Hans-Jörg Höxer <hshoe...@genua.de> wrote:
> On Mon, Oct 28, 2024 at 09:00:25AM -0600, Theo de Raadt wrote:
> > Hans-Jörg Höxer <hshoe...@genua.de> wrote:
> >
> > > With GENERIC.MP the pattern below seems to work (but of course no tgz
> > > available yet on firmware.openbsd.org). psp(4) is not in RAMDISK.
> >
> > Since it is not in RAMDISK, you want to find a pattern so that it is
> > during RAMDISK. Otherwise, it will never be automatically installed.
>
> the SEV firmware is needed, when starting vmd(8) (when launching a
> SEV guest). Until then psp(4) is not used at all.
>
> I'd say the AMD SEV firmware is similar to the SeaBIOS which is
> needs to be installed before launching vmd(8). It is installed as
> /etc/firmware/vmm-bios which happens on first boot after successful
> installation. So I think, the same approach as for vmm(4) should be ok
> for psp(4), no?
yes, but also no. The earlier we get firmwares. I prefer if we add
any possible reason for a firmware fetch to the patterns. If there is
a cpu feature which indicates this, then we can added it to the patterns also.
> ---------------------------------------------------------------
> commit 37d69bb81e128fd9f085d142ceee075e0b74bf91
> Author: Hans-Joerg Hoexer <hshoe...@genua.de>
> Date: Mon Oct 28 13:07:50 2024 +0100
>
> fw_update: add pattern for AMD SEV firmware used with psp(4)
>
> diff --git a/usr.sbin/fw_update/patterns.c b/usr.sbin/fw_update/patterns.c
> index 868a2308b15..921513b9164 100644
> --- a/usr.sbin/fw_update/patterns.c
> +++ b/usr.sbin/fw_update/patterns.c
> @@ -94,6 +94,7 @@ main(void)
> print_devices("amdgpu", amdgpu_devices, nitems(amdgpu_devices));
> printf("%s\n", "amdgpu ^vga*vendor \"ATI\", unknown product");
> printf("%s\n", "amdgpu ^vendor \"ATI\", unknown product*class display");
> + printf("%s\n", "amdsev psp");
> printf("%s\n", "apple-boot ^cpu0*Apple");
> printf("%s\n", "arm64-qcom-dtb ^qcgpio0");
> printf("%s\n", "athn");
