Bug 215447

Created attachment 300217 [details]
SUSE Tumbleweed kernel with fix

notable by omission of messages (only error conditions are logged by mptscsih)

Created attachment 300218 [details]
my understanding of a typical scsi messages / error codes from the logs

Created attachment 300219 [details]
add use_10_for_ms default to sr.c

self-ping my list-email

Is anyone able to comment on the sr_mod / cdrom / mptsas issue I'm
experiencing (see details in bug)?
I fixed my issue with a one line patch, but I'd like to know if it's
correct and what I should do to have it integrated upstream.

Thanks, Chris

On Sat, 15 Jan 2022 at 11:42, <bugzilla-daemon@bugzilla.kernel.org> wrote:
>
> https://bugzilla.kernel.org/show_bug.cgi?id=215447
>
> --- Comment #4 from Chris Horler (cshorler@googlemail.com) ---
> self-ping my list-email
>
> --
> You may reply to this email to add a comment.
>
> You are receiving this mail because:
> You are watching the assignee of the bug.

Christopher,

> Is anyone able to comment on the sr_mod / cdrom / mptsas issue I'm
> experiencing (see details in bug)?  I fixed my issue with a one line
> patch, but I'd like to know if it's correct and what I should do to
> have it integrated upstream.

My concern is that switching to MODE SENSE(10) by default could break
things for other devices.

There are compelling arguments that we should use MODE SENSE(10). Most
ROM devices appear to favor it. The specs allow both but MMC3 (2001)
mentions MODE SENSE(10) as "shall implement" although it doesn't go as
far as marking it as mandatory in the SCSI command table.

In the current code we fall back to the MODE SENSE(6) command if a MODE
SENSE(10) fails. So if we change the default, unless a device hangs when
we send the 10-byte command, we should be OK.

Another option is to allow a fallback to the 10-byte command if the
6-byte command fails. We currently don't do that because MODE SENSE(6)
was required for so many years. This way we could accommodate devices
such as yours without the risk of changing the default.

The good news is that the "consumer" transports (ATA, FireWire, USB) all
use MODE SENSE(10) by default. So we are really only talking about
changing things for SPI-attached devices which typically are
well-behaved. So my hunch is that switching the default is probably OK,
although I would like the 6/10-byte fallback mechanism to work in both
directions as well.

is it risky to implement the fallback mechanism in both directions? is there a chance of forever retry? (fail 6, fail 10, fail 6, fail 10...)

scsi_lib.c:scsi_mode_sense(): ...

        if (!scsi_status_is_good(result) &&
            driver_byte(result) == DRIVER_SENSE) {
                if (scsi_sense_valid(sshdr)) {
                        if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
                            (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
                                /* 
                                 * Invalid command operation code
                                 */
                                sdev->use_10_for_ms = !sdev->use_10_for_ms;
                                goto retry;
                        }
                }
        }

I guess I could increase the default retry_count at initialisation and also decrementing it here.

> is it risky to implement the fallback mechanism in both directions? is
> there a chance of forever retry? (fail 6, fail 10, fail 6, fail 10...)

Just a heads-up that I am still working on a reasonable way to go about
this. It is not entirely trivial to perform the transition from MODE
SENSE(6) to (10) given how the SCSI disk driver works.

okay, thanks for the update!

any update on this?