Hi Geert!
On Tue, 2023-04-18 at 09:14 +0200, Geert Uytterhoeven wrote:
> Hi Adrian,
>
> On Tue, Apr 18, 2023 at 8:36 AM John Paul Adrian Glaubitz
> <glaubitz@xxxxxxxxxxxxxxxxxxx> wrote:
> > Thanks for your patch. The changes look good to me. However, I have
> > one question, see below.
> >
> > On Sun, 2023-04-16 at 21:05 +0200, Christophe JAILLET wrote:
> > > Using the bitmap API is less verbose than hand writing them.
> > > It also improves the semantic.
> > >
> > > Signed-off-by: Christophe JAILLET <christophe.jaillet@xxxxxxxxxx>
>
> > > --- a/arch/sh/kernel/cpu/sh4/sq.c
> > > +++ b/arch/sh/kernel/cpu/sh4/sq.c
> > > @@ -372,7 +372,6 @@ static struct subsys_interface sq_interface = {
> > > static int __init sq_api_init(void)
> > > {
> > > unsigned int nr_pages = 0x04000000 >> PAGE_SHIFT;
> > > - unsigned int size = (nr_pages + (BITS_PER_LONG - 1)) / BITS_PER_LONG;
> > > int ret = -ENOMEM;
> > >
> > > printk(KERN_NOTICE "sq: Registering store queue API.\n");
> > > @@ -382,7 +381,7 @@ static int __init sq_api_init(void)
> > > if (unlikely(!sq_cache))
> > > return ret;
> > >
> > > - sq_bitmap = kzalloc(size, GFP_KERNEL);
> > > + sq_bitmap = bitmap_zalloc(nr_pages, GFP_KERNEL);
> > > if (unlikely(!sq_bitmap))
> > > goto out;
> > >
> >
> > I have look through other patches where k{z,c,m}alloc() were replaced with
> > bitmap_zalloc() and I noticed that in the other cases such as [1], kcalloc()
> > was used instead of kzalloc() in our cases with the element size set to
> > sizeof(long) while kzalloc() is using an element size equal to a byte.
> >
> > Wouldn't that mean that the current code in sq is allocating a buffer that is
> > too small by a factor of 1/sizeof(long) or am I missing something?
> >
> > @Geert: Do you have any idea?
>
> Nice catch!
>
> Looking more deeply at the code, the intention is to allocate a bitmap
> with nr_pages bits, so the code fater Christophe's patch is correct.
> However, the old code is indeed wrong:
>
> (nr_pages + (BITS_PER_LONG - 1)) / BITS_PER_LONG
>
> The aim is to calculate the size in bytes, rounded up to an integral
> number of longs, but it lacks a final multiplication by BITS_PER_BYTE,
> so it's off by a factor of 4.
Yeah, that's what I understood from reading the code which is why I was
wondering why the factor was missing.
> Fixes: d7c30c682a278abe ("sh: Store Queue API rework.")
>
> As we didn't have bitmap_zalloc() until commit c42b65e363ce97a8
> ("bitmap: Add bitmap_alloc(), bitmap_zalloc() and bitmap_free()")
> in v4.19, it would be good to fix the bug first in a separate patch,
> not using
I agree. Do you want to send a patch I can review?
> BTW, interesting how this got missed when fixing the other out-of-range
> bug in commit 9f650cf2b811cfb6 ("sh: Fix store queue bitmap end.",
> s/marc.theaimsgroup.com/marc.info/ when following the link).
Yeah.
PS: Sorry for the slightly messy grammar in my previous mail, didn't have
a coffee yet that early in the morning :-).
Adrian
--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913
