Re: [PATCH] x86: prevent gcc from emitting SSE into boot code

[Dmitry Vyukov]

On Mon, Jun 27, 2016 at 8:53 PM, Ingo Molnar <mingo@xxxxxxxxxx> wrote:
>
> * Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>
>> On Wed, Jun 22, 2016 at 7:37 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>> > Latest gcc manages to emit MOVDQA into boot code
>> > which leads to crashes. Add more -mno-see flags.
>
> Which GCC version is that?
>
>> >
>> > Signed-off-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
>> > ---
>> >  arch/x86/boot/compressed/Makefile | 2 +-
>> >  1 file changed, 1 insertion(+), 1 deletion(-)
>> >
>> > diff --git a/arch/x86/boot/compressed/Makefile b/arch/x86/boot/compressed/Makefile
>> > index f135688..7eaafb8 100644
>> > --- a/arch/x86/boot/compressed/Makefile
>> > +++ b/arch/x86/boot/compressed/Makefile
>> > @@ -31,7 +31,7 @@ KBUILD_CFLAGS += -DDISABLE_BRANCH_PROFILING
>> >  cflags-$(CONFIG_X86_32) := -march=i386
>> >  cflags-$(CONFIG_X86_64) := -mcmodel=small
>> >  KBUILD_CFLAGS += $(cflags-y)
>> > -KBUILD_CFLAGS += -mno-mmx -mno-sse
>> > +KBUILD_CFLAGS += -mno-mmx -mno-sse -mno-sse2 -mno-sse3
>> >  KBUILD_CFLAGS += $(call cc-option,-ffreestanding)
>> >  KBUILD_CFLAGS += $(call cc-option,-fno-stack-protector)
>> >
>> > --
>> > 2.8.0.rc3.226.g39d4020
>>
>>
>> I withdraw this patch.
>>
>> From off-list discussion:
>>
>> "-mno-sse alone should disable all SSE/AVX instructions. If not,
>> please file a GCC bug report".
>
> So the GCC documentation says:
>
>  -mmmx
>  -mno-mmx
>  -msse
>  -mno-sse
>  -msse2
>  -mno-sse2
>  -msse3
>  -mno-sse3
>  -m3dnow
>  -mno-3dnow
>     These switches enable or disable the use of built-in functions that allow
>     direct access to the MMX, SSE, SSE2, SSE3 and 3Dnow extensions of the
>     instruction set.
>
> Technically -mno-mmx won't turn off SSE instructions because they are independent.
> Do we really want to rely on an undocumented property of -mno-sse (clearly broken
> in at least one released GCC version), especially as the fix is so easy?
>
> Thanks,
>
>         Ingo


Unfortunately part of this discussion gone private. Bringing it public.
This was not necessary a release version of gcc.

=====
The full story is that I spent quite some time debugging it with some
non-release gcc5 version a while ago, and then I was just sitting on
this patch. Now I am using release gcc6. I've just checked and the
kernel boots without these new flags. I thought that -mno-see does not
have to prevent newer SSE instructions, so this patch will save
somebody debugging in future. If the existing -mno-see should prevent
all SSE instructions, then I guess we don't need this patch.
And if not, then, no, I don't have a test case right now.
=====