[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
Re: kern/45946: Kernel locks up in VMEM system
The following reply was made to PR kern/45946; it has been noted by GNATS.
From: Lars Heidieker <lars%heidieker.de@localhost>
Subject: Re: kern/45946: Kernel locks up in VMEM system
Date: Wed, 08 Feb 2012 15:09:45 +0100
On 02/08/2012 02:05 PM, Matthias Scheler wrote:
> The following reply was made to PR kern/45946; it has been noted by GNATS.
> From: Matthias Scheler <tron%zhadum.org.uk@localhost>
> To: NetBSD GNATS <gnats-bugs%NetBSD.org@localhost>
> Subject: Re: kern/45946: Kernel locks up in VMEM system
> Date: Wed, 8 Feb 2012 13:04:52 +0000
> On Wed, Feb 08, 2012 at 08:05:03AM +0000, Matthias Scheler wrote:
> > > can you check the version of src/sys/arch/i386/include/param.h if it
> > > is 1.75 because a patch sizing the arena went in 3 1/2 days ago.
> > > The behavior you see is that the kmem_arena runs out of space and the
> > > pagedaemon tries hard too free something.
> > I'm afraid that I already have version 1.75:
> > tron@lyssa:~>grep NetBSD /usr/src/sys/arch/i386/include/param.h
> > /* $NetBSD: param.h,v 1.75 2012/02/04 17:56:16 para Exp $ */
> > I've also tried to reduce the amount of memory of the virtual machine
> > to 2GB and it still locked up.
> Lars Heidieker suggested the following change:
> Index: param.h
> RCS file: /cvsroot/src/sys/arch/i386/include/param.h,v
> retrieving revision 1.75
> diff -u -r1.75 param.h
> --- param.h 4 Feb 2012 17:56:16 -0000 1.75
> +++ param.h 8 Feb 2012 13:00:59 -0000
> @@ -129,7 +129,7 @@
> * logical pages.
> #define NKMEMPAGES_MIN_DEFAULT ((8 * 1024 * 1024) >> PAGE_SHIFT)
> -#define NKMEMPAGES_MAX_DEFAULT ((280 * 1024 * 1024) >> PAGE_SHIFT)
> +#define NKMEMPAGES_MAX_DEFAULT ((360 * 1024 * 1024) >> PAGE_SHIFT)
> * Mach derived conversion macros
> This change indeed fixes the problem on my machine with 3GB and 6 cores.
> I have however no idea what the impact on different configurations in
> particukar with only little memory will be.
The problem shows up on machines with large physical memory and limited
kva, where physical memory >> kva.
We might want to consider to do less quantum caching on the
kmem_va_arena for such machines.
Main Index |
Thread Index |