The underlying variable of the counter innodb_buffer_pool_wait_free is only modified in buf_LRU_get_free_block(), where we could easily do that after acquiring buf_pool.mutex.

Furthermore, the variable is being duplicated for no good reason:

	export_vars.innodb_buffer_pool_wait_free =

		srv_stats.buf_pool_wait_free;

Because this counter should not be in a time-critical execution path, I think that we can replace both of the above variables with buf_pool.stat.LRU_waits, which would be incremented while holding buf_pool.mutex, to make the counter more accurate. I think that we can allow the reads to remain dirty, not acquiring the mutex.