Re: [PATCH 3/3] ptdesc: Account page tables to memcgs again

[Matthew Wilcox <willy@infradead.org>]

On Wed, Feb 25, 2026 at 04:22:17PM +0000, Matthew Wilcox (Oracle) wrote:
>  static inline void __pagetable_ctor(struct ptdesc *ptdesc)
>  {
>  	pg_data_t *pgdat = NODE_DATA(memdesc_nid(ptdesc->pt_flags));
> +	struct mem_cgroup *memcg = pagetable_memcg(ptdesc);
>  
>  	__SetPageTable(ptdesc_page(ptdesc));
> -	mod_node_page_state(pgdat, NR_PAGETABLE, ptdesc_nr_pages(ptdesc));
> +	memcg_stat_mod(memcg, pgdat, NR_PAGETABLE, ptdesc_nr_pages(ptdesc));
>  }

It occurs to me that we're not holding the rcu_read_lock() here
(whereas we do for the other two callers).  I'm not quite clear
on what the rcu read lock is protecting here -- can it be that the
memcg is rcu-freed while a page table belongs to it?  Or does the task
existing prevent the memcg from being freed?

(is there documentation on this that I've been unable to find?)