RE: [PATCH net v6 3/4] net: ethernet: oa_tc6: Disable tx queues on fatal error

[Selvamani Rajagopal]

> -----Original Message-----
> From: Jakub Kicinski <kuba@xxxxxxxxxx>
> Sent: Tuesday, August 18, 2026 9:07 AM
> Subject: Re: [PATCH net v6 3/4] net: ethernet: oa_tc6: Disable tx queues on fatal error
> 
> On Wed, 12 Aug 2026 20:00:11 -0700 Selvamani Rajagopal via B4 Relay
> wrote:
> > + /* disable_traffic, when set, is a point of no return to working
> > + * state. TX queues are disabled. In some cases, upper layer or
> > + * vendor code may inadvertently enable the queue. Intention of
> > + * the disable_traffic flag is to stop traffic from flowing.
> > + */
> 
> I don't get this comment and why this isn't part of patch 1.
> Patch 1 is the place where the stop is removed:

Thanks for the review. I could have made part of patch 1. Since this changes the behavior of
aa_tc6_disable_traffic, I thought I should have it separately so that it is clear for the reviewer.


> 
> - if (tc6->disable_traffic || tc6->waiting_tx_skb) {
> - netif_stop_queue(tc6->netdev);
> - return NETDEV_TX_BUSY;
> - }
> 
> Also the "vendor code may inadvertently enable the queue"
> is some vague indication of a race? I have no idea what "vendor code"
> means here, this "library" has one user. If there is a race please fix
> it. If there's not - please don't add confusing comments.

It is not a race. If vendor's code (or even from command line) marks the interface "down" followed by "up",
queue may be re-enabled. This use case was raised by AI review during the previous submission, which is a 
valid question. So I thought I should capture this in the comment.