> On Oct 8, 2021, at 1:05 AM, David Hildenbrand <david@xxxxxxxxxx> wrote: > > On 08.10.21 01:50, Nadav Amit wrote: >> From: Nadav Amit <namit@xxxxxxxxxx> >> Userfaultfd is supposed to provide the full address (i.e., unmasked) of >> the faulting access back to userspace. However, that is not the case for >> quite some time. >> Even running "userfaultfd_demo" from the userfaultfd man page provides >> the wrong output (and contradicts the man page). Notice that >> "UFFD_EVENT_PAGEFAULT event" shows the masked address. >> Address returned by mmap() = 0x7fc5e30b3000 >> fault_handler_thread(): >> poll() returns: nready = 1; POLLIN = 1; POLLERR = 0 >> UFFD_EVENT_PAGEFAULT event: flags = 0; address = 7fc5e30b3000 >> (uffdio_copy.copy returned 4096) >> Read address 0x7fc5e30b300f in main(): A >> Read address 0x7fc5e30b340f in main(): A >> Read address 0x7fc5e30b380f in main(): A >> Read address 0x7fc5e30b3c0f in main(): A >> Add a new "real_address" field to vmf to hold the unmasked address. It >> is possible to keep the unmasked address in the existing address field >> (and mask whenever necessary) instead, but this is likely to cause >> backporting problems of this patch. > > Can we be sure that no existing users will rely on this behavior that has been the case since end of 2016 IIRC, one year after UFFD was upstreamed? Let me to blow off your mind: how do you be sure that the current behavior does not make applications to misbehave? It might cause performance issues as it did for me or hidden correctness issues. > I do wonder what the official ABI nowadays is, because man pages aren't necessarily the source of truth. Documentation/admin-guide/mm/userfaultfd.rst says: "You get the address of the access that triggered the missing page event”. So it is a bug.
