Hi Abhishek,
> QCA_ROME doesn't have support for the reset gpio but sometimes gets into
> a state where it is unresponsive to commands. When this happens, reset
> the port to attempt to revive the chip.
>
> Signed-off-by: Abhishek Pandit-Subedi <abhishekpandit@xxxxxxxxxxxx>
> ---
> On Chromebooks with this chipset, we encountered cmd_timeout after
> running suspend stress test for hundreds of iterations. Without
> a recovery mechanism, continued cmd_timeout failures eventually caused
> the suspend stress test to fail.
>
> This change will just reset the port that the Bluetooth chip is on when
> cmd_timeout is encountered. At the very least, the driver will unload
> and stop affecting suspend. It doesn't seem to restore the BT controller
> to a good state however (this still requires a power cycle).
>
> drivers/bluetooth/btusb.c | 17 +++++++++++++++++
> 1 file changed, 17 insertions(+)
>
> diff --git a/drivers/bluetooth/btusb.c b/drivers/bluetooth/btusb.c
> index e42fdd625eb023..b806a88702328b 100644
> --- a/drivers/bluetooth/btusb.c
> +++ b/drivers/bluetooth/btusb.c
> @@ -573,6 +573,22 @@ static void btusb_rtl_cmd_timeout(struct hci_dev *hdev)
> gpiod_set_value_cansleep(reset_gpio, 0);
> }
>
> +static void btusb_generic_usb_cmd_timeout(struct hci_dev *hdev)
> +{
> + struct btusb_data *data = hci_get_drvdata(hdev);
> + int err;
> +
> + if (++data->cmd_timeout_cnt < 5)
> + return;
> +
> + bt_dev_err(hdev, "Multiple cmd timeouts seen. Resetting usb device.");
> + err = usb_autopm_get_interface(data->intf);
> + if (!err)
> + usb_queue_reset_device(data->intf);
> + else
> + bt_dev_err(hdev, "Failed usb_autopm_get_interface with %d", err);
> +}
> +
> static inline void btusb_free_frags(struct btusb_data *data)
> {
> unsigned long flags;
> @@ -3964,6 +3980,7 @@ static int btusb_probe(struct usb_interface *intf,
> if (id->driver_info & BTUSB_QCA_ROME) {
> data->setup_on_usb = btusb_setup_qca;
> hdev->set_bdaddr = btusb_set_bdaddr_ath3012;
> + hdev->cmd_timeout = btusb_generic_usb_cmd_timeout;
lets give this a btusb_qca_cmd_timeout name. If it ever gets generic, we can rename it, but right now it is specific to QCA.
Regards
Marcel
