Re: [PATCH] powerpc: Fix booting P9 hash with CONFIG_PPC_RADIX_MMU=N

["Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>]

On Thursday 25 May 2017 11:46 AM, Michael Ellerman wrote:
> Michael Neuling <mikey@neuling.org> writes:
> 
>> On Wed, 2017-05-24 at 14:26 +0530, Aneesh Kumar K.V wrote:
>>> Michael Neuling <mikey@neuling.org> writes:
>>>
>>>> Currently if you disable CONFIG_PPC_RADIX_MMU you'll crash on boot on
>>>> a P9. This is because we still set MMU_FTR_TYPE_RADIX via
>>>> ibm,pa-features and MMU_FTR_TYPE_RADIX is what's used for code patching
>>>> in much of the asm code (ie. slb_miss_realmode)
>>>>
>>>> This patch fixes the problem by stopping MMU_FTR_TYPE_RADIX from being
>>>> set from ibm.pa-features.
>>>>
>>>> We may eventually end up removing the CONFIG_PPC_RADIX_MMU option
>>>> completely but until then this fixes the issue.
>>>>
>>>> Signed-off-by: Michael Neuling <mikey@neuling.org>
>>>> ---
>>>>   arch/powerpc/kernel/prom.c | 2 ++
>>>>   1 file changed, 2 insertions(+)
>>>>
>>>> diff --git a/arch/powerpc/kernel/prom.c b/arch/powerpc/kernel/prom.c
>>>> index 40c4887c27..f830562974 100644
>>>> --- a/arch/powerpc/kernel/prom.c
>>>> +++ b/arch/powerpc/kernel/prom.c
>>>> @@ -161,7 +161,9 @@ static struct ibm_pa_feature {
>>>>   	{ .pabyte = 0,  .pabit = 3, .cpu_features  = CPU_FTR_CTRL },
>>>>   	{ .pabyte = 0,  .pabit = 6, .cpu_features  = CPU_FTR_NOEXECUTE },
>>>>   	{ .pabyte = 1,  .pabit = 2, .mmu_features  = MMU_FTR_CI_LARGE_PAGE
>>>> },
>>>> +#ifdef CONFIG_PPC_RADIX_MMU
>>>>   	{ .pabyte = 40, .pabit = 0, .mmu_features  = MMU_FTR_TYPE_RADIX },
>>>> +#endif
>>>>   	{ .pabyte = 1,  .pabit = 1, .invert = 1, .cpu_features =
>>>> CPU_FTR_NODSISRALIGN },
>>>>   	{ .pabyte = 5,  .pabit = 0, .cpu_features  = CPU_FTR_REAL_LE,
>>>>   				    .cpu_user_ftrs = PPC_FEATURE_TRUE_LE },
>>>> -- 
>>>> 2.11.0
>>>
>>> Instead can we do that feature removal in mmu_early_init_devtree. ie,
>>> something like
>>
>> It looks like mmu_early_init_devtree() gets called after parsing ibm,pa_features
>> so that should work.
> 
> But why is doing it later preferable?
> 
> Mikey's patch means it will never be set at any point during boot, which
> seems obviously better to me.

My suggestion was w.r.t consolidating different ways of clearing 
RADIX_MMU feature flag into one place. Also I was not sure we want to 
conditionally parse the pa-feature device tree. Instead we parse it 
without any #ifdef in there and clear things which are not supported by 
the kernel either via command line or via Kconfig.

-aneesh