Re: [Xenomai-core] [PATCH v4 02/25] RTDM: Add rtdm_rt_capable() service

[Jan Kiszka <jan.kiszka@domain.hid>]

[-- Attachment #1: Type: text/plain, Size: 2377 bytes --]

Philippe Gerum wrote:
> On Mon, 2010-04-19 at 12:28 +0200, Jan Kiszka wrote:
>> This adds rtdm_rt_capable(), a function that can be used by drivers to
>> detect callers that could issue a service request also from the
>> (typically preferred) real-time context. If that is the case, the driver
>> can trigger a restart of the request if the current context is not
>> real-time.
>>
>> CC: Philippe Gerum <rpm@xenomai.org>
>> CC: Alexis Berlemont <berlemont.hauw@domain.hid>
>> Signed-off-by: Jan Kiszka <jan.kiszka@domain.hid>
>> ---
>>  include/rtdm/rtdm_driver.h |    6 ++++++
>>  ksrc/skins/rtdm/drvlib.c   |   25 +++++++++++++++++++++++++
>>  2 files changed, 31 insertions(+), 0 deletions(-)
>>
>> diff --git a/include/rtdm/rtdm_driver.h b/include/rtdm/rtdm_driver.h
>> index 0fc1496..d124157 100644
>> --- a/include/rtdm/rtdm_driver.h
>> +++ b/include/rtdm/rtdm_driver.h
>> @@ -1296,6 +1296,12 @@ static inline int rtdm_in_rt_context(void)
>>  {
>>  	return (rthal_current_domain != rthal_root_domain);
>>  }
>> +
>> +static inline int rtdm_rt_capable(void)
>> +{
>> +	return (!xnpod_root_p() || xnshadow_thread(current) != NULL);
>> +}
>> +
> 
> This patch still has fundamental flaw, the service would not work
> correctly and would actually return the wrong value, in case it is used
> deeply buried into some helper code, traversed by asynchronous Xenomai
> contexts.

Using it from that contexts would be the fundamental flaw. We can catch
this, though.

> At the very least, you should avoid this by binding that
> service to a valid rtdm_dev_context, so that you know that it is called
> from a context that is correct in the first place, where a
> rtdm_dev_context reference is available.
> 
> I.e. 
> static inline int rtdm_rt_capable(struct rtdm_dev_context *context), and
> extract the information from that context structure. That would avoid a
> lot of headaches down the road to your user base.

The context is not that helpful here.

But instead of using current, we can force the caller to pass the
user_info which is only available in helper context:

static inline int rtdm_rt_capable(rtdm_user_info_t *user_info)
{
	XENO_ASSERT(RTDM, !xnpod_asynch_p(), return 0;);

	return (!xnpod_root_p() ||
		(user_info && xnshadow_thread(user_info) != NULL));
}

Pushed a corresponding patch.

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]