Re: [PATCH] earlycon: prevent multiple register_console()

[Michael Walle]

Hi,

Am 2022-06-01 18:39, schrieb Rob Herring:
> On Wed, Jun 1, 2022 at 8:55 AM Michael Walle <michael@xxxxxxxx> wrote:
> > If the earlycon parameter is given twice, the kernel will spit out a
> > WARN() in register_console() because it was already registered. The
> > non-dt variant setup_earlycon() already handles that gracefully. The 
> > dt
> > variant of_setup_earlycon() doesn't. Add the check there and propagate
> > it through early_init_dt_scan_chosen_stdout().
> >
> > FWIW, this doesn't happen if CONFIG_ACPI_SPCR_TABLE is set. In that 
> > case
> > the registration is delayed until after earlycon parameter(s) are
> > parsed.
> >
> > Signed-off-by: Michael Walle <michael@xxxxxxxx>
> > ---


Scrap this version. Somehow I missed to add the actual check in
of_setup_earlycon() to this patch.

> > I'm not sure if this should have a Fixes tag or not. If so I guess it
> > should be the very first commit which introduced the support (commit
> > fb11ffe74c79 ("of/fdt: add FDT serial scanning for earlycon")).
>
> Considering no one noticed or cared in 8 years and that earlycon is
> supposed to be a debug option, I'd say no.
>
> > For the curious, here is the backtrace:
> >
> > [    0.000000] ------------[ cut here ]------------
> > [    0.000000] WARNING: CPU: 0 PID: 0 at kernel/printk/printk.c:3328 
> > register_console+0x2b4/0x364
> > [    0.000000] console 'atmel_serial0' already registered
> > [    0.000000] Modules linked in:
> > [    0.000000] CPU: 0 PID: 0 Comm: swapper Not tainted 
> > 5.18.0-next-20220601+ #652
> > [    0.000000] Hardware name: Generic DT based system
> > [    0.000000] Backtrace:
> > [    0.000000]  dump_backtrace from show_stack+0x18/0x1c
> > [    0.000000]  show_stack from dump_stack_lvl+0x48/0x54
> > [    0.000000]  dump_stack_lvl from dump_stack+0x18/0x1c
> > [    0.000000]  dump_stack from __warn+0xd0/0x148
> > [    0.000000]  __warn from warn_slowpath_fmt+0x9c/0xc4
> > [    0.000000]  warn_slowpath_fmt from register_console+0x2b4/0x364
> > [    0.000000]  register_console from of_setup_earlycon+0x29c/0x2ac
> > [    0.000000]  of_setup_earlycon from 
> > early_init_dt_scan_chosen_stdout+0x154/0x18c
> > [    0.000000]  early_init_dt_scan_chosen_stdout from 
> > param_setup_earlycon+0x40/0x48
> > [    0.000000]  param_setup_earlycon from do_early_param+0x88/0xc4
> > [    0.000000]  do_early_param from parse_args+0x1a4/0x404
> > [    0.000000]  parse_args from parse_early_options+0x40/0x48
> > [    0.000000]  parse_early_options from parse_early_param+0x38/0x48
> > [    0.000000]  parse_early_param from setup_arch+0x114/0x7a4
> > [    0.000000]  setup_arch from start_kernel+0x74/0x6dc
> > [    0.000000]  start_kernel from 0x0
> > [    0.000000] ---[ end trace 0000000000000000 ]---
> >
> >  drivers/of/fdt.c              | 6 ++++--
> >  drivers/tty/serial/earlycon.c | 5 ++++-
> >  2 files changed, 8 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/of/fdt.c b/drivers/of/fdt.c
> > index a8f5b6532165..7f3524213b43 100644
> > --- a/drivers/of/fdt.c
> > +++ b/drivers/of/fdt.c
> > @@ -1025,6 +1025,7 @@ int __init 
> > early_init_dt_scan_chosen_stdout(void)
> >         int l;
> >         const struct earlycon_id *match;
> >         const void *fdt = initial_boot_params;
> > +       int ret;
> >
> >         offset = fdt_path_offset(fdt, "/chosen");
> >         if (offset < 0)
> > @@ -1057,8 +1058,9 @@ int __init 
> > early_init_dt_scan_chosen_stdout(void)
> >                 if (fdt_node_check_compatible(fdt, offset, 
> > match->compatible))
> >                         continue;
> >
> > -               if (of_setup_earlycon(match, offset, options) == 0)
> > -                       return 0;
> > +               ret = of_setup_earlycon(match, offset, options);
> > +               if (!ret || ret == -EALREADY)
> > +                       return ret;
>
> Wouldn't just doing a 'return 0' here be simpler? The only other
> caller of this function doesn't check the error.

Ok.

-michael