Re: [PATCH] iov_iter: fix copy_page_from_iter_atomic() if KMAP_LOCAL_FORCE_MAP

[Matthew Wilcox <willy@infradead.org>]

On Sun, Oct 27, 2024 at 03:23:23PM -0700, Hugh Dickins wrote:
> generic/077 on x86_32 CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP=y with highmem,
> on huge=always tmpfs, issues a warning and then hangs (interruptibly):

> +++ b/lib/iov_iter.c
> @@ -461,6 +461,8 @@ size_t copy_page_from_iter_atomic(struct page *page, size_t offset,
>  		size_t bytes, struct iov_iter *i)
>  {
>  	size_t n, copied = 0;
> +	bool uses_kmap = IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) ||
> +			 PageHighMem(page);
>  
>  	if (!page_copy_sane(page, offset, bytes))
>  		return 0;
> @@ -471,7 +473,7 @@ size_t copy_page_from_iter_atomic(struct page *page, size_t offset,
>  		char *p;
>  
>  		n = bytes - copied;
> -		if (PageHighMem(page)) {
> +		if (uses_kmap) {
>  			page += offset / PAGE_SIZE;
>  			offset %= PAGE_SIZE;
>  			n = min_t(size_t, n, PAGE_SIZE - offset);

Urgh.  I've done this same optimisation elsewhere.

memcpy_from_folio:
                if (folio_test_highmem(folio) &&
                    chunk > PAGE_SIZE - offset_in_page(offset))
                        chunk = PAGE_SIZE - offset_in_page(offset);

also memcpy_to_folio(), folio_zero_tail(), folio_fill_tail(),
memcpy_from_file_folio()

I think that means we need a new predicate.  I don't have a good name
yet.  folio_kmap_can_access_multiple_pages() is a bit too wordy.  Anyone
think of a good one?