[lkml]   [2004]   [May]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    Patch in this message
    SubjectRe: [PATCH] ppc64: Fix possible race with set_pte on a present PTE

    On Wed, 26 May 2004, Benjamin Herrenschmidt wrote:
    > I think we are using ptep_establish for more than just setting those
    > 2 bits (like for setting up the new PTE in break_cow and such while
    > the current implementationL/definition is more like just setting
    > those bits and nothing else....

    You're right. We do use it on the do_wp_page() path, and there we actually
    use a whole new page in the "break_cow()" case. That case is in fact
    fundamentally different from the other ones.

    So we should probably break up the "ptep_establish()" into its two pieces,
    since the callers don't actually want to do the same thing. One really
    wants to do a "clear old one, set a totally new one", and the two other
    places want to actually update just the dirty and accessed bits.

    In fact, the only non-generic user of "ptep_establish()" (s390) didn't
    want to use the generic version exactly because of this very conceptual
    bug. It uses "ptep_clear_flush()" for the replacement case, which actually
    makes sense.

    So does it work if you do this appended patch first? This is a real
    cleanup, and I think it will allow us to get rid of the s390-specific code
    in ptep_establish(). Along with hopefully fixing your problem too.

    After this, we should be able to have a BUG() in "set_pte()" if the entry
    wasn't clear before (assuming the arch doesn't use set_pte() for the dirty
    updates etc).


    ===== mm/memory.c 1.177 vs edited =====
    --- 1.177/mm/memory.c Tue May 25 12:37:09 2004
    +++ edited/mm/memory.c Tue May 25 21:04:49 2004
    @@ -1004,7 +1004,10 @@
    flush_cache_page(vma, address);
    entry = maybe_mkwrite(pte_mkdirty(mk_pte(new_page, vma->vm_page_prot)),
    - ptep_establish(vma, address, page_table, entry, 1);
    + /* Get rid of the old entry, replace with new */
    + ptep_clear_flush(vma, address, page_table);
    + set_pte(page_table, entry);
    update_mmu_cache(vma, address, entry);

    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    More majordomo info at
    Please read the FAQ at
     \ /
      Last update: 2009-11-18 23:46    [W:0.036 / U:9.720 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site