Re: [PATCH] common: board_r: Initialize interrupts before watchdog

["J. Neuschäfer" <j.ne@posteo.net>]

(CC'ing Mario Six regarding MPC83xx timer driver in U-Boot)

On Thu, Feb 20, 2025 at 06:49:58AM -0700, Simon Glass wrote:
> Hi J,
> 
> On Tue, 18 Feb 2025 at 08:55, J. Neuschäfer via B4 Relay
> <devnull+j.ne.posteo.net@kernel.org> wrote:
> >
> > From: "J. Neuschäfer" <j.ne@posteo.net>
> >
> > On some platforms, initializing the watchdog driver enables a timer
> > interrupt. This of course requires the interrupt handlers to be
> > properly initialized, otherwise U-Boot may crash or run the timer
> > interrupt handler of a previous bootloader stage.
> >
> > To account for such systems, always initialize interrupts
> > (arch_initr_trap) before the watchdog (initr_watchdog).
> >
> > This problem was observed on a PowerPC MPC83xx board.
> >
> > Signed-off-by: J. Neuschäfer <j.ne@posteo.net>
> > ---
> > NOTE: This approach seems safe and fine to me, but an argument could be
> >       made that this should be fixed in the platform-specific drivers
> >       instead.  Please let me know what you think.
> >
> >
> > Rough stack trace (not sure if it should be part of the commit message):
> >
> >   initr_watchdog                (drivers/watchdog/wdt-uclass.c)
> >     device_probe(wdt@200)
> >       device_probe(timer)
> >         mpc8xxx_wdt_start       (drivers/watchdog/mpc8xxx_wdt.c)
> >           set_msr(get_msr() | MSR_EE);
> >
> >   arch_initr_trap               (arch/powerpc/lib/traps.c)
> >     trap_init                   (arch/powerpc/cpu/mpc83xx/start.S)
> > ---
> >  common/board_r.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/common/board_r.c b/common/board_r.c
> > index 179259b00de81f7ba9802fc5288e7c2b6e6f381a..f711cd237ae76d80ca2413017bfe131656f44180 100644
> > --- a/common/board_r.c
> > +++ b/common/board_r.c
> > @@ -652,11 +652,11 @@ static init_fnc_t init_sequence_r[] = {
> >         serial_initialize,
> >         initr_announce,
> >         dm_announce,
> > +       arch_initr_trap,
> >  #if CONFIG_IS_ENABLED(WDT)
> >         initr_watchdog,
> >  #endif
> >         INIT_FUNC_WATCHDOG_RESET
> > -       arch_initr_trap,
> >  #if defined(CONFIG_BOARD_EARLY_INIT_R)
> >         board_early_init_r,
> >  #endif
> >
> > ---
> > base-commit: 064556910e61044f1295162ceaad600582b66cda
> > change-id: 20250218-init-dab1fc72abd2
> >
> > Best regards,
> > --
> > J. Neuschäfer <j.ne@posteo.net>
> >
> >
> 
> The driver model way of doing this would be that your UCLASS_WDT
> driver calls uclass_first_device(UCLASS_IRQ) to make sure interrupts
> are ready.

This sounds good in principle, but would require some reworking of the
MPC83xx interrupt infrastructure, because it currently doesn't declare a
UCLASS_IRQ driver.

> The arch_initr_trap thing should probably not be used in new code.
> Also, we have interrupt_init() which sounds like it is similar?

This isn't really *new* code, the MPC83xx platform support has been in
U-Boot for a long time (I am touching it now due to a newly to be added
board, though).

I see various functions called interrupt_init(), one of them in
mpc83xx_timer.c (i.e. relevant to my problems):

  /*
   * TODO(mario.six@gdsys.cc): This should really be done by timer_init, and the
   * interrupt init should go into a interrupt driver.
   */
  int interrupt_init(void)
  { ... }

it seems someone has stumbled on this oddity before.

Another version of interrupt_init is in arch/powerpc/lib/interrupts.c.
Fortunately, as I just found out, my PowerPC MPC8314E board works fine
with the generic timer/interrupt support in arch/powerpc, with
CONFIG_TIMER=n. In other words, the problem is localized to
CONFIG_MPC83XX_TIMER.


J. Neuschäfer