[PATCH] Fix booting some PPC32 machines

[PATCH] Fix booting some PPC32 machines

[Tom Rini]

Hello.  The following patch fixes booting on some PPC32 machines with
OpenFirmware, when booted without the aid of an additional bootloader.
The problem is that the linker script for the 'zImage' type targets was
put into the list of dependancies which objcopy would parse as a list of
files to copy into the resulting image.  The fix is to make the phony
zImage targets depend on the linker script.

===== arch/ppc/boot/openfirmware/Makefile 1.22 vs edited =====
--- 1.22/arch/ppc/boot/openfirmware/Makefile	Thu Apr  1 08:16:57 2004
+++ edited/arch/ppc/boot/openfirmware/Makefile	Mon May  3 10:24:35 2004
@@ -104,10 +104,10 @@
 			$(HACKCOFF) $@ && \
 			ln -sf $(notdir $@) $(images)/zImage$(initrd).pmac
 
-$(images)/vmlinux.coff: $(obj)/coffboot $(boot)/ld.script
+$(images)/vmlinux.coff: $(obj)/coffboot
 	$(call cmd,gen-coff)
 
-$(images)/vmlinux.initrd.coff: $(obj)/coffboot.initrd $(boot)/ld.script
+$(images)/vmlinux.initrd.coff: $(obj)/coffboot.initrd
 	$(call cmd,gen-coff)
 
 quiet_cmd_gen-elf-pmac = ELF     $@
@@ -116,19 +116,19 @@
 			$(OBJCOPY) $@ $@ --add-section=.note=$(obj)/note \
 					 -R .comment $(del-ramdisk-sec)
 
-$(images)/vmlinux.elf-pmac: $(obj)/image.o $(NEWWORLDOBJS) $(LIBS) $(obj)/note $(boot)/ld.script
+$(images)/vmlinux.elf-pmac: $(obj)/image.o $(NEWWORLDOBJS) $(LIBS) $(obj)/note
 	$(call cmd,gen-elf-pmac)
 $(images)/vmlinux.initrd.elf-pmac: $(obj)/image.initrd.o $(NEWWORLDOBJS) \
-				   $(LIBS) $(obj)/note $(boot)/ld.script
+				   $(LIBS) $(obj)/note
 	$(call cmd,gen-elf-pmac)
 
 quiet_cmd_gen-chrp = CHRP    $@
       cmd_gen-chrp = $(LD) $(CHRP_LD_ARGS) -o $@ $^ && \
 			$(OBJCOPY) $@ $@ -R .comment $(del-ramdisk-sec)
 
-$(images)/zImage.chrp: $(CHRPOBJS) $(obj)/image.o $(LIBS) $(boot)/ld.script
+$(images)/zImage.chrp: $(CHRPOBJS) $(obj)/image.o $(LIBS)
 	$(call cmd,gen-chrp)
-$(images)/zImage.initrd.chrp: $(CHRPOBJS) $(obj)/image.initrd.o $(LIBS) $(boot)/ld.script
+$(images)/zImage.initrd.chrp: $(CHRPOBJS) $(obj)/image.initrd.o $(LIBS)
 	$(call cmd,gen-chrp)
 
 quiet_cmd_addnote = ADDNOTE $@
@@ -153,13 +153,15 @@
 		 $(images)/vmlinux.elf-pmac	\
 		 $(images)/zImage.chrp		\
 		 $(images)/zImage.chrp-rs6k	\
-		 $(images)/miboot.image
+		 $(images)/miboot.image		\
+		 $(boot)/ld.script
 	@echo '  kernel: $@ is ready ($<)'
 zImage.initrd:	 $(images)/vmlinux.initrd.coff 		\
 		 $(images)/vmlinux.initrd.elf-pmac	\
 		 $(images)/zImage.initrd.chrp		\
 		 $(images)/zImage.initrd.chrp-rs6k	\
-		 $(images)/miboot.initrd.image
+		 $(images)/miboot.initrd.image		\
+		 $(boot)/ld.script
 	@echo '  kernel: $@ is ready ($<)'
 
 TFTPIMAGE	:= /tftpboot/zImage

-- 
Tom Rini
http://gate.crashing.org/~trini/

Re: [PATCH] Fix booting some PPC32 machines

[Paul Mackerras]

Tom Rini writes:

> Hello.  The following patch fixes booting on some PPC32 machines with
> OpenFirmware, when booted without the aid of an additional bootloader.
> The problem is that the linker script for the 'zImage' type targets was
> put into the list of dependancies which objcopy would parse as a list of
> files to copy into the resulting image.  The fix is to make the phony
> zImage targets depend on the linker script.

I don't like this part of the fix:

>  $(images)/vmlinux.initrd.elf-pmac: $(obj)/image.initrd.o $(NEWWORLDOBJS) \
> -				   $(LIBS) $(obj)/note $(boot)/ld.script
> +				   $(LIBS) $(obj)/note
>  	$(call cmd,gen-elf-pmac)

because the linking process (gen-elf-pmac) really does depend on
ld.script.  Since gen-elf-pmac uses $<, not $^, it should be OK with
the ld.script in the dependencies.  Did you verify that there was a
problem with this rule or did you just rip out all of the ld.script
dependencies?

