Subject: Re: Spectacular Train Wreck: WD DMA bad on tweaked Alpha...
To: Manuel Bouyer <bouyer@antioche.lip6.fr>
From: Matthew Jacob <mjacob@feral.com>
List: port-alpha
Date: 05/29/2001 10:31:58
> On Sun, May 27, 2001 at 05:15:19PM -0700, Matthew Jacob wrote:
> > 
> > (Top of Tree -current)
> > 
> > Hmm- I changed MAXBSIZE to 128*1024 && MAXPHYS to 1024*1024.....
> > 
> > but then when I was doing some large filesystem XFER stuff on a SCSI based
> > filesystem (I boot off of IDE on this PC164), apparently in the swap area, I
> > got....
> > 
> > What's a poor fella to do :-)?
> > 
> > I rather suspect that some limit checks about breaking up large transfers for
> > DMA is in order here...
> 
> IDE can't handle xfers larger than 64K, and this is a hardware limitation
> (the size of the sector count register). I suspect ATAPI doesn't have this
> limitation, though.
> I suspect some older SCSI adapters have a hardware limit too.
> 
> What we need here is get rid of MAXBSIZE & MAXPHYS as #define, and have
> a per-device info for this. Maybe get rid of minphys completely in
> physio, and remplace it by a maxphys variable in struct disk or struct
> dkdriver. Disks users could then limit their xfer against this variable.
> 
> But for now I fear we're stuck at 64k :(

I think you need to do your own breakup in IDE.

-matt