Douglas Gilbert wrote:
Stefan Richter wrote:
usb-storage and sbp2 already implement the leanest possible HCIL mapping
(1 logical unit to 1 H).
Stefan,
Here is a counter example from a USB multicard reader:
[...]
That is 4 logical units to 1 host. [Since I'm often
critical of USB mass storage device SCSI support, in
this case it is good to see some thought went into the
INQUIRY string namings.]
Indeed. Thanks for the correction. It is 1 target = 1 H then. BTW I
prefer to call it "H" or "scsi_host", not "host". :-)
Sbp2 currently transforms the logical unit number into scsi core's "L"
too but allocates a separate H for each logical unit anyway AFAICS. That
is certainly not very logical. But on the other hand, real target names
and logical unit numbers are not exposed directly to userspace yet (or
is there a direct, transport-independent way?), so why bother with more
refined mapping for now.
There is an argument for the SCSI subsystem to keep
at least the "l" part of the maligned <h:c:i:l> tuple
because the "l" part really is SCSI (and should be u8[8]).
SATA-II is beginning to struggle along that path as well
with port multipliers of up to 15 devices. SCSI luns
can easily cope with that.
Absolutely. There should be "logical units" (with proper LUNs) which
belong to "targets" (with proper target names exposed to userspace).
The concepts of "host adapter" and "channel" may exist at the level of
transport/ interconnect drivers, but they may be absent on any higher level.
--
Stefan Richter
-=====-=-=-= =-=- =---=
http://arcgraph.de/sr/
-
: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
