Re: [PATCH v2 3/5] fork: Remove assumption that vm_area->nr_pages equals to THREAD_SIZE

[Mateusz Guzik <mjguzik@gmail.com>]

On Wed, May 07, 2025 at 02:46:29PM +0200, Linus Walleij wrote:
> From: Pasha Tatashin <pasha.tatashin@soleen.com>
> 
> In many places number of pages in the stack is detremined via
> (THREAD_SIZE / PAGE_SIZE). There is also a BUG_ON() that ensures that
> (THREAD_SIZE / PAGE_SIZE) is indeed equal to vm_area->nr_pages.
> 
> Consistently use vm_area->nr_pages to determine the actual number
> of pages allocated in the stack instead, so it is clear what is
> going on here.
> 
> The assignment of a local variable nr_pages in
> memcg_charge_kernel_stack() takes roughly the same amount of time
> as the BUG() assertion, and the two other sites arguably should
> have had a BUG() assertion as well.
> 
> Signed-off-by: Pasha Tatashin <pasha.tatashin@soleen.com>
> Link: https://lore.kernel.org/20240311164638.2015063-5-pasha.tatashin@soleen.com
> [linus.walleij@linaro.orh: Rebased, initialized helpers variables in declaration]
> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
> ---
>  kernel/fork.c | 14 +++++++-------
>  1 file changed, 7 insertions(+), 7 deletions(-)
> 
> diff --git a/kernel/fork.c b/kernel/fork.c
> index 8b8457562740c114c640a8cc230876f6a286b246..c60a0ec61a421324e3733b506d99531c5965cbc6 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -253,10 +253,9 @@ static int memcg_charge_kernel_stack(struct vm_struct *vm_area)
>  	int i;
>  	int ret;
>  	int nr_charged = 0;
> +	int nr_pages = vm_area->nr_pages;
>  
> -	BUG_ON(vm_area->nr_pages != THREAD_SIZE / PAGE_SIZE);
> -
> -	for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
> +	for (i = 0; i < nr_pages; i++) {
>  		ret = memcg_kmem_charge_page(vm_area->pages[i], GFP_KERNEL, 0);
>  		if (ret)
>  			goto err;
> @@ -516,9 +515,10 @@ static void account_kernel_stack(struct task_struct *tsk, int account)
>  {
>  	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
>  		struct vm_struct *vm_area = task_stack_vm_area(tsk);
> +		int nr_pages = vm_area->nr_pages;
>  		int i;
>  
> -		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
> +		for (i = 0; i < nr_pages; i++)
>  			mod_lruvec_page_state(vm_area->pages[i], NR_KERNEL_STACK_KB,
>  					      account * (PAGE_SIZE / 1024));
>  	} else {
> @@ -535,11 +535,11 @@ void exit_task_stack_account(struct task_struct *tsk)
>  	account_kernel_stack(tsk, -1);
>  
>  	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
> -		struct vm_struct *vm_area;
> +		struct vm_struct *vm_area = task_stack_vm_area(tsk);
> +		int nr_pages = vm_area->nr_pages;
>  		int i;
>  
> -		vm_area = task_stack_vm_area(tsk);
> -		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
> +		for (i = 0; i < nr_pages; i++)
>  			memcg_kmem_uncharge_page(vm_area->pages[i], 0);
>  	}
>  }
> 
> -- 
> 2.49.0
> 
> 

I think the change makes it very much less clear what's going on here as it
suggests the page count for stacks is variable instead of fixed.

I can agree "THREAD_SIZE / PAGE_SIZE" open-coded is not the best way to
express it though.

How about THREAD_STACK_PAGE_COUNT or similar to hide the above division?