Subject: Re: Request for review on scsipi changes
To: James Chacon <>
From: Manuel Bouyer <>
List: tech-kern
Date: 09/17/2002 19:44:37
On Mon, Sep 16, 2002 at 05:52:04PM -0400, James Chacon wrote:
> >
> >On Mon, Sep 16, 2002 at 01:28:44PM -0400, James Chacon wrote:
> >> [...]
> >> Attached are patches that do #2. Of the 2 sides to scsipi only the scsi end
> >> needed this as atapi doesn't have this problem. To account for "jitter" and
> >> to make sure all devices always attach in the same order a queue is used to
> >> force all probes to occur in bus attach order.
> >
> >Looks good to me.
> >Maybe there's a chance to make the 2s SCSI delay before probe global instead
> >of per-channel ? It's annoying when you have a lot of channels.
> Doesn't that work by sending a reset to the HBA and then waiting 2s? Anything
> sent to the HBA should be done in the appropriate bus's thread to avoid
> possible issues.

The bus should be probed when the hba is attached. I think the only one
which doesn't do this right now is aic7xxx (the reset is issued at the
first command, which is broken even with the current sheme).

> >The problem is that we have to keep it for hot-plug SCSI adapters.
> >I think the mechanism you implemented can help, with a global flag
> >set in scsibusattach() and cleared at the first SCSI_DELAY.
> Thats also why I think leaving it as is inside the probe is the best thing.

It's a problem when you have a lot of busses. 

Manuel Bouyer <>