Re: [PATCH -tip 1/2] x86/alternative: Sync bp_patching update for avoiding NULL pointer exception

[Peter Zijlstra]

On Wed, Dec 11, 2019 at 01:44:01AM +0900, Masami Hiramatsu wrote:

> This looks good, but I feel it is a bit complicated.
> 
> If we use atomic (and spin-wait) here, can we use atomic_inc_not_zero()
> in the poke_int3_handler() at first for making sure the bp_batching is
> under operation or not?
> I think it makes things simpler, like below.
> 
> ---------
> atomic_t bp_refcnt;
> 
> poke_int3_handler()
> {
> 	smp_rmb();
> 	if (!READ_ONCE(bp_patching.nr_entries))
> 		return 0;
> 	if (!atomic_inc_not_zero(&bp_refcnt))
> 		return 0;
> 	smp_mb__after_atomic();
> 	[use bp_patching]
> 	atomic_dec(&bp_refcnt);
> }
> 
> text_poke_bp_batch()
> {
> 	bp_patching.vec = tp;
> 	bp_patching.nr_entries = nr_entries;
> 	smp_wmb();
> 	atomic_inc(&bp_refcnt);
> 	...
> 	atomic_dec(&bp_refcnt);
> 	/* wait for all running poke_int3_handler(). */
> 	atomic_cond_read_acquire(&bp_refcnt, !VAL);
> 	bp_patching.vec = NULL;
> 	bp_patching.nr_entries = 0;
> }

I feel that is actually more complicated...  Let me try to see if I can
simplify things.