svn commit: r265418 - head/sys/vm

[Ian Lepore]

On Tue, 2014-05-06 at 18:59 -0500, Alan Cox wrote:
> On 05/06/2014 07:39, Ian Lepore wrote:
> > On Tue, 2014-05-06 at 03:42 +0000, Alan Cox wrote:
> >> Author: alc
> >> Date: Tue May  6 03:42:04 2014
> >> New Revision: 265418
> >> URL: http://svnweb.freebsd.org/changeset/base/265418
> >>
> >> Log:
> >>   Prior to r254304, a separate function, vm_pageout_page_stats(), was used to
> >>   periodically update the reference status of the active pages.  This function
> >>   was called, instead of vm_pageout_scan(), when memory was not scarce.  The
> >>   objective was to provide up to date reference status for active pages in
> >>   case memory did become scarce and active pages needed to be deactivated.
> >>   
> >>   The active page queue scan performed by vm_pageout_page_stats() was
> >>   virtually identical to that performed by vm_pageout_scan(), and so r254304
> >>   eliminated vm_pageout_page_stats().  Instead, vm_pageout_scan() is
> >>   called with the parameter "pass" set to zero.  The intention was that when
> >>   pass is zero, vm_pageout_scan() would only scan the active queue.  However,
> >>   the variable page_shortage can still be greater than zero when memory is not
> >>   scarce and vm_pageout_scan() is called with pass equal to zero.
> >>   Consequently, the inactive queue may be scanned and dirty pages laundered
> >>   even though that was not intended by r254304.  This revision fixes that.
> >>   
> >>   Reported by:	avg
> >>   MFC after:	1 week
> >>   Sponsored by:	EMC / Isilon Storage Division
> >>
> >> Modified:
> >>   head/sys/vm/vm_pageout.c
> >>
> >> Modified: head/sys/vm/vm_pageout.c
> >> ==============================================================================
> >> --- head/sys/vm/vm_pageout.c	Tue May  6 03:38:04 2014	(r265417)
> >> +++ head/sys/vm/vm_pageout.c	Tue May  6 03:42:04 2014	(r265418)
> >> @@ -942,13 +942,15 @@ vm_pageout_scan(struct vm_domain *vmd, i
> >>  	 */
> >>  	addl_page_shortage = 0;
> >>  
> >> -	deficit = atomic_readandclear_int(&vm_pageout_deficit);
> >> -
> >>  	/*
> >>  	 * Calculate the number of pages we want to either free or move
> >>  	 * to the cache.
> >>  	 */
> >> -	page_shortage = vm_paging_target() + deficit;
> >> +	if (pass > 0) {
> >> +		deficit = atomic_readandclear_int(&vm_pageout_deficit);
> >> +		page_shortage = vm_paging_target() + deficit;
> >> +	} else
> >> +		page_shortage = deficit = 0;
> >>  
> >>  	/*
> >>  	 * maxlaunder limits the number of dirty pages we flush per scan.
> >>
> > Does this address the observation that several folks have made on
> > current@ that pages are being pushed to swap much more agressively than
> > in the past, even when the system doesn't seem short of memory?
> >
> 
> I hope so.  Please let me know if you see any difference.
> 
> I also suspect that another side-effect of r254304 is that we are
> swapping out idle processes sooner, before memory has become scarce. 
> Has anyone observed this behavior?
> 
> Alan
> 

I haven't seen the problem myself (I generally use option NO_SWAPPING on
my ARM stuff), but there was some mailing-list talk about it.  This is
the main thread I was thinking of:

http://lists.freebsd.org/pipermail/freebsd-stable/2014-April/078361.html

I wonder if this one could be related too?

http://lists.freebsd.org/pipermail/freebsd-stable/2014-January/077044.html

-- Ian