Re: [PATCH bpf-next v9 1/6] locking/local_lock: Introduce localtry_lock_t

[Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>]

On 2025-03-11 16:44:30 [+0100], Mateusz Guzik wrote:
> On Fri, Feb 21, 2025 at 06:44:22PM -0800, Alexei Starovoitov wrote:
> > +#define __localtry_lock(lock)					\
> > +	do {							\
> > +		localtry_lock_t *lt;				\
> > +		preempt_disable();				\
> > +		lt = this_cpu_ptr(lock);			\
> > +		local_lock_acquire(&lt->llock);			\
> > +		WRITE_ONCE(lt->acquired, 1);			\
> > +	} while (0)
> 
> I think these need compiler barriers.
> 
> I checked with gcc docs (https://gcc.gnu.org/onlinedocs/gcc/Volatiles.html)
> and found this as confirmation:
> > Accesses to non-volatile objects are not ordered with respect to volatile accesses.
> 
> Unless the Linux kernel is built with some magic to render this moot(?).

You say we need a barrier() after the WRITE_ONCE()? If so, we need it in
the whole file…

Sebastian