From mboxrd@z Thu Jan 1 00:00:00 1970 From: "David S. Miller" Subject: Re: [PATCH 2.6.12-rc2] bonding: partially back out dev_set_mac_address Date: Thu, 7 Apr 2005 13:31:51 -0700 Message-ID: <20050407133151.5887c946.davem@davemloft.net> References: <200504071959.j37JxZ8g003626@death.nxdomain.ibm.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Cc: netdev@oss.sgi.com, jgarzik@pobox.com Return-path: To: Jay Vosburgh In-Reply-To: <200504071959.j37JxZ8g003626@death.nxdomain.ibm.com> Sender: netdev-bounce@oss.sgi.com Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org On Thu, 07 Apr 2005 12:59:35 -0700 Jay Vosburgh wrote: > This patch backs out some of the calls to dev_set_mac_address > and replaces them with calls to a similar function that does not call > notifier_call_chain. > > The reason for this is that the rtnetlink event handler and its > descendents make GFP_KERNEL memory allocation requests, and the bonding > driver makes some of its MAC address change calls from timer context > with a lock held (notably the ALB mode). > > Rearranging the bonding driver to not call this way is a fairly > involved change; this patch merely reverts one part of bonding to the > way it used to be. You can't remove that notifier call, you will break ipv4 ARP, ipv6 neighbour discovery, and bridging if you do that.