Hello Bart,
> have you tried to perform a bisect?
We did finally complete the bisect, which led to this commit:
# first bad commit: [bb4f552a598d40624343bc0815e594efd4d6ec3d] Merge branch 'for-4.16-console-waiter-logic' into for-4.16
The commit has to do with printk, so it doesn't seem like an obvious candidate for introducing the behavior we've observed. Also, since the failures are intermittent, there's unfortunately some possibility of error during the bisect. However, there are some noticeable differences in the initiator syslog between that commit and the previous one. Before the commit, SCSI device setup messages appear more serially in the log, i.e., /dev/sdb setup completes start to finish before setup starts for /dev/sdc. After the commit, the device setup messages appear more in parallel - setup starts for /dev/sdb, /dev/sdc/, etc., continues in parallel, and all of the "Attached SCSI disk" messages appear consecutively in the log at the end. When the log messages are more parallel, we tend to see the intermittent delay.
So although I'm not sure the bisect definitely proves anything, I still wanted to follow up with our continued observations.
Thanks,
Jess
��.n��������+%������w��{.n�����{���fk��ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f
