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

[Nick Piggin <npiggin@suse.de>]

On Mon, May 05, 2008 at 10:28:48AM -0700, Christoph Lameter wrote:
> On Mon, 5 May 2008, Nick Piggin wrote:
> 
> > AFAIK, any filesystem which may not lock the page under read IO should
> > have PG_private set. In which case, if they don't have buffers they
> > should have a releasepage method. Otherwise, how would we ever reclaim
> > !uptodate && !buffers pages?
> 
> Hmmm.. Ok mpage.c does:
> 
> static void mpage_end_io_write(struct bio *bio, int err)
> {
>         const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
>         struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
> 
>         do {
>                 struct page *page = bvec->bv_page;
> 
>                 if (--bvec >= bio->bi_io_vec)
>                         prefetchw(&bvec->bv_page->flags);
> 
>                 if (!uptodate){
>                         SetPageError(page);
>                         if (page->mapping)
>                                 set_bit(AS_EIO, &page->mapping->flags);
>                 }
>                 end_page_writeback(page);
>         } while (bvec >= bio->bi_io_vec);
>         bio_put(bio);
> }
> 
> So it seems the page is always locked if !Uptodate.

Well if you weren't sure, you'd have to go through all filesystems because
any of them can do anything they like with their own pagecache really.
But in the end you've got the page count check which is the same as
reclaim. So I don't think there is really any doubt about this. The actual
page private / buffer migrations paths are obviously the less tested ones
in that code.

  
> > So I don't think we need this patch.
> 
> Ok.
> 
> Is there any easy way to check if any of the buffers are locked? It would 
> be good if we could skip the pages with pending I/O on the first migration 
> passes and only get to them after most of the others have been migrated. 
> The taking of the buffer locks instead of the page lock defeats the scheme 
> to defer the difficult migrations till later.

Can't you just test whether the buffers are locked?

--
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>