Re: 2.6.24-git7: section mismatches woes

[Sam Ravnborg]

On Sat, Feb 02, 2008 at 11:47:29PM +0100, Rafael J. Wysocki wrote:
> On Saturday, 2 of February 2008, Sam Ravnborg wrote:
> > On Wed, Jan 30, 2008 at 10:32:52PM +0100, Rafael J. Wysocki wrote:
> > > On Wednesday, 30 of January 2008, Sam Ravnborg wrote:
> > > > On Wed, Jan 30, 2008 at 07:50:43PM +0100, Rafael J. Wysocki wrote:
> > > > > Hi,
> > > > > 
> > > > > I get these messages, the majority of which seem to be false-positives:
> > > > ...
> > > > > modpost: Found 35 section mismatch(es).
> > > > > To see additional details select "Enable full Section mismatch analysis"
> > > > > in the Kernel Hacking menu (CONFIG_SECTION_MISMATCH).
> > > > Looking in to these atm.
> > > > 
> > > > > 
> > > > > and if I compile the kernel with CONFIG_SECTION_MISMATCH, it breaks resuming
> > > > > from RAM.
> > > > 
> > > > The only functional difference when you enable CONFIG_SECTION_MISMATCH is the
> > > > addition of the -fno-inline-functions-called-once to CFLAGS.
> > > > So we have some code somewhere that breaks if it is not inlined by gcc.
> > > > 
> > > > It would be nice to sort out where.
> > > > If you have a rough idea where to look
> > > 
> > > No, I don't.
> > > 
> > > It looks like there's somewhere in arch/x86, since I ruled out kernel/power and
> > > drivers/acpi already.
> > 
> > Hi Rafael.
> 
> Hi,
> 
> > Do you plan to look closer into this or do you have an easy receipe so
> > I can test myself (on a x86 64 bit box)?
> 
> Well, I really don't know how to approach this.  Do you have an x86-64 box with
> suspend to RAM working?
I dunno - never used it I'm afraid. And do not know how to do it either.
Feel free to call me ignorant :-(

I have in latest kbuild.git made DEBUG_SECTION_MISMATCH so it cannot be selected,
partly due to this regression but mostly due to the noise it creates.

The way to approach it is straightforward but boring.

You can add:
ccflags-y += -fno-inline-functions-called-once
to the Makefiles where you think this can have caused troubles until
you find the problematic directory.
kbuild will pick up that gcc options changed and rebuild
all relevant files.

When the problematic directory is located then
remove the ccflags-y assignment and do it file-by-file
using the CFLAGS_xxx.o syntax:

CFLAGS_file.o := -fno-inline-functions-called-once

This is probarly not the highest priority thing to look into
but I am afraid that it could show up under other circumstances too.


	Sam
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/