Re: [Xenomai] Redhat packaging patch

[Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>]

On 01/18/2013 01:07 PM, Gilles Chanteperdrix wrote:

> On 01/18/2013 09:43 AM, John Morris wrote:
> 
>> On 01/18/2013 02:18 AM, Gilles Chanteperdrix wrote:
>>> On 01/18/2013 09:16 AM, John Morris wrote:
>>>
>>>> Here's a two-liner patch to Makefile.in that disables /dev entry
>>>> creation when 'make NODEVICES=1 install'.
>>>
>>>
>>> What is wrong with "make install-user" ?
>>>
>>>>
>>>> RedHat packages are built as normal users, and there is no fakeroot
>>>> mechanism like Debian has, so devices must be recreated from a separate
>>>> script run after the rest of the package is unpacked.
>>>>
>>>> Incorporating this patch into Xenomai makes it a little cleaner for RH
>>>> package maintainers, as we don't have to keep a separate patch in the
>>>> RPM sources.  Unfortunately, we still must continue maintaining the
>>>> %post script to create the devices.  :P
>>>
>>>
>>> Why not using udev rules like debian?
>>
>> Why, those are exactly what I need.  Simplifies the package well, thanks!
>>
>> Just ran into this regression test failure, wouldn't repeat itself.  How
>> much cause for alarm?
>>
>> ++ /usr/lib64/xenomai/regression/native/tsc
>> Checking tsc for 1 minute(s)
>> min: 8, max: 18, avg: 10.0001
>> 00000056e5d97ffa -> 00000056e5d0eef9
>> tsc not monotonic after 2217251653 ticks, jumped back 561409 tick
> 
> 
> I believe it is a false positive, the test is running as an ordinary
> linux thread, so, it can be migrated by the kernel between different
> cpus. What you have here is simply the case where the test was migrated,
> on a machine with unsynced tscs.
> 
> To avoid this, we have to either set the affinity, or to shadow the main
> thread. The second solution is easier, since it will work on machines
> with UP glibc which do not support setting affinity, without having to


More exactly, with old libcs.

-- 
                                                                Gilles.