Re: [linux-stable-rc:linux-5.15.y 6874/8723] arch/arm/lib/xor-neon.c:30:2: warning: This code requires at least version 4.6 of GCC

From: Ard Biesheuvel
Date: Mon Jul 25 2022 - 08:42:02 EST


On Mon, 25 Jul 2022 at 10:49, Greg Kroah-Hartman
<gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Jul 25, 2022 at 07:42:07AM +0200, Ard Biesheuvel wrote:
> > On Mon, 25 Jul 2022 at 00:40, Jason A. Donenfeld <Jason@xxxxxxxxx> wrote:
> > >
> > > Hi Ard,
> > >
> > > On Mon, Jul 25, 2022 at 04:14:48AM +0800, kernel test robot wrote:
> > > > All warnings (new ones prefixed by >>):
> > > >
> > > > >> arch/arm/lib/xor-neon.c:30:2: warning: This code requires at least version 4.6 of GCC [-W#warnings]
> > > > #warning This code requires at least version 4.6 of GCC
> > > > ^
> > > > 1 warning generated.
> > > >
> > > >
> > > > vim +30 arch/arm/lib/xor-neon.c
> > > >
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 16
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 17 /*
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 18 * Pull in the reference implementations while instructing GCC (through
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 19 * -ftree-vectorize) to attempt to exploit implicit parallelism and emit
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 20 * NEON instructions.
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 21 */
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 22 #if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6)
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 23 #pragma GCC optimize "tree-vectorize"
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 24 #else
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 25 /*
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 26 * While older versions of GCC do not generate incorrect code, they fail to
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 27 * recognize the parallel nature of these functions, and emit plain ARM code,
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 28 * which is known to be slower than the optimized ARM code in asm-arm/xor.h.
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 29 */
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 @30 #warning This code requires at least version 4.6 of GCC
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 31 #endif
> > > > 01956597cbc46d Ard Biesheuvel 2013-05-17 32
> > >
> > > Does this file need a depends on ≥4.6 thing in Kconfig? Or is something
> > > else happening here since 4.6 isn't even supported by the kernel these
> > > days?
> > >
> >
> > Hi Jason,
> >
> > This is explained/fixed in
> >
> > commit a69cb445f7d129abf7c50d48c8a8eca7c8d5df15
> > Author: Ard Biesheuvel <ardb@xxxxxxxxxx>
> > Date: Sat Feb 5 16:23:46 2022 +0100
> >
> > crypto: arm/xor - make vectorized C code Clang-friendly
> >
> > The ARM version of the accelerated XOR routines are simply the 8-way C
> > routines passed through the auto-vectorizer with SIMD codegen enabled.
> > This used to require GCC version 4.6 at least, but given that 5.1 is now
> > the baseline, this check is no longer necessary, and actually
> > misidentifies Clang as GCC < 4.6 as Clang defines the GCC major/minor as
> > well, but makes no attempt at doing this in a way that conveys feature
> > parity with a certain version of GCC (which would not be a great idea in
> > the first place).
> >
> > which landed in v5.18. Note that the patch depends on its parent,
> > which touches many different architectures, so it may not be
> > backportable cleanly.
>
> I would love to backport this as it bugs me to no end on some various
> gerrit systems all the time. I'll look into it later this week unless
> someone beats me to it :)
>

If backporting is an issue, we could just drop this warning - GCC
before 4.6 hasn't been supported for a while anyway, and Clang never
had the SIMD accelerated XOR to begin with (it needs the __restrict
annotations that the parent patch introduces)