On Thu, 8 Sep 2005, James Bottomley wrote: > Actually, no, that's why we have the parallel EH states ... let me put > in the events that trigger state transitions so you can see what > happens: > > > EH thread finishes > <--------------- > > EH thread begins > ----------------> > > > <--------- > running ---------> recovery > | | > | | scsi_remove_host() called > v <---------- v > cancel ----------> recover/cancel > | | > | | scsi_remove_host finishes visibility removal > v v > del <------------ recover/del > > So the EH is allowed to activate in either running or cancel states, but > goes through its own state transition eventually coming back to del when > it finishes. Once the EH gets into recover/cancel it can never > transition back to running.
Why allow cancel -> recover/cancel? Once the device is in the cancel state, there isn't anything useful the error handler can accomplish, is there? Failed or timed-out commands should simply return an error. And if you accept that, then what point is there in distinguishing between cancel and recover/cancel? As far as I can see, the only significant difference is that in recover/cancel the error handler is running (but not accomplishing anything). Is this related to 1) below? > > At least, that's how it would work if you allow the RECOVERY -> CANCEL > > transition. Either way you end up in the correct state. So what's wrong > > with the old (current) system? > > It's just nasty on two counts: > > 1) we have an incorrect bifurcation in the state model and > 2) we never actually enforce any of the state transitions. Can you explain 1) more fully? I don't really understand what you're getting at. As for 2), what do you mean? In 2.6.13, scsi_device_set_state does not change the state if the transition is illegal. Alan Stern - To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
