[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <84frcx842e.fsf@jogness.linutronix.de>
Date: Mon, 08 Sep 2025 15:36:17 +0206
From: John Ogness <john.ogness@...utronix.de>
To: Mike Galbraith <efault@....de>, Breno Leitao <leitao@...ian.org>, Simon
Horman <horms@...nel.org>, kuba@...nel.org, calvin@...nvd.org
Cc: Pavel Begunkov <asml.silence@...il.com>, Johannes Berg
<johannes@...solutions.net>, paulmck@...nel.org, LKML
<linux-kernel@...r.kernel.org>, netdev@...r.kernel.org,
boqun.feng@...il.com
Subject: Re: netconsole: HARDIRQ-safe -> HARDIRQ-unsafe lock order warning
On 2025-09-06, Mike Galbraith <efault@....de> wrote:
>> The ->write_atomic() callback is intended to perform immediate
>> transmission. It is called with hardware interrupts disabled and is even
>> expected to work from NMI context. If you are not able to implement
>> these requirements, do not implement ->write_atomic(). Implementing some
>> sort of deferrment mechanism is inappropriate. Such a mechanism already
>> exists based on ->write_thread().
>
> Truly atomic packet blasting would be a case of happiness, but barring
> that, deferment is way better than the nothing that's available to both
> RT and !RT+wireless here/now. With a .write_atomic that's really just
> .write_thread, both RT and !RT+wireless managed to successfully send a
> death rattle with the WIP nbcon patch.. a progression for each of them.
Just be aware that ->write_atomic() will be called from _any_ context
(including scheduler and NMI) and a console implementing this callback
must handle the message or the message is lost for that console.
I request that I am added CC to the next incarnation of this
series. Thanks!
John Ogness
Powered by blists - more mailing lists