Re: [PATCH net v3] netpoll: Fix deadlock in memory allocation under spinlock

[Breno Leitao]

On Thu, Oct 16, 2025 at 04:23:23PM -0700, Jakub Kicinski wrote:
> On Tue, 14 Oct 2025 09:37:50 -0700 Breno Leitao wrote:
> > +	while (1) {
> > +		spin_lock_irqsave(&skb_pool->lock, flags);
> > +		if (skb_pool->qlen >= MAX_SKBS)
> > +			goto unlock;
> > +		spin_unlock_irqrestore(&skb_pool->lock, flags);
> 
> No need for the lock here:
> 
> 	if (READ_ONCE(..) >= MAX_SKBS)
> 
> >  		skb = alloc_skb(MAX_SKB_SIZE, GFP_ATOMIC);
> >  		if (!skb)
> > -			break;
> > +			return;
> >  
> > +		spin_lock_irqsave(&skb_pool->lock, flags);
> > +		if (skb_pool->qlen >= MAX_SKBS)
> > +			/* Discard if len got increased (TOCTOU) */
> > +			goto discard;
> 
> Not sure this is strictly needed, the number 32 (MAX_SKBS) was not
> chosen super scientifically anyway, doesn't matter if we go over a
> little. 

Agree. I will take this approach them, since it is not going to hurt at
all.

Thanks for the review,
--breno