On 12/3/21 3:52 AM, Manish Rangankar wrote: > When issued LUN reset under heavy i/o, we hit the qedi WARN_ON > because of a mismatch in firmware i/o cmd cleanup request count > and i/o cmd cleanup response count received. The mismatch is > because of the race caused by the postfix increment of > cmd_cleanup_cmpl. > > [qedi_clearsq:1295]:18: fatal error, need hard reset, cid=0x0 > WARNING: CPU: 48 PID: 110963 at drivers/scsi/qedi/qedi_fw.c:1296 qedi_clearsq+0xa5/0xd0 [qedi] > CPU: 48 PID: 110963 Comm: kworker/u130:0 Kdump: loaded Tainted: G W > Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 04/15/2020 > Workqueue: iscsi_conn_cleanup iscsi_cleanup_conn_work_fn [scsi_transport_iscsi] > RIP: 0010:qedi_clearsq+0xa5/0xd0 [qedi] > RSP: 0018:ffffac2162c7fd98 EFLAGS: 00010246 > RAX: 0000000000000000 RBX: ffff975213c40ab8 RCX: 0000000000000000 > RDX: 0000000000000000 RSI: ffff9761bf816858 RDI: ffff9761bf816858 > RBP: ffff975247018628 R08: 000000000000522c R09: 000000000000005b > R10: 0000000000000000 R11: ffffac2162c7fbd8 R12: ffff97522e1b2be8 > R13: 0000000000000000 R14: ffff97522e1b2800 R15: 0000000000000001 > FS: 0000000000000000(0000) GS:ffff9761bf800000(0000) knlGS:0000000000000000 > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > CR2: 00007f1a34e3e1a0 CR3: 0000000108bb2000 CR4: 0000000000350ee0 > Call Trace: > qedi_ep_disconnect+0x533/0x550 [qedi] > ? iscsi_dbg_trace+0x63/0x80 [scsi_transport_iscsi] > ? _cond_resched+0x15/0x30 > ? iscsi_suspend_queue+0x19/0x40 [libiscsi] > iscsi_ep_disconnect+0xb0/0x130 [scsi_transport_iscsi] > iscsi_cleanup_conn_work_fn+0x82/0x130 [scsi_transport_iscsi] > process_one_work+0x1a7/0x360 > ? create_worker+0x1a0/0x1a0 > worker_thread+0x30/0x390 > ? create_worker+0x1a0/0x1a0 > kthread+0x116/0x130 > ? kthread_flush_work_fn+0x10/0x10 > ret_from_fork+0x22/0x40 > ---[ end trace 5f1441f59082235c ]--- > > Signed-off-by: Manish Rangankar <mrangankar@xxxxxxxxxxx> > --- > v1 -> v2: > - Changing cmd_cleanup_cmpl variable to atomic > - In completion path instead pre-increment use atomic inc. > Reviewed-by: Mike Christie <michael.christie@xxxxxxxxxx>
