On Wed, Sep 11, 2024 at 5:22 PM Jakub Kicinski <kuba@xxxxxxxxxx> wrote:
>
> On Thu, 12 Sep 2024 00:53:31 +0900 Taehee Yoo wrote:
> > > if (netif_running(dev)) {
> > > bnxt_close_nic(bp, false, false);
> > > bp->rx_copybreak = rx_copybreak;
> > > bnxt_set_ring_params(bp);
> > > bnxt_open_nic(bp, false, false);
> > > } else {
> > > bp->rx_copybreak = rx_copybreak;
> > > }
> >
> > I think your suggestion is much safer!
> > I will use your suggestion in the v3 patch.
>
> This is better but Andy mentioned on another thread that queue reset
> should work, so instead of full close / open maybe we can just do:
>
> for (/* all Rx queues */) {
> bnxt_queue_stop();
> bnxt_queue_start();
> }
>
> when the device is already running?
If the copybreak value changes, I don't think queue restart will work.
We need to size and allocate the buffers differently than before, so I
think we need to do close/open.
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature
