Re: [PATCH 1/6] lib: Add two-byte cmpxchg emulation function

[Paul E. McKenney]

On Sat, Sep 12, 2026 at 10:30:47AM +0100, David Laight wrote:
> On Fri, 11 Sep 2026 16:09:16 -0700
> "Paul E. McKenney" <paulmck@xxxxxxxxxx> wrote:
> 
> > On Fri, Sep 11, 2026 at 11:10:17PM +0100, David Laight wrote:
> > > On Fri, 11 Sep 2026 19:25:34 +0000
> > > Bradley Morgan <brads@xxxxxxxxxxxxxx> wrote:
> > >   
> > > > cmpxchg_emu_u8() emulates one-byte cmpxchg() in terms of four-byte
> > > > cmpxchg() for the architectures lacking native one-byte atomics.
> > > > The same architectures also lack native two-byte cmpxchg(), where
> > > > such an operation is not supported and either fails to compile via
> > > > BUILD_BUG() or fails to link, because the bad pointer sentinels
> > > > these architectures declare are never defined.
> > > > 
> > > > Add cmpxchg_emu_u16(), the two-byte sibling. It reads the enclosing
> > > > word with READ_ONCE(), splices the two target bytes through a union
> > > > and loops on cmpxchg() of the full word until the compare succeeds.
> > > > Like cmpxchg_emu_u8() it is fully ordered.
> > > > 
> > > > The Kconfig symbol gating this file is renamed from
> > > > ARCH_NEED_CMPXCHG_1_EMU to ARCH_NEED_CMPXCHG_1_2_EMU, as it now
> > > > selects both the one-byte and the two-byte emulation.
> > > > 
> > > > Suggested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> > > > Signed-off-by: Bradley Morgan <brads@xxxxxxxxxxxxxx>
> > > > ---
> > > >  arch/Kconfig                |  2 +-
> > > >  arch/arc/Kconfig            |  2 +-
> > > >  arch/arm/Kconfig            |  2 +-
> > > >  arch/csky/Kconfig           |  2 +-
> > > >  arch/sh/Kconfig             |  2 +-
> > > >  arch/xtensa/Kconfig         |  2 +-
> > > >  include/linux/cmpxchg-emu.h |  1 +
> > > >  lib/Makefile                |  2 +-
> > > >  lib/cmpxchg-emu.c           | 35 ++++++++++++++++++++++++++++++++---
> > > >  9 files changed, 40 insertions(+), 10 deletions(-)
> > > >   
> > > ...  
> > > > diff --git a/include/linux/cmpxchg-emu.h b/include/linux/cmpxchg-emu.h
> > > > index 998deec67740..fee8171fa05e 100644
> > > > --- a/include/linux/cmpxchg-emu.h
> > > > +++ b/include/linux/cmpxchg-emu.h
> > > > @@ -11,5 +11,6 @@
> > > >  #define __LINUX_CMPXCHG_EMU_H
> > > >  
> > > >  uintptr_t cmpxchg_emu_u8(volatile u8 *p, uintptr_t old, uintptr_t new);
> > > > +uintptr_t cmpxchg_emu_u16(volatile u16 *p, uintptr_t old, uintptr_t new);  
> > > 
> > > Why uintptr_t? Shouldn't it just be u16?
> > > (Which probably means the code would better if it was just 'unsigned int')  
> > 
> > I suspect that Bradley is just following my cmpxchg_emu_u8() example,
> > which also returns uintptr_t.
> > 
> > I remember that *something* broke when I made this be u8, but I cannot
> > recall what the problem was.
> > 
> > Bradley, could you please try making it be u16 as David suggests just to
> > see what happens?  Who knows?  Maybe it was a compiler issue that has
> > since been fixed.  Or maybe the macros and asms using cmpxchg_emu_u8()
> > need that uintptr_t for some reason.
> 
> I think the uintptr (unsigned long) cast is needed to stop a compile
> error when exchanging pointers.
> But that is an issue with the #define not the called functions.
> 
> Possibly changing the #define to have:
> 	unsigned long ul_old = (unsigned long)(old);
> Or even, with the type check from:
> 	unsigned long ul_old = (unsigned long)(0 ? *(ptr) : (old));
> (with the same for 'new')
> and the removing all the casts where the value are used might be better.

It does sound worth a try, especially since we have the uintptr_t version
to fall back on should that fail, or in case it somehow makes things
more complicated.

But we have been living with uintptr_t for some years with the 8-bit
code, so this should be able to be a follow-on patch.

							Thanx, Paul