Re: [PATCH v2 1/2] libata: Add support for SEND/RECEIVE FPDMA QUEUED

[Tejun Heo <tj@kernel.org>]

On Thu, Aug 08, 2013 at 11:03:55PM +0400, Sergei Shtylyov wrote:
> Hello.
> 
> On 08/08/2013 10:58 PM, Marc C wrote:
> 
> >>    How about non-AHCI FIS-based controllers?
> 
> >Right. Since it's cost prohibitive for me to test exhaustively on
> >non-AHCI FIS-based controllers, do you think it would be acceptable to
> 
>    You can mark your patch as RFT (request for testing) in this case.
> 
> >add a new ATA host flag... something like, ATA_FLAG_AHCI, which would
> >denote the controller as being AHCI-based? Then the flag could be used
> >to gate processing of READ/WRITE/SEND/RECEIVE FPDMA commands that have
> >the 'auxiliary' field set. Or, I could add a big fat warning print
> >whenever an ata_queued_cmd is passed to the drivers with a non-zero
> >'auxiliary' value.
> 
>    I was rather thinking about a flag (ATA_FLAG_FIS_BASED, maybe)
> marking a controller as FIS-based, so that libata would know whether
> it can issue the new commands using the 'auxiliary' value.

I'm not sure whether checking whether a controller is FIS based would
be enough.  sil24 is clearly FIS based but it snoops the command code
and likely to choke on commands which it doesn't know about.  I think
the best way to deal with it would be having a feature flag on the
host / port and setting it on controllers known to work.  In practice,
just enabling it on ahci's should cover most use cases from now on
anyway.

Thanks.

-- 
tejun