[PATCH v2 1/8] ARM: replace PROCINFO embedded branch with relative offset

[Ard Biesheuvel]

> On 19 apr. 2015, at 19:08, Russell King - ARM Linux <linux at arm.linux.org.uk> wrote:
> 
>> On Sun, Apr 19, 2015 at 06:59:45PM +0200, Joachim Eastwood wrote:
>> Hi Ard,
>>> On 13 March 2015 at 13:07, Ard Biesheuvel <ard.biesheuvel at linaro.org> wrote:
>>> This patch replaces the 'branch to setup()' instructions embedded
>>> in the PROCINFO structs with the offset to that setup function
>>> relative to the base of the struct. This preserves the position
>>> independent nature of that field, but uses a data item rather
>>> than an instruction.
>>> 
>>> This is mainly done to prevent linker failures on large kernels,
>>> where the setup function is out of reach for the branch.
>> 
>> This commit (bf35706f3d09 in Linus master) breaks booting on ARMv7-M.
>> 
>> When I try to boot Linus master now on my NXP LPC4357 (Cortex-M4) dev
>> kit I get the following message from u-boot.
>> ## Booting kernel from Legacy Image at 29000000 ...
>> Image Name: Linux
>> Image Type: ARM Linux Kernel Image (uncompressed)
>> Data Size: 1412318 Bytes = 1.3 MB
>> Load Address: 28008000
>> Entry Point: 28008001
>> Verifying Checksum ... OK
>> Loading Kernel Image ... OK
>> OK
>> 
>> Starting kernel ...
>> 
>> UNHANDLED EXCEPTION: HARD FAULT
>> R0 = ffffffff R1 = 00001038
>> R2 = 281d8711 R3 = 00000000
>> R12 = 2822092c LR = 28008023
>> PC = 2822092e PSR = 21000000
>> 
>> Reverting bf35706f3d09 (plus fixing a small conflict) makes Linus
>> master boot again.
>> 
>> I am using the following compiler:
>> gcc version 4.9.2 20140904 (prerelease) (crosstool-NG
>> linaro-1.13.1-4.9-2014.09 - Linaro GCC 4.9-2014.09
>> 
>> The ARMv7-M machine that I am using is not upstream yet, but you can
>> find the patch set on the mailing list.
> 
> Interesting... it works here with stock gcc 4.9.2.  Maybe it's a bug in
> the Linaro gcc?
> 
> Could you mail me (privately) your vmlinux file (the one in the root
> directory) for analysis please?

I am away from my work pc so i can't check but i wonder if all setup functions are correctly annotated as thumb2 when built in thumb2 mode. If not, it would explain why a plain branch works but doing arithmetic on the address doesn't.

Perhaps a bsym() around the setup function in question is sufficient to solve this?

Ard.