Subject: Re: R10k errors (was: Re: R10k O2 won't boot)
To: Christopher SEKIYA <wileyc@rezrov.net>
From: Jared D. McNeill <jmcneill@invisible.ca>
List: port-sgimips
Date: 04/16/2007 08:09:34
On 16-Apr-07, at 2:35 AM, Christopher SEKIYA wrote:

> On Sun, Apr 15, 2007 at 11:30:12PM -0500, david l goodrich wrote:
>
>> Okay, I rolled my own kernel (crossbuilt on i386) for fun, and
>> tried checking out pkgsrc again.
>>
>> This time it rebooted spontaneously in the middle of checkout.
>
> We're still not doing bounce-buffer stuff for DMA on r10k.  The  
> behavior that
> you are observing is unfortunate but not unexpected.
>
> For fun, you might want to try the patch I posted to the list  
> awhile back
> at http://mail-index.netbsd.org/port-sgimips/2005/12/13/0000.html
>
> As Tsutsui-san pointed out, it doesn't handle all cases where DMA  
> would be
> performed below the magic 8M barrier.  However, it might be enough to
> stabilize your machine -- and give you a good starting point if you  
> wanted
> to flesh it out and thereby contribute to the project.

I was wondering about the missing wbflush() calls around the MACE  
bus_space operations myself -- is that something that we should add  
anyway (even for R5k)? Can you do this, or do you want me to?

Cheers,
Jared