WARNING: CPU: 1 PID: 1 at net/core/netpoll.c:370 netpoll_send_skb+0x1fc/0x20c at boot when netconsole is enabled (kernel v6.9-rc5, v6.8.7, sungem, PowerMac G4 DP)

[Jakub Kicinski]

On Wed, 8 May 2024 10:55:05 +0200 Erhard Furtner wrote:
> I could do that with the explanation you stated. But should any
> further questions arise in this process I would also lack the
> technical background to deal with them. ;)

Alright, submitted :)

> I also noticed a similar #ifdef CONFIG_NET_POLL_CONTROLLER logic shows up in
> many network drivers, e.g. net/ethernet/realtek/8139too.c:
> 
> #ifdef CONFIG_NET_POLL_CONTROLLER
> static void rtl8139_poll_controller(struct net_device *dev);
> #endif
> [...]
> #ifdef CONFIG_NET_POLL_CONTROLLER
> /*
>  * Polling receive - used by netconsole and other diagnostic tools
>  * to allow network i/o with interrupts disabled.
>  */
> static void rtl8139_poll_controller(struct net_device *dev)
> {
>         struct rtl8139_private *tp = netdev_priv(dev);
>        	const int irq = tp->pci_dev->irq;
> 
>        	disable_irq_nosync(irq);
>        	rtl8139_interrupt(irq, dev);
>        	enable_irq(irq);
> }
> #endif
> [...]
> #ifdef CONFIG_NET_POLL_CONTROLLER
>        	.ndo_poll_controller    = rtl8139_poll_controller,
> #endif
> 
> 
> Should it be removed here too? This would be more cards I can test.
> So far I only see this on my G4 and I think something similar on an
> old Pentium4 box I no longer have. 

That one looks legit. Note the _nosync() which solves the immediate
IRQ masking / deadlock issue. And the rtl8139_interrupt() function
actually does the packet cleanup in case of 8139too.