Hi, We wanted to share an issue we were having with the following setup and it's solution Kernel 3.14.43 (Vendor) Xenomai 3.0.2 Cobalt core. ARM cortex A7 (SMP) based board.
We have mapped (in numerous ways/APIs) a register space of some device to user space using various methods For IO mapping - mmap rtdm_iomap_to_user rtdm_mmap_iomem etc. And for unknown at the time reason we have seen page faults on writes to these mapping which resulted In mode switches no matter what method of mapping APIs we used. It is important to note that mapping the same registrer space in Kernel i.e. devm_ioremap_resource for example did not result in page faults. After some investigation we found this had to do with the actual mapping permissions being read-only on initial mapping to user space and after the page fault handling - each faulted page changed permission to read-write. The solution to for this was to add L_PTE_DIRTY mask to pgprot for that mapping and in retrospect we found this patch https://lkml.org/lkml/2015/9/12/29 which addressed the issue. I'm unsure whether this patch found its way to 3.14, I think for Xenomai on ARM this can be very relevant so please consider adding to the mapping flow. Thanks, Alex Plits. _______________________________________________ Xenomai mailing list Xenomai@xenomai.org https://xenomai.org/mailman/listinfo/xenomai
