[parisc-linux] Re: kernel>=2.6.4-rc3 hung or panic on C1[18]0 [was: 2.6.5-rc2-pa2 boot panic on c110 :(]

[Joel Soete]

James Bottomley wrote:
> On Sun, 2004-04-18 at 09:39, Joel Soete wrote:
> 
>>I presume that wrong stuff come from the ccio_fill_pdir() or ccio_coalesce_chunks() merge with lba one?
>>But I don't have yet more accurate idea on what went wrong here (difference between functions are important).
>>
>>(would it help to rebuild this same kernel tree with gcc-3.0 32bit would help? right now is was build with latest gcc-3.3.3.)
>>
>>Grant, James any idea?
> 
> 
> Well, actually, the problem can't be in the code you cite, otherwise my
> raven wouldn't work either and it's been fine.
> 
> However, it's entirely possible that the effects of the patch are
> causing issues in the ncr driver.  What it does is correctly coalesce
> segments in the iommu.  Before this, the parisc iommus rarely did
> coalescing, so our SG lists were usually lots of page sized entities. 
> Now the individual entries can be up to 256k long.
> 
> I suspect, since your C110 has a 53c720 (using the ncr driver) and my
> C360 has a 53c875 (using sym_2) that the ncr driver can't cope with sg
> lists whose entries are so long.
> 
Yes that's also the main difference I noticed between c110 and c360.

> The problems are probably due to some sort of fixed length assumption on
> sg elements in the ncr driver.
> 
Ok, I now better understand inter-action between ccio and ncr drivers and I check in more detail this code.

Thanks a lot for all,
	Joel