以只读模式线程锁定结构（C11 stdatomic）

Question

I've created a mutex-like lock function with stdatomic that basically does

do atomic_store(&zero, 0);
while (!atomic_compare_exchange_weak(&mystructure->address, &zero, threadlocal_address));
atomic_fetch_add(&structure->locklevel, 1);

and another unlock function that decrements locklevel and sets the address back to zero ONLY IF locklevel==0 (this allows locking before calling a function which calls another lock, allowing it to still remain locked)

this works fine for RW locking, but I also wanted to make a R-lock function (structure can be read by multiple functions, but trying to write it [calling a RW-lock] will put it to wait) and I tried by simply using a public address instead of a threadlocal one.

The problem with this approach is it would lead to starvation of RW operations if many threads call R-lock in a loop (as locklevel would never [or hardly ever] get to zero)... any ideas?

Answer 1

You can try to use priority for read-write operations. I am assuming that rw/read ratio is low in your system. Implement a basic queue for rw requests and while there is a rw request waiting in the queue, prevent read requests to acquire the lock. After current read requests completed, acquire lock for rw and wait for it to complete. So as long as there is a rw request waiting in the queue or currently running, block read requests. When rw thread unlocks and there is no more rw requests waiting in the queue allow read requests to acquire lock.

You may also need to use condition variables and signaling them depending on your implementation.