[PATCH v7 2/3] kho: fix deferred init of kho scratch
Michał Cłapiński
mclapinski at google.com
Thu Mar 19 11:17:48 PDT 2026
On Thu, Mar 19, 2026 at 8:54 AM Mike Rapoport <rppt at kernel.org> wrote:
>
> Hi,
>
> On Wed, Mar 18, 2026 at 01:36:07PM -0400, Zi Yan wrote:
> > On 18 Mar 2026, at 13:19, Michał Cłapiński wrote:
> > > On Wed, Mar 18, 2026 at 6:08 PM Zi Yan <ziy at nvidia.com> wrote:
> > >>
> > >> ## Call site analysis
> > >>
> > >> init_pageblock_migratetype() has nine call sites. The init call ordering
> > >> relevant to scratch is:
> > >>
> > >> ```
> > >> setup_arch()
> > >> zone_sizes_init() -> free_area_init() -> memmap_init_range() [1]
>
> Hmm, this is slightly outdated, but largely correct :)
>
> > >>
> > >> mm_init_free_all() / start_kernel():
> > >> kho_memory_init() -> kho_release_scratch() [2]
> > >> memblock_free_all()
> > >> free_low_memory_core_early()
> > >> memmap_init_reserved_pages()
> > >> reserve_bootmem_region() -> __init_deferred_page()
> > >> -> __init_page_from_nid() [3]
> > >> deferred init kthreads -> __init_page_from_nid() [4]
>
> And this is wrong, deferred init does not call __init_page_from_nid, only
> reserve_bootmem_region() does.
>
> And there's a case claude missed:
>
> hugetlb_bootmem_free_invalid_page() -> __init_page_from_nid() that
> shouldn't check for KHO. Well, at least until we have support for hugetlb
> persistence and most probably even afterwards.
>
> I don't think we should modify reserve_bootmem_region(). If there are
> reserved pages in a pageblock, it does not matter if it's initialized to
> MIGRATE_CMA. It only becomes important if the reserved pages freed, so we
> can update pageblock migrate type in free_reserved_area().
> When we boot with KHO, all memblock allocations come from scratch, so
> anything freed in free_reserved_area() should become CMA again.
What happens if the reserved area covers one page and that page is
pageblock aligned? Then it won't be marked as CMA until it is freed
and unmovable allocation might appear in that pageblock, right?
>
> > >> ```
> > >
> > > I don't understand this. deferred_free_pages() doesn't call
> > > __init_page_from_nid(). So I would clearly need to modify both
> > > deferred_free_pages and __init_page_from_nid.
>
> For deferred_free_pages() we don't need kho_scratch_overlap(), we already
> have memblock_region (almost) at hand and it's enough to check if it's
> MEMBLOCK_KHO_SCRATCH.
>
> Something along these lines (compile tested only) should do the trick:
>
> diff --git a/include/linux/memblock.h b/include/linux/memblock.h
> index 3e217414e12d..b9b1e0991ec8 100644
> --- a/include/linux/memblock.h
> +++ b/include/linux/memblock.h
> @@ -275,6 +275,8 @@ static inline void __next_physmem_range(u64 *idx, struct memblock_type *type,
> __for_each_mem_range(i, &memblock.reserved, NULL, NUMA_NO_NODE, \
> MEMBLOCK_NONE, p_start, p_end, NULL)
>
> +struct memblock_region *memblock_region_from_iter(u64 iterator);
> +
> static inline bool memblock_is_hotpluggable(struct memblock_region *m)
> {
> return m->flags & MEMBLOCK_HOTPLUG;
> diff --git a/mm/memblock.c b/mm/memblock.c
> index ae6a5af46bd7..9cf99f32279f 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -1359,6 +1359,16 @@ void __init_memblock __next_mem_range_rev(u64 *idx, int nid,
> *idx = ULLONG_MAX;
> }
>
> +__init_memblock struct memblock_region *memblock_region_from_iter(u64 iterator)
> +{
> + int index = iterator & 0xffffffff;
I'm not sure about this. __next_mem_range() has this code:
/*
* The region which ends first is
* advanced for the next iteration.
*/
if (m_end <= r_end)
idx_a++;
else
idx_b++;
Therefore, the index you get from this might be correct or it might
already be incremented.
> +
> + if (index < 0 || index >= memblock.memory.cnt)
> + return NULL;
> +
> + return &memblock.memory.regions[index];
> +}
> +
> /*
> * Common iterator interface used to define for_each_mem_pfn_range().
> */
> diff --git a/mm/mm_init.c b/mm/mm_init.c
> index cec7bb758bdd..96b25895ffbe 100644
> --- a/mm/mm_init.c
> +++ b/mm/mm_init.c
> @@ -1996,7 +1996,7 @@ unsigned long __init node_map_pfn_alignment(void)
>
> #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
> static void __init deferred_free_pages(unsigned long pfn,
> - unsigned long nr_pages)
> + unsigned long nr_pages, enum migratetype mt)
> {
> struct page *page;
> unsigned long i;
> @@ -2009,8 +2009,7 @@ static void __init deferred_free_pages(unsigned long pfn,
> /* Free a large naturally-aligned chunk if possible */
> if (nr_pages == MAX_ORDER_NR_PAGES && IS_MAX_ORDER_ALIGNED(pfn)) {
> for (i = 0; i < nr_pages; i += pageblock_nr_pages)
> - init_pageblock_migratetype(page + i, MIGRATE_MOVABLE,
> - false);
> + init_pageblock_migratetype(page + i, mt, false);
> __free_pages_core(page, MAX_PAGE_ORDER, MEMINIT_EARLY);
> return;
> }
> @@ -2020,8 +2019,7 @@ static void __init deferred_free_pages(unsigned long pfn,
>
> for (i = 0; i < nr_pages; i++, page++, pfn++) {
> if (pageblock_aligned(pfn))
> - init_pageblock_migratetype(page, MIGRATE_MOVABLE,
> - false);
> + init_pageblock_migratetype(page, mt, false);
> __free_pages_core(page, 0, MEMINIT_EARLY);
> }
> }
> @@ -2077,6 +2075,8 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
> u64 i = 0;
>
> for_each_free_mem_range(i, nid, 0, &start, &end, NULL) {
> + struct memblock_region *region = memblock_region_from_iter(i);
> + enum migratetype mt = MIGRATE_MOVABLE;
> unsigned long spfn = PFN_UP(start);
> unsigned long epfn = PFN_DOWN(end);
>
> @@ -2086,12 +2086,15 @@ deferred_init_memmap_chunk(unsigned long start_pfn, unsigned long end_pfn,
> spfn = max(spfn, start_pfn);
> epfn = min(epfn, end_pfn);
>
> + if (memblock_is_kho_scratch(region))
> + mt = MIGRATE_CMA;
> +
> while (spfn < epfn) {
> unsigned long mo_pfn = ALIGN(spfn + 1, MAX_ORDER_NR_PAGES);
> unsigned long chunk_end = min(mo_pfn, epfn);
>
> nr_pages += deferred_init_pages(zone, spfn, chunk_end);
> - deferred_free_pages(spfn, chunk_end - spfn);
> + deferred_free_pages(spfn, chunk_end - spfn, mt);
>
> spfn = chunk_end;
>
> --
> Sincerely yours,
> Mike.
More information about the kexec
mailing list