On Tue, 2017-03-07 at 23:34 -0800, Nicholas A. Bellinger wrote: > Btw, the regression reported here in v2: > > http://www.spinics.net/lists/target-devel/msg14348.html > > is completely different from what you've reported here.
The call traces differ but the root cause is probably the same. > It would be useful to explain how you reproduced this, instead of just > posting backtrace with zero context..? > > Can we at least identify which patch in this series is causing this..? > > Also, I assume you are running this on stock v4.11-rc1 with only this > qla2xxx series applied, and not all of your other stuff, right..? The test I ran against v4.11-rc1 + this patch series is to start LIO on a system equipped with two back-to-back connected QLogic FC HBAs (no switch inbetween), to load the tcm_qla2xxx driver, to configure LUNs and to wait until the SCSI stack reports that these LUNs have appeared. What I see in the lsscsi output with both v2 and v3 of this patch series is that these LUNs appear briefly and then disappear and that a little bit later the kernel reports that a hang occurred. Without this patch series the LUNs are detected and do not disappear automatically and no hang is reported. I think the next step is that Cavium verifies whether they can reproduce this behavior and if they can reproduce it to run a bisect. BTW, since there are login-related patches in this series I wouldn't be surprised if one of these patches introduced the regression. Bart.
