Re: [PATCH] netns: deinline net_generic()

[Denys Vlasenko <dvlasenk@redhat.com>]

On 04/16/2015 02:38 PM, Eric Dumazet wrote:
> On Thu, 2015-04-16 at 13:14 +0200, Denys Vlasenko wrote:
> 
>> However, without BUG_ONs, function is still a bit big
>> on PREEMPT configs.
> 
> Only on allyesconfig builds, that nobody use but to prove some points
> about code size.

How do you expect one to find excessively large inlines,
if not on allyesconfig build?

Only by using allyesconfig, I can measure how many calls
are there in the kernel. (grepping source is utterly unreliable
due to nested inlines and macros).

For the record: I am not using the _full_ allyesconfig,
I do disable some debugging options which clearly aren't
ever enabled on production systems. E.g. in my config:

# CONFIG_DEBUG_KMEMLEAK_TEST is not set
# CONFIG_KASAN is not set

etc.

> If you look at net_generic(), it is mostly used from code that is
> normally in 3 modules. How many people really load them ?
> 
> net/tipc : 91 call sites
> net/sunrpc : 57
> fs/nfsd & fs/lockd : 183
> 
> Then few remaining uses in tunnels.

Grepping is far from reliable. The above missed more than half
of all calls. I disassembed vmlinux after deinlining, there are
nearly 500 calls of net_generic().

> As we suggested, please just remove the BUG_ON().

Going to send the patch in a minute.
-- 
vda