This seems the a possible root-cause...
[11367.785938] kworker/5:1H S 0 768 2 0x80004000 [11367.786401] Workqueue: ib-comp-wq ib_cq_poll_work [ib_core] [11367.786958] Call Trace: [11367.787446] ? __schedule+0x31a/0x850 [11367.787988] schedule+0x36/0xc0 [11367.788488] iscsit_allocate_cmd+0x246/0x2c0 [iscsi_target_mod] [11367.789042] ? finish_wait+0x80/0x80 [11367.789564] isert_allocate_cmd+0x1d/0x60 [ib_isert] [11367.790120] isert_recv_done+0x282/0xa30 [ib_isert] [11367.790620] __ib_process_cq+0x76/0xc0 [ib_core] [11367.791181] ib_cq_poll_work+0x26/0x80 [ib_core] [11367.791697] process_one_work+0x235/0x5d0 [11367.792300] worker_thread+0x3c/0x390 [11367.792818] ? process_one_work+0x5d0/0x5d0 [11367.793366] kthread+0x11f/0x140 [11367.793850] ? kthread_park+0x90/0x90 [11367.794440] ret_from_fork+0x24/0x30
The only way that the workqueue cannot process the drain completion is if it is stuck. This appears to be the cq workqueue blocks on a command allocation that apparently never completes. The question is why does that happen only for 511MB luns, and how come this allocation never completes...
