Re: [with-PATCH-really] highmem deadlock removal, balancing & cleanup

[Rik van Riel]

On Fri, 25 May 2001, Linus Torvalds wrote:

> So I think I'll buy some experimentation. That HIGHMEM change is
> too ugly to live, though, I'd really like to hear more about why
> something that ugly is necessary.

If you mean the "GFP_BUFFER allocations should fail instead
of looping forever" thing, it is because:

1) GFP_BUFFER allocations are made in order to try to flush
   (and free) pages and allocate highmem pages.

2) This is the page flushing equivalent of PF_MEMALLOC, in
   the sense that we should not go and try to "recursively"
   flush more random pages until we find one that succeeds
   without an allocation; instead, we should just break the
   loop and let the caller deal with it.

If you mean the change to nr_free_buffer_pages(), that one
is needed because that function should return what the name
implies ... returning "2 GB of memory available for dirty
buffers" makes for a completely filled up ZONE_NORMAL and
processes which never get throttled for writing (and instead,
end up looping for more memory and killing performance for
the rest of the system).

regards,

Rik
--
Linux MM bugzilla: http://linux-mm.org/bugzilla.shtml

Virtual memory is like a game you can't win;
However, without VM there's truly nothing to lose...

		http://www.surriel.com/
http://www.conectiva.com/	http://distro.conectiva.com/