cpu/hotplug: Make target state writeable
authorThomas Gleixner <tglx@linutronix.de>
Fri, 26 Feb 2016 18:43:32 +0000 (18:43 +0000)
committerThomas Gleixner <tglx@linutronix.de>
Tue, 1 Mar 2016 19:36:55 +0000 (20:36 +0100)
Make it possible to write a target state to the per cpu state file, so we can
switch between states.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-arch@vger.kernel.org
Cc: Rik van Riel <riel@redhat.com>
Cc: Rafael Wysocki <rafael.j.wysocki@intel.com>
Cc: "Srivatsa S. Bhat" <srivatsa@mit.edu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Sebastian Siewior <bigeasy@linutronix.de>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul Turner <pjt@google.com>
Link: http://lkml.kernel.org/r/20160226182341.022814799@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
kernel/cpu.c
lib/Kconfig.debug

index 1979b8927b8662215e3248bbf1c7918d1395586d..be9335da82f1999840dcf8ee2ac8d5efed387871 100644 (file)
@@ -48,12 +48,14 @@ static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state);
  * @teardown:  Teardown function of the step
  * @skip_onerr:        Do not invoke the functions on error rollback
  *             Will go away once the notifiers are gone
+ * @cant_stop: Bringup/teardown can't be stopped at this step
  */
 struct cpuhp_step {
        const char      *name;
        int             (*startup)(unsigned int cpu);
        int             (*teardown)(unsigned int cpu);
        bool            skip_onerr;
+       bool            cant_stop;
 };
 
 static DEFINE_MUTEX(cpuhp_state_mutex);
@@ -558,7 +560,7 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
        if (num_online_cpus() == 1)
                return -EBUSY;
 
-       if (!cpu_online(cpu))
+       if (!cpu_present(cpu))
                return -EINVAL;
 
        cpu_hotplug_begin();
@@ -683,16 +685,25 @@ static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
 
        cpu_hotplug_begin();
 
-       if (cpu_online(cpu) || !cpu_present(cpu)) {
+       if (!cpu_present(cpu)) {
                ret = -EINVAL;
                goto out;
        }
 
-       /* Let it fail before we try to bring the cpu up */
-       idle = idle_thread_get(cpu);
-       if (IS_ERR(idle)) {
-               ret = PTR_ERR(idle);
+       /*
+        * The caller of do_cpu_up might have raced with another
+        * caller. Ignore it for now.
+        */
+       if (st->state >= target)
                goto out;
+
+       if (st->state == CPUHP_OFFLINE) {
+               /* Let it fail before we try to bring the cpu up */
+               idle = idle_thread_get(cpu);
+               if (IS_ERR(idle)) {
+                       ret = PTR_ERR(idle);
+                       goto out;
+               }
        }
 
        cpuhp_tasks_frozen = tasks_frozen;
@@ -909,27 +920,32 @@ static struct cpuhp_step cpuhp_bp_states[] = {
                .name                   = "threads:create",
                .startup                = smpboot_create_threads,
                .teardown               = NULL,
+               .cant_stop              = true,
        },
        [CPUHP_NOTIFY_PREPARE] = {
                .name                   = "notify:prepare",
                .startup                = notify_prepare,
                .teardown               = notify_dead,
                .skip_onerr             = true,
+               .cant_stop              = true,
        },
        [CPUHP_BRINGUP_CPU] = {
                .name                   = "cpu:bringup",
                .startup                = bringup_cpu,
                .teardown               = NULL,
+               .cant_stop              = true,
        },
        [CPUHP_TEARDOWN_CPU] = {
                .name                   = "cpu:teardown",
                .startup                = NULL,
                .teardown               = takedown_cpu,
+               .cant_stop              = true,
        },
        [CPUHP_NOTIFY_ONLINE] = {
                .name                   = "notify:online",
                .startup                = notify_online,
                .teardown               = notify_down_prepare,
+               .cant_stop              = true,
        },
 #endif
        [CPUHP_ONLINE] = {
@@ -947,6 +963,7 @@ static struct cpuhp_step cpuhp_ap_states[] = {
                .startup                = notify_starting,
                .teardown               = notify_dying,
                .skip_onerr             = true,
+               .cant_stop              = true,
        },
 #endif
        [CPUHP_ONLINE] = {
@@ -979,6 +996,46 @@ static ssize_t show_cpuhp_state(struct device *dev,
 }
 static DEVICE_ATTR(state, 0444, show_cpuhp_state, NULL);
 
+static ssize_t write_cpuhp_target(struct device *dev,
+                                 struct device_attribute *attr,
+                                 const char *buf, size_t count)
+{
+       struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, dev->id);
+       struct cpuhp_step *sp;
+       int target, ret;
+
+       ret = kstrtoint(buf, 10, &target);
+       if (ret)
+               return ret;
+
+#ifdef CONFIG_CPU_HOTPLUG_STATE_CONTROL
+       if (target < CPUHP_OFFLINE || target > CPUHP_ONLINE)
+               return -EINVAL;
+#else
+       if (target != CPUHP_OFFLINE && target != CPUHP_ONLINE)
+               return -EINVAL;
+#endif
+
+       ret = lock_device_hotplug_sysfs();
+       if (ret)
+               return ret;
+
+       mutex_lock(&cpuhp_state_mutex);
+       sp = cpuhp_get_step(target);
+       ret = !sp->name || sp->cant_stop ? -EINVAL : 0;
+       mutex_unlock(&cpuhp_state_mutex);
+       if (ret)
+               return ret;
+
+       if (st->state < target)
+               ret = do_cpu_up(dev->id, target);
+       else
+               ret = do_cpu_down(dev->id, target);
+
+       unlock_device_hotplug();
+       return ret ? ret : count;
+}
+
 static ssize_t show_cpuhp_target(struct device *dev,
                                 struct device_attribute *attr, char *buf)
 {
@@ -986,7 +1043,7 @@ static ssize_t show_cpuhp_target(struct device *dev,
 
        return sprintf(buf, "%d\n", st->target);
 }
-static DEVICE_ATTR(target, 0444, show_cpuhp_target, NULL);
+static DEVICE_ATTR(target, 0644, show_cpuhp_target, write_cpuhp_target);
 
 static struct attribute *cpuhp_cpu_attrs[] = {
        &dev_attr_state.attr,
@@ -1007,7 +1064,7 @@ static ssize_t show_cpuhp_states(struct device *dev,
        int i;
 
        mutex_lock(&cpuhp_state_mutex);
-       for (i = 0; i <= CPUHP_ONLINE; i++) {
+       for (i = CPUHP_OFFLINE; i <= CPUHP_ONLINE; i++) {
                struct cpuhp_step *sp = cpuhp_get_step(i);
 
                if (sp->name) {
index 8bfd1aca7a3d01a9887700b3f90d1d5697a9dab4..f28f7fad452fe462b324761898a4a686581ef90c 100644 (file)
@@ -1442,6 +1442,19 @@ config DEBUG_BLOCK_EXT_DEVT
 
          Say N if you are unsure.
 
+config CPU_HOTPLUG_STATE_CONTROL
+       bool "Enable CPU hotplug state control"
+       depends on DEBUG_KERNEL
+       depends on HOTPLUG_CPU
+       default n
+       help
+         Allows to write steps between "offline" and "online" to the CPUs
+         sysfs target file so states can be stepped granular. This is a debug
+         option for now as the hotplug machinery cannot be stopped and
+         restarted at arbitrary points yet.
+
+         Say N if your are unsure.
+
 config NOTIFIER_ERROR_INJECTION
        tristate "Notifier error injection"
        depends on DEBUG_KERNEL