Memory address alignment and atomic operations.

[Tetsuo Handa <from-kernelnewbies@xxxxxxxxxxxxxxxxxxx>]

Hello.

I want to use something like RCU (read copy update).
I want to "read a one-way linked list without a lock"
and "add to the tail of the list".

 1: typedef struct foo {
 2:   struct foo *next;
 3:   char c;
 4:   int i;
 5: } FOO;
 6:
 7: static FOO first = { NULL, 0, 0 };
 8: 
 9: void add_tail(FOO *next) {
10:   static spinlock_t lock = SPIN_LOCK_UNLOCKED;
11:   FOO *f = &first;
12:   spin_lock(&lock);
13:   while (f->next) f = f->next;
14:   next->next = NULL;
15:   wmb();
16:   f->next = next;
17:   spin_unlock(&lock);
18: }

And I want to confirm some requirements for above code work properly.

(1) Is "first" always located as &first % sizeof(int) == 0 ?

(2) Is "first.i" always located as &first.i % sizeof(int) == 0 ?
    (Do I need to use #pragma ?)

(3) Are there more requirements other than the following two
    for atomic read/write operations ?
    #1: 1 <= sizeof(var1) <= sizeof(int)
    #2: &var1 % sizeof(int) == 0
    #3: (Please mention if any.)

Regards.
--
   Tetsuo Handa

--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive:       http://mail.nl.linux.org/kernelnewbies/
FAQ:           http://kernelnewbies.org/faq/