Page fault, GEOM problem??

[Michael S. Eubanks]

On Mon, 2006-01-23 at 06:43 +0100, Johan Ström wrote:
> On 23 jan 2006, at 01.17, Michael S. Eubanks wrote:
> 
> 
> > On Sun, 2006-01-22 at 23:51 +0100, Johan Ström wrote:
> >
> > ...snip...
> >
> >
> >> On 22 jan 2006, at 22.58, Michael S. Eubanks wrote:
> >> This card does afaik dont have raid functionalitys (I've never read
> >> anything about it either on the web, the cards box or anywhere  
> >> else..).
> >> I'm running GENERIC, which does include ataraid..
> >> What does your dmesg identify your card as?
> >>
> >> atapci0: <Promise PDC40518 SATA150 controller> port 0xb800-0xb87f,
> >> 0xb400-0xb4ff mem 0xfb800000-0xfb800fff,0xfb000000-0xfb01ffff irq 19
> >> at device 12.0 on pci0
> >>
> >> Is it the same PDC chipset?
> >>
> >> --
> >> Johan
> >>
> >>
> >
> > No, I have a different controller.  My mistake.  I think what is
> > happening is the DMA read command is failing, therefore causing the
> > device to be disconnected, and the kernel can't write to the disk from
> > that point on (this is somewhat obvious given the output below).
> >
> >
> >>> Nov 29 20:36:54 elfi kernel: subdisk10: detached
> >>> Nov 29 20:36:54 elfi kernel: ad10: detached
> >>> Nov 29 20:36:54 elfi kernel: unknown: TIMEOUT - READ_DMA48 retrying
> >>> (1 retry left) LBA=426562704
> >>> Nov 29 20:36:54 elfi kernel: GEOM_MIRROR: Device gm0s1: provider
> >>> ad10s1 disconnected.
> >>>
> >
> > The message seen from the last line above is generated in any of the
> > following scenarios (from g_mirror.c):
> >   1. Device wasn't running yet, but disk disappear.
> >   2. Disk was active and disapppear.
> >   3. Disk disappear during synchronization process.
> >
> >
> >>> Nov 29 20:36:54 elfi kernel: GEOM_MIRROR: Request failed (error=6).
> >>> ad10s1[WRITE(offset=134356992, length=16384)]
> >>>
> >
> > As far as recovering the disk, I remember seeing something about  
> > booting
> > to single user mode and using fsck after a core dump in a previous  
> > post.
> > I'm assuming the disks worked initially and that you were able to  
> > label
> > them etc?  Is there any possibility that the disk state may be altered
> > by a power saving feature or setting in the BIOS and FreeBSD just
> > doesn't know when it happens until the next time it tries to access  
> > the
> > disk?
> >
> 
> For recovering, i've always done a direct reboot, the gmirror  
> rebuilds the mirror and fsck is run.
> No problems reading labels etc, and never has been, only problem has  
> been these sporadic crashes.. And the read/write performance (see  
> earlier in thread)...
> 
> This is a server, so all bios setting for powersaving is (should be)  
> shut of. Bios should thus never make the disk go to sleep.
> 
> Thanks for trying to help!

Wish I could be of more help. :)  Have you tried to toggle the sysctl
dma flags?  I've seen similar posts in the past with read timeouts
caused from dma being enabled.

# sysctl -a | grep dma
...
hw.ata.ata_dma: 1      <=== Try turning this one off (1 ==> 0).
hw.ata.atapi_dma: 1
...

-Michael