From: Andy Shevchenko
> Sent: 24 February 2022 19:51
>
> On Thu, Feb 24, 2022 at 2:51 PM Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx> wrote:
> >
> > On Wed, 23 Feb 2022 18:44:20 +0200
> > Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> wrote:
> >
> > > While in this particular case it would not be a (critical) issue,
> > > the pattern itself is bad and error prone in case somebody blindly
> > > copies to their code.
> > >
> > > Don't cast parameter to unsigned long pointer in the bit operations.
> > > Instead copy to a local variable on stack of a proper type and use.
>
> ...
>
> > > + struct { /* as a 256-bit bitmap */
> > > + DECLARE_BITMAP(b, 256);
> > > + } bitmap;
> > > + struct { /* as a set of 64-bit words */
> > > u64 word[4];
> > > } u64;
>
> > > - set_bit_inv(IPM_BIT_OFFSET + gisc, (unsigned long *) gisa);
> > > + set_bit_inv(IPM_BIT_OFFSET + gisc, gisa->bitmap.b);
> >
> > wouldn't it be enough to pass gisa->u64.word here?
> > then no cast would be necessary
>
> No, it will have the same hidden bugs. As I stated in the commit
> message, the pattern is quite bad even if in particular code it would
> work.
>
> Thanks, Michael, for pointing out other places. They all need to be fixed.
It may even be worth writing some alternate bitmap functions
that use u64[] and unlocked operations?
Although I think I'd still want to encapsulate the actual array
(somehow) so that what is defined has to be the bitmap type.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
