[Crash-utility] vmlist initialize fix

[Dave Anderson]

Kazuo Moriwaka wrote:

> Hi Dave,
>
> Crash was stopped with following messages when I try to treat very small
> dump image.
> It was caused in vmalloc_start initialise when vmalloc is not used
> (vmlist == 0).  This patch return address 0 for this case.
>
> > <readmem: c02a99a8, KVADDR, "vmlist", 4, (FOE), afdae5f4>
> > <readmem: 0, KVADDR, "first vmlist addr", 4, (ROE), afdae5f0>
> > crash: invalid kernel virtual address: 0  type: "first vmlist addr"
>
> regards,
> --
> Kazuo Moriwaka <moriwaka at valinux.co.jp>
>
> --- crash-4.0-2.31.orig/memory.c        2006-06-27 23:15:32.000000000 +0900
> +++ crash-4.0-2.31/memory.c     2006-07-10 18:24:54.000000000 +0900
> @@ -11049,10 +11049,13 @@
>          ulong vmlist, addr;
>
>          get_symbol_data("vmlist", sizeof(void *), &vmlist);
> -
> -        if (!readmem(vmlist+OFFSET(vm_struct_addr), KVADDR, &addr,
> -           sizeof(void *), "first vmlist addr", RETURN_ON_ERROR))
> -               non_matching_kernel();
> +        if (vmlist != 0x0) {
> +            if (!readmem(vmlist+OFFSET(vm_struct_addr), KVADDR, &addr,
> +               sizeof(void *), "first vmlist addr", RETURN_ON_ERROR))
> +                   non_matching_kernel();
> +        } else {
> +            addr = 0;
> +        }
>
>          return addr;
>  }
>

Hi Kazuo,

The problem is that the return vmalloc address of zero eventually
gets stored in vt->vmalloc_start, which, among a few other places,
is used here:

  #define IS_VMALLOC_ADDR(X) ((ulong)(X) >= vt->vmalloc_start)

Can you verify that setting it to zero will not cause problems in
the macro above, and the other places that it's used directly?

Upon a quick examination, it does looks safe enough in the relevant
vtop routines, but for example, the search command's use of next_kpage()
looks like it might fail.

Perhaps IS_VMALLOC_ADDRESS() itself should also verify
that vt->vmalloc_start is non-zero, and the other places that
use vt->vmalloc_start directly should be verified.  (Of course
we don't need to do this kind of check for the processors
that have hardwired vmalloc addresses).

Thanks,
  Dave





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/crash-utility/attachments/20060710/6d8c0c34/attachment.htm>