[PATCH] powerpc/boot: Build wrapper for an appropriate CPU

[Joel Stanley]

On Wed, 30 Mar 2022 at 11:33, Christophe Leroy
<christophe.leroy at csgroup.eu> wrote:
>
>
>
> Le 30/03/2022 à 13:24, Joel Stanley a écrit :
> > Currently the boot wrapper lacks a -mcpu option, so it will be built for
> > the toolchain's default cpu. This is a problem if the toolchain defaults
> > to a cpu with newer instructions.
> >
> > We could wire in TARGET_CPU but instead use the oldest supported option
> > so the wrapper runs anywhere.
> >
> > The GCC documentation stays that -mcpu=powerpc64le will give us a
> > generic 64 bit powerpc machine:
> >
> >   -mcpu=powerpc, -mcpu=powerpc64, and -mcpu=powerpc64le specify pure
> >   32-bit PowerPC (either endian), 64-bit big endian PowerPC and 64-bit
> >   little endian PowerPC architecture machine types, with an appropriate,
> >   generic processor model assumed for scheduling purposes.
> >
> > So do that for each of the three machines.
> >
> > This bug was found when building the kernel with a toolchain that
> > defaulted to powre10, resulting in a pcrel enabled wrapper which fails
> > to link:
> >
> >   arch/powerpc/boot/wrapper.a(crt0.o): in function `p_base':
> >   (.text+0x150): call to `platform_init' lacks nop, can't restore toc; (toc save/adjust stub)
> >   (.text+0x154): call to `start' lacks nop, can't restore toc; (toc save/adjust stub)
> >   powerpc64le-buildroot-linux-gnu-ld: final link failed: bad value
> >
> > Even with tha bug worked around the resulting kernel would crash on a
> > power9 box:
> >
> >   $ qemu-system-ppc64 -nographic -nodefaults -M powernv9 -kernel arch/powerpc/boot/zImage.epapr -serial mon:stdio
> >   [    7.069331356,5] INIT: Starting kernel at 0x20010020, fdt at 0x3068c628 25694 bytes
> >   [    7.130374661,3] ***********************************************
> >   [    7.131072886,3] Fatal Exception 0xe40 at 00000000200101e4    MSR 9000000000000001
> >   [    7.131290613,3] CFAR : 000000002001027c MSR  : 9000000000000001
> >   [    7.131433759,3] SRR0 : 0000000020010050 SRR1 : 9000000000000001
> >   [    7.131577775,3] HSRR0: 00000000200101e4 HSRR1: 9000000000000001
> >   [    7.131733687,3] DSISR: 00000000         DAR  : 0000000000000000
> >   [    7.131905162,3] LR   : 0000000020010280 CTR  : 0000000000000000
> >   [    7.132068356,3] CR   : 44002004         XER  : 00000000
> >
> > Link: https://github.com/linuxppc/issues/issues/400
> > Signed-off-by: Joel Stanley <joel at jms.id.au>
> > ---
> > Tested:
> >
> >   - ppc64le_defconfig
> >   - pseries and powernv qemu, for power8, power9, power10 cpus
> >   - buildroot compiler that defaults to -mcpu=power10 (gcc 10.3.0, ld 2.36.1)
> >   -  RHEL9 cross compilers (gcc 11.2.1-1, ld 2.35.2-17.el9)
> >
> > All decompressed and made it into the kernel ok.
> >
> > ppc64_defconfig did not work, as we've got a regression when the wrapper
> > is built for big endian. It hasn't worked for zImage.pseries for a long
> > time (at least v4.14), and broke some time between v5.4 and v5.17 for
> > zImage.epapr.
> >
> >   arch/powerpc/boot/Makefile | 8 ++++++--
> >   1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/powerpc/boot/Makefile b/arch/powerpc/boot/Makefile
> > index 9993c6256ad2..1f5cc401bfc0 100644
> > --- a/arch/powerpc/boot/Makefile
> > +++ b/arch/powerpc/boot/Makefile
> > @@ -38,9 +38,13 @@ BOOTCFLAGS    := -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs \
> >                $(LINUXINCLUDE)
> >
> >   ifdef CONFIG_PPC64_BOOT_WRAPPER
> > -BOOTCFLAGS   += -m64
> > +ifdef CONFIG_CPU_LITTLE_ENDIAN
> > +BOOTCFLAGS   += -m64 -mcpu=powerpc64le
> >   else
> > -BOOTCFLAGS   += -m32
> > +BOOTCFLAGS   += -m64 -mcpu=powerpc64
> > +endif
> > +else
> > +BOOTCFLAGS   += -m32 -mcpu=powerpc
>
> How does that interracts with the following lines ? Isn't it an issue to
> have two -mcpu ?
>
> arch/powerpc/boot/Makefile:$(obj)/4xx.o: BOOTCFLAGS += -mcpu=405
> arch/powerpc/boot/Makefile:$(obj)/ebony.o: BOOTCFLAGS += -mcpu=440
> arch/powerpc/boot/Makefile:$(obj)/cuboot-hotfoot.o: BOOTCFLAGS += -mcpu=405
> arch/powerpc/boot/Makefile:$(obj)/cuboot-taishan.o: BOOTCFLAGS += -mcpu=440
> arch/powerpc/boot/Makefile:$(obj)/cuboot-katmai.o: BOOTCFLAGS += -mcpu=440
> arch/powerpc/boot/Makefile:$(obj)/cuboot-acadia.o: BOOTCFLAGS += -mcpu=405
> arch/powerpc/boot/Makefile:$(obj)/treeboot-iss4xx.o: BOOTCFLAGS += -mcpu=405
> arch/powerpc/boot/Makefile:$(obj)/treeboot-currituck.o: BOOTCFLAGS +=
> -mcpu=405
> arch/powerpc/boot/Makefile:$(obj)/treeboot-akebono.o: BOOTCFLAGS +=
> -mcpu=405

Good point, I didn't test the other wrappers.

Last one wins as far as -mcpu lines goes, from a quick test. But it
might lead to less confusion if I dropped the -mcpu=powerpc change.