The routine disable_nonboot_cpus() shuts down cpus sequentially
using for_each_online_cpu(cpu) to call cpu_down() one cpu at
a time. cpu_down() calls __stop_machine() which stops all
the cpus while it disables one. Then it re-enables the remaining
cpus, only to do it all over again for the next cpu. The
result is that it takes 16 minutes on a 1024 cpu system to
disable 1023 cpus.
This patch changes disable_nonboot_cpus() to pass a bitmask
of cpus to cpu_down() and modifies cpu_down() to only call
__stop_machine() once.
On a 1024 cpu system this reduces the time it takes to disable
all but one cpu from 16 minutes down to 4 minutes.
Signed-off-by: Russ Anderson <r...@sgi.com>
To: Andrew Morton <a...@linux-foundation.org>
Cc: "H. Peter Anvin" <h...@zytor.com>
Cc: Ingo Molnar <mi...@redhat.com>
Cc: Robin Holt <h...@sgi.com>
Cc: Russ Anderson <r...@sgi.com>
---
kernel/cpu.c | 104 ++++++++++++++++++++++++++++++++++-------------------------
1 file changed, 61 insertions(+), 43 deletions(-)
Index: linux/kernel/cpu.c
===================================================================
--- linux.orig/kernel/cpu.c 2013-05-03 09:56:31.145508321 -0500
+++ linux/kernel/cpu.c 2013-05-03 17:01:20.652959400 -0500
@@ -241,13 +241,13 @@ static inline void check_for_tasks(int c
struct take_cpu_down_param {
unsigned long mod;
- void *hcpu;
};
/* Take this CPU down. */
static int __ref take_cpu_down(void *_param)
{
struct take_cpu_down_param *param = _param;
+ void *hcpu = (void *)(long)smp_processor_id();
int err;
/* Ensure this CPU doesn't handle any more interrupts. */
@@ -255,21 +255,21 @@ static int __ref take_cpu_down(void *_pa
if (err < 0)
return err;
- cpu_notify(CPU_DYING | param->mod, param->hcpu);
+ cpu_notify(CPU_DYING | param->mod, hcpu);
/* Park the stopper thread */
kthread_park(current);
return 0;
}
/* Requires cpu_add_remove_lock to be held */
-static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
+static int __ref _cpu_down(const cpumask_t *cpus_to_offline, int tasks_frozen)
{
- int err, nr_calls = 0;
+ int err = 0, cpu = 0, nr_calls = 0;
void *hcpu = (void *)(long)cpu;
+ cpumask_var_t cpus_offlined;
unsigned long mod = tasks_frozen ? CPU_TASKS_FROZEN : 0;
struct take_cpu_down_param tcd_param = {
.mod = mod,
- .hcpu = hcpu,
};
if (num_online_cpus() == 1)
@@ -278,46 +278,67 @@ static int __ref _cpu_down(unsigned int
if (!cpu_online(cpu))
return -EINVAL;
+ if (!alloc_cpumask_var(&cpus_offlined, GFP_KERNEL))
+ return -ENOMEM;
+
cpu_hotplug_begin();
+ cpumask_copy(cpus_offlined, cpus_to_offline);
- err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
- if (err) {
- nr_calls--;
- __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls, NULL);
- printk("%s: attempt to take down CPU %u failed\n",
+ for_each_cpu_mask(cpu, *cpus_to_offline) {
+ if (!cpu_online(cpu))
+ continue;
+ hcpu = (void *)(long)cpu;
+ err = __cpu_notify(CPU_DOWN_PREPARE | mod, hcpu, -1, &nr_calls);
+ if (err) {
+ nr_calls--;
+ __cpu_notify(CPU_DOWN_FAILED | mod, hcpu, nr_calls,
NULL);
+ pr_err("%s: attempt to take down CPU %u failed\n",
__func__, cpu);
- goto out_release;
+ goto out_release;
+ }
+ smpboot_park_threads(cpu);
}
- smpboot_park_threads(cpu);
- err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
+ err = __stop_machine(take_cpu_down, &tcd_param, cpus_to_offline);
if (err) {
/* CPU didn't die: tell everyone. Can't complain. */
- smpboot_unpark_threads(cpu);
- cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
+ for_each_cpu_mask(cpu, *cpus_to_offline) {
+ hcpu = (void *)(long)cpu;
+ smpboot_unpark_threads(cpu);
+ cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
+ }
goto out_release;
}
- BUG_ON(cpu_online(cpu));
/*
* The migration_call() CPU_DYING callback will have removed all
* runnable tasks from the cpu, there's only the idle task left now
* that the migration thread is done doing the stop_machine thing.
