Re: [RFC 1/4] libata: cache device select

[Jeff Garzik]

On 02/17/2010 08:10 AM, Alan Cox wrote:
> Avoid the device select overhead on every qc_issue (>  10uS) by caching the
> currently selected device. This shows up on profiles under load. Best case
> this costs us 10uS for the delay, worst case with a dumb interface it's
> costing us about *1mS* a command.
>
> I believe the logic here is sufficient, but would welcome some second reviews
> as its not something you want to get wrong !
>
>
> Signed-off-by: Alan Cox<alan@xxxxxxxxxxxxxxx>
> ---
>
>   drivers/ata/libata-sff.c |    8 ++++++--
>   include/linux/libata.h   |    1 +
>   2 files changed, 7 insertions(+), 2 deletions(-)
>
>
> diff --git a/drivers/ata/libata-sff.c b/drivers/ata/libata-sff.c
> index 63d9c6a..cf0332a 100644
> --- a/drivers/ata/libata-sff.c
> +++ b/drivers/ata/libata-sff.c
> @@ -469,6 +469,7 @@ void ata_sff_dev_select(struct ata_port *ap, unsigned int device)
>
>   	iowrite8(tmp, ap->ioaddr.device_addr);
>   	ata_sff_pause(ap);	/* needed; also flushes, for mmio */
> +	ap->sff_selected = device;
>   }
>   EXPORT_SYMBOL_GPL(ata_sff_dev_select);
>
> @@ -1538,7 +1539,8 @@ unsigned int ata_sff_qc_issue(struct ata_queued_cmd *qc)
>   	}
>
>   	/* select the device */
> -	ata_dev_select(ap, qc->dev->devno, 1, 0);
> +	if (qc->dev->devno != ap->sff_selected)
> +        	ata_dev_select(ap, qc->dev->devno, 1, 0);
>
>   	/* start the command */
>   	switch (qc->tf.protocol) {

My main worry here is that this logic excises the 150ms wait in 
ata_dev_select() that has been used effectively to allow ATAPI devices 
to "collect themselves" after waiting for idle, prior to command issuance.

	Jeff




--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/