Alan Cox wrote:
>> trying to fix it for some time now && finally got a machine with
>> pata_ali and non-working MWDMA2 ATAPI device on my desk a few weeks ago.
> 
> Oh good. I've been through a pile of ones that just work.
> 
> Which North and south does it have and what revision ?

Attached full output of 'lspci -nnvvvxxx' just in case.

>> Then, I made libata probing sequence very close to IDE such that no
>> reset is performed and NIEN isn't tempered with.  No go.
> 
> Ditto, and removing the taskfile writing optimisations

Yeah, I did that too.  I also converted pata_ali to use 32bit PIO for
CDB transfer.  :-/

>> So, I'm out of ideas at this point.  Any suggestions?  Ideas?
> 
> Short of doing a printk of each access to the PCI and I/O space and
> comparing them no. I ran out of ideas as well. I guess some kind of weird
> chance alignment bug but who knows, and the docs I have don't explain it.

The problem is not on the PCI bus as the harddisk works just fine.  The
drive too works fine if connected to ata_piix.  I'm thinking of ordering
a slim UDMA capable combo drive and see whether UDMA works; then, we can
tell whether the controller doesn't like ATAPI command sequence used by
libata (I made it behave the same way as IDE, but still...).

Thanks.

-- 
tejun