[PATCH v2 8/9] ARM: Thumb-2: Fix long-distance conditional branches in head.S for Thumb-2.

[Nicolas Pitre]

On Mon, 22 Nov 2010, Russell King - ARM Linux wrote:

> On Mon, Nov 22, 2010 at 02:03:51PM -0500, Nicolas Pitre wrote:
> > On Mon, 22 Nov 2010, Dave Martin wrote:
> > 
> > > The 32-bit conditional branches in Thumb-2 have a shorter range (+/-512K)
> > > than their ARM counterparts (+/-32MB).  The linker does not currently
> > > generate trampolines to extend the range of these Thumb-2 conditional
> > > branches, resulting in link errors when vmlinux is sufficiently large, e.g.:
> > > 
> > > head.o:(.text+0x464): relocation truncated to fit: R_ARM_THM_JUMP19
> > > 
> > > This patch forces the longer-range, unconditional branch encoding by
> > > use of an explicit IT instruction.  The resulting branches are triggered on
> > > the same conditions as before.
> > > 
> > > Signed-off-by: Dave Martin <dave.martin at linaro.org>
> > 
> > I'm afraid this patch is only making an actual bug visible.  There is no 
> > reason why __error_p or __error_a ought to be farther than 512K away 
> > from their call sites as they are all supposed to live in the __init 
> > section close together.  It looks like the refactoring done in commit 
> > 75d90832d5 has dropped the __init attribute from those moved functions, 
> > and adding it back would be the real fix.
> 
> No, you're not thinking right.  It was _intentional_ that __error_p
> ended up in a different section to __init.  With CPU hotplug, as it
> is called from the hotplug CPU initialization path, it needs to be
> available for that path to call.

Well... agreed for the hotplug CPU case.  The section selection is good 
as it is for that.

However I doubt this was intentional when commit 75d90832d5 was created, 
which is when the actual section move took place.


Nicolas