I think we may need some more investigation into it.
Here's one thing I think would be helpful to assess the impact of this problem. Can we take a survey of Linux distros to see what preemption settings they use? I will even make some predictions, as I did above. I'll predict Ubuntu uses voluntary preemption, and Fedora turns it off. Prove me wrong.
Ok, signal 8 is just a symptom, the reason can be many things: overclocking, undervolting, overheating, or, as in this case, a kernel bug resulting in FPU stack corruption.
I guess a FAQ entry should state that Linux kernels between 2.6.20 and probably 2.6.27 (or whatever kernel will get this fix) with CONFIG_PREEMPT=y have a certain bug that might trigger validation errors or signal 8 exceptions on any BOINC app (in fact all applications)
All other possible reasons like overclocking/undervolting/overheating seem to produce a mix of signal 8 and SIGSEGV segmentation faults (error code 99)
The fix for the Linux kernel bug that was discovered when E@H kept failing under certain circumstances has now made it into kernel release candidate 2.6.26-rc5
So if you have experienced frequent "signal 8" (error code 38) problems, are using a Linux kernel >= 2.6.20 and the kernel was compiled with the setting "CONFIG_PREEMPT=y" , you can either wait for the final release of kernel 2.6.26 (estimated to happen within the next couple of weeks) or try the 2.6.26-rc5 kernel version.
Ageless wrote: I think we may
)
Here's one thing I think would be helpful to assess the impact of this problem. Can we take a survey of Linux distros to see what preemption settings they use? I will even make some predictions, as I did above. I'll predict Ubuntu uses voluntary preemption, and Fedora turns it off. Prove me wrong.
- Eric Myers
Ok, signal 8 is just a
)
Ok, signal 8 is just a symptom, the reason can be many things: overclocking, undervolting, overheating, or, as in this case, a kernel bug resulting in FPU stack corruption.
I guess a FAQ entry should state that Linux kernels between 2.6.20 and probably 2.6.27 (or whatever kernel will get this fix) with CONFIG_PREEMPT=y have a certain bug that might trigger validation errors or signal 8 exceptions on any BOINC app (in fact all applications)
All other possible reasons like overclocking/undervolting/overheating seem to produce a mix of signal 8 and SIGSEGV segmentation faults (error code 99)
CU
Bikeman
Hi! The fix for the Linux
)
Hi!
The fix for the Linux kernel bug that was discovered when E@H kept failing under certain circumstances has now made it into kernel release candidate 2.6.26-rc5
http://thread.gmane.org/gmane.linux.kernel/690130
So if you have experienced frequent "signal 8" (error code 38) problems, are using a Linux kernel >= 2.6.20 and the kernel was compiled with the setting "CONFIG_PREEMPT=y" , you can either wait for the final release of kernel 2.6.26 (estimated to happen within the next couple of weeks) or try the 2.6.26-rc5 kernel version.
CU
Bikeman