Re: [PATCH v5 09/17] mm/sparse: initialize memory sections earlier

[Muchun Song <muchun.song@linux.dev>]

> On Sep 9, 2026, at 17:56, David Hildenbrand (Arm) <david@kernel.org> wrote:
> 
> On 8/25/26 10:46, Muchun Song wrote:
>> Upcoming HugeTLB bootmem changes need sparsemem section metadata before
>> the HugeTLB bootmem allocation path runs. The memory sections are
>> initialized from sparse_init(), which is called too late for that setup.
>> 
>> Move the code that initializes sparsemem section metadata for memblock
>> ranges into mm_core_init_early(), before free_area_init() and the HugeTLB
>> bootmem setup. Rename the helper to sparse_sections_init() so the new
>> caller describes the sparsemem-specific initialization step.
>> 
>> This is a preparatory change.
>> 
>> Signed-off-by: Muchun Song <songmuchun@bytedance.com>
>> Reviewed-by: Mike Rapoport (Microsoft) <rppt@kernel.org>
>> ---
>> v5:
>> - Collect Reviewed-by from Mike Rapoport
>> 
>> v2:
>> - Rename the helper to sparse_sections_init() to describe the section
>>  metadata initialization (suggested by Mike Rapoport)
>> - Fix the !SPARSEMEM stub name so SPARSEMEM=n builds compile
>>  (reported by Sashiko)
>> ---
>> mm/mm_init.c |  1 +
>> mm/sparse.c  | 10 ++--------
>> mm/sparse.h  |  2 ++
>> 3 files changed, 5 insertions(+), 8 deletions(-)
>> 
>> diff --git a/mm/mm_init.c b/mm/mm_init.c
>> index 0317fb781e77..e2a16d833635 100644
>> --- a/mm/mm_init.c
>> +++ b/mm/mm_init.c
>> @@ -2642,6 +2642,7 @@ void __init mm_core_init_early(void)
>> {
>> 	kho_memory_init_early();
>> 
>> + 	sparse_sections_init();
>> 	free_area_init();
>> 
>> hugetlb_cma_reserve();
>> diff --git a/mm/sparse.c b/mm/sparse.c
>> index e6cb67ca9c8d..439802e6a6ad 100644
>> --- a/mm/sparse.c
>> +++ b/mm/sparse.c
>> @@ -191,12 +191,8 @@ static void __init memory_present(int nid, unsigned long start, unsigned long en
>> }
>> }
>> 
>> -/*
>> - * Mark all memblocks as present using memory_present().
>> - * This is a convenience function that is useful to mark all of the systems
>> - * memory as present during initialization.
>> - */
>> -static void __init memblocks_present(void)
>> +/* Initialize memory section metadata for all system memory. */
> 
> I'd drop this comment, it's not entirely accurate. A lot more initialization of
> the sections happens later.

I can drop this line comment.

> 
> What this function mostly does is
> 
> (a) Allocate sparse-exteme metadata
> (b) Temporarily store the nid and mark the section for being present and online

Yes.

> 
> I have patches to get rid of __section_mark_present() and just set them all to
> SECTION_IS_EARLY|SECTION_IS_ONLINE such that we can just iterate early sections
> and get rid of that PRESENT monstrosity.
> 
> Apart from that LGTM, even though I don't particularly enjoy splitting us now
> having two init entry points
> 
> sparse_init

Maybe this one should be sparse_memmap_init? If yes, it could be another separate
cleanup patch.

Thanks,
Muchun

> 
> and
> 
> sparse_sections_init
> 
> 
> Where it's rather unclear from the name what is actually happening.
> 
> -- 
> Cheers,
> 
> David