lkml.org 
[lkml]   [2009]   [Apr]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [BUG][-tip] kprobes on module functions hits kernel BUG in text_poke on x86-32
    Masami Hiramatsu wrote:
    > Masami Hiramatsu wrote:
    >> Mathieu Desnoyers wrote:
    >>> * Masami Hiramatsu (mhiramat@redhat.com) wrote:
    >>>> Hi,
    >>>>
    >>>> I found text_poke() problem on x86-32 with the latest-tip tree.
    >>>> When I put a kprobe on a module function, text_poke() hit a BUG.
    >>>>
    >>>> This bug can be reproduced on x86-32, but not on x86-64.
    >>>> And inserting kprobes on a kernel-core function is OK.
    >>>>
    >>>> Thank you,
    >>>>
    >>> Hi Masami,
    >>>
    >>> OK, so text_poke safety net saves the day :)
    >>>
    >>> Basically, what we have here is the BUG_ON I have put :
    >>>
    >>> for (i = 0; i < len; i++)
    >>> BUG_ON(((char *)addr)[i] != ((char *)opcode)[i]);
    >>>
    >>> Which checks that the modification is really preceivable in the kernel
    >>> code, triggers this bug. Only for modules you say.
    >>>
    >>> It might not be this, but.. let's try something simple (this could be
    >>> completely unrelated, but won't take long to test): can you try to add a
    >>> vmalloc_sync_all() at the beginning of text_poke ? This would make sure
    >>> that lazily-populated TLB entries, which include module code and data on
    >>> x86, will be populated. I wonder if we hit this problem because
    >>> vmalloc_to_page would be returning a mapping to a yet unpopulated TLB
    >>> entry, if it is ever possible.
    >> Hmm, from the result of my test, vmalloc_sync_all() didn't change anything...
    >>
    >>> If that's not this, then I guess we have some problem with setting a
    >>> fixmap to a page returned by vmalloc on x86 32.

    Hmm, ok. AFAICS, fixmap is only for lowmem, and pkmap is only for highmem.

    So, I think we have some options;

    A) Separate text_poke into __text_poke and __text_poke_highmem. And
    use pkmap_atomic in __text_poke_highmem. This way doesn't require
    any additional change except adding KM_TEXT_POKE0/1 in km_type.

    B) Add set_fixmap_page and use it in text_poke. This will require
    changes in paravirt_ops and pgtable.c. We need to ensure there is
    no side effects.

    C) Change pkmap_atomic_prot to map lowmem only if the page's pgprot
    is different from user specified pgprot. And use it instead of
    fixmap. This also requires KM_TEXT_POKE0/1, however we can
    remove FIX_TEXT_POKE0/1.

    etc...

    I think A) is for short-term solution. I guess it will be acceptable
    for next release. But for long-term, C) might be better.

    Thank you,

    --
    Masami Hiramatsu

    Software Engineer
    Hitachi Computer Products (America) Inc.
    Software Solutions Division

    e-mail: mhiramat@redhat.com



    \
     
     \ /
      Last update: 2009-04-05 05:49    [W:2.447 / U:0.176 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site