Re: [PATCH v2 0/5] fork: Page operation cleanups in the fork code

[Mateusz Guzik <mjguzik@gmail.com>]

On Wed, May 07, 2025 at 02:46:26PM +0200, Linus Walleij wrote:
> This patch set consists of outtakes from a 1 year+ old
> patch set from Pasha, which all stand on their own.
> See:
> https://lore.kernel.org/all/20240311164638.2015063-1-pasha.tatashin@soleen.com/
> 
> What the code mainly does is make the fork.c file more
> oriented around pages and remove reliance on THREAD_SIZE.
> 
> These are good cleanups for readability and in one case
> (last patch using clear_page()) a performance improvement,
> so I split these off, rebased on v6.15-rc1, addressed review
> comments and send them separately.
> 
> All mentions of dynamic stack are removed from the patch
> set as we have no idea whether that will go anywhere.
> 
> This is mostly MM related so when the patches are ready
> I expect they would land in Andrew's patch stack.
> 

The ifdef clean up looks fine.

If readability is concerned I think a big problem is code duplication in
alloc_thread_stack_node().

The code instead obtain the stack space and only then do the memcg
charge and tsk var popoulation work.

The real problem here is that the cache is incredibly primitive and in
fact with a performance bug (ignoring NUMA affinity -- pages get added
to the cache and removed from it without any regard for that aspect or
the node parameter!)