Re: [RESEND net-next 1/2] net: hns3: add support for PTP

On 2021/6/3 21:14, Richard Cochran wrote:

On Wed, Jun 02, 2021 at 10:57:43AM +0800, Guangbin Huang wrote:

quoted

@@ -4342,12 +4352,34 @@ static void hclge_periodic_service_task(struct hclge_dev *hdev)
  	hclge_task_schedule(hdev, delta);
  }
  
+static void hclge_ptp_service_task(struct hclge_dev *hdev)
+{
+	if (!test_bit(HCLGE_STATE_PTP_EN, &hdev->state) ||
+	    !test_bit(HCLGE_STATE_PTP_TX_HANDLING, &hdev->state) ||
+	    !time_is_before_jiffies(hdev->ptp->tx_start + HZ))
+		return;
+
+	/* to prevent concurrence with the irq handler, disable vector0
+	 * before handling ptp service task.
+	 */
+	disable_irq(hdev->misc_vector.vector_irq);

This won't work.  After all, the ISR thread might already be running.
Use a proper spinlock instead.

Thanks for review. Using spinlock in irq_handler looks heavy, what about
adding a new flag HCLGE_STATE_PTP_CLEANING_TX_HWTS for hclge_ptp_clean_tx_hwts()?
Function hclge_ptp_clean_tx_hwts() test and set this flag at the beginning
and clean it in the end. Do you think it is Ok?

Thanks,
Guangbin
.

quoted

+	/* check HCLGE_STATE_PTP_TX_HANDLING here again, since the irq
+	 * handler may handle it just before disable_irq().
+	 */
+	if (test_bit(HCLGE_STATE_PTP_TX_HANDLING, &hdev->state))
+		hclge_ptp_clean_tx_hwts(hdev);
+
+	enable_irq(hdev->misc_vector.vector_irq);
+}

Thanks,
Richard
.