Re: linux-next: Tree for Jan 18

[Ard Biesheuvel]

(+ Arnd)

On 18 January 2017 at 20:44, Paul Gortmaker
<paul.gortmaker@xxxxxxxxxxxxx> wrote:
> On Wed, Jan 18, 2017 at 12:32 AM, Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx> wrote:
>> Hi all,
>>
>> Changes since 20170117:
>>
>> The arm-soc tree gained a conflict against the arm tree.
>>
>> The amlogic tree gained a conflict against the arm-soc tree.
>>
>> Non-merge commits (relative to Linus' tree): 3676
>>  4511 files changed, 139928 insertions(+), 85028 deletions(-)
>
> Crypto now fails on arm allmodconfig:
>
> arch/arm/crypto/Makefile:25: These ARMv8 Crypto Extensions modules
> need binutils 2.23 or higher
> arch/arm/crypto/Makefile:26:  aes-arm-ce.o sha1-arm-ce.o sha2-arm-ce.o
> ghash-arm-ce.o crct10dif-arm-ce.o crc32-arm-ce.o
>   AS [M]  arch/arm/crypto/aes-neonbs-core.o
>   CC [M]  arch/arm/crypto/aes-neonbs-glue.o
> /home/paul/git/linux-head/arch/arm/crypto/aes-neonbs-core.S: Assembler messages:
> /home/paul/git/linux-head/arch/arm/crypto/aes-neonbs-core.S:817:
> Error: bad type for scalar -- `vmov q1h[1],r10'
> /home/paul/git/linux-head/arch/arm/crypto/aes-neonbs-core.S:817:
> Error: bad type for scalar -- `vmov q1h[0],r9'
> /home/paul/git/linux-head/arch/arm/crypto/aes-neonbs-core.S:817:
> Error: bad type for scalar -- `vmov q1l[1],r8'
> /home/paul/git/linux-head/arch/arm/crypto/aes-neonbs-core.S:817:
> Error: bad type for scalar -- `vmov q1l[0],r7'
> /home/paul/git/linux-head/arch/arm/crypto/aes-neonbs-core.S:818:
> Error: bad type for scalar -- `vmov q2h[1],r10'
> /home/paul/git/linux-head/arch/arm/crypto/aes-neonbs-core.S:818:
> Error: bad type for scalar -- `vmov q2h[0],r9'
>
> [many more similar scalar errors deleted]
>
> The 2.2.3 warning was there in the past, but it wasn't fatal.  I'm using
> an older binutils 2.22 from the kernel.org crosstool stuff.
>
> Seems sfr's builds get the same issue:
>
> http://kisskb.ellerman.id.au/kisskb/buildresult/12910945/
>
> A mindless "git bisect run" comes up with:
>
> cc477bf645736739e69d31fdf715281ef0dd5f9b is the first bad commit
> commit cc477bf645736739e69d31fdf715281ef0dd5f9b
> Author: Ard Biesheuvel <ard.biesheuvel@xxxxxxxxxx>
> Date:   Wed Jan 11 16:41:54 2017 +0000
>
>     crypto: arm/aes - replace bit-sliced OpenSSL NEON code
>
> Maybe binnutils 2.22 is finally too old?
>

I am not sure what the current policy is (Arnd?), but if all errors
pertain to vmovs from scalar to NEON register, I suppose I could work
around it. But perhaps the best solution is the simplest one: not
build the module for < 2.23, like we do for the Crypto Extensions
modules.