Hi Jakub,
On Fri, 6 Oct 2023 14:47:02 -0700
Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
> On Thu, 28 Sep 2023 09:06:28 +0200 Herve Codina wrote:
> > +static int qmc_hdlc_close(struct net_device *netdev)
> > +{
> > + struct qmc_hdlc *qmc_hdlc = netdev_to_qmc_hdlc(netdev);
> > + struct qmc_hdlc_desc *desc;
> > + int i;
> > +
> > + netif_stop_queue(netdev);
> > +
> > + qmc_chan_stop(qmc_hdlc->qmc_chan, QMC_CHAN_ALL);
> > + qmc_chan_reset(qmc_hdlc->qmc_chan, QMC_CHAN_ALL);
>
> stopping the queue looks a bit racy, a completion may come in
> and restart the queue
Indeed, qmc_hdlc_xmit_complete() completion can restart the queue.
I will call netif_stop_queue() after calling qmc_chan_stop/reset().
This is the simple way to fix that without adding an internal flag.
Thanks for pointing that.
Best regards,
Hervé
>
> > + for (i = 0; i < ARRAY_SIZE(qmc_hdlc->tx_descs); i++) {
> > + desc = &qmc_hdlc->tx_descs[i];
> > + if (!desc->skb)
> > + continue;
> > + dma_unmap_single(qmc_hdlc->dev, desc->dma_addr, desc->dma_size,
> > + DMA_TO_DEVICE);
> > + kfree_skb(desc->skb);
> > + desc->skb = NULL;
> > + }
> > +
> > + for (i = 0; i < ARRAY_SIZE(qmc_hdlc->rx_descs); i++) {
> > + desc = &qmc_hdlc->rx_descs[i];
> > + if (!desc->skb)
> > + continue;
> > + dma_unmap_single(qmc_hdlc->dev, desc->dma_addr, desc->dma_size,
> > + DMA_FROM_DEVICE);
> > + kfree_skb(desc->skb);
> > + desc->skb = NULL;
> > + }
> > +
> > + hdlc_close(netdev);
