On 2/28/2025 11:21 PM, Ackerley Tng wrote:
> Vlastimil Babka <vbabka@xxxxxxx> writes:
>
>> On 2/26/25 09:25, Shivank Garg wrote:
>>> From: Shivansh Dhiman <shivansh.dhiman@xxxxxxx>
>>>
>>> Add NUMA mempolicy support to the filemap allocation path by introducing
>>> new APIs that take a mempolicy argument:
>>> - filemap_grab_folio_mpol()
>>> - filemap_alloc_folio_mpol()
>>> - __filemap_get_folio_mpol()
>>>
>>> These APIs allow callers to specify a NUMA policy during page cache
>>> allocations, enabling fine-grained control over memory placement. This is
>>> particularly needed by KVM when using guest-memfd memory backends, where
>>> the guest memory needs to be allocated according to the NUMA policy
>>> specified by VMM.
>>>
>>> The existing non-mempolicy APIs remain unchanged and continue to use the
>>> default allocation behavior.
>>>
>>> Signed-off-by: Shivansh Dhiman <shivansh.dhiman@xxxxxxx>
>>> Signed-off-by: Shivank Garg <shivankg@xxxxxxx>
>>
>> <snip>
>>
>>> --- a/mm/filemap.c
>>> +++ b/mm/filemap.c
>>> @@ -1001,11 +1001,17 @@ int filemap_add_folio(struct address_space *mapping, struct folio *folio,
>>> EXPORT_SYMBOL_GPL(filemap_add_folio);
>>>
>>> #ifdef CONFIG_NUMA
>>> -struct folio *filemap_alloc_folio_noprof(gfp_t gfp, unsigned int order)
>>> +struct folio *filemap_alloc_folio_mpol_noprof(gfp_t gfp, unsigned int order,
>>> + struct mempolicy *mpol)
>>> {
>>> int n;
>>> struct folio *folio;
>>>
>>> + if (mpol)
>>> + return folio_alloc_mpol_noprof(gfp, order, mpol,
>>> + NO_INTERLEAVE_INDEX,
>
> Could we pass in the interleave index instead of hard-coding it?
Good point.
I'll modify this to allow passing the interleave index.
>
>>> + numa_node_id());
>>> +
>>> if (cpuset_do_page_mem_spread()) {
>>> unsigned int cpuset_mems_cookie;
>>> do {
>>> @@ -1018,6 +1024,12 @@ struct folio *filemap_alloc_folio_noprof(gfp_t gfp, unsigned int order)
>>> }
>>> return folio_alloc_noprof(gfp, order);
>>> }
>>> +EXPORT_SYMBOL(filemap_alloc_folio_mpol_noprof);
>>> +
>>> +struct folio *filemap_alloc_folio_noprof(gfp_t gfp, unsigned int order)
>>> +{
>>> + return filemap_alloc_folio_mpol_noprof(gfp, order, NULL);
>>> +}
>>> EXPORT_SYMBOL(filemap_alloc_folio_noprof);
>>> #endif
>>
>> Here it seems to me:
>>
>> - filemap_alloc_folio_noprof() could stay unchanged
>> - filemap_alloc_folio_mpol_noprof() would
>> - call folio_alloc_mpol_noprof() if (mpol)
>> - call filemap_alloc_folio_noprof() otherwise
>>
>> The code would be a bit more clearly structured that way?
>>
>
> I feel that the original proposal makes it clearer that for all filemap
> folio allocations, if mpol is defined, anything to do with cpuset's page
> spread is overridden. Just a slight preference though. I do also agree
> that having filemap_alloc_folio_mpol_noprof() call
> filemap_alloc_folio_noprof() would result in fewer changes.
>
Your proposed structure makes sense.
I'll update the patch to add these suggestions in the next version.
Thanks,
Shivank
>>> @@ -1881,11 +1893,12 @@ void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
>>> }
>>>
>>> /**
>>> - * __filemap_get_folio - Find and get a reference to a folio.
>>> + * __filemap_get_folio_mpol - Find and get a reference to a folio.
>>> * @mapping: The address_space to search.
>>> * @index: The page index.
>>> * @fgp_flags: %FGP flags modify how the folio is returned.
>>> * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
>>> + * @mpol: The mempolicy to apply when allocating a new folio.
>>> *
>>> * Looks up the page cache entry at @mapping & @index.
>>> *
>>> @@ -1896,8 +1909,8 @@ void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
>>> *
>>> * Return: The found folio or an ERR_PTR() otherwise.
>>> */
>>> -struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
>>> - fgf_t fgp_flags, gfp_t gfp)
>>> +struct folio *__filemap_get_folio_mpol(struct address_space *mapping, pgoff_t index,
>>> + fgf_t fgp_flags, gfp_t gfp, struct mempolicy *mpol)
>>> {
>>> struct folio *folio;
>>>
>>> @@ -1967,7 +1980,7 @@ struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
>>> err = -ENOMEM;
>>> if (order > min_order)
>>> alloc_gfp |= __GFP_NORETRY | __GFP_NOWARN;
>>> - folio = filemap_alloc_folio(alloc_gfp, order);
>>> + folio = filemap_alloc_folio_mpol(alloc_gfp, order, mpol);
>>> if (!folio)
>>> continue;
>>>
>>> @@ -2003,6 +2016,13 @@ struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
>>> folio_clear_dropbehind(folio);
>>> return folio;
>>> }
>>> +EXPORT_SYMBOL(__filemap_get_folio_mpol);
>>> +
>>> +struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
>>> + fgf_t fgp_flags, gfp_t gfp)
>>> +{
>>> + return __filemap_get_folio_mpol(mapping, index, fgp_flags, gfp, NULL);
>>> +}
>>> EXPORT_SYMBOL(__filemap_get_folio);
>>>
>>> static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
