Sorry about that, my rtai-modules package was broken, fixed it, please re-download the new rtai-modules deb, and I will also bump the CPU count to 16 by default tomorrow.
On Wednesday, April 1, 2026 at 01:50:32 AM CDT, Luca Toniolo <[email protected]> wrote: Hey Alec, Two things to report: First, rtai_hal.ko was refusing to load because my system has 12 cores but the package was built with CONFIG_RTAI_CPUS=8. The error message was misleading ("Operation not permitted"), but dmesg showed "RTAI CONFIGURED WITH LESS THAN NUM ONLINE CPUS". Fixed it by booting with maxcpus=8. After that, the RTAI testsuite works fine: ``` RTH| lat min| ovl min| lat avg| lat max| ovl max| overruns RTD| -655| -655| -81| 5586| 5586| 0 ``` The problem is rtai_math.ko. It fails to load: ``` rtai_math: Unknown symbol __math_divzero (err -2) rtai_math: Unknown symbol __log_data (err -2) rtai_math: Unknown symbol fma (err -2) ``` I ran nm on it and those three symbols are listed as U (undefined). They're internal musl helpers that get called by the higher-level math functions (pow, log, etc.) but they weren't linked into the module. The musl code itself is fine, the issue is just that these objects didn't make it into rtai_math.ko during the build. The RTAI testsuite doesn't load rtai_math.ko so it passes, but LinuxCNC needs it for all its RT modules (motmod, tpmod). Without it nothing can run. I can see from your screenshots that latency-histogram runs on your system with HAL threads, so rtai_math.ko must load fine there. On my Trixie install (GCC 14.2.0-19, same as the kernel headers) it doesn't. What distro/version did you build the packages on? Maybe there's a binutils or linker difference causing the symbols not to get pulled in. Luca On 4/1/26 1:13 PM, Alec Ari via Emc-developers wrote: > Yes, 5.4.280 works but parallel cards have not yet been tested with it. The > reason I went with musl is due to correctness and portability. Adding glibc > to kernel space is overkill and adds way too much. You actually need to strip > out a lot from the glibc library in order for it to work whereas musl is a > lot more modular. > > Don't worry about the musl math library at all, that stuff is well tested and > true. > > Two things... Make sure the RTAI testsuite works: > > sudo /usr/realtime-5.4.280-rtai-amd64/testsuite/run > > and `halrun -U` should unload the RTAI modules. > > On Wednesday, April 1, 2026 at 12:03:12 AM CDT, Luca > Toniolo<[email protected]> wrote: > > > > > > is the 5.4.280 supposed to work? I cannot fix the rtai_hal.ko not loading > > Why are you not using Trixie glibc/libm? Was the package not for debian 13? > > > On April 1, 2026 12:54:03 PM GMT+08:00, Alec Ari via > Emc-developers<[email protected]> wrote: >> Not very... Each major kernel release requires a lot of work. IPIPE is from >> Xenomai but slightly modified for RTAI which is then called >> "hal-linux."AlecOn Tuesday, March 31, 2026 at 07:34:33 PM CDT, andy >> pugh<[email protected]> wrote: On Tue, 31 Mar 2026 at 22:39, Alec Ari via >> Emc-developers<[email protected]> wrote:Sadly, IPIPE >> isn't patched that far ahead. 5.4 or 4.19 are the only choices right now.Who >> does IPIPE? How non-portable is the patch between kernels? > > _______________________________________________ > Emc-developers mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/emc-developers _______________________________________________ Emc-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-developers _______________________________________________ Emc-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-developers
