Re: [PATCH 1/2] sched/wait: Break up long wake list walk

From: Linus Torvalds
Date: Wed Aug 23 2017 - 19:30:58 EST


On Wed, Aug 23, 2017 at 11:17 AM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Wed, Aug 23, 2017 at 8:58 AM, Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx> wrote:
>>
>> Will you still consider the original patch as a fail safe mechanism?
>
> I don't think we have much choice, although I would *really* want to
> get this root-caused rather than just papering over the symptoms.

Oh well. Apparently we're not making progress on that, so I looked at
the patch again.

Can we fix it up a bit? In particular, the "bookmark_wake_function()"
thing added no value, and definitely shouldn't have been exported.
Just use NULL instead.

And the WAITQUEUE_WALK_BREAK_CNT thing should be internal to
__wake_up_common(), not in some common header file. Again, there's no
value in exporting it to anybody else.

And doing

if (curr->flags & WQ_FLAG_BOOKMARK)

looks odd, when we just did

unsigned flags = curr->flags;

one line earlier, so that can be just simplified.

So can you test that simplified version of the patch? I'm attaching my
suggested edited patch, but you may just want to do those changes
directly to your tree instead.

Hmm?

Linus
include/linux/wait.h | 1 +
kernel/sched/wait.c | 74 ++++++++++++++++++++++++++++++++++++++++++----------
2 files changed, 61 insertions(+), 14 deletions(-)

diff --git a/include/linux/wait.h b/include/linux/wait.h
index dc19880c02f5..78401ef02d29 100644
--- a/include/linux/wait.h
+++ b/include/linux/wait.h
@@ -18,6 +18,7 @@ int default_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int
/* wait_queue_entry::flags */
#define WQ_FLAG_EXCLUSIVE 0x01
#define WQ_FLAG_WOKEN 0x02
+#define WQ_FLAG_BOOKMARK 0x04

/*
* A single wait-queue entry structure:
diff --git a/kernel/sched/wait.c b/kernel/sched/wait.c
index 17f11c6b0a9f..789dc24a323d 100644
--- a/kernel/sched/wait.c
+++ b/kernel/sched/wait.c
@@ -53,6 +53,12 @@ void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry
}
EXPORT_SYMBOL(remove_wait_queue);

+/*
+ * Scan threshold to break wait queue walk.
+ * This allows a waker to take a break from holding the
+ * wait queue lock during the wait queue walk.
+ */
+#define WAITQUEUE_WALK_BREAK_CNT 64

/*
* The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
@@ -63,17 +69,64 @@ EXPORT_SYMBOL(remove_wait_queue);
* started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
* zero in this (rare) case, and we handle it by continuing to scan the queue.
*/
-static void __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
- int nr_exclusive, int wake_flags, void *key)
+static int __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
+ int nr_exclusive, int wake_flags, void *key,
+ wait_queue_entry_t *bookmark)
{
wait_queue_entry_t *curr, *next;
+ int cnt = 0;
+
+ if (bookmark && (bookmark->flags & WQ_FLAG_BOOKMARK)) {
+ curr = list_next_entry(bookmark, entry);

- list_for_each_entry_safe(curr, next, &wq_head->head, entry) {
+ list_del(&bookmark->entry);
+ bookmark->flags = 0;
+ } else
+ curr = list_first_entry(&wq_head->head, wait_queue_entry_t, entry);
+
+ if (&curr->entry == &wq_head->head)
+ return nr_exclusive;
+
+ list_for_each_entry_safe_from(curr, next, &wq_head->head, entry) {
unsigned flags = curr->flags;

+ if (flags & WQ_FLAG_BOOKMARK)
+ continue;
+
if (curr->func(curr, mode, wake_flags, key) &&
(flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
break;
+
+ if (bookmark && (++cnt > WAITQUEUE_WALK_BREAK_CNT) &&
+ (&next->entry != &wq_head->head)) {
+ bookmark->flags = WQ_FLAG_BOOKMARK;
+ list_add_tail(&bookmark->entry, &next->entry);
+ break;
+ }
+ }
+ return nr_exclusive;
+}
+
+static void __wake_up_common_lock(struct wait_queue_head *wq_head, unsigned int mode,
+ int nr_exclusive, int wake_flags, void *key)
+{
+ unsigned long flags;
+ wait_queue_entry_t bookmark;
+
+ bookmark.flags = 0;
+ bookmark.private = NULL;
+ bookmark.func = NULL;
+ INIT_LIST_HEAD(&bookmark.entry);
+
+ spin_lock_irqsave(&wq_head->lock, flags);
+ nr_exclusive = __wake_up_common(wq_head, mode, nr_exclusive, wake_flags, key, &bookmark);
+ spin_unlock_irqrestore(&wq_head->lock, flags);
+
+ while (bookmark.flags & WQ_FLAG_BOOKMARK) {
+ spin_lock_irqsave(&wq_head->lock, flags);
+ nr_exclusive = __wake_up_common(wq_head, mode, nr_exclusive,
+ wake_flags, key, &bookmark);
+ spin_unlock_irqrestore(&wq_head->lock, flags);
}
}

@@ -90,11 +143,7 @@ static void __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
void __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
int nr_exclusive, void *key)
{
- unsigned long flags;
-
- spin_lock_irqsave(&wq_head->lock, flags);
- __wake_up_common(wq_head, mode, nr_exclusive, 0, key);
- spin_unlock_irqrestore(&wq_head->lock, flags);
+ __wake_up_common_lock(wq_head, mode, nr_exclusive, 0, key);
}
EXPORT_SYMBOL(__wake_up);

@@ -103,13 +152,13 @@ EXPORT_SYMBOL(__wake_up);
*/
void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
{
- __wake_up_common(wq_head, mode, nr, 0, NULL);
+ __wake_up_common(wq_head, mode, nr, 0, NULL, NULL);
}
EXPORT_SYMBOL_GPL(__wake_up_locked);

void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
{
- __wake_up_common(wq_head, mode, 1, 0, key);
+ __wake_up_common(wq_head, mode, 1, 0, key, NULL);
}
EXPORT_SYMBOL_GPL(__wake_up_locked_key);

@@ -133,7 +182,6 @@ EXPORT_SYMBOL_GPL(__wake_up_locked_key);
void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
int nr_exclusive, void *key)
{
- unsigned long flags;
int wake_flags = 1; /* XXX WF_SYNC */

if (unlikely(!wq_head))
@@ -142,9 +190,7 @@ void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
if (unlikely(nr_exclusive != 1))
wake_flags = 0;

- spin_lock_irqsave(&wq_head->lock, flags);
- __wake_up_common(wq_head, mode, nr_exclusive, wake_flags, key);
- spin_unlock_irqrestore(&wq_head->lock, flags);
+ __wake_up_common_lock(wq_head, mode, nr_exclusive, wake_flags, key);
}
EXPORT_SYMBOL_GPL(__wake_up_sync_key);