Re: [PATCH 1/7] cleanup: Introduce DEFINE_ACQUIRE() a CLASS() for conditional locking

[<dan.j.williams@intel.com>]

Peter Zijlstra wrote:
> On Mon, May 12, 2025 at 08:25:59PM +0200, Peter Zijlstra wrote:
> > On Mon, May 12, 2025 at 12:50:26PM +0200, Peter Zijlstra wrote:
> > 
> > > +#define __GUARD_ERR(_ptr) \
> > > +	({ long _rc = (__force unsigned long)(_ptr); \
> > > +	   if (!_rc) { _rc = -EBUSY; } if (!IS_ERR_VALUE(_rc)) { _rc = 0; } \
> > > +	   _rc; })
> > > +
> > 
> > >  #define DEFINE_GUARD(_name, _type, _lock, _unlock) \
> > > -	DEFINE_CLASS(_name, _type, if (_T) { _unlock; }, ({ _lock; _T; }), _type _T); \
> > > +	DEFINE_CLASS(_name, _type, if (!__GUARD_ERR(_T)) { _unlock; }, ({ _lock; _T; }), _type _T); \
> > >  	DEFINE_CLASS_IS_GUARD(_name)
> > 
> > GCC is 'stupid' and this generates atrocious code. I'll play with it.
> 
> PRE:
>     bf9e:       48 85 db                test   %rbx,%rbx
>     bfa1:       74 1a                   je     bfbd <foo+0x5d>
>     bfa3:       48 81 fb 00 f0 ff ff    cmp    $0xfffffffffffff000,%rbx
>     bfaa:       77 11                   ja     bfbd <foo+0x5d>
> 
> POST:
>     bf9e:       48 8d 43 ff             lea    -0x1(%rbx),%rax
>     bfa2:       48 3d ff ef ff ff       cmp    $0xffffffffffffefff,%rax
>     bfa8:       77 11                   ja     bfbb <foo+0x5b>
> 

FWIW this looks good to me, and the conversion to drop all explicit
locking passed tests. I threw it out on a branch to get some bot
coverage in the meantime.

https://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl.git/log/?h=cxl-acquire