svn commit: r211176 - in head/sys: amd64/amd64 i386/i386

[Attilio Rao]

2010/8/11 John Baldwin <jhb at freebsd.org>:
> Attilio Rao wrote:
>>
>> 2010/8/11 John Baldwin <jhb at freebsd.org>:
>>>
>>> Attilio Rao wrote:
>>>>
>>>> Author: attilio
>>>> Date: Wed Aug 11 10:51:27 2010
>>>> New Revision: 211176
>>>> URL: http://svn.freebsd.org/changeset/base/211176
>>>>
>>>> Log:
>>>>  IPI handlers may run generally with interrupts disabled because they
>>>>  are served via an interrupt gate.
>>>>   However, that doesn't explicitly prevent preemption and thread
>>>>  migration thus scheduler pinning may be necessary in some handlers.
>>>>  Fix that.
>>>>   Tested by:  gianni
>>>>  MFC after:    1 month
>>>
>>> Actually that does prevent preemption if you do not call any code that
>>> would
>>> schedule a thread.  I think this change is all safe to revert.
>>
>> Do you recall, then, why lapic_handle_timer() does critical section?
>> It seems to be catered by interrupt gate as well, and I don't see any
>> point re-enabling them explicitly.
>
> Because hardclock() explicitly calls sched_add() via swi_sched() when
> scheduling the softclock swi.  The critical section there is just to ensure
> that the preemption to softclock happens after all of the clock interrupt
> handlers have finished.  However, PCPU_GET() does not call sched_add(), so
> it does not need to be protected.

Oh, sorry, was looking at the -CURRENT code which had no explicit call
to hardclock (it uses the new timer infrastructure, so that was not so
obvious).

I understand now, so I will rever the change asap.

Thanks,
Attilio


-- 
Peace can only be achieved by understanding - A. Einstein