cpu/hotplug: Convert to a state machine for the control processor
authorThomas Gleixner <tglx@linutronix.de>
Fri, 26 Feb 2016 18:43:28 +0000 (18:43 +0000)
committerThomas Gleixner <tglx@linutronix.de>
Tue, 1 Mar 2016 19:36:54 +0000 (20:36 +0100)
Move the split out steps into a callback array and let the cpu_up/down
code iterate through the array functions. For now most of the
callbacks are asymmetric to resemble the current hotplug maze.

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/20160226182340.671816690@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
include/linux/cpu.h
include/linux/cpuhotplug.h [new file with mode: 0644]
init/main.c
kernel/cpu.c

index f2fb54938ee6a19cf1a52e88297b4b87d67b7e9f..78989f20420f5fac31b2a33c6c88ea94f071235c 100644 (file)
@@ -16,6 +16,7 @@
 #include <linux/node.h>
 #include <linux/compiler.h>
 #include <linux/cpumask.h>
+#include <linux/cpuhotplug.h>
 
 struct device;
 struct device_node;
@@ -27,6 +28,9 @@ struct cpu {
        struct device dev;
 };
 
+extern void boot_cpu_init(void);
+extern void boot_cpu_state_init(void);
+
 extern int register_cpu(struct cpu *cpu, int num);
 extern struct device *get_cpu_device(unsigned cpu);
 extern bool cpu_is_hotpluggable(unsigned cpu);
@@ -267,11 +271,6 @@ static inline int disable_nonboot_cpus(void) { return 0; }
 static inline void enable_nonboot_cpus(void) {}
 #endif /* !CONFIG_PM_SLEEP_SMP */
 
-enum cpuhp_state {
-       CPUHP_OFFLINE,
-       CPUHP_ONLINE,
-};
-
 void cpu_startup_entry(enum cpuhp_state state);
 
 void cpu_idle_poll_ctrl(bool enable);
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
new file mode 100644 (file)
index 0000000..d55c9e6
--- /dev/null
@@ -0,0 +1,13 @@
+#ifndef __CPUHOTPLUG_H
+#define __CPUHOTPLUG_H
+
+enum cpuhp_state {
+       CPUHP_OFFLINE,
+       CPUHP_CREATE_THREADS,
+       CPUHP_NOTIFY_PREPARE,
+       CPUHP_BRINGUP_CPU,
+       CPUHP_NOTIFY_ONLINE,
+       CPUHP_ONLINE,
+};
+
+#endif
index 58c9e374704bb20cfff41fa92afcf7cafe498c32..c2ea72362ee3ad38351b2b66d9d280db488652b3 100644 (file)
@@ -452,20 +452,6 @@ void __init parse_early_param(void)
        done = 1;
 }
 
-/*
- *     Activate the first processor.
- */
-
-static void __init boot_cpu_init(void)
-{
-       int cpu = smp_processor_id();
-       /* Mark the boot cpu "present", "online" etc for SMP and UP case */
-       set_cpu_online(cpu, true);
-       set_cpu_active(cpu, true);
-       set_cpu_present(cpu, true);
-       set_cpu_possible(cpu, true);
-}
-
 void __init __weak smp_setup_processor_id(void)
 {
 }
@@ -530,6 +516,7 @@ asmlinkage __visible void __init start_kernel(void)
        setup_command_line(command_line);
        setup_nr_cpu_ids();
        setup_per_cpu_areas();
+       boot_cpu_state_init();
        smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
 
        build_all_zonelists(NULL, NULL);
index 0b5d2596f3ec6c55cc345b11463acf1d1e430438..301851974b8ddcebe08612c11e3767c208bddf53 100644 (file)
 #include <linux/lockdep.h>
 #include <linux/tick.h>
 #include <linux/irq.h>
+
 #include <trace/events/power.h>
+#define CREATE_TRACE_POINTS
+#include <trace/events/cpuhp.h>
 
 #include "smpboot.h"
 
