lkml.org 
[lkml]   [2007]   [Apr]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: Why kmem_cache_free occupy CPU for more than 10 seconds?
    On Thu, 12 Apr 2007 09:39:25 +0200 Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:

    > On Wed, 2007-04-11 at 15:30 -0700, Andrew Morton wrote:
    >
    > > There used to be a cond_resched() in invalidate_mapping_pages() which would
    > > have prevented this, but I rudely removed it to support
    > > /proc/sys/vm/drop_caches (which needs to call invalidate_inode_pages()
    > > under spinlock).
    > >
    > > We could resurrect that cond_resched() by passing in some flag, I guess.
    > > Or change the code to poke the softlockup detector. The former would be
    > > better.
    >
    > cond_resched() is conditional on __resched_legal(0)

    What's __resched_legal()?

    > which should take
    > care of being called under a spinlock.

    We only increment preempt_count() in spin_lock() if CONFIG_PREEMPT.

    > so I guess we can just reinstate the call in invalidate_mapping_pages()
    >
    > (still waiting on the compile to finish...)
    > ---
    > invalidate_mapping_pages() is called under locks (usually preemptable)
    > but can do a _lot_ of work, stick in a voluntary preemption point to
    > avoid excessive latencies (over 10 seconds was reported by softlockup).
    >
    > Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
    > ---
    > mm/truncate.c | 2 ++
    > 1 file changed, 2 insertions(+)
    >
    > Index: linux-2.6-mm/mm/truncate.c
    > ===================================================================
    > --- linux-2.6-mm.orig/mm/truncate.c
    > +++ linux-2.6-mm/mm/truncate.c
    > @@ -292,6 +292,8 @@ unsigned long invalidate_mapping_pages(s
    > pgoff_t index;
    > int lock_failed;
    >
    > + cond_resched();
    > +
    > lock_failed = TestSetPageLocked(page);

    Is deadlocky on the drop_caches path and if CONFIG_PREEMPT we'll get
    scheduling-in-spinlock warnings.

    For the blkdev_close() path the change is unneeded if CONFIG_PREEMPT and
    will fix things if !CONFIG_PREEMPT.

    We can presumably just remove the invalidate_mapping_pages() call from the
    kill_bdev() path (at least) - kill_bdev()'s truncate_inode_pages() will do
    the same thing.

    It might be time to remove that unused-for-six-years destroy_dirty_buffers
    too.

    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2007-04-12 10:23    [W:3.775 / U:0.044 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site