Re: [PATCH 2/2] hw/i386: support loading OVMF using -bios too

["Michael S. Tsirkin" <mst@redhat.com>]

On Sun, Jan 16, 2022 at 10:05:58PM +0100, Philippe Mathieu-Daudé wrote:
> On 14/1/22 19:07, Michael S. Tsirkin wrote:
> > On Thu, Jan 13, 2022 at 04:55:11PM +0000, Daniel P. Berrangé wrote:
> > > Traditionally the OVMF firmware has been loaded using the pflash
> > > mechanism. This is because it is usually provided as a pair of
> > > files, one read-only containing the code and one writable to
> > > provided persistence of non-volatile firmware variables.
> > > 
> > > The AMD SEV build of EDK2, however, is provided as a single
> > > file that contains only the code. This is intended to be used
> > > read-only and explicitly does not provide any ability for
> > > persistance of non-volatile firmware variables. While it is
> > > possible to configure this with the pflash mechanism, by only
> > > providing one of the 2 pflash blobs, conceptually it is a
> > > little strange to use pflash if there won't be any persistent
> > > data.
> 
> It certainly would be simpler to have a ROM for the CODE part.
> IIUC using CFI pflash allows the firmware to poll the underlying
> device size.
> 
> > > A stateless OVMF build can be loaded with -bios, however, QEMU
> > > does not currently initialize SEV in that scenario. This patch
> > > introduces the call needed for SEV initialization of the
> > > firmware.
> > > 
> > > Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
> > > ---
> > >   hw/i386/x86.c | 5 +++++
> > >   1 file changed, 5 insertions(+)
> > > 
> > > diff --git a/hw/i386/x86.c b/hw/i386/x86.c
> > > index b84840a1bb..c79d84936f 100644
> > > --- a/hw/i386/x86.c
> > > +++ b/hw/i386/x86.c
> > > @@ -45,6 +45,7 @@
> > >   #include "target/i386/cpu.h"
> > >   #include "hw/i386/topology.h"
> > >   #include "hw/i386/fw_cfg.h"
> > > +#include "hw/i386/pc.h"
> > >   #include "hw/intc/i8259.h"
> > >   #include "hw/rtc/mc146818rtc.h"
> > >   #include "target/i386/sev.h"
> > 
> > This builds fine because there's a stub in pc_sysfw_ovmf-stubs.c
> > 
> > The unfortunate thing about this however is that it's too easy to pull
> > in a PC dependency, and people building with CONFIG_PC will not notice
> > until it breaks for others.
> > 
> > Is it time we split pc.h further and had pc_sysfw_ovmf.h ?
> 
> While "pc*" is specific to the PC machines, "x86*" is shared between
> PC and microvm. "pc.h" must not be included in "x86.c". The shared
> method introduced in the previous patch becomes
> x86_system_ovmf_initialize_sev(). The dual pflash mechanism is proper
> to OVMF, so having this method in "x86.h" seems correct.
> 
> Phil.

Well.  E.g. pc_system_parse_ovmf_flash is defined in hw/i386/pc_sysfw_ovmf.c
and declared in pc.h. If you want to move all pc_sysfw_ovmf.c
declarations to x86.h that might be fine, but please do not
split declarations between multiple headers, that's messy.
And really, when in doubt do a separate header is a good rule.

-- 
MST