Re: Problems forcing async SCSI

[email protected] (John Klos) writes:

>I then realized that both the A2091 and the Amiga 3000 use the 33C93 SCSI 
>controller. The A3000 has the -8 version of the chip, so I'm guessing it's 
>not a problematic SCSI chip.

All 33c93 chips are problematic, some are just worse.

Part of that comes from not sanitizing the input signals, which
can bring its internal state machine to invalid (and undocumented)
states.


>[    1.0000030] ahsc0 at mainbus0
>[    1.0000030] scsibus0 at ahsc0: 8 targets, 8 luns per target
>...
>[    1.0355564] scsibus0: waiting 2 seconds for devices to settle...
>[    3.5655092] ahsc0: target 1 now synchronous, period=208ns, offset=12.

Both the initiator and the target can request synchronous mode.
The -I option only prevents the initiator from starting the negotiation.

ahsc0 at mainbus0
scsibus0 at ahsc0: 8 targets, 8 luns per target

ahsc0: target 0 now synchronous, period=208ns, offset=12.
sd0 at scsibus0 target 0 lun 0: <ATA, Samsung SSD 850, 119P> disk fixed
sd0: fabricating a geometry
sd0: 119 GB, 122104 cyl, 64 head, 32 sec, 512 bytes/sect x 250069680 sectors

The SSD is behind a SCSI->ATA converter.

Sometimes the kernel still hangs when probing the device, but
if that succeeds, it is stable.

N.B. we have two 33c93 drivers in tree, I tend to believe the MI one (used
by sgimips and acorn32?) has more complete error and recovery handling, but
lacks an Amiga attachment.