lkml.org 
[lkml]   [2011]   [Jun]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
DateMon, 06 Jun 2011 18:45:23 +0200
FromArne Jansen <>
SubjectRe: [debug patch] printk: Add a printk killswitch to robustify NMI watchdog messages
On 06.06.2011 18:38, Arne Jansen wrote:
> On 06.06.2011 18:17, Ingo Molnar wrote:
>>
>> * Peter Zijlstra<peterz@infradead.org> wrote:
>>
>>> On Mon, 2011-06-06 at 18:08 +0200, Ingo Molnar wrote:
>>>> * Peter Zijlstra<peterz@infradead.org> wrote:
>>>>
>>>>> On Mon, 2011-06-06 at 17:52 +0200, Ingo Molnar wrote:
>>>>>> * Peter Zijlstra<peterz@infradead.org> wrote:
>>>>>>
>>>>>>> Needs more staring at, preferably by someone who actually
>>>>>>> understands that horrid mess :/ Also, this all still doesn't make
>>>>>>> printk() work reliably while holding rq->lock.
>>>>>>
>>>>>> So, what about my suggestion to just *remove* the wakeup from there
>>>>>> and use the deferred wakeup mechanism that klogd uses.
>>>>>>
>>>>>> That would make printk() *visibly* more robust in practice.
>>>>>
>>>>> That's currently done from the jiffy tick, do you want to effectively
>>>>> delay releasing the console_sem for the better part of a jiffy?
>>>>
>>>> Yes, and we already do it in some other circumstances.
>>>
>>> We do?
>>
>> Yes, see the whole printk_pending logic, it delays:
>>
>> wake_up_interruptible(&log_wait);
>>
>> to the next jiffies tick.
>>
>>>> Can you see
>>>> any problem with that? klogd is an utter slowpath anyway.
>>>
>>> but console_sem isn't klogd. We delay klogd and that's perfectly
>>> fine, but afaict we don't delay console_sem.
>>
>> But console_sem is really a similar special case as klogd. See, it's
>> about a *printk*. That's rare by definition.
>>
>> If someone on the console sees it he'll be startled by at least 10
>> msecs ;-) So delaying the wakeup to the next jiffy really fits into
>> the same approach as we already do with&log_wait, hm?
>
> As long as it doesn't scramble the order of the messages, the delay
> imho doesn't matter even in very printk-heavy debugging sessions.

And, as important, doesn't reduce the throughput of printk. Having only
100 wakeups/s sounds like the throughput is limited to 100xsizeof(ring
buffer).

>
>>
>> This would solve a real nightmare that has plagued us ever since
>> printk() has done wakeups directly - i.e. like forever.
>>
>> Thanks,
>>
>> Ingo
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/



\
 
 \ /
  Last update: 2011-06-06 18:47    [from the cache]
©2003-2010