Re: [PATCH] enforce function inlining for hot functions

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Fri, Apr 24, 2015 at 12:49:05PM -0700, Andrew Morton wrote:
> On Thu, 23 Apr 2015 23:40:01 +0200 Hagen Paul Pfeifer <hagen@jauu.net> wrote:
> 
> > GCC inlining heuristics are sometimes quizzical. Especially with inline
> > assembler constructs GCC seems to have issues. A allyesconfig show a rather
> > long list of functions where GCC inlining decisions are questionable (not
> > inlined).
> 
> I can't reproduce this with either gcc-4.8.2 or gcc-4.4.4.  The patch
> makes zero difference to `size vmlinux' and a bit of poking around with
> nm doesn't show any out-of-lined versions of the functions you
> identify.
> 
> So.  More details, please.  How to demonstrate this, gcc versions, etc.
> 
> > Furthermore, because the functions are declared with static
> > linkage each function is copied n times - and n can be rather high:
> > 
> >               atomic_inc: 544 duplicates
> >          rcu_read_unlock: 453 duplicates
> >            rcu_read_lock: 383 duplicates

Hmmm...  allyesconfig would have PROVE_RCU=y, which would mean that the
above two would contain lockdep calls that might in some cases defeat
inlining.  With the more typical production choice of PROVE_RCU=n, I would
expect these to just be a call instruction, which should get inlined.

							Thanx, Paul

> >              get_dma_ops: 271 duplicates
> >   arch_local_irq_restore: 258 duplicates
> >               atomic_dec: 215 duplicates
> >                  kzalloc: 185 duplicates
> >            cpumask_check: 157 duplicates
> >         test_and_set_bit: 156 duplicates
> >             cpumask_next: 146 duplicates
> >                 list_del: 131 duplicates
> >                 kref_get: 126 duplicates
> 
> That's pretty pathetic.
>