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

[Xiang W]

在 2023-12-25星期一的 14:54 +0530，Anup Patel写道：
> On Thu, Dec 21, 2023 at 7:14 AM Xiang W <wxjstz at 126.com> wrote:
> > 
> > There is a problem with judging whether the current hart belongs to
> > hmask. If cur_hartid minus hbase is greater than BITS_PER_LONG, the
> > previous hmask will also have a bit cleared incorrectly, which will
> > cause some harts to lose ipi.
> > 
> > Signed-off-by: Xiang W <wxjstz at 126.com>
> > ---
> >  lib/sbi/sbi_system.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/lib/sbi/sbi_system.c b/lib/sbi/sbi_system.c
> > index d1fa349..e930272 100644
> > --- a/lib/sbi/sbi_system.c
> > +++ b/lib/sbi/sbi_system.c
> > @@ -72,7 +72,8 @@ void __noreturn sbi_system_reset(u32 reset_type, u32 reset_reason)
> > 
> >         /* Send HALT IPI to every hart other than the current hart */
> >         while (!sbi_hsm_hart_interruptible_mask(dom, hbase, &hmask)) {
> > -               if (hbase <= cur_hartid)
> > +               if ((hbase <= cur_hartid)
> > +                         && (cur_hartid < hbase + BITS_PER_LONG))
> 
> If "cur_hartid < hbase + BITS_PER_LONG" then
> "1UL << (cur_hartid - hbase) == 0x0" so hmask
> won't be impacted.
> 
> Do we still need this additional check ?
Yes, needed.

1UL << (cur_hartid - hbase) is equal to 1UL << ((cur_hartid - hbase) % XLEN)
is not equal to 0.

Regards,
Xiang W
> 
> >                         hmask &= ~(1UL << (cur_hartid - hbase));
> >                 if (hmask)
> >                         sbi_ipi_send_halt(hmask, hbase);
> > --
> > 2.43.0
> > 
> > 
> > --
> > opensbi mailing list
> > opensbi at lists.infradead.org
> > http://lists.infradead.org/mailman/listinfo/opensbi
> 
> Regards,
> Anup