Re: [PATCH net-next 0/3] net: ethernet: cortina: Use page_pool for Gemini RX buffers

[Jakub Kicinski <kuba@kernel.org>]

On Thu, 24 Sep 2026 09:40:24 +0200 Linus Walleij wrote:
> > > Move the Gemini RX buffer management to a parent-owned page_pool.  
> >
> > TBH sounds like weird piece of HW. IDK if we want to burden the generic
> > page pool code with weird, 20+ yo drivers+HW :/ Well, I know that
> > I definitely don't, I don't know if others agree :)  
> 
> The current AI reviews actively encourages people to go and fix
> really old bugs,  or code that hasn't seen a lot of love in recent
> years, pre-existing issues you know.
> 
> That's how the patch came about that this patch is a "let's do it
> properly in that case" reply to:
> https://lore.kernel.org/netdev/20260917192835.58126-1-mhun512@gmail.com/
> 
> We have had, for years, a debug print in this driver:
> 
>                 /* Freeq pointers are one page off */
>                 gpage = gmac_get_queue_page(geth, port, mapping + PAGE_SIZE);
>                 if (!gpage) {
>                         dev_err(geth->dev, "could not find page\n");
>                         continue;
> 
> and it seems this patch set actually finally fixes that bug that was
> bugging me (heh) for years and years.
> 
> So there is a bit of incentive to actually fix it.
> 
> (I also have actual users of the code, it's not therapeutic coding.)

Understood, but none of this really necessitate the page pool.
Supporting weird drivers makes it much harder for us to create
reasonable uAPI and protect these things with fine-granularity
locks.