Re: [PATCH 0/2] do not select KALLSYMS_ALL

[Artem Bityutskiy <dedekind1@gmail.com>]

On Fri, 2011-04-01 at 16:34 +0100, Paulo Marques wrote:
> Artem Bityutskiy wrote:
> > On Fri, 2011-04-01 at 17:56 +0300, Artem Bityutskiy wrote:
> >> Well, ok, I've measured how much is this "a lot". On an embedded arm
> >> platform this makes the kernel only 1.5% larger. 
> > 
> > But in absolute numbers this is 70KiB in my case, which is indeed
> > considerable amount. So, I agree that it makes sense to keep this as a
> > separate option.
> 
> Yes, and this depends a lot on the kernel configuration options you've
> selected (and that 1.5% of the kernel size, might be 50%~100% of the
> kallsyms table).
> 
> IIRC, I had configurations where the KALLSYMS_ALL option was increasing
> the kallsyms table from something like 150kB to above 500kB (before
> compression). This was a long time ago though and I can't say exactly
> what was the configuration that made this happen.

Yes, thanks, I agree that having 2 separate options is OK.

> As for the CONFIG_KALLSYMS_EXTRA_PASS, I think the original idea (it was
> not mine) was that it should be off by default and then the user would
> need to turn it on when something went wrong.

Yes, but my point is that this feature is about the build and it does
not affect run-time, so it should not be in Kconfig.

> With an automatic makefile mechanism, the problem would go unnoticed and
> it just wouldn't be fixed, increasing the kernel compile time for
> everyone who hits the same troublesome configuration.

Yes, the build system may print a message:

Your symbols table is screwed, this is a bug, report about it. As a
temporary workaround use "make KALLSYMS_EXTRA_PASS=1"

Or something like that.

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)