Re: [PATCH v2 1/2] mm: huge_memory: disallow hugepages if the system-wide THP sysfs settings are disabled

[Lorenzo Stoakes <lorenzo.stoakes@oracle.com>]

On Thu, Jun 12, 2025 at 10:13:40PM +0800, Baolin Wang wrote:
>
>
> On 2025/6/12 21:29, Lorenzo Stoakes wrote:
> > On Thu, Jun 12, 2025 at 02:27:06PM +0100, Lorenzo Stoakes wrote:
> > [snip]
> >
> > > I propose a compromise as I rather like your 'exclude never' negation bit.
> > >
> > > So:
> > >
> > > /* Strictly mask requested anonymous orders according to sysfs settings. */
> > > static inline unsigned long __thp_mask_anon_orders(unsigned long vm_flags,
> > >                  unsigned long tva_flags, unsigned long orders)
> > > {
> > >          const unsigned long always = READ_ONCE(huge_anon_orders_always);
> > >          const unsigned long madvise = READ_ONCE(huge_anon_orders_madvise);
> > >          const unsigned long inherit = READ_ONCE(huge_anon_orders_inherit);;
> > > 	const unsigned long never = ~(always | madvise | inherit);
> > >          const bool inherit_enabled = hugepage_global_enabled();
> > >
> > > 	/* Disallow orders that are set to NEVER directly ... */
> > > 	orders &= ~never;
> > >
> > > 	/* ... or through inheritance (global == NEVER). */
> > > 	if (!inherit_enabled)
> > > 		orders &= ~inherit;
> > >
> > > 	/*
> > > 	 * Otherwise, we only enforce sysfs settings if asked. In addition,
> > > 	 * if the user sets a sysfs mode of madvise and if TVA_ENFORCE_SYSFS
> > > 	 * is not set, we don't bother checking whether the VMA has VM_HUGEPAGE
> > > 	 * set.
> > > 	 */
> > > 	if (!(tva_flags & TVA_ENFORCE_SYSFS))
> > > 		return orders;
> > >
> > > 	if (hugepage_global_always())
> > > 		return orders & (always | inherit);
> > >
> > > 	/* We already excluded never inherit above. */
> > > 	if (vm_flags & VM_HUGEPAGE)
> > > 		return orders & (always | madvise | inherit);
> >
> > Of course... I immediately made a mistake... swap these two statements around. I
> > thought it'd be 'neater' to do the first one first, but of course it means
> > madvise (rather than inherit) orders don't get selected.
> >
> > This WHOLE THING needs refactoring.
>
> Personally, I think the 'exclude never' logic becomes more complicated. I
> made a simpler change without adding a new helper. What do you think?
>
> static inline
> unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
>                                        unsigned long vm_flags,
>                                        unsigned long tva_flags,
>                                        unsigned long orders)
> {
>         /* Optimization to check if required orders are enabled early. */
>         if (vma_is_anonymous(vma)) {

I hate the level of indentation here. There's really no reason not to have this
as a helper as this just solves this problem and any sane compiler will inline.

>                 unsigned long mask = READ_ONCE(huge_anon_orders_always);
>                 bool huge_enforce = !(tva_flags & TVA_ENFORCE_SYSFS);

Huge enforce is when we don't have enforce flag set? This is super confusing.

>                 bool has_madvise =  vm_flags & VM_HUGEPAGE;
>
>                 /*
>                  * if the user sets a sysfs mode of madvise and if
> TVA_ENFORCE_SYSFS
>                  * is not set, we don't bother checking whether the VMA has
> VM_HUGEPAGE
>                  * set.
>                  */
>                 if (huge_enforce || has_madvise)
>                         mask |= READ_ONCE(huge_anon_orders_madvise);

I find this more confusing, honestly.

I far prefer having the never checks up front, and I prefer David's 'explicitly
deal with never through negations' approach.

I also think adding these READ_ONCE()'s here adds a ton of noise.

>                 if (hugepage_global_always() ||
>                     ((has_madvise || huge_enforce) &&
> hugepage_global_enabled()))

This combination of conditions is just horribly confusing. And why are you
giving an explanation above but not for this one...

This is just combining a ton of logic in a confusing way.

>                         mask |= READ_ONCE(huge_anon_orders_inherit);
>
>                 orders &= mask;
>                 if (!orders)
>                         return 0;
>         }
>
>         return __thp_vma_allowable_orders(vma, vm_flags, tva_flags, orders);
> }

Thanks for the suggestion but I do prefer the proposed compromise solution.

>
> >
> > >
> > > 	return orders & always;
> > > }
> > >
> > > What do you think?
> > >
> > >
> > > > +       return orders;
> > > > +}
> > > > +
> > > >   /**
> > > >    * thp_vma_allowable_orders - determine hugepage orders that are allowed for vma
> > > >    * @vma:  the vm area to check
> > > > @@ -287,16 +323,8 @@ unsigned long thp_vma_allowable_orders(struct vm_area_struct *vma,
> > > >                                         unsigned long orders)
> > > >   {
> > > >          /* Optimization to check if required orders are enabled early. */
> > > > -       if ((tva_flags & TVA_ENFORCE_SYSFS) && vma_is_anonymous(vma)) {
> > > > -               unsigned long mask = READ_ONCE(huge_anon_orders_always);
> > > > -
> > > > -               if (vm_flags & VM_HUGEPAGE)
> > > > -                       mask |= READ_ONCE(huge_anon_orders_madvise);
> > > > -               if (hugepage_global_always() ||
> > > > -                   ((vm_flags & VM_HUGEPAGE) && hugepage_global_enabled()))
> > > > -                       mask |= READ_ONCE(huge_anon_orders_inherit);
> > > > -
> > > > -               orders &= mask;
> > > > +       if (vma_is_anonymous(vma)) {
> > > > +               orders = __thp_mask_anon_orders(vm_flags, tva_flags, orders);
> > > >                  if (!orders)
> > > >                          return 0;
> > >
> > > I pointed out to Baolin that __thp_vma_allowable_orders() handles the orders ==
> > > 0 case almost immediately so there's no need to do this, it just makes the code
> > > noisier.
> > >
> > > I mean we _could_ keep it but I think it's better not to for cleanliness, what
> > > do you think?
> > >
> > > >          }
> > > >
> > > >
> > > > --
> > > > Cheers,
> > > >
> > > > David / dhildenb
> > > >
> > >