[PATCH] ARM: exynos_defconfig: Enable big.LITTLE CPUidle support

[Krzysztof Kozlowski]

2015-08-29 19:07 GMT+09:00 Javier Martinez Canillas <javier at dowhile0.org>:
> Hello Krzysztof,
>
> On Sat, Aug 29, 2015 at 11:55 AM, Krzysztof Kozlowski
> <k.kozlowski at samsung.com> wrote:
>> 2015-08-29 18:33 GMT+09:00 Javier Martinez Canillas <javier at osg.samsung.com>:
>>> Hello Krzysztof,
>>>
>>> On 08/29/2015 11:01 AM, Krzysztof Kozlowski wrote:
>>>> W dniu 28.08.2015 o 17:16, Javier Martinez Canillas pisze:
>>>>> Some Exynos big.LITTLE boards (i.e: Exynos5420 and Exynos5800 based
>>>>> Chromebooks) have proper firmware that allow the big.LITTLE CPUidle
>>>>> driver to work correctly, so enable support for this.
>>>>>
>>>>> Signed-off-by: Javier Martinez Canillas <javier at osg.samsung.com>
>>>>>
>>>>> ---
>>>>> Kukjin and Krzysztof,
>>>>>
>>>>> As you know there are other boards like the Exynos5422 based Odroid XU{3,4}
>>>>> whose firmware is broken due leaving CCI in secure mode which means that the
>>>>> kernel MCPM support can't properly manage CCI.
>>>>>
>>>>> So if you pick this patch, it should be tested in kernelci before appearing
>>>>> in linux-next to prevent any boot issues.
>>>>>
>>>>> But if that happens, I believe that is better to do a fix / workaround in
>>>>> those broken platforms since nothing prevents users to enable this option
>>>>> anyways. For example the CCI device node could be disabled in the DTS.
>>>>>
>>>>>  arch/arm/configs/exynos_defconfig | 1 +
>>>>>  1 file changed, 1 insertion(+)
>>>>
>>>> On Odroid XU3L (next-20150828, Hardkernel u-boot) boot hangs just after:
>>>>
>>>
>>> Thanks for testing, I was expecting that is just that I don't have a
>>> Odroid XU{3,4} board for test here, I guess I should get one.
>>>
>>>> [    2.568650] dwmmc_exynos 12200000.mmc: num-slots property not found,
>>>> assuming 1 slot is available
>>>>
>>>> ... so no. NACK :). First the boards, firmware, bootloader or kernel
>>>
>>> Agreed with the nack :)
>>>
>>>> code have to be fixed.
>>>>
>>>
>>> Or disable CCI, could you please test the following patch [0] so I
>>> can post it properly?
>>
>> It fixes the boot hang but causes other issues. Not all CPUs boot (I
>
> Thanks for testing.
>
>> tested it on Chanho Park's patch for fixing CPU boot with SWRESET):
>> [    0.010781] CPU0: update cpu_capacity 448
>> [    0.010839] CPU0: thread -1, cpu 0, socket 1, mpidr 80000100
>> [    0.011098] Setting up static identity map for 0x40008280 - 0x400082d8
>> [    0.056329] CPU1: update cpu_capacity 448
>> [    0.056337] CPU1: thread -1, cpu 1, socket 0, mpidr 80000001
>> [    0.071100] CPU2: update cpu_capacity 448
>> [    0.071107] CPU2: thread -1, cpu 2, socket 0, mpidr 80000002
>> [    0.086103] CPU3: update cpu_capacity 448
>> [    0.086111] CPU3: thread -1, cpu 3, socket 0, mpidr 80000003
>> [    0.101100] CPU4: update cpu_capacity 1535
>> [    0.101108] CPU4: thread -1, cpu 0, socket 0, mpidr 80000000
>> [    1.115009] CPU5: failed to boot: -110
>> [    2.130019] CPU6: failed to boot: -110
>> [    3.145049] CPU7: failed to boot: -110
>> [    3.145151] Brought up 5 CPUs
>> [    3.145196] SMP: Total of 5 processors activated (240.00 BogoMIPS).
>> [    3.145251] CPU: WARNING: CPU(s) started in wrong/inconsistent
>> modes (primary CPU mode 0x13)
>> [    3.145327] CPU: This may indicate a broken bootloader or firmware.
>> [    3.149347] devtmpfs: initialized
>>
>
> Sigh, such an inconsistent behavior between Exynos5420/5422/5800 boards :-/

I put all my hope into Przemyslaw's work :)

>
> Any ideas? I agree that the b.L CPUidle support shouldn't be enabled
> by default in exynos_defconfig but as I said nothing prevents the user
> to enable that option and make the board to hang on boot.

I think these are actually two different issues:
1. If user enables b.L cpuidle manually, some of the Exynos542x boards
would work (Peach Pi, AFAIU?) and some would not (Odroid XU3, probably
also Arndale Octa). If we cannot fix firmware then we could implement
a workaround in Exynos code. However it shouldn't be DTS but rather
the mach code.
2. The default config should be a reference config so it should work
on all boards and should offer all necessary features on them (or even
all features). Enabling the b.L in it makes sense only if it does not
introduce regressions to other boards. If we cannot achieve that
(Odroid XU3 will be affected negatively) then still we could fix issue
#1 so user's manual setting would work.

BR,
Krzysztof