James and others:
One of the things scsi_host_dev_release() does is to call
kthread_stop(shost->ehandler). This is all well and good, and to be
expected.
But...
What if a host is hotunplugged in the middle of error recovery, and the
error-handler thread holds the last reference? Then when the
error-handler gives up and does its final put, it will end up waiting
endlessly for itself to terminate.
I have managed to trigger this a couple of times now while testing
other things. Here's the relevant part of a stack trace:
scsi_eh_6 D 00000136 0 7975 2
c87cfd88 00000096 17070554 00000136 00000000 00000008 cfdd8fec cfdd8a90
00000000 cfdd8a90 cfdd8bc4 c138ac00 00000000 c02ab06a c87cfda4 cfa33280
c013a35e c03d89ac c02a8818 00000046 c03d89ac c03d89ac c03d89ac c03d89a8
Call Trace:
[<c02a881d>] wait_for_completion+0x74/0xaa
[<c0131b3e>] kthread_stop+0x6b/0x8b
[<d09b8e6a>] scsi_host_dev_release+0x3e/0xa4 [scsi_mod]
[<c02212ae>] device_release+0x3c/0x7e
[<c01d2786>] kobject_cleanup+0x3c/0x4d
[<c01d27a2>] kobject_release+0xb/0xd
[<c01d33f1>] kref_put+0x79/0x88
[<c01d2748>] kobject_put+0x14/0x16
[<c0221400>] put_device+0x11/0x13
[<d09be37f>] scsi_target_dev_release+0x19/0x1c [scsi_mod]
[<c02212ae>] device_release+0x3c/0x7e
[<c01d2786>] kobject_cleanup+0x3c/0x4d
[<c01d27a2>] kobject_release+0xb/0xd
[<c01d33f1>] kref_put+0x79/0x88
[<c01d2748>] kobject_put+0x14/0x16
[<c0221400>] put_device+0x11/0x13
[<d09c11db>] scsi_device_dev_release_usercontext+0xfd/0x106 [scsi_mod]
[<c012f052>] execute_in_process_context+0x19/0x3d
[<d09c0305>] scsi_device_dev_release+0x13/0x15 [scsi_mod]
[<c02212ae>] device_release+0x3c/0x7e
[<c01d2786>] kobject_cleanup+0x3c/0x4d
[<c01d27a2>] kobject_release+0xb/0xd
[<c01d33f1>] kref_put+0x79/0x88
[<c01d2748>] kobject_put+0x14/0x16
[<c0221400>] put_device+0x11/0x13
[<d09b80d1>] scsi_device_put+0x32/0x36 [scsi_mod]
[<d09b8268>] __scsi_iterate_devices+0x57/0x60 [scsi_mod]
[<d09bd2d3>] scsi_run_host_queues+0x1c/0x26 [scsi_mod]
[<d09bc105>] scsi_error_handler+0x40d/0x46c [scsi_mod]
[<c0131d9a>] kthread+0x3b/0x61
[<c0104d53>] kernel_thread_helper+0x7/0x10
=======================
What's the solution?
Alan Stern
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
