Re: [PATCH printk v1 07/10] console: add write_atomic interface

From: Sergey Senozhatsky <senozhatsky@chromium.org>
To: John Ogness <john.ogness@linutronix.de>
Cc: Petr Mladek <pmladek@suse.com>,
	Sergey Senozhatsky <senozhatsky@chromium.org>,
	Steven Rostedt <rostedt@goodmis.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	linux-kernel@vger.kernel.org,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Masahiro Yamada <masahiroy@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Nick Desaulniers <ndesaulniers@google.com>,
	Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
	"Paul E. McKenney" <paulmck@kernel.org>,
	Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>,
	Vitor Massaru Iha <vitor@massaru.org>,
	Sedat Dilek <sedat.dilek@gmail.com>,
	Changbin Du <changbin.du@intel.com>
Subject: Re: [PATCH printk v1 07/10] console: add write_atomic interface
Date: Tue, 31 Aug 2021 11:55:01 +0900	[thread overview]
Message-ID: <YS2aBeymxuiW3CIw@google.com> (raw)
In-Reply-To: <20210803131301.5588-8-john.ogness@linutronix.de>

On (21/08/03 15:18), John Ogness wrote:
[..]
> @@ -1993,6 +1993,16 @@ static int console_trylock_spinning(void)
>  	bool spin = false;
>  	unsigned long flags;
>  
> +#ifdef CONFIG_SMP
> +	/*
> +	 * CPUs holding the printk cpulock must not spin on any lock. Even
> +	 * console_trylock() must not be called because its implementation
> +	 * uses spinlocks.
> +	 */
> +	if (atomic_read(&printk_cpulock_owner) == smp_processor_id())
> +		return 0;
> +#endif
> +
>  	if (console_trylock())
>  		return 1;
>  
> @@ -2719,7 +2729,17 @@ static int have_callable_console(void)
>   */
>  static inline int can_use_console(void)
>  {
> -	return cpu_online(raw_smp_processor_id()) || have_callable_console();
> +	int cpu = raw_smp_processor_id();
> +#ifdef CONFIG_SMP
> +	/*
> +	 * CPUs holding the printk cpulock must not spin on any lock.
> +	 * Allowing console usage could call into the spinlocks of the
> +	 * various console drivers.
> +	 */
> +	if (atomic_read(&printk_cpulock_owner) == cpu)
> +		return 0;

I guess the only reason this is done in can_use_console() is
console_flush_on_panic()?

Because otherwise, I think, we can move this check to vprintk_emit().

can_use_console() can be called from preemptible() context. But
if it's called from preemptible() then we know that this is not
printk()/NMI path (but console_device() and friends instead) and
that this CPU is definitely not holding printk CPU lock.

console_trylock_spinning() and console_unlock()->can_use_console()
follow each other

	if (console_trylock_spinning())
		console_unlock();

so a single `atomic_read(&printk_cpulock_owner) == cpu` can suffice.

Now we get to the console_flush_on_panic(), which still calls console_unlock(),
iterates over messages, but when called by the CPU that owns printk_lock,
just skips all the messages. But there is no point in calling console_unlock()
in such a case, we can check if we're printk_cpulock_owner and bail out if so.

Or am I missing something?