svn commit: r220387 - head/sys/vm

[Kostik Belousov]

On Wed, Apr 06, 2011 at 07:04:47PM +0200, Edward Tomasz Napiera?a wrote:
> Wiadomo?? napisana przez Garrett Cooper w dniu 2011-04-06, o godz. 18:57:
> > On Wed, Apr 6, 2011 at 9:27 AM, Edward Tomasz Napierala
> > <trasz at freebsd.org> wrote:
> >> Author: trasz
> >> Date: Wed Apr  6 16:27:04 2011
> >> New Revision: 220387
> >> URL: http://svn.freebsd.org/changeset/base/220387
> >> 
> >> Log:
> >>  In vm_daemon(), do not skip processes stopped with SIGSTOP.
> > 
> >    Did you run this by anyone else before you committed the change?
> 
> The whole racct patchset was reviewed by kib@, and I seem to remember
> that he said this might cause problems.  However, I didn't encounter
> any problems with this, neither did any person testing the patchset.
> 
> So, what's wrong with this?
I remember that I disliked the whole approach of handling RSS limits,
and still hold the same opinion.
I said something about honoring the limit at the time of page allocation
or page-in, and not `offline' as it is committed, by periodic scans
by daemon.

I do not think that r220387 is much problematic. It performs forced
pageout of usermode pages, that it. On the other hand, a commit message
is not very accurate, thread inhibitor TDI_SUSPENDED might be set e.g.
due to single-threading event, and then we could end up swapping out
the vmspace that is going to be destroyed due to exec(2).
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/svn-src-all/attachments/20110406/46741f69/attachment.pgp