On 1/13/15 1:57 PM, David Miller wrote:
From: Cong Wang <cwang@xxxxxxxxxxxxxxxx>
Date: Tue, 13 Jan 2015 11:25:45 -0800
On Tue, Jan 13, 2015 at 6:10 AM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
When IPV4 support is disabled, we cannot call arp_send from
the bridge code, which would result in a kernel link error:
net/built-in.o: In function `br_handle_frame_finish':
:(.text+0x59914): undefined reference to `arp_send'
:(.text+0x59a50): undefined reference to `arp_tbl'
This makes the newly added proxy ARP support in the bridge
code depend on the CONFIG_INET symbol and lets the compiler
optimize the code out to avoid the link error.
Not sure how much sense to make CONFIG_BRIDGE depend
on CONFIG_INET, at least CONFIG_BONDING does.
It depends upon whether we want to provide and consider
as a valid configuration bridging without INET. Probably
we do.
Rather than connect CONFIG_BRIDGE to CONFIG_INET, why not make
br_do_proxy_arp (and setting BR_PROXYARP flag) a no-op if CONFIG_INET is
not set?
#ifdef CONFIG_INET
#else
static inline void br_do_proxy_arp(...args...)
{
}
#endif
That covers both arp_tbl and arp_send.
David