A bottom half handler (where is already implicit as bh's are currently
atomic w.r.t each other and themselves), and user space where it will
prevent a BH from executing during the marked section (and on SMP in 2.1.x
wait for a BH to complete - since a BH could be running on another CPU
in parallel in 2.1.x but not 2.0.x).
> I roll my own? Also, when I look at the defs for the
> start/end_bh_atomic() calls, they reference a function called barrier()
> that is #defined as:
>
> #define barrier() __asm__("": : :"memory")
The instruction is "" (ie nothing) and the "memory" is the thing it affects
- so barrier is a null instruction that arbitarily affects memory space.
Its used to persuade gcc to write back and invalidate any register temporary
variables
Without it something like
start_bh_atomic();
x=list->head;
list->head=list->head->next;
end_bh_atomic();
might be re-ordered by the compiler either so the x=list->head moves before
the beginning of the atomic area or the list->head= is moved past the end.
Its a better alternative than using volatile in most cases
Alan
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/