On Thu, Dec 19, 2024 at 5:02 PM Uros Bizjak <ubizjak@gmail.com> wrote:

> > > The intermediate variable in the PERCPU_PTR() macro results in a kernel
> > > panic on boot [1] due to a compiler bug seen when compiling the kernel
> > > (+ KASAN) with gcc 11.3.1, but not when compiling with latest gcc
> > > (v14.2)/clang(v18.1).
> > >
> > > To solve it, remove the intermediate variable (which is not needed) and
> > > keep the casting that resolves the address space checks.

[...]

> > >  include/linux/percpu-defs.h | 3 +--
> > >  1 file changed, 1 insertion(+), 2 deletions(-)
> > >
> > > diff --git a/include/linux/percpu-defs.h b/include/linux/percpu-defs.h
> > > index 35842d1e3879..573adb643d90 100644
> > > --- a/include/linux/percpu-defs.h
> > > +++ b/include/linux/percpu-defs.h
> > > @@ -222,8 +222,7 @@ do {                                                                        \
> > >
> > >  #define PERCPU_PTR(__p)                                                        \
> > >  ({                                                                     \
> > > -       unsigned long __pcpu_ptr = (__force unsigned long)(__p);        \
> > > -       (typeof(*(__p)) __force __kernel *)(__pcpu_ptr);                \
> > > +       (typeof(*(__p)) __force __kernel *)((__force unsigned long)(__p)); \
> > >  })
>
> Actually, you can simplify the above a bit by writing it as:
>
> #define PERCPU_PTR(__p)                            \
>     ((typeof(*(__p)) __force __kernel *)(__force unsigned long)(__p)) \

Andrew, please find attached a substitute patch "[PATCH 4/6] percpu:
Use TYPEOF_UNQUAL() in *_cpu_ptr() accessors" for your MM tree
relative to the above hotfix. The whole patch series (+ hotfix) has
been re-tested against the current mainline defconfig (+ KASAN),
compiled once with gcc-11.4.1 and once with gcc-14.2.1.

Uros.