How to make sd retry a command?

How to make sd retry a command?

[Alan Stern]

I've seen a few examples of USB storage devices that return clearly bogus 
error information on occasion.  For instance, the device might return 
CHECK CONDITION status with SK=5 (Illegal request) and ASC=20 (Invalid 
command) for a WRITE_10, even though it has accepted many other READ_10 
and WRITE_10 commands previously.  I don't know why the device does this.

When this happens, rather than failing the transfer entirely it makes 
sense to retry it.  Right now sd will switch over from WRITE_10 to 
WRITE_6, which is definitely the wrong thing to do.  I can prevent that 
easily enough.  The question is how to force a retry.

What information should sd_rw_intr() pass to scsi_io_completion() to
request a retry?  Or is there some other way of doing it?

And how can the number of retries be limited, say to 3?

Alan Stern

Re: How to make sd retry a command?

[Brian King]

Alan Stern wrote:
> I've seen a few examples of USB storage devices that return clearly bogus 
> error information on occasion.  For instance, the device might return 
> CHECK CONDITION status with SK=5 (Illegal request) and ASC=20 (Invalid 
> command) for a WRITE_10, even though it has accepted many other READ_10 
> and WRITE_10 commands previously.  I don't know why the device does this.
> 
> When this happens, rather than failing the transfer entirely it makes 
> sense to retry it.  Right now sd will switch over from WRITE_10 to 
> WRITE_6, which is definitely the wrong thing to do.  I can prevent that 
> easily enough.  The question is how to force a retry.
> 
> What information should sd_rw_intr() pass to scsi_io_completion() to
> request a retry?  Or is there some other way of doing it?
> 
> And how can the number of retries be limited, say to 3?

scsi_decide_disposition in scsi_error.c is what controls whether or not
a failed command gets retried. If you know that you might want a failed
command retried, you can set DID_ERROR, which may force a retry,
depending on if the retry counter has been exceeded yet or not.

Brian

-- 
Brian King
eServer Storage I/O
IBM Linux Technology Center