On Wed, 2024-03-13 at 12:42 -0700, Oliver Upton wrote: > I do not believe using the SMCCC filter to take SYSTEM_OFF2 to userspace > would necessitate a _full_ userspace reimplementation. You're free to > leave the default handler in place for functions you don't care about. > Forwarding PSCI_VERSION, PSCI_FEATURES, and SYSTEM_OFF2 would be sufficient > to get this off the ground, and the VMM can still advertise the rest of > the hypercalls implemented by KVM. Right... so we'd intercept PSCI_FEATURES *just* to indicate support for the one call we implement in userspace, and pass all other PSCI_FEATURES calls through to the kernel to handle the others? And then we'd implement SYSTEM_OFF2, hooking it up to do whatever our existing KVM_EXIT_SYSTEM_EVENT handler *already* does for a standard power-off, just with the extra flag to show it's a hibernate. This concept does not fill me with joy. > That might get you where you want to go a bit faster, since it'd avoid > any concerns about implementing a draft ABI in the kernel. I'd be more concerned about supporting a draft ABI in a public cloud provider, TBH. Having it as just one more downstream kernel patch, posted upstream but not yet merged before the final publication of the spec, would be the least of my worries.
Attachment:
smime.p7s
Description: S/MIME cryptographic signature
