[xen master] xen: RCU: don't let a CPU with a callback go idle.

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view

[xen master] xen: RCU: don't let a CPU with a callback go idle.

commit 38ad8151f563db9166e7b4133273ab27d8a3730e
Author:     Dario Faggioli <[hidden email]>
AuthorDate: Wed Aug 30 12:06:21 2017 +0100
Commit:     George Dunlap <[hidden email]>
CommitDate: Wed Aug 30 12:06:21 2017 +0100

    xen: RCU: don't let a CPU with a callback go idle.
    If a CPU has a callback queued, it must be ready to invoke
    it, as soon as all the other CPUs involved in the grace period
    has gone through a quiescent state.
    But if we let such CPU go idle, we can't really tell when (if!)
    it will realize that it is actually time to invoke the callback.
    To solve this problem, a CPU that has a callback queued (and has
    already gone through a quiescent state itself) will stay online,
    until the grace period ends, and the callback can be invoked.
    This is similar to what Linux does, and is the second and last
    step for fixing the overly long (or infinite!) grace periods.
    The problem, though, is that, within Linux, we have the tick,
    so, all that is necessary is to not stop the tick for the CPU
    (even if it has gone idle). In Xen, there's no tick, so we must
    avoid for the CPU to go idle entirely, and let it spin on
    rcu_pending(), consuming power and causing overhead.
    In this commit, we implement the above, using rcu_needs_cpu(),
    in a way similar to how it is used in Linux. This it correct,
    useful and not wasteful for CPUs that participate in grace
    period, but have not a callback queued. For the ones that
    has callbacks, an optimization that avoids having to spin is
    introduced in a subsequent change.
    Signed-off-by: Dario Faggioli <[hidden email]>
    Reviewed-by: Jan Beulich <[hidden email]>
    Acked-by: George Dunlap <[hidden email]>
 xen/include/xen/sched.h | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
index 5828a01..c116604 100644
--- a/xen/include/xen/sched.h
+++ b/xen/include/xen/sched.h
@@ -847,7 +847,8 @@ uint64_t get_cpu_idle_time(unsigned int cpu);
  * Used by idle loop to decide whether there is work to do:
- *  (1) Run softirqs; or (2) Play dead; or (3) Run tasklets.
+ *  (1) Deal with RCU; (2) or run softirqs; or (3) Play dead;
+ *  or (4) Run tasklets.
  * About (3), if a tasklet is enqueued, it will be scheduled
  * really really soon, and hence it's pointless to try to
@@ -855,7 +856,8 @@ uint64_t get_cpu_idle_time(unsigned int cpu);
  * the tasklet_work_to_do() helper).
 #define cpu_is_haltable(cpu)                    \
-    (!softirq_pending(cpu) &&                   \
+    (!rcu_needs_cpu(cpu) &&                     \
+     !softirq_pending(cpu) &&                   \
      cpu_online(cpu) &&                         \
      !per_cpu(tasklet_work_to_do, cpu))
generated by git-patchbot for /home/xen/git/xen.git#master

Xen-changelog mailing list
[hidden email]