irq_work: Add generic hardirq context callbacks

Provide a mechanism that allows running code in IRQ context. It is
most useful for NMI code that needs to interact with the rest of the
system -- like wakeup a task to drain buffers.

Perf currently has such a mechanism, so extract that and provide it as
a generic feature, independent of perf so that others may also
benefit.

The IRQ context callback is generated through self-IPIs where
possible, or on architectures like powerpc the decrementer (the
built-in timer facility) is set to generate an interrupt immediately.

Architectures that don't have anything like this get to do with a
callback from the timer tick. These architectures can call
irq_work_run() at the tail of any IRQ handlers that might enqueue such
work (like the perf IRQ handler) to avoid undue latencies in
processing the work.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Kyle McMartin <kyle@mcmartin.ca>
Acked-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
[ various fixes ]
Signed-off-by: Huang Ying <ying.huang@intel.com>
LKML-Reference: <1287036094.7768.291.camel@yhuang-dev>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

4 files changed, 176 insertions, 101 deletions

diff --git a/kernel/Makefile b/kernel/Makefile
index d52b473c99a1..4d9bf5f8531f 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -23,6 +23,7 @@ CFLAGS_REMOVE_rtmutex-debug.o = -pg
 CFLAGS_REMOVE_cgroup-debug.o = -pg
 CFLAGS_REMOVE_sched_clock.o = -pg
 CFLAGS_REMOVE_perf_event.o = -pg
+CFLAGS_REMOVE_irq_work.o = -pg
 endif
 
