2019-09-04 04:35:23

by Harish Bandi

[permalink] [raw]
Subject: [PATCH v1] bluetooth: hci_qca: disable irqs when spinlock is acquired

Looks like Deadlock is observed in hci_qca while performing
stress and stability tests. Since same lock is getting
acquired from qca_wq_awake_rx and hci_ibs_tx_idle_timeout
seeing spinlock recursion, irqs should be disable while
acquiring the spinlock always.

Signed-off-by: Harish Bandi <[email protected]>
---
drivers/bluetooth/hci_qca.c | 10 ++++++----
1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/drivers/bluetooth/hci_qca.c b/drivers/bluetooth/hci_qca.c
index d33828f..e3164c2 100644
--- a/drivers/bluetooth/hci_qca.c
+++ b/drivers/bluetooth/hci_qca.c
@@ -309,13 +309,14 @@ static void qca_wq_awake_device(struct work_struct *work)
ws_awake_device);
struct hci_uart *hu = qca->hu;
unsigned long retrans_delay;
+ unsigned long flags;

BT_DBG("hu %p wq awake device", hu);

/* Vote for serial clock */
serial_clock_vote(HCI_IBS_TX_VOTE_CLOCK_ON, hu);

- spin_lock(&qca->hci_ibs_lock);
+ spin_lock_irqsave(&qca->hci_ibs_lock, flags);

/* Send wake indication to device */
if (send_hci_ibs_cmd(HCI_IBS_WAKE_IND, hu) < 0)
@@ -327,7 +328,7 @@ static void qca_wq_awake_device(struct work_struct *work)
retrans_delay = msecs_to_jiffies(qca->wake_retrans);
mod_timer(&qca->wake_retrans_timer, jiffies + retrans_delay);

- spin_unlock(&qca->hci_ibs_lock);
+ spin_unlock_irqrestore(&qca->hci_ibs_lock, flags);

/* Actually send the packets */
hci_uart_tx_wakeup(hu);
@@ -338,12 +339,13 @@ static void qca_wq_awake_rx(struct work_struct *work)
struct qca_data *qca = container_of(work, struct qca_data,
ws_awake_rx);
struct hci_uart *hu = qca->hu;
+ unsigned long flags;

BT_DBG("hu %p wq awake rx", hu);

serial_clock_vote(HCI_IBS_RX_VOTE_CLOCK_ON, hu);

- spin_lock(&qca->hci_ibs_lock);
+ spin_lock_irqsave(&qca->hci_ibs_lock, flags);
qca->rx_ibs_state = HCI_IBS_RX_AWAKE;

/* Always acknowledge device wake up,
@@ -354,7 +356,7 @@ static void qca_wq_awake_rx(struct work_struct *work)

qca->ibs_sent_wacks++;

- spin_unlock(&qca->hci_ibs_lock);
+ spin_unlock_irqrestore(&qca->hci_ibs_lock, flags);

/* Actually send the packets */
hci_uart_tx_wakeup(hu);
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project


2019-09-04 14:07:24

by Marcel Holtmann

[permalink] [raw]
Subject: Re: [PATCH v1] bluetooth: hci_qca: disable irqs when spinlock is acquired

Hi Harish,

> Looks like Deadlock is observed in hci_qca while performing
> stress and stability tests. Since same lock is getting
> acquired from qca_wq_awake_rx and hci_ibs_tx_idle_timeout
> seeing spinlock recursion, irqs should be disable while
> acquiring the spinlock always.
>
> Signed-off-by: Harish Bandi <[email protected]>
> ---
> drivers/bluetooth/hci_qca.c | 10 ++++++----
> 1 file changed, 6 insertions(+), 4 deletions(-)

patch has been applied to bluetooth-stable tree.

Regards

Marcel