2022-04-22 08:32:34

by John Ogness

Subject: [PATCH printk v3 03/15] printk: add missing memory barrier to wake_up_klogd()

It is important that any new records are visible to preparing
waiters before the waker checks if the wait queue is empty.
Otherwise it is possible that:

- there are new records available
- the waker sees an empty wait queue and does not wake
- the preparing waiter sees no new records and begins to wait

This is exactly the problem that the function description of
waitqueue_active() warns about.

Use wq_has_sleeper() instead of waitqueue_active() because it
includes the necessary full memory barrier.

Signed-off-by: John Ogness <[email protected]>
---
kernel/printk/printk.c | 39 ++++++++++++++++++++++++++++++++++++---
1 file changed, 36 insertions(+), 3 deletions(-)

diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
index 13a1eebe72af..f817dfb4852d 100644
--- a/kernel/printk/printk.c
+++ b/kernel/printk/printk.c
@@ -746,8 +746,19 @@ static ssize_t devkmsg_read(struct file *file, char __user *buf,
goto out;
}

+ /*
+ * Guarantee this task is visible on the waitqueue before
+ * checking the wake condition.
+ *
+ * The full memory barrier within set_current_state() of
+ * prepare_to_wait_event() pairs with the full memory barrier
+ * within wq_has_sleeper().
+ *
+ * This pairs with wake_up_klogd:A.
+ */
ret = wait_event_interruptible(log_wait,
- prb_read_valid(prb, atomic64_read(&user->seq), r));
+ prb_read_valid(prb,
+ atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
if (ret)
goto out;
}
@@ -1513,7 +1524,18 @@ static int syslog_print(char __user *buf, int size)
seq = syslog_seq;

mutex_unlock(&syslog_lock);
- len = wait_event_interruptible(log_wait, prb_read_valid(prb, seq, NULL));
+ /*
+ * Guarantee this task is visible on the waitqueue before
+ * checking the wake condition.
+ *
+ * The full memory barrier within set_current_state() of
+ * prepare_to_wait_event() pairs with the full memory barrier
+ * within wq_has_sleeper().
+ *
+ * This pairs with wake_up_klogd:A.
+ */
+ len = wait_event_interruptible(log_wait,
+ prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
mutex_lock(&syslog_lock);

if (len)
@@ -3316,7 +3338,18 @@ void wake_up_klogd(void)
return;

preempt_disable();
- if (waitqueue_active(&log_wait)) {
+ /*
+ * Guarantee any new records can be seen by tasks preparing to wait
+ * before this context checks if the wait queue is empty.
+ *
+ * The full memory barrier within wq_has_sleeper() pairs with the full
+ * memory barrier within set_current_state() of
+ * prepare_to_wait_event(), which is called after ___wait_event() adds
+ * the waiter but before it has checked the wait condition.
+ *
+ * This pairs with devkmsg_read:A and syslog_print:A.
+ */
+ if (wq_has_sleeper(&log_wait)) { /* LMM(wake_up_klogd:A) */
this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
}
--
2.30.2

2022-04-22 19:38:17

by Petr Mladek

Subject: Re: [PATCH printk v3 03/15] printk: add missing memory barrier to wake_up_klogd()

On Wed 2022-04-20 01:52:25, John Ogness wrote:
> It is important that any new records are visible to preparing
> waiters before the waker checks if the wait queue is empty.
> Otherwise it is possible that:
>
> - there are new records available
> - the waker sees an empty wait queue and does not wake
> - the preparing waiter sees no new records and begins to wait
>
> This is exactly the problem that the function description of
> waitqueue_active() warns about.
>
> Use wq_has_sleeper() instead of waitqueue_active() because it
> includes the necessary full memory barrier.
>
> Signed-off-by: John Ogness <[email protected]>

Great catch! Looks good to me.

Reviewed-by: Petr Mladek <[email protected]>

Best Regards,
Petr