This fixes the issue reported by Robert Richter where the fact that the node id of struct pages covered by NOMAP regions is not initialized, triggering a VM_BUG_ON() in the mm code. I know that this approach is the least preferred option by Robert, but it has been used successfully in the downstream Linaro Enterprise kernel, running on HiSilicon D05, which suffered from the same issue as Cavium ThunderX where it was originally reported. Given that the other proposed solutions either fail to solve the issue completely, or cause regressions in other code (hibernate), I think this issue is appropriate for merging now, and backported to -stable. If there are performance concerns, we can try to improve on this solution, which could include reverting patch #2 altogether, for all I care. Patch #1 fixes a bug in the generic mm code where a struct page is dereferenced before pfn_valid() is called. This should probably go to stable regardless of where the arm64 discussion goes. Patch #2 enables CONFIG_HOLES_IN_ZONE for arm64 numa, causing the kernel to no longer assume that all pages in a zone have valid struct pages associated with them. Ard Biesheuvel (2): mm: don't dereference struct page fields of invalid pages arm64: mm: enable CONFIG_HOLES_IN_ZONE for NUMA arch/arm64/Kconfig | 4 ++++ mm/page_alloc.c | 6 +++--- 2 files changed, 7 insertions(+), 3 deletions(-) -- 2.7.4 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>