Re: [PATCH 5/5] mm/thp: Split huge pmds/puds if they're pinned when fork()

[John Hubbard]

On 9/21/20 2:20 PM, Peter Xu wrote:
...
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 7ff29cc3d55c..c40aac0ad87e 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -1074,6 +1074,23 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
>   
>   	src_page = pmd_page(pmd);
>   	VM_BUG_ON_PAGE(!PageHead(src_page), src_page);
> +
> +	/*
> +	 * If this page is a potentially pinned page, split and retry the fault
> +	 * with smaller page size.  Normally this should not happen because the
> +	 * userspace should use MADV_DONTFORK upon pinned regions.  This is a
> +	 * best effort that the pinned pages won't be replaced by another
> +	 * random page during the coming copy-on-write.
> +	 */
> +	if (unlikely(READ_ONCE(src_mm->has_pinned) &&
> +		     page_maybe_dma_pinned(src_page))) {

This condition would make a good static inline function. It's used in 3 places,
and the condition is quite special and worth documenting, and having a separate
function helps with that, because the function name adds to the story. I'd suggest
approximately:

    page_likely_dma_pinned()

for the name.

> +		pte_free(dst_mm, pgtable);
> +		spin_unlock(src_ptl);
> +		spin_unlock(dst_ptl);
> +		__split_huge_pmd(vma, src_pmd, addr, false, NULL);
> +		return -EAGAIN;
> +	}


Why wait until we are so deep into this routine to detect this and unwind?
It seems like if you could do a check near the beginning of this routine, and
handle it there, with less unwinding? In fact, after taking only the src_ptl,
the check could be made, right?


thanks,
--
John Hubbard
NVIDIA