Re: [PATCH 10/17] RISC-V: Atomic and Locking Code

[Peter Zijlstra]

On Wed, Jul 12, 2017 at 08:40:49PM +0800, Boqun Feng wrote:
> > +/**
> > + * set_bit - Atomically set a bit in memory
> > + * @nr: the bit to set
> > + * @addr: the address to start counting from
> > + *
> > + * This function is atomic and may not be reordered.  See __set_bit()
> 
> This is incorrect, {set,change,clear}_bit() can be reordered, see
> Documentation/memory-barriers.txt, they are just relaxed atomics. But I
> think you just copy this from x86 code, so maybe x86 code needs help
> too, at least claim that's only x86-specific guarantee.

Yeah, I suspect that's an x86 special (all our atomics are fully
ordered).

> > +/**
> > + * test_and_set_bit_lock - Set a bit and return its old value, for lock
> > + * @nr: Bit to set
> > + * @addr: Address to count from
> > + *
> > + * This operation is atomic and provides acquire barrier semantics.
> > + * It can be used to implement bit locks.
> > + */
> > +static inline int test_and_set_bit_lock(
> > +	unsigned long nr, volatile unsigned long *addr)
> > +{
> > +	return test_and_set_bit(nr, addr);
> 
> If you want, you can open code an "amoor.aq" here, because
> test_and_set_bit_lock() only needs an acquire barrier.
> 
> > +}
> > +
> > +/**
> > + * clear_bit_unlock - Clear a bit in memory, for unlock
> > + * @nr: the bit to set
> > + * @addr: the address to start counting from
> > + *
> > + * This operation is atomic and provides release barrier semantics.
> > + */
> > +static inline void clear_bit_unlock(
> > +	unsigned long nr, volatile unsigned long *addr)
> > +{
> 
> You need a smp_mb__before_atomic() here, because clear_bit() is only
> relaxed atomic. And clear_bit_unlock() is a release.

alternatively you can do "amoand.rl".