Re: Mass-building defconfigs: many fail with assembler errors

From: Segher Boessenkool <segher@kernel.crashing.org>
To: Jan-Benedict Glaw <jbglaw@lug-owl.de>
Cc: linuxppc-dev@lists.ozlabs.org
Subject: Re: Mass-building defconfigs: many fail with assembler errors
Date: Wed, 14 Dec 2022 14:08:42 -0600	[thread overview]
Message-ID: <20221214200842.GP25951@gate.crashing.org> (raw)
In-Reply-To: <20221214183632.reqfqdhtatog2vmu@lug-owl.de>

Hi!

On Wed, Dec 14, 2022 at 07:36:32PM +0100, Jan-Benedict Glaw wrote:
> So we have these remaining build issues:
> 
> linux-powerpc-cell_defconfig           bad asm (arch/powerpc/boot/pseries-head.S)
> linux-powerpc-mvme5100_defconfig       bad asm (arch/powerpc/kernel/epapr_hcalls.S)
> linux-powerpc-asp8347_defconfig        bad asm (arch/powerpc/kernel/pmc.c)
> linux-powerpc-ppc6xx_defconfig         bad asm (arch/powerpc/kernel/pmc.c)
> linux-powerpc-ppc64e_defconfig         bad asm (arch/powerpc/kernel/vdso/gettimeofday.S)
> linux-powerpc-corenet64_smp_defconfig  bad asm (arch/powerpc/kernel/vdso/gettimeofday.S)
> 
> I do *not* have CROSS32_COMPILE=... set for my builds. Maybe that
> could cure at least the issues within the ./boot and ./kernel/vdso
> directories?

I never set that, -m32 does the trick, every powerpc compiler is
biarch :-)

> Let's try that...  But I guess that won't help for the
> other two remaining files (arch/powerpc/kernel/{epapr_hcalls.S,pmc.c).

Not likely no.  Can you show the error of those again?

> linux-powerpc-pseries_defconfig         sstep (out of array bounds)
> linux-powerpc-powernv_defconfig         sstep
> linux-powerpc-ppc64_defconfig           sstep
> linux-powerpc-pseries_le_defconfig      sstep
> linux-powerpc-ppc64le_defconfig         sstep
> linux-powerpc-ppc64le_guest_defconfig   sstep
> linux-powerpc-ppc64_guest_defconfig     sstep
> linux-powerpc-powernv_be_defconfig      sstep
> 
> My first guess on these is that it's a wrong warning. The union's
> `u8 b[2 * sizeof(double)]` seems to be large enough.

A false positive, yes.  Which is *not* wrong.  What is wrong is using
-Werror in any unknown environment.  I have a stack of patches I use for
all my kernel builds, and half of those are eradicating harmful -Werror
instances.

> linux-powerpc-akebono_defconfig         ahci (BUILD_BUG_ON failed: sizeof(_s) > sizeof(long))
> linux-powerpc-xes_mpc85xx_defconfig     ahci
> linux-powerpc-ge_imp3a_defconfig        ahci
> linux-powerpc-mpc85xx_defconfig         ahci
> linux-powerpc-mpc85xx_smp_defconfig     ahci
> linux-powerpc-corenet32_smp_defconfig   ahci
> linux-powerpc-mpc86xx_defconfig         ahci
> linux-powerpc-mpc86xx_smp_defconfig     ahci
> 
> I've seen the AHCI issue on other (non-powerpc) builds as well,
> haven't looked into this so I won't guess about whether this is a real
> bug or a compiler issue.

It is a real bug afaics.

Segher