Work at SourceForge, help us to make it a better place! We have an immediate need for a Support Technician in our San Francisco or Denver office.

Close

#173 CDBs to /dev/sg* get aborts issued without timing out

4.0.2
closed
4
2005-04-20
2005-04-12
Dave Wysochanski
No

This didn't used to happen (not sure how far back), but
now I'm
seeing this odd behavior with the latest 4-0 CVS code and
my 2.6.9* kernel.

Anyone else testing /dev/sg* devices?

The tool I have issues a few commands like TUR,
INQUIRY, etc,
to get some info about the device, then issues a vendor
specific
command to get more info. From the trace, it looks
like the
first TUR completes then right after that a TMF ABORT TASK
is issued on the command, followed by the targets' response
of "task not in task set". A few more commands are issued,
some of which complete just fine with no TMF's issued
by the
driver, but then the driver issues another TMF for one of
the commands. It's not consistent though. You can see
some info in /var/log/messages as well.

Attached is the ethereal trace and the
/var/log/messages file.

Discussion

  • Snippit of /var/log/messages

     
  • Logged In: YES
    user_id=752546

    Investigation in progress

     
    • assigned_to: bhojas --> wysochanski
     
  • Logged In: YES
    user_id=752546

    This is looking like a kernel bug, unrelated to iSCSI, and
    has to do with a MAX_UINT value being given in the timeout
    field of sg_io_hdr_t. According to the comment in the
    header file, this is supposed to indicate "no timeout", and
    this used to work. However, it looks like the kernel I'm
    using to test has a bug in this area.

     
    • status: open --> closed