[PATCH v4 1/2] cobalt: switch hand over status to -EADV for non-RTDM fd

Philippe Gerum rpm at xenomai.org
Fri Aug 30 19:36:31 CEST 2019


On 8/30/19 7:13 PM, Philippe Gerum via Xenomai wrote:
> On 8/30/19 6:58 PM, Jan Kiszka wrote:
>> On 30.08.19 18:03, Philippe Gerum wrote:
>>> Having the RTDM core return -EBADF to indicate that it does not manage
>>> a file descriptor is a problem, as several drivers also raise this
>>> error to notify userland about an aborted wait due to a connection
>>> being dismantled (e.g. RTnet). In this case, libcobalt ends up
>>> forwarding the aborted request to the glibc, which is wrong.
>>>
>>> Switch from -EBADF to -EADV to notify userland that RTDM does not
>>> manage a file descriptor, which cannot conflict with any sensible
>>> error code the Cobalt core or any RTDM driver may return.

<snip>

>>
>> I'm seeing a few more suspicious occurrences of EBADF in the core:
>>
>> kernel/cobalt/posix/io.c:191:   return -EBADF;
>> kernel/cobalt/posix/io.c:289:                   return -EBADF;
>> kernel/cobalt/posix/syscall32.c:742:                    return -EBADF;
> 
> We still want to receive -EBADF on wrong fildes appearing in select()
> descriptors.

FWIW, the rationale for this change is that forwarding a request to
glibc iff the first fd found in a set is a regular one is really fragile
and actually never worked properly: what if there is a mixed set of
rtdm/regular fds, with the first one belonging to RTDM?

So people should make sure to call libcobalt's version of select() if
they want to monitor RTDM fds, __STD(select()) otherwise. Relying on a
silent fallback from libcobalt to glibc solely based on fairly weak
heuristics is not helping anyone in the end.

-- 
Philippe.



More information about the Xenomai mailing list