define check_gcc before used (Re: CVS Update@-mips.org: linux)

define check_gcc before used (Re: CVS Update@-mips.org: linux)

[Atsushi Nemoto]

>>>>> On Wed, 22 Oct 2003 00:25:50 +0100, ralf@linux-mips.org said:
ralf> CVSROOT:	/home/cvs
ralf> Module name:	linux
ralf> Changes by:	ralf@ftp.linux-mips.org	03/10/22 00:25:50

ralf> Modified files:
ralf> 	arch/mips      : Tag: linux_2_4 Makefile 
ralf> 	arch/mips64    : Tag: linux_2_4 Makefile 

ralf> Log message:
ralf> 	Make gcc try inlining functions really hard.

It seems mips64 Makefile does not pass "-finline-limit=100000" to gcc.
The "check_gcc" must be defined before used ?

diff -u linux-mips-cvs/arch/mips64/Makefile linux/arch/mips64/Makefile
--- linux-mips-cvs/arch/mips64/Makefile	Tue Nov  4 16:57:37 2003
+++ linux/arch/mips64/Makefile	Wed Nov  5 16:50:40 2003
@@ -24,6 +24,8 @@
 CROSS_COMPILE	= $(tool-prefix)
 endif
 
+check_gcc = $(shell if $(CC) $(1) -S -o /dev/null -xc /dev/null > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi)
+
 #
 # The ELF GCC uses -G 0 -mabicalls -fpic as default.  We don't need PIC
 # code in the kernel since it only slows down the whole thing.  For the
@@ -47,8 +49,6 @@
 endif
 endif
 
-check_gcc = $(shell if $(CC) $(1) -S -o /dev/null -xc /dev/null > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi)
-
 #
 # CPU-dependent compiler/assembler options for optimization.
 #
---
Atsushi Nemoto

[PATCH, 2.4] Fix bad check_gcc order for mips64, make offset.h creation more robust

[Thiemo Seufer]

Hallo All,

current 2.4 64bit kernels fail to build for newer toolchains because
the -finline-limit=100000 option is ignored. The symptom is some bogus
offset.h content which stops the build in arch/mips64/mm/tlbex-r4k.S.

This patch fixes it by moving check_gcc in front of its first use,
it also removes some stale files if offset.h needs to be regenerated.


Thiemo


Index: arch/mips/Makefile
===================================================================
RCS file: /home/cvs/linux/arch/mips/Makefile,v
retrieving revision 1.78.2.50
diff -a -d -u -p -r1.78.2.50 Makefile
--- arch/mips/Makefile	11 Jan 2004 00:11:35 -0000	1.78.2.50
+++ arch/mips/Makefile	22 Jan 2004 03:18:57 -0000
@@ -746,5 +746,6 @@ archmrproper:
 archdep:
 	if [ ! -f $(TOPDIR)/include/asm-$(ARCH)/offset.h ]; then \
 		touch $(TOPDIR)/include/asm-$(ARCH)/offset.h; \
+		$(MAKE) -C arch/mips/tools clean; \
 	fi;
 	@$(MAKEBOOT) dep
Index: arch/mips64/Makefile
===================================================================
RCS file: /home/cvs/linux/arch/mips64/Attic/Makefile,v
retrieving revision 1.22.2.43
diff -a -d -u -p -r1.22.2.43 Makefile
--- arch/mips64/Makefile	3 Jan 2004 02:07:52 -0000	1.22.2.43
+++ arch/mips64/Makefile	22 Jan 2004 03:19:23 -0000
@@ -26,6 +26,9 @@ ifdef CONFIG_CROSSCOMPILE
 CROSS_COMPILE	= $(tool-prefix)
 endif
 
+check_gcc = $(shell if $(CC) $(1) -S -o /dev/null -xc /dev/null > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi)
+check_gas = $(shell if $(CC) $(1) -Wa,-Z -c -o /dev/null -xassembler /dev/null > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi)
+
 #
 # The ELF GCC uses -G 0 -mabicalls -fpic as default.  We don't need PIC
 # code in the kernel since it only slows down the whole thing.  For the
@@ -49,9 +52,6 @@ GCCFLAGS	+= -mno-sched-prolog -fno-omit-
 endif
 endif
 
-check_gcc = $(shell if $(CC) $(1) -S -o /dev/null -xc /dev/null > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi)
-check_gas = $(shell if $(CC) $(1) -Wa,-Z -c -o /dev/null -xassembler /dev/null > /dev/null 2>&1; then echo "$(1)"; else echo "$(2)"; fi)
-
 #
 # Use: $(call set_gccflags,<cpu0>,<isa0>,<cpu1>,<isa1>)
 #
@@ -402,5 +406,6 @@ archmrproper:
 archdep:
 	if [ ! -f $(TOPDIR)/include/asm-$(ARCH)/offset.h ]; then \
 		touch $(TOPDIR)/include/asm-$(ARCH)/offset.h; \
+		$(MAKE) -C arch/mips/tools clean; \
 	fi;
 	@$(MAKEBOOT) dep

Re: [PATCH, 2.4] Fix bad check_gcc order for mips64, make offset.h creation more robust

[Atsushi Nemoto]

Could you remember this two months old patch?

>>>>> On Thu, 22 Jan 2004 05:07:59 +0100, Thiemo Seufer <ica2_ts@csv.ica.uni-stuttgart.de> said:
Thiemo> current 2.4 64bit kernels fail to build for newer toolchains
Thiemo> because the -finline-limit=100000 option is ignored. The
Thiemo> symptom is some bogus offset.h content which stops the build
Thiemo> in arch/mips64/mm/tlbex-r4k.S.

Or this four months old patch?

>>>>> On Wed, 05 Nov 2003 17:17:01 +0900 (JST), Atsushi Nemoto <anemo@mba.ocn.ne.jp> said:
anemo> It seems mips64 Makefile does not pass "-finline-limit=100000"
anemo> to gcc.  The "check_gcc" must be defined before used ?


I think this fix is definitely required for 2.4 mips64 tree.  Please
commit to CVS if no problem with this.

---
Atsushi Nemoto

Re: [PATCH, 2.4] Fix bad check_gcc order for mips64, make offset.h creation more robust

[Ralf Baechle]

On Mon, Mar 22, 2004 at 04:06:27PM +0900, Atsushi Nemoto wrote:

> Could you remember this two months old patch?

Sorry ...  Applied.

  Ralf