[RFC] RISC-V: Don't set CLONE_BACKWARDS

[RFC] RISC-V: Don't set CLONE_BACKWARDS

[Palmer Dabbelt]

During the glibc upstreaming it was suggested that CLONE_BACKWARDS was a
deprecated ABI decision.  I think we just copied it from ARM, but I
don't see any reason to favor one over the other.

While we haven't released yet so I think it's still legal to change our
ABI, I'd actually kind of prefer to avoid changing our ABI this late in
the game.  I guess this is more of an RFC than a patch: is there a
reason to avoid CLONE_BACKWARDS?

Note that I haven't tried any of this -- I'll give it some thourough
testing and submit an actual patch if this is the way we want to go.

CC: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
---
 arch/riscv/Kconfig | 1 -
 1 file changed, 1 deletion(-)

diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index 2c6adf12713a..02076f3a2883 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -10,7 +10,6 @@ config RISCV
 	select OF_IRQ
 	select ARCH_HAS_ATOMIC64_DEC_IF_POSITIVE
 	select ARCH_WANT_FRAME_POINTERS
-	select CLONE_BACKWARDS
 	select COMMON_CLK
 	select GENERIC_CLOCKEVENTS
 	select GENERIC_CPU_DEVICES
-- 
2.13.6

Re: [patches] [RFC] RISC-V: Don't set CLONE_BACKWARDS

[Christoph Hellwig]

On Mon, Jan 08, 2018 at 05:27:56PM -0800, Palmer Dabbelt wrote:
> During the glibc upstreaming it was suggested that CLONE_BACKWARDS was a
> deprecated ABI decision.  I think we just copied it from ARM, but I
> don't see any reason to favor one over the other.
> 
> While we haven't released yet so I think it's still legal to change our
> ABI, I'd actually kind of prefer to avoid changing our ABI this late in
> the game.  I guess this is more of an RFC than a patch: is there a
> reason to avoid CLONE_BACKWARDS?
> 
> Note that I haven't tried any of this -- I'll give it some thourough
> testing and submit an actual patch if this is the way we want to go.

I see absolutely no reason to change this.  Linux currently has 30
architecture port, out of which 10 (including riscv, i386, arm and arm64)
set CLONE_BACKWARDS.

There are no performance benefits of doing it one way or another, and
changing it now will break all the riscv enablement that's been going
on.

Re: [patches] [RFC] RISC-V: Don't set CLONE_BACKWARDS

[Palmer Dabbelt]

On Tue, 09 Jan 2018 00:11:45 PST (-0800), hch@lst.de wrote:
> On Mon, Jan 08, 2018 at 05:27:56PM -0800, Palmer Dabbelt wrote:
>> During the glibc upstreaming it was suggested that CLONE_BACKWARDS was a
>> deprecated ABI decision.  I think we just copied it from ARM, but I
>> don't see any reason to favor one over the other.
>>
>> While we haven't released yet so I think it's still legal to change our
>> ABI, I'd actually kind of prefer to avoid changing our ABI this late in
>> the game.  I guess this is more of an RFC than a patch: is there a
>> reason to avoid CLONE_BACKWARDS?
>>
>> Note that I haven't tried any of this -- I'll give it some thourough
>> testing and submit an actual patch if this is the way we want to go.
>
> I see absolutely no reason to change this.  Linux currently has 30
> architecture port, out of which 10 (including riscv, i386, arm and arm64)
> set CLONE_BACKWARDS.
>
> There are no performance benefits of doing it one way or another, and
> changing it now will break all the riscv enablement that's been going
> on.

OK, works for me!  Unless anyone has a strong argument against CLONE_BACKWARDS 
we're just going to leave it alone.