ARC: [SMP] optimize IPI send and receive

* Don't send an IPI if receiver already has a pending IPI.
  Atomically piggyback the new msg with pending msg.

* IPI receiver looping on xchg() not required

References: https://lkml.org/lkml/2013/11/25/232
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>

diff --git a/arch/arc/kernel/smp.c b/arch/arc/kernel/smp.c
index c00c612e8dd3946ec35a81389b9b52e5fecba560..40859e5619f914115ba010641094aa3a3ca5a8b3 100644 (file)
--- a/arch/arc/kernel/smp.c
+++ b/arch/arc/kernel/smp.c
@@ -215,16 +215,31 @@ static DEFINE_PER_CPU(unsigned long, ipi_data);
 static void ipi_send_msg_one(int cpu, enum ipi_msg_type msg)
 {
        unsigned long __percpu *ipi_data_ptr = per_cpu_ptr(&ipi_data, cpu);
+       unsigned long old, new;
        unsigned long flags;
 
        pr_debug("%d Sending msg [%d] to %d\n", smp_processor_id(), msg, cpu);
 
        local_irq_save(flags);
 
-       set_bit(msg, ipi_data_ptr);
+       /*
+        * Atomically write new msg bit (in case others are writing too),
+        * and read back old value
+        */
+       do {
+               new = old = *ipi_data_ptr;
+               new |= 1U << msg;
+       } while (cmpxchg(ipi_data_ptr, old, new) != old);
 
-       /* Call the platform specific cross-CPU call function  */
-       if (plat_smp_ops.ipi_send)
+       /*
+        * Call the platform specific IPI kick function, but avoid if possible:
+        * Only do so if there's no pending msg from other concurrent sender(s).
+        * Otherwise, recevier will see this msg as well when it takes the
+        * IPI corresponding to that msg. This is true, even if it is already in
+        * IPI handler, because !@old means it has not yet dequeued the msg(s)
+        * so @new msg can be a free-loader
+        */
+       if (plat_smp_ops.ipi_send && !old)
                plat_smp_ops.ipi_send(cpu);
 
        local_irq_restore(flags);
@@ -269,31 +284,23 @@ static void ipi_cpu_stop(void)
        machine_halt();
 }
 
-static inline void __do_IPI(unsigned long pending)
+static inline void __do_IPI(unsigned long msg)
 {
-       while (pending) {
-
-               unsigned long msg = __ffs(pending);
-
-               switch (msg) {
-               case IPI_RESCHEDULE:
-                       scheduler_ipi();
-                       break;
+       switch (msg) {
+       case IPI_RESCHEDULE:
+               scheduler_ipi();
+               break;
 
-               case IPI_CALL_FUNC:
-                       generic_smp_call_function_interrupt();
-                       break;
-
-               case IPI_CPU_STOP:
-                       ipi_cpu_stop();
-                       break;
-
-               default:
-                       pr_warn("IPI missing msg\n");
+       case IPI_CALL_FUNC:
+               generic_smp_call_function_interrupt();
+               break;
 
-               }
+       case IPI_CPU_STOP:
+               ipi_cpu_stop();
+               break;
 
-               pending &= ~(1U << msg);
+       default:
+               pr_warn("IPI with unexpected msg %ld\n", msg);
        }
 }
 
@@ -312,11 +319,16 @@ irqreturn_t do_IPI(int irq, void *dev_id)
                plat_smp_ops.ipi_clear(irq);
 
        /*
-        * XXX: is this loop really needed
-        * And do we need to move ipi_clean inside
+        * "dequeue" the msg corresponding to this IPI (and possibly other
+        * piggybacked msg from elided IPIs: see ipi_send_msg_one() above)
         */
-       while ((pending = xchg(this_cpu_ptr(&ipi_data), 0)) != 0)
-               __do_IPI(pending);
+       pending = xchg(this_cpu_ptr(&ipi_data), 0);
+
+       do {
+               unsigned long msg = __ffs(pending);
+               __do_IPI(msg);
+               pending &= ~(1U << msg);
+       } while (pending);
 
        return IRQ_HANDLED;
 }