Win32线程

Question

My application uses CreateThread to create a secondary (and only) thread that executes every 10ms , like the following pseudocode :

map<string, int32_t> a_map;

DWORD WINAPI Table::manual_action_execute_thread(LPVOID lpParameter) {
   while(Table::manual_action_execute_thread_run) {
      ...
      if (!Table::automatic_action_execute_inprogress) {
         ...
      }
      ...
      if (a_map["blah"] == 0) {
         ...
      }
      ...
      Sleep(10);
   }
   return 0;
}

The variables are declared as following:

static volatile bool manual_action_execute_thread_run;
static volatile bool automatic_action_execute_inprogress;

The first one takes the value of true before even start my thread so i don't use locking on this. The second takes a false at first.

I am using ::automatic_action_execute_inprogress to control some behaviour on second thread which only changes on the main thread .

QUESTION(s):

1) Since i only update ::automatic_action_execute_inprogress on main thread and just reading it on second thread i still need to lock it first using EnterCriticalSection ? Or the locking is only restricted to the shared variables that are changed on both threads?

2) What about a <map> that's used on multiple threads and modified by just one? Certainly i have to lock it with EnterCriticalSection whenever changes but what about read access? Should i lock it when i read from it (like if (a_map["foo"] == 0) ) if can change even by a single thread? Like this for example?

EnterCriticalSection(&cs);
   bool val = a_map["foo"];
LeaveCriticalSection(&cs);

   if (val == 0) {
      ...
   }

Answer 1

Since you only update your variable in one thread and just read it in other threads, and since the type of the variable is atomic , you do not need any locking mechanism. All you need to do is declare it volatile , which you have already done. If the variable was of a non-atomic type, (say, a map , or even just a 64-bit integer on a 32-bit architecture,) then you would need locking.

The C & C++ standard does not presume atomicity of operations even on simple machine-word operations like reading an int , but if you are using winapi then you are on x86 , which guarantees it. (So, if you want to write portable code, use locking even if you are writing an int on one thread and only reading it from other threads.)