[PATCH v2] powerpc/smp: poll cpu_callin_map more aggressively in __cpu_up()

[Nathan Lynch]

It is not necessary to delay or sleep between polls of cpu_callin_map when
waiting for a kicked CPU to come up. We can use spin_until_cond(),
combining the boot and hotplug paths while preserving the intended timeout.

Without the msleep(1) in the hotplug path, the time it takes to online a
CPU on a P9 PowerVM LPAR goes from roughly 30ms to 4ms or less when
exercised via thaw_secondary_cpus().

Signed-off-by: Nathan Lynch <nathanl at linux.ibm.com>
---

Notes:
    Changes since v1:
    * Do not poll indefinitely; restore the original 5sec timeout
    
    "Benchmark" method, on a LPAR with 24 CPUs:
    $ echo processors > /sys/power/pm_test
    $ echo mem > /sys/power/state
    $ dmesg --reltime | grep -A23 'Enabling non-boot CPUs'
    
    Before:
    [  +5.000003] Enabling non-boot CPUs ...
    [  +0.047537] CPU1 is up
    [  +0.030177] CPU2 is up
    [  +0.030082] CPU3 is up
    [  +0.030146] CPU4 is up
    [  +0.030114] CPU5 is up
    [  +0.030108] CPU6 is up
    [  +0.030134] CPU7 is up
    [  +0.040272] CPU8 is up
    [  +0.030352] CPU9 is up
    [  +0.030169] CPU10 is up
    [  +0.040224] CPU11 is up
    [  +0.030186] CPU12 is up
    [  +0.030173] CPU13 is up
    [  +0.030181] CPU14 is up
    [  +0.030179] CPU15 is up
    [  +0.030236] CPU16 is up
    [  +0.030012] CPU17 is up
    [  +0.030188] CPU18 is up
    [  +0.030233] CPU19 is up
    [  +0.030182] CPU20 is up
    [  +0.030160] CPU21 is up
    [  +0.030226] CPU22 is up
    [  +0.030125] CPU23 is up
    
    After:
    [  +5.000004] Enabling non-boot CPUs ...
    [  +0.000956] CPU1 is up
    [  +0.000568] CPU2 is up
    [  +0.000682] CPU3 is up
    [  +0.000836] CPU4 is up
    [  +0.000914] CPU5 is up
    [  +0.001054] CPU6 is up
    [  +0.001184] CPU7 is up
    [  +0.001533] CPU8 is up
    [  +0.001779] CPU9 is up
    [  +0.001935] CPU10 is up
    [  +0.002151] CPU11 is up
    [  +0.002358] CPU12 is up
    [  +0.002597] CPU13 is up
    [  +0.002713] CPU14 is up
    [  +0.002891] CPU15 is up
    [  +0.003209] CPU16 is up
    [  +0.003149] CPU17 is up
    [  +0.003337] CPU18 is up
    [  +0.003552] CPU19 is up
    [  +0.003765] CPU20 is up
    [  +0.003922] CPU21 is up
    [  +0.004104] CPU22 is up
    [  +0.004266] CPU23 is up

 arch/powerpc/kernel/smp.c | 23 ++++++-----------------
 1 file changed, 6 insertions(+), 17 deletions(-)

diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index bcefab484ea6..c8431074d590 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -23,6 +23,7 @@
 #include <linux/delay.h>
 #include <linux/init.h>
 #include <linux/spinlock.h>
+#include <linux/jiffies.h>
 #include <linux/cache.h>
 #include <linux/err.h>
 #include <linux/device.h>
@@ -1268,7 +1269,8 @@ static void cpu_idle_thread_init(unsigned int cpu, struct task_struct *idle)
 
 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 {
-	int rc, c;
+	unsigned long deadline;
+	int rc;
 
 	/*
 	 * Don't allow secondary threads to come online if inhibited
@@ -1313,23 +1315,10 @@ int __cpu_up(unsigned int cpu, struct task_struct *tidle)
 	}
 
 	/*
-	 * wait to see if the cpu made a callin (is actually up).
-	 * use this value that I found through experimentation.
-	 * -- Cort
+	 * Give the remote CPU five seconds to enter the kernel.
 	 */
-	if (system_state < SYSTEM_RUNNING)
-		for (c = 50000; c && !cpu_callin_map[cpu]; c--)
-			udelay(100);
-#ifdef CONFIG_HOTPLUG_CPU
-	else
-		/*
-		 * CPUs can take much longer to come up in the
-		 * hotplug case.  Wait five seconds.
-		 */
-		for (c = 5000; c && !cpu_callin_map[cpu]; c--)
-			msleep(1);
-#endif
-
+	deadline = jiffies + msecs_to_jiffies(5000);
+	spin_until_cond(cpu_callin_map[cpu] != 0 || time_after(jiffies, deadline));
 	if (!cpu_callin_map[cpu]) {
 		printk(KERN_ERR "Processor %u is stuck.\n", cpu);
 		return -ENOENT;
-- 
2.36.1