On 12/7/22 15:44, Peter Xu wrote:
The part that matters in the comment is "need to be before setting current state". blocking_state = userfaultfd_get_blocking_state(vmf->flags); if (is_vm_hugetlb_page(vma)) hugetlb_vma_lock_read(vma); set_current_state(blocking_state); down_read() can sleep and also modify the task state, we cannot take the lock after that point because otherwise the task state will be messed up.
OK, how about: /* * Take the vma lock now, in order to safely call * userfaultfd_huge_must_wait() a little bit later. Because acquiring * the (sleepable) vma lock potentially modifies the current task state, * that must be before explicitly calling set_current_state(). */ Other than that, the patch looks good, so: Reviewed-by: John Hubbard <jhubbard@xxxxxxxxxx> thanks, -- John Hubbard NVIDIA
