Hi YunQiang,
On Fri, Dec 14, 2018 at 11:56:05PM +0800, YunQiang Su wrote:
> On the Loongson-2G/2H/3A/3B there is a hardware flaw that ll/sc and
> lld/scd is very weak ordering. We should add sync instructions before
> each ll/lld and after the last sc/scd to workaround. Otherwise, this
> flaw will cause deadlock occationally (e.g. when doing heavy load test
> with LTP).
>
> We introduced an gcc/as option "-mfix-loongson3-llsc", this option
> inserts sync before ll, and so some addresses in __ex_table will need
> to be shift.
>
> This is based on the patch from Huacai Chen.
I much prefer this less invasive version of the patch - thanks! I'd like
to know that the gcc & binutils parts are definitely going to be merged
before I take this though. Please let me know when that happens.
> diff --git a/arch/mips/Makefile b/arch/mips/Makefile
> index b6303e48d..360ee1c30 100644
> --- a/arch/mips/Makefile
> +++ b/arch/mips/Makefile
> @@ -194,6 +194,11 @@ cflags-$(CONFIG_CPU_CAVIUM_OCTEON) += -Wa,-march=octeon
> endif
> cflags-$(CONFIG_CAVIUM_CN63XXP1) += -Wa,-mfix-cn63xxp1
> cflags-$(CONFIG_CPU_BMIPS) += -march=mips32 -Wa,-mips32 -Wa,--trap
> +ifeq ($(CONFIG_CPU_LOONGSON3),y)
> +cflags-y += $(call cc-option,-mfix-loongson3-llsc,)
> +else
> +cflags-y += $(call cc-option,-mno-fix-loongson3-llsc,)
> +endif
The fix-loongson3-llsc option will be default, right? At least for
everything other than -march=octeon3. As such the above should just be:
cflags-$(CONFIG_CPU_LOONGSON3) += $(call cc-option,-mfix-loongson3-llsc,)
Or possibly, if we decide that we only want to support loongson3 kernels
built with a compiler that supports this which it seems may make sense,
it could just be:
cflags-$(CONFIG_CPU_LOONGSON3) += -mfix-loongson3-llsc
Thanks,
Paul
