On Sat, Jan 08, 2022 at 01:37:09AM +0000, Matthew Wilcox wrote: > I think there is sufficient protection; either we have interrupts > disabled in gup-fast (which prevents RCU from finishing a grace > period), or we have the mmap_sem held in gup-slow (which prevents > pages from being removed from the page tables). mmap sem doesn't prevent pages from being removed, for instance unmap_mapping_pages() doesn't hold it. It is safe because gup slow holds the PTLs when it calls this. No idea about how the get_gate_page() works Jason