On Thu, Sep 11, 2025 at 10:25:04PM +0800, Fan Gong wrote: > On 9/11/2025 8:33 PM, Simon Horman wrote: > > > > + err = hinic3_get_link_status(nic_dev->hwdev, &link_status_up); > > > + if (!err && link_status_up) > > > + netif_carrier_on(netdev); > > > + > > > + return 0; > > > + > > > +err_flush_qps_res: > > > + hinic3_flush_qps_res(nic_dev->hwdev); > > > + /* wait to guarantee that no packets will be sent to host */ > > > + msleep(100); > > > > I realise that Jakub's feedback on msleep() in his review of v3 was > > in a different code path. But I do wonder if there is a better way. > > ... > > > > + hinic3_flush_txqs(netdev); > > > + /* wait to guarantee that no packets will be sent to host */ > > > + msleep(100); > > > > Likewise, here. > > Thanks for your review, Simon. > > Firstly, The main issue on the code of Jakub's feedback on msleep() is > duplicate code function. The msleep() in hinic3_vport_down and > hinic3_free_hwdev is repetitive because of our oversight. So we removed > msleep() in hinic3_free_hwdev in v04 patch. > > Secondly, there is no better way indeed. As our HW bad decision, HW > didn't have an accurate way of checking if rq has been flushed. The > only way is to close the func & port . Then we wait for HW to process > the pkts and upload them to driver. > The sleep time is determined through our testing. The two calls of > msleep() are the same issue. Thanks for the clarification, much appreciated. > Finally, we have received your reviews on other patches and we will > fix them soon in the next version. >
