Re: [PATCH] fix id computation in scsi_eh_target_reset()

[Mike Christie <michaelc@cs.wisc.edu>]

On 10/25/2010 03:53 PM, James Bottomley wrote:
>
> diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
> index 1de30eb..484b128 100644
> --- a/drivers/scsi/scsi_error.c
> +++ b/drivers/scsi/scsi_error.c
> @@ -1156,51 +1156,40 @@ static int scsi_eh_target_reset(struct Scsi_Host *shost,
>   				struct list_head *work_q,
>   				struct list_head *done_q)
>   {
> -	struct scsi_cmnd *scmd, *tgtr_scmd, *next;
> -	unsigned int id = 0;
> -	int rtn;
> +	LIST_HEAD(tmp_list);
>
> -	do {
> -		tgtr_scmd = NULL;
> -		list_for_each_entry(scmd, work_q, eh_entry) {
> -			if (id == scmd_id(scmd)) {
> -				tgtr_scmd = scmd;
> -				break;
> -			}
> -		}
> -		if (!tgtr_scmd) {
> -			/* not one exactly equal; find the next highest */
> -			list_for_each_entry(scmd, work_q, eh_entry) {
> -				if (scmd_id(scmd)>  id&&
> -				    (!tgtr_scmd ||
> -				     scmd_id(tgtr_scmd)>  scmd_id(scmd)))
> -						tgtr_scmd = scmd;
> -			}
> -		}
> -		if (!tgtr_scmd)
> -			/* no more commands, that's it */
> -			break;
> +	list_splice(work_q,&tmp_list);


I think I am hitting multiple bugs. One might be with this patch, but 
strangely they both hit __list_add bugs in the scsi eh. I think this 
needs to be list_splice_init().

> +		list_for_each_entry_safe(scmd, next,&tmp_list, eh_entry) {
> +			if (scmd_id(scmd) != id)
> +				continue;
> +
> +			if ((rtn == SUCCESS || rtn == FAST_IO_FAIL)
> +			&&  (!scsi_device_online(scmd->device) ||
> +				 rtn == FAST_IO_FAIL || !scsi_eh_tur(scmd)))
> +				scsi_eh_finish_cmd(scmd, done_q);

Because if the target resets work and move all the commands to the 
done_q here.


> +			else
> +				/* push back on work queue for further processing */
> +				list_move(&scmd->eh_entry, work_q);
> +		}
>
>   	return list_empty(work_q);
>   }


Then this work_q is showing garbage in it from the splice I think. 
Without the list_splice_init I get that trace I cut and pasted in the 
other mail when I just force the scsi eh to run the target reset code.

In the test, the target reset succeeds. The scsi_eh_finish_cmd call 
above moves all the cmds to the done_q. But the work_q still has junk in 
it, and so we return that we need to do more work and eventually we end 
up running scsi_eh_offline_sdevs which calls scsi_eh_finish_cmd on a 
command we already moved to the done_q.