kenneth topp wrote:
>
> This may be a sore subject, but does this new sg interface settle all the
> issues raised by this document?:
> http://www.fokus.gmd.de/research/cc/glone/employees/joerg.schilling/private/sg.text
>
> I hope so,
In any case the subject should be discussed. I'll take Joerg
Schilling's points one by one (indenting his with ">>>").
"It" refers to the "sg" SCSI device driver, most likely as
it is found in the 2.0 series of kernels.
vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv
>>> It has the following problems:
>>> It cannot see if a SCSI command could not be sent at all.
I'm not sure what is being referred to here. No one has ever
demonstrated this to me.
>>> It cannot get the SCSI status byte. [snip]
This refers to the original driver in the 2.0 series and up to 2.2.5 .
Joerg's interface, the one from 2.2.6 onwards and the new interface
yield the SCSI status value.
>>> It cannot get real DMA count of transfer. [snip]
Changes accepted into the 2.3 series of kernels make allowance
for a DMA residual count. The only adapter driver that currently
supports this is Advansys version 3.2I . Hopefully more will
follow. Here is an example:
# sg_dd if=/dev/sgb of=/dev/null count=1k bs=515
1024+0 records in
1024+0 records out
>> Non-zero sum of residual counts=3072
In this case /dev/sgb is a disk with 512 byte sectors.
>>> It cannot get number of bytes valid in auto sense data. [snip]
The new interface allows the user to specify the maximum number
of sense buffer bytes and yields the actual number output.
>>> It fetches to[o] few data in auto request sense
>>> (CCS/SCSI-2/SCSI-3 needs >= 18). [snip]
Changes accepted into the 2.3 series of kernels increase the
sense buffer held by the SCSI mid level structures from 16 to
64 bytes. Evidentally SCSI 3 in some situations outputs to
byte 56.
>>> It cannot send SCSI commands with vendor unique lengths.
This can be specified directly in the new interface and
awkwardly (via an ioctl) in kernels 2.2.10 and above. The
SCSI mid level still limits the maximum SCSI command length
to 12 bytes. Ken Preslan <[EMAIL PROTECTED]> posted a
patch to extend this to 16 bytes. If the impact isn't too
great on existing adapter drivers then it might be possible
to squeeze it into to 2.3 before before the "freeze"
solidifies.
------------------------------------------------------------
And today Joerg Schilling posted a positive note about the new
interface on the [EMAIL PROTECTED] reflector:
"The priciple idea seeems to be OK, however the actual structure
contains unneccesary members. I cleaned it up a bit:" ....
I hope this doesn't mean another round of alternate interfaces
sitting atop a common implementation :-)
Doug Gilbert
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
