Re: [PATCH v6] hw/i386: pass RNG seed via setup_data entry

[Paolo Bonzini]

On 7/21/22 14:27, Daniel P. Berrangé wrote:
> Pre-existing guests which have merely upgraded their QEMU instance
> won't get the feature, because they'll be fixed on the old machine
> type to guarantee no guest ABI change. This isn't a problem, as
> such pre-existing guests likely won't have the new Linux code to
> consume the seed anyway.

In fact this isn't a given either and depends on how these pre-existing 
guests are managed.  People that do not use Libvirt, or that just start 
their guests with "virsh create" and some on-disk XML file, _will_ get 
the new feature.

The way QEMU does things is that fw_cfg is part of guest ABI, and this 
is _not_ up for discussion.  You're not the first one to be confused[1] 
and you probably will not be the last, so this means that the whole 
concept of guest ABI should be documented better.

(By the way, even though I agree that the failure is completely 
theoretical apart from the record-replay case, it can actually be 
reproduced easily by sticking a long-enough sleep in 
pc-bios/optionrom/linuxboot_dma.c).

Paolo