[lkml]   [2012]   [Apr]   [19]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH] clocksource, prevent overflow in clocksource_cyc2ns

On 04/19/2012 08:52 AM, Thomas Gleixner wrote:
> On Thu, 19 Apr 2012, Thomas Gleixner wrote:
>> On Wed, 18 Apr 2012, John Stultz wrote:
>>> On 04/18/2012 04:59 PM, Prarit Bhargava wrote:

>> No. The show_state() part prints into the buffer. But it's not
>> guaranteed that the buffer is flushed right away. It could be flushed
>> later as well in a different context. And of course the flush code
>> runs with interrupts disabled and dumping out a gazillion of lines
>> over serial will cause the same hickup. Just planting random
>> touch_watchdog() calls into the code is not the right approach,
>> really.
>> We should think about the reasons why we have interrupts disabled for
>> so much time. Is that really, really necessary ?

In the case of the sysrq-t, I would argue that it is. The whole point behind
the sysrq-t is that we're capturing the *current* state of the system. Having
that output effected by interrupts seems like a bad idea.

> I'm not against making the clocksource code more robust, but I don't
> want to add crap there just to cope with complete madness elsewhere.

Maybe I came off the wrong way but I completely agree with that sentiment. Like
yourself, I'm looking for a correct fix rather than a fast fix.

Sorry that I haven't provided any debug info but I'm still in the gathering data
stage atm. It was just John's ping that made me "brain dump" the current info I


 \ /
  Last update: 2012-04-19 15:09    [W:0.089 / U:4.548 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site