[Bug 191717] [iscsi] smartctl -H gives "ATA output registers missing" for a disk using the isci driver

[Jim Harris]

On Tue, Jul 8, 2014 at 10:43 AM, Douglas Gilbert <dgilbert at interlog.com>
wrote:

> On 14-07-08 07:45 AM, Martin Simmons wrote:
>
>> On Mon, 07 Jul 2014 23:12:27 -0400, Douglas Gilbert said:
>>>>>>>
>>>>>>
>>> On 14-07-07 09:45 PM, bugzilla-noreply at freebsd.org wrote:
>>>
>>>> https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=191717
>>>>
>>>> Mark Linimon <linimon at FreeBSD.org> changed:
>>>>
>>>>              What    |Removed                     |Added
>>>> ------------------------------------------------------------
>>>> ----------------
>>>>              Assignee|freebsd-bugs at FreeBSD.org    |freebsd-scsi at
>>>> FreeBSD.org
>>>>               Summary|smartctl -H gives "ATA      |[iscsi] smartctl -H
>>>> gives
>>>>                      |output registers missing"   |"ATA output registers
>>>>                      |for a disk using the isci   |missing" for a disk
>>>> using
>>>>                      |driver                      |the isci driver
>>>>
>>>> --- Comment #1 from Mark Linimon <linimon at FreeBSD.org> ---
>>>> Over to maintainers.
>>>>
>>>>
>>> At the point of failure "whatever" produces this SCSI sense data:
>>>      f0 00 01 00 50 40 00 00  00 c2 4f 00 00 1d 00 00 00 00
>>>
>>> FreeBSD is wrong to print out 18 bytes because that is an 8
>>> byte (deferred, fixed type) buffer because byte 7 (the
>>> additional length) is 0. Whatever produced that broken
>>> sense data is the probably culprit.
>>>
>>> It is trying to say there is "ATA pass-through information
>>> available" but fails to get its message across.
>>>
>>
>> Hi,
>>
>> I'm the original reporter.
>>
>> The sense data is smartctl's interpretation of the ccb union.
>>
>> In particular, the bytes are from ccb->csio.sense_data and it calculates
>> 18
>> from ccb->csio.sense_len - ccb->csio.sense_resid (32 - 14).  I don't know
>> if
>> that is correct or not.
>>
>> FWIW, here is the output from CentOS 6.3 on the same machine:
>>
>> REPORT-IOCTL: Device=/dev/sdc Command=SMART STATUS CHECK
>>   Input:   FR=0xda, SC=...., LL=...., LM=0x4f, LH=0xc2, DEV=...., CMD=0xb0
>>   [ata pass-through(16): 85 06 2c 00 da 00 00 00 00 00 4f 00 c2 00 b0 00 ]
>>    scsi_status=0x2, host_status=0x0, driver_status=0x8
>>    info=0x1  duration=17 milliseconds  resid=0
>>    >>> Sense buffer, len=22:
>>   00     72 00 00 00 00 00 00 0e  09 0c 00 00 00 00 00 00
>>   10     00 4f 00 c2 40 50
>>    status=2: [desc] sense_key=0 asc=0 ascq=0
>> Values from ATA Return Descriptor are:
>>   00     09 0c 00 00 00 00 00 00  00 4f 00 c2 40 50
>>   [Duration: 0.016s]
>>   Output: ERR=0x00, SC=0x00, LL=0x00, LM=0x4f, LH=0xc2, DEV=0x40, STS=0x50
>> REPORT-IOCTL: Device=/dev/sdc Command=SMART STATUS CHECK returned 0
>>
>> It appears to have the same ata pass-through command but completely
>> different
>> sense data.
>>
>
> The SAT standard originally only defined "descriptor" sense
> data format for passing back ATA errors and warning. That is
> what is being done properly in the Centos 6.3 output that
> you have shown above.
>
> More recently someone at T10 objected that SAT ignored the D_SENSE
> flag which allows an application client to choose whether it wants
> "fixed" or "descriptor" sense data format. It looks like the FreeBSD
> case is trying to produce the equivalent "fixed" sense data, but it
> fails to format it properly.
>
> So in the FreeBSD case you are looking for the SAT Layer (SATL). It
> could be in the FreeBSD CAM mid-level, the driver code or firmware,
> or in a remote device like a USB bridge. However if it was in something
> like a USB bridge then you would expect FreeBSD and Linux to react
> the same way. So it looks like a FreeBSD bug.
>
> Doug Gilbert
>
>
Apologies for missing this thread until now.

The SATL that is generating this fixed format sense data is in the isci
driver itself.  I think isci is returning the fixed format sense data
correctly.  My reading of the SPC spec is that fixed format sense data is
18 bytes + additional sense length, so if sense_data[7] == 0, then the
sense data should indeed be 18 bytes long.  SAT-3 also indicates that for
fixed format sense data for ATA PASSTHROUGH commands, there are no
additional sense bytes - everything is returned in the INFORMATION and
COMMAND-SPECIFIC INFORMATION bytes which are part of the first 18 bytes.

I have looked a little at the smartctl code, and it does not look like it
supports fixed format sense data - at least for ATA PASSTHROUGH.  It
expects additional sense length to be non-zero.  smartctl could be fixed,
but I think for compatibility, I should have isci return descriptor sense
for ATA PASSTHROUGH commands instead.

Martin - could you please test the attached patch?  This passes your test
case using smartctl 6.3 r3939.

Thanks,

-Jim
-------------- next part --------------
A non-text attachment was scrubbed...
Name: sati_passthrough_sense.patch
Type: application/octet-stream
Size: 5179 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-scsi/attachments/20140709/4016a83f/attachment.obj>