[PATCH -next] riscv: TRANSPARENT_HUGEPAGE: depends on MMU

[Palmer Dabbelt]

On Tue, 25 May 2021 08:40:10 PDT (-0700), rdunlap at infradead.org wrote:
> On 5/25/21 3:33 AM, Mike Rapoport wrote:
>> On Mon, May 24, 2021 at 05:13:02PM -0700, Randy Dunlap wrote:
>>> Fix a Kconfig warning and many build errors:
>>>
>>> WARNING: unmet direct dependencies detected for COMPACTION
>>>   Depends on [n]: MMU [=n]
>>>   Selected by [y]:
>>>   - TRANSPARENT_HUGEPAGE [=y] && HAVE_ARCH_TRANSPARENT_HUGEPAGE [=y]
>>>
>>> and the subseqent thousands of build errors and warnings.
>>>
>>> Fixes: e88b333142e4 ("riscv: mm: add THP support on 64-bit")
>>> Signed-off-by: Randy Dunlap <rdunlap at infradead.org>
>>> Cc: Paul Walmsley <paul.walmsley at sifive.com>
>>> Cc: Palmer Dabbelt <palmer at dabbelt.com>
>>> Cc: Albert Ou <aou at eecs.berkeley.edu>
>>> Cc: linux-riscv at lists.infradead.org
>>> Cc: Andrew Morton <akpm at linux-foundation.org>
>>> Cc: Nanyong Sun <sunnanyong at huawei.com>
>>
>> Acked-by: Mike Rapoport <rppt at linux.ibm.com>
>>
>>> ---
>>> I really think that mm/Kconfig should have "depends on MMU" for
>>> TRANSPARENT_HUGEPAGE, but commit 15626062f4a9 from 2012
>>> ("thp, x86: introduce HAVE_ARCH_TRANSPARENT_HUGEPAGE") thought
>>> differently. I.e., mm/Kconfig should be protecting all arches
>>> against this Kconfig problem.
>>
>> We could
>>
>> diff --git a/mm/Kconfig b/mm/Kconfig
>> index 02d44e3420f5..85bc8bf32e7b 100644
>> --- a/mm/Kconfig
>> +++ b/mm/Kconfig
>> @@ -403,7 +403,7 @@ config NOMMU_INITIAL_TRIM_EXCESS
>>
>>  config TRANSPARENT_HUGEPAGE
>>  	bool "Transparent Hugepage Support"
>> -	depends on HAVE_ARCH_TRANSPARENT_HUGEPAGE
>> +	depends on MMU && HAVE_ARCH_TRANSPARENT_HUGEPAGE
>>  	select COMPACTION
>>  	select XARRAY_MULTI
>>  	help
>>
>> but in a sense HAVE_ARCH_TRANSPARENT_HUGEPAGE should imply that MMU is
>> selected :)
>
> Yes, I had this patch before the one that I sent but decided against it
> based on history.

IMO 15626062f4a9 ("thp, x86: introduce HAVE_ARCH_TRANSPARENT_HUGEPAGE") 
is really saying "the arches that have transparent huge pages are MMU=y, 
so we don't need the extra check".  That seems reasonable as of when the 
code landed, and while arm has MMU=n and THP it already depends on MMU 
(via ARM_LPAE, which seems kind of natural there).

The generic version seems fine to me, but I'm going to take this one now 
so we can avoid the build issues.

>>>  arch/riscv/Kconfig |    2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> --- linux-next-20210524.orig/arch/riscv/Kconfig
>>> +++ linux-next-20210524/arch/riscv/Kconfig
>>> @@ -106,7 +106,7 @@ config RISCV
>>>  	select SYSCTL_EXCEPTION_TRACE
>>>  	select THREAD_INFO_IN_TASK
>>>  	select UACCESS_MEMCPY if !MMU
>>> -	select HAVE_ARCH_TRANSPARENT_HUGEPAGE if 64BIT
>>> +	select HAVE_ARCH_TRANSPARENT_HUGEPAGE if 64BIT && MMU
>>>
>>>  config ARCH_MMAP_RND_BITS_MIN
>>>  	default 18 if 64BIT
>>>
>>
>
> thanks.