Re: Lockup on USB and block devices

From: Alan Stern
Date: Tue Oct 01 2019 - 14:31:49 EST


On Tue, 1 Oct 2019, Steven Rostedt wrote:

> Not sure who to blame, but my server locked up when upgraded (accessing
> volume group information), and echoing in "w" into sysrq-trigger showed
> a bit of information.
>
> First, looking at my dmesg I see that my usb-storage is hung up, for
> whatever reason. Thus, this could be the source of all issues.
>
>
> [5434447.145737] INFO: task usb-storage:32246 blocked for more than 120 seconds.
> [5434447.145740] Not tainted 5.2.4-custom #4
>
> (BTW, I was upgrading to my 5.2.17 kernel when this happened)
>
> [5434447.145741] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> [5434447.145743] usb-storage D 0 32246 2 0x80004000
> [5434447.145745] Call Trace:
> [5434447.145749] ? __schedule+0x1e8/0x600
> [5434447.145752] schedule+0x29/0x90
> [5434447.145755] schedule_timeout+0x208/0x300
> [5434447.145765] ? usb_hcd_submit_urb+0xbe/0xb90 [usbcore]
> [5434447.145773] ? usb_alloc_urb+0x23/0x70 [usbcore]
> [5434447.145782] ? usb_sg_init+0x92/0x2b0 [usbcore]
> [5434447.145791] ? usb_hcd_submit_urb+0xbe/0xb90 [usbcore]
> [5434447.145795] ? __switch_to_asm+0x34/0x70
> [5434447.145798] wait_for_completion+0x111/0x180
> [5434447.145800] ? wake_up_q+0x70/0x70
> [5434447.145809] usb_sg_wait+0xfa/0x150 [usbcore]
> [5434447.145814] usb_stor_bulk_transfer_sglist.part.4+0x64/0xb0 [usb_storage]
> [5434447.145818] usb_stor_bulk_srb+0x49/0x80 [usb_storage]
> [5434447.145821] usb_stor_Bulk_transport+0x167/0x3e0 [usb_storage]
> [5434447.145824] ? schedule+0x29/0x90
> [5434447.145828] ? usb_stor_disconnect+0xb0/0xb0 [usb_storage]
> [5434447.145832] usb_stor_invoke_transport+0x3a/0x4e0 [usb_storage]
> [5434447.145835] ? wait_for_completion_interruptible+0x12d/0x1d0
> [5434447.145837] ? wake_up_q+0x70/0x70
> [5434447.145841] usb_stor_control_thread+0x1c5/0x270 [usb_storage]
> [5434447.145845] kthread+0x116/0x130
> [5434447.145847] ? kthread_create_worker_on_cpu+0x70/0x70
> [5434447.145851] ret_from_fork+0x35/0x40

It looks like a problem with your xHCI USB host controller. Normally a
usb-storage transfer would be aborted after 90 seconds. But if the
host controller (or its driver) isn't working right, and the abort
never completes, you end up with a situation like this -- usb-storage
and the higher SCSI and block layers waiting indefinitely for an event
that won't occur.

In theory, unplugging the USB mass-storage device would unblock
everything (unless xhci-hcd is too badly wedged). But of course, doing
that is likely to mess up whatever you were working on.

Alan Stern