2021-04-09 05:56:40

by Anshuman Khandual

[permalink] [raw]
Subject: [PATCH] mm/page_alloc: Ensure that HUGETLB_PAGE_ORDER is less than MAX_ORDER

pageblock_order must always be less than MAX_ORDER, otherwise it might lead
to an warning during boot. A similar problem got fixed on arm64 platform
with the commit 79cc2ed5a716 ("arm64/mm: Drop THP conditionality from
FORCE_MAX_ZONEORDER"). Assert the above condition before HUGETLB_PAGE_ORDER
gets assigned as pageblock_order. This will help detect the problem earlier
on platforms where HUGETLB_PAGE_SIZE_VARIABLE is enabled.

Cc: Andrew Morton <[email protected]>
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Anshuman Khandual <[email protected]>
---
mm/page_alloc.c | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 604dcd69397b..81b7460e1228 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -7068,10 +7068,17 @@ void __init set_pageblock_order(void)
if (pageblock_order)
return;

- if (HPAGE_SHIFT > PAGE_SHIFT)
+ if (HPAGE_SHIFT > PAGE_SHIFT) {
+ /*
+ * pageblock_order must always be less than
+ * MAX_ORDER. So does HUGETLB_PAGE_ORDER if
+ * that is being assigned here.
+ */
+ WARN_ON(HUGETLB_PAGE_ORDER >= MAX_ORDER);
order = HUGETLB_PAGE_ORDER;
- else
+ } else {
order = MAX_ORDER - 1;
+ }

/*
* Assume the largest contiguous order of interest is a huge page.
--
2.20.1


2021-04-09 08:25:52

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH] mm/page_alloc: Ensure that HUGETLB_PAGE_ORDER is less than MAX_ORDER

On 09.04.21 07:55, Anshuman Khandual wrote:
> pageblock_order must always be less than MAX_ORDER, otherwise it might lead
> to an warning during boot. A similar problem got fixed on arm64 platform
> with the commit 79cc2ed5a716 ("arm64/mm: Drop THP conditionality from
> FORCE_MAX_ZONEORDER"). Assert the above condition before HUGETLB_PAGE_ORDER
> gets assigned as pageblock_order. This will help detect the problem earlier
> on platforms where HUGETLB_PAGE_SIZE_VARIABLE is enabled.
>
> Cc: Andrew Morton <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Signed-off-by: Anshuman Khandual <[email protected]>
> ---
> mm/page_alloc.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 604dcd69397b..81b7460e1228 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -7068,10 +7068,17 @@ void __init set_pageblock_order(void)
> if (pageblock_order)
> return;
>
> - if (HPAGE_SHIFT > PAGE_SHIFT)
> + if (HPAGE_SHIFT > PAGE_SHIFT) {
> + /*
> + * pageblock_order must always be less than
> + * MAX_ORDER. So does HUGETLB_PAGE_ORDER if
> + * that is being assigned here.
> + */
> + WARN_ON(HUGETLB_PAGE_ORDER >= MAX_ORDER);

Can't that be a BUILD_BUG_ON() ?

> order = HUGETLB_PAGE_ORDER;
> - else
> + } else {
> order = MAX_ORDER - 1;
> + }
>
> /*
> * Assume the largest contiguous order of interest is a huge page.
>


--
Thanks,

David / dhildenb

2021-04-09 08:40:29

by Anshuman Khandual

[permalink] [raw]
Subject: Re: [PATCH] mm/page_alloc: Ensure that HUGETLB_PAGE_ORDER is less than MAX_ORDER



On 4/9/21 1:54 PM, David Hildenbrand wrote:
> On 09.04.21 07:55, Anshuman Khandual wrote:
>> pageblock_order must always be less than MAX_ORDER, otherwise it might lead
>> to an warning during boot. A similar problem got fixed on arm64 platform
>> with the commit 79cc2ed5a716 ("arm64/mm: Drop THP conditionality from
>> FORCE_MAX_ZONEORDER"). Assert the above condition before HUGETLB_PAGE_ORDER
>> gets assigned as pageblock_order. This will help detect the problem earlier
>> on platforms where HUGETLB_PAGE_SIZE_VARIABLE is enabled.
>>
>> Cc: Andrew Morton <[email protected]>
>> Cc: [email protected]
>> Cc: [email protected]
>> Signed-off-by: Anshuman Khandual <[email protected]>
>> ---
>>   mm/page_alloc.c | 11 +++++++++--
>>   1 file changed, 9 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 604dcd69397b..81b7460e1228 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -7068,10 +7068,17 @@ void __init set_pageblock_order(void)
>>       if (pageblock_order)
>>           return;
>>   -    if (HPAGE_SHIFT > PAGE_SHIFT)
>> +    if (HPAGE_SHIFT > PAGE_SHIFT) {
>> +        /*
>> +         * pageblock_order must always be less than
>> +         * MAX_ORDER. So does HUGETLB_PAGE_ORDER if
>> +         * that is being assigned here.
>> +         */
>> +        WARN_ON(HUGETLB_PAGE_ORDER >= MAX_ORDER);
>
> Can't that be a BUILD_BUG_ON() ?

Yes, it can be. Probably might be appropriate as well, given that both
the arguments here are compile time constants. Okay, will change.

>
>>           order = HUGETLB_PAGE_ORDER;
>> -    else
>> +    } else {
>>           order = MAX_ORDER - 1;
>> +    }
>>         /*
>>        * Assume the largest contiguous order of interest is a huge page.
>>
>
>