Re: [PATCH 4/4] libata sata_qstor conversion to new error handling (EH).

[Mark Lord <liml@rtr.ca>]

Jeff Garzik wrote:
> On Wed, Nov 07, 2007 at 06:36:13PM -0500, Mark Lord wrote:
>> Mark Lord wrote:
>> ..
>>> and SG_IO WRITEs don't work.  That's really no different from
>>> the existing 2.6.24 version.  Regular I/O looks fine.
>> ...
>>
>> I should qualify that:  SG_IO *PIO* WRITEs don't work.
>> SG_IO PIO READs do work, and SG_IO DMA R/W both work.
> 
> multi-sector or single-sector PIO?
..

I was issuing single-sector PIO writes for one sector.
Something gets confused somewhere by that.

It really didn't seem worth tracking down further,
as regular kernel initiated I/O is fine (DMA though),
and this chip doesn't have much of a user/testing base.

I suppose it has to do with the initial DRQ interrupt
from a PIO WRITE --> the chip designers actually never
tested anything other than READ and IDENTIFY in register
PIO mode, so it could even be a chip issue.

The full-fledged qstor driver uses the chip's packet mode
for *everything*, including "PIO" R/W, so this is actually
the first time that direct PIO has even been attempted.
No surprise that it might fail.

If the chip were more commonly available/used, then I would
probably update sata_qstor to use packet mode for PIO as well.
But that's more effort than it's worth, really.

Cheers