Bluetooth: Add ncmd=0 recovery handling
authorManish Mandlik <mmandlik@google.com>
Thu, 29 Apr 2021 17:24:22 +0000 (10:24 -0700)
committerMarcel Holtmann <marcel@holtmann.org>
Sat, 26 Jun 2021 05:12:27 +0000 (07:12 +0200)
During command status or command complete event, the controller may set
ncmd=0 indicating that it is not accepting any more commands. In such a
case, host holds off sending any more commands to the controller. If the
controller doesn't recover from such condition, host will wait forever,
until the user decides that the Bluetooth is broken and may power cycles
the Bluetooth.

This patch triggers the hardware error to reset the controller and
driver when it gets into such state as there is no other wat out.

Reviewed-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
Signed-off-by: Manish Mandlik <mmandlik@google.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
include/net/bluetooth/hci.h
include/net/bluetooth/hci_core.h
net/bluetooth/hci_core.c
net/bluetooth/hci_event.c

index ea4ae55..c4b0650 100644 (file)
@@ -339,6 +339,7 @@ enum {
 #define HCI_PAIRING_TIMEOUT    msecs_to_jiffies(60000) /* 60 seconds */
 #define HCI_INIT_TIMEOUT       msecs_to_jiffies(10000) /* 10 seconds */
 #define HCI_CMD_TIMEOUT                msecs_to_jiffies(2000)  /* 2 seconds */
+#define HCI_NCMD_TIMEOUT       msecs_to_jiffies(4000)  /* 4 seconds */
 #define HCI_ACL_TX_TIMEOUT     msecs_to_jiffies(45000) /* 45 seconds */
 #define HCI_AUTO_OFF_TIMEOUT   msecs_to_jiffies(2000)  /* 2 seconds */
 #define HCI_POWER_OFF_TIMEOUT  msecs_to_jiffies(5000)  /* 5 seconds */
index 8f5f390..43b08be 100644 (file)
@@ -470,6 +470,7 @@ struct hci_dev {
        struct delayed_work     service_cache;
 
        struct delayed_work     cmd_timer;
+       struct delayed_work     ncmd_timer;
 
        struct work_struct      rx_work;
        struct work_struct      cmd_work;
index 25484bb..572f236 100644 (file)
@@ -1730,6 +1730,7 @@ int hci_dev_do_close(struct hci_dev *hdev)
        }
 
        cancel_delayed_work(&hdev->power_off);
+       cancel_delayed_work(&hdev->ncmd_timer);
 
        hci_request_cancel_all(hdev);
        hci_req_sync_lock(hdev);
@@ -2777,6 +2778,24 @@ static void hci_cmd_timeout(struct work_struct *work)
        queue_work(hdev->workqueue, &hdev->cmd_work);
 }
 
+/* HCI ncmd timer function */
+static void hci_ncmd_timeout(struct work_struct *work)
+{
+       struct hci_dev *hdev = container_of(work, struct hci_dev,
+                                           ncmd_timer.work);
+
+       bt_dev_err(hdev, "Controller not accepting commands anymore: ncmd = 0");
+
+       /* During HCI_INIT phase no events can be injected if the ncmd timer
+        * triggers since the procedure has its own timeout handling.
+        */
+       if (test_bit(HCI_INIT, &hdev->flags))
+               return;
+
+       /* This is an irrecoverable state, inject hardware error event */
+       hci_reset_dev(hdev);
+}
+
 struct oob_data *hci_find_remote_oob_data(struct hci_dev *hdev,
                                          bdaddr_t *bdaddr, u8 bdaddr_type)
 {
@@ -3841,6 +3860,7 @@ struct hci_dev *hci_alloc_dev(void)
        init_waitqueue_head(&hdev->suspend_wait_q);
 
        INIT_DELAYED_WORK(&hdev->cmd_timer, hci_cmd_timeout);
+       INIT_DELAYED_WORK(&hdev->ncmd_timer, hci_ncmd_timeout);
 
        hci_request_setup(hdev);
 
@@ -4078,6 +4098,8 @@ int hci_reset_dev(struct hci_dev *hdev)
        hci_skb_pkt_type(skb) = HCI_EVENT_PKT;
        skb_put_data(skb, hw_err, 3);
 
+       bt_dev_err(hdev, "Injecting HCI hardware error event");
+
        /* Send Hardware Error to upper stack */
        return hci_recv_frame(hdev, skb);
 }
index 59c5329..18339eb 100644 (file)
@@ -3268,6 +3268,23 @@ unlock:
        hci_dev_unlock(hdev);
 }
 
+static inline void handle_cmd_cnt_and_timer(struct hci_dev *hdev,
+                                           u16 opcode, u8 ncmd)
+{
+       if (opcode != HCI_OP_NOP)
+               cancel_delayed_work(&hdev->cmd_timer);
+
+       if (!test_bit(HCI_RESET, &hdev->flags)) {
+               if (ncmd) {
+                       cancel_delayed_work(&hdev->ncmd_timer);
+                       atomic_set(&hdev->cmd_cnt, 1);
+               } else {
+                       schedule_delayed_work(&hdev->ncmd_timer,
+                                             HCI_NCMD_TIMEOUT);
+               }
+       }
+}
+
 static void hci_cmd_complete_evt(struct hci_dev *hdev, struct sk_buff *skb,
                                 u16 *opcode, u8 *status,
                                 hci_req_complete_t *req_complete,
@@ -3630,11 +3647,7 @@ static void hci_cmd_complete_evt(struct hci_dev *hdev, struct sk_buff *skb,
                break;
        }
 
-       if (*opcode != HCI_OP_NOP)
-               cancel_delayed_work(&hdev->cmd_timer);
-
-       if (ev->ncmd && !test_bit(HCI_RESET, &hdev->flags))
-               atomic_set(&hdev->cmd_cnt, 1);
+       handle_cmd_cnt_and_timer(hdev, *opcode, ev->ncmd);
 
        hci_req_cmd_complete(hdev, *opcode, *status, req_complete,
                             req_complete_skb);
@@ -3735,11 +3748,7 @@ static void hci_cmd_status_evt(struct hci_dev *hdev, struct sk_buff *skb,
                break;
        }
 
-       if (*opcode != HCI_OP_NOP)
-               cancel_delayed_work(&hdev->cmd_timer);
-
-       if (ev->ncmd && !test_bit(HCI_RESET, &hdev->flags))
-               atomic_set(&hdev->cmd_cnt, 1);
+       handle_cmd_cnt_and_timer(hdev, *opcode, ev->ncmd);
 
        /* Indicate request completion if the command failed. Also, if
         * we're not waiting for a special event and we get a success