[PATCH 0/2] Bluetooth: Suspend related bugfixes

Hi Marcel,

After further automated testing of the upstreamed suspend patches,
I found two issues:
- A failure in PM_SUSPEND_PREPARE wasn't calling PM_POST_SUSPEND.
  I misread the docs and thought it would call it for all notifiers
  already run but it only does so for the ones that returned
  successfully from PM_SUSPEND_PREPARE.
- hci_conn_complete_evt wasn't completing on auto-connects (an else
  block was removed during a refactor incorrectly)

With the following patches, I've run a suspend stress test on a couple
of Chromebooks for several dozen iterations (each) successfully.

Thanks
Abhishek



Abhishek Pandit-Subedi (2):
  Bluetooth: Restore running state if suspend fails
  Bluetooth: Fix incorrect branch in connection complete

 net/bluetooth/hci_core.c  | 39 ++++++++++++++++++++-------------------
 net/bluetooth/hci_event.c | 17 +++++++++--------
 2 files changed, 29 insertions(+), 27 deletions(-)

-- 
2.25.1.696.g5e7596f4ac-goog

When handling auto-connected devices, we should execute the rest of the
connection complete when it was previously discovered and it is an ACL
connection.

Signed-off-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
---

 net/bluetooth/hci_event.c | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c
index 20408d386268..cd3d7d90029b 100644
--- a/net/bluetooth/hci_event.c
+++ b/net/bluetooth/hci_event.c

@@ -2539,16 +2539,17 @@ static void hci_conn_complete_evt(struct hci_dev *hdev, struct sk_buff *skb)
 				bt_dev_err(hdev, "no memory for new conn");
 				goto unlock;
 			}
-		}
-
-		if (ev->link_type != SCO_LINK)
-			goto unlock;
+		} else {
+			if (ev->link_type != SCO_LINK)
+				goto unlock;
 
-		conn = hci_conn_hash_lookup_ba(hdev, ESCO_LINK, &ev->bdaddr);
-		if (!conn)
-			goto unlock;
+			conn = hci_conn_hash_lookup_ba(hdev, ESCO_LINK,
+						       &ev->bdaddr);
+			if (!conn)
+				goto unlock;
 
-		conn->type = SCO_LINK;
+			conn->type = SCO_LINK;
+		}
 	}
 
 	if (!ev->status) {

-- 
2.25.1.696.g5e7596f4ac-goog

If Bluetooth fails to enter the suspended state correctly, restore the
state to running (re-enabling scans). PM_POST_SUSPEND is only sent to
notifiers that successfully return from PM_PREPARE_SUSPEND notification
so we should recover gracefully if it fails.

Signed-off-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
---

 net/bluetooth/hci_core.c | 39 ++++++++++++++++++++-------------------
 1 file changed, 20 insertions(+), 19 deletions(-)

diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index dbd2ad3a26ed..2e7bc2da8371 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c

@@ -3305,6 +3305,15 @@ static void hci_prepare_suspend(struct work_struct *work)
 	hci_dev_unlock(hdev);
 }
 
+static int hci_change_suspend_state(struct hci_dev *hdev,
+				    enum suspended_state next)
+{
+	hdev->suspend_state_next = next;
+	set_bit(SUSPEND_PREPARE_NOTIFIER, hdev->suspend_tasks);
+	queue_work(hdev->req_workqueue, &hdev->suspend_prepare);
+	return hci_suspend_wait_event(hdev);
+}
+
 static int hci_suspend_notifier(struct notifier_block *nb, unsigned long action,
 				void *data)
 {

@@ -3330,32 +3339,24 @@ static int hci_suspend_notifier(struct notifier_block *nb, unsigned long action,
 		 *    connectable (disabling scanning)
 		 *  - Second, program event filter/whitelist and enable scan
 		 */
-		hdev->suspend_state_next = BT_SUSPEND_DISCONNECT;
-		set_bit(SUSPEND_PREPARE_NOTIFIER, hdev->suspend_tasks);
-		queue_work(hdev->req_workqueue, &hdev->suspend_prepare);
-		ret = hci_suspend_wait_event(hdev);
+		ret = hci_change_suspend_state(hdev, BT_SUSPEND_DISCONNECT);
 
