On 11/02/2025 00:30, Nico Pache wrote:
> khugepaged scans PMD ranges for potential collapse to a hugepage. To add
> mTHP support we use this scan to instead record chunks of fully utilized
> sections of the PMD.
>
> create a bitmap to represent a PMD in order MTHP_MIN_ORDER chunks.
> by default we will set this to order 3. The reasoning is that for 4K 512
I'm still a bit confused by this (hopefully to be resolved as I'm about to read
the code); Does this imply that the smallest order you can collapse to is order
3? Because that would be different from the fault handler. For anon memory we
can support order-2 and above. I believe that these days files can support order-1.
There is a case for wanting to support order-2 for arm64. We have a (not yet
well deployed) technology called Hardware Page Aggregation (HPA) which can
automatically (transparent to SW) aggregate (usually) 4 contiguous pages into a
single TLB. I'd like the solution to be compatible with that.
> PMD size this results in a 64 bit bitmap which has some optimizations.
> For other arches like ARM64 64K, we can set a larger order if needed.
>
> khugepaged_scan_bitmap uses a stack struct to recursively scan a bitmap
> that represents chunks of utilized regions. We can then determine what
> mTHP size fits best and in the following patch, we set this bitmap while
> scanning the PMD.
>
> max_ptes_none is used as a scale to determine how "full" an order must
> be before being considered for collapse.
>
> If a order is set to "always" lets always collapse to that order in a
> greedy manner.
This is not the policy that the fault handler uses, and I think we should use
the same policy in both places.
The fault handler gets a list of orders that are permitted for the VMA, then
prefers the highest orders in that list.
I don't think we should be preferring a smaller "always" order over a larger
"madvise" order if MADV_HUGEPAGE is set for the VMA (if that's what your
statement was suggesting).
>
> Signed-off-by: Nico Pache <npache@xxxxxxxxxx>
> ---
> include/linux/khugepaged.h | 4 ++
> mm/khugepaged.c | 89 +++++++++++++++++++++++++++++++++++---
> 2 files changed, 86 insertions(+), 7 deletions(-)
>
> diff --git a/include/linux/khugepaged.h b/include/linux/khugepaged.h
> index 1f46046080f5..1fe0c4fc9d37 100644
> --- a/include/linux/khugepaged.h
> +++ b/include/linux/khugepaged.h
> @@ -1,6 +1,10 @@
> /* SPDX-License-Identifier: GPL-2.0 */
> #ifndef _LINUX_KHUGEPAGED_H
> #define _LINUX_KHUGEPAGED_H
> +#define MIN_MTHP_ORDER 3
> +#define MIN_MTHP_NR (1<<MIN_MTHP_ORDER)
> +#define MAX_MTHP_BITMAP_SIZE (1 << (ilog2(MAX_PTRS_PER_PTE * PAGE_SIZE) - MIN_MTHP_ORDER))
I don't think you want "* PAGE_SIZE" here? I think MAX_MTHP_BITMAP_SIZE wants to
specify the maximum number of groups of MIN_MTHP_NR pte entries in a PTE table?
In that case, MAX_PTRS_PER_PTE will be 512 on x86_64. Your current formula will
give 262144 (which is 32KB!). I think you just need:
#define MAX_MTHP_BITMAP_SIZE (1 << (ilog2(MAX_PTRS_PER_PTE) - MIN_MTHP_ORDER))
> +#define MTHP_BITMAP_SIZE (1 << (HPAGE_PMD_ORDER - MIN_MTHP_ORDER))
Perhaps all of these macros need a KHUGEPAGED_ prefix? Otherwise MIN_MTHP_ORDER,
especially, is misleading. The min MTHP order is not 3.
>
> extern unsigned int khugepaged_max_ptes_none __read_mostly;
> #ifdef CONFIG_TRANSPARENT_HUGEPAGE
> diff --git a/mm/khugepaged.c b/mm/khugepaged.c
> index 3776055bd477..c8048d9ec7fb 100644
> --- a/mm/khugepaged.c
> +++ b/mm/khugepaged.c
> @@ -94,6 +94,11 @@ static DEFINE_READ_MOSTLY_HASHTABLE(mm_slots_hash, MM_SLOTS_HASH_BITS);
>
> static struct kmem_cache *mm_slot_cache __ro_after_init;
>
> +struct scan_bit_state {
> + u8 order;
> + u16 offset;
> +};
> +
> struct collapse_control {
> bool is_khugepaged;
>
> @@ -102,6 +107,15 @@ struct collapse_control {
>
> /* nodemask for allocation fallback */
> nodemask_t alloc_nmask;
> +
> + /* bitmap used to collapse mTHP sizes. 1bit = order MIN_MTHP_ORDER mTHP */
> + DECLARE_BITMAP(mthp_bitmap, MAX_MTHP_BITMAP_SIZE);
> + DECLARE_BITMAP(mthp_bitmap_temp, MAX_MTHP_BITMAP_SIZE);
> + struct scan_bit_state mthp_bitmap_stack[MAX_MTHP_BITMAP_SIZE];
> +};
> +
> +struct collapse_control khugepaged_collapse_control = {
> + .is_khugepaged = true,
> };
>
> /**
> @@ -851,10 +865,6 @@ static void khugepaged_alloc_sleep(void)
> remove_wait_queue(&khugepaged_wait, &wait);
> }
>
> -struct collapse_control khugepaged_collapse_control = {
> - .is_khugepaged = true,
> -};
> -
> static bool khugepaged_scan_abort(int nid, struct collapse_control *cc)
> {
> int i;
> @@ -1112,7 +1122,8 @@ static int alloc_charge_folio(struct folio **foliop, struct mm_struct *mm,
>
> static int collapse_huge_page(struct mm_struct *mm, unsigned long address,
> int referenced, int unmapped,
> - struct collapse_control *cc)
> + struct collapse_control *cc, bool *mmap_locked,
> + u8 order, u16 offset)
> {
> LIST_HEAD(compound_pagelist);
> pmd_t *pmd, _pmd;
> @@ -1130,8 +1141,12 @@ static int collapse_huge_page(struct mm_struct *mm, unsigned long address,
> * The allocation can take potentially a long time if it involves
> * sync compaction, and we do not need to hold the mmap_lock during
> * that. We will recheck the vma after taking it again in write mode.
> + * If collapsing mTHPs we may have already released the read_lock.
> */
> - mmap_read_unlock(mm);
> + if (*mmap_locked) {
> + mmap_read_unlock(mm);
> + *mmap_locked = false;
> + }
>
> result = alloc_charge_folio(&folio, mm, cc, HPAGE_PMD_ORDER);
> if (result != SCAN_SUCCEED)
> @@ -1266,12 +1281,71 @@ static int collapse_huge_page(struct mm_struct *mm, unsigned long address,
> out_up_write:
> mmap_write_unlock(mm);
> out_nolock:
> + *mmap_locked = false;
> if (folio)
> folio_put(folio);
> trace_mm_collapse_huge_page(mm, result == SCAN_SUCCEED, result);
> return result;
> }
>
> +// Recursive function to consume the bitmap
> +static int khugepaged_scan_bitmap(struct mm_struct *mm, unsigned long address,
> + int referenced, int unmapped, struct collapse_control *cc,
> + bool *mmap_locked, unsigned long enabled_orders)
> +{
> + u8 order, next_order;
> + u16 offset, mid_offset;
> + int num_chunks;
> + int bits_set, threshold_bits;
> + int top = -1;
> + int collapsed = 0;
> + int ret;
> + struct scan_bit_state state;
> +
> + cc->mthp_bitmap_stack[++top] = (struct scan_bit_state)
> + { HPAGE_PMD_ORDER - MIN_MTHP_ORDER, 0 };
> +
> + while (top >= 0) {
> + state = cc->mthp_bitmap_stack[top--];
> + order = state.order + MIN_MTHP_ORDER;
> + offset = state.offset;
> + num_chunks = 1 << (state.order);
> + // Skip mTHP orders that are not enabled
> + if (!test_bit(order, &enabled_orders))
> + goto next;
> +
> + // copy the relavant section to a new bitmap
> + bitmap_shift_right(cc->mthp_bitmap_temp, cc->mthp_bitmap, offset,
> + MTHP_BITMAP_SIZE);
> +
> + bits_set = bitmap_weight(cc->mthp_bitmap_temp, num_chunks);
> + threshold_bits = (HPAGE_PMD_NR - khugepaged_max_ptes_none - 1)
> + >> (HPAGE_PMD_ORDER - state.order);
> +
> + //Check if the region is "almost full" based on the threshold
> + if (bits_set > threshold_bits
> + || test_bit(order, &huge_anon_orders_always)) {
> + ret = collapse_huge_page(mm, address, referenced, unmapped, cc,
> + mmap_locked, order, offset * MIN_MTHP_NR);
> + if (ret == SCAN_SUCCEED) {
> + collapsed += (1 << order);
> + continue;
> + }
> + }
> +
> +next:
> + if (state.order > 0) {
> + next_order = state.order - 1;
> + mid_offset = offset + (num_chunks / 2);
> + cc->mthp_bitmap_stack[++top] = (struct scan_bit_state)
> + { next_order, mid_offset };
> + cc->mthp_bitmap_stack[++top] = (struct scan_bit_state)
> + { next_order, offset };
> + }
> + }
> + return collapsed;
> +}
I'm struggling to understand the details of this function. I'll come back to it
when I have more time.
> +
> static int khugepaged_scan_pmd(struct mm_struct *mm,
> struct vm_area_struct *vma,
> unsigned long address, bool *mmap_locked,
> @@ -1440,7 +1514,7 @@ static int khugepaged_scan_pmd(struct mm_struct *mm,
> pte_unmap_unlock(pte, ptl);
> if (result == SCAN_SUCCEED) {
> result = collapse_huge_page(mm, address, referenced,
> - unmapped, cc);
> + unmapped, cc, mmap_locked, HPAGE_PMD_ORDER, 0);
> /* collapse_huge_page will return with the mmap_lock released */
> *mmap_locked = false;
Given that collapse_huge_page() now takes mmap_locked and sets it to false on
return, I don't think we need this line here any longer?
> }
> @@ -2856,6 +2930,7 @@ int madvise_collapse(struct vm_area_struct *vma, struct vm_area_struct **prev,
> mmdrop(mm);
> kfree(cc);
>
> +
> return thps == ((hend - hstart) >> HPAGE_PMD_SHIFT) ? 0
> : madvise_collapse_errno(last_fail);
> }
