Re: CPU Hotplug rework

["Srivatsa S. Bhat" <srivatsa.bhat@linux.vnet.ibm.com>]

On 04/10/2012 09:16 PM, Paul E. McKenney wrote:

> On Tue, Apr 10, 2012 at 07:11:50PM +0530, Srivatsa S. Bhat wrote:


[This is a quick reply to give the links you requested. I'll reply to the
 other things after I read what you wrote more thoroughly.]

>> Why does my approach help?
> 
> At this point, I must confess that I have lost track of exactly what
> your approach is...


The same old "incomplete" patchset ;-)
(Note that the patch 1/3 is complete. The "incomplete" tag is just because
it is followed by changes only to powerpc (2/3) and sparc (3/3), while
actually, many other places need to be changed. But the first patch in the
series is definitely in full form.

https://lkml.org/lkml/2012/3/1/39
https://lkml.org/lkml/2012/3/1/40
https://lkml.org/lkml/2012/3/1/41

> 
>> It ensures that do_setup() will never occur in parallel with CPU hotplug,
>> at any time. Hence the individual notifiers need not watch their back -
>> they can continue to be non-reentrant and still everything will work fine
>> because we fix it at the callback registration level itself.
>>
>> Honestly, I wrote this patchset to fix issues opened up by the async booting
>> patch[1]. That patch caused boot failures in powerpc [2] because of CPU
>> Hotplug notifier races. And I believe the solution I proposed will fix it.
>>
>> Without the async booting patch, this was more or less a theoretical race.
>> That patch made it not only real but also severe enough to cause boot
>> failures.
>>
>> So, if the async booting design is not being pushed any further, then I
>> guess we can simply ignore this theoretical race altogether and focus on
>> more important issues (I am totally OK with that) ... and possibly revisit
>> this race whenever it bites us again ;-)
>>
>> What do you think?
>>
>> [1]. http://thread.gmane.org/gmane.linux.kernel/1246209
>> [2]. http://thread.gmane.org/gmane.linux.kernel.next/20726/focus=20757
> 
> Neither of the above two URLs points to a patch,


??
Well, the first one points to the async booting patch and the second one points
to a verbal root-cause analysis of the boot failure on powerpc, caused by that
patch.

Let me give equivalent links from lkml.org:
[1]. https://lkml.org/lkml/2012/1/31/286
[2]. https://lkml.org/lkml/2012/2/13/383

 
Regards,
Srivatsa S. Bhat