[PATCH] lib: sbi: Fix shift bug in sbi_system_reset

[Anup Patel]

On Wed, Dec 27, 2023 at 7:25 AM Xiang W <wxjstz at 126.com> wrote:
>
> 在 2023-12-26星期二的 12:03 -0600，Samuel Holland写道：
> > On 2023-12-26 9:51 AM, Anup Patel wrote:
> > > On Mon, Dec 25, 2023 at 3:19 PM Andreas Schwab <schwab at linux-m68k.org> wrote:
> > > >
> > > > On Dez 25 2023, Anup Patel wrote:
> > > >
> > > > > If "cur_hartid < hbase + BITS_PER_LONG" then
> > > > > "1UL << (cur_hartid - hbase) == 0x0"
> > > >
> > > > If cur_hartid - hbase < BITS_PER_LONG, then 1UL << (cur_hartid - hbase)
> > > > will not be 0.  If cur_hartid - hbase >= BITS_PER_LONG, then 1UL <<
> > > > (cur_hartid - hbase) will overflow and be undefined.
> > >
> > > I meant "cur_hartid >= hbase + BITS_PER_LONG" which is
> > > the overflow case.
> > >
> > > Since the overflow behavior is undefined, we have the
> > > explicit check added by this patch.
> >
> > As pointed out by Andreas, "cur_hartid - hbase < BITS_PER_LONG" is the only
> > check needed here. Unsigned wraparound will take care of the case where
> > cur_hartid < hbase.
> cur_hartid < hbase is not always true.
>
> For example, cur_hartid is 3 and 0-100 belongs to a domain. Here the loop
> is executed twice, the first time with hbase 0 and the second time with
> hbase 64. The second loop will have cur_hartid < hbase.

The real issue is that overflow behaviour of the left shift operator is
undefined. No need for further justification on this issue.

Regards,
Anup

>
> Regards,
> Xiang W
> >
> > Regards,
> > Samuel
>
>
> --
> opensbi mailing list
> opensbi at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/opensbi