Re: [PATCH RFC net-next v2 7/7] net: skbuff: always try to recycle PP pages directly when in softirq

[Jakub Kicinski <kuba@kernel.org>]

On Thu, 20 Jul 2023 21:33:40 +0200 Alexander Lobakin wrote:
> > We can as well check
> > 	(in_softirq() && !irqs_disabled() && !in_hardirq())
> > ?  
> 
> Yes, something like that. Messy, but I see no other options...
> 
> So, I guess you want to add an assertion to make sure that we're *not*
> in this:
> 
> in_hardirq() || irqs_disabled()
> 
> Does this mean that after it's added, my patch is sane? :p

Well... it's acceptable. Make sure you add a good, informative
but concise comment :)

> > The interrupt_context_level() thing is fairly new, I think.
> > Who knows what happens to it going forward...  
> 
> Well, it counts the number of active hard interrupts, but doesn't take
> into account that if there are no hardirqs we can still disable them
> manually. Meh.
> Should I try to send a patch for it? :D

Depends on how you like to send your time :)

> > netcons or anyone who freed socket-less skbs from hardirq.
> > Until pp recycling was added freeing an skb from hardirq was legal,
> > AFAICT.  
> 
> I don't think so. Why do we have dev_kfree_skb_any() then? It checks for
> 
> in_hardirq() || irqs_disabled()
> 
> and if it's true, defers the skb to process it by backlog task.
> "Regular" skb freeing functions don't do that. The _any() variant lives
> here for a long time IIRC, so it's not something recent.

Drivers (or any other users of dev_kfree_skb_any()) should be fine. 
I'm only paranoid about some unknown bits of code which thought they
can be clever and call kfree_skb() directly, as long as !skb->sk.

But if you add the hard irq checks to your patch then you're strictly
safer than the existing code. Hopefully the checks are not too
expensive.