-
Takashi Iwai authored
The recent change in lockdep for read lock caused the deadlock warnings in ALSA control code which uses the read_lock() for notification and else while write_lock_irqsave() is used for adding and removing the list entry. Although a deadlock would practically never hit in a real usage (the addition and the deletion can't happen with the notification), it's better to fix the read_lock() usage in a semantically correct way. This patch replaces the read_lock() calls with read_lock_irqsave() version for avoiding a reported deadlock. The notification code path takes the irq disablement in anyway, and other code paths are very short execution, hence there shouldn't be any big performance hit by this change. Fixes: e9181886 ("locking: More accurate annotations for read_lock()") Reported-by:
<syzbot+561a74f84100162990b2@syzkaller.appspotmail.com> Link: https://lore.kernel.org/r/20200922084953.29018-1-tiwai@suse.de Signed-off-by:
Takashi Iwai <tiwai@suse.de>
6564d0adTakashi Iwai authoredThe recent change in lockdep for read lock caused the deadlock warnings in ALSA control code which uses the read_lock() for notification and else while write_lock_irqsave() is used for adding and removing the list entry. Although a deadlock would practically never hit in a real usage (the addition and the deletion can't happen with the notification), it's better to fix the read_lock() usage in a semantically correct way. This patch replaces the read_lock() calls with read_lock_irqsave() version for avoiding a reported deadlock. The notification code path takes the irq disablement in anyway, and other code paths are very short execution, hence there shouldn't be any big performance hit by this change. Fixes: e9181886 ("locking: More accurate annotations for read_lock()") Reported-by:
Loading