Re: [PATCH 0/3] mm: treewide: make get_free_pages() and return void *

[Vlastimil Babka <vbabka@suse.cz>]

On 10/20/25 11:13, David Hildenbrand wrote:
> On 20.10.25 11:08, Jiri Slaby wrote:
>> On 20. 10. 25, 11:02, David Hildenbrand wrote:
>>> Regarding the metadata overhead, in 2015 Linus wrote in that thread:
>>>
>>> "Long ago, allocating a page using kmalloc() was a bad idea, because
>>> there was overhead for it in the allocation and the code.
>>>
>>> These days, kmalloc() not only doesn't have the allocation overhead,
>>> but may actually scale better too, thanks to percpu caches etc."
>>>
>>> What's that status of that 10 years later?
>> 
>> AFAI skimmed through the code, for allocations > 2 pages
>> (KMALLOC_MAX_CACHE_SIZE) -- if size is a constant -- slub resorts to
>> alloc_pages().
>> 
>> For smaller ones (1 and 2 pages), there is a very little overhead in
>> struct slab -- mm people, please correct me if I am wrong.
> 
> If it's really only "struct slab", then there is currently no overhead. 
> Once it is decoupled from "struct page", there would be some.

Yes, but there's potentially better scalability and more debugging
possibilities as benefits.

> IIUC, I'm surprised that larger allocations wouldn't currently end up in 
> PageSlab() pages.
Can you elaborate why surprised?