Re: out_of_memory() heuristic broken for different mem configurations

[Marcelo Tosatti]

On Thu, 8 Nov 2001, Rik van Riel wrote:

> On Thu, 8 Nov 2001, Linus Torvalds wrote:
> > On Thu, 8 Nov 2001, Marcelo Tosatti wrote:
> > >
> > > I guess you forgot to apply the following patch on 2.4.15-pre1, right ?
> >
> > The thing is, I _really_ think it is broken.
> >
> > The way to make it fail is to have many large SHARED mappings
> 
> ISTR that you wanted swap_out() changed into something which
> only scans a portion of the ptes and doesn't have any return
> value for a related reason in early 2.4 ... ;)

Rik,

I remember Linus had a reasoning for the "scan _ALL_ ptes until success"
behaviour.

Linus, was that due to zone-specific (eg DMA shortage on bigmem machine)
shortages or ?

That _is_ one good argument (I'm still seeing the network driver not being
able to allocate memory from the DMA zone on the 16GB boxen), I think.

However, the current code breaks badly as I've tested on the 16GB boxen.