Alessandro Pittaluga wrote:
Hi,
In the native skin rt_task_create receives a pointer to a RT_TASK struct
which must be allocated in permanent memory.
So I thought that the address of this struct would be the task
identifier used to be passed to the other rt_task functions.
And it works in this way.
The strange thing is that rt_task_self does not return the same value of
RT_TASK*, but allocates a new RT_TASK structure the first time it is
called in a new task context.
Why?
Because the relevant bit of information is not the descriptor struct per
se, but rather its contents. This way, one can point to the same object
using a descriptor struct, regardless of the allocation context of such
struct. This is particularly useful for sharing objects between
different execution spaces: regardless of the layout of the descriptor
or the context it has been created, one only has to know that the
descriptor contains enough information to point back to the registered
object; IOW, the descriptor address is not part of the indexing information.
Using this property, rt_task_self() can simply fill any descriptor
_struct_ once, at the time of the first call for any given thread. The
struct itself is part of TSD/PTD information, and has been pre-allocated
at thread creation. This way, the kernel is only asked once for filling
the struct on a pre-allocated memory area, and subsequent calls to
rt_task_self() just return the PTD value associated with the descriptor
struct key.
Many thanks
Alessandro
--
Philippe.
_______________________________________________
Xenomai-help mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-help
