I agree with your previous conclusion that there are no race conditions during probing. However this patch is regarding the function probe_table and a possible init failure. Note that this race condition will never occur if all init's are successful (as was probably the case in your previous testing).

Consider the scenario where the same position is probed from two separate threads and init has not yet been done for the corresponding tb. If init_table fails it sets tbHash[hashIdx].ptr to NULL but the second thread checks the variable tbHash[hashIdx].ptr and then uses it if it is not NULL. This is a race condition.

We could either lock each time when checking the tbHash[hashIdx].ptr variable, but this is unnecessarily slow. My patch introduces a variable called error which handles this scenario.