Re: [RFC] pgflags_t

[Vlastimil Babka <vbabka@suse.cz>]

On 10/6/21 16:58, Matthew Wilcox wrote:
> David expressed some unease about the lack of typesafety in patches
> 1 & 2 of the page->slab conversion [1], and I'll admit to not being
> particularly a fan of passing around an unsigned long.  That crystallised
> in a discussion with Kent [2] about how to lock a page when you don't know
> its type (solution: every memory descriptor type starts with a
> pgflags_t)
> 
> So this patch is a step towards typesafety for pgflags_lock() by
> changing the type of page->flags from a bare unsigned long to a
> struct encapsulating an unsigned long.  A few users can already benefit
> from passing around a pgflags_t, but most just append '.f' to get at
> the unsigned long.  Also, most of them only do it for logging/tracing.
> 
> [1] https://lore.kernel.org/linux-mm/02a055cd-19d6-6e1d-59bb-e9e5f9f1da5b@redhat.com/
> [2] https://lore.kernel.org/linux-mm/YVyQpPuwIGFSSEQ8@casper.infradead.org/
> 
...
> diff --git a/mm/debug.c b/mm/debug.c
> index fae0f81ad831..48af9829aee9 100644
> --- a/mm/debug.c
> +++ b/mm/debug.c
> @@ -162,7 +162,7 @@ static void __dump_page(struct page *page)
>  out_mapping:
>  	BUILD_BUG_ON(ARRAY_SIZE(pageflag_names) != __NR_PAGEFLAGS + 1);
>  
> -	pr_warn("%sflags: %#lx(%pGp)%s\n", type, head->flags, &head->flags,
> +	pr_warn("%sflags: %#lx(%pGp)%s\n", type, head->flags.f, &head->flags.f,

The %pGp case (here and elsewhere) could perhaps take the new type, no?
Would need to change format_page_flags() and flags_string() in lib/vsprintf.c