On Tue, 28 Jul 2015, Michal Hocko wrote:
> [I am sorry but I didn't get to this sooner.]
>
> On Mon 27-07-15 10:54:09, Eric B Munson wrote:
> > Now that VM_LOCKONFAULT is a modifier to VM_LOCKED and
> > cannot be specified independentally, it might make more sense to mirror
> > that relationship to userspace. Which would lead to soemthing like the
> > following:
>
> A modifier makes more sense.
>
> > To lock and populate a region:
> > mlock2(start, len, 0);
> >
> > To lock on fault a region:
> > mlock2(start, len, MLOCK_ONFAULT);
> >
> > If LOCKONFAULT is seen as a modifier to mlock, then having the flags
> > argument as 0 mean do mlock classic makes more sense to me.
> >
> > To mlock current on fault only:
> > mlockall(MCL_CURRENT | MCL_ONFAULT);
> >
> > To mlock future on fault only:
> > mlockall(MCL_FUTURE | MCL_ONFAULT);
> >
> > To lock everything on fault:
> > mlockall(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT);
>
> Makes sense to me. The only remaining and still tricky part would be
> the munlock{all}(flags) behavior. What should munlock(MLOCK_ONFAULT)
> do? Keep locked and poppulate the range or simply ignore the flag an
> just unlock?
>
> I can see some sense to allow munlockall(MCL_FUTURE[|MLOCK_ONFAULT]),
> munlockall(MCL_CURRENT) resp. munlockall(MCL_CURRENT|MCL_FUTURE) but
> other combinations sound weird to me.
>
> Anyway munlock with flags opens new doors of trickiness.
In the current revision there are no new munlock[all] system calls
introduced. munlockall() unconditionally cleared both MCL_CURRENT and
MCL_FUTURE before the set and now unconditionally clears all three.
munlock() does the same for VM_LOCK and VM_LOCKONFAULT. If the user
wants to adjust mlockall flags today, they need to call mlockall a
second time with the new flags, this remains true for mlockall after
this set and the same behavior is mirrored in mlock2. The only
remaining question I have is should we have 2 new mlockall flags so that
the caller can explicitly set VM_LOCKONFAULT in the mm->def_flags vs
locking all current VMAs on fault. I ask because if the user wants to
lock all current VMAs the old way, but all future VMAs on fault they
have to call mlockall() twice:
mlockall(MCL_CURRENT);
mlockall(MCL_CURRENT | MCL_FUTURE | MCL_ONFAULT);
This has the side effect of converting all the current VMAs to
VM_LOCKONFAULT, but because they were all made present and locked in the
first call, this should not matter in most cases. The catch is that,
like mmap(MAP_LOCKED), mlockall() does not communicate if mm_populate()
fails. This has been true of mlockall() from the beginning so I don't
know if it needs more than an entry in the man page to clarify (which I
will add when I add documentation for MCL_ONFAULT). In a much less
likely corner case, it is not possible in the current setup to request
all current VMAs be VM_LOCKONFAULT and all future be VM_LOCKED.
Attachment:
signature.asc
Description: Digital signature