Messages in this thread | | | From | Joel Fernandes <> | Subject | Re: [PATCH 0/2] fix vma->anon_vma check for per-VMA locking; fix anon_vma memory ordering | Date | Fri, 28 Jul 2023 15:50:23 -0400 |
| |
> On Jul 28, 2023, at 2:18 PM, Paul E. McKenney <paulmck@kernel.org> wrote: > > On Fri, Jul 28, 2023 at 02:03:09PM -0400, Joel Fernandes wrote: >>> On Fri, Jul 28, 2023 at 1:51 PM Alan Stern <stern@rowland.harvard.edu> wrote: >>> >>> On Fri, Jul 28, 2023 at 01:35:43PM -0400, Joel Fernandes wrote: >>>> On Fri, Jul 28, 2023 at 8:44 AM Will Deacon <will@kernel.org> wrote: >>>>> >>>>> On Thu, Jul 27, 2023 at 12:34:44PM -0400, Joel Fernandes wrote: >>>>>>> On Jul 27, 2023, at 10:57 AM, Will Deacon <will@kernel.org> wrote: >>>>>>> On Thu, Jul 27, 2023 at 04:39:34PM +0200, Jann Horn wrote: >>>>>>>> if (READ_ONCE(vma->anon_vma) != NULL) { >>>>>>>> // we now know that vma->anon_vma cannot change anymore >>>>>>>> >>>>>>>> // access the same memory location again with a plain load >>>>>>>> struct anon_vma *a = vma->anon_vma; >>>>>>>> >>>>>>>> // this needs to be address-dependency-ordered against one of >>>>>>>> // the loads from vma->anon_vma >>>>>>>> struct anon_vma *root = a->root; >>>>>>>> } >>>>>>>> >>>>>>>> >>>>>>>> Is this fine? If it is not fine just because the compiler might >>>>>>>> reorder the plain load of vma->anon_vma before the READ_ONCE() load, >>>>>>>> would it be fine after adding a barrier() directly after the >>>>>>>> READ_ONCE()? >>>>>>> >>>>>>> I'm _very_ wary of mixing READ_ONCE() and plain loads to the same variable, >>>>>>> as I've run into cases where you have sequences such as: >>>>>>> >>>>>>> // Assume *ptr is initially 0 and somebody else writes it to 1 >>>>>>> // concurrently >>>>>>> >>>>>>> foo = *ptr; >>>>>>> bar = READ_ONCE(*ptr); >>>>>>> baz = *ptr; >>>>>>> >>>>>>> and you can get foo == baz == 0 but bar == 1 because the compiler only >>>>>>> ends up reading from memory twice. >>>>>>> >>>>>>> That was the root cause behind f069faba6887 ("arm64: mm: Use READ_ONCE >>>>>>> when dereferencing pointer to pte table"), which was very unpleasant to >>>>>>> debug. >>>>>> >>>>>> Will, Unless I am missing something fundamental, this case is different though. >>>>>> This case does not care about fewer reads. As long as the first read is volatile, the subsequent loads (even plain) >>>>>> should work fine, no? >>>>>> I am not seeing how the compiler can screw that up, so please do enlighten :). >>>>> >>>>> I guess the thing I'm worried about is if there is some previous read of >>>>> 'vma->anon_vma' which didn't use READ_ONCE() and the compiler kept the >>>>> result around in a register. In that case, 'a' could be NULL, even if >>>>> the READ_ONCE(vma->anon_vma) returned non-NULL. >>>> >>>> If I can be a bit brave enough to say -- that appears to be a compiler >>>> bug to me. It seems that the compiler in such an instance violates the >>>> "Sequential Consistency Per Variable" rule? I mean if it can't even >>>> keep SCPV true for a same memory-location load (plain or not) for a >>>> sequence of code, how can it expect the hardware to. >>> >>> It's not a compiler bug. In this example, some other thread performs a >>> write that changes vma->anon_vma from NULL to non-NULL. This write >>> races with the plain reads, and compilers are not required to obey the >>> "Sequential Consistency Per Variable" rule (or indeed, any rule) when >>> there is a data race. >> >> So you're saying the following code behavior is OK? >> >> /* Say anon_vma can only ever transition from NULL to non-NULL values */ >> a = vma->anon_vma; // Reads NULL >> b = READ_ONCE(vma->anon_vma); // Reads non-NULL >> c = vma->anon_vma; // Reads NULL!!! >> if (b) { >> c->some_attribute++; // Oopsie >> } > > Is there some way to obtain (a && !b) that does not involve a data race, > and they carte blanche for the compiler to do whatever it pleases? > I am not seeing one. > > What am I missing?
Probably nothing. I think I was living briefly in a fantasy world where I expected predictable compiler behavior on same-memory accesses amidst data traces. It is good to come back to reality.
thanks,
- Joel
> > Thanx, Paul
| |