On 8/30/19 3:17 PM, David Miller wrote:

From: Jakub Kicinski <redacted>
Date: Fri, 30 Aug 2019 15:16:04 -0700

quoted

On Fri, 30 Aug 2019 12:31:07 -0700, Shannon Nelson wrote:

quoted

On 8/29/19 3:52 PM, Jakub Kicinski wrote:

quoted

On Thu, 29 Aug 2019 11:27:08 -0700, Shannon Nelson wrote:

quoted

+static void ionic_lif_qcq_deinit(struct ionic_lif *lif, struct ionic_qcq *qcq)
+{
+	struct ionic_dev *idev = &lif->ionic->idev;
+	struct device *dev = lif->ionic->dev;
+
+	if (!qcq)
+		return;
+
+	ionic_debugfs_del_qcq(qcq);
+
+	if (!(qcq->flags & IONIC_QCQ_F_INITED))
+		return;
+
+	if (qcq->flags & IONIC_QCQ_F_INTR) {
+		ionic_intr_mask(idev->intr_ctrl, qcq->intr.index,
+				IONIC_INTR_MASK_SET);
+		synchronize_irq(qcq->intr.vector);
+		devm_free_irq(dev, qcq->intr.vector, &qcq->napi);

Doesn't free_irq() basically imply synchronize_irq()?

The synchronize_irq() waits for any threaded handlers to finish, while
free_irq() only waits for HW handling.  This helps makes sure we don't
have anything still running before we remove resources.

mm.. I'm no IRQ expert but it strikes me as surprising as that'd mean
every single driver would always have to run synchronize_irq() on
module exit, no?

I see there is a kthread_stop() in __free_irq(), you sure it doesn't
wait for threaded IRQs?

I'm pretty sure it does.

Yes, deeper in there are the kthread_stop() calls that make the 
synchronize_irq() unneccessary.  I'll pull it out.

Thanks,
sln