[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Ye9Rk4y9Ck7Bg5pv@dev-arch.archlinux-ax161>
Date: Mon, 24 Jan 2022 18:25:39 -0700
From: Nathan Chancellor <nathan@...nel.org>
To: Jiaxun Yang <jiaxun.yang@...goat.com>
Cc: Nick Desaulniers <ndesaulniers@...gle.com>,
Thomas Bogendoerfer <tsbogend@...ha.franken.de>,
Huacai Chen <chenhuacai@...nel.org>,
"linux-mips@...r.kernel.org" <linux-mips@...r.kernel.org>,
linux-kernel@...r.kernel.org, llvm@...ts.linux.dev
Subject: Re: [PATCH 1/2] MIPS: Loongson64: Clean up use of cc-ifversion
On Tue, Jan 25, 2022 at 12:03:32AM +0000, Jiaxun Yang wrote:
>
>
> 在2022年1月24日一月 下午8:31,Nick Desaulniers写道:
> > On Thu, Jan 20, 2022 at 1:40 PM Nathan Chancellor <nathan@...nel.org> wrote:
> >>
> >> This Makefile checks that GCC is 4.9 or newer, which is redundant after
> >> commit 76ae847497bc ("Documentation: raise minimum supported version of
> >> GCC to 5.1"), so cc-option can be removed.
> >>
> >> Clang does not support -march=loongson3a so it needs to continue to use
> >> -march=mips64r2, along with binutils less than 2.25, so check that both
> >> GCC and binutils 2.25 or newer are being used before using
> >> -march=loongson3a. These flags do not need to be checked with cc-option
> >> anymore because all GCC versions support -march=loongson3a and
> >> -march=mips64r2 and all clang versions support -march=mips64r2.
> >>
> >> Signed-off-by: Nathan Chancellor <nathan@...nel.org>
> >
> > Thanks for the patch. I wonder why `_MIPS_ISA` only seems to be set at
> > all for `-march=loongson3a` AFAICT, though that question is orthogonal
> > to this patch. Perhaps the Loongson or MIPS maintainers know more?
> > Otherwise seems like most uses of _MIPS_ISA can either be deleted or
> > simplified now.
>
> This is because earlier GCC mistakenly set loongson3a to MIPS64 not MIPS64R2.
>
> But given that it's earlier than the minimal requirement GCC of kernel today, I think it should be safe to just move the whole logic.
>
> Thanks.
So something like this (completely untested)?
diff --git a/arch/mips/loongson64/Platform b/arch/mips/loongson64/Platform
index 3e660d6d3c2b..36ab2fe00835 100644
--- a/arch/mips/loongson64/Platform
+++ b/arch/mips/loongson64/Platform
@@ -5,24 +5,9 @@
cflags-$(CONFIG_CPU_LOONGSON64) += -Wa,--trap
-#
-# binutils from v2.25 on and gcc starting from v4.9.0 treat -march=loongson3a
-# as MIPS64 R2; older versions as just R1. This leaves the possibility open
-# that GCC might generate R2 code for -march=loongson3a which then is rejected
-# by GAS. The cc-option can't probe for this behaviour so -march=loongson3a
-# can't easily be used safely within the kbuild framework.
-#
-ifeq ($(call cc-ifversion, -ge, 0409, y), y)
- ifeq ($(call ld-ifversion, -ge, 22500, y), y)
- cflags-$(CONFIG_CPU_LOONGSON64) += \
- $(call cc-option,-march=loongson3a -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS64)
- else
- cflags-$(CONFIG_CPU_LOONGSON64) += \
- $(call cc-option,-march=mips64r2,-mips64r2 -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS64)
- endif
-else
- cflags-$(CONFIG_CPU_LOONGSON64) += \
- $(call cc-option,-march=mips64r2,-mips64r2 -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS64)
+ifdef CONFIG_CPU_LOONGSON64
+cflags-$(CONFIG_CC_IS_CLANG) += -march=mips64r2
+cflags-$(CONFIG_CC_IS_GCC) += -march=loongson3a
endif
# Some -march= flags enable MMI instructions, and GCC complains about that
> >
> > Reviewed-by: Nick Desaulniers <ndesaulniers@...gle.com>
> >
> >> ---
> >> arch/mips/loongson64/Platform | 13 +++----------
> >> 1 file changed, 3 insertions(+), 10 deletions(-)
> >>
> >> diff --git a/arch/mips/loongson64/Platform b/arch/mips/loongson64/Platform
> >> index 3e660d6d3c2b..981d3abc150e 100644
> >> --- a/arch/mips/loongson64/Platform
> >> +++ b/arch/mips/loongson64/Platform
> >> @@ -12,17 +12,10 @@ cflags-$(CONFIG_CPU_LOONGSON64) += -Wa,--trap
> >> # by GAS. The cc-option can't probe for this behaviour so -march=loongson3a
> >> # can't easily be used safely within the kbuild framework.
> >> #
> >> -ifeq ($(call cc-ifversion, -ge, 0409, y), y)
> >> - ifeq ($(call ld-ifversion, -ge, 22500, y), y)
> >> - cflags-$(CONFIG_CPU_LOONGSON64) += \
> >> - $(call cc-option,-march=loongson3a -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS64)
> >> - else
> >> - cflags-$(CONFIG_CPU_LOONGSON64) += \
> >> - $(call cc-option,-march=mips64r2,-mips64r2 -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS64)
> >> - endif
> >> +ifeq ($(CONFIG_CC_IS_GCC)$(call ld-ifversion, -ge, 22500, y), yy)
> >> + cflags-$(CONFIG_CPU_LOONGSON64) += -march=loongson3a -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS64
> >> else
> >> - cflags-$(CONFIG_CPU_LOONGSON64) += \
> >> - $(call cc-option,-march=mips64r2,-mips64r2 -U_MIPS_ISA -D_MIPS_ISA=_MIPS_ISA_MIPS64)
> >> + cflags-$(CONFIG_CPU_LOONGSON64) += -march=mips64r2
> >> endif
> >>
> >> # Some -march= flags enable MMI instructions, and GCC complains about that
> >>
> >> base-commit: 2c271fe77d52a0555161926c232cd5bc07178b39
> >> --
> >> 2.34.1
> >>
> >
> >
> > --
> > Thanks,
> > ~Nick Desaulniers
>
> --
> - Jiaxun
Powered by blists - more mailing lists