-		/* If the disconnect portion failed, don't attempt to complete
-		 * by configuring the whitelist. The suspend notifier will
-		 * follow a cancelled suspend with a PM_POST_SUSPEND
-		 * notification.
-		 */
-		if (!ret) {
-			hdev->suspend_state_next = BT_SUSPEND_COMPLETE;
-			set_bit(SUSPEND_PREPARE_NOTIFIER, hdev->suspend_tasks);
-			queue_work(hdev->req_workqueue, &hdev->suspend_prepare);
-			ret = hci_suspend_wait_event(hdev);
-		}
+		/* Only configure whitelist if disconnect succeeded */
+		if (!ret)
+			ret = hci_change_suspend_state(hdev,
+						       BT_SUSPEND_COMPLETE);
 	} else if (action == PM_POST_SUSPEND) {
-		hdev->suspend_state_next = BT_RUNNING;
-		set_bit(SUSPEND_PREPARE_NOTIFIER, hdev->suspend_tasks);
-		queue_work(hdev->req_workqueue, &hdev->suspend_prepare);
-		ret = hci_suspend_wait_event(hdev);
+		ret = hci_change_suspend_state(hdev, BT_RUNNING);
 	}
 
+	/* If suspend failed, restore it to running */
+	if (ret && action == PM_SUSPEND_PREPARE)
+		hci_change_suspend_state(hdev, BT_RUNNING);
+
 done:
 	return ret ? notifier_from_errno(-EBUSY) : NOTIFY_STOP;
 }
+
 /* Alloc HCI device */
 struct hci_dev *hci_alloc_dev(void)
 {

-- 
2.25.1.696.g5e7596f4ac-goog

Hi Abhishek,

After further automated testing of the upstreamed suspend patches,
I found two issues:
- A failure in PM_SUSPEND_PREPARE wasn't calling PM_POST_SUSPEND.
 I misread the docs and thought it would call it for all notifiers
 already run but it only does so for the ones that returned
 successfully from PM_SUSPEND_PREPARE.
- hci_conn_complete_evt wasn't completing on auto-connects (an else
 block was removed during a refactor incorrectly)

With the following patches, I've run a suspend stress test on a couple
of Chromebooks for several dozen iterations (each) successfully.

Thanks
Abhishek



Abhishek Pandit-Subedi (2):
 Bluetooth: Restore running state if suspend fails
 Bluetooth: Fix incorrect branch in connection complete

net/bluetooth/hci_core.c  | 39 ++++++++++++++++++++-------------------
net/bluetooth/hci_event.c | 17 +++++++++--------
2 files changed, 29 insertions(+), 27 deletions(-)

both patches have been applied to bluetooth-next tree.

Regards

Marcel

Thanks Marcel.

On Mon, Mar 23, 2020 at 10:50 AM Marcel Holtmann [off-list ref] wrote:

Hi Abhishek,

quoted

After further automated testing of the upstreamed suspend patches,
I found two issues:
- A failure in PM_SUSPEND_PREPARE wasn't calling PM_POST_SUSPEND.
 I misread the docs and thought it would call it for all notifiers
 already run but it only does so for the ones that returned
 successfully from PM_SUSPEND_PREPARE.
- hci_conn_complete_evt wasn't completing on auto-connects (an else
 block was removed during a refactor incorrectly)

With the following patches, I've run a suspend stress test on a couple
of Chromebooks for several dozen iterations (each) successfully.

Thanks
Abhishek



Abhishek Pandit-Subedi (2):
 Bluetooth: Restore running state if suspend fails
 Bluetooth: Fix incorrect branch in connection complete

net/bluetooth/hci_core.c  | 39 ++++++++++++++++++++-------------------
net/bluetooth/hci_event.c | 17 +++++++++--------
2 files changed, 29 insertions(+), 27 deletions(-)

both patches have been applied to bluetooth-next tree.

Regards

Marcel