netdev - Re: [PATCH net 2/2] net: gro: Let the timeout timer expire in softirq context with `threadirqs'

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <871rcdzx8w.fsf@nanos.tec.linutronix.de>
Date:   Wed, 17 Mar 2021 15:10:55 +0100
From:   Thomas Gleixner <tglx@...utronix.de>
To:     Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
        Eric Dumazet <edumazet@...gle.com>
Cc:     netdev <netdev@...r.kernel.org>,
        "David S. Miller" <davem@...emloft.net>
Subject: Re: [PATCH net 2/2] net: gro: Let the timeout timer expire in softirq context with `threadirqs'

On Thu, Apr 16 2020 at 15:59, Sebastian Andrzej Siewior wrote:
> any comments from the timer department?

Yes.

> On 2019-11-27 18:37:19 [+0100], To Eric Dumazet wrote:
>> On 2019-11-27 09:11:40 [-0800], Eric Dumazet wrote:
>> > Resent in non HTML mode :/
>> don't worry, mutt handles both :)
>> 
>> > Long story short, why hrtimer are not by default using threaded mode
>> > in threadirqs mode ?
>> 
>> Because it is only documented to thread only interrupts. Not sure if we
>> want change this.
>> In RT we expire most of the hrtimers in softirq context for other
>> reasons. A subset of them still expire in hardirq context.
>>
>> > Idea of having some (but not all of them) hard irq handlers' now being
>> > run from BH mode,
>> > is rather scary.
>> 
>> As I explained in my previous email: All IRQ-handlers fire in
>> threaded-mode if enabled. Only the hrtimer is not affected by this
>> change.
>> 
>> > Also, hrtimers got the SOFT thing only in 4.16, while the GRO patch
>> > went in linux-3.19
>> > 
>> > What would be the plan for stable trees ?
>> No idea yet. We could let __napi_schedule_irqoff() behave like
>> __napi_schedule(). 

It's not really a timer departement problem. It's an interrupt problem.

With force threaded interrupts we don't call the handler with interrupts
disabled. What sounded a good idea long ago, is actually bad.

See https://lore.kernel.org/r/87eegdzzez.fsf@nanos.tec.linutronix.de

Any leftover issues on a RT kernel are a different story, but for !RT
this is the proper fix.

I'll spin up a proper patch and tag it for stable...

Thanks,

        tglx