 obj-$(CONFIG_FREEZER) += freezer.o
@@ -100,6 +101,7 @@ obj-$(CONFIG_TRACING) += trace/
 obj-$(CONFIG_X86_DS) += trace/
 obj-$(CONFIG_RING_BUFFER) += trace/
 obj-$(CONFIG_SMP) += sched_cpupri.o
+obj-$(CONFIG_IRQ_WORK) += irq_work.o
 obj-$(CONFIG_PERF_EVENTS) += perf_event.o
 obj-$(CONFIG_HAVE_HW_BREAKPOINT) += hw_breakpoint.o
 obj-$(CONFIG_USER_RETURN_NOTIFIER) += user-return-notifier.o
diff --git a/kernel/irq_work.c b/kernel/irq_work.c
new file mode 100644
index 000000000000..f16763ff8481
--- /dev/null
+++ b/kernel/irq_work.c
@@ -0,0 +1,164 @@
+/*
+ * Copyright (C) 2010 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
+ *
+ * Provides a framework for enqueueing and running callbacks from hardirq
+ * context. The enqueueing is NMI-safe.
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/irq_work.h>
+#include <linux/hardirq.h>
+
+/*
+ * An entry can be in one of four states:
+ *
+ * free      NULL, 0 -> {claimed}       : free to be used
+ * claimed   NULL, 3 -> {pending}       : claimed to be enqueued
+ * pending   next, 3 -> {busy}          : queued, pending callback
+ * busy      NULL, 2 -> {free, claimed} : callback in progress, can be claimed
+ *
+ * We use the lower two bits of the next pointer to keep PENDING and BUSY
+ * flags.
+ */
+
+#define IRQ_WORK_PENDING        1UL
+#define IRQ_WORK_BUSY           2UL
+#define IRQ_WORK_FLAGS          3UL
+
+static inline bool irq_work_is_set(struct irq_work *entry, int flags)
+{
+        return (unsigned long)entry->next & flags;
+}
+
+static inline struct irq_work *irq_work_next(struct irq_work *entry)
+{
+        unsigned long next = (unsigned long)entry->next;
+        next &= ~IRQ_WORK_FLAGS;
+        return (struct irq_work *)next;
+}
+
+static inline struct irq_work *next_flags(struct irq_work *entry, int flags)
+{
+        unsigned long next = (unsigned long)entry;
+        next |= flags;
+        return (struct irq_work *)next;
+}
+
+static DEFINE_PER_CPU(struct irq_work *, irq_work_list);
+
+/*
+ * Claim the entry so that no one else will poke at it.
+ */
+static bool irq_work_claim(struct irq_work *entry)
+{
+        struct irq_work *next, *nflags;
+
+        do {
+                next = entry->next;
+                if ((unsigned long)next & IRQ_WORK_PENDING)
+                        return false;
+                nflags = next_flags(next, IRQ_WORK_FLAGS);
+        } while (cmpxchg(&entry->next, next, nflags) != next);
+
+        return true;
+}
+
+
+void __weak arch_irq_work_raise(void)
+{
+        /*
+         * Lame architectures will get the timer tick callback
+         */
+}
+
+/*
+ * Queue the entry and raise the IPI if needed.
+ */
+static void __irq_work_queue(struct irq_work *entry)
+{
+        struct irq_work **head, *next;
+
+        head = &get_cpu_var(irq_work_list);
+
+        do {
+                next = *head;
+                /* Can assign non-atomic because we keep the flags set. */
+                entry->next = next_flags(next, IRQ_WORK_FLAGS);
+        } while (cmpxchg(head, next, entry) != next);
+
+        /* The list was empty, raise self-interrupt to start processing. */
+        if (!irq_work_next(entry))
+                arch_irq_work_raise();
+
+        put_cpu_var(irq_work_list);
+}
+
+/*
+ * Enqueue the irq_work @entry, returns true on success, failure when the
+ * @entry was already enqueued by someone else.
+ *
+ * Can be re-enqueued while the callback is still in progress.
+ */
+bool irq_work_queue(struct irq_work *entry)
+{
+        if (!irq_work_claim(entry)) {
+                /*
+                 * Already enqueued, can't do!
+                 */
+                return false;
+        }
+
+        __irq_work_queue(entry);
+        return true;
+}
+EXPORT_SYMBOL_GPL(irq_work_queue);
+
+/*
+ * Run the irq_work entries on this cpu. Requires to be ran from hardirq
+ * context with local IRQs disabled.
+ */
+void irq_work_run(void)
+{
+        struct irq_work *list, **head;
+
+        head = &__get_cpu_var(irq_work_list);
+        if (*head == NULL)
+                return;
+
+        BUG_ON(!in_irq());
+        BUG_ON(!irqs_disabled());
+
+        list = xchg(head, NULL);
+        while (list != NULL) {
+                struct irq_work *entry = list;
+
+                list = irq_work_next(list);
+
+                /*
+                 * Clear the PENDING bit, after this point the @entry
+                 * can be re-used.
+                 */
+                entry->next = next_flags(NULL, IRQ_WORK_BUSY);
+                entry->func(entry);
+                /*
+                 * Clear the BUSY bit and return to the free state if
+                 * no-one else claimed it meanwhile.
+                 */
+                cmpxchg(&entry->next, next_flags(NULL, IRQ_WORK_BUSY), NULL);
+        }
+}
+EXPORT_SYMBOL_GPL(irq_work_run);
+
+/*
+ * Synchronize against the irq_work @entry, ensures the entry is not
+ * currently in use.
+ */
+void irq_work_sync(struct irq_work *entry)
+{
+        WARN_ON_ONCE(irqs_disabled());
+
+        while (irq_work_is_set(entry, IRQ_WORK_BUSY))
+                cpu_relax();
+}
+EXPORT_SYMBOL_GPL(irq_work_sync);
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 634f86a4b2f9..99b9700e74d0 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -2206,12 +2206,11 @@ static void free_event_rcu(struct rcu_head *head)
         kfree(event);
 }
 
-static void perf_pending_sync(struct perf_event *event);
 static void perf_buffer_put(struct perf_buffer *buffer);
 
 static void free_event(struct perf_event *event)
 {
-        perf_pending_sync(event);
+        irq_work_sync(&event->pending);
 
         if (!event->parent) {
                 atomic_dec(&nr_events);
@@ -3162,16 +3161,7 @@ void perf_event_wakeup(struct perf_event *event)
         }
 }
 
-/*
- * Pending wakeups
- *
- * Handle the case where we need to wakeup up from NMI (or rq->lock) context.
- *
- * The NMI bit means we cannot possibly take locks. Therefore, maintain a
- * single linked list and use cmpxchg() to add entries lockless.
- */
-
-static void perf_pending_event(struct perf_pending_entry *entry)
+static void perf_pending_event(struct irq_work *entry)
 {
         struct perf_event *event = container_of(entry,
                         struct perf_event, pending);
@@ -3187,89 +3177,6 @@ static void perf_pending_event(struct perf_pending_entry *entry)
         }
 }
 
