On Fri, 2003-03-07 at 16:43, Mike Anderson wrote:
> James Bottomley [James.Bottomley@steeleye.com] wrote:
> 
> > @@ -702,8 +701,14 @@
> >  		 * if the result was normal, then just pass it along to the
> >  		 * upper level.
> >  		 */
> > -		if (rtn == SUCCESS)
> > +		if (rtn == SUCCESS) {
> > +			/* we don't want this command reissued, just
> > +			 * finished with the sense data, so set
> > +			 * retries to the max allowed to ensure it
> > +			 * won't get reissued */
> > +			scmd->retries = scmd->allowed;
> >  			scsi_eh_finish_cmd(scmd, done_q);
> > +		}
> >  		if (rtn != NEEDS_RETRY)
> >  			continue;
> >  
> 
> We might need to cover the case down below if we succeed on retry without
> reaching allowed.
> 
> Another option is to look into re-sending the command from the
> scsi_queue_insert handler like we do with timeouts. The interface to the
> device should be the same from the LLDD's point of view just delayed
> some.

The attached should sweep up all of this.  For NEEDS_RETRY, we will
retry until we reach the max retry count.

James