Re: Warning on memory offline (and possible in usual migration?)

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Wed, 16 Apr 2008 11:36:42 -0700
Andrew Morton <akpm@linux-foundation.org> wrote:

> > To set a page as Uptodate, all buffers must be uptodate.
> > 
> > But *all* buffers to this page is not necessary to be uptodate, here. 
> > Then, the page can be not-up-to-date after commit-write.
> > 
> > At page offlining, all buffers on the page seems to be marked as Uptodate
> > (by printk) but the page itself isn't. This seems strange.
> > 
> > But I don't found who set Uptodate to the buffers. 
> > And why page isn't up-to-date while all buffers are marked as up-to-date.
> 
> That would imply that someone brought a buffer uptodate and didn't mark the
> page uptodate.  That can happen if a read reads the buffer from disk or
> memsets all of it.  Or if a write memsets all of it, or does
> copy_from_user() into all of it.
> 
ok, I'll pay attention to codes for read.

> > still chasing.
> 
> umm..
> 
> If you had some code which does
> 
> 	pread(fd, buf, 1, 0);
> 	pread(fd, buf, 1, 4096);
> 	pread(fd, buf, 1, 8192);
> 	pread(fd, buf, 1, 12288);
> 
> then I'd expect that each read would read a single buffer so we end up with
> four uptodate buffers, but nobody brings the entire page uptodate.
> 
> readahead will hide this most of the time by reading entire pages, but if
> for some reason readahead has collapsed the window to zero then it could
> happen.
> 
> I'd expect that you could reproduce this by disabling readahead with
> fadvise(POSIX_FADV_RANDOM) and then issuing the above four reads.
> 
Thank you for advice. I'll try.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>