Re: [PATCH] Ensure that walk_page_range()'s start and end arepage-aligned

[Andrew Morton]

On Fri, 24 Feb 2012 11:19:25 -0800
Dan Smith <danms@xxxxxxxxxx> wrote:

>
> ...
>
>     The inner function walk_pte_range() increments "addr" by PAGE_SIZE after
>     each pte is processed, and only exits the loop if the result is equal to
>     "end". Current, if either (or both of) the starting or ending addresses
>     passed to walk_page_range() are not page-aligned, then we will never
>     satisfy that exit condition and begin calling the pte_entry handler with
>     bad data.
>     
>     To be sure that we will land in the right spot, this patch checks that
>     both "addr" and "end" are page-aligned in walk_page_range() before starting
>     the traversal.
>     
> ...
>
> --- a/mm/pagewalk.c
> +++ b/mm/pagewalk.c
> @@ -196,6 +196,11 @@ int walk_page_range(unsigned long addr, unsigned long end,
>  	if (addr >= end)
>  		return err;
>  
> +	if (WARN_ONCE((addr & ~PAGE_MASK) || (end & ~PAGE_MASK),
> +		      "address range is not page-aligned")) {
> +		return -EINVAL;
> +	}
> +
>  	if (!walk->mm)
>  		return -EINVAL;

Well...  why should we apply the patch?  Is there some buggy code which
is triggering the problem?  Do you intend to write some buggy code to
trigger the problem?  ;) 

IOW, what benefit is there to this change?

Also, as it's a developer-only thing we should arrange for the overhead
to vanish when CONFIG_DEBUG_VM=n?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/