powerpc/e500: WARNING: at mm/hugetlb.c:4755 hugetlb_add_hstate

[Christophe Leroy]

Le 10/11/2025 à 11:10, David Hildenbrand (Red Hat) a écrit :
> [fighting with mail transitioning, for some reason I did not receive
> the mails from Christophe, so replying here]
> 
>>>> diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
>>>> index e24f4d88885ae..55c3626c86273 100644
>>>> --- a/arch/powerpc/Kconfig
>>>> +++ b/arch/powerpc/Kconfig
>>>> @@ -137,6 +137,7 @@ config PPC
>>>>        select ARCH_HAS_DMA_OPS            if PPC64
>>>>        select ARCH_HAS_FORTIFY_SOURCE
>>>>        select ARCH_HAS_GCOV_PROFILE_ALL
>>>> +    select ARCH_HAS_GIGANTIC_PAGE        if PPC64
>>
>>
>> The patch looks good from PPC64 perspective, it also fixes the problem
>> reported on corenet64_smp_defconfig...
>>
>>>
>>> Problem is not only on PPC64, it is on PPC32 as well, for instance
>>> corenet32_smp_defconfig has the problem as well.
>>>
>>
>> However on looking deeper into it - I agree with Christophe that this
>> problem might still exist on PPC32.
> 
> Ah, I missed that. I thought it would be a ppc64 thing. :(
> 
>>
>> I did try the patch on corenet32_smp_defconfig and I can see the WARN_ON
>> still triggering. You can check the logs here..
>>
>> https://eur01.safelinks.protection.outlook.com/? 
>> url=https%3A%2F%2Fgithub.com%2Friteshharjani%2Flinux- 
>> ci%2Factions%2Fruns%2F19169468405%2Fjob%2F54799498288&data=05%7C02%7Cchristophe.leroy%40csgroup.eu%7Cf2e19b221ba740b2034e08de204158de%7C8b87af7d86474dc78df45f69a2011bb5%7C0%7C0%7C638983662203106300%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=UKQnlJWDKPfNCiYL8W7d2%2FTAhMhGbmxx8IDvy8jTbNQ%3D&reserved=0
>>
>>
>>>
>>> So I think what you want instead is:
>>>
>>> diff --git a/arch/powerpc/platforms/Kconfig.cputype
>>> b/arch/powerpc/platforms/Kconfig.cputype
>>> index 7b527d18aa5ee..1f5a1e587740c 100644
>>> --- a/arch/powerpc/platforms/Kconfig.cputype
>>> +++ b/arch/powerpc/platforms/Kconfig.cputype
>>> @@ -276,6 +276,7 @@ config PPC_E500
>>>           select FSL_EMB_PERFMON
>>>           bool
>>>           select ARCH_SUPPORTS_HUGETLBFS if PHYS_64BIT || PPC64
>>> +       select ARCH_HAS_GIGANTIC_PAGE if ARCH_SUPPORTS_HUGETLBFS
>>>           select PPC_SMP_MUXED_IPI
>>>           select PPC_DOORBELL
>>>           select PPC_KUEP
>>>
>>>
>>>
>>
>> @Christophe,
>>
>> I don't think even the above diff will fix the warning on PPC32.
>> The patch defines MAX_FOLIO_ORDER as P4D_ORDER...
>>
>> +#define MAX_FOLIO_ORDER        P4D_ORDER
>> +#define P4D_ORDER              (P4D_SHIFT - PAGE_SHIFT)
>>
>> and for ppc32 in..
>> include/asm-generic/pgtable-nop4d.h
>>      #define P4D_SHIFT        PGDIR_SHIFT
>>
>> Then in..
>> arch/powerpc/include/asm/nohash/32/pgtable.h
>>      #define PGDIR_SHIFT    (PAGE_SHIFT + PTE_INDEX_SIZE)
>>      #define PTE_INDEX_SIZE    PTE_SHIFT
>>
>> in...
>> arch/powerpc/include/asm/page_32.h
>>      #define PTE_SHIFT    (PAGE_SHIFT - PTE_T_LOG2)    /* full page */
>>
>>      #define PTE_T_LOG2    (__builtin_ffs(sizeof(pte_t)) - 1)
>>
>>
>> So if you see from above P4D_ORDER is coming down to PTE_INDEX_SIZE
>>
>> IIUC, that will cause MAX_FOLIO_ORDER to be 9 in case of e500mc 
>> machine type right?
>>
>> Can you please confirm if the above analysis looks correct to you?
> 
> Cristophe wrote
> 
> "
> Ah you are right, that's not enough. I was thinking that PGDIR_ORDER was
> the highest possible value ever but in fact not. PGDIR_SIZE is 4Mbytes
> so any page larger than that still triggers the warning. Here are the
> warnings I get on QEMU with corenet32_smp_defconfig
> "
> 
> And then we get
> 
> HugeTLB: registered 1.00 GiB page size, pre-allocated 0 pages
> HugeTLB: 0 KiB vmemmap can be freed for a 1.00 GiB page
> HugeTLB: registered 64.0 MiB page size, pre-allocated 1 pages
> HugeTLB: 0 KiB vmemmap can be freed for a 64.0 MiB page
> HugeTLB: registered 256 MiB page size, pre-allocated 1 pages
> HugeTLB: 0 KiB vmemmap can be freed for a 256 MiB page
> HugeTLB: registered 4.00 MiB page size, pre-allocated 0 pages
> HugeTLB: 0 KiB vmemmap can be freed for a 4.00 MiB page
> HugeTLB: registered 16.0 MiB page size, pre-allocated 0 pages
> HugeTLB: 0 KiB vmemmap can be freed for a 16.0 MiB page
> 
> 
> How could any of these larger sizes possibly ever get mapped into a page 
> table on 32bit? I'm probably missing something important :)
> 

Using contiguous entries in a table to describe larger pages.

See commit 7c44202e3609 ("powerpc/e500: use contiguous PMD instead of 
hugepd")

That's similar to what ARM64 does as far as I understand, see commit 
66b3923a1a0f ("arm64: hugetlb: add support for PTE contiguous bit")

Christophe