Re: [patch RFC 06/29] printk: Protect [un]register_console() with a mutex

[John Ogness]

On 2022-09-14, Sergey Senozhatsky <senozhatsky@xxxxxxxxxxxx> wrote:
> On (22/09/11 00:27), Thomas Gleixner wrote:
> [..]
>> + * console_sem protects the console_drivers list, and also provides
>> + * serialization for access to the entire console driver system.
>> + *
>> + * console_mutex serializes register/unregister. console_sem has to be
>> + * taken for any list manipulation inside the console_mutex locked
>> + * section to keep the console BKL machinery happy.
>>   */
>> +static DEFINE_MUTEX(console_mutex);
>>  static DEFINE_SEMAPHORE(console_sem);
> [..]
>>  /*
>>   * Helper macros to handle lockdep when locking/unlocking console_sem. We use
>>   * macros instead of functions so that _RET_IP_ contains useful information.
>> @@ -2978,17 +3008,21 @@ struct tty_driver *console_device(int *i
>>  void console_stop(struct console *console)
>>  {
>>  	__pr_flush(console, 1000, true);
>> +	console_list_lock();
>>  	console_lock();
>>  	console->flags &= ~CON_ENABLED;
>>  	console_unlock();
>> +	console_list_unlock();
>>  }
>>  EXPORT_SYMBOL(console_stop);
>>  
>>  void console_start(struct console *console)
>>  {
>> +	console_list_lock();
>>  	console_lock();
>>  	console->flags |= CON_ENABLED;
>>  	console_unlock();
>> +	console_list_unlock();
>>  	__pr_flush(console, 1000, true);
>>  }
>>  EXPORT_SYMBOL(console_start);
>
> So the comment says that list lock (console_mutex) is to serialize
> register/unregister, but then we take it in stop/start as well. What
> does list lock protect us against in start/stop? console->flags reader
> (console_is_usable()) does not take list lock, it's called under console
> lock and console->flags writers (console_unregister() and console_stop())
> modify console->flags under console_lock.

Currently all writers to console->flags are holding the
console_lock. However, there are console->flags readers that do _not_
hold the console_lock (register_console, unregister_console,
printk_late_init).

Aside from adding list synchronization, the list lock also provides the
missing console->flags synchronization. Now all console->flags writers
hold the list lock _and_ console_lock. A console->flags reader can hold
either the list lock or the console_lock.

Since console_start and console_stop are console->flags writers, they
also need to take the list lock. I agree that this should be mentioned
in the commit message and code comments.

The follow-up patch in the series only deals with list/flags
readers. Therefore I think the change to console_stop/console_start
belongs in this patch, which focusses on fixing synchronization.

John