ia64: salinfo: use a waitqueue instead a sema down/up combo
authorSebastian Andrzej Siewior <bigeasy@linutronix.de>
Mon, 25 Jul 2016 08:43:13 +0000 (10:43 +0200)
committerTony Luck <tony.luck@intel.com>
Mon, 25 Jul 2016 16:55:32 +0000 (09:55 -0700)
The only purpose of down_try_lock() followed by up() seems to be to wake
up a possible reader. This patch replaces it with a wake-queue. There is
no locking around cpumask_empty() and the test is re-done in case there
was no hit.
With wait_event_interruptible_lock_irq(,&data_saved_lock) we would probably
be able to get rid of the `retry` label. However we still can return CPU
X which is valid now but later (after the lock dropped) the event may
have been removed because the CPU went offline.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Tony Luck <tony.luck@intel.com>
arch/ia64/kernel/salinfo.c

index 1eeffb7fbb16b1e35d7e0abbe72dbebe9d1b5c04..5313007d5423a23622192f00218399806b421844 100644 (file)
@@ -141,7 +141,7 @@ enum salinfo_state {
 
 struct salinfo_data {
        cpumask_t               cpu_event;      /* which cpus have outstanding events */
-       struct semaphore        mutex;
+       wait_queue_head_t       read_wait;
        u8                      *log_buffer;
        u64                     log_size;
        u8                      *oemdata;       /* decoded oem data */
@@ -182,21 +182,6 @@ struct salinfo_platform_oemdata_parms {
        int ret;
 };
 
-/* Kick the mutex that tells user space that there is work to do.  Instead of
- * trying to track the state of the mutex across multiple cpus, in user
- * context, interrupt context, non-maskable interrupt context and hotplug cpu,
- * it is far easier just to grab the mutex if it is free then release it.
- *
- * This routine must be called with data_saved_lock held, to make the down/up
- * operation atomic.
- */
-static void
-salinfo_work_to_do(struct salinfo_data *data)
-{
-       (void)(down_trylock(&data->mutex) ?: 0);
-       up(&data->mutex);
-}
-
 static void
 salinfo_platform_oemdata_cpu(void *context)
 {
@@ -258,7 +243,7 @@ salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
        }
        cpumask_set_cpu(smp_processor_id(), &data->cpu_event);
        if (irqsafe) {
-               salinfo_work_to_do(data);
+               wake_up_interruptible(&data->read_wait);
                spin_unlock_irqrestore(&data_saved_lock, flags);
        }
 }
@@ -271,14 +256,10 @@ extern void ia64_mlogbuf_dump(void);
 static void
 salinfo_timeout_check(struct salinfo_data *data)
 {
-       unsigned long flags;
        if (!data->open)
                return;
-       if (!cpumask_empty(&data->cpu_event)) {
-               spin_lock_irqsave(&data_saved_lock, flags);
-               salinfo_work_to_do(data);
-               spin_unlock_irqrestore(&data_saved_lock, flags);
-       }
+       if (!cpumask_empty(&data->cpu_event))
+               wake_up_interruptible(&data->read_wait);
 }
 
 static void
@@ -308,10 +289,11 @@ salinfo_event_read(struct file *file, char __user *buffer, size_t count, loff_t
        int i, n, cpu = -1;
 
 retry:
-       if (cpumask_empty(&data->cpu_event) && down_trylock(&data->mutex)) {
+       if (cpumask_empty(&data->cpu_event)) {
                if (file->f_flags & O_NONBLOCK)
                        return -EAGAIN;
-               if (down_interruptible(&data->mutex))
+               if (wait_event_interruptible(data->read_wait,
+                                            !cpumask_empty(&data->cpu_event)))
                        return -EINTR;
        }
 
@@ -510,7 +492,7 @@ salinfo_log_clear(struct salinfo_data *data, int cpu)
        if (data->state == STATE_LOG_RECORD) {
                spin_lock_irqsave(&data_saved_lock, flags);
                cpumask_set_cpu(cpu, &data->cpu_event);
-               salinfo_work_to_do(data);
+               wake_up_interruptible(&data->read_wait);
                spin_unlock_irqrestore(&data_saved_lock, flags);
        }
        return 0;
@@ -582,7 +564,7 @@ salinfo_cpu_callback(struct notifier_block *nb, unsigned long action, void *hcpu
                     i < ARRAY_SIZE(salinfo_data);
                     ++i, ++data) {
                        cpumask_set_cpu(cpu, &data->cpu_event);
-                       salinfo_work_to_do(data);
+                       wake_up_interruptible(&data->read_wait);
                }
                spin_unlock_irqrestore(&data_saved_lock, flags);
                break;
@@ -640,7 +622,7 @@ salinfo_init(void)
        for (i = 0; i < ARRAY_SIZE(salinfo_log_name); i++) {
                data = salinfo_data + i;
                data->type = i;
-               sema_init(&data->mutex, 1);
+               init_waitqueue_head(&data->read_wait);
                dir = proc_mkdir(salinfo_log_name[i], salinfo_dir);
                if (!dir)
                        continue;