Re: [PATCH] x86: Align jump targets to 1 byte boundaries

[Josh Triplett <josh@joshtriplett.org>]

On Sun, May 17, 2015 at 07:34:29AM +0200, Ingo Molnar wrote:
> 
> * Josh Triplett <josh@joshtriplett.org> wrote:
> 
> > On Mon, Sep 17, 2001 at 07:00:00AM +0000, Ingo Molnar wrote:
> > > * Denys Vlasenko <dvlasenk@redhat.com> wrote:
> > > > > What do you guys think about this? I think we should seriously 
> > > > > consider relaxing our alignment defaults.
> > > > 
> > > > Looks like nobody objected. I think it's ok to submit
> > > > this patch for real.
> > > 
> > > Yeah, so my plan is to apply the following three changes from that 
> > > discussion:
> > > 
> > > --- tip.orig/arch/x86/Makefile
> > > +++ tip/arch/x86/Makefile
> > >  <at>  <at>  -77,6 +77,15  <at>  <at>  else
> > >          KBUILD_AFLAGS += -m64
> > >          KBUILD_CFLAGS += -m64
> > > 
> > > +        # Pack jump targets tightly, don't align them to the default 16 bytes:
> > > +        KBUILD_CFLAGS += -falign-jumps=1
> > > +
> > > +        # Pack functions tightly as well:
> > > +        KBUILD_CFLAGS += -falign-functions=1
> > > +
> > > +        # Pack loops tightly as well:
> > > +        KBUILD_CFLAGS += -falign-loops=1
> > > +
> > >          # Don't autogenerate traditional x87 instructions
> > >          KBUILD_CFLAGS += $(call cc-option,-mno-80387)
> > >          KBUILD_CFLAGS += $(call cc-option,-mno-fp-ret-in-387)
> > 
> > It looks like the patch you applied to the tip tree only included one of
> > these (-falign-junmps=1), not the other two.
> 
> It's three separate patches, in case there are any regressions.

Fair enough.  At the time I sent my mail, only the first of the three
had shown up on LKML.

> > Also, you've only applied these to 64-bit; could you please apply 
> > them to both 32-bit and 64-bit, since many embedded systems aiming 
> > for small code size use 32-bit?  (Unless 32-bit already defaults to 
> > these.)
> 
> First things first - 64-bit is getting far more testing these days 
> than 32-bit.

What testing do you want to see on these patches before applying them to
32-bit as well?

> > Have you considered including -falign-labels=1 as well?  Does that 
> > make a difference on top of the other three.
> 
> So isn't the default on x86 for -falign-labels already 1?

GCC's manual says that -O2 and above turn on -falign-labels, which has a
machine-specific default alignment.

A fair bit of digging turned up gcc/config/i386/i386.c, which does seem
to have processor-specific defaults for the other three but not for
align-labels.  So it looks like it does indeed use the general default
of 1.  Nevermind.

- Josh Triplett