Re: [MINI-OS PATCH 02/12] kexec: add final kexec stage

[Jason Andryuk <jandryuk@xxxxxxxxx>]

On Fri, Mar 21, 2025 at 5:25 AM Juergen Gross <jgross@xxxxxxxx> wrote:
>
> Add the code and data definitions of the final kexec stage.
>
> Put the code and related data into a dedicated section in order to be
> able to copy it to another location. For this reason there must be no
> absolute relocations being used in the code or data.
>
> Being functionally related, add a function for adding a final kexec
> action.
>
> Signed-off-by: Juergen Gross <jgross@xxxxxxxx>

> --- /dev/null
> +++ b/arch/x86/kexec.c
> @@ -0,0 +1,109 @@

> +
> +/*
> + * Final stage of kexec. Copies all data to the final destinations, zeroes
> + * .bss and activates new kernel.
> + * Must be called with interrupts off. Stack, code and data must be
> + * accessible via identity mapped virtual addresses (virt == phys). Copying
> + * and zeroing is done using virtual addresses.
> + * No relocations inside the function are allowed, as it is copied to an
> + * allocated page before being executed.

"page" is stated here.  Do we need an ASSERT later?

> + */

> +void do_kexec(void *kexec_page)
> +{
> +    unsigned long actions;
> +    unsigned long stack;
> +    unsigned long final;
> +    unsigned long phys;
> +
> +    actions = get_kexec_addr(kexec_page, kexec_actions);
> +    stack = get_kexec_addr(kexec_page, kexec_stack + KEXEC_STACK_LONGS);
> +    final = get_kexec_addr(kexec_page, kexec_final);
> +    phys = get_kexec_addr(kexec_page, kexec_phys);
> +
> +    memcpy(kexec_page, _kexec_start, KEXEC_SECSIZE);
> +    asm("cli\n\t"
> +        "mov %0, %%"ASM_SP"\n\t"
> +        "mov %1, %%"ASM_ARG1"\n\t"
> +        "mov %2, %%"ASM_ARG2"\n\t"
> +        "jmp *%3"
> +        :"=m" (stack), "=m" (actions), "=m" (phys)

Aren't these inputs and not outputs?

> +        :"m" (final));
> +}
> +
> +#endif /* CONFIG_KEXEC */


> diff --git a/include/kexec.h b/include/kexec.h
> index 6fd96774..722be456 100644
> --- a/include/kexec.h
> +++ b/include/kexec.h
> @@ -1,7 +1,34 @@

> +
> +int kexec_add_action(int action, void *dest, void *src, unsigned int len);
> +
> +#define KEXEC_SECSIZE ((unsigned long)_kexec_end - (unsigned 
> long)_kexec_start)

Add a build assertion here?  Or maybe the correct amount is allocated
and it doesn't matter.

Generally looks good.

Regards,
Jason