[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <5c04998fb225430e8e4966cc95f8a0ea@AcuMS.aculab.com>
Date: Tue, 20 Jun 2023 10:41:41 +0000
From: David Laight <David.Laight@...LAB.COM>
To: 'Andrew Morton' <akpm@...ux-foundation.org>,
Wang Ming <machel@...o.com>
CC: Mike Rapoport <rppt@...nel.org>,
"linux-mm@...ck.org" <linux-mm@...ck.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"opensource.kernel@...o.com" <opensource.kernel@...o.com>
Subject: RE: [PATCH v1] mm:Using div64_ul replaces do_div() function
From: Andrew Morton
> Sent: 16 June 2023 20:52
>
> On Thu, 15 Jun 2023 15:51:14 +0800 Wang Ming <machel@...o.com> wrote:
>
> > Fix the following coccicheck warning:
> >
> > mm/mm_init.c:2492:2-8: WARNING: do_div() does a
> > 64-by-32 division, please consider using div64_ul instead.
>
> I guess coccicheck was fooled by the incorrect comment over do_div().
> do_div() returns an unsigned long.
IIRC do_div() does an in-place update, but div64_ul() returns the result.
So the 'fixed' code is entirely broken.
The coccicheck message should really be 'check the domain/type of the divisor'.
> alloc_large_system_hash() should be using a u64 for `max'. Nobody uses
> unsigned long long any more.
Probably makes no difference.
The issue is with 'buckesize' - probably 'long' so 64bit on 64bit systems.
I guess 'max' can overflow 'long' on 32bit PAE.
Although using min(nr_all_pages, 0x80000000UL >> PAGE_SHIFT >> 4)
and doing everything as 'unsigned long' might be better.
David
>
> > --- a/mm/mm_init.c
> > +++ b/mm/mm_init.c
> > @@ -2508,7 +2508,7 @@ void *__init alloc_large_system_hash(const char *tablename,
> > /* limit allocation size to 1/16 total memory by default */
> > if (max == 0) {
> > max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
> > - do_div(max, bucketsize);
> > + div64_ul(max, bucketsize);
> > }
> > max = min(max, 0x80000000ULL);
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
Powered by blists - more mailing lists