[3383] in linux-scsi channel archive
Re: problems while converting to the new scsi code
daemon@ATHENA.MIT.EDU (Dario_Ballabio)
Sun Feb 22 03:50:39 1998
Date: Sun, 22 Feb 1998 09:48:19 +0100
From: Dario_Ballabio@milano.europe.dg.com (Dario_Ballabio)
To: linux-scsi@vger.rutgers.edu, eric@andante.jic.com
As a matter of fact what is required is that, on some specific type
of errors, the low level driver should be responsible for setting
the appropriate number of retries. 5 could be good in average, but
for example when the u14f returns error 0x92, this error is normally
recovered retrying up to 30 times. Since this behaviour is pretty
stable, why should we give up while the low level driver knows
perfectly what is going on? Obviously by "infinite-retry" I meant
only from the upper layer point of view: in this case is must be
the low level driver responsability to take a decision and possibly
give up in a finite number of attempts. This is what both
the u14-34f and the eata drivers are doing successfully since
years. As I wrote before, an alternative could be to implement
this in the upper layer, returning DID_SOFT_ERROR whit a higher
or configurable number of retries associated to it.
Cheers,
- db
