performance measurements: extend perf meas state ipc

Implement actual functionality for perf meas state ipc handling. This
enables changing the state of global performance measurements.

Signed-off-by: Tobiasz Dryjanski <tobiaszx.dryjanski@intel.com>

Author: tobonex

src/audio/base_fw.c

@@ -413,13 +413,21 @@ int set_perf_meas_state(const char *data)
 
 	switch (state) {
 	case IPC4_PERF_MEASUREMENTS_DISABLED:
+		disable_performance_counters();
+		perf_meas_set_state(IPC4_PERF_MEASUREMENTS_DISABLED);
 		break;
 	case IPC4_PERF_MEASUREMENTS_STOPPED:
-		for (int i = 0; i < CONFIG_MAX_CORE_COUNT; i++)
-			systick_info[i].peak_utilization = 0;
+		enable_performance_counters();
+		reset_performance_counters();
+		perf_meas_set_state(IPC4_PERF_MEASUREMENTS_STOPPED);
 		break;
 	case IPC4_PERF_MEASUREMENTS_STARTED:
+		enable_performance_counters();
+		perf_meas_set_state(IPC4_PERF_MEASUREMENTS_STARTED);
+		break;
 	case IPC4_PERF_MEASUREMENTS_PAUSED:
+		enable_performance_counters();
+		perf_meas_set_state(IPC4_PERF_MEASUREMENTS_PAUSED);
 		break;
 	default:
 		return -EINVAL;

src/audio/component.c

@@ -542,14 +542,18 @@ bool update_peak_of_measured_cpc(struct comp_dev *dev, size_t measured_cpc)
 bool comp_update_performance_data(struct comp_dev *dev, uint32_t cycles_used)
 {
 	struct perf_data_item_comp *item = dev->perf_data.perf_data_item;
-	/* we divide by ibs so we need to check if its set */
-	if (item && dev->ibs != 0) {
-		item->total_iteration_count++;
-		item->total_cycles_consumed += cycles_used;
-		item->item.avg_kcps = item->total_cycles_consumed * dev->ll_chunk_size_
+
+	if (perf_meas_get_state() == IPC4_PERF_MEASUREMENTS_STARTED) {
+		/* we divide by ibs so we need to check if its set */
+		if (item && dev->ibs != 0) {
+			item->total_iteration_count++;
+			item->total_cycles_consumed += cycles_used;
+			item->item.avg_kcps = item->total_cycles_consumed * dev->ll_chunk_size_
 				/ (dev->ibs * item->total_iteration_count);
-		item->item.peak_kcps = MAX(item->item.peak_kcps, (cycles_used * dev->ll_chunk_size_)
+			item->item.peak_kcps =
+				MAX(item->item.peak_kcps, (cycles_used * dev->ll_chunk_size_)
 				/ dev->ibs);
+		}
 	}
 	return update_peak_of_measured_cpc(dev, cycles_used);
 }

src/debug/telemetry/telemetry.c

@@ -21,6 +21,9 @@
 #include <stddef.h>
 #include <stdint.h>
 
+#include <sof/lib_manager.h>
+#include <ipc4/base_fw_vendor.h>
+
 LOG_MODULE_DECLARE(ipc, CONFIG_SOF_LOG_LEVEL);
 
 #define PERFORMANCE_DATA_ENTRIES_COUNT (CONFIG_MEMORY_WIN_3_SIZE / sizeof(struct perf_data_item_comp))
@@ -37,6 +40,12 @@ struct perf_bitmap performance_data_bitmap;
 
 volatile struct perf_data_item_comp *perf_data_;
 
+/* Note that ref. FW used one state per core, all set together to the same state
+ * by one IPC but only for active cores. It may work slightly different in case
+ * where we enable a core while perf meas is started.
+ */
+enum ipc4_perf_measurements_state_set perf_measurements_state = IPC4_PERF_MEASUREMENTS_DISABLED;
+
 int perf_bitmap_init(struct perf_bitmap * const bitmap, sys_bitarray_t *array, size_t size)
 {
 	bitmap->array = array;
@@ -198,9 +207,15 @@ int free_performance_data(struct perf_data_item_comp *item)
 
 	if (item) {
 		item->item.is_removed = true;
-		ret = perf_data_free(item);
-		if (ret < 0)
-			return -EINVAL;
+		/* if we don't get the disabled state now, item will be
+		 * deleted on next disable perf meas message
+		 */
+		if (perf_measurements_state == IPC4_PERF_MEASUREMENTS_DISABLED) {
+			int ret = perf_data_free(item);
+
+			if (ret < 0)
+				return -EINVAL;
+		}
 	}
 	return IPC4_SUCCESS;
 }
@@ -266,7 +281,91 @@ int get_extended_performance_data(struct extended_global_perf_data