[PATCH v2 2/4] async: make async_synchronize_full() flush all workregardless of domain

[Dan Williams]

In response to an async related regression James noted:

  "My theory is that this is an init problem: The assumption in a lot of
   our code is that async_synchronize_full() waits for everything ... even
   the domain specific async schedules, which isn't true."

...so make this assumption true.

Each domain, including the default one, registers itself on a global domain
list when work is scheduled.  Once all entries complete it exits that
list.  Waiting for the list to be empty syncs all in-flight work across
all domains.

Domains can opt-out of global syncing if they are declared as exclusive
ASYNC_DOMAIN_EXCLUSIVE().  All stack-based domains have been declared
exclusive since the domain may go out of scope as soon as the last work
item completes.

Statically declared domains are mostly ok, but async_unregister_domain()
is there to close any theoretical races with pending
async_synchronize_full waiters at module removal time.

Cc: Arjan van de Ven <arjan@xxxxxxxxxxxxxxx>
Cc: Len Brown <len.brown@xxxxxxxxx>
Cc: Rafael J. Wysocki <rjw@xxxxxxx>
Cc: James Bottomley <JBottomley@xxxxxxxxxxxxx>
Reported-by: Meelis Roos <mroos@xxxxxxxx>
Reported-by: Eldad Zack <eldadzack@xxxxxxxxx>
Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
---
 drivers/scsi/scsi.c   |    1 +
 include/linux/async.h |    1 +
 kernel/async.c        |   43 +++++++++++++++++++++++++++++++++++++++++--
 3 files changed, 43 insertions(+), 2 deletions(-)

diff --git a/drivers/scsi/scsi.c b/drivers/scsi/scsi.c
index 2f92007..2dd3c93 100644
--- a/drivers/scsi/scsi.c
+++ b/drivers/scsi/scsi.c
@@ -1355,6 +1355,7 @@ static void __exit exit_scsi(void)
 	scsi_exit_devinfo();
 	scsi_exit_procfs();
 	scsi_exit_queue();
+	async_unregister_domain(&scsi_sd_probe_domain);
 }
 
 subsys_initcall(init_scsi);
diff --git a/include/linux/async.h b/include/linux/async.h
index 364e7ff..7a24fe9 100644
--- a/include/linux/async.h
+++ b/include/linux/async.h
@@ -46,6 +46,7 @@ struct async_domain {
 extern async_cookie_t async_schedule(async_func_ptr *ptr, void *data);
 extern async_cookie_t async_schedule_domain(async_func_ptr *ptr, void *data,
 					    struct async_domain *domain);
+void async_unregister_domain(struct async_domain *domain);
 extern void async_synchronize_full(void);
 extern void async_synchronize_full_domain(struct async_domain *domain);
 extern void async_synchronize_cookie(async_cookie_t cookie);
diff --git a/kernel/async.c b/kernel/async.c
index ba5491d..9d31183 100644
--- a/kernel/async.c
+++ b/kernel/async.c
@@ -63,7 +63,9 @@ static async_cookie_t next_cookie = 1;
 
 static LIST_HEAD(async_pending);
 static ASYNC_DOMAIN(async_running);
+static LIST_HEAD(async_domains);
 static DEFINE_SPINLOCK(async_lock);
+static DEFINE_MUTEX(async_register_mutex);
 
 struct async_entry {
 	struct list_head	list;
@@ -145,6 +147,8 @@ static void async_run_entry_fn(struct work_struct *work)
 	/* 3) remove self from the running queue */
 	spin_lock_irqsave(&async_lock, flags);
 	list_del(&entry->list);
+	if (running->registered && --running->count == 0)
+		list_del_init(&running->node);
 
 	/* 4) free the entry */
 	kfree(entry);
@@ -187,6 +191,8 @@ static async_cookie_t __async_schedule(async_func_ptr *ptr, void *data, struct a
 	spin_lock_irqsave(&async_lock, flags);
 	newcookie = entry->cookie = next_cookie++;
 	list_add_tail(&entry->list, &async_pending);
+	if (running->registered && running->count++ == 0)
+		list_add_tail(&running->node, &async_domains);
 	atomic_inc(&entry_count);
 	spin_unlock_irqrestore(&async_lock, flags);
 
@@ -236,13 +242,43 @@ EXPORT_SYMBOL_GPL(async_schedule_domain);
  */
 void async_synchronize_full(void)
 {
+	mutex_lock(&async_register_mutex);
 	do {
-		async_synchronize_cookie(next_cookie);
-	} while (!list_empty(&async_running.domain) || !list_empty(&async_pending));
+		struct async_domain *domain = NULL;
+
+		spin_lock_irq(&async_lock);
+		if (!list_empty(&async_domains))
+			domain = list_first_entry(&async_domains, typeof(*domain), node);
+		spin_unlock_irq(&async_lock);
+
+		async_synchronize_cookie_domain(next_cookie, domain);
+	} while (!list_empty(&async_domains));
+	mutex_unlock(&async_register_mutex);
 }
 EXPORT_SYMBOL_GPL(async_synchronize_full);
 
 /**
+ * async_unregister_domain - ensure no more anonymous waiters on this domain
+ * @domain: idle domain to flush out of any async_synchronize_full instances
+ *
+ * async_synchronize_{cookie|full}_domain() are not flushed since callers
+ * of these routines should know the lifetime of @domain
+ *
+ * Prefer ASYNC_DOMAIN_EXCLUSIVE() declarations over flushing
+ */
+void async_unregister_domain(struct async_domain *domain)
+{
+	mutex_lock(&async_register_mutex);
+	spin_lock_irq(&async_lock);
+	WARN_ON(!domain->registered || !list_empty(&domain->node) ||
+		!list_empty(&domain->domain));
+	domain->registered = 0;
+	spin_unlock_irq(&async_lock);
+	mutex_unlock(&async_register_mutex);
+}
+EXPORT_SYMBOL_GPL(async_unregister_domain);
+
+/**
  * async_synchronize_full_domain - synchronize all asynchronous function within a certain domain
  * @domain: running list to synchronize on
  *
@@ -268,6 +304,9 @@ void async_synchronize_cookie_domain(async_cookie_t cookie, struct async_domain
 {
 	ktime_t uninitialized_var(starttime), delta, endtime;
 
+	if (!running)
+		return;
+
 	if (initcall_debug && system_state == SYSTEM_BOOTING) {
 		printk(KERN_DEBUG "async_waiting @ %i\n", task_pid_nr(current));
 		starttime = ktime_get();

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/