Re: kern/54818: 9.0_RC1 pagedaemon spins

[Andrew Doran <ad%netbsd.org@localhost>]

The following reply was made to PR kern/54818; it has been noted by GNATS.

From: Andrew Doran <ad%netbsd.org@localhost>
To: gnats-bugs%netbsd.org@localhost, tsutsui%ceres.dti.ne.jp@localhost
Cc: martin%netbsd.org@localhost
Subject: Re: kern/54818: 9.0_RC1 pagedaemon spins
Date: Wed, 15 Jan 2020 13:19:48 +0000

 On Thu, Jan 02, 2020 at 10:30:01AM +0000, Izumi Tsutsui wrote:
 
 > The following reply was made to PR kern/54818; it has been noted by GNATS.
 > 
 > From: Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>
 > To: ad%netbsd.org@localhost
 > Cc: gnats-bugs%netbsd.org@localhost, tsutsui%ceres.dti.ne.jp@localhost
 > Subject: Re: kern/54818: 9.0_RC1 pagedaemon spins
 > Date: Thu, 2 Jan 2020 19:25:00 +0900
 > 
 >  >  Could you please try the following?
 >  >  
 >  >  	http://www.netbsd.org/~ad/2020/pdaemon-9.diff
 >  
 >  I guess uvm/uvm_pdpolicy_clockpro.c part (empty uvmpdpol_scanfini())
 >  isn't necessary, right?
 
 It's needed to make it compile, but it does not work on 9.0 or -current.
   
 >  Unfortunately the spin still occurs with the above diff on load:
 >  (Firefox with HTML5 video + build.sh -j4 release + serial terminal output)
 
 Thank you for the vmstat output, that was very helpful.  I think it trying
 to balance the queues but your other processes are competing with it, and
 because of uvm_pageqlock nobody can make progress.  This is not a problem on
 -current any more because uvm_pageqlock is gone..  I have updated the patch
 to try and help this situation.  I have not compiled it because I don't have
 a 9.0 system at the moment.  Could you please try it again?
 
 	http://www.netbsd.org/~ad/2020/uvm-9.diff 
 
 Thanks,
 Andrew