PSCI: Introduce cache and barrier wrappers

The PSCI implementation performs cache maintenance operations on its
data structures to ensure their visibility to both cache-coherent and
non-cache-coherent participants. These cache maintenance operations
can be skipped if all PSCI participants are cache-coherent. When
HW_ASSISTED_COHERENCY build option is enabled, we assume PSCI
participants are cache-coherent.

For usage abstraction, this patch introduces wrappers for PSCI cache
maintenance and barrier operations used for state coordination: they are
effectively NOPs when HW_ASSISTED_COHERENCY is enabled, but are
applied otherwise.

Also refactor local state usage and associated cache operations to make
it clearer.

Change-Id: I77f17a90cba41085b7188c1345fe5731c99fad87
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

lib/psci/psci_common.c

@@ -247,6 +247,50 @@ static plat_local_state_t *psci_get_req_local_pwr_states(unsigned int pwrlvl,
 	return &psci_req_local_pwr_states[pwrlvl - 1][cpu_idx];
 }
 
+/*
+ * psci_non_cpu_pd_nodes can be placed either in normal memory or coherent
+ * memory.
+ *
+ * With !USE_COHERENT_MEM, psci_non_cpu_pd_nodes is placed in normal memory,
+ * it's accessed by both cached and non-cached participants. To serve the common
+ * minimum, perform a cache flush before read and after write so that non-cached
+ * participants operate on latest data in main memory.
+ *
+ * When USE_COHERENT_MEM is used, psci_non_cpu_pd_nodes is placed in coherent
+ * memory. With HW_ASSISTED_COHERENCY, all PSCI participants are cache-coherent.
+ * In both cases, no cache operations are required.
+ */
+
+/*
+ * Retrieve local state of non-CPU power domain node from a non-cached CPU,
+ * after any required cache maintenance operation.
+ */
+static plat_local_state_t get_non_cpu_pd_node_local_state(
+		unsigned int parent_idx)
+{
+#if !USE_COHERENT_MEM || !HW_ASSISTED_COHERENCY
+	flush_dcache_range(
+			(uintptr_t) &psci_non_cpu_pd_nodes[parent_idx],
+			sizeof(psci_non_cpu_pd_nodes[parent_idx]));
+#endif
+	return psci_non_cpu_pd_nodes[parent_idx].local_state;
+}
+
+/*
+ * Update local state of non-CPU power domain node from a cached CPU; perform
+ * any required cache maintenance operation afterwards.
+ */
+static void set_non_cpu_pd_node_local_state(unsigned int parent_idx,
+		plat_local_state_t state)
+{
+	psci_non_cpu_pd_nodes[parent_idx].local_state = state;
+#if !USE_COHERENT_MEM || !HW_ASSISTED_COHERENCY
+	flush_dcache_range(
+			(uintptr_t) &psci_non_cpu_pd_nodes[parent_idx],
+			sizeof(psci_non_cpu_pd_nodes[parent_idx]));
+#endif
+}
+
 /******************************************************************************
  * Helper function to return the current local power state of each power domain
  * from the current cpu power domain to its ancestor at the 'end_pwrlvl'. This
@@ -264,18 +308,7 @@ void psci_get_target_local_pwr_states(unsigned int end_pwrlvl,
 
 	/* Copy the local power state from node to state_info */
 	for (lvl = PSCI_CPU_PWR_LVL + 1; lvl <= end_pwrlvl; lvl++) {
-#if !USE_COHERENT_MEM
+		pd_state[lvl] = get_non_cpu_pd_node_local_state(parent_idx);
-		/*
-		 * If using normal memory for psci_non_cpu_pd_nodes, we need
-		 * to flush before reading the local power state as another
-		 * cpu in the same power domain could have updated it and this
-		 * code runs before caches are enabled.
-		 */
-		flush_dcache_range(
-				(uintptr_t) &psci_non_cpu_pd_nodes[parent_idx],
-				sizeof(psci_non_cpu_pd_nodes[parent_idx]));
-#endif
-		pd_state[lvl] =	psci_non_cpu_pd_nodes[parent_idx].local_state;
 		parent_idx = psci_non_cpu_pd_nodes[parent_idx].parent_node;
 	}
 
@@ -299,21 +332,16 @@ static void psci_set_target_local_pwr_states(unsigned int end_pwrlvl,
 	psci_set_cpu_local_state(pd_state[PSCI_CPU_PWR_LVL]);
 
 	/*
-	 * Need to flush as local_state will be accessed with Data Cache
+	 * Need to flush as local_state might be accessed with Data Cache
 	 * disabled during power on
 	 */
-	flush_cpu_data(psci_svc_cpu_data.local_state);
+	psci_flush_cpu_data(psci_svc_cpu_data.local_state);
 
 	parent_idx = psci_cpu_pd_nodes[plat_my_core_pos()].parent_node;
 
 	/* Copy the local_state from state_info */
 	for (lvl = 1; lvl <= end_pwrlvl; lvl++) {
-		psci_non_cpu_pd_nodes[parent_idx].local_state =	pd_state[lvl];
+		set_non_cpu_pd_node_local_state(parent_idx, pd_state[lvl]);
-#if !USE_COHERENT_MEM
-		flush_dcache_range(
-				(uintptr_t)&psci_non_cpu_pd_nodes[parent_idx],
-				sizeof(psci_non_cpu_pd_nodes[parent_idx]));
-#endif
 		parent_idx = psci_non_cpu_pd_nodes[parent_idx].parent_node;
 	}
 }
@@ -347,13 +375,8 @@ void psci_set_pwr_domains_to_run(unsigned int end_pwrlvl)
 
