Re: [PATCH v6 6/6] mm/page_isolation: remove migratetype parameter from more functions.

[Zi Yan <ziy@nvidia.com>]

On 2 Jun 2025, at 12:25, David Hildenbrand wrote:

> On 02.06.25 16:18, Zi Yan wrote:
>> On 30 May 2025, at 16:55, David Hildenbrand wrote:
>>
>>> On 30.05.25 22:46, Zi Yan wrote:
>>>> On 30 May 2025, at 16:08, David Hildenbrand wrote:
>>>>
>>>>> On 30.05.25 21:58, Zi Yan wrote:
>>>>>> On 30 May 2025, at 15:56, David Hildenbrand wrote:
>>>>>>
>>>>>>> On 30.05.25 18:22, 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 provide the information. At the
>>>>>>>> same time change isolation flags to enum pb_isolate_mode
>>>>>>>> (PB_ISOLATE_MODE_MEM_OFFLINE, PB_ISOLATE_MODE_CMA_ALLOC,
>>>>>>>> PB_ISOLATE_MODE_OTHER). Remove REPORT_FAILURE and check
>>>>>>>> MEMORY_OFFLINE instead, since only PB_ISOLATE_MODE_MEM_OFFLINE reports
>>>>>>>> isolation failures.
>>>>>>>>
>>>>>>>> alloc_contig_range() no longer needs migratetype. Replace it with
>>>>>>>> enum acr_flags_t to tell if an allocation is for CMA. So does
>>>>>>>> __alloc_contig_migrate_range().
>>>>>>>>
>>>>>>>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>>>>>>>> ---
>>>>>>>>      drivers/virtio/virtio_mem.c    |  2 +-
>>>>>>>>      include/linux/gfp.h            |  9 ++++-
>>>>>>>>      include/linux/page-isolation.h | 20 ++++++++--
>>>>>>>>      include/trace/events/kmem.h    | 14 ++++---
>>>>>>>>      mm/cma.c                       |  2 +-
>>>>>>>>      mm/memory_hotplug.c            |  6 +--
>>>>>>>>      mm/page_alloc.c                | 27 ++++++-------
>>>>>>>>      mm/page_isolation.c            | 70 +++++++++++++++-------------------
>>>>>>>>      8 files changed, 82 insertions(+), 68 deletions(-)
>>>>>>>>
>>>>>>>> diff --git a/drivers/virtio/virtio_mem.c b/drivers/virtio/virtio_mem.c
>>>>>>>> index 56d0dbe62163..6bce70b139b2 100644
>>>>>>>> --- a/drivers/virtio/virtio_mem.c
>>>>>>>> +++ b/drivers/virtio/virtio_mem.c
>>>>>>>> @@ -1243,7 +1243,7 @@ static int virtio_mem_fake_offline(struct virtio_mem *vm, unsigned long pfn,
>>>>>>>>      		if (atomic_read(&vm->config_changed))
>>>>>>>>      			return -EAGAIN;
>>>>>>>>     -		rc = alloc_contig_range(pfn, pfn + nr_pages, MIGRATE_MOVABLE,
>>>>>>>> +		rc = alloc_contig_range(pfn, pfn + nr_pages, ACR_OTHER,
>>>>>>>>      					GFP_KERNEL);
>>>>>>>>      		if (rc == -ENOMEM)
>>>>>>>>      			/* whoops, out of memory */
>>>>>>>> diff --git a/include/linux/gfp.h b/include/linux/gfp.h
>>>>>>>> index be160e8d8bcb..51990d571e3e 100644
>>>>>>>> --- a/include/linux/gfp.h
>>>>>>>> +++ b/include/linux/gfp.h
>>>>>>>> @@ -423,9 +423,16 @@ static inline bool gfp_compaction_allowed(gfp_t gfp_mask)
>>>>>>>>      extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma);
>>>>>>>>       #ifdef CONFIG_CONTIG_ALLOC
>>>>>>>> +
>>>>>>>> +enum acr_flags_t {
>>>>>>>> +	ACR_CMA,	// CMA allocation
>>>>>>>> +	ACR_OTHER,	// other allocation
>>>>>>>> +};
>>>>>>>
>>>>>>> Hm, enum != flags.
>>>>>>>
>>>>>>> If you want to use flags, then just have ACR_CMA. ACR_OTHER is implied if not set.
