Problems booting exynos5420 with >1 CPU

[Abhilash Kesavan]

Hi Doug,

On Fri, Jun 6, 2014 at 11:32 PM, Doug Anderson <dianders at google.com> wrote:
> Abhilash,
>
> On Fri, Jun 6, 2014 at 10:36 AM, Abhilash Kesavan
> <kesavan.abhilash at gmail.com> wrote:
>> Hi Doug,
>>
>> The first change in the kernel (clearing an iRAM location) is needed
>> because of an unnecessary change that we are carrying in the Chrome
>> U-boot. There is no reason for us to have the workaround in the
>> mainline kernel. Rather, we should remove the check from our u-boot.
>> However AFAIR a clean-up patch that I had posted internally was not
>> accepted as we had frozen the SPL at the time.
>
> Ah, is that this one, or a different one?
>
> https://chromium-review.googlesource.com/#/c/66049/
Yes, this along with a kernel side change.
>
>
> If we land that patch now it won't help since nobody is going to be
> updating their read-only firmware.  We'll need to put code somewhere
> that fixes it.
We just carry the workaround fix locally until we migrate to mainline
u-boot for 5420 where the unnecessay check will not be present.
>
>
>> The second change is to enable snoops for boot cluster. Internally, we
>> were disabling the snoops for both the clusters at power off and
>> enabling it in power_up_setup and power_up. However, I dropped the
>> approach due to problems pointed out by Nicolas here
>> http://www.spinics.net/lists/arm-kernel/msg324091.html related to
>> cpuidle. Hence, we turn it on at the u-boot.
>
> I think I followed all that.  What you're saying is that our kernel
> dynamically enables and disables snoops as needed, but Nicolas pointed
> out that it was unsafe (though apparently we're not seeing problems in
> our usage).
We did not see any problems as CPUIdle was one of the problematic
scenarios which we have not got enabled.
>
> ...so now the kernel doesn't touch the snoops and assumes that U-Boot
> turned them on.
The kernel does turn on the snoops for the incoming cluster.
>
>
> Ugh.


Regards,
Abhilash
>
> -Doug