kernel/sched/membarrier.c

5  * membarrier system call
9  * For documentation purposes, here are some membarrier ordering
12  * A) Userspace thread execution after IPI vs membarrier's memory
19  * The memory barrier at the start of membarrier() on CPU0 is necessary in
21  * the membarrier() is executed will be visible to any code executing on
27  *         membarrier():
42  * and r2 == 0.  This violates the guarantee that membarrier() is
48  * B) Userspace thread execution before IPI vs membarrier's memory
55  * The memory barrier at the end of membarrier() on CPU0 is necessary in
57  * the membarrier() is executed will be visible to any code executing on
58  * CPU0 after the membarrier():
66  *         membarrier():
77  * the guarantee that membarrier() is supposed to provide.
82  * C) Scheduling userspace thread -> kthread -> userspace thread vs membarrier
86  *           membarrier():
97  * D) exit_mm vs membarrier
106  *           membarrier():
116  * E) kthread_{use,unuse}_mm vs membarrier
120  *           membarrier():
176 	 * The smp_mb() in membarrier after all the IPIs is supposed to
178 	 * become visible to membarrier()'s caller -- see scenario B in
183 	 * after membarrier()'s smp_mb().
396 		 * For regular membarrier, we can save a few cycles by
405 		 * membarrier, we will end up with some thread in the mm
461 	 * scheduler executions will observe @mm's new membarrier
468 	 * @mm's membarrier state set bits are also set in the runqueue's
469 	 * membarrier state. This ensures that a runqueue scheduling
470 	 * between threads which are users of @mm has its membarrier state
625 SYSCALL_DEFINE3(membarrier, int, cmd, unsigned int, flags, int, cpu_id)