+/**
+ * cpuhp_cpu_state - Per cpu hotplug state storage
+ * @state:     The current cpu state
+ * @target:    The target state
+ */
+struct cpuhp_cpu_state {
+       enum cpuhp_state        state;
+       enum cpuhp_state        target;
+};
+
+static DEFINE_PER_CPU(struct cpuhp_cpu_state, cpuhp_state);
+
+/**
+ * cpuhp_step - Hotplug state machine step
+ * @name:      Name of the step
+ * @startup:   Startup function of the step
+ * @teardown:  Teardown function of the step
+ * @skip_onerr:        Do not invoke the functions on error rollback
+ *             Will go away once the notifiers are gone
+ */
+struct cpuhp_step {
+       const char      *name;
+       int             (*startup)(unsigned int cpu);
+       int             (*teardown)(unsigned int cpu);
+       bool            skip_onerr;
+};
+
+static struct cpuhp_step cpuhp_bp_states[];
+
+/**
+ * cpuhp_invoke_callback _ Invoke the callbacks for a given state
+ * @cpu:       The cpu for which the callback should be invoked
+ * @step:      The step in the state machine
+ * @cb:                The callback function to invoke
+ *
+ * Called from cpu hotplug and from the state register machinery
+ */
+static int cpuhp_invoke_callback(unsigned int cpu, enum cpuhp_state step,
+                                int (*cb)(unsigned int))
+{
+       struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
+       int ret = 0;
+
+       if (cb) {
+               trace_cpuhp_enter(cpu, st->target, step, cb);
+               ret = cb(cpu);
+               trace_cpuhp_exit(cpu, st->state, step, ret);
+       }
+       return ret;
+}
+
 #ifdef CONFIG_SMP
 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
 static DEFINE_MUTEX(cpu_add_remove_lock);
@@ -454,10 +508,29 @@ static int notify_dead(unsigned int cpu)
        return 0;
 }
 
+#else
+#define notify_down_prepare    NULL
+#define takedown_cpu           NULL
+#define notify_dead            NULL
+#endif
+
+#ifdef CONFIG_HOTPLUG_CPU
+static void undo_cpu_down(unsigned int cpu, struct cpuhp_cpu_state *st)
+{
+       for (st->state++; st->state < st->target; st->state++) {
+               struct cpuhp_step *step = cpuhp_bp_states + st->state;
+
+               if (!step->skip_onerr)
+                       cpuhp_invoke_callback(cpu, st->state, step->startup);
+       }
+}
+
 /* Requires cpu_add_remove_lock to be held */
 static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 {
-       int err;
+       struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
+       int prev_state, ret = 0;
+       bool hasdied = false;
 
        if (num_online_cpus() == 1)
                return -EBUSY;
@@ -469,20 +542,25 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
 
        cpuhp_tasks_frozen = tasks_frozen;
 
-       err = notify_down_prepare(cpu);
-       if (err)
-               goto out_release;
-       err = takedown_cpu(cpu);
-       if (err)
-               goto out_release;
+       prev_state = st->state;
+       st->target = CPUHP_OFFLINE;
+       for (; st->state > st->target; st->state--) {
+               struct cpuhp_step *step = cpuhp_bp_states + st->state;
 
-       notify_dead(cpu);
+               ret = cpuhp_invoke_callback(cpu, st->state, step->teardown);
+               if (ret) {
+                       st->target = prev_state;
+                       undo_cpu_down(cpu, st);
+                       break;
+               }
+       }
+       hasdied = prev_state != st->state && st->state == CPUHP_OFFLINE;
 
-out_release:
        cpu_hotplug_done();
-       if (!err)
+       /* This post dead nonsense must die */
+       if (!ret && hasdied)
                cpu_notify_nofail(CPU_POST_DEAD, cpu);
-       return err;
+       return ret;
 }
 
 int cpu_down(unsigned int cpu)
@@ -537,11 +615,22 @@ void smpboot_thread_init(void)
        register_cpu_notifier(&smpboot_thread_notifier);
 }
 
+static void undo_cpu_up(unsigned int cpu, struct cpuhp_cpu_state *st)
+{
+       for (st->state--; st->state > st->target; st->state--) {
+               struct cpuhp_step *step = cpuhp_bp_states + st->state;
+
+               if (!step->skip_onerr)
+                       cpuhp_invoke_callback(cpu, st->state, step->teardown);
+       }
+}
+
 /* Requires cpu_add_remove_lock to be held */
 static int _cpu_up(unsigned int cpu, int tasks_frozen)
 {
+       struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
        struct task_struct *idle;
-       int ret;
+       int prev_state, ret = 0;
 
        cpu_hotplug_begin();
 
@@ -550,6 +639,7 @@ static int _cpu_up(unsigned int cpu, int tasks_frozen)
                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);
@@ -558,22 +648,22 @@ static int _cpu_up(unsigned int cpu, int tasks_frozen)
 
        cpuhp_tasks_frozen = tasks_frozen;
 
-       ret = smpboot_create_threads(cpu);
-       if (ret)
-               goto out;
-
-       ret = notify_prepare(cpu);
-       if (ret)
-               goto out;
-
-       ret = bringup_cpu(cpu);
-       if (ret)
-               goto out;
-
-       notify_online(cpu);
+       prev_state = st->state;
+       st->target = CPUHP_ONLINE;
+       while (st->state < st->target) {
+               struct cpuhp_step *step;
+
+               st->state++;
+               step = cpuhp_bp_states + st->state;
+               ret = cpuhp_invoke_callback(cpu, st->state, step->startup);
+               if (ret) {
+                       st->target = prev_state;
+                       undo_cpu_up(cpu, st);
+                       break;
+               }
+       }
 out:
        cpu_hotplug_done();
-
        return ret;
 }
 
@@ -767,6 +857,44 @@ void notify_cpu_starting(unsigned int cpu)
 
 #endif /* CONFIG_SMP */
 
+/* Boot processor state steps */
+static struct cpuhp_step cpuhp_bp_states[] = {
+       [CPUHP_OFFLINE] = {
+               .name                   = "offline",
+               .startup                = NULL,
+               .teardown               = NULL,
+       },
+#ifdef CONFIG_SMP
+       [CPUHP_CREATE_THREADS]= {
+               .name                   = "threads:create",
+               .startup                = smpboot_create_threads,
+               .teardown               = NULL,
+       },
+       [CPUHP_NOTIFY_PREPARE] = {
+               .name                   = "notify:prepare",
+               .startup                = notify_prepare,
+               .teardown               = notify_dead,
+               .skip_onerr             = true,
+       },
+       [CPUHP_BRINGUP_CPU] = {
+               .name                   = "cpu:bringup",
+               .startup                = bringup_cpu,
+               .teardown               = takedown_cpu,
+               .skip_onerr             = true,
+       },
+       [CPUHP_NOTIFY_ONLINE] = {
+               .name                   = "notify:online",
+               .startup                = notify_online,
+               .teardown               = notify_down_prepare,
+       },
+#endif
+       [CPUHP_ONLINE] = {
+               .name                   = "online",
+               .startup                = NULL,
+               .teardown               = NULL,
+       },
+};
+
 /*
  * cpu_bit_bitmap[] is a special, "compressed" data structure that
  * represents all NR_CPUS bits binary values of 1<<nr.
@@ -826,3 +954,25 @@ void init_cpu_online(const struct cpumask *src)
 {
        cpumask_copy(&__cpu_online_mask, src);
 }
+
+/*
+ * Activate the first processor.
+ */
+void __init boot_cpu_init(void)
+{
+       int cpu = smp_processor_id();
+
+       /* Mark the boot cpu "present", "online" etc for SMP and UP case */
+       set_cpu_online(cpu, true);
+       set_cpu_active(cpu, true);
+       set_cpu_present(cpu, true);
+       set_cpu_possible(cpu, true);
+}
+
+/*
+ * Must be called _AFTER_ setting up the per_cpu areas
+ */
+void __init boot_cpu_state_init(void)
+{
+       per_cpu_ptr(&cpuhp_state, smp_processor_id())->state = CPUHP_ONLINE;
+}