Re: Playing with SATA NCQ

[Jens Axboe <axboe@suse.de>]

On Thu, May 26 2005, Jens Axboe wrote:
> On Thu, May 26 2005, Jeff Garzik wrote:
> > > static void ahci_qc_prep(struct ata_queued_cmd *qc)
> > > {
> > > 	struct ahci_port_priv *pp = qc->ap->private_data;
> > >-	u32 opts;
> > >+	void *port_mmio = (void *) qc->ap->ioaddr.cmd_addr;
> > > 	const u32 cmd_fis_len = 5; /* five dwords */
> > >+	dma_addr_t cmd_tbl_dma;
> > >+	u32 opts;
> > >+	int offset;
> > >+
> > >+	if (qc->flags & ATA_QCFLAG_NCQ) {
> > >+		pp->sactive |= (1 << qc->tag);
> > >+
> > >+		writel(1 << qc->tag, port_mmio + PORT_SCR_ACT);
> > >+		readl(port_mmio + PORT_SCR_ACT);	/* flush */
> > >+	}
> > 
> > Wrong, you should do this in ahci_qc_issue not here.
> 
> Are you sure, I moved this on purpose? I think the reason I did this was
> the wording at the back of the the sata-ii spec (appendix b) that says
> something ala 'preset the active bit and transmit a register FIS'. Feel
> free to point me at the authoritative wording in the ACHI spec.
> 
> One thing that I definitely think _was_ wrong with the sactive bit
> before, is that you set it unconditionally of whether this was an NCQ
> command or not. The maxtor drives don't clear sactive on non-fpdma
> commands, which confused me at first.

Re-reading AHCI spec, it does indicate that you want to set SActive
after building the command. I'll move it back, but keep the conditional
of setting SActive on queued commands.

-- 
Jens Axboe