Re: [PATCH] stop swapoff 3/3 OOMkill

[Hugh Dickins]

On Thu, 17 Apr 2003, Andrew Morton wrote:
> 
> __GFP_NORETRY:
> 
> 	Don't oom-kill and don't retry.   For swapoff.
> 
> I've implemented a __GFP_REPEAT, and don't like it, because it blurs the
> __GFP_REPEAT and __GFP_NOFAIL requirements.  I'll add __GFP_NORETRY and we
> can then pass that into read_swap_cache_async() and handle the error.
> 
> Sound good?

Probably not.  I did try something like that over a year ago, and it
didn't work as well as I'd expected.  One problem, I think, is that
while it's going through pages already in the swap cache, swapoff
may not need to allocate memory itself; but meanwhile other tasks
are trying to allocate memory and getting OOMkilled.  I think that
argues that swapoff does need to register itself for the duration
with the OOMkiller: a PF_ flag achieves that but a __GFP flag does not.

Hugh