Re: [BUG] PDC20268 crashing during DMA setup on stock Debian 2.6.12-1-powerpc

[Kyle Moffett <mrmacman_g4@mac.com>]

On Oct 19, 2005, at 18:13:44, Benjamin Herrenschmidt wrote:
> On Wed, 2005-10-19 at 13:48 -0400, Kyle Moffett wrote:
>
> > Do you have any other ideas WRT this bug?  I've been browsing  
> > around in the code a bit, and I plan to try diffing my 2.6.8.1  
> > version of the files against the latest Debian to see what  
> > changed, although I suspect it will be a relatively fat hunk of  
> > changes.  Thanks for your
> > help!
>
> Nope. The lspci output looks perfectly normal. I looks like a  
> mixture of issues with BM DMA being disabled for a reason I haven't  
> figured out and then the code crashing because it doesn't like BM- 
> DMA being disabled ...

BusMaster-DMA definitely should be enabled on that card.  After a lot  
of looking through icky IDE code, I've determined that the reason for  
the crash is that if there is a "mate", or another IDE bus on the  
same card, then hwif->dma_master is set to hwif->mate->dma_base on  
the secondary channel.  Since DMA explicitly wasn't enabled on the  
primary channel, hwif->dma_master on the secondary is 0 even though  
dma is enabled, and therefore we hit that BUG().

Therefore it seems that the only issue is ide_get_or_set_dma_base is  
returning 0 when it should return a valid DMA base.  In that  
function, the only ways that function can even theoretically return 0  
without printing any weird error messages is "if (hwif->mmio)" or "if  
(hwif->mate && hwif->mate->dma_base)".  The latter can't happen  
before hwif->mate is set up (since the problem is while initializing  
the primary).  Could hwif->mmio be nonzero somehow?  The only drivers  
that seem to set it are pci/sgiioc4, pci/siimage, ppc/pmac, and a  
couple misc arch drivers.

I see a couple theoretical possibilities:
  * hwif->mate and hwif->mate->dma_base are set for the primary  
while still initializing it (before any secondary is set up.
  * hwif->mmio is set somehow even though it shouldn't be, is the  
value ever pre-initialized to 0?

> Again, best is you pour printk's all over setup-pci.c and ide-dma.c  
> to figure out what's going on...

I wish I could but the machine is remote and in-production, so it's  
hard to have time to do much with it.  I'm trying as best I can to  
walk through the sources by hand, specifically with regards to  
changes between the two.  I'm hoping I can come up with a good enough  
guess by Thanksgiving, so that when I have a week near the server to  
test things out I can make significant progress.

Cheers,
Kyle Moffett

--
I have yet to see any problem, however complicated, which, when you  
looked at it in the right way, did not become still more complicated.
  -- Poul Anderson



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/