Re: [Question] Dead code in copy_pud_range() for copy_huge_pud() error check?

[Joshua Hahn <joshua.hahnjy@gmail.com>]

On Sun,  1 Feb 2026 02:28:54 +0900 Ingyu Jang <ingyujang25@korea.ac.kr> wrote:

Hello Ingyu,

I hope you are doing well!

I'm by no means an expert in this area, but thought I would reply with
what I found in the commit log.

> Hi,
> 
> I noticed that in mm/memory.c, the function copy_pud_range() checks
> if copy_huge_pud() returns -ENOMEM (at line 1421):
> 
>     err = copy_huge_pud(dst_mm, src_mm,
>                         dst_pud, src_pud, addr, src_vma);
>     if (err == -ENOMEM)
>         return -ENOMEM;
> 
> However, looking at copy_huge_pud() in mm/huge_memory.c (line 1966),
> it only returns two values:
>   - 0 on success (line 1994)
>   - -EAGAIN when !pud_trans_huge(pud) (line 1978)
> 
> There is no code path that returns -ENOMEM, which means the check
> "if (err == -ENOMEM)" at line 1421 is unreachable dead code.
> 
> Is this intentional defensive coding for potential future changes,
> or could this unreachable error check be cleaned up?

It seems like commit a00cc7d9dd93 ("mm, x86: add support for PUD-sized
transparent hugepages") introduced both the if (err == -ENOMEM) check and
copy_huge_pud, back in 2017.

And at the time these were introduced, copy_huge_pud only had return
values for -EAGAIN and 0, as you pointed out. So it doesn't seem like
there's any bug or missed case, just an extra case.

My guess is that copy_pud_range tried to imitate copy_pmd_range and added
the -ENOMEM check. copy_huge_pmd can return -ENOMEM though, so it seems
like this is what was missed when copying the function over.

If you have other changes in this area, this might be a good candidate
for a small fix-up to be included in the series? : -) The whole if
block inside pud_trans_huge(...) can be simplified quite a bit:

	do {
		...

		if (pud_trans_huge(*src_pud)) {
			VM_BUG_ON_VMA(next-addr != HPAGE_PUD_SIZE, src_vma);
			if (!copy_huge_pud(dst_mm, src_mm, dst_pud, src_pud,
					   addr, src_vma))
				continue;
		}

		...
	}

Anyways... Hope this answers your question! Have a great day,
Joshua