Re: [PATCH] mm/thp: Make ARCH_ENABLE_SPLIT_PMD_PTLOCK dependent on PGTABLE_LEVELS > 2

[Anshuman Khandual]

On 5/10/21 3:46 PM, Anshuman Khandual wrote:
> 
> 
> On 5/10/21 3:40 PM, Oscar Salvador wrote:
>> On Mon, May 10, 2021 at 03:36:29PM +0530, Anshuman Khandual wrote:
>>>
>>>
>>> On 5/10/21 2:23 PM, Peter Zijlstra wrote:
>>>> On Mon, May 10, 2021 at 10:05:45AM +0530, Anshuman Khandual wrote:
>>>>> -	select ARCH_ENABLE_SPLIT_PMD_PTLOCK if X86_64 || X86_PAE
>>>>> +	select ARCH_ENABLE_SPLIT_PMD_PTLOCK if (PGTABLE_LEVELS > 2) && (X86_64 || X86_PAE)
>>>>
>>>> It's still very early on a Monday, but IIRC this new condition is
>>>> identical to the pre-existing one.
>>>
>>> Did not get it, could you please elaborate ?
>>
>> When using x86_PAE, you must have more than two pgtable levels, right?
>> And not speaking of x86_64.
> 
> arch/x86/Kconfig..
> 
> config PGTABLE_LEVELS
>         int
>         default 5 if X86_5LEVEL
>         default 4 if X86_64
>         default 3 if X86_PAE
>         default 2
> 
> Both X86_PAE and X86_64 will always have page table levels > 2 ? But
> regardless, it might be still useful to assert (PGTABLE_LEVELS > 2)
> before selecting ARCH_ENABLE_SPLIT_PMD_PTLOCK.

PGTABLE_LEVELS > 2 is a necessary condition for this PMD split lock
config. The problem is that for arch selectable configs like this,
conditional statements would not work properly when defined along
with the config. Otherwise the following change would have been
sufficient.

diff --git a/mm/Kconfig b/mm/Kconfig
index 02d44e3420f5..5830ea7746b3 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -214,6 +214,7 @@ config SPLIT_PTLOCK_CPUS
 
 config ARCH_ENABLE_SPLIT_PMD_PTLOCK
        bool
+       depends on PGTABLE_LEVELS > 2

Hence this just moves the condition to all subscribing platforms
while making the selection for ARCH_ENABLE_SPLIT_PMD_PTLOCK.