RE: [PATCH] locking/atomic: Make test_and_*_bit() ordered on failure

[David Laight]

...
>  	p += BIT_WORD(nr);
> -	if (READ_ONCE(*p) & mask)
> -		return 1;
> -
>  	old = arch_atomic_long_fetch_or(mask, (atomic_long_t *)p);
>  	return !!(old & mask);
>  }

This looks like the same pattern (attempting to avoid a
locked bus cycle) that caused the qdisc code to sit on
transmit packets (even on x86).
That had some barriers in it (possibly nops on x86) that
didn't help - although the comments suggested otherwise.

I wonder if the pattern has been used anywhere else?

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)