From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <stable-owner@vger.kernel.org>
Received: from mail-pl0-f67.google.com ([209.85.160.67]:40850 "EHLO
        mail-pl0-f67.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
        with ESMTP id S1751644AbeERCHK (ORCPT
        <rfc822;stable@vger.kernel.org>); Thu, 17 May 2018 22:07:10 -0400
Date: Fri, 18 May 2018 11:07:05 +0900
From: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>
To: Petr Mladek <pmladek@suse.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>,
        Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>,
        linux-kernel@vger.kernel.org, "4 . 13+" <stable@vger.kernel.org>
Subject: Re: [PATCH] printk/nmi: Prevent deadlock when serializing NMI
 backtraces
Message-ID: <20180518020705.GB1160@jagdpanzerIV>
References: <20180517143903.19339-1-pmladek@suse.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180517143903.19339-1-pmladek@suse.com>
Sender: stable-owner@vger.kernel.org
List-ID: <stable.vger.kernel.org>

On (05/17/18 16:39), Petr Mladek wrote:
> 
> CPU0			CPU1			CPU2
> 
> printk()
>   vprintk_emit()
>     spin_lock(&logbuf_lock)
> 
> 						trigger_all_cpu_backtrace()
> 						  raise()
> 
> 			nmi_enter()
> 			  printk_nmi_enter()
> 			    if (this_cpu_read(printk_context)
> 			      & PRINTK_SAFE_CONTEXT_MASK)
> 			      // false
> 			    else
> 			      // looks safe to use printk_deferred()
> 			      this_cpu_or(printk_context,
> 				PRINTK_NMI_DEFERRED_CONTEXT_MASK);
> 
> 			  nmi_cpu_backtrace()
> 			    arch_spin_lock(&lock);
> 			      show_regs()
> 
> nmi_enter()
>   nmi_cpu_backtrace()
>     arch_spin_lock(&lock);
> 
> 			      printk()
> 				vprintk_func()
> 				  vprintk_deferred()
> 				    vprintk_emit()
> 				      spin_lock(&logbuf_lock)
> 
> DEADLOCK: between &logbuf_lock from vprintk_emit() and
> 		  &lock from nmi_cpu_backtrace().
> 
> CPU0			CPU1
> lock(logbuf_lock)	lock(lock)
>   lock(lock)		  lock(logbuf_lock)
> 
[..]
> Signed-off-by: Petr Mladek <pmladek@suse.com>

This is a pretty cool find!

Acked-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>

> -	if ((this_cpu_read(printk_context) & PRINTK_SAFE_CONTEXT_MASK) &&
> -	    raw_spin_is_locked(&logbuf_lock)) {
> +	if (raw_spin_is_locked(&logbuf_lock))
>  		this_cpu_or(printk_context, PRINTK_NMI_CONTEXT_MASK);
> -	} else {
> +	else
>  		this_cpu_or(printk_context, PRINTK_NMI_DEFERRED_CONTEXT_MASK);
> -	}

A question - can we switch to a bitwise OR?

if (this_cpu_read(printk_context) & PRINTK_SAFE_CONTEXT_MASK) ||
	raw_spin_is_locked(&logbuf_lock)

just to check per-CPU `printk_context' first and only afterwards
access the global `logbuf_lock'. printk_nmi_enter() happens on
every CPU, so maybe we can avoid some overhead by checking the
local per-CPU data first.

	-ss