[U-Boot] [PATCH 2/2] fdt_support: Don't panic if unable to perform stdout fixup

[Scott Wood]

On Tue, 2015-09-01 at 21:10 -0600, Simon Glass wrote:
> Hi Scott,
> 
> On 1 September 2015 at 21:00, Scott Wood <scottwood at freescale.com> wrote:
> > On Tue, 2015-09-01 at 20:48 -0600, Simon Glass wrote:
> > > Hi Scott,
> > > 
> > > On 31 August 2015 at 21:16, Scott Wood <scottwood at freescale.com> wrote:
> > > > On Mon, 2015-08-31 at 21:13 -0600, Simon Glass wrote:
> > > > > Hi Scott,
> > > > > 
> > > > > On 31 August 2015 at 20:11, Scott Wood <scottwood at freescale.com> 
> > > > > wrote:
> > > > > > Currently, using fdt_fixup_stdout() on a device tree that is 
> > > > > > missing
> > > > > > the relevant alias results in this:
> > > > > > 
> > > > > > WARNING: could not set linux,stdout-path FDT_ERR_NOTFOUND.
> > > > > > ERROR: /chosen node create failed
> > > > > >  - must RESET the board to recover.
> > > > > > 
> > > > > > FDT creation failed! hanging...### ERROR ### Please RESET the 
> > > > > > board
> > > > > > ###
> > > > > > 
> > > > > > There is no reason for this to be a fatal error rather than a 
> > > > > > warning,
> > > > > > and removing this allows for a smooth transition on a platform 
> > > > > > where
> > > > > > the device tree currently lacks the correct aliases but will have 
> > > > > > them
> > > > > > in the future.
> > > > > 
> > > > > Why do we need this patch - what platform?
> > > > 
> > > > LS2085A
> > > > 
> > > > > 
> > > > > > 
> > > > > > Signed-off-by: Scott Wood <scottwood at freescale.com>
> > > > > > Cc: Kumar Gala <galak at kernel.crashing.org>
> > > > > > Cc: Simon Glass <sjg at chromium.org>
> > > > > > ---
> > > > > > Resent with correct address for Simon Glass.
> > > > > > 
> > > > > >  common/fdt_support.c | 3 ++-
> > > > > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > > > > 
> > > > > > diff --git a/common/fdt_support.c b/common/fdt_support.c
> > > > > > index f86365e..6052c77 100644
> > > > > > --- a/common/fdt_support.c
> > > > > > +++ b/common/fdt_support.c
> > > > > > @@ -308,7 +308,8 @@ int fdt_chosen(void *fdt)
> > > > > >                 }
> > > > > >         }
> > > > > > 
> > > > > > -       return fdt_fixup_stdout(fdt, nodeoffset);
> > > > > > +       fdt_fixup_stdout(fdt, nodeoffset);
> > > > > 
> > > > > Will some platforms will not boot correctly with this failing? 
> > > > > Should
> > > > > we make your new feature a Kconfig options perhaps? I worry that it
> > > > > will become the default behaviour and then it will be hard to remove
> > > > > later.
> > > > 
> > > > A warning will still be printed.  I'm not sure how "### ERROR ### 
> > > > Please
> > > > RESET the board ###" is more useful than trying to continue and 
> > > > possibly
> > > > failing.
> > > 
> > > Only that if it indicates a fatal error the board code can at least
> > > find out about it and deal with it. Perhaps booting will just result
> > > in a hang?
> > 
> > If booting results in a hang, then you're no worse off than the current
> > 
> > situation.  Either way, the user sees a message that indicates the 
> > problem.
> > 
> > > I think ignoring errors is fine but here we make it impossible to
> > > detect a failure. So I think that a Kconfig is the best idea, so we
> > > can remove it later.
> > 
> > It seems excessive...  Config options aren't free, from a maintenance
> > 
> > perspective, and I have a hard time imagining a scenario where proceeding 
> > to
> > 
> > boot causes a real problem.
> > 
> > Also, from a consistency perspective:
> >  - The OF_STDOUT_PATH version doesn't cause a panic if the destination 
> > path
> > doesn't exist.
> >  - None of the callers of fdt_status_<foo>_by_alias() panic if the alias 
> > is
> > not found.
> >  - do_fixup_by_path() doesn't panic if the node is not found.
> >  - Neither does do_fixup_by_compatible().
> >  - Neither does fdt_fixup_ethernet().
> 
> Yes this code needs cleaning up.

I wasn't citing them as problems. :-P

> The particular problem you have is that there are no aliases, right?
> How about returning 0 from fdt_fixup_stdout() in that case? Then you
> will not affect the behaviour of other boards which perhaps do have
> /aliases but do not have space in their tree for the fdt_setprop().

That would be fine.

-Scott