On 24 October 2024 17:44:49 CEST, Oliver Upton <oliver.up...@linux.dev> wrote: >Hi, > >On Thu, Oct 24, 2024 at 03:48:26PM +0200, David Woodhouse wrote: >> On 24 October 2024 14:54:41 CEST, Miguel Luis <miguel.l...@oracle.com> wrote: >> >Perhaps spec. F.b. could be accommodated by first invoking SYSTEM_OFF2 with >> >PSCI_1_3_OFF_TYPE_HIBERNATE_OFF and checking its return value in case of a >> >fallback to an invocation with 0x0 ? > >This already complies with F.b. > >The PSCI implementation is required to accept either 0 or 1 for >HIBERNATE_OFF. Using 0 seems like a good choice for compatibility since ... > >> I wasn't aware there was any point. Are there any hypervisors which actually >> implemented it that way? Amazon Linux and Ubuntu guests already just use >> zero. >> >> We could add it later if such a hypervisor (now in violation of F.b) turns >> up, I suppose? > >IIUC, you're really wanting to 0x0 because there are hypervisors out >there that violate the final spec and *only* accept this value. > >That's perfectly fine, but it'd help avoid confusion if the supporting >comment was a bit more direct: > > /* > * If no hibernate type is specified SYSTEM_OFF2 defaults to > * selecting HIBERNATE_OFF. > * > * There are hypervisors in the wild that violate the spec and > * reject calls that explicitly provide a hibernate type. For > * compatibility with these nonstandard implementations, pass 0 > * as the type. > */ > if (system_entering_hibernation()) > invoke_psci_fn(PSCI_FN_NATIVE(1_3, SYSTEM_OFF2), 0 , 0, 0);
By the time this makes it into released versions of the guest Linux kernel, that comment won't be true any more.
