[PATCH] ARM: EXYNOS: Avoid early use of of_machine_is_compatible()

Wed Nov 28 12:44:12 EST 2012

On Wed, Nov 28, 2012 at 8:16 AM, Doug Anderson <dianders at chromium.org> wrote:
> Olof / Kukjin,
>
> On Tue, Nov 27, 2012 at 10:05 PM, Olof Johansson <olof at lixom.net> wrote:
>> On Wed, Nov 28, 2012 at 02:23:09PM +0900, Kukjin Kim wrote:
>>> Olof Johansson wrote:
>>> >
>>> > On Tue, Nov 27, 2012 at 2:27 PM, Kukjin Kim <kgene.kim at samsung.com> wrote:
>>> > > On 11/28/12 07:11, Olof Johansson wrote:
>>> > >>
>>> > >> On Tue, Nov 27, 2012 at 11:53 AM, Doug Anderson<dianders at chromium.org>
>>> > >> wrote:
>>> > >>>
>>> > >>> The recent commit "ARM: EXYNOS: add support for EXYNOS5440 SoC" broke
>>> > >>> support for exynos5250 because of_machine_is_compatible() was used too
>>> > >>> early in the boot process.  It also probably meant that the exynos5440
>>> > >>> failed to use the proper iotable.  Switch to use
>>> > >>> of_flat_dt_is_compatible() in both of these cases.
>>> > >>>
>>> > >>> The failure I was seeing in exynos5250 because of this was:
>>> > >>>    Division by zero in kernel.
>>> > >>>    [<80015ed4>] (unwind_backtrace+0x0/0xec) from [<8045c7a4>]
>>> > >>> (dump_stack+0x20/0x24)
>>> > >>>    [<8045c7a4>] (dump_stack+0x20/0x24) from [<80012990>]
>>> > >>> (__div0+0x20/0x28)
>>> > >>>    [<80012990>] (__div0+0x20/0x28) from [<8021ab04>]
>>> (Ldiv0_64+0x8/0x18)
>>> > >>>    [<8021ab04>] (Ldiv0_64+0x8/0x18) from [<80068560>]
>>> > >>> (__clocksource_updatefreq_scale+0x54/0x134)
>>> > >>>    [<80068560>] (__clocksource_updatefreq_scale+0x54/0x134) from
>>> > >>> [<8006865c>] (__clocksource_register_scale+0x1c/0x54)
>>> > >>>    [<8006865c>] (__clocksource_register_scale+0x1c/0x54) from
>>> > >>> [<80612a18>] (exynos_timer_init+0x100/0x1e8)
>>> > >>>    [<80612a18>] (exynos_timer_init+0x100/0x1e8) from [<8060d184>]
>>> > >>> (time_init+0x28/0x38)
>>> > >>>    [<8060d184>] (time_init+0x28/0x38) from [<8060a754>]
>>> > >>> (start_kernel+0x1e0/0x3c8)
>>> > >>>    [<8060a754>] (start_kernel+0x1e0/0x3c8) from [<40008078>]
>>> > (0x40008078)
>>> > >>>
>>> > >>> Signed-off-by: Doug Anderson<dianders at chromium.org>
>>> > >>
>>> > >>
>>> > >> Thanks Doug.
>>> > >>
>>> > >> Kukjin, I'll apply this directly on top of the previous branch in
>>> > >> arm-soc, if that's OK with you.
>>> > >>
>>> > > Sure, go ahead with my ack if you want,
>>> > >
>>> > > Acked-by: Kukjin Kim <kgene.kim at samsung.com>
>>> > >
>>> > > Note, actually there was a fix which uses soc_is_exynos5440() in my
>>> > local
>>> > > :-) I'm not sure which one is better at this moment, but I'm OK on this.
>>> >
>>> > Ok, applied. Thanks all.
>>> >
>>> Olof, just note, happens build error with exynos4_defconfig because of
>>> non-DT.
>>
>> Ick, thanks for catching that.
>
> Sorry for this!  I will try to be more diligent about trying
> exynos4_defconfig before submitting future patches to these files.
>
>>>
>>> Following can resolve it or we should create null function for
>>> of_get_flat_dt_root() and of_flat_dt_is_compatible()...
>>>
>>> 8<---------------------------------------
>>> From: Kukjin Kim <kgene.kim at samsung.com>
>>> Subject: ARM: EXYNOS: fix a build error with non-DT for exynos4
>>>
>>> This fixes following in case of non-DT:
>>> arch/arm/mach-exynos/common.c: In function 'exynos_init_io':
>>> arch/arm/mach-exynos/common.c:339: error: implicit declaration of function
>>> 'of_get_flat_dt_root'
>>> arch/arm/mach-exynos/common.c:342: error: implicit declaration of function
>>> 'of_flat_dt_is_compatible'
>>> make[1]: *** [arch/arm/mach-exynos/common.o] Error 1
>>>
>>> Signed-off-by: Kukjin Kim <kgene.kim at samsung.com>
>>> ---
>>> diff --git a/arch/arm/mach-exynos/common.c b/arch/arm/mach-exynos/common.c
>>> index b919f5f..2110091 100644
>>> --- a/arch/arm/mach-exynos/common.c
>>> +++ b/arch/arm/mach-exynos/common.c
>>> @@ -336,12 +336,14 @@ void __init exynos_init_late(void)
>>>
>>>  void __init exynos_init_io(struct map_desc *mach_desc, int size)
>>>  {
>>> +#ifdef CONFIG_OF
>>>       unsigned long root = of_get_flat_dt_root();
>>>
>>>       /* initialize the io descriptors we need for initialization */
>>>       if (of_flat_dt_is_compatible(root, "samsung,exynos5440"))
>>>               iotable_init(exynos5440_iodesc,
>>> ARRAY_SIZE(exynos5440_iodesc));
>>>       else
>>> +#endif
>>>               iotable_init(exynos_iodesc, ARRAY_SIZE(exynos_iodesc));
>>
>> I really don't like splitting an if/else with an ifdef like this, it's fragile
>> code and can be hard to follow.
>
> Agree.
>
>> There's also a second build error with exynos_defconfig in the
>> exynos5-dt.c board file due to a missing include. Teaches me to just apply
>> patches without trying to build. :(
>
> In the tree I was testing against (the arm-soc/for-next branch at
> 659b19ca3a77e2ac32fe84d95242653c75dd07c7) I see the include file in
> the exynos5-dt.c file already.  In my tree it was added by "2eae613b:
> ARM: EXYNOS: Add MFC device tree support".
>
> Your patch applies cleanly on mine but I end up with:
>
>   #include <linux/of_platform.h>
>   #include <linux/of_fdt.h>
>   #include <linux/serial_core.h>
>   #include <linux/io.h>
>   #include <linux/memblock.h>
>   #include <linux/of_fdt.h>

