kern/165863

[Ryan Stone]

The following reply was made to PR kern/165863; it has been noted by GNATS.

From: Ryan Stone <rysto32 at gmail.com>
To: Eric van Gyzen <eric at vangyzen.net>
Cc: Gleb Smirnoff <glebius at freebsd.org>, Eric van Gyzen <eric_van_gyzen at dell.com>, emaste at freebsd.org, 
	bug-followup at freebsd.org
Subject: Re: kern/165863
Date: Tue, 27 Mar 2012 17:24:37 -0400

 2012/3/9 Eric van Gyzen <eric at vangyzen.net>:
 > On 03/09/12 03:20, Gleb Smirnoff wrote:
 >>
 >> =A0 Hello, Eric and Ed.
 >>
 >> =A0 Can you look at this patch? I decided to utilize newer callout API,
 >> that allows to delegate lock retrieval to the callout subsystem, and
 >> this makes things simplier. Hope that should work.
 >>
 >> =A0 Patch is against head.
 >
 >
 > Doesn't arptimer() still need to acquire the if_afdata_lock in order to f=
 ree
 > the entry in the normal case (when the llentry is still in the hash bucke=
 t
 > list)?
 
 Oops, on reviewing the code I believe that this is correct.  My test
 case wouldn't have had arp entries timing out so I wouldn't have hit
 this case.  I'll try to come up with a test case for this.
 
 Unfortunately it's not as quite as simple just acquiring
 if_afdata_lock because of lock ordering problems.  I think that we'll
 have to revert the usage of callout_init_rw so that arptimer can
 acquire the if_afdata_lock before acquiring the lock LLE_LOCK.