Re: [PATCH] net: ti: icssg-prueth: Check return value to avoid a kernel oops

[Benedikt Spranger <b.spranger@linutronix.de>]

On Mon, 24 Mar 2025 16:44:20 +0200
Roger Quadros <rogerq@kernel.org> wrote:

> On 23/03/2025 17:18, Benedikt Spranger wrote:
> > On Sun, 23 Mar 2025 09:19:35 +0200
> > Roger Quadros <rogerq@kernel.org> wrote:
> >   
> >> Did you actually get a kernel oops?  
> > Yes. And I would like to attach the kernel output, but I do not have
> > access to the board ATM.
> >   
> >> If yes, which part of code produces the oops.  
> > I get an NULL pointer dereference in is_multicast_ether_addr().
> > It happens here:
> > 
> >     u32 a = *(const u32 *)addr;  
> 
> But this should not happen. Because ndev->addr (pointer) should not
> be zero. Driver allocated ndev with alloc_etherdev_mq() which
> allocates memory for ndev->addr using dev_addr_init(dev)).
Emphasis on *should* :)
OK, got your point. Dig deeper into that.

> >> Even if it fails we do set a random MAC address and do not return
> >> error. So above statement is false.  
> > I doubt that. of_get_ethdev_address() do not set a random MAC
> > address in case of a failure. It simply returns -ENODEV. Since
> > is_valid_ether_addr() fails with a NULL pointer dereference in
> > is_multicast_ether_addr() on the other hand, no random MAC address
> > is set.   
> 
> What I meant was we set random address using eth_hw_addr_random().
But that happens after the failing check. So evaluating the return of
of_get_ethdev_address() seem to be a good thing in the first place.

I my understanding (for now) it is nessesary to check both: the return
of of_get_ethdev_address() *and* !is_valid_ether_addr(). If any of these
checks fail eth_hw_addr_random() should be called and therefore a
random MAC address be set.

Regards
    Bene