How to invalidate cache on mmap'ed shared memory (multi processor machine)

Question

I have two processes, sharing a memory area for fast communication. Both processes run the following (error check removed):

int fd = open("/run/afile", O_RDWR );
void *ptr = mmap(NULL, 4096, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
volatile struct data *shared_data = (struct data *) ptr;

The mmap is backed by the previously created file /run/afile, ( /run is mounted as tempfs, I am assuming 'afile' lives in RAM)

Process A writes data like this:

data->field1 = 34;
data->field2 = 231;
data->new_data_flag = 1;

Process B, at certain points, checks for data->new_data_flag, and then reads data->field1 and data->field2

The question is, how can I force the shared memory area to be updated so the changes are visible for Process B?

I understand msync() will update /run/afile; is this automatically invalidating Process B cache on its own mapping of /run/afile?

Side question: I am using volatile on struct data to prevent the compiler from reordering the writes into field1, field2 and new_data_flag. Is this correct?