Re: more kswapd swap death ideas

George (greerga@nidhogg.ham.muohio.edu)
Tue, 23 Jun 1998 18:25:51 -0400 (EDT)


On 22 Jun 1998, Eric W. Biederman wrote:

>There is one more case to consider. Swap cached pages.
>
>The story: A page was swapped out, it was later read in, but hasn't
>been changed so it has a swap page associated with it. These pages
>can still be removed from memory, and only kswapd can do that.

The test for !nr_swap_pages could be moved after that check, but it hasn't
mattered for me whether I did or not in my testing, nothing still runs out
of memory.

Using my !nr_swap_pages test, I've loaded the machine down to the point
where it has nothing free except 1 128k page. It prints 'Out of memory for
process' a couple times but nothing ever gets the 'virtual memory
exhausted' message (which I can't find in the source) unless I'm lucky.
I've never seen an ALT+SysRQ+M printout where it gave out that last 128k
page, even when 0 swap and 0 other pages were left.

-George

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu