Re: [PATCH v7 mm-new 02/10] mm: thp: add support for BPF based THP order selection

[Lance Yang <lance.yang@linux.dev>]

On 2025/9/10 20:54, Lance Yang wrote:
> On Wed, Sep 10, 2025 at 8:42 PM Lance Yang <lance.yang@linux.dev> wrote:
>>
>> Hey Yafang,
>>
>> On Wed, Sep 10, 2025 at 10:53 AM Yafang Shao <laoar.shao@gmail.com> wrote:
>>>
>>> This patch introduces a new BPF struct_ops called bpf_thp_ops for dynamic
>>> THP tuning. It includes a hook bpf_hook_thp_get_order(), allowing BPF
>>> programs to influence THP order selection based on factors such as:
>>> - Workload identity
>>>    For example, workloads running in specific containers or cgroups.
>>> - Allocation context
>>>    Whether the allocation occurs during a page fault, khugepaged, swap or
>>>    other paths.
>>> - VMA's memory advice settings
>>>    MADV_HUGEPAGE or MADV_NOHUGEPAGE
>>> - Memory pressure
>>>    PSI system data or associated cgroup PSI metrics
>>>
>>> The kernel API of this new BPF hook is as follows,
>>>
>>> /**
>>>   * @thp_order_fn_t: Get the suggested THP orders from a BPF program for allocation
>>>   * @vma: vm_area_struct associated with the THP allocation
>>>   * @vma_type: The VMA type, such as BPF_THP_VM_HUGEPAGE if VM_HUGEPAGE is set
>>>   *            BPF_THP_VM_NOHUGEPAGE if VM_NOHUGEPAGE is set, or BPF_THP_VM_NONE if
>>>   *            neither is set.
>>>   * @tva_type: TVA type for current @vma
>>>   * @orders: Bitmask of requested THP orders for this allocation
>>>   *          - PMD-mapped allocation if PMD_ORDER is set
>>>   *          - mTHP allocation otherwise
>>>   *
>>>   * Return: The suggested THP order from the BPF program for allocation. It will
>>>   *         not exceed the highest requested order in @orders. Return -1 to
>>>   *         indicate that the original requested @orders should remain unchanged.
>>>   */
>>> typedef int thp_order_fn_t(struct vm_area_struct *vma,
>>>                             enum bpf_thp_vma_type vma_type,
>>>                             enum tva_type tva_type,
>>>                             unsigned long orders);
>>>
>>> Only a single BPF program can be attached at any given time, though it can
>>> be dynamically updated to adjust the policy. The implementation supports
>>> anonymous THP, shmem THP, and mTHP, with future extensions planned for
>>> file-backed THP.
>>>
>>> This functionality is only active when system-wide THP is configured to
>>> madvise or always mode. It remains disabled in never mode. Additionally,
>>> if THP is explicitly disabled for a specific task via prctl(), this BPF
>>> functionality will also be unavailable for that task.
>>>
>>> This feature requires CONFIG_BPF_GET_THP_ORDER (marked EXPERIMENTAL) to be
>>> enabled. Note that this capability is currently unstable and may undergo
>>> significant changes—including potential removal—in future kernel versions.
>>>
>>> Suggested-by: David Hildenbrand <david@redhat.com>
>>> Suggested-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
>>> Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
>>> ---
>> [...]
>>> diff --git a/mm/huge_memory_bpf.c b/mm/huge_memory_bpf.c
>>> new file mode 100644
>>> index 000000000000..525ee22ab598
>>> --- /dev/null
>>> +++ b/mm/huge_memory_bpf.c
>>> @@ -0,0 +1,243 @@
>>> +// SPDX-License-Identifier: GPL-2.0
>>> +/*
>>> + * BPF-based THP policy management
>>> + *
>>> + * Author: Yafang Shao <laoar.shao@gmail.com>
>>> + */
>>> +
>>> +#include <linux/bpf.h>
>>> +#include <linux/btf.h>
>>> +#include <linux/huge_mm.h>
>>> +#include <linux/khugepaged.h>
>>> +
>>> +enum bpf_thp_vma_type {
>>> +       BPF_THP_VM_NONE = 0,
>>> +       BPF_THP_VM_HUGEPAGE,    /* VM_HUGEPAGE */
>>> +       BPF_THP_VM_NOHUGEPAGE,  /* VM_NOHUGEPAGE */
>>> +};
>>> +
>>> +/**
>>> + * @thp_order_fn_t: Get the suggested THP orders from a BPF program for allocation
>>> + * @vma: vm_area_struct associated with the THP allocation
>>> + * @vma_type: The VMA type, such as BPF_THP_VM_HUGEPAGE if VM_HUGEPAGE is set
>>> + *            BPF_THP_VM_NOHUGEPAGE if VM_NOHUGEPAGE is set, or BPF_THP_VM_NONE if
>>> + *            neither is set.
>>> + * @tva_type: TVA type for current @vma
>>> + * @orders: Bitmask of requested THP orders for this allocation
>>> + *          - PMD-mapped allocation if PMD_ORDER is set
>>> + *          - mTHP allocation otherwise
>>> + *
>>> + * Return: The suggested THP order from the BPF program for allocation. It will
>>> + *         not exceed the highest requested order in @orders. Return -1 to
>>> + *         indicate that the original requested @orders should remain unchanged.
>>
>> A minor documentation nit: the comment says "Return -1 to indicate that the
>> original requested @orders should remain unchanged". It might be slightly
>> clearer to say "Return a negative value to fall back to the original
>> behavior". This would cover all error codes as well ;)
>>
>>> + */
>>> +typedef int thp_order_fn_t(struct vm_area_struct *vma,
>>> +                          enum bpf_thp_vma_type vma_type,
>>> +                          enum tva_type tva_type,
>>> +                          unsigned long orders);
>>
>> Sorry if I'm missing some context here since I haven't tracked the whole
>> series closely.
>>
>> Regarding the return value for thp_order_fn_t: right now it returns a
>> single int order. I was thinking, what if we let it return an unsigned
>> long bitmask of orders instead? This seems like it would be more flexible
>> down the road, especially if we get more mTHP sizes to choose from. It
>> would also make the API more consistent, as bpf_hook_thp_get_orders()
>> itself returns an unsigned long ;)
> 
> I just realized a flaw in my previous suggestion :(
> 
> Changing the return type of thp_order_fn_t to unsigned long for consistency
> and flexibility. However, I completely overlooked that this would prevent
> the BPF program from returning negative error codes ...
> 
> Thanks,
> Lance
> 
>>
>> Also, for future extensions, it might be a good idea to add a reserved
>> flags argument to the thp_order_fn_t signature.
>>
>> For example thp_order_fn_t(..., unsigned long flags).
>>
>> This would give us aforward-compatible way to add new semantics later
>> without breaking the ABI and needing a v2. We could just require it to be
>> 0 for now.
>>
>> Thanks for the great work!
>> Lance


Forgot to add:

Noticed that if the hook returns 0, bpf_hook_thp_get_orders() falls
back to 'orders', preventing us from dynamically disabling mTHP
allocations.

Honoring a return of 0 is critical for our use case, which is to
dynamically disable mTHP for low-priority containers when memory gets
low in mixed workloads.

And then re-enable it for them when memory is back above the low
watermark.