Re: [PATCH printk 4/5] printk: remove logbuf_lock, add syslog_lock

[John Ogness]

On 2020-09-24, Sergey Senozhatsky <sergey.senozhatsky@xxxxxxxxx> wrote:
> [..]
>>  void printk_safe_flush_on_panic(void)
>>  {
>> -	/*
>> -	 * Make sure that we could access the main ring buffer.
>> -	 * Do not risk a double release when more CPUs are up.
>> -	 */
>> -	if (raw_spin_is_locked(&logbuf_lock)) {
>> -		if (num_online_cpus() > 1)
>> -			return;
>> -
>> -		debug_locks_off();
>> -		raw_spin_lock_init(&logbuf_lock);
>
> If we panic() with syslog_lock locked then we need to zap syslog_lock
> and disable debug_locks for kmsg_dump() to succeed. Otherwise panic()
> may deadlock:
>
> 	syslog_lock_irq()
> 	-> NMI
> 		panic()
> 		...
> 			printk_safe_flush_on_panic();
> 			kmsg_dump(KMSG_DUMP_PANIC);	// syslog_lock_irqsave(flags)
> 		...
>
> Did I miss the point where we make sure that kmsg_dump() will not
> deadlock?

You are correct. Thanks.

@clear_seq should either become atomic64 or kmsg_dump() should start
from the oldest available message (seq=0).

John Ogness