Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Michael Ellerman <mpe@ellerman.id.au>
To:     Leonardo Bras <leonardo@linux.ibm.com>,
        Benjamin Herrenschmidt <benh@kernel.crashing.org>,
        Paul Mackerras <paulus@samba.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Enrico Weigelt <info@metux.net>,
        Thomas Gleixner <tglx@linutronix.de>,
        Allison Randal <allison@lohutok.net>,
        Christophe Leroy <christophe.leroy@c-s.fr>,
        Leonardo Bras <leonardo@linux.ibm.com>,
        Nicholas Piggin <npiggin@gmail.com>
Cc:     linuxppc-dev@lists.ozlabs.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/1] powerpc/crash: Use NMI context for printk after crashing other CPUs
In-Reply-To: <20200406174058.686436-1-leonardo@linux.ibm.com>
References: <20200406174058.686436-1-leonardo@linux.ibm.com>
Date:   Wed, 08 Apr 2020 22:13:31 +1000
Message-ID: <87y2r6jhuc.fsf@mpe.ellerman.id.au>
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Leonardo Bras <leonardo@linux.ibm.com> writes:
> Currently, if printk lock (logbuf_lock) is held by other thread during
> crash, there is a chance of deadlocking the crash on next printk, and
> blocking a possibly desired kdump.
>
> After sending IPI to all other CPUs, make printk enter in NMI context,
> as it will use per-cpu buffers to store the message, and avoid locking
> logbuf_lock.
>
> Signed-off-by: Leonardo Bras <leonardo@linux.ibm.com>
> ---
>  arch/powerpc/kexec/crash.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/arch/powerpc/kexec/crash.c b/arch/powerpc/kexec/crash.c
> index d488311efab1..9b73e3991bf4 100644
> --- a/arch/powerpc/kexec/crash.c
> +++ b/arch/powerpc/kexec/crash.c
> @@ -115,6 +115,7 @@ static void crash_kexec_prepare_cpus(int cpu)

Added context:

	printk(KERN_EMERG "Sending IPI to other CPUs\n");

	if (crash_wake_offline)
		ncpus = num_present_cpus() - 1;

>  
>  	crash_send_ipi(crash_ipi_callback);
>  	smp_wmb();
> +	printk_nmi_enter();
  
Why did you decide to put it there, rather than at the start of
default_machine_crash_shutdown() like I did?

The printk() above could have already deadlocked if another CPU is stuck
with the logbuf lock held.

cheers