[PATCH] driver core: Use mod_delayed_work to prevent lost deferred probe work

[Zhang Yuwei]

The deferred_probe_timeout_work may be permanently and unexpectedly
canceled when deferred_probe_extend_timeout() executes concurrently.
Starting with deferred_probe_timeout_work pending, the problem can
occur after the following sequence:

  CPU0                                 CPU1
deferred_probe_extend_timeout
  -> cancel_delayed_work() => true
                                   deferred_probe_extend_timeout
                                     -> cancel_delayed_work()
                                       -> __cancel_work()
                                         -> try_grab_pending()
  -> schedule_delayed_work()
    -> queue_delayed_work_on()
(Since the pending bit is grabbed,
 it just returns without queuing)
                                         -> set_work_pool_and_clear_pending()
                                  (This __cancel_work() returns false and
                                     the work will never be queued again)

The root cause is that the WORK_STRUCT_PENDING_BIT of the work_struct
is set temporarily in __cancel_work() (via try_grab_pending()). This
transient state prevents the work_struct from being successfully queued
by another CPU.

To fix this, replace the original non-atomic cancel and schedule
mechanism with mod_delayed_work(). This ensures the modification is
handled atomically and guarantees that the work is not lost.

Fixes: 2b28a1a84a0e ("driver core: Extend deferred probe timeout on driver registration")
Signed-off-by: Zhang Yuwei <zhangyuwei20@xxxxxxxxxx>
---
Link to previous discussion
Link:https://lore.kernel.org/all/40fa16cf-950b-4ca7-9935-dbce75e46eb9@xxxxxxxxxx/

 drivers/base/dd.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/base/dd.c b/drivers/base/dd.c
index 37c7e54e0e4c..fac019f567ae 100644
--- a/drivers/base/dd.c
+++ b/drivers/base/dd.c
@@ -327,12 +327,10 @@ void deferred_probe_extend_timeout(void)
 	 * If the work hasn't been queued yet or if the work expired, don't
 	 * start a new one.
 	 */
-	if (cancel_delayed_work(&deferred_probe_timeout_work)) {
-		schedule_delayed_work(&deferred_probe_timeout_work,
-				driver_deferred_probe_timeout * HZ);
+	if (mod_delayed_work(system_wq, &deferred_probe_timeout_work,
+						 driver_deferred_probe_timeout))
 		pr_debug("Extended deferred probe timeout by %d secs\n",
 					driver_deferred_probe_timeout);
-	}
 }
 
 /**
-- 
2.43.0