Re: [PATCH 2/2] Add CONFIG_READABLE_ASM

[Sam Ravnborg <sam@ravnborg.org>]

On Sat, May 12, 2012 at 10:23:40PM +0200, Geert Uytterhoeven wrote:
> On Tue, May 8, 2012 at 9:36 PM, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> > On Thu, Apr 12, 2012 at 10:35 PM, Sam Ravnborg <sam@ravnborg.org> wrote:
> >> On Thu, Apr 12, 2012 at 10:17:07PM +0200, Geert Uytterhoeven wrote:
> >>> On Thu, Apr 12, 2012 at 22:07, Sam Ravnborg <sam@ravnborg.org> wrote:
> >>> > Could you try this:
> >>> >
> >>> > cc-disable-opt = $(call try-run,\
> >>> >        $(CC) $(KBUILD_CPPFLAGS) $(KBUILD_CFLAGS) -f$(strip $(1)) -c -xc /dev/null -o "$TMP",-fno-$(strip $(1)))
> >>> >
> >>> > KBUILD_CFLAGS += $(call cc-disable-opt, ipa-cp-clone)
> >>> >
> >>> >
> >>> > It will try if "gcc -fipa-cp-clone" is supported,
> >>> > and add -fno-ipa-cp-clone in that case.
> >>>
> >>> Unfortunately not:
> >>>
> >>> m68k-linux-gnu-gcc -Wp,-MD,kernel/.bounds.s.d  -nostdinc -isystem
> >>> /usr/lib/gcc/m68k-linux-gnu/4.1.2/include
> >>> -I/scratch/geert/linux/linux-next/arch/m68k/include
> >>> Iarch/m68k/include/generated -Iinclude  -include
> >>> /scratch/geert/linux/linux-next/include/linux/kconfig.h -D__KERNEL__
> >>> -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs -fno-strict-aliasing
> >>> -fno-common -Werror-implicit-function-declaration -Wno-format-security
> >>> -fno-delete-null-pointer-checks -Os -fno-reorder-blocks
> >>> -fno-ipa-cp-clone -pipe -fno-strength-reduce -ffixed-a2
> >>> -fno-omit-frame-pointer -fno-optimize-sibling-calls -g
> >>> -D"KBUILD_STR(s)=#s" -D"KBUILD_BASENAME=KBUILD_STR(bounds)"
> >>> -D"KBUILD_MODNAME=KBUILD_STR(bounds)" -fverbose-asm -S -o
> >>> kernel/bounds.s kernel/bounds.c
> >>> cc1: error: unrecognized command line option "-fno-ipa-cp-clone"
> >>> make[1]: *** [kernel/bounds.s] Error 1
> >>> make: *** [prepare0] Error 2
> >>
> >> So with the above command-line gcc fails with an exit code for -fno-ipa-cp-clone.
> >> But with cc-option it does not.
> >>
> >> Can you dig out the command-line used for cc-option?
> >> We may be able to analyze why it fails in the above case, but
> >> not with cc-option.
> >
> > gcc -D__KERNEL__ -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs
> > -fno-strict-aliasing -fno-common -Werror-implicit-function-declaration
> > -Wno-format-security -fno-delete-null-pointer-checks -Os
> > -fno-ipa-cp-clone -c -xc /dev/null -o .6376.tmp
> >
> > So it's running "gcc" instead of "m68k-linux-gnu-gcc".
> 
> Found it using the patch below to log all try-run invocations:
> 
> diff --git a/scripts/Kbuild.include b/scripts/Kbuild.include
> index 6a3ee98..12eb839 100644
> --- a/scripts/Kbuild.include
> +++ b/scripts/Kbuild.include
> @@ -88,6 +88,7 @@ TMPOUT := $(if $(KBUILD_EXTMOD),$(firstword
> $(KBUILD_EXTMOD))/)
>  try-run = $(shell set -e;              \
>         TMP="$(TMPOUT).$$$$.tmp";       \
>         TMPO="$(TMPOUT).$$$$.o";        \
> +       echo "$(1)" >> /tmp/x;          \
>         if ($(1)) >/dev/null 2>&1;      \
>         then echo "$(2)";               \
>         else echo "$(3)";               \
> 
> | --- a/Makefile
> | +++ b/Makefile
> | @@ -564,6 +564,16 @@ else
> |  KBUILD_CFLAGS  += -O2
> |  endif
> |
> | +ifdef CONFIG_READABLE_ASM
> | +# Disable optimizations that make assembler listings hard to read.
> | +# reorder blocks reorders the control in the function
> | +# ipa clone creates specialized cloned functions
> | +# partial inlining inlines only parts of functions
> | +KBUILD_CFLAGS += $(call cc-option,-fno-reorder-blocks,) \
> | +                 $(call cc-option,-fno-ipa-cp-clone,) \
> | +                 $(call cc-option,-fno-partial-inlining)
> | +endif
> | +
> 
> The above calls cc-option using CC = gcc, as my cross compiler
> is only setup in the next line:
> 
> | include $(srctree)/arch/$(SRCARCH)/Makefile
> 
> This affects all architectures that setup CROSS_COMPILE in their arch/*/Makefile
> (blackfin, h8300, m68k, mips, parisc, score, sh, tile, unicore32, xtensa).
> 
> Worse, m68k, sh, and xtensa call cc-option in their arch/*/Makefile
> before CROSS_COMPILE is setup, causing those checks to be done against
> the host compiler when cross-compiling.
> 
> Sending out patches, as soon as the (automatic) "git gc" completes...

I expected something like this - but I got distraced by some sparc32 work and
forgot to return to you.

Great that you found it so we can get this fixed!

	Sam