generall RTEMS questions

[Joel Sherrill]

On September 26, 2014 12:36:20 AM CDT, Sebastian Huber <sebastian.huber at embedded-brains.de> wrote:
>Hello Marco,
>
>On 19/09/14 11:15, Hoefle Marco wrote:
>> Hello,
>>
>> we have currently the RTEMS sources from Gaisler in use (as they have
>their
>> drivers not in mainline RTEMS, unfortunately).
>>
>> The Gaisler RTEMS kernel is based on 4.10.
>>
>> However, a few useful features are only available in 4.11 (JFFS2 File
>system
>> for example).
>>
>> We are thinking on moving to 4.11 but we are wondering if there is a
>proper
>> tested stable version available. In 4-10 are tags in 4-11 I did not
>find one.
>>
>> Would you recommend using 4.11 in a production environment?
>
>RTEMS 4.11 is work in progress.  If you are lucky and select the right
>commit, 
>then it is a production ready version.  It helps to rely on companies
>working 
>daily with the system doing regular tests if you want to do serious
>business 
>with RTEMS.
>
>>
>> We use rtems source builder which works fine. However, we still do
>not
>> understand why is there such a dependency toolchain – rtems kernel?
>
>There is an interaction between Newlib (the C library) and RTEMS.  Some
>parts 
>are defined in Newlib header files and implemented in RTEMS source
>files.

And changing the GCC version gets you out of sync up with testing. Different gcc versions can have new optimizations that break the code, bugs, or generate new warnings.

>>
>> Why can’t the standard gcc be used to compile the rtems bsp in an
>archive and
>> linked to the application afterwards?
>
>The problem is not GCC.  GCC and Newlib are built together.
.
But the GCC configuration is tweaked to be RTEMS specific and not generic elf or Linux. The runtime libraries are compiled using Newlib in the RTEMS configuration. There is not a large difference between say sparc-elf and sparc-rtems at the toolset source level but the resulting GCC and Newlib are different. 

Ultimately a test with a specific configuration of tools and if you deviate from that, it can be OK or not.