On Fri, Nov 02, 2007 at 09:33:21AM +0800, Fengguang Wu wrote:
> > I will try that with a USB disk - I hope that won't make a difference.
> 
> Thank you. I guess a reiserfs on loop file would also be OK.
> 
> > > btw, what's the exact kernel version you are running?
> > 
> > I noticed it with the kernel in the $SUBJECT, as reported by 'git
> > describe'.  I have pulled in new changesets since then.
> 
> And with the following patch applied?
> 
> ---
>  fs/reiserfs/stree.c |    3 ---
>  1 file changed, 3 deletions(-)
> 
> --- linux-2.6.24-git17.orig/fs/reiserfs/stree.c
> +++ linux-2.6.24-git17/fs/reiserfs/stree.c
> @@ -1458,9 +1458,6 @@ static void unmap_buffers(struct page *p
>  				}
>  				bh = next;
>  			} while (bh != head);
> -			if (PAGE_SIZE == bh->b_size) {
> -				cancel_dirty_page(page, PAGE_CACHE_SIZE);
> -			}
>  		}
>  	}
>  }

... and with the above patch applied.

Copying 300 MB from root (ext3) to the new file system did not trigger
the pdflush condition.  But then I did a
   cd $MOUNTPOINT && find . -exec md5sum {} \;
and that brought one cpu to 75% iowait.

I have attached my .config, if it helps.

Cheers,
florin

-- 
Bruce Schneier expects the Spanish Inquisition.
      http://geekz.co.uk/schneierfacts/fact/163