Major SMP Improvements

Daniel Hellstrom daniel at gaisler.com
Fri May 15 13:01:54 UTC 2015


Hello Sebastian,

Thanks for your work! I havn't looked at your work but I have some comments.

On 05/15/2015 01:41 PM, Sebastian Huber wrote:
> https://devel.rtems.org/ticket/2130
> https://devel.rtems.org/ticket/2131
> https://devel.rtems.org/ticket/2271
> https://devel.rtems.org/ticket/2273
> https://devel.rtems.org/ticket/2307
>
> It provides a major performance boost for SMP configurations, see
> https://devel.rtems.org/ticket/2271 and https://devel.rtems.org/ticket/2273.
>
> It was tested on the Altera Cyclone V, ESA/Gaisler NGMP and Freescale T4240
> platforms using the standard RTEMS testsuite, a randomized test case that
> covers the thread queue and watchdog implementation (it run several days) and
> with heavy load network traffic using the new network stack (it uses hundreds
> of locks). After two years of work all tests pass now on the NGMP platform in
> uni-processor and SMP configurations.
Sound very promising.

> The BSPs compile and link the samples using the latest RSB based tool chain
> (except targets with a broken compiler, e.g. AVR, Moxie).
>
> With this patch set the SMP support of RTEMS can be considered to be
> practically useful on targets with more than two processors.  This is
> definitely not the case for the current Git master (possible RTEMS 4.11).  Also
> timestamps are broken on SMP in the current Git master.  So it may make sense
> to include this patch set in the RTEMS 4.11 release.  We don't have additional
> RTEMS work in the next three months planned so this is the last big change set
> for a while.
I just want to confirm this issue with the master branch. In order to use it on LEON in SMP configuration we had to make a hack in cpukit code to call the LEON timer code to make it work. It would of 
course be nice to have a fix for this in 4.11.

Daniel



More information about the devel mailing list