kern/129053: lock order reversal

[Brad Huntting]

>Number:         129053
>Category:       kern
>Synopsis:       lock order reversal
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    freebsd-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Fri Nov 21 16:40:02 UTC 2008
>Closed-Date:
>Last-Modified:
>Originator:     Brad Huntting
>Release:        7.1-PRERELEASE
>Organization:
>Environment:
FreeBSD spork.glarp.com 7.1-PRERELEASE FreeBSD 7.1-PRERELEASE #4: Thu Nov 13 11:34:12 MST 2008     root at spork.glarp.com:/usr/src/sys/i386/compile/SPORK  i386

>Description:
My device driver (a proprietary USB device called 'scmicro') appears to have tickled a lock order reversal.  When witness.watch is enabled, all goes well until my driver detach()es.  At which point my system panics with:

lock order reversal:
 1st 0xc0c932a0 syscons video lock (syscons video lock) @ dev/syscons/syscons.c:2534
 2nd 0xc0c23400 sched lock 0 (sched lock) @ kern/kern_clock.c:286

panic: blockable sleep lock (sleep mutex) sellck @ kern/sys_generic.c:1127

My driver creates two taskqueue(9) threads in attach() and destroys them in detach().  It has 4 mutexes which it uses to protect it's own data structures.  In addition, it holds Giant around any code section which calls usb functions.  With the exception of wakeup(9), cv_signal(9), cv_destroy(9) and taskqueue_enqueue_fast(), it always release it's own locks (except for Giant) before it calls external code.

Of course, this does not happen when my devices uses the ugen(4) driver.

Is this a known issue?  Am I doing something wrong?

>How-To-Repeat:
Sorry, my client wont let me release details of the driver.
>Fix:


>Release-Note:
>Audit-Trail:
>Unformatted: