On Thu, 11 Oct 2001, Matthias Andree wrote:
>
> It swaps blazingly fast, but one strange observation. After the efenced
> application had died at approx 300 MB in RAM and 180 MB of swap, I had
> somewhat around 130 MB in swap and like 250 MB "USED+SHAR" as per
> xosview. That looked too high a number, so I did swapoff -av, and after
> that, I had 90 MB used. The swapoff was rather fast, compared with older
> 2.4.x vanilla kernels.
>
> It may well be a cosmetic issue, but it's irritating that switching the
> swap off looks like freeing main memory as well, one might expect pages
> are swapped back into RAM, so USED increases.
That's because when your application exits, zap_pte_range frees page and
swap for the present ptes, but only swap_free for the non-present ptes:
sometimes that brings the swap count down to 1 (still used), but that 1
corresponds to page remaining in the swap cache which could now be freed.
There's no lookup in that case, intentionally. A patch was posted a few
months ago to do so, but Linus preferred not to add such unmap overhead.
Indeed, for a while we didn't even free swap for the present ptes, but a
number of problems arose from that (maybe now fixed in other ways, but I
don't think we dare to reopen that wormcan).
In due course, when memory pressure demands, reclaim_page (Alan+Rik)
or shrink_cache (Linus+Andrea) will discover those pages and make them
available. Or, as you found, swapoff will free them: one of the reasons
swapoff is now faster is that it no longers searches mms in that case.
Hugh
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Oct 15 2001 - 21:00:46 EST