Re: [PATCH] RISC-V: Include clone3() on rv32

From: Christian Brauner
Date: Tue Oct 05 2021 - 05:14:57 EST


On Mon, Oct 04, 2021 at 05:35:40PM -0700, Palmer Dabbelt wrote:
> On Mon, 04 Oct 2021 04:17:58 PDT (-0700), christian.brauner@xxxxxxxxxx wrote:
> > On Sun, Oct 03, 2021 at 05:30:24PM +0200, Arnd Bergmann wrote:
> > > On Sun, Oct 3, 2021 at 2:58 AM Palmer Dabbelt <palmer@xxxxxxxxxxx> wrote:
> > > >
> > > > From: Palmer Dabbelt <palmerdabbelt@xxxxxxxxxx>
> > > >
> > > > As far as I can tell this should be enabled on rv32 as well, I'm not
> > > > sure why it's rv64-only. checksyscalls is complaining about our lack of
> > > > clone3() on rv32.
> > > >
> > > > Fixes: 56ac5e213933 ("riscv: enable sys_clone3 syscall for rv64")
> > > > Signed-off-by: Palmer Dabbelt <palmerdabbelt@xxxxxxxxxx>
> > >
> > > We should probably reverse the polarity of this symbol and force
> > > architectures that don't implement it properly to say they don't
> > > have it, but for now, it definitely makes sense to treat this the same
> > > way on 32-bit and 64-bit risc-v.
> >
> > I think we had that discussion back when I added it and I think I even
> > proposed that or you did but then we settled on __ARCH_WANT_SYS_CLONE3.
> > Most likely because it fell in line with the other
> > __ARCH_WANT_SYS_{CLONE,FORK}.
> >
> > I think at this point its alpha, ia64, nios, sparc, and sh that don't
> > implement it. For some it looks trivial at first glance at least (Fwiw,
> > nios implements sys_clone() but doesn't select __ARCH_WANT_SYS_CLONE3):
> >
> > diff --git a/arch/nios2/include/uapi/asm/unistd.h b/arch/nios2/include/uapi/asm/unistd.h
> > index 0b4bb1d41b28..6c4f45abd3ab 100644
> > --- a/arch/nios2/include/uapi/asm/unistd.h
> > +++ b/arch/nios2/include/uapi/asm/unistd.h
> > @@ -18,6 +18,7 @@
> >
> > #define sys_mmap2 sys_mmap_pgoff
> >
> > +#define __ARCH_WANT_SYS_CLONE3
> > #define __ARCH_WANT_RENAMEAT
> > #define __ARCH_WANT_STAT64
> > #define __ARCH_WANT_SET_GET_RLIMIT
> > diff --git a/arch/nios2/kernel/entry.S b/arch/nios2/kernel/entry.S
> > index 0794cd7803df..c1804bda8259 100644
> > --- a/arch/nios2/kernel/entry.S
> > +++ b/arch/nios2/kernel/entry.S
> > @@ -396,6 +396,15 @@ ENTRY(sys_clone)
> > RESTORE_SWITCH_STACK
> > ret
> >
> > +/*
> > + * int clone3(struct clone_args __user *, uargs, size_t, size)
> > + */
> > +ENTRY(sys_clone3)
> > + SAVE_SWITCH_STACK
> > + call sys_clone3
> > + RESTORE_SWITCH_STACK
> > + ret
> > +
> > ENTRY(sys_rt_sigreturn)
> > SAVE_SWITCH_STACK
> > mov r4, sp
>
> Thanks.
>
> I've put this on fixes, but if you're trying to do that refactoring I've
> merged it in as a single patch on top of 5.15-rc1. That's on
> palmer/riscv-clone3, in case it helps someone avoid a conflict when doing
> that refactoring. I'd usually offer to do the refactoring, but I'm super
> buried right now with all the RISC-V stuff ;)
>
> I want to call this fix because it's breaking my builds, these checksyscall
> warnings have recently turned into errors.

Oh yeah, please just merge your fix here for this issue. I think
flipping the symbol isn't that pressing right now and fixing your builds
is more urgent! :)