-#define PENDING_TAIL ((struct perf_pending_entry *)-1UL)
-
-static DEFINE_PER_CPU(struct perf_pending_entry *, perf_pending_head) = {
-        PENDING_TAIL,
-};
-
-static void perf_pending_queue(struct perf_pending_entry *entry,
-                               void (*func)(struct perf_pending_entry *))
-{
-        struct perf_pending_entry **head;
-
-        if (cmpxchg(&entry->next, NULL, PENDING_TAIL) != NULL)
-                return;
-
-        entry->func = func;
-
-        head = &get_cpu_var(perf_pending_head);
-
-        do {
-                entry->next = *head;
-        } while (cmpxchg(head, entry->next, entry) != entry->next);
-
-        set_perf_event_pending();
-
-        put_cpu_var(perf_pending_head);
-}
-
-static int __perf_pending_run(void)
-{
-        struct perf_pending_entry *list;
-        int nr = 0;
-
-        list = xchg(&__get_cpu_var(perf_pending_head), PENDING_TAIL);
-        while (list != PENDING_TAIL) {
-                void (*func)(struct perf_pending_entry *);
-                struct perf_pending_entry *entry = list;
-
-                list = list->next;
-
-                func = entry->func;
-                entry->next = NULL;
-                /*
-                 * Ensure we observe the unqueue before we issue the wakeup,
-                 * so that we won't be waiting forever.
-                 * -- see perf_not_pending().
-                 */
-                smp_wmb();
-
-                func(entry);
-                nr++;
-        }
-
-        return nr;
-}
-
-static inline int perf_not_pending(struct perf_event *event)
-{
-        /*
-         * If we flush on whatever cpu we run, there is a chance we don't
-         * need to wait.
-         */
-        get_cpu();
-        __perf_pending_run();
-        put_cpu();
-
-        /*
-         * Ensure we see the proper queue state before going to sleep
-         * so that we do not miss the wakeup. -- see perf_pending_handle()
-         */
-        smp_rmb();
-        return event->pending.next == NULL;
-}
-
-static void perf_pending_sync(struct perf_event *event)
-{
-        wait_event(event->waitq, perf_not_pending(event));
-}
-
-void perf_event_do_pending(void)
-{
-        __perf_pending_run();
-}
-
 /*
  * We assume there is only KVM supporting the callbacks.
  * Later on, we might change it to a list if there is
@@ -3319,8 +3226,7 @@ static void perf_output_wakeup(struct perf_output_handle *handle)
 
         if (handle->nmi) {
                 handle->event->pending_wakeup = 1;
-                perf_pending_queue(&handle->event->pending,
-                                   perf_pending_event);
+                irq_work_queue(&handle->event->pending);
         } else
                 perf_event_wakeup(handle->event);
 }
@@ -4356,8 +4262,7 @@ static int __perf_event_overflow(struct perf_event *event, int nmi,
                 event->pending_kill = POLL_HUP;
                 if (nmi) {
                         event->pending_disable = 1;
-                        perf_pending_queue(&event->pending,
-                                           perf_pending_event);
+                        irq_work_queue(&event->pending);
                 } else
                         perf_event_disable(event);
         }
@@ -5374,6 +5279,7 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
         INIT_LIST_HEAD(&event->event_entry);
         INIT_LIST_HEAD(&event->sibling_list);
         init_waitqueue_head(&event->waitq);
+        init_irq_work(&event->pending, perf_pending_event);
 
         mutex_init(&event->mmap_mutex);
 
diff --git a/kernel/timer.c b/kernel/timer.c
index 97bf05baade7..68a9ae7679b7 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -37,7 +37,7 @@
 #include <linux/delay.h>
 #include <linux/tick.h>
 #include <linux/kallsyms.h>
-#include <linux/perf_event.h>
+#include <linux/irq_work.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
 
@@ -1279,7 +1279,10 @@ void update_process_times(int user_tick)
         run_local_timers();
         rcu_check_callbacks(cpu, user_tick);
         printk_tick();
-        perf_event_do_pending();
+#ifdef CONFIG_IRQ_WORK
+        if (in_irq())
+                irq_work_run();
+#endif
         scheduler_tick();
         run_posix_cpu_timers(p);
 }