sched: membarrier: document memory ordering scenarios

Document membarrier ordering scenarios in membarrier.c. Thanks to Alan
Stern for refreshing my memory. Now that I have those in mind, it seems
appropriate to serialize them to comments for posterity.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20201020134715.13909-4-mathieu.desnoyers@efficios.com

kernel/sched/membarrier.c

@@ -6,6 +6,134 @@
  */
 #include "sched.h"
 
+/*
+ * For documentation purposes, here are some membarrier ordering
+ * scenarios to keep in mind:
+ *
+ * A) Userspace thread execution after IPI vs membarrier's memory
+ *    barrier before sending the IPI
+ *
+ * Userspace variables:
+ *
+ * int x = 0, y = 0;
+ *
+ * The memory barrier at the start of membarrier() on CPU0 is necessary in
+ * order to enforce the guarantee that any writes occurring on CPU0 before
+ * the membarrier() is executed will be visible to any code executing on
+ * CPU1 after the IPI-induced memory barrier:
+ *
+ *         CPU0                              CPU1
+ *
+ *         x = 1
+ *         membarrier():
+ *           a: smp_mb()
+ *           b: send IPI                       IPI-induced mb
+ *           c: smp_mb()
+ *         r2 = y
+ *                                           y = 1
+ *                                           barrier()
+ *                                           r1 = x
+ *
+ *                     BUG_ON(r1 == 0 && r2 == 0)
+ *
+ * The write to y and load from x by CPU1 are unordered by the hardware,
+ * so it's possible to have "r1 = x" reordered before "y = 1" at any
+ * point after (b).  If the memory barrier at (a) is omitted, then "x = 1"
+ * can be reordered after (a) (although not after (c)), so we get r1 == 0
+ * and r2 == 0.  This violates the guarantee that membarrier() is
+ * supposed by provide.
+ *
+ * The timing of the memory barrier at (a) has to ensure that it executes
+ * before the IPI-induced memory barrier on CPU1.
+ *
+ * B) Userspace thread execution before IPI vs membarrier's memory
+ *    barrier after completing the IPI
+ *
+ * Userspace variables:
+ *
+ * int x = 0, y = 0;
+ *
+ * The memory barrier at the end of membarrier() on CPU0 is necessary in
+ * order to enforce the guarantee that any writes occurring on CPU1 before
+ * the membarrier() is executed will be visible to any code executing on
+ * CPU0 after the membarrier():
+ *
+ *         CPU0                              CPU1
+ *
+ *                                           x = 1
+ *                                           barrier()
+ *                                           y = 1
+ *         r2 = y
+ *         membarrier():
+ *           a: smp_mb()
+ *           b: send IPI                       IPI-induced mb
+ *           c: smp_mb()
+ *         r1 = x
+ *         BUG_ON(r1 == 0 && r2 == 1)
+ *
+ * The writes to x and y are unordered by the hardware, so it's possible to
+ * have "r2 = 1" even though the write to x doesn't execute until (b).  If
+ * the memory barrier at (c) is omitted then "r1 = x" can be reordered
+ * before (b) (although not before (a)), so we get "r1 = 0".  This violates
+ * the guarantee that membarrier() is supposed to provide.
+ *
+ * The timing of the memory barrier at (c) has to ensure that it executes
+ * after the IPI-induced memory barrier on CPU1.
+ *
+ * C) Scheduling userspace thread -> kthread -> userspace thread vs membarrier
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *           a: smp_mb()
+ *                                           d: switch to kthread (includes mb)
+ *           b: read rq->curr->mm == NULL
+ *                                           e: switch to user (includes mb)
+ *           c: smp_mb()
+ *
+ * Using the scenario from (A), we can show that (a) needs to be paired
+ * with (e). Using the scenario from (B), we can show that (c) needs to
+ * be paired with (d).
+ *
+ * D) exit_mm vs membarrier
+ *
+ * Two thread groups are created, A and B.  Thread group B is created by
+ * issuing clone from group A with flag CLONE_VM set, but not CLONE_THREAD.
+ * Let's assume we have a single thread within each thread group (Thread A
+ * and Thread B).  Thread A runs on CPU0, Thread B runs on CPU1.
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *             a: smp_mb()
+ *                                           exit_mm():
+ *                                             d: smp_mb()
+ *                                             e: current->mm = NULL
+ *             b: read rq->curr->mm == NULL
+ *             c: smp_mb()
+ *
+ * Using scenario (B), we can show that (c) needs to be paired with (d).
+ *
+ * E) kthread_{use,unuse}_mm vs membarrier
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *           a: smp_mb()
+ *                                           kthread_unuse_mm()
+ *                                             d: smp_mb()
+ *                                             e: current->mm = NULL
+ *           b: read rq->curr->mm == NULL
+ *                                           kthread_use_mm()
+ *                                             f: current->mm = mm
+ *                                             g: smp_mb()
+ *           c: smp_mb()
+ *
+ * Using the scenario from (A), we can show that (a) needs to be paired
+ * with (g). Using the scenario from (B), we can show that (c) needs to
+ * be paired with (d).
+ */
+
 /*
  * Bitmask made from a "or" of all commands within enum membarrier_cmd,
  * except MEMBARRIER_CMD_QUERY.