Re: [PATCH v6 6/6] locking/lockdep: Reuse freed chain_hlocks entries

[Peter Zijlstra]

On Thu, Feb 06, 2020 at 12:08:20PM -0500, Waiman Long wrote:
> On 2/6/20 11:16 AM, Peter Zijlstra wrote:
> > On Thu, Feb 06, 2020 at 10:24:08AM -0500, Waiman Long wrote:
> >> +static int alloc_chain_hlocks(int req)
> >> +{
> >> +	int bucket, curr, size;
> >> +
> >> +	/*
> >> +	 * We rely on the MSB to act as an escape bit to denote freelist
> >> +	 * pointers. Make sure this bit isn't set in 'normal' class_idx usage.
> >> +	 */
> >> +	BUILD_BUG_ON((MAX_LOCKDEP_KEYS-1) & CHAIN_BLK_FLAG);
> >> +
> >> +	init_data_structures_once();
> >> +
> >> +	if (nr_free_chain_hlocks < req)
> >> +		return -1;
> >> +
> >> +	/*
> >> +	 * We require a minimum of 2 (u16) entries to encode a freelist
> >> +	 * 'pointer'.
> >> +	 */
> >> +	req = max(req, 2);
> >> +	bucket = size_to_bucket(req);
> >> +	curr = chain_block_buckets[bucket];
> >> +
> >> +	if (bucket && (curr >= 0)) {
> >> +		del_chain_block(bucket, req, chain_block_next(curr));
> >> +		return curr;
> >> +	} else if (bucket) {
> >> +		/* Try bucket 0 */
> >> +		curr = chain_block_buckets[0];
> >> +	}
> > 	if (bucket) {
> > 		if (curr >= 0) {
> > 			del_chain_block(bucket, req, chain_block_next(curr));
> > 			return curr;
> > 		}
> > 		/* Try bucket 0 */
> > 		curr = chain_block_bucket[0];
> > 	}
> >
> > reads much easier IMO.
> 
> Yes, that is simpler. I can send out an updated patch if you want, or
> you can apply the change when you pull the patch.

I'll frob it. Thanks!