[Xenomai] CONFIG_CONTEXT_TRACKING

[Xenomai] CONFIG_CONTEXT_TRACKING

[Gilles Chanteperdrix]

Hi,

this new option popped up in ARM's entry.S, what do other arches do, do
we support it? or do we forbid the option with CONFIG_IPIPE?

Regards.

-- 
                                                                Gilles.

Re: [Xenomai] CONFIG_CONTEXT_TRACKING

[Philippe Gerum]

On 08/31/2013 01:50 PM, Gilles Chanteperdrix wrote:
>
> Hi,
>
> this new option popped up in ARM's entry.S, what do other arches do, do
> we support it? or do we forbid the option with CONFIG_IPIPE?
>

Full nohz is desirable for an increasing number of users who have 
requirements on power consumption, so enabling context tracking for our 
supported archs seems desirable too.

-- 
Philippe.

Re: [Xenomai] CONFIG_CONTEXT_TRACKING

[Gilles Chanteperdrix]

On 08/31/2013 03:56 PM, Philippe Gerum wrote:
> On 08/31/2013 01:50 PM, Gilles Chanteperdrix wrote:
>>
>> Hi,
>>
>> this new option popped up in ARM's entry.S, what do other arches do, do
>> we support it? or do we forbid the option with CONFIG_IPIPE?
>>
> 
> Full nohz is desirable for an increasing number of users who have 
> requirements on power consumption, so enabling context tracking for our 
> supported archs seems desirable too.
> 

Is replacing the test

if (in_interrupt())

in user_enter and user_exit with

if (ipipe_root_p == 0 || in_interrupt())

sufficient? Or do we want to also track user context for tasks running
in Xenomai domain.

-- 
                                                                Gilles.

Re: [Xenomai] CONFIG_CONTEXT_TRACKING

[Gilles Chanteperdrix]

On 08/31/2013 03:57 PM, Gilles Chanteperdrix wrote:
> On 08/31/2013 03:56 PM, Philippe Gerum wrote:
>> On 08/31/2013 01:50 PM, Gilles Chanteperdrix wrote:
>>>
>>> Hi,
>>>
>>> this new option popped up in ARM's entry.S, what do other arches do, do
>>> we support it? or do we forbid the option with CONFIG_IPIPE?
>>>
>>
>> Full nohz is desirable for an increasing number of users who have 
>> requirements on power consumption, so enabling context tracking for our 
>> supported archs seems desirable too.
>>
> 
> Is replacing the test
> 
> if (in_interrupt())
> 
> in user_enter and user_exit with
> 
> if (ipipe_root_p == 0 || in_interrupt())
> 
> sufficient? Or do we want to also track user context for tasks running
> in Xenomai domain.
> 
The reason I ask is because avoiding the call to user_enter for a task
running in Xenomai domain is possible, however avoiding the call to
user_exit implies calling an equivalent of ipipe_root_p for each kernel
entry point, so, entering user_exit and adding the call to ipipe_root_p
there seems easier, simply adding a function call on the way.

-- 
                                                                Gilles.