On Monday 16 May 2005 11:21, Timothy R. Chavez wrote:
 Hello,
 Serge pointed out something to me on an internal list here and I wanted to
 poll for opinions.  Which would you prefer?  I think adding another
 parameter to audit_send_reply to specify GFP_ATOMIC/KERNEL would be
 simplest.  But is it wise to do so?
 Here's what he had to say:
 Hey Tim,
 it looks like w_master is now adequately protected by the spinlock.
 But I'm not sure whether netlink_unicast is a problem (sure looks like
 it - it puts current on a waitqueue for one thing) or what you should do
 about it.  You might just need to build a list of things to send under
 the rcu_read_lock(), then send each element once you rcu_read_unlock(). 
Ok, I took this advice, and just went ahead and implemented something very 
similar.  Good idea Serge.  Thanks.  I'm going to do some more extensive 
testing here before I release a patch since people are apparently having 
problems with the code (plus I'm going to patch up to rc4-mm2).
-tim