On 14/07/2020 06.20, Sean Christopherson wrote:
> Revert to using 'unsigned long' instead of 'size_t' for free_pages() and
> get_order(). The recent change to size_t for free_pages() breaks i386
> with -Werror as the assert_msg() formats expect unsigned longs, whereas
> size_t is an 'unsigned int' on i386 (though both longs and ints are 4
> bytes).
>
> Message formatting aside, unsigned long is the correct choice given the
> current code base as alloc_pages() and free_pages_by_order() explicitly
> expect, work on, and/or assert on the size being an unsigned long.
>
> Fixes: 73f4b202beb39 ("lib/alloc_page: change some parameter types")
> Cc: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
> Cc: Andrew Jones <drjones@xxxxxxxxxx>
> Cc: Jim Mattson <jmattson@xxxxxxxxxx>
> Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> ---
[...]
> diff --git a/lib/bitops.h b/lib/bitops.h
> index 308aa86..dd015e8 100644
> --- a/lib/bitops.h
> +++ b/lib/bitops.h
> @@ -79,7 +79,7 @@ static inline bool is_power_of_2(unsigned long n)
> return n && !(n & (n - 1));
> }
>
> -static inline unsigned int get_order(size_t size)
> +static inline unsigned int get_order(unsigned long size)
> {
> return size ? fls(size) + !is_power_of_2(size) : 0;
> }
>
get_order() already used size_t when it was introduced in commit
f22e527df02ffaba ... is it necessary to switch it to unsigned long now?
Apart from that, this patch fixes the compilation problems, indeed, I
just checked it in the travis-CI.
Tested-by: Thomas Huth <thuth@xxxxxxxxxx>
