Re: [PATCH 4/4] Bluetooth: hci_ldisc: make sure we don't loose HCI_UART_TX_WAKEUP events

On Tue, 30 Aug 2016 09:53:53 -0700
Marcel Holtmann [off-list ref] wrote:

Hi Boris,

quoted

The HCI_UART_TX_WAKEUP flag checking is racy and some HCI_UART_TX_WAKEUP
events can be lost.

Signed-off-by: Boris Brezillon <redacted>
---
drivers/bluetooth/hci_ldisc.c | 11 +++++++++++
1 file changed, 11 insertions(+)

diff --git a/drivers/bluetooth/hci_ldisc.c b/drivers/bluetooth/hci_ldisc.c
index 27f73294edcb..ee7b25f1c6ce 100644
--- a/drivers/bluetooth/hci_ldisc.c
+++ b/drivers/bluetooth/hci_ldisc.c

@@ -172,6 +172,17 @@ restart:

		goto restart;

	clear_bit(HCI_UART_SENDING, &hu->tx_state);
+
+	/*
+	 * One last check to make sure hci_uart_tx_wakeup() did not set
+	 * HCI_UART_TX_WAKEUP while we where clearing HCI_UART_SENDING.
+	 * The work might have been scheduled by someone else in the
+	 * meantime, in this case we return directly.
+	 */
+	if (test_bit(HCI_UART_TX_WAKEUP, &hu->tx_state) &&
+	    !test_and_set_bit(HCI_UART_SENDING, &hu->tx_state))
+		goto restart;
+  

I know this is correct, but I would actually make it visually different.

	if (test_bit(UART_TX_WAKEUP, ..) {
		/* comment goes here
		 */
		if (!test_and_set_bit(UART_SENDING, ..)
			goto restart;
	}

For me with a proper comment that is a lot easier to read and grok that it is correct.

Sure, I'll address that.