Received-SPF: pass (google.com: domain of linux-kernel+bounces-71557-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Message-ID: <759adfbf-b238-44fe-b538-418b43701539@redhat.com>
Date: Mon, 19 Feb 2024 10:07:35 -0500
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH printk v2 26/26] lockdep: Mark emergency section in
 lockdep splats
Content-Language: en-US
To: John Ogness <john.ogness@linutronix.de>, Petr Mladek <pmladek@suse.com>
Cc: Sergey Senozhatsky <senozhatsky@chromium.org>,
 Steven Rostedt <rostedt@goodmis.org>, Thomas Gleixner <tglx@linutronix.de>,
 linux-kernel@vger.kernel.org, Peter Zijlstra <peterz@infradead.org>,
 Ingo Molnar <mingo@redhat.com>, Will Deacon <will@kernel.org>,
 Boqun Feng <boqun.feng@gmail.com>
References: <20240218185726.1994771-1-john.ogness@linutronix.de>
 <20240218185726.1994771-27-john.ogness@linutronix.de>
 <6d8f5126-3595-4abd-9b05-b19a659d1cea@redhat.com>
 <878r3gbtev.fsf@jogness.linutronix.de>
From: Waiman Long <longman@redhat.com>
In-Reply-To: <878r3gbtev.fsf@jogness.linutronix.de>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit


On 2/19/24 06:11, John Ogness wrote:
> On 2024-02-18, Waiman Long <longman@redhat.com> wrote:
>> On 2/18/24 13:57, John Ogness wrote:
>> lockdep.c has multiple functions that print stuff to the console, like
>>
>>    - print_circular_bug_header()
>>    - print_bad_irq_dependency()
>>    - print_deadlock_bug()
>>    - print_collision()
>>    - print_usage_bug()
>>    - print_irq_inversion_bug()
>>    - print_lock_invalid_wait_context()
>>    - print_lock_nested_lock_not_held()
>>    - print_unlock_imbalance_bug()
>>    - print_lock_contention_bug()
>>    - print_freed_lock_bug()
>>    - print_held_locks_bug()
>>    - lockdep_rcu_suspicious()
>>
>> So what is special about print_usage_bug() that it needs this
>> emergency treatment but not the other ones?
> I do not expect to be able to identify all "emergency printing" paths in
> the kernel from the beginning. This series initially marks some sections
> that are IMHO interesting for the feature.
That is what I like to see in the changelog. I am aware that this patch 
is probably not complete, but you need to set the right expectation that 
similar changes will have to be done elsewhere in lockdep to complete 
the change. We can make the other necessary changes after this patch 
series have been merged. It also helps if you can document what 
undesirable thing may happen if printk() is called without setting the 
emergency mode.
>   
>
> As you are implying, for lockdep probably all printing should be
> considered emergency. Is it preferred to place the markers outside the
> high-level print functions, for example:
>
> diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
> index 00465373d358..7a4e4f4a9156 100644
> --- a/kernel/locking/lockdep.c
> +++ b/kernel/locking/lockdep.c
> @@ -2182,10 +2182,12 @@ check_noncircular(struct held_lock *src, struct held_lock *target,
>   			*trace = save_trace();
>   		}
>   
> +		nbcon_cpu_emergency_enter();
>   		if (src->class_idx == target->class_idx)
>   			print_deadlock_bug(current, src, target);
>   		else
>   			print_circular_bug(&src_entry, target_entry, src, target);
> +		nbcon_cpu_emergency_exit();
>   	}
>   
>   	return ret;
>
> Or is it preferred to put them directly around the various pr_warn()
> blocks (as the patch in this series is doing)?

There are pros and cons for both. It will depend on how expensive is the 
nbcon_cpu_emergency_{enter|exit}() call as printing won't happen if 
lockdep is turned off somehow. Since lockdep is for debugging and 
efficiency isn't that important, putting the emergency enter/exit 
markers outside the high level print functions will make it a bit easier 
to read.

My 2 cents.

Cheers,
Longman


>
> John Ogness
>