Date:   Wed, 08 Feb 2023 09:56:06 -0800 (PST)
Subject:     Re: [PATCH] riscv: avoid enabling vectorized code generation
In-Reply-To: <20221216185012.2342675-1-abdulras@google.com>
CC:     Paul Walmsley <paul.walmsley@sifive.com>, aou@eecs.berkeley.edu,
        linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
        abdulras@google.com
From:   Palmer Dabbelt <palmer@dabbelt.com>
To:     abdulras@google.com
Message-ID: <mhng-4c71ada6-003c-414f-9a74-efa3ccd2856b@palmer-ri-x1c9>
Mime-Version: 1.0 (MHng)
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On Fri, 16 Dec 2022 10:50:12 PST (-0800), abdulras@google.com wrote:
> The compiler is free to generate vectorized operations for zero'ing
> memory.  The kernel does not use the vector unit on RISCV, similar to
> architectures such as x86 where we use `-mno-mmx` et al to prevent the
> implicit vectorization.  Perform a similar check for
> `-mno-implicit-float` to avoid this on RISC-V targets.
>
> Signed-off-by: Saleem Abdulrasool <abdulras@google.com>
> ---
>  arch/riscv/Makefile | 4 ++++
>  1 file changed, 4 insertions(+)
>
> diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile
> index 0d13b597cb55..68433476a96e 100644
> --- a/arch/riscv/Makefile
> +++ b/arch/riscv/Makefile
> @@ -89,6 +89,10 @@ KBUILD_AFLAGS_MODULE += $(call as-option,-Wa$(comma)-mno-relax)
>  # architectures.  It's faster to have GCC emit only aligned accesses.
>  KBUILD_CFLAGS += $(call cc-option,-mstrict-align)
>
> +# Ensure that we do not vectorize the kernel code when the `v` extension is
> +# enabled.  This mirrors the `-mno-mmx` et al on x86.
> +KBUILD_CFLAGS += $(call cc-option,-mno-implicit-float)
> +
>  ifeq ($(CONFIG_STACKPROTECTOR_PER_TASK),y)
>  prepare: stack_protector_prepare
>  stack_protector_prepare: prepare0

Sorry to just restart the thread, but there's been discussions on this 
in a bunch of places.  From my understanding, we don't actually need 
this: we have this tricky line in the Makefile

    KBUILD_CFLAGS += -march=$(subst fd,,$(riscv-march-y))

that removes the floating-point extenions from what the kernel is built 
with, so adding `-mno-implicit-float` doesn't actually do anything (and 
we'd end up with essentially the same thing for V when it gets added).

So unless I'm missing something, we don't need this.