Re: [PATCH] double flush_page_to_ram

[Andrea Arcangeli <andrea@suse.de>]

On Thu, May 09, 2002 at 04:56:43AM -0700, David S. Miller wrote:
>    From: Hugh Dickins <hugh@veritas.com>
>    Date: Thu, 9 May 2002 13:03:52 +0100 (BST)
> 
>    filemap_nopage and shmem_nopage do flush_page_to_ram before returning
>    page, but do_no_page also does flush_page_to_ram on any page it slots
>    into the user address space.  It's memory.c's business, remove it from
>    shmem and filemap (and cut outdated comment from when filemap copied).
>    
> Wrong, consider the case where we do early COW in do_no_page, you miss
> a flush on the new-new page.

so you mean we need a flush_page_to_ram also before the
copy_user_highpage to be sure we copy uptodate contents of the
pagecache? (possibly mapped writeable elsewhere in the user address
space?)

If not then I don't see how non-flushed pagecache can be mapped into
user address space.

Andrea