svn commit: r327168 - head/sys/vm

[Peter Holm]

On Mon, Dec 25, 2017 at 11:51:17AM -0800, Mark Johnston wrote:
> On Mon, Dec 25, 2017 at 06:49:46PM +0100, Peter Holm wrote:
> > On Sun, Dec 24, 2017 at 07:45:16PM +0000, Mark Johnston wrote:
> > > Author: markj
> > > Date: Sun Dec 24 19:45:16 2017
> > > New Revision: 327168
> > > URL: https://svnweb.freebsd.org/changeset/base/327168
> > > 
> > > Log:
> > >   Fix two problems with the page daemon control loop.
> > >   
> > >   Both issues caused the page daemon to erroneously go to sleep when
> > >   applications are consuming free pages at a high rate, leaving the
> > >   application threads blocked in VM_WAIT.
> > >   
> > >   1) After completing an inactive queue scan, concurrent allocations may
> > >      have prevented the page daemon from meeting the v_free_min threshold.
> > >      In this case, the page daemon was going to sleep even when the
> > >      inactive queue contained plenty of clean pages.
> > >   2) pagedaemon_wakeup() may be called without the free queues lock held.
> > >      This can lead to a lost wakeup if a call occurs after the page daemon
> > >      clears vm_pageout_wanted but before going to sleep.
> > >   
> > >   Fix 1) by ensuring that we start a new inactive queue scan immediately
> > >   if v_free_count < v_free_min after a prior scan.
> > >   
> > >   Fix 2) by adding a new subroutine, pagedaemon_wait(), called from
> > >   vm_wait() and vm_waitpfault(). It wakes up the page daemon if either
> > >   vm_pages_needed or vm_pageout_wanted is false, and atomically sleeps
> > >   on v_free_count.
> > >   
> > >   Reported by:	jeff
> > >   Reviewed by:	alc
> > >   MFC after:	2 weeks
> > >   Differential Revision:	https://reviews.freebsd.org/D13424
> > > 
> > > Modified:
> > >   head/sys/vm/vm_page.c
> > >   head/sys/vm/vm_pageout.c
> > >   head/sys/vm/vm_pageout.h
> > > 
> > 
> > I now see threads blocked in VM_WAIT?
> > 
> > https://people.freebsd.org/~pho/stress/log/mark013.txt
> 
> Hmm, I don't see anything obviously wrong here. v_free_count is 38, well
> below the minimum threshold at which threads will block in VM_WAIT. The
> page daemon was running and vm_pages_needed is 1, which is expected.
> This just looks like a case of the page daemon failing to keep up with
> memory pressure. Do you not see threads entering VM_WAIT with this
> revision reverted?

I have been running the same test on two hosts for one hours without
seeing any problems. I'll keep the test run thru the night.

-- 
Peter