Re: [PATCH] x86: only modify setup_data if the boot protocol indicates safety

[Gerd Hoffmann <kraxel@redhat.com>]

On Sun, Sep 04, 2022 at 06:50:58PM +0200, Jason A. Donenfeld wrote:
> This reverts 3824e25db1 ("x86: disable rng seeding via setup_data"), and
> then makes the use of setup_data safe. It does so by checking the boot
> protocol version. If it's sufficient, then it means EFI boots won't
> crash. While we're at it, gate this on SEV too.

> @@ -463,6 +462,7 @@ static void pc_i440fx_7_0_machine_options(MachineClass *m)

> +    pcmc->legacy_no_rng_seed = true;

This needs go into the pc_i440fx_7_1_machine_options function, otherwise
legacy_no_rng_seed gets flipped from false to true for 7.1 machine types
which breaks compatibility.

> @@ -398,6 +397,7 @@ static void pc_q35_7_0_machine_options(MachineClass *m)

> +    pcmc->legacy_no_rng_seed = true;

Same here.

> --- a/hw/i386/x86.c
> +++ b/hw/i386/x86.c
> @@ -1088,8 +1088,15 @@ void x86_load_linux(X86MachineState *x86ms,
>          qemu_guest_getrandom_nofail(setup_data->data, RNG_SEED_LENGTH);
>      }
>  
> -    /* Offset 0x250 is a pointer to the first setup_data link. */
> -    stq_p(header + 0x250, first_setup_data);
> +    /*
> +     * Only modify the header if doing so won't crash EFI boot, which is the
> +     * case only for newer boot protocols, and don't do so either if SEV is
> +     * enabled.
> +     */
> +    if (protocol >= 0x210 && !sev_enabled()) {
> +        /* Offset 0x250 is a pointer to the first setup_data link. */
> +        stq_p(header + 0x250, first_setup_data);
> +    }

This should better go into a separate patch.

take care,
  Gerd