Re: [PATCH] arm: add default value for PHYS_OFFSET Kconfig setting

[Nicolas Pitre]

On Fri, 2 Dec 2011, Nicolas Pitre wrote:

> On Thu, 1 Dec 2011, Paul Gortmaker wrote:
> 
> > On Thu, Dec 1, 2011 at 9:19 AM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
> > > On Thursday 01 December 2011, Paul Gortmaker wrote:
> > >> Commit 1b9f95f8ade9efc2bd49f0e7b9dc61a038ac3eef introduced the new
> > >> Kconfig for PHYS_OFFSET but it did not specify a default value.
> > >>
> > >> This has the undesired side effect of breaking an implictly used
> > >> trick for updating defconfigs on the fly for routine buildtesting
> > >> across all arch and all platforms, i.e.
> > >>
> > >>   cp /path/to/somedefconfig .config ; yes "" | make oldconfig
> > >>
> > >
> > > FWIW, a better way to achieve the above is using
> > >
> > > make KCONFIG_ALLCONFIG=/path/to/somedefconfig listnewconfig alldefconfig
> > >
> > > which does not have this problem.
> > >
> > > Your patch looks still ok, but you might want to update your scripts
> > > to be more robust.
> > 
> > Thanks -- but on the other hand, the above use case does actually
> > emulate exactly what happens when someone runs the
> > "make oldconfig" and simply holds down the Enter key. (yes it
> > happens probably way more than we want to know...)
> > 
> > So, I guess what I'm trying to say is that the above "script" is
> > intentionally fragile.  :)  As a use case, we shouldn't break it.
> 
> Well... I took a while before replying because I'm also modifying things 
> around this, but those changes are good for a merge window not a -rc 
> period.
> 
> Still, I don't think your patch is good, even as a temporary measure.  
> The fact is that we can't just pick a random default value since it is 
> likely to be wrong for the majority of the boards as the right answer 
> for this option is quite varied.  So the fact that the config system 
> insists on your entering a proper value is actually a good thing.
> 
> In the at91x40_defconfig case, it has CONFIG_EMBEDDED=y.  If it was not 
> selected then you'd get CONFIG_ARM_PATCH_PHYS_VIRT=y by default and then 
> no CONFIG_PHYS_OFFSET would be presented.
> 
> So the right fix here would actually consist of adding to 
> at91x40_defconfig either CONFIG_ARM_PATCH_PHYS_VIRT=y, or a default 
> value for CONFIG_PHYS_OFFSET which is the right value for that platform.

And it seems that at91x40 is the only defconfig with MMU=n, so the only 
option is really to provide a default CONFIG_PHYS_OFFSET.  So what about 
this:

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index e084b7e981..a24672924f 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -222,6 +222,7 @@ config NEED_MACH_MEMORY_H
 config PHYS_OFFSET
 	hex "Physical address of main memory"
 	depends on !ARM_PATCH_PHYS_VIRT && !NEED_MACH_MEMORY_H
+	default DRAM_BASE if !MMU
 	help
 	  Please provide the physical address corresponding to the
 	  location of main memory in your system.

Eventually the DRAM_BASE config option could simply be replaced by 
PHYS_OFFSET directly, but that's a larger change.


Nicolas