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

From: Jakub Kicinski <kuba@kernel.org>
To: Alexander Lobakin <aleksander.lobakin@intel.com>
Cc: "David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>,
	Paolo Abeni <pabeni@redhat.com>,
	Maciej Fijalkowski <maciej.fijalkowski@intel.com>,
	Larysa Zaremba <larysa.zaremba@intel.com>,
	Yunsheng Lin <linyunsheng@huawei.com>,
	Alexander Duyck <alexanderduyck@fb.com>,
	Jesper Dangaard Brouer <hawk@kernel.org>,
	"Ilias Apalodimas" <ilias.apalodimas@linaro.org>,
	<netdev@vger.kernel.org>, <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH RFC net-next v2 7/7] net: skbuff: always try to recycle PP pages directly when in softirq
Date: Thu, 20 Jul 2023 12:20:15 -0700	[thread overview]
Message-ID: <20230720122015.1e7efc21@kernel.org> (raw)
In-Reply-To: <b3884ff9-d903-948d-797a-1830a39b1e71@intel.com>

On Thu, 20 Jul 2023 20:13:07 +0200 Alexander Lobakin wrote:
> IOW, it reports we're in softirq no bloody matter if interrupts are
> enabled or not. Either I did something wrong or the entire in_*irq()
> family, including interrupt_context_level(), doesn't protect from
> anything at all and doesn't work the way that most devs expect it to work?
> 
> (or was it just me? :D)
> 
> I guess the only way to be sure is to always check irqs_disabled() when
> in_softirq() returns true.

We can as well check
	(in_softirq() && !irqs_disabled() && !in_hardirq())
?

The interrupt_context_level() thing is fairly new, I think.
Who knows what happens to it going forward...

> >> Right now page pool only supports BH and process contexts. IOW the
> >> "else" branch of if (in_softirq()) in page pool is expecting to be
> >> in process context.
> >>
> >> Supporting hard irq would mean we need to switch to _irqsave() locking.
> >> That's likely way too costly.
> >>
> >> Or stash the freed pages away and free them lazily.
> >>
> >> Or add a lockdep warning and hope nobody will ever free a page-pool
> >> backed skb from hard IRQ context :)  
> > 
> > I told you under the previous version that this function is not supposed
> > to be called under hardirq context, so we don't need to check for it :D
> > But I was assuming nobody would try to do that. Seems like not really
> > (netcons) if you want to sanitize this...

netcons or anyone who freed socket-less skbs from hardirq.
Until pp recycling was added freeing an skb from hardirq was legal,
AFAICT.