Hi, building Xenomai for an 8-way box and deriving its setup from an existing .config, I fortunately noticed some warning in ipipe.c. It pointed out an overflow in IPIPE_IRQMASK_VIRT when NR_CPUS gets too large. This causes NR_IRQS to explode, which then lets the virtual IRQ mask overflow.
I bet no one currently wants to run Xenomai on such large systems, but the problem is that the dependency above may nevertheless slip through some setups unnoticed. Therefore, the attached patch catches this invalid setup and bails out at compile time (does anyone have a nicer idea for a config-time safety bag?). Jan -- Siemens AG, Corporate Technology, CT SE 2 Corporate Competence Center Embedded Linux
--- arch/x86_64/kernel/ipipe.c | 6 ++++++ 1 file changed, 6 insertions(+) Index: linux-2.6.23.12-xeno/arch/x86_64/kernel/ipipe.c =================================================================== --- linux-2.6.23.12-xeno.orig/arch/x86_64/kernel/ipipe.c +++ linux-2.6.23.12-xeno/arch/x86_64/kernel/ipipe.c @@ -44,6 +44,12 @@ #include <asm/ipi.h> #include <asm/mach_apic.h> +/* Sanity check: IPIPE_VIRQ_BASE depends on CONFIG_NR_CPUS, and if this + gets too large, we run into troubles with the ipipe IRQ log. */ +#if IPIPE_VIRQ_BASE / BITS_PER_LONG > BITS_PER_LONG +# error CONFIG_NR_CPUS is too large, please lower it. +#endif + asmlinkage void preempt_schedule_irq(void); int __ipipe_tick_irq; /* =0: 8254 */
_______________________________________________ Adeos-main mailing list [email protected] https://mail.gna.org/listinfo/adeos-main