- *
- * Wait for the stop thread to go away.
*/
- while (!idle_cpu(cpu))
- cpu_relax();
-
- /* This actually kills the CPU. */
- __cpu_die(cpu);
+ for_each_cpu_mask(cpu, *cpus_offlined) {
+ BUG_ON(cpu_online(cpu));
- /* CPU is completely dead: tell everyone. Too late to complain. */
- cpu_notify_nofail(CPU_DEAD | mod, hcpu);
-
- check_for_tasks(cpu);
+ /*
+ * Wait for the stop thread to go away.
+ */
+ while (!idle_cpu(cpu))
+ cpu_relax();
+
+ /*
+ * This actually kills the CPU.
+ */
+ __cpu_die(cpu);
+
+ /*
+ * CPU is completely dead: tell everyone. Too late to complain.
+ */
+ hcpu = (void *)(long)cpu;
+ cpu_notify_nofail(CPU_DEAD | mod, hcpu);
+ check_for_tasks(cpu);
+ }
out_release:
+ free_cpumask_var(cpus_offlined);
cpu_hotplug_done();
if (!err)
cpu_notify_nofail(CPU_POST_DEAD | mod, hcpu);
@@ -327,6 +348,7 @@ out_release:
int __ref cpu_down(unsigned int cpu)
{
int err;
+ cpumask_var_t cpumask;
cpu_maps_update_begin();
@@ -335,7 +357,11 @@ int __ref cpu_down(unsigned int cpu)
goto out;
}
- err = _cpu_down(cpu, 0);
+ if (!alloc_cpumask_var(&cpumask, GFP_KERNEL))
+ return -ENOMEM;
+ cpumask_set_cpu(cpu, cpumask);
+ err = _cpu_down(cpumask, 0);
+ free_cpumask_var(cpumask);
out:
cpu_maps_update_done();
@@ -459,7 +485,7 @@ static cpumask_var_t frozen_cpus;
int disable_nonboot_cpus(void)
{
- int cpu, first_cpu, error = 0;
+ int first_cpu, error = 0;
cpu_maps_update_begin();
first_cpu = cpumask_first(cpu_online_mask);
@@ -467,21 +493,13 @@ int disable_nonboot_cpus(void)
* We take down all of the non-boot CPUs in one shot to avoid races
* with the userspace trying to use the CPU hotplug at the same time
*/
- cpumask_clear(frozen_cpus);
-
printk("Disabling non-boot CPUs ...\n");
- for_each_online_cpu(cpu) {
- if (cpu == first_cpu)
- continue;
- error = _cpu_down(cpu, 1);
- if (!error)
- cpumask_set_cpu(cpu, frozen_cpus);
- else {
- printk(KERN_ERR "Error taking CPU%d down: %d\n",
- cpu, error);
- break;
- }
- }
+ cpumask_copy(frozen_cpus, cpu_online_mask);
+ cpumask_clear_cpu(first_cpu, frozen_cpus); /* all but one cpu*/
+
+ error = _cpu_down(frozen_cpus, 1);
+ if (error)
+ pr_err("Error %d stopping cpus\n", error);
if (!error) {
BUG_ON(num_online_cpus() > 1);
--
Russ Anderson, OS RAS/Partitioning Project Lead
SGI - Silicon Graphics Inc r...@sgi.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