>>>>>>>
>>>>>>> And ACR_CMA would then have to be "1" etc.
>>>>>>
>>>>>> I have a fixup to change acr_flags_t to acr_mode.
>>>>>>
>>>>>>>
>>>>>>>> +
>>>>>>>>      /* The below functions must be run on a range from a single zone. */
>>>>>>>>      extern int alloc_contig_range_noprof(unsigned long start, unsigned long end,
>>>>>>>> -			      unsigned migratetype, gfp_t gfp_mask);
>>>>>>>> +				     enum acr_flags_t alloc_flags,
>>>>>>>> +				     gfp_t gfp_mask);
>>>>>>>>      #define alloc_contig_range(...)			alloc_hooks(alloc_contig_range_noprof(__VA_ARGS__))
>>>>>>>>       extern struct page *alloc_contig_pages_noprof(unsigned long nr_pages, gfp_t gfp_mask,
>>>>>>>> diff --git a/include/linux/page-isolation.h b/include/linux/page-isolation.h
>>>>>>>> index 7a681a49e73c..3e2f960e166c 100644
>>>>>>>> --- a/include/linux/page-isolation.h
>>>>>>>> +++ b/include/linux/page-isolation.h
>>>>>>>> @@ -38,8 +38,20 @@ static inline void set_pageblock_isolate(struct page *page)
>>>>>>>>      }
>>>>>>>>      #endif
>>>>>>>>     -#define MEMORY_OFFLINE	0x1
>>>>>>>> -#define REPORT_FAILURE	0x2
>>>>>>>> +/*
>>>>>>>> + * Pageblock isolation modes:
>>>>>>>> + * PB_ISOLATE_MODE_MEM_OFFLINE - isolate to offline (!allocate) memory
>>>>>>>> + *				 e.g., skip over PageHWPoison() pages and
>>>>>>>> + *				 PageOffline() pages. Unmovable pages will be
>>>>>>>> + *				 reported in this mode.
>>>>>>>> + * PB_ISOLATE_MODE_CMA_ALLOC   - isolate for CMA allocations
>>>>>>>> + * PB_ISOLATE_MODE_OTHER       - isolate for other purposes
>>>>>>>> + */
>>>>>>>> +enum pb_isolate_mode {
>>>>>>>> +	PB_ISOLATE_MODE_MEM_OFFLINE,
>>>>>>>> +	PB_ISOLATE_MODE_CMA_ALLOC,
>>>>>>>> +	PB_ISOLATE_MODE_OTHER,
>>>>>>>> +};
>>>>>>>
>>>>>>> It's late on friady, but it looks like we are duplicating things here.
>>>>>>>
>>>>>>> Let me think about that once my brain is recharged :)
>>>>>>
>>>>>> Sure. Take your time.
>>>>>
>>>>> Could we abstract both settings and use a single one? Then, we could simply reject if MEM_OFFLINE is passed into alloc_contig_range().
>>>>>
>>>>> alloc_contig_pages and page isolation, hmmmm, MEM_OFFLINE is kind-of an allocation. CMA is an allocation.
>>>>>
>>>>> Just an idea, not sure ...
>>>>
>>>> I think so.
>>>>
>>>> This is the fixup of removing acr_flags_t. It is on top of the original
>>>> patch. Take your time. I guess I will send V7 with all fixups next week.
>>>
>>> Only wondering if we should rename "pb_isolate_mode" to something more abstract, that covers both use cases clearer.
>>>
>>> Hmmm.
>>
>> It is specifying the purpose of either alloc_contig_range() or
>> test_pages_isolated(), but these two use cases do not have high level
>> connection AFAICT.
>
> Hm, then maybe just keep it as is for now, and have the translation from ACP -> isolatetype.
>
> isolation modes should probably be an enum.
>
> acp might long-term benefit from flags, where we would essentially for now only have a single flag ("ACP_CMA").

OK, let me send a fixup to my V7.

Best Regards,
Yan, Zi