Messages in this thread | | | Subject | Re: [PATCH RESEND 0/1] AHCI: Optimize interrupt processing | From | "Nicholas A. Bellinger" <> | Date | Mon, 29 Jul 2013 12:11:06 -0700 |
| |
On Mon, 2013-07-29 at 07:50 -0400, Tejun Heo wrote: > Yo, > > On Fri, Jul 26, 2013 at 05:43:13PM -0700, Nicholas A. Bellinger wrote: > > Considering there can be more than a single ata_device hanging off each > > ata_port, the '*sdev = ap->link.device[0].sdev' in __ata_qc_from_tag() > > is definitely bogus, but I'm not sure how else to correlate > > blk-mq/scsi-mq per device descriptors to existing code expecting > > ata_port->qcmd[] descriptors to be shared across multiple devices.. > > > > Tejun..? > > I have no idea. Let's please just do simpler conversion and worry > about embedding qc's into scsi_cmnds later. libata isn't a normal > SCSI driver and has a rather its own thick midlayer doing the > impedance matching inbetween && I really don't think there is too much > benefit to be reaped from embedding qc's into scsi_cmnds. >
That is essentially the same conclusion that I came to, but wanted to at least give you a chance to comment here. ;)
So that said, I'll include a simple conversion for libata into the scsi-mq WIP branch, and folks who are interested in more detailed conversions can pursue them as separate items.
--nab
| |