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

[Baolin Wang <baolin.wang@linux.alibaba.com>]

On 2025/6/12 16:51, David Hildenbrand wrote:
> On 12.06.25 09:51, Baolin Wang wrote:
>>
>>
>> On 2025/6/11 20:34, David Hildenbrand wrote:
>>> On 05.06.25 10:00, Baolin Wang wrote:
>>>> The MADV_COLLAPSE will ignore the system-wide Anon THP sysfs settings,
>>>> which
>>>> means that even though we have disabled the Anon THP configuration,
>>>> MADV_COLLAPSE
>>>> will still attempt to collapse into a Anon THP. This violates the rule
>>>> we have
>>>> agreed upon: never means never.
>>>>
>>>> Another rule for madvise, referring to David's suggestion: “allowing
>>>> for collapsing
>>>> in a VM without VM_HUGEPAGE in the "madvise" mode would be fine".
>>>>
>>>> To address this issue, should check whether the Anon THP configuration
>>>> is disabled
>>>> in thp_vma_allowable_orders(), even when the TVA_ENFORCE_SYSFS flag is
>>>> set.
>>>>
>>>> In summary, the current strategy is:
>>>>
>>>> 1. If always & orders == 0, and madvise & orders == 0, and
>>>> hugepage_global_enabled() == false
>>>> (global THP settings are not enabled), it means mTHP of that orders
>>>> are prohibited
>>>> from being used, then madvise_collapse() is forbidden for that orders.
>>>>
>>>> 2. If always & orders == 0, and madvise & orders == 0, and
>>>> hugepage_global_enabled() == true
>>>> (global THP settings are enabled), and inherit & orders == 0, it means
>>>> mTHP of that
>>>> orders are still prohibited from being used, thus madvise_collapse()
>>>> is not allowed
>>>> for that orders.
>>>>
>>>> Reviewed-by: Zi Yan <ziy@nvidia.com>
>>>> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
>>>> ---
>>>>    include/linux/huge_mm.h | 23 +++++++++++++++++++----
>>>>    1 file changed, 19 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>>>> index 2f190c90192d..199ddc9f04a1 100644
>>>> --- a/include/linux/huge_mm.h
>>>> +++ b/include/linux/huge_mm.h
>>>> @@ -287,20 +287,35 @@ 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 (vma_is_anonymous(vma)) {
>>>> +        unsigned long always = READ_ONCE(huge_anon_orders_always);
>>>> +        unsigned long madvise = READ_ONCE(huge_anon_orders_madvise);
>>>> +        unsigned long inherit = READ_ONCE(huge_anon_orders_inherit);
>>>> +        unsigned long mask = always | madvise;
>>>> +
>>>> +        /*
>>>> +         * If the system-wide THP/mTHP sysfs settings are disabled,
>>>> +         * then we should never allow hugepages.
>>>   > +         */> +        if (!(mask & orders) &&
>>> !(hugepage_global_enabled() && (inherit & orders)))
>>>> +            return 0;
>>>
>>> I'm still trying to digest that. Isn't there a way for us to work with
>>> the orders,
>>> essentially masking off all orders that are forbidden globally. Similar
>>> to below, if !orders, then return 0?
>>> /* Orders disabled directly. */
>>> orders &= ~TODO;
>>> /* Orders disabled by inheriting from the global toggle. */
>>> if (!hugepage_global_enabled())
>>>       orders &= ~READ_ONCE(huge_anon_orders_inherit);
>>>
>>> TODO is probably a -1ULL and then clearing always/madvise/inherit. Could
>>> add a simple helper for that
>>>
>>> huge_anon_orders_never
>>
>> I followed Lorenzo's suggestion to simplify the logic. Does that look
>> more readable?
>>
>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>> index 2f190c90192d..3087ac7631e0 100644
>> --- a/include/linux/huge_mm.h
>> +++ b/include/linux/huge_mm.h
>> @@ -265,6 +265,43 @@ unsigned long __thp_vma_allowable_orders(struct
>> vm_area_struct *vma,
>>                                            unsigned long tva_flags,
>>                                            unsigned long orders);
>>
>> +/* 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)
>> +{
>> +       unsigned long always = READ_ONCE(huge_anon_orders_always);
>> +       unsigned long madvise = READ_ONCE(huge_anon_orders_madvise);
>> +       unsigned long inherit = READ_ONCE(huge_anon_orders_inherit);
>> +       bool inherit_enabled = hugepage_global_enabled();
>> +       bool has_madvise =  vm_flags & VM_HUGEPAGE;
>> +       unsigned long mask = always | madvise;
>> +
>> +       mask = always | madvise;
>> +       if (inherit_enabled)
>> +               mask |= inherit;
>> +
>> +       /* All set to/inherit NEVER - never means never globally, 
>> abort. */
>> +       if (!(mask & orders))
>> +               return 0;
> 
> Still confusing. I am not sure if we would properly catch when someone 
> specifies e.g., 2M and 1M, while we only have 2M disabled.

IIUC, Yes. In your case, we will only allow order 8 (1M mTHP).

> I would rewrite the function to only ever substract from "orders".
> 
> ...
> 
> /* Disallow orders that are set to NEVER directly ... */
> order &= (always | madvise | inherit);
> 
> /* ... or through inheritance. */
> if (inherit_enabled)
>      orders &= ~inherit;

Sorry, I didn't get you here.

If orders = THP_ORDERS_ALL_ANON, inherit = 0x200 (order 9), always and 
madvise are 0, and inherit_enabled = true. Then orders will be 0 with 
your logic. But we should allow order 9, right?

> 
> /*
>   * 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 (!orders || !(tva_flags & TVA_ENFORCE_SYSFS))
>      return orders;
> 
>> +
>> +       /*
>> +        * 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;
>> +
>> +       mask = always;
>> +       if (has_madvise)
>> +               mask |= madvise;
>> +       if (hugepage_global_always() || (has_madvise && inherit_enabled))
>> +               mask |= inherit;
> 
> Similarly, this can maybe become (not 100% sure if I got it right, the 
> condition above is confusing)

IMO, this is the original logic.

> if (!has_madvise) {
>      /*
>       * Without VM_HUGEPAGE, only allow orders that are set to
>       * ALWAYS directly ...
>        */
>      order &= (always | inherit);
>      /* ... or through inheritance. */
>      if (!hugepage_global_always())
>          orders &= ~inherit;

Ditto. I can not understand this too.

> }
>