Re: [PATCH] x86/mm: Do not verify W^X at boot up

[Linus Torvalds]

On Tue, Oct 25, 2022 at 3:16 AM Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> This seems to boot...

This looks much better, thanks.

But this:

> diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
> @@ -801,8 +803,9 @@ void __init poking_init(void)
>         spinlock_t *ptl;
>         pte_t *ptep;
>
> -       poking_mm = copy_init_mm();
> -       BUG_ON(!poking_mm);
> +       __poking_mm = init_mm;
> +       mm_init(&__poking_mm, NULL, __poking_mm.user_ns);
> +       poking_mm = &__poking_mm;

Should probably be just

        poking_mm = mm_alloc();

because we shouldn't be messing with 'mm_init()' in places like this,
and we shouldn't be exporting it either:

> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -1104,7 +1104,7 @@ static void mm_init_uprobes_state(struct mm_struct *mm)
>  #endif
>  }
>
> -static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
> +struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,

since "mm_alloc()" would seem to be exactly what we need, and it's
what execve() already uses. It creates a new VM with nothing attached,
ready to be populated.

Or is there some reason why mm_alloc() doesn't work? It does require
"current" and "current->user_ns" to be set up, but that should be true
even during _very_ early boot.

                Linus