Sigh, this is because people add includes out of alphabetical order.
We'll just have to fix it up later, if we don't add of_fdt.h in the
exynos5440 branch, the code will not be bisectable.

>> I'll squash this into Doug's original patch, if that's OK?
>
> No objection to squashing a fix and your CL is better than what I
> have, but see below for an issue.
>
>
>> diff --git a/arch/arm/mach-exynos/common.c b/arch/arm/mach-exynos/common.c
>> index 796e0c9..77e7c5b 100644
>> --- a/arch/arm/mach-exynos/common.c
>> +++ b/arch/arm/mach-exynos/common.c
>> @@ -122,6 +122,7 @@ static struct map_desc exynos_iodesc[] __initdata = {
>>         },
>>  };
>>
>> +#ifdef CONFIG_ARCH_EXYNOS5
>
> Are you sure you want this #ifdef?  If so it should match the ifdef
> used below.  With your patch applied I can get a compile error with:
>
> make exynos_defconfig
> echo '# CONFIG_ARCH_EXYNOS5 is not set' >> .config
>
> In other words your code will fail if someone wants a FDT-enabled exynos4 build.

Yep, the above needs to be:
#if defined(CONFIG_OF) && defined(CONFIG_ARCH_EXYNOS5)

I've pushed out the branch with the patch applied (with the above changed).

-Olof