On 10/15/2019 04:15 PM, Michal Hocko wrote:
> On Tue 15-10-19 14:51:41, Anshuman Khandual wrote:
> [...]
>> +/**
>> + * alloc_gigantic_page_order() -- tries to allocate given order of pages
>> + * @order: allocation order (greater than MAX_ORDER)
>> + * @gfp_mask: GFP mask to use during compaction
>> + * @nid: allocation node
>> + * @nodemask: allocation nodemask
>> + *
>> + * This routine is an wrapper around alloc_contig_range() which scans over
>> + * all zones on an applicable zonelist to find a contiguous pfn range which
>> + * can the be allocated with alloc_contig_range(). This routine is intended
>> + * to be used for allocations greater than MAX_ORDER.
>> + *
>> + * Return: page on success or NULL on failure. On success a memory block
>> + * of 'order' starting with 'page' has been allocated successfully. Memory
>> + * allocated here needs to be freed with free_contig_range().
>> + */
>> +struct page *alloc_gigantic_page_order(unsigned int order, gfp_t gfp_mask,
>> + int nid, nodemask_t *nodemask)
>
> One of the objections when Mike has proposed a similar thing last year
> was that the interface shouldn't be order bases
> http://lkml.kernel.org/r/20180423000943.go17...@dhcp22.suse.cz
>
> Order based API makes sense for the buddy allocator but why should we
> restrict sizes like that for an allocator that is capable to allocate
> arbitrary page sized requests?
Fair enough, will change it. Anyways we calculate nr_pages from the order
argument at the very beginning.
