lkml.org 
[lkml]   [2010]   [Mar]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [PATCH 4/4] cpuset,mm: use rwlock to protect task->mempolicy and mems_allowed
    on 2010-3-5 20:03, Paul Menage wrote:
    > On Wed, Mar 3, 2010 at 2:52 AM, Miao Xie <miaox@cn.fujitsu.com> wrote:
    >> if MAX_NUMNODES > BITS_PER_LONG, loading/storing task->mems_allowed or mems_allowed in
    >> task->mempolicy are not atomic operations, and the kernel page allocator gets an empty
    >> mems_allowed when updating task->mems_allowed or mems_allowed in task->mempolicy. So we
    >> use a rwlock to protect them to fix this probelm.
    >
    > Rather than adding locks, if the intention is just to avoid the
    > allocator seeing an empty nodemask couldn't we instead do the
    > equivalent of:
    >
    > current->mems_allowed |= new_mask;
    > current->mems_allowed = new_mask;
    >
    > i.e. effectively set all new bits in the nodemask first, and then
    > clear all old bits that are no longer in the new mask. The only
    > downside of this is that a page allocation that races with the update
    > could potentially allocate from any node in the union of the old and
    > new nodemasks - but that's the case anyway for an allocation that
    > races with an update, so I don't see that it's any worse.

    Before applying this patch, cpuset updates task->mems_allowed just like
    what you said. But the allocator is still likely to see an empty nodemask.
    This problem have been pointed out by Nick Piggin.

    The problem is following:
    The size of nodemask_t is greater than the size of long integer, so loading
    and storing of nodemask_t are not atomic operations. If task->mems_allowed
    don't intersect with new_mask, such as the first word of the mask is empty
    and only the first word of new_mask is not empty. When the allocator
    loads a word of the mask before

    current->mems_allowed |= new_mask;

    and then loads another word of the mask after

    current->mems_allowed = new_mask;

    the allocator gets an empty nodemask.

    I make a new patch to fix this problem now.
    Considering the change of task->mems_allowed is not frequent, so in the new
    patch, I use variables as a tag to indicate whether task->mems_allowed need
    be update or not. And before setting the tag, cpuset caches the new mask of
    every task at somewhere.

    When the allocator want to access task->mems_allowed, it must check updated-tag
    first. If the tag is set, the allocator enters the slow path and updates
    task->mems_allowed.

    Thanks!
    Miao



    \
     
     \ /
      Last update: 2010-03-07 03:35    [W:0.022 / U:95.652 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site