Matthew Dillon wrote:
  Fix issues with spinlocks being held through blocking conditions (because
  DragonFly spinlocks aren't FreeBSD mutexes) by replacing the spinlocks
  with lockmgr locks.

I'm afraid this breaks more than it fixes.  I get panics on module unload with 
this.  What I guess happens is this:

1. detach() called
2. obtains lock
3.                      interrupt arrives
4.                      tries to obtain lock, sleeps (stupid for intr)
5. removes interrupt
6. releases lock
7. destroys lock
8. unloads module
9.                      wakes up
10.                     tries to obtain lock (other lock or so)
11.                     panic, lock in invalid state:

mixer: ops removed
dsp(30)[000000ff/00000003]: Warning: dev_ops_remove() called while 10 device 
refs still exist!
dsp(30)[000000ff/00000005]: Warning: dev_ops_remove() called while 6 device 
refs still exist!
dsp(30)[000000ff/00010003]: Warning: dev_ops_remove() called while 5 device 
refs still exist!
dsp(30)[000000ff/00010005]: Warning: dev_ops_remove() called while 4 device 
refs still exist!
dsp(30)[000000ff/0001000b]: Warning: dev_ops_remove() called while 3 device 
refs still exist!
dsp(30)[000000ff/00020003]: Warning: dev_ops_remove() called while 2 device 
refs still exist!
dsp(30)[000000ff/00020005]: Warning: dev_ops_remove() called while 1 device 
refs still exist!
pcm0: detached
panic: lockmgr: non-zero exclusive count

how do freebsd deal with this sleeping thing?  I have to review the code, maybe they only 
access one "hardware" lock, then we should use serializers for this.

cheers
 simon

--
Serve - BSD     +++  RENT this banner advert  +++    ASCII Ribbon   /"\
Work - Mac      +++  space for low €€€ NOW!1  +++      Campaign     \ /
Party Enjoy Relax   |   http://dragonflybsd.org      Against  HTML   \
Dude 2c 2 the max   !   http://golden-apple.biz       Mail + News   / \

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to