lkml.org 
[lkml]   [2009]   [Mar]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: PATCH: Introduce struct vma_link_info
    Em Fri, 20 Mar 2009 21:36:29 +0100
    Peter Zijlstra <peterz@infradead.org> escreveu:

    | On Fri, 2009-03-20 at 10:34 -0300, Luiz Fernando N. Capitulino wrote:
    | > Andrew,
    | >
    | > Currently find_vma_prepare() and low-level VMA functions (eg. __vma_link())
    | > require callers to provide three parameters to return/pass "link" information
    | > (pprev, rb_link and rb_parent):
    | >
    | > static struct vm_area_struct *
    | > find_vma_prepare(struct mm_struct *mm, unsigned long addr,
    | > struct vm_area_struct **pprev, struct rb_node ***rb_link,
    | > struct rb_node ** rb_parent);
    | >
    | > With this patch callers can pass a struct vma_link_info instead:
    | >
    | > static struct vm_area_struct *
    | > find_vma_prepare(struct mm_struct *mm, unsigned long addr,
    | > struct vma_link_info *link_info);
    | >
    | > The code gets simpler and it should be better because less variables
    | > are pushed into the stack/registers. As shown by the following
    | > kernel build test:
    | >
    | > kernel real user sys
    | >
    | > 2.6.29-rc8-vanilla 1136.64 1033.38 82.88
    | > 2.6.29-rc8-linfo 1135.07 1032.44 82.92
    | >
    | > I have also ran hackbench, but I can't understand why its result
    | > indicates a regression:
    | >
    | > kernel Avarage of three runs (25 processes groups)
    | >
    | > 2.6.29.rc8-vanilla 2.03
    | > 2.6.29.rc8-linfo 2.12
    | >
    | > Rik has said to me that this could be inside error margin. So, I'm
    | > submitting the patch for inclusion.
    | >
    | > Signed-off-by: Luiz Fernando N. Capitulino <lcapitulino@mandriva.com.br>
    |
    | I'd rather we look into using the threaded RB-tree to get rid of all
    | this prev crap.

    Okay, it makes sense. Also, Eduardo has a point for the hackbench's
    regression: the patch is probably dropping some of gcc's optimizations
    on the variables that got packed into the struct (although I haven't
    checked the assembly yet).

    So, better to forget this one.

    Are there patches for the threaded tree available already?

    --
    Luiz Fernando N. Capitulino


    \
     
     \ /
      Last update: 2009-03-23 14:37    [W:0.023 / U:150.116 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site