On Fri, Jun 24, 2016 at 09:55:20AM +0530, Vikas Sajjan wrote:

On Thu, Jun 23, 2016 at 1:06 AM, Lina Iyer [off-list ref] wrote:

quoted

+static int __init psci_1_0_init(struct device_node *np)
+{
+       int ret;
+
+       ret = psci_0_2_init(np);
+       if (ret)
+               return ret;
+
+       /* Check if PSCI OSI mode is available */
+       ret = psci_features(psci_function_id[PSCI_FN_CPU_SUSPEND]);
+       if (ret & PSCI_1_0_OS_INITIATED) {
+               ret = psci_features(PSCI_1_0_FN_SET_SUSPEND_MODE);
+               if (!ret)
+                       psci_has_osi_pd = true;

IMHO, its better to have this done in psci_init_cpu_suspend() itself
for 2 reasons

a] psci_init_cpu_suspend() already calls
psci_features(psci_function_id[PSCI_FN_CPU_SUSPEND])
b] by moving this in psci_init_cpu_suspend() we make this support
available even for ACPI platforms, since psci_acpi_init() calls
psci_probe() and this calls  psci_init_cpu_suspend() for
PSCI_VERSION_MAJOR(ver) >= 1

For ACPI platforms it is necessary to go through a handshake to
determine whether LPI can use OSI (see 6.2.11.2 in the ACPI 6.1 spec).
So there will have to be some ACPI-specific code to determine whether
OSI should be used.

We will probably have to an ACPI-specific wrapper for
psci_init_cpu_suspend to cater for that. I don't think that
psci_init_cpu_suspend itself should be in charge of deciding whether to
enable OSI.

Thanks,
Mark.