First, I thought that it was wrong to enter a critical section from
one thread and then leave the same critical section from another
thread. I wrote a small program and it indicates that there is no
error to do so:
#include <windows.h>
#include <process.h>
#include <stdio.h>
struct threadparameter {
CRITICAL_SECTION *criticalsection;
int threadid;
};
// enter / leave critical section in different threads
void testcriticalsection(void *params){
threadparameter *parameter = (threadparameter*)params;
int threadid = parameter->threadid;
switch (threadid) {
case 0:
EnterCriticalSection(parameter->criticalsection);
printf("thread %d entered critical section, error code: %d\n",
threadid,GetLastError());
Sleep(9000);
break;
case 1:
LeaveCriticalSection(parameter->criticalsection);
printf("thread %d left critical section, error code: %d\n",
threadid,GetLastError());
Sleep(5000);
break;
case 2:
printf("thread %d enter critical section\n", threadid);
EnterCriticalSection(parameter->criticalsection);
printf("thread %d entered critical section, error code: %d\n",
threadid,GetLastError());
break;
}
printf("thread %d is done\n", threadid);
}
int main() {
CRITICAL_SECTION criticalsection;
InitializeCriticalSection(&criticalsection);
printf("testing different thread enter / leave critical section\n");
const int threadcount = 3;
threadparameter param[threadcount];
HANDLE th[threadcount];
for (int i = 0; i<threadcount; ++i) {
param[i].threadid = i;
param[i].criticalsection=&criticalsection;
th[i] = (HANDLE)_beginthread(testcriticalsection,0, ¶m[i]);
}
Sleep(1000);
WaitForMultipleObjects(2, th, TRUE, INFINITE);
DeleteCriticalSection(&criticalsection);
printf("main exits\n");
return 0;
}
The first thread call EnterCriticalSection, the second calls Leave and
third calls Enter again. If the Leave call in the second thread has no
effect (since it doesn't own it), the third thread should be blocked.
But it was NOT. If the second thread doesn't call Leave, the third
thread IS blocked which is as expected. To prevent critical section
become undefined (when its owner thread dies), I put some sleep call
in the code (certainly, real application should not use or minimize
Sleep).
Why a thread doesn't own a critical section could release the lock? Is
there something wrong in the code so this is not a valid test?
Another issue illustrated in the code is WaitForMultipleObject. The
first parameter is the number of objects in handle array. in the above
example, the object count should be 3. However, if I put it 3 over
there, the call always returns as soon as ONE thread is done.
So, my second question is why WaitForMultipleObject doesn't wait for
all objects to signal?
It is quite possible that the small program has a fundamental fault
and hence, above issues are just my fault. In that case, I am eagerly
waiting for someone to point it out.
Thanks.
J