On Wed, 3 Dec 2025 08:33:53 -0500 "Michael S. Tsirkin" <[email protected]> wrote:
> On Tue, Dec 02, 2025 at 03:55:39PM +0000, Simon Horman wrote:
> > On Wed, Nov 26, 2025 at 01:35:38PM -0600, Daniel Jurgens wrote:
> >
> > ...
> >
> > > @@ -6005,6 +6085,11 @@ static void parse_ip4(struct iphdr *mask, struct
> > > iphdr *key,
> > > mask->tos = l3_mask->tos;
> > > key->tos = l3_val->tos;
> > > }
> > > +
> > > + if (l3_mask->proto) {
> > > + mask->protocol = l3_mask->proto;
> > > + key->protocol = l3_val->proto;
> > > + }
> > > }
> >
> > Hi Daniel,
> >
> > Claude Code with review-prompts flags an issue here,
> > which I can't convince myself is not the case.
> >
> > If parse_ip4() is called for a IP_USER_FLOW, which use ethtool_usrip4_spec,
> > as does this function, then all is well.
> >
> > However, it seems that it may also be called for TCP_V4_FLOW and UDP_V4_FLOW
> > flows, in which case accessing .proto will overrun the mask and key which
> > are actually struct ethtool_tcpip4_spec.
> >
> > https://netdev-ai.bots.linux.dev/ai-review.html?id=51d97b85-5ca3-4cb8-a96a-0d6eab5e7196#patch-10
>
>
> Oh I didn't know about this one. Is there any data on how does it work?
> Which model/prompt/etc?
I'm not actually sure if the netdev usage is written up somewhere?
The automation is running claude, but (hopefully) there's nothing specific to
claude in the prompts, it's just what I've been developing against.
The prompts are:
https://github.com/masoncl/review-prompts
Jakub also wired up semcode indexing, which isn't required but does
make it easier for claude to find code:
https://github.com/facebookexperimental/semcode
I'm still working on docs and easy setup for semcode and the review prompts,
but please feel free to send questions.
-chris