Similarly we need the ld.script dependency in here:

> -$(images)/zImage.chrp: $(CHRPOBJS) $(obj)/image.o $(LIBS) $(boot)/ld.script
> +$(images)/zImage.chrp: $(CHRPOBJS) $(obj)/image.o $(LIBS)
>  	$(call cmd,gen-chrp)
> -$(images)/zImage.initrd.chrp: $(CHRPOBJS) $(obj)/image.initrd.o $(LIBS) $(boot)/ld.script
> +$(images)/zImage.initrd.chrp: $(CHRPOBJS) $(obj)/image.initrd.o $(LIBS)
>  	$(call cmd,gen-chrp)

Here, gen-chrp does have a problem because it uses $^, but it looks to
me that with a minor change it could just use $< and avoid the
problem.

The other places where you have removed ld.script from the
dependencies look OK to me.

Regards,
Paul.

Re: [PATCH] Fix booting some PPC32 machines

[Tom Rini]

On Tue, May 04, 2004 at 10:34:35AM +1000, Paul Mackerras wrote:

> Tom Rini writes:
> 
> > Hello.  The following patch fixes booting on some PPC32 machines with
> > OpenFirmware, when booted without the aid of an additional bootloader.
> > The problem is that the linker script for the 'zImage' type targets was
> > put into the list of dependancies which objcopy would parse as a list of
> > files to copy into the resulting image.  The fix is to make the phony
> > zImage targets depend on the linker script.
> 
> I don't like this part of the fix:
> 
> >  $(images)/vmlinux.initrd.elf-pmac: $(obj)/image.initrd.o $(NEWWORLDOBJS) \
> > -				   $(LIBS) $(obj)/note $(boot)/ld.script
> > +				   $(LIBS) $(obj)/note
> >  	$(call cmd,gen-elf-pmac)
> 
> because the linking process (gen-elf-pmac) really does depend on
> ld.script.  Since gen-elf-pmac uses $<, not $^, it should be OK with
> the ld.script in the dependencies.  Did you verify that there was a
> problem with this rule or did you just rip out all of the ld.script
> dependencies?

I did just remove all of the previous dependancies and place them on the
zImage target (we don't mention that you can do 'make
arch/ppc/boot/images/foo.image, and I don't know if that would work, so
I don't think that'd be a problem).

> Similarly we need the ld.script dependency in here:
> 
> > -$(images)/zImage.chrp: $(CHRPOBJS) $(obj)/image.o $(LIBS) $(boot)/ld.script
> > +$(images)/zImage.chrp: $(CHRPOBJS) $(obj)/image.o $(LIBS)
> >  	$(call cmd,gen-chrp)
> > -$(images)/zImage.initrd.chrp: $(CHRPOBJS) $(obj)/image.initrd.o $(LIBS) $(boot)/ld.script
> > +$(images)/zImage.initrd.chrp: $(CHRPOBJS) $(obj)/image.initrd.o $(LIBS)
> >  	$(call cmd,gen-chrp)
> 
> Here, gen-chrp does have a problem because it uses $^, but it looks to
> me that with a minor change it could just use $< and avoid the
> problem.

IIRC, changing to $< means we have to compilcate the rules a bit more,
or thinking a bit harder, re-arrange the dependancies s.t.
$(obj)/image*.o is first, and gen-chrp (and gen-pmac) both know to use
$(CHRPOBJS) and $(LIBS).

-- 
Tom Rini
http://gate.crashing.org/~trini/

Re: [PATCH] Fix booting some PPC32 machines

[Olaf Hering]

 On Mon, May 03, Tom Rini wrote:

> Hello.  The following patch fixes booting on some PPC32 machines with
> OpenFirmware, when booted without the aid of an additional bootloader.
> The problem is that the linker script for the 'zImage' type targets was
> put into the list of dependancies which objcopy would parse as a list of
> files to copy into the resulting image.  The fix is to make the phony
> zImage targets depend on the linker script.

This fixes netbooting on my B50. But it breaks the dependency to ld.script,
which was the whole point of the previous patch.

-- 
USB is for mice, FireWire is for men!

sUse lINUX ag, nÜRNBERG

Re: [PATCH] Fix booting some PPC32 machines

[Tom Rini]

On Tue, May 04, 2004 at 09:06:06AM +0200, Olaf Hering wrote:
>  On Mon, May 03, Tom Rini wrote:
> 
> > Hello.  The following patch fixes booting on some PPC32 machines with
> > OpenFirmware, when booted without the aid of an additional bootloader.
> > The problem is that the linker script for the 'zImage' type targets was
> > put into the list of dependancies which objcopy would parse as a list of
> > files to copy into the resulting image.  The fix is to make the phony
> > zImage targets depend on the linker script.
> 
> This fixes netbooting on my B50. But it breaks the dependency to ld.script,
> which was the whole point of the previous patch.

As I mentioned on IRC, but don't have time ATM to verify, I don't think
a 'touch' of the ld.script is sufficient a test in this case since it's
a call-if-changed thing so you'd have to modify ld.script.  Can you do
that and see if it's really a change?  Thanks.

-- 
Tom Rini
http://gate.crashing.org/~trini/