Re: [PATCH v5 1/4] x86: record SSP at non-guest entry points

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 09/11/2026 1:08 pm, Jan Beulich wrote:
> We will want to use that value for call trace generation, and likely
> also to eliminate the somewhat fragile shadow stack searching done in
> fixup_exception_return(). For those purposes, guest-only entry points do
> not need to record that value.
>
> To keep the saving code simple, record our own SSP that corresponds to
> an exception frame, pointing to the top of the shadow stack counterpart
> of what the CPU has saved on the regular stack. Consuming code can then
> work its way from there.
>
> In SAVE_ALL / RESTORE_ALL simply drop the use of UREGS_r15. We want the
> full size in all cases, so what exactly the top-of-stack field is going
> to be (whose UREGS_* is 0 anyway) doesn't matter this much there. This
> way we don't need to distinguish between XEN_SHSTK=y and XEN_SHSTK=n.
>
> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>

The code is now fine, but the commit message is quite stale.

You need to drop "non-guest" in at least two places, and change
SAVE_ALL/RESTORE_ALL.  In fact that whole final paragraph can go.

With the commit message suitably adjusted, Reviewed-by: Andrew Cooper
<andrew.cooper3@xxxxxxxxxx>

> ---
> v5: Drop ssp parameter of PUSH_AND_CLEAR_GPRS again.
> v4: Re-base.
> v3: Put new field at the front of struct cpu_user_regs.
> v2: Add comment ahead of SAVE_ALL. Add comma between its parameters.
>     Re-base.
>
> --- a/xen/arch/x86/include/asm/asm_defns.h
> +++ b/xen/arch/x86/include/asm/asm_defns.h
> @@ -259,12 +259,19 @@ static always_inline void stac(void)
>          xor   %r14d, %r14d
>          push  %r15
>          xor   %r15d, %r15d
> +#ifdef CONFIG_XEN_SHSTK
> +        rdsspq %rcx
> +        push  %rcx
> +#endif

It might be worth having /* Zero (from above) if CET-SS is not enabled
*/ on the push here to make it clear what value is stored.

~Andrew