On Fri, 2024-09-27 at 12:43 +0000, Miguel Luis wrote:
The common factor being the bit offset in the bitmap for SYSTEM_OFF2 discovery and argument to call SYSTEM_OFF2 as well. Would it be clearer something like:
#define PSCI_1_3_HIBERNATE_TYPE_OFF BIT(0)
I've updated the tree at https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/psci-hib... to do it that way.
As I did so, I realised that KVM *does* care about the argument to SYSTEM_OFF2. This is a straight copy of the SYSTEM_RESET2 handling. Although it doesn't pass the argument up to userspace (presumably userspace is expected to look at the registers if it cares), it *does* check it's within the range of permitted values and return PSCI_RET_INVALID_PARAMS if not.
I've changed that check to:
arg = smccc_get_arg1(vcpu); /* * Permit zero to mean HIBERNATE_OFF as well as the bitmap * form which was introduced in PSCI v1.3 beta. */ if (arg && arg != PSCI_1_3_HIBERNATE_TYPE_OFF) { val = PSCI_RET_INVALID_PARAMS; break; } kvm_psci_system_off2(vcpu);
On the guest side, I've changed the invocation to:
static int psci_sys_hibernate(struct sys_off_data *data) { /* * Zero is an acceptable alternative to PSCI_1_3_HIBERNATE_TYPE_OFF * and is supported by hypervisors implementing an earlier version * of the pSCI v1.3 spec. */ if (system_entering_hibernation()) invoke_psci_fn(PSCI_FN_NATIVE(1_3, SYSTEM_OFF2), 0 /*PSCI_1_3_HIBERNATE_TYPE_OFF*/, 0, 0); return NOTIFY_DONE; }
I'm going to do some careful interop testing with existing and new hypervisors before reposting this version.