two bugs in Chapter 28

[98zyx]

1. the usage of setpark() in 28.14 I think the code of setpark() should be: queue_add(m-q, gettid()); m->guard=0; setpark();

rather than the code: queue_add(m-q, gettid()); setpark(); m->guard=0;

Because if we adopt the second way, like the book, the guard lock will be possessed by the thread that calls setpark() until the thread unpark(). In the sleeping time, any other thread can not call unpark() in unlock(), because only the thread which possesses the guard lock can call unpark(), but meanwhile the guard lock is possessed by the sleep thread, which may degenrate to a spin lock.

2. the code in mutex_unlock() in 28.15 I think the code of mutex_unlock() should be: if (!atomic_add_zero(mutex, 0x80000000))

rather than: if (atomic_add_zero(mutex, 0x80000000))

If my guess is correct, the atomic_add_zero() should return the new value of mutex. So, if the old value of mutex is 0x80000000, which means there is no waiting thread, mutex_unlock should return immediately, but the new value of mutex is 0x80000000+0x80000000=0x00000000 and the expression is false.

[remzi-arpacidusseau]

thanks. 1 - if you set the guard to zero first, you have the same race unfortunately. 2 - check out https://xiayingp.gitbook.io/build_a_os/lock/untitled-3