Re: [PATCH 1/1] Xen ARINC 653 Scheduler (updated to add support for CPU pools)

[Keir Fraser <keir.fraser@eu.citrix.com>]

On 16/06/2010 19:03, "Kathy Hadley" <Kathy.Hadley@dornerworks.com> wrote:

> That sounds reasonable to me.

Fixed as of changeset 21626, in the staging tree
(http://xenbits.xensource.com/staging/xen-unstable.hg).

 K.

> Kathy
> 
>> -----Original Message-----
>> From: Keir Fraser [mailto:keir.fraser@eu.citrix.com]
>> Sent: Wednesday, June 16, 2010 12:50 PM
>> To: Kathy Hadley; George Dunlap
>> Cc: xen-devel@lists.xensource.com; Juergen Gross
>> Subject: Re: [Xen-devel] [PATCH 1/1] Xen ARINC 653 Scheduler (updated
>> to add support for CPU pools)
>> 
>> Oh, I see. Well, the cause is that the
>> common/schedule.c:sched_adjust_global() is broken. But, what should it
>> actually do, given that multiple schedulers of same or differing types
>> may
>> exist in a system now? Perhaps the sysctl should take a cpupool id, to
>> uniquely identify the scheduler instance to be adjusted?
>> 
>>  -- Keir
>> 
>> On 16/06/2010 17:40, "Kathy Hadley" <Kathy.Hadley@dornerworks.com>
>> wrote:
>> 
>>> Keir, George, et. al.,
>>>   I definitely saw two "ops" values.  When the .init function was
>> called, ops
>>> = 0xFF213DC0; I then used xmalloc() to allocate memory for the
>> scheduler data
>>> structure and set ops->sched_data equal to the address of that memory
>> block
>>> (similar to what is done in csched_init in sched_credit.c).  When the
>>> .adjust_global function was called, ops = 0xFF2112D0 and ops-
>>> sched_data was
>>> not equal to the address of the memory block allocated in the .init
>> function
>>> (it was equal to the value set when "sched_arinc653_def" was
>> declared).
>>> 
>>>   Regards,
>>> Kathy
>>> 
>>>> -----Original Message-----
>>>> From: Keir Fraser [mailto:keir.fraser@eu.citrix.com]
>>>> Sent: Wednesday, June 16, 2010 12:32 PM
>>>> To: Kathy Hadley; George Dunlap
>>>> Cc: xen-devel@lists.xensource.com; Juergen Gross
>>>> Subject: Re: [Xen-devel] [PATCH 1/1] Xen ARINC 653 Scheduler
>> (updated
>>>> to add support for CPU pools)
>>>> 
>>>> On 16/06/2010 17:25, "Kathy Hadley" <Kathy.Hadley@dornerworks.com>
>>>> wrote:
>>>> 
>>>>> Keir,
>>>>>   I only saw the .init function called once.  I downloaded xen-
>>>> unstable on May
>>>>> 27.  Were your updates after that?
>>>> 
>>>> My changes were done before May 27, and that ties in with you seeing
>>>> .init
>>>> called only once. That being the case, you should not see multiple
>>>> different
>>>> ops structures ('struct scheduler' instances). The only ops struct
>> that
>>>> should exist in the system in this case should be the one statically
>>>> defined
>>>> near the top of common/schedule.c.
>>>> 
>>>>  -- Keir
>>>> 
>>>>>   Thanks,
>>>>> Kathy Hadley
>>>>> 
>>>>> 
>>>>>> -----Original Message-----
>>>>>> From: Keir Fraser [mailto:keir.fraser@eu.citrix.com]
>>>>>> Sent: Wednesday, June 16, 2010 12:20 PM
>>>>>> To: George Dunlap; Kathy Hadley
>>>>>> Cc: xen-devel@lists.xensource.com; Juergen Gross
>>>>>> Subject: Re: [Xen-devel] [PATCH 1/1] Xen ARINC 653 Scheduler
>>>> (updated
>>>>>> to add support for CPU pools)
>>>>>> 
>>>>>> On 16/06/2010 17:14, "George Dunlap" <George.Dunlap@eu.citrix.com>
>>>>>> wrote:
>>>>>> 
>>>>>>>>  I actually tried the xmalloc() method first.  I found that when
>>>> the
>>>>>>>> .adjust_global function was called, the address of the "ops"
>> data
>>>>>> structure
>>>>>>>> passed to that function was different from the address of the
>>>> "ops"
>>>>>> data
>>>>>>>> structure when the .init function was called.  I wanted to use
>>>>>> .adjust_global
>>>>>>>> to modify the data structure that was created when the .init
>>>>>> function was
>>>>>>>> called, but I could not figure out a way to get the address of
>> the
>>>>>> second
>>>>>>>> data structure.  Suggestions?
>>>>>>> 
>>>>>>> It's been a month or two since I trawled through the cpupools
>> code;
>>>>>>> but I seem to recall that .init is called twice -- once for the
>>>>>>> "default pool" (cpupool0), and once for an actually in-use pool.
>>>>>>> (Juergen, can you correct me if I'm wrong?)  Is it possible that
>>>>>>> that's the difference in the pointers that you're seeing?
>>>>>> 
>>>>>> Oh yes, that was the old behaviour. I took a hatchet to the
>>>>>> scheduler/cpupool interfaces a few weeks ago and now we should
>> only
>>>>>> initialise the scheduler once, unless extra cpupools are manually
>>>>>> created.
>>>>>> The fact that Kathy is seeing two different ops structures
>> probably
>>>>>> indicates that her xen-unstable tree is very out of date. Which
>> may
>>>>>> also
>>>>>> mean that the patch will not apply to current tip.
>>>>>> 
>>>>>>  -- Keir
>>>>>> 
>>>>> 
>>>> 
>>> 
>> 
>