_HOST ") (" LINUX_COMPILER ") " UTS_VERSION "\n";
>
> Symbol table entry for linux_banner:
> c0407000   129 OBJECT  GLOBAL DEFAULT    3 linux_banner
>
> But in 2.6.20-rc, it has been made static. Hence scope of this symbol is no
> more global and now it is local, hence 'r' instead of 'R'.
>
> static const char linux_banner[] =
>         "Linux version " UTS_RELEASE
>         " (" LINUX_COMPILE_BY "@" LINUX_COMPILE_HOST ")"
>         " (" LINUX_COMPILER ")"
>         " " UTS_VERSION "\n";
>
> Symbol table entry for linux_banner:
>
> c0411000   143 OBJECT  LOCAL  DEFAULT    7 linux_banner
>
> Attached test patch works for me.
>
> Just a thought, can debug info section tell us what is the type of
> linux_banner? I mean something to differentiate between above two cases
> where linux_banner itself is a string or it contains a pointer to string.
>

I suppose so.  In the kernels I have checked, the readonly version
is a char[#], as opposed to a char * if it's in the data section.  The
reason it's been done as above is because originally the call to
kernel_init() was broken into two parts, one before the gdb session
was initialized, and again afterwards.  And the linux_banner check
used to be done in the pre-gdb timeframe.

In any case, your patch should work just fine.

Thanks,
  Dave


>
> Thanks
> Vivek
>
> Signed-off-by: Vivek Goyal <vgoyal at in.ibm.com>
> ---
>
>  kernel.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff -puN kernel.c~crash-unable-to-read-linux-banner-fix kernel.c
> --- crash-4.0-3.16/kernel.c~crash-unable-to-read-linux-banner-fix       2007-01-04 14:05:23.000000000 +0530
> +++ crash-4.0-3.16-root/kernel.c        2007-01-04 14:07:15.000000000 +0530
> @@ -480,7 +480,7 @@ verify_version(void)
>
>         if (!(sp = symbol_search("linux_banner")))
>                 error(FATAL, "linux_banner symbol does not exist?\n");
> -       else if (sp->type == 'R')
> +       else if ((sp->type == 'R') || (sp->type == 'r'))
>                 linux_banner = symbol_value("linux_banner");
>         else
>                 get_symbol_data("linux_banner", sizeof(ulong), &linux_banner);
> _