Messages in this thread | | | Date | Fri, 25 Jun 2010 05:48:04 -0400 | From | Jeff Garzik <> | Subject | Re: [PATCH 11/12] libata: use IRQ expecting |
| |
On 06/25/2010 03:44 AM, Tejun Heo wrote: > Hello, Jeff. > > On 06/25/2010 02:22 AM, Jeff Garzik wrote: >>> @@ -4972,6 +4972,8 @@ void ata_qc_complete(struct ata_queued_cmd *qc) >>> { >>> struct ata_port *ap = qc->ap; >>> >>> + unexpect_irq(ap->irq_expect, false); >>> + >>> /* XXX: New EH and old EH use different mechanisms to >>> * synchronize EH with regular execution path. >>> * >> >> Unconditional use of unexpect_irq() here seems incorrect for some cases, >> such as sata_mv's use, where ata_qc_complete() is called multiple times >> rather than a singleton ata_qc_complete_multiple() call. > > Indeed, sata_mv is calling ata_qc_complete() directly multiple times. > I still think calling unexpect_irq() from ata_qc_complete() is correct > as ata_qc_complete() is always a good indicator of completion events.
My basic point is that you are implicitly changing the entire ata_qc_complete() API, and associated underlying assumptions.
The existing assumption, since libata day #0, is that ata_qc_complete() works entirely within the scope of a single qc -- thus enabling multiple calls for a single controller interrupt. Your change greatly widens the scope to an entire port.
This isn't just an issue with sata_mv, that was just the easy example I remember off the top of my head. sata_fsl and sata_nv also make the same assumption. And it's a reasonable assumption, IMO.
I think an unexpect_irq() call is more appropriate outside ata_qc_complete().
Jeff
| |