Re: [RFC PATCH 4/4] mm/page_isolation: remove migratetype parameter from more functions.

[Zi Yan <ziy@nvidia.com>]

[-- Attachment #1: Type: text/plain, Size: 5450 bytes --]

On 4 Sep 2024, at 4:50, David Hildenbrand wrote:

> On 04.09.24 04:02, Zi Yan wrote:
>> On 2 Sep 2024, at 12:42, David Hildenbrand wrote:
>>
>>> On 02.09.24 17:34, Zi Yan wrote:
>>>> On 2 Sep 2024, at 5:06, David Hildenbrand wrote:
>>>>
>>>>> On 28.08.24 22:22, Zi Yan wrote:
>>>>>> migratetype is no longer overwritten during pageblock isolation,
>>>>>> start_isolate_page_range(), has_unmovable_pages(), and
>>>>>> set_migratetype_isolate() no longer need which migratetype to restore
>>>>>> during isolation failure. For has_unmoable_pages(), it needs to know if
>>>>>> the isolation is for CMA allocation, so adding CMA_ALLOCATION to isolation
>>>>>> flags to provide the information.
>>>>>>
>>>>>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>>>>>> ---
>>>>>>     include/linux/page-isolation.h |  3 ++-
>>>>>>     mm/memory_hotplug.c            |  1 -
>>>>>>     mm/page_alloc.c                |  4 +++-
>>>>>>     mm/page_isolation.c            | 27 +++++++++++----------------
>>>>>>     4 files changed, 16 insertions(+), 19 deletions(-)
>>>>>>
>>>>>> diff --git a/include/linux/page-isolation.h b/include/linux/page-isolation.h
>>>>>> index c2a1bd621561..e94117101b6c 100644
>>>>>> --- a/include/linux/page-isolation.h
>>>>>> +++ b/include/linux/page-isolation.h
>>>>>> @@ -32,13 +32,14 @@ static inline bool is_migrate_isolate(int migratetype)
>>>>>>      #define MEMORY_OFFLINE	0x1
>>>>>>     #define REPORT_FAILURE	0x2
>>>>>> +#define CMA_ALLOCATION	0x4
>>>>>>      void set_pageblock_migratetype(struct page *page, int migratetype);
>>>>>>      bool move_freepages_block_isolate(struct zone *zone, struct page *page);
>>>>>>      int start_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn,
>>>>>> -			     int migratetype, int flags, gfp_t gfp_flags);
>>>>>> +			     int flags, gfp_t gfp_flags);
>>>>>>      void undo_isolate_page_range(unsigned long start_pfn, unsigned long end_pfn);
>>>>>>    diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>>>>>> index 4265272faf4c..fe0b71e0f307 100644
>>>>>> --- a/mm/memory_hotplug.c
>>>>>> +++ b/mm/memory_hotplug.c
>>>>>> @@ -1993,7 +1993,6 @@ int offline_pages(unsigned long start_pfn, unsigned long nr_pages,
>>>>>>      	/* set above range as isolated */
>>>>>>     	ret = start_isolate_page_range(start_pfn, end_pfn,
>>>>>> -				       MIGRATE_MOVABLE,
>>>>>>     				       MEMORY_OFFLINE | REPORT_FAILURE,
>>>>>>     				       GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL);
>>>>>>     	if (ret) {
>>>>>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>>>>>> index 4d06932ba69a..c60bb95d7e65 100644
>>>>>> --- a/mm/page_alloc.c
>>>>>> +++ b/mm/page_alloc.c
>>>>>> @@ -6607,7 +6607,9 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
>>>>>>     	 * put back to page allocator so that buddy can use them.
>>>>>>     	 */
>>>>>>    -	ret = start_isolate_page_range(start, end, migratetype, 0, gfp_mask);
>>>>>> +	ret = start_isolate_page_range(start, end,
>>>>>> +			migratetype == MIGRATE_CMA ? CMA_ALLOCATION : 0,
>>>>>
>>>>> Can we have flags for alloc_contig_range() instead of passing in a (weird) migratetype?
>>>>>
>>>>> Then, we should make sure that we warn if we try a CMA allocation on any pageblock that is not of type CMA.
>>>>
>>>> Sure. I will expose the existing isolation flags (MEMORY_OFFLINE, REPORT_FAILURE,
>>>> and CMA_ALLOCATION) as alloc_contig_range() parameter to replace migratetype one.
>>>>
>>>
>>> Maybe we want some proper, distinct alloc_contig_range() falgs "acr_flags_t". Might be cleanest, to express anything that doesn't fall into the gfp_t flag category.
>>>
>>> Exposing MEMORY_OFFLINE feels wrong, for example.
>>
>> OK, it seems that I mixed up of start_isolate_page_range() flags with
>> alloc_contig_range() flags. Let me clarify them below.
>>
>> For start_isolate_page_range(), memory offline calls it separately and
>> needs MEMORY_OFFLINE and REPORT_FAILURE; CMA allocation uses it via
>> alloc_contig_range() and needs a flag (like CMA_ALLOCATION) for its
>> own checks.
>>
>> BTW, it seems to me that MEMORY_OFFLINE and REPORT_FAILURE can be merged,
>> since they are always used together. Let me know if you disagree.
>
> I think there was a discussion about possibly using REPORT_FAILURE in other cases, but I agree that we might just merge them at this point.
>
>>
>> For alloc_contig_range(), migratetype parameter is what you are talking about
>> above. There are two callers: cma_alloc() and alloc_contig_pages().
>> The acr_flags_t is basically a caller id. Something like?
>> enum acr_flags_t {
>> 	ACR_CMA_ALLOC,
>> 	ACR_CONTIG_PAGES,
>> };
>
> I'd do something like:
>
> typedef unsigned int __bitwise acr_flags_t;
>
> #define ACR_CMA		((__force acr_flags_t)BIT(0))
>
> No need for "ACR_CONTIG_PAGES", it's implicit if the CMA flag is not set.

Got it. Will use this in the next version.
>
>
>>
>> And ACR_CMA_ALLOC needs to be translated to CMA_ALLOCATION when
>> start_isolate_page_range() is called.
>
> Yes.
>
>>
>> BTW, after removing migratetype parameter from alloc_contig_range(),
>> the tracepoint in __alloc_contig_migrate_range() needs to be changed to
>> use acr_flags_t, since I do not think we want to convert acr_flags_t
>> back to migratetype.
>
> Sure, feel free to modify these tracepoints as it suits you.

Thanks. :)

--
Best Regards,
Yan, Zi

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 854 bytes --]