Re: [PATCH-2.6.0-tiny] "uninline" {lock,release}_sock

[Linus Torvalds]

On Sun, 28 Dec 2003, Arnaldo Carvalho de Melo wrote:
> 
> 	Please apply on top of your 2.6.0-tiny1 tree, CC to netdev for
> eventual comments.

Please don't do it this way.

This is quite possibly faster even _normally_, so it might make more sense 
to just do it globally instead of having a CONFIG_NEX_SMALL.

Basically, inline functions tend to win only when inlining them is smaller
and simpler than actually calling a function. The most common cause of
that is that some argument is commonly constant (and thus gets simplified
away by inlining), or the function itself literally expands to just a few 
instructions (the list functions, the inline asms for things like "cli" 
etc).

We use a lot too many inline functions for other reasons: one reason to 
use them is that they are sometimes more convenient than it is to find a 
good place for the non-inline version. Another common reason is that the 
thing started out smaller than it eventually became - and the inline just 
stuck around.

But if you do things like this for a CONFIG_SMALL, then the convenience 
argument obviously isn't true any more, and you'd be a lot better off just 
unconditionally making it a real function call.

Function calls aren't all that expensive, especially with FASTCALL() etc 
to show that you don't have to follow the common calling conventions. 
Right now I think FASTCALL() only matters on x86, but some other 
architectures could make it mean "smaller call clobbered list" or similar.

Have you benchmarked with the smaller kernel? 

		Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/