Le Mardi, Octobre 01, 2019 16:37 CEST, Jan Kiszka <[email protected]> a 
écrit:
 On 01.10.19 15:26, François Legal wrote:
> Hello,
>
> Could you please clarify what "upstream" is ? RTNet project ? There does not
> seem to be activity there since 2012 ?

RTnet was merged into Xenomai 2014. I thought you knew when you were posting
here ;).That's what I though, but I was surprised by the mention to "upstream".
> I'll be happy to contribute on this. You mention "glue scripting", could you 
> be
> more precise to let me try to fix that part too.

xenomai/utils/net/rtnet.in. But I suspect that will be rather simple.So the 
attached patch (against master) allows one to get the RTNet code builtin, plus 
the rtnet script modified to take that into account. I can make a patch to 
3.0.9 available if that's needed.
About the stable init that does not depend on link order, I'll be glad to work 
on that, but I need some hints on where to start if possible.
More tricky is a stable init ordering, ideally something that does not depend on
link ordering. And maybe there is more.

>
> Shall I send the patch to this list ?

Exactly. Against the master branch. If it's a fix, we can consider backporting
to 3.0.x and 3.1.x (3.1 is already in feature freeze state).

Jan

>
> Thanks
>
> François
>
> Le Mardi, Octobre 01, 2019 15:10 CEST, Jan Kiszka <[email protected]> a 
> écrit:
>> On 01.10.19 15:04, François Legal via Xenomai wrote:
>> >
>> > Hello list,
>> >
>> > I've been struggling to get the RTNet + RTIPv4 part to work on a Zynq 7000
>> board. Using source code from stable Xenomai 3.0.9 tarball, and patching a
>> vanilla 4.4.189 linux.
>> > I consistently had the kernel crash (0 pointer dereference error) at 
>> > startup
>> when enabling either RTIPv4, RTCFG or RTMAC.
>> >
>> > I finally found that there "might" be a problem with the Makefile
>> in kernel/drivers/net/stack.
>> > It turns out that (on my system at least), the order in which the protocol
>> stacks appear in the Makefile is wrong because (I running a kernel with
>> everything builtin, no modules) the protocols are loaded in that order, and
>> thus, (in my case) RTIPv4 gets initialized before stack_mgr does which cause
>> it to fail while registering (calling __rtdev_add_pack with a not
>> initialized rt_packets array).
>> >
>> > When I reorder the protocols in the Makefile, everything works flawlessly.
>> >
>> > I can provide a patch if this needs to be fixed.
>> >
>>
>> Upstream does not support built-in RTnet. I suspect we need to sort this and
>> more issues out systematically in order to permit that. Also, the glue 
>> scripting
>> will need adjustments because of assumptions that RTnet comes in modules.
>>
>> That said, contributions in that directions are welcome!
>>
>> Jan
>>
>> --
>> Siemens AG, Corporate Technology, CT RDA IOT SES-DE
>> Corporate Competence Center Embedded Linux
>
>
>


--
Siemens AG, Corporate Technology, CT RDA IOT SES-DE
Corporate Competence Center Embedded Linux


 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: patch-builtin
Type: application/octet-stream
Size: 6156 bytes
Desc: not available
URL: 
<http://xenomai.org/pipermail/xenomai/attachments/20191002/5549df57/attachment.obj>

Reply via email to