On Tue, Jun 15, 2004 at 10:49:08PM +0200, Manuel Bouyer wrote:
[SNIP]
> I think this just hides the problem. We should find why we don't get an error
> while the data was not updated. I suspect it's related to a Unit Attention
> condition that would be cleared by the scsipi_test_unit_ready(), but I
> don't see how it could be a problem.
> Can you test the attached patch ? It should emit messages when a command ends
> with Unit Attention and has XS_CTL_IGNORE_MEDIA_CHANGE set, and another
> message when such a command is restarted.

Well this does not solve the original "problem".

Here is what your patch does:
NetBSD 2.0_BETA (GENERIC) #1: Wed Jun 16 00:33:34 CEST 2004
        root@voyager:/u1/src/sys/arch/sparc64/compile/GENERIC
...
sd1 at scsibus0 target 1 lun 0: <SEAGATE, ST336706LC, 010A> disk fixed
sd1(esiop0:0:1:0): got Unit Attention, returning ERESTART
sd1(esiop0:0:1:0): restarting command
sd1: 35003 MB, 26302 cyl, 4 head, 681 sec, 512 bytes/sect x 71687370 sectors
...
cd0(esiop0:0:6:0): restarting command
ch0 at scsibus1 target 4 lun 0: <QUALSTAR, RLS-4221, 0047> changer removable
ch0: 0 slots, 0 drives, 0 pickers, 0 portals
...
st0 at scsibus1 target 5 lun 0: <SONY, SDX-500C, 0204> tape removable
st0: st0(esiop1:0:5:0): restarting command
density code 49, 512-byte blocks, write-enabled
...

Cheers, Stoned.