Re: [PATCH v5] alpha: fix memory barriers so that they conform to the specification

[Mikulas Patocka <mpatocka@redhat.com>]

On Mon, 25 May 2020, Maciej W. Rozycki wrote:

> On Mon, 25 May 2020, Mikulas Patocka wrote:
> 
> > > > The functions __raw_read* are already extern inline, so the compiler will 
> > > > inline/noinline them depending on the macros trivial_io_bw and 
> > > > trivial_io_lq - so we can just call them from read*_relaxed without 
> > > > repeating the extern inline pattern.
> > > 
> > >  The whole point of this peculiar arrangement for cooked accessors is to 
> > > avoid having barriers inserted inline around out-of-line calls to raw 
> > > accessors,
> > 
> > I see, but why do we want to avoid that? Linux kernel has no binary 
> > compatibility, so it doesn't matter if the barriers are inlined in the 
> > drivers or not.
> 
>  It does matter as it expands code unnecessarily (at all call sites), as I 
> noted in the original review.  This has nothing to do with compatibility.
> 
> > Anyway, I've sent a next version of the patch that makes read*_relaxed 
> > extern inline.
> 
>  Thanks, I'll have a look.  And now that you have this update, please run 
> `size' on ALPHA_GENERIC `vmlinux', preferably monolithic, to see what the 
> difference is between `read*_relaxed' handlers `static inline' and keyed 
> with `*trivial_rw_*'.
> 
>   Maciej

The patch with static inline:
   text    data     bss     dec     hex filename
124207762       75953010        5426432 205587204       c410304 vmlinux

The patch with extern inline:
   text    data     bss     dec     hex filename
124184422       75953474        5426432 205564328       c40a9a8 vmlinux

(I've sent version 7 of the patch because I misnamed the "write_relaxed" 
function in version 6).

Mikulas