Hey Toshi, i am working on the xenomai project where we recently rebased our patches on 4.1. The fact that io-regions may be mapped as huge pages caused some problem in our patch.
Whenever memory is vmalloced or ioremapped we make sure it is synched into all mm's to avoid page faults in that area. We use vmalloc_sync_one to do that. Now vmalloc_sync_one can not handle huge pages (i am talking about x86 32+64). Hence for areas where you introduced huge pages vmalloc_fault should never happen in any mm, otherwise your patchset would have introduced a regression in mainline. Can you help me understand by which mechanism these io-regions are always mapped into all mm's, ruling out calls to vmalloc_fault and vmalloc_sync_one? Because calling this on these regions will treat you device memory as pagetables when huge pages are used. Best regards, Henning _______________________________________________ Xenomai mailing list [email protected] http://xenomai.org/mailman/listinfo/xenomai
