Re: [PATCH 3/3] x86/sev-es: Improve comments in and around __sev_es_ist_enter/exit()

From: Borislav Petkov
Date: Wed Feb 17 2021 - 13:01:48 EST


On Wed, Feb 17, 2021 at 01:01:43PM +0100, Joerg Roedel wrote:
> From: Joerg Roedel <jroedel@xxxxxxx>
>
> Better explain why this code is necessary and what it is doing.
>
> Signed-off-by: Joerg Roedel <jroedel@xxxxxxx>
> ---
> arch/x86/kernel/sev-es.c | 23 ++++++++++++++++-------
> 1 file changed, 16 insertions(+), 7 deletions(-)
>
> diff --git a/arch/x86/kernel/sev-es.c b/arch/x86/kernel/sev-es.c
> index 0df38b185d53..79241bc45f25 100644
> --- a/arch/x86/kernel/sev-es.c
> +++ b/arch/x86/kernel/sev-es.c
> @@ -127,14 +127,20 @@ static __always_inline bool on_vc_stack(unsigned long sp)
> }
>
> /*
> - * This function handles the case when an NMI is raised in the #VC exception
> - * handler entry code. In this case, the IST entry for #VC must be adjusted, so
> - * that any subsequent #VC exception will not overwrite the stack contents of the
> - * interrupted #VC handler.
> + * This function handles the case when an NMI is raised in the #VC
> + * exception handler entry code, before the #VC handler has switched off
> + * its IST stack. In this case, the IST entry for #VC must be adjusted,
> + * so that any nested #VC exception will not overwrite the stack
> + * contents of the interrupted #VC handler.
> *
> * The IST entry is adjusted unconditionally so that it can be also be
> - * unconditionally adjusted back in sev_es_ist_exit(). Otherwise a nested
> - * sev_es_ist_exit() call may adjust back the IST entry too early.
> + * unconditionally adjusted back in __sev_es_ist_exit(). Otherwise a
> + * nested sev_es_ist_exit() call may adjust back the IST entry too
> + * early.
> + *
> + * The __sev_es_ist_enter() and __sev_es_ist_exit() functions always run
> + * on the NMI IST stack, as they are only called from NMI handling code
> + * right now.
> */
> void noinstr __sev_es_ist_enter(struct pt_regs *regs)
> {
> @@ -143,7 +149,10 @@ void noinstr __sev_es_ist_enter(struct pt_regs *regs)
> /* Read old IST entry */
> old_ist = __this_cpu_read(cpu_tss_rw.x86_tss.ist[IST_INDEX_VC]);
>
> - /* Make room on the IST stack */
> + /*
> + * Make room on the IST stack - Reserve 8 bytes to store the old
> + * IST entry.
> + */
> if (on_vc_stack(regs->sp) &&
> !user_mode(regs) &&
> !from_syscall_gap(regs))
> --

Yah, and then we probably should simplify this __sev_es_ist_enter()
function even more as it is not easy to grok.

For example, the ALIGN_DOWN(regs->sp, 8) is not really needed, right?

Also, both branches do "- sizeof(old_ist);" so you can just as well do
it unconditionally.

And the sizeof(old_ist) is just a confusing way to write 8, right? We're
64-bit only so there's no need for that, I'd say.

And then you probably should change the comments from

/* Store old IST entry */

and

/* Set new IST entry */

to something like:

/*
* If on the #VC IST stack, new_ist gets set to point one stack slot
* further down from the #VC interrupt frame which has been pushed on
* it during the first #VC exception entry.
*
* If not, simply the next slot on the #VC IST stack is set to point...

and here I'm not even sure why we're doing it?

The else branch, when we're not on the #VC stack, why are we doing

new_ist = old_ist - sizeof(old_ist);

?

I mean, if the NMI handler causes a #VC exception, it will simply run on
the #VC IST stack so why do we have to do that - 8 thing at all?

Thx.

--
Regards/Gruss,
Boris.

https://people.kernel.org/tglx/notes-about-netiquette