Working from comments, not yet grep, ...
...
a) Mysteries: ... BLIST_NOSTARTONADD ... BLIST_BORKEN ...
BLIST_KEY ... BLIST_NO_ULD_ATTACH ... BLIST_SELECT_NO_ATN ...
...
b) Scsi Simplifications:
12 #define BLIST_NOTQ 0x020 /* Buggy Tagged
Command Queuing */
...
NOTQ is for all requests
All requests to the LUN or to the target?
All requests to the target was my guess, though clicking thru
BLIST_NOTQ might tell us more ...
Are tagged requests for different LUNs on the same target handled
separately?
Suppose instead we ask, are requests for different Lba's on the same
memory card handled separately?
We'd give the same answer: usually not, but in general it depends.
Yes the device may choose to react arbitrarily to any bits that
distinguish one request from another.
Only some devices accept multiple tagged requests simultaneously,
only some devices have multiple Lun's, only some devices service
different Lun's independently. Combine all those only's, and we
don't visit this corner often. But on visiting this corner might we
then find that a device handled tags differently for one Lun than for
another? Yes, we might, because "an address bit is an address bit is
an address bit"
... could there be simultaneous outstanding requests having the
same tag value because they are for separate LUNs?
That answer varies by transport ... and varies by host for most
transports?
Usually the host does the work of acquiring a distinct tag to
construct each request and releasing the tag for reuse only after
completing a request. If the host acquires tags from a pool per
device, then each incomplete request has a distinct tag. If the host
acquires tags from a pool per Lun, then each incomplete request of
the Lun has a distinct tag, but requests of the device may collide.
The merely standard definition of BBB transport of Scsi in Usb, for
example, runs only one request at a time thru the wire, so colliding
tags hurt only the host, never the device. Vendor-specific
definitions then vary over whether the bits of the Lun do or do not
extend the tag to avoid collision.
The T10.org Sam (Scsi Architectural Model) probably has opinions,
while the other Scsi definitions are incomplete, in disagreement, or
unstated.
Am I making much sense yet? Does logic not force these answers on us?
-
: 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
