On Fri, 2024-03-22 at 16:02 +0000, Marc Zyngier wrote:
> On Tue, 19 Mar 2024 12:59:06 +0000,
> David Woodhouse <dwmw2@xxxxxxxxxxxxx> wrote:
>
> [...]
>
> > +static void __init psci_init_system_off2(void)
> > +{
> > + int ret;
> > +
> > + ret = psci_features(PSCI_FN_NATIVE(1_3, SYSTEM_OFF2));
> > +
> > + if (ret != PSCI_RET_NOT_SUPPORTED)
> > + psci_system_off2_supported = true;
>
> It'd be worth considering the (slightly broken) case where SYSTEM_OFF2
> is supported, but HIBERNATE_OFF is not set in the response, as the
> spec doesn't say that this bit is mandatory (it seems legal to
> implement SYSTEM_OFF2 without any hibernate type, making it similar to
> SYSTEM_OFF).
Such is not my understanding. If SYSTEM_OFF2 is supported, then
HIBERNATE_OFF *is* mandatory.
The next update to the spec is turning the PSCI_FEATURES response into
a *bitmap* of the available features, and I believe it will mandate
that bit zero is set.
And if for whatever reason that SYSTEM_OFF2/HIBERNATE_OFF call
*doesn't* work, Linux will end up doing a 'real' poweroff, first
through EFI and then finally as a last resort with a PSCI SYSTEM_OFF.
So it would be OK to have false positives in the detection.
Attachment:
smime.p7s
Description: S/MIME cryptographic signature
