RTEMS | Should all PPC BSPs build with -mstrict-align? (#3767)

Sat Feb 15 13:28:56 UTC 2025

Giles Knap commented: https://gitlab.rtems.org/rtems/rtos/rtems/-/issues/3767#note_120046

I've just tried adding this flag to my BSP build and it has definitely solved the intermittent problems I was having. 

I'm building the first RTEMS Vacuum EPICS IOC for a beamline at Diamond Light Source with RTEMS6.1, epics 7.0 (HEAD). Our hardware is the mvme5500 and I'm using the beatnik BSP.

The IOC is fairly complex and is linking in many EPICS support modules. It has been working well but is slightly unreliable on startup with about 5% of boots crashing early. When I added the autosave support module into the mix the failure rate rose to 90%. On the occasions where the IOC boots OK, it then seems to function reliably, indefinitely. The stack trace on failure always looks like the attached file.

[bad.txt](/uploads/674550257730aa31e7b7153a46bd9700/bad.txt)

The trace shows that at an exception is occurring while freeing some heap during a thread exit. I don't understand much more detail than that. 

An identical IOC except with --mstrict-align used in the BSP build build completely fixed intermittent boot failures.

As I write this, my stress test has hit 55 reboots without a failure. I've never seem this level of reliability before.

For info my (containerized) BSP build is here https://github.com/epics-containers/rtems-bsp/tree/strict-align

-- 
View it on GitLab: https://gitlab.rtems.org/rtems/rtos/rtems/-/issues/3767#note_120046
You're receiving this email because of your account on gitlab.rtems.org.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/bugs/attachments/20250215/840609bd/attachment.htm>