On Thu, 19 Nov 2015 09:25:24 +0100
Christian Borntraeger <borntraeger@xxxxxxxxxx> wrote:
> On 11/19/2015 09:18 AM, Martin Schwidefsky wrote:
> > On Thu, 19 Nov 2015 00:49:58 +0100
> > Dominik Dingel <dingel@xxxxxxxxxxxxxxxxxx> wrote:
> >
> >> The userfaultfd does need FAULT_FLAG_ALLOW_RETRY to not return
> >> VM_FAULT_SIGBUS. So we improve the gmap code to handle one
> >> VM_FAULT_RETRY.
> >>
> >> Signed-off-by: Dominik Dingel <dingel@xxxxxxxxxxxxxxxxxx>
> >> ---
> >> arch/s390/mm/pgtable.c | 28 ++++++++++++++++++++++++----
> >> 1 file changed, 24 insertions(+), 4 deletions(-)
> >>
> >> diff --git a/arch/s390/mm/pgtable.c b/arch/s390/mm/pgtable.c
> >> index 54ef3bc..8a0025d 100644
> >> --- a/arch/s390/mm/pgtable.c
> >> +++ b/arch/s390/mm/pgtable.c
> >> @@ -577,15 +577,22 @@ int gmap_fault(struct gmap *gmap, unsigned long gaddr,
> >> unsigned int fault_flags)
> >> {
> >> unsigned long vmaddr;
> >> - int rc;
> >> + int rc, fault;
> >>
> >> + fault_flags |= FAULT_FLAG_ALLOW_RETRY;
> >> +retry:
> >> down_read(&gmap->mm->mmap_sem);
> >> vmaddr = __gmap_translate(gmap, gaddr);
> >> if (IS_ERR_VALUE(vmaddr)) {
> >> rc = vmaddr;
> >> goto out_up;
> >> }
> >> - if (fixup_user_fault(current, gmap->mm, vmaddr, fault_flags)) {
> >> + fault = fixup_user_fault(current, gmap->mm, vmaddr, fault_flags);
> >> + if (fault & VM_FAULT_RETRY) {
> >> + fault_flags &= ~FAULT_FLAG_ALLOW_RETRY;
> >> + fault_flags |= FAULT_FLAG_TRIED;
> >> + goto retry;
> >> + } else if (fault) {
> >> rc = -EFAULT;
> >> goto out_up;
> >> }
> >
> > Me thinks that you want to add the retry code into fixup_user_fault itself.
> > You basically have the same code around the three calls to fixup_user_fault.
> > Yes, it will be a common code patch but I guess that it will be acceptable
> > given userfaultfd as a reason.
>
> That makes a lot of sense. In an earlier discussion (a followup of Jasons
> mm: Loosen MADV_NOHUGEPAGE to enable Qemu postcopy on s390) patch.
>
> Andrea suggested the following:
>
> It's probably better to add a fixup_user_fault_unlocked that will work
> like get_user_pages_unlocked. I.e. leaves the details of the mmap_sem
> locking internally to the function, and will handle VM_FAULT_RETRY
> automatically by re-taking the mmap_sem and repeating the
> fixup_user_fault after updating the FAULT_FLAG_ALLOW_RETRY to
> FAULT_FLAG_TRIED.
I know, I saw his mail. But within the gmap code we need to take the mmap_sem before calling fixup_user_fault as well as holding it for later on like __gmap_link.
We could introduce a new wrapper arround fixup_user_fault, like:
fixup_user_fault_retry, which would take care of the retry logic, but does not encapsulate the complete mmap_sem logic.
@Kirill would that be acceptable for you as well?
--
To unsubscribe from this list: send the line "unsubscribe linux-s390" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
