Re: [PATCH] mm-treewide-redefine-max_order-sanely-fix.txt

[Mike Rapoport]

On Thu, Apr 06, 2023 at 06:10:15PM +0300, Kirill A. Shutemov wrote:
> On Thu, Apr 06, 2023 at 06:57:41AM -0700, Guenter Roeck wrote:
> > On 4/6/23 00:25, Kirill A. Shutemov wrote:
> > > On Wed, Apr 05, 2023 at 10:20:26PM -0700, Guenter Roeck wrote:
> > > > Hi,
> > > > 
> > > > On Wed, Mar 15, 2023 at 06:38:00PM +0300, Kirill A. Shutemov wrote:
> > > > > fix min() warning
> > > > > 
> > > > > Link: https://lkml.kernel.org/r/20230315153800.32wib3n5rickolvh@box
> > > > > Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > > > >    Link: https://lore.kernel.org/oe-kbuild-all/202303152343.D93IbJmn-lkp@xxxxxxxxx/
> > > > > Signed-off-by: "Kirill A. Shutemov" <kirill@xxxxxxxxxxxxx>
> > > > > Cc: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx>
> > > > > Cc: Zi Yan <ziy@xxxxxxxxxx>
> > > > > Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > > > 
> > > > This patch results in various boot failures (hang) on arm targets
> > > > in linux-next. Debug messages reveal the reason.
> > > > 
> > > > ########### MAX_ORDER=10 start=0 __ffs(start)=-1 min()=10 min_t=-1
> > > >                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > > > 
> > > > If start==0, __ffs(start) returns 0xfffffff or (as int) -1, which min_t()
> > > > interprets as such, while min() apparently uses the returned unsigned long
> > > > value. Obviously a negative order isn't received well by the rest of the
> > > > code.
> > > 
> > > Actually, __ffs() is not defined for 0.
> > > 
> > > Maybe something like this?
> > > 
> > > diff --git a/mm/memblock.c b/mm/memblock.c
> > > index 7911224b1ed3..63603b943bd0 100644
> > > --- a/mm/memblock.c
> > > +++ b/mm/memblock.c
> > > @@ -2043,7 +2043,11 @@ static void __init __free_pages_memory(unsigned long start, unsigned long end)
> > >   	int order;
> > >   	while (start < end) {
> > > -		order = min_t(int, MAX_ORDER, __ffs(start));
> > > +		/* __ffs() behaviour is undefined for 0 */
> > > +		if (start)
> > > +			order = min_t(int, MAX_ORDER, __ffs(start));
> > > +		else
> > > +			order = MAX_ORDER;
> > 
> > Shouldn't that be
> > 		else
> > 			order = 0;
> > ?
> 
> +Mike.
> 
> No. start == 0 is MAX_ORDER-aligned. We want to free the pages in the
> largest chunks alignment allows.

Right. Before the changes to MAX_ORDER it was

		order = min(MAX_ORDER - 1UL, __ffs(start));

which would evaluate to 10.

I'd just prefer the comment to include the explanation about why we choose
MAX_ORDER for start == 0. Say

	/*
	 * __ffs() behaviour is undefined for 0 and we want to free the
	 * pages in the largest chunks alignment allows, so set order to
	 * MAX_ORDER when start == 0
	 */

> -- 
>   Kiryl Shutsemau / Kirill A. Shutemov

-- 
Sincerely yours,
Mike.