 	/* Reset the local_state to RUN for the non cpu power domains. */
 	for (lvl = PSCI_CPU_PWR_LVL + 1; lvl <= end_pwrlvl; lvl++) {
-		psci_non_cpu_pd_nodes[parent_idx].local_state =
+		set_non_cpu_pd_node_local_state(parent_idx,
-				PSCI_LOCAL_STATE_RUN;
+				PSCI_LOCAL_STATE_RUN);
-#if !USE_COHERENT_MEM
-		flush_dcache_range(
-				(uintptr_t) &psci_non_cpu_pd_nodes[parent_idx],
-				sizeof(psci_non_cpu_pd_nodes[parent_idx]));
-#endif
 		psci_set_req_local_pwr_state(lvl,
 					     cpu_idx,
 					     PSCI_LOCAL_STATE_RUN);
@@ -364,7 +387,7 @@ void psci_set_pwr_domains_to_run(unsigned int end_pwrlvl)
 	psci_set_aff_info_state(AFF_STATE_ON);
 
 	psci_set_cpu_local_state(PSCI_LOCAL_STATE_RUN);
-	flush_cpu_data(psci_svc_cpu_data);
+	psci_flush_cpu_data(psci_svc_cpu_data);
 }
 
 /******************************************************************************

lib/psci/psci_off.c

@@ -154,17 +154,17 @@ exit:
 	 */
 	if (rc == PSCI_E_SUCCESS) {
 		/*
-		 * Set the affinity info state to OFF. This writes directly to
+		 * Set the affinity info state to OFF. When caches are disabled,
-		 * main memory as caches are disabled, so cache maintenance is
+		 * this writes directly to main memory, so cache maintenance is
 		 * required to ensure that later cached reads of aff_info_state
-		 * return AFF_STATE_OFF.  A dsbish() ensures ordering of the
+		 * return AFF_STATE_OFF. A dsbish() ensures ordering of the
 		 * update to the affinity info state prior to cache line
 		 * invalidation.
 		 */
-		flush_cpu_data(psci_svc_cpu_data.aff_info_state);
+		psci_flush_cpu_data(psci_svc_cpu_data.aff_info_state);
 		psci_set_aff_info_state(AFF_STATE_OFF);
-		dsbish();
+		psci_dsbish();
-		inv_cpu_data(psci_svc_cpu_data.aff_info_state);
+		psci_inv_cpu_data(psci_svc_cpu_data.aff_info_state);
 
 #if ENABLE_RUNTIME_INSTRUMENTATION
 

lib/psci/psci_private.h

@@ -38,6 +38,29 @@
 #include <psci.h>
 #include <spinlock.h>
 
+#if HW_ASSISTED_COHERENCY
+/*
+ * On systems with hardware-assisted coherency, make PSCI cache operations NOP,
+ * as PSCI participants are cache-coherent, and there's no need for explicit
+ * cache maintenance operations or barriers to coordinate their state.
+ */
+#define psci_flush_dcache_range(addr, size)
+#define psci_flush_cpu_data(member)
+#define psci_inv_cpu_data(member)
+
+#define psci_dsbish()
+#else
+/*
+ * If not all PSCI participants are cache-coherent, perform cache maintenance
+ * and issue barriers wherever required to coordinate state.
+ */
+#define psci_flush_dcache_range(addr, size)	flush_dcache_range(addr, size)
+#define psci_flush_cpu_data(member)		flush_cpu_data(member)
+#define psci_inv_cpu_data(member)		inv_cpu_data(member)
+
+#define psci_dsbish()				dsbish()
+#endif
+
 /*
  * The following helper macros abstract the interface to the Bakery
  * Lock API.

lib/psci/psci_setup.c

@@ -86,7 +86,7 @@ static void psci_init_pwr_domain_node(unsigned int node_idx,
 		/* Set the power state to OFF state */
 		svc_cpu_data->local_state = PLAT_MAX_OFF_STATE;
 
-		flush_dcache_range((uintptr_t)svc_cpu_data,
+		psci_flush_dcache_range((uintptr_t)svc_cpu_data,
 						 sizeof(*svc_cpu_data));
 
 		cm_set_context_by_index(node_idx,
@@ -242,9 +242,9 @@ int psci_setup(const psci_lib_args_t *lib_args)
 
 	/*
 	 * Flush `psci_plat_pm_ops` as it will be accessed by secondary CPUs
-	 * during warm boot before data cache is enabled.
+	 * during warm boot, possibly before data cache is enabled.
 	 */
-	flush_dcache_range((uintptr_t)&psci_plat_pm_ops,
+	psci_flush_dcache_range((uintptr_t)&psci_plat_pm_ops,
 					sizeof(psci_plat_pm_ops));
 
 	/* Initialize the psci capability */

lib/psci/psci_suspend.c

@@ -91,10 +91,10 @@ static void psci_suspend_to_pwrdown_start(unsigned int end_pwrlvl,
 	psci_set_suspend_pwrlvl(end_pwrlvl);
 
 	/*
-	 * Flush the target power level as it will be accessed on power up with
+	 * Flush the target power level as it might be accessed on power up with
 	 * Data cache disabled.
 	 */
-	flush_cpu_data(psci_svc_cpu_data.target_pwrlvl);
+	psci_flush_cpu_data(psci_svc_cpu_data.target_pwrlvl);
 
 	/*
 	 * Call the cpu suspend handler registered by the Secure Payload