lib: psci: early suspend handler for platforms

This patch adds an early suspend handler, that executes with
SMP and data cache enabled. This handler allows platforms to
perform any early actions during the CPU suspend entry sequence.

This handler is optional and platforms can choose to implement it
depending on their needs. The `pwr_domain_suspend` handler still
exists and platforms can keep on using it without any side effects.

Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>

2 files changed, 24 insertions, 6 deletions

diff --git a/docs/platform-migration-guide.rst b/docs/platform-migration-guide.rst
index 5e8eeba1..638033e4 100644
--- a/docs/platform-migration-guide.rst
+++ b/docs/platform-migration-guide.rst
@@ -146,6 +146,7 @@ for the ``plat_psci_ops`` structure which is declared as :
         void (*cpu_standby)(plat_local_state_t cpu_state);
         int (*pwr_domain_on)(u_register_t mpidr);
         void (*pwr_domain_off)(const psci_power_state_t *target_state);
+        void (*pwr_domain_suspend_early)(const psci_power_state_t *target_state);
         void (*pwr_domain_suspend)(const psci_power_state_t *target_state);
         void (*pwr_domain_on_finish)(const psci_power_state_t *target_state);
         void (*pwr_domain_suspend_finish)(
@@ -170,12 +171,12 @@ convert the power-state parameter (possibly encoding a composite power state)
 passed in a PSCI ``CPU_SUSPEND`` to the ``psci_power_state`` format. This handler
 is now mandatory for PSCI ``CPU_SUSPEND`` support.
 
-The ``plat_psci_ops`` handlers, ``pwr_domain_off`` and ``pwr_domain_suspend``, are
-passed the target local state for each affected power domain. The platform
-must execute operations specific to these target states. Similarly,
-``pwr_domain_on_finish`` and ``pwr_domain_suspend_finish`` are passed the local
-states of the affected power domains before wakeup. The platform
-must execute actions to restore these power domains from these specific
+The ``plat_psci_ops`` handlers, ``pwr_domain_off``, ``pwr_domain_suspend_early``
+and ``pwr_domain_suspend``, are passed the target local state for each affected
+power domain. The platform must execute operations specific to these target
+states. Similarly, ``pwr_domain_on_finish`` and ``pwr_domain_suspend_finish``
+are passed the local states of the affected power domains before wakeup. The
+platform must execute actions to restore these power domains from these specific
 local states.
 
 -  Difference in invocation
diff --git a/docs/porting-guide.rst b/docs/porting-guide.rst
index 4d4cdcd6..bf8dea7b 100644
--- a/docs/porting-guide.rst
+++ b/docs/porting-guide.rst
@@ -2060,6 +2060,23 @@ power down state where as it could be either power down, retention or run state
 for the higher power domain levels depending on the result of state
 coordination. The generic code expects the handler to succeed.
 
+plat\_psci\_ops.pwr\_domain\_suspend\_pwrdown\_early() [optional]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+This optional function may be used as a performance optimization to replace
+or complement pwr_domain_suspend() on some platforms. Its calling semantics
+are identical to pwr_domain_suspend(), except the PSCI implementation only
+calls this function when suspending to a power down state, and it guarantees
+that data caches are enabled.
+
+When HW_ASSISTED_COHERENCY = 0, the PSCI implementation disables data caches
+before calling pwr_domain_suspend(). If the target_state corresponds to a
+power down state and it is safe to perform some or all of the platform
+specific actions in that function with data caches enabled, it may be more
+efficient to move those actions to this function. When HW_ASSISTED_COHERENCY
+= 1, data caches remain enabled throughout, and so there is no advantage to
+moving platform specific actions to this function.
+
 plat\_psci\_ops.pwr\_domain\_suspend()
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~