[Bug 89531] New: mpt2sas: only every second disk working with PCIe x4 connection

[Bug 89531] New: mpt2sas: only every second disk working with PCIe x4 connection

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=89531

            Bug ID: 89531
           Summary: mpt2sas: only every second disk working with PCIe x4
                    connection
           Product: SCSI Drivers
           Version: 2.5
    Kernel Version: 3.16.5
          Hardware: x86-64
                OS: Linux
              Tree: Mainline
            Status: NEW
          Severity: blocking
          Priority: P1
         Component: Other
          Assignee: scsi_drivers-other@kernel-bugs.osdl.org
          Reporter: abusse@cs.tu-berlin.de
        Regression: No

Created attachment 160351
  --> https://bugzilla.kernel.org/attachment.cgi?id=160351&action=edit
syslog

I have an SAS controller with LSI SAS 2008 controller in a PCIe slot with only
four lanes where the card can use up to eight lanes. I have the following
issue. Lets assume there are four drives attached to the controller
(sda...sdd). I use the following simple benchmark "cat /dev/sdX | pipebench >
/dev/null". When accessing every drive individually, every thing works fine.
However when I run the experiment on sda and sdb concurrently, it still works
on sda but the benchmarks stalls on sdb. When I kill the benchmark after some
seconds I see the error messages in the attachment. 

However, it gets more strange. When I run the benchmark concurrently on sda and
sdc every thing works as expected. When trying to read sdb or sdd concurrently
it always results in the error described above.

I also tried to change the hardware setup and only used every second connection
of the card. However the result stayed the same: concurrent access to every
second disk fails.

In the same machine I have the same SAS controller model in a PCIe x16 slot and
there every thing works like a charm. I also swaped the cards and the issue
stays with the PCIe x4 slot. So a defective card as the source of the problem
can be ruled out.

Thank you

-- 
You are receiving this mail because:
You are watching the assignee of the bug.