Re: [PATCH] SCSI: add check for host busy in scsi_run_queue()

[James Bottomley <James.Bottomley@suse.de>]

On Sun, 2010-09-19 at 20:53 +0800, Hillf Danton wrote:
> It seems that check for host busy is deserved before scanning the starved list.
> And list operation is simplified.
> 
> Signed-off-by: Hillf Danton <dhillf@gmail.com>
> ---
> 
> --- o/linux-2.6.36-rc4/drivers/scsi/scsi_lib.c	2010-09-13
> 07:07:38.000000000 +0800
> +++ m/linux-2.6.36-rc4/drivers/scsi/scsi_lib.c	2010-09-19
> 20:32:34.000000000 +0800
> @@ -399,18 +399,20 @@ static inline int scsi_host_is_busy(stru
>   */
>  static void scsi_run_queue(struct request_queue *q)
>  {
> -	struct scsi_device *sdev = q->queuedata;
> +	struct scsi_device *sdev = q->queuedata, *safe;
>  	struct Scsi_Host *shost = sdev->host;
> -	LIST_HEAD(starved_list);
>  	unsigned long flags;
> 
>  	if (scsi_target(sdev)->single_lun)
>  		scsi_single_lun_run(sdev);
> 
>  	spin_lock_irqsave(shost->host_lock, flags);
> -	list_splice_init(&shost->starved_list, &starved_list);
> -
> -	while (!list_empty(&starved_list)) {
> +	if (scsi_host_is_busy(shost)) {
> +		spin_unlock_irqrestore(shost->host_lock, flags);
> +		return;
> +	}

To be honest, there's no real gain from this micro-optimisation.

> +	list_for_each_entry_safe(sdev, safe, &shost->starved_list,
> +					starved_entry) {

And this is definitely wrong.  The list is altered in flight so the
whole purpose of the static copy is to process only what existed before
we began; otherwise you can get a livelock.

James