Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <1602673931-28782-1-git-send-email-sumit.garg@linaro.org>
 <1602673931-28782-2-git-send-email-sumit.garg@linaro.org> <fbf6bfecf344a06af94e17ccabb817dd@kernel.org>
 <CAFA6WYPEw5x_3zDZSnkL23YjFWbF=VkUKefM6Ts46JLVaj2sDA@mail.gmail.com>
 <f3596e3ed70737d36f72056827d7d441@kernel.org> <CAFA6WYMEJb0o+PMmvLh1wDhZ+M4rChP3a1cNNoEsoMS1kDWwKA@mail.gmail.com>
 <20201020122535.paym32gksuhcqtwq@holly.lan> <abe358694e06a6076fb5838d1231eee6@kernel.org>
In-Reply-To: <abe358694e06a6076fb5838d1231eee6@kernel.org>
From:   Sumit Garg <sumit.garg@linaro.org>
Date:   Wed, 21 Oct 2020 10:52:10 +0530
Message-ID: <CAFA6WYNG3EMuDSkp7P9s7_pfw0jAC+-CZcH6f6sRnpvLsyQE6Q@mail.gmail.com>
Subject: Re: [PATCH v5 1/5] arm64: Add framework to turn IPI as NMI
To:     Marc Zyngier <maz@kernel.org>,
        Daniel Thompson <daniel.thompson@linaro.org>
Cc:     Catalin Marinas <catalin.marinas@arm.com>,
        Will Deacon <will@kernel.org>,
        linux-arm-kernel <linux-arm-kernel@lists.infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Jason Cooper <jason@lakedaemon.net>,
        Mark Rutland <mark.rutland@arm.com>,
        julien.thierry.kdev@gmail.com,
        Douglas Anderson <dianders@chromium.org>,
        Jason Wessel <jason.wessel@windriver.com>,
        Masayoshi Mizuma <msys.mizuma@gmail.com>,
        ito-yuichi@fujitsu.com, kgdb-bugreport@lists.sourceforge.net,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, 20 Oct 2020 at 18:02, Marc Zyngier <maz@kernel.org> wrote:
>
> On 2020-10-20 13:25, Daniel Thompson wrote:
> > On Tue, Oct 20, 2020 at 04:52:43PM +0530, Sumit Garg wrote:
>
> [...]
>
> >> So in general, IPI as a normal IRQ is still useful for debugging but
> >> it can't debug a core which is stuck in deadlock with interrupts
> >> disabled.
> >>
> >> And since we choose override default implementations for pseudo NMI
> >> support, we need to be backwards compatible for platforms which don't
> >> possess pseudo NMI support.
> >
> > Do the fallback implementations require a new IPI? The fallbacks
> > could rely on existing mechanisms such as the smp_call_function
> > family.
>
> Indeed. I'd be worried of using that mechanism for NMIs, but normal
> IPIs should stick to the normal cross-call stuff.

Yes, the fallback implementations could rely on existing cross-call
stuff but current framework only allows this fallback choice at
compile time and to make this choice at runtime, we need additional
framework changes like allowing arch_trigger_cpumask_backtrace() to
return a boolean in order to switch to fallback mode, similar would be
the case for kgdb.

I think this should be doable but I am still not sure regarding the
advantages of using existing IPI vs new IPI (which we will already use
in case of pseudo NMI support) as normal IRQ.

>
> The jury is still out on why this is a good idea, given that it
> doesn't work in the only interesting case (deadlocked CPUs).

I think CPU soft-lockups (interrupts enabled) is an interesting case
to debug as well.

-Sumit

>
>           M.
> --
> Jazz is not dead. It just smells funny...