2023-05-08 06:55:46

by Haifeng Xu

Subject: [PATCH] fork: optimize memcg_charge_kernel_stack() a bit

Since commit f1c1a9ee00e4 ("fork: Move memcg_charge_kernel_stack()
into CONFIG_VMAP_STACK"), memcg_charge_kernel_stack() has been moved
into CONFIG_VMAP_STACK block, so the CONFIG_VMAP_STACK check can be
removed.

Furthermore, memcg_charge_kernel_stack() is only invoked by
alloc_thread_stack_node() instead of dup_task_struct(). If
memcg_kmem_charge_page() fails, the uncharge process is handled in
memcg_charge_kernel_stack() itself instead of free_thread_stack(),
so remove the incorrect comments.

If memcg_charge_kernel_stack() fails to charge pages used by kernel
stack, only charged pages need to be uncharged. It's unnecessary to
uncharge those pages which memory cgroup pointer is NULL.

Signed-off-by: Haifeng Xu <[email protected]>
---
kernel/fork.c | 11 ++++-------
1 file changed, 4 insertions(+), 7 deletions(-)

diff --git a/kernel/fork.c b/kernel/fork.c
index ea332319dffe..50762a41aa4e 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -250,23 +250,20 @@ static int memcg_charge_kernel_stack(struct vm_struct *vm)
{
int i;
int ret;
+ int nr_charged = 0;

- BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
+ BUILD_BUG_ON(PAGE_SIZE % 1024 != 0);
BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);

for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
if (ret)
goto err;
+ nr_charged++;
}
return 0;
err:
- /*
- * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
- * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
- * ignore this page.
- */
- for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
+ for (i = 0; i < nr_charged; i++)
memcg_kmem_uncharge_page(vm->pages[i], 0);
return ret;
}
--
2.25.1

2023-05-13 00:36:44

by Andrew Morton

Subject: Re: [PATCH] fork: optimize memcg_charge_kernel_stack() a bit

On Mon, 8 May 2023 06:44:58 +0000 Haifeng Xu <[email protected]> wrote:

> Since commit f1c1a9ee00e4 ("fork: Move memcg_charge_kernel_stack()
> into CONFIG_VMAP_STACK"), memcg_charge_kernel_stack() has been moved
> into CONFIG_VMAP_STACK block, so the CONFIG_VMAP_STACK check can be
> removed.
>
> Furthermore, memcg_charge_kernel_stack() is only invoked by
> alloc_thread_stack_node() instead of dup_task_struct(). If
> memcg_kmem_charge_page() fails, the uncharge process is handled in
> memcg_charge_kernel_stack() itself instead of free_thread_stack(),
> so remove the incorrect comments.
>
> If memcg_charge_kernel_stack() fails to charge pages used by kernel
> stack, only charged pages need to be uncharged. It's unnecessary to
> uncharge those pages which memory cgroup pointer is NULL.
>
> ...
>
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -250,23 +250,20 @@ static int memcg_charge_kernel_stack(struct vm_struct *vm)
> {
> int i;
> int ret;
> + int nr_charged = 0;
>
> - BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
> + BUILD_BUG_ON(PAGE_SIZE % 1024 != 0);

This check now deserves some sort of award. I'll remove it.

> BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
>
> for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
> ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
> if (ret)
> goto err;
> + nr_charged++;
> }
> return 0;
> err:
> - /*
> - * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
> - * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
> - * ignore this page.
> - */
> - for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
> + for (i = 0; i < nr_charged; i++)
> memcg_kmem_uncharge_page(vm->pages[i], 0);
> return ret;

We don't really need nr_charged - we could just do

while (--i >= 0)

but nr_charged is straightforward, which is always a good choice.