Re: [PATCH] spi: mark device nodes only in case of successful instantiation

[Wolfram Sang]

On Sun, Oct 16, 2016 at 10:49:11AM +0200, Geert Uytterhoeven wrote:
> Hi Ralf,
> 
> (Cc i2c)

Thanks for letting me know! Adding Pantelis to CC, as he is the original
author of OF_DYNAMIC. Please keep me in the loop.

> 
> On Fri, Oct 14, 2016 at 9:31 PM, Ralf Ramsauer
> <ralf@xxxxxxxxxxxxxxxxxxxxxx> wrote:
> > Instantiated SPI device nodes are marked with OF_POPULATE. This was
> > introduced in bd6c164. On unloading, loaded device nodes will of course
> > be unmarked. The problem are nodes the fail during initialisation: If a
> > node failed during registration, it won't be unloaded and hence never be
> > unmarked again.
> >
> > So if a SPI driver module is unloaded and reloaded, it will skip nodes
> > that failed before.
> >
> > Skip device nodes that are already populated and mark them only in case
> > of success.
> >
> > Fixes: bd6c164 ("spi: Mark instantiated device nodes with OF_POPULATE")
> > Signed-off-by: Ralf Ramsauer <ralf@xxxxxxxxxxxxxxxxxxxxxx>
> > Cc: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> > ---
> > Hi,
> >
> > imagine the following situation: you loaded a spi driver as module, but
> > it fails to instantiate, because of some reasons (e.g. some resources,
> > like gpios, might be in use in userspace).
> >
> > When reloading the driver, _all_ nodes, including previously failed
> > ones, should be probed again. This is not the case at the moment.
> > Current behaviour only re-registers nodes that were previously
> > successfully loaded.
> >
> > This small patches fixes this behaviour. I stumbled over this while
> > working on a spi driver.
> 
> Thanks for your patch!
> 
> >  drivers/spi/spi.c | 7 +++++--
> >  1 file changed, 5 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/spi/spi.c b/drivers/spi/spi.c
> > index 200ca22..f96a04e 100644
> > --- a/drivers/spi/spi.c
> > +++ b/drivers/spi/spi.c
> > @@ -1604,12 +1604,15 @@ static void of_register_spi_devices(struct spi_master *master)
> >                 return;
> >
> >         for_each_available_child_of_node(master->dev.of_node, nc) {
> > -               if (of_node_test_and_set_flag(nc, OF_POPULATED))
> > +               if (of_node_check_flag(nc, OF_POPULATED))
> >                         continue;
> >                 spi = of_register_spi_device(master, nc);
> > -               if (IS_ERR(spi))
> > +               if (IS_ERR(spi)) {
> >                         dev_warn(&master->dev, "Failed to create SPI device for %s\n",
> >                                 nc->full_name);
> > +                       continue;
> > +               }
> > +               of_node_set_flag(nc, OF_POPULATED);
> 
> I think it's safer to keep the atomic test-and-set, but clear the flag on
> failure, cfr. of_platform_device_create_pdata() and of_amba_device_create().
> 
> Shouldn't of_spi_notify() be fixed, too?
> 
> The same issue exists for i2c in of_i2c_register_devices() and of_i2c_notify(),
> which is what I had used as an example.
> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
> 
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds

Attachment: signature.asc
Description: PGP signature