]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
stop_machine: use workqueues instead of kernel threads
authorHeiko Carstens <heiko.carstens@de.ibm.com>
Mon, 13 Oct 2008 21:50:10 +0000 (23:50 +0200)
committerRusty Russell <rusty@rustcorp.com.au>
Tue, 21 Oct 2008 23:00:26 +0000 (10:00 +1100)
Convert stop_machine to a workqueue based approach. Instead of using kernel
threads for stop_machine we now use a an rt workqueue to synchronize all
cpus.
This has the advantage that all needed per cpu threads are already created
when stop_machine gets called. And therefore a call to stop_machine won't
fail anymore. This is needed for s390 which needs a mechanism to synchronize
all cpus without allocating any memory.
As Rusty pointed out free_module() needs a non-failing stop_machine interface
as well.

As a side effect the stop_machine code gets simplified.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
kernel/stop_machine.c

index af3c7cea258b00d57762913a8986978495c2a81e..0e688c6a1a63d22d6b034de5af7b2891c1df2aad 100644 (file)
@@ -37,9 +37,13 @@ struct stop_machine_data {
 /* Like num_online_cpus(), but hotplug cpu uses us, so we need this. */
 static unsigned int num_threads;
 static atomic_t thread_ack;
-static struct completion finished;
 static DEFINE_MUTEX(lock);
 
+static struct workqueue_struct *stop_machine_wq;
+static struct stop_machine_data active, idle;
+static const cpumask_t *active_cpus;
+static void *stop_machine_work;
+
 static void set_state(enum stopmachine_state newstate)
 {
        /* Reset ack counter. */
@@ -51,21 +55,25 @@ static void set_state(enum stopmachine_state newstate)
 /* Last one to ack a state moves to the next state. */
 static void ack_state(void)
 {
-       if (atomic_dec_and_test(&thread_ack)) {
-               /* If we're the last one to ack the EXIT, we're finished. */
-               if (state == STOPMACHINE_EXIT)
-                       complete(&finished);
-               else
-                       set_state(state + 1);
-       }
+       if (atomic_dec_and_test(&thread_ack))
+               set_state(state + 1);
 }
 
-/* This is the actual thread which stops the CPU.  It exits by itself rather
- * than waiting for kthread_stop(), because it's easier for hotplug CPU. */
-static int stop_cpu(struct stop_machine_data *smdata)
+/* This is the actual function which stops the CPU. It runs
+ * in the context of a dedicated stopmachine workqueue. */
+static void stop_cpu(struct work_struct *unused)
 {
        enum stopmachine_state curstate = STOPMACHINE_NONE;
-
+       struct stop_machine_data *smdata = &idle;
+       int cpu = smp_processor_id();
+
+       if (!active_cpus) {
+               if (cpu == first_cpu(cpu_online_map))
+                       smdata = &active;
+       } else {
+               if (cpu_isset(cpu, *active_cpus))
+                       smdata = &active;
+       }
        /* Simple state machine */
        do {
                /* Chill out and ensure we re-read stopmachine_state. */
@@ -90,7 +98,6 @@ static int stop_cpu(struct stop_machine_data *smdata)
        } while (curstate != STOPMACHINE_EXIT);
 
        local_irq_enable();
-       do_exit(0);
 }
 
 /* Callback for CPUs which aren't supposed to do anything. */
@@ -101,78 +108,34 @@ static int chill(void *unused)
 
 int __stop_machine(int (*fn)(void *), void *data, const cpumask_t *cpus)
 {
-       int i, err;
-       struct stop_machine_data active, idle;
-       struct task_struct **threads;
+       struct work_struct *sm_work;
+       int i;
 
+       /* Set up initial state. */
+       mutex_lock(&lock);
+       num_threads = num_online_cpus();
+       active_cpus = cpus;
        active.fn = fn;
        active.data = data;
        active.fnret = 0;
        idle.fn = chill;
        idle.data = NULL;
 
-       /* This could be too big for stack on large machines. */
-       threads = kcalloc(NR_CPUS, sizeof(threads[0]), GFP_KERNEL);
-       if (!threads)
-               return -ENOMEM;
-
-       /* Set up initial state. */
-       mutex_lock(&lock);
-       init_completion(&finished);
-       num_threads = num_online_cpus();
        set_state(STOPMACHINE_PREPARE);
 
-       for_each_online_cpu(i) {
-               struct stop_machine_data *smdata = &idle;
-               struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
-
-               if (!cpus) {
-                       if (i == first_cpu(cpu_online_map))
-                               smdata = &active;
-               } else {
-                       if (cpu_isset(i, *cpus))
-                               smdata = &active;
-               }
-
-               threads[i] = kthread_create((void *)stop_cpu, smdata, "kstop%u",
-                                           i);
-               if (IS_ERR(threads[i])) {
-                       err = PTR_ERR(threads[i]);
-                       threads[i] = NULL;
-                       goto kill_threads;
-               }
-
-               /* Place it onto correct cpu. */
-               kthread_bind(threads[i], i);
-
-               /* Make it highest prio. */
-               if (sched_setscheduler_nocheck(threads[i], SCHED_FIFO, &param))
-                       BUG();
-       }
-
-       /* We've created all the threads.  Wake them all: hold this CPU so one
+       /* Schedule the stop_cpu work on all cpus: hold this CPU so one
         * doesn't hit this CPU until we're ready. */
        get_cpu();
-       for_each_online_cpu(i)
-               wake_up_process(threads[i]);
-
+       for_each_online_cpu(i) {
+               sm_work = percpu_ptr(stop_machine_work, i);
+               INIT_WORK(sm_work, stop_cpu);
+               queue_work_on(i, stop_machine_wq, sm_work);
+       }
        /* This will release the thread on our CPU. */
        put_cpu();
-       wait_for_completion(&finished);
+       flush_workqueue(stop_machine_wq);
        mutex_unlock(&lock);
-
-       kfree(threads);
-
        return active.fnret;
-
-kill_threads:
-       for_each_online_cpu(i)
-               if (threads[i])
-                       kthread_stop(threads[i]);
-       mutex_unlock(&lock);
-
-       kfree(threads);
-       return err;
 }
 
 int stop_machine(int (*fn)(void *), void *data, const cpumask_t *cpus)
@@ -187,3 +150,11 @@ int stop_machine(int (*fn)(void *), void *data, const cpumask_t *cpus)
        return ret;
 }
 EXPORT_SYMBOL_GPL(stop_machine);
+
+static int __init stop_machine_init(void)
+{
+       stop_machine_wq = create_rt_workqueue("kstop");
+       stop_machine_work = alloc_percpu(struct work_struct);
+       return 0;
+}
+early_initcall(stop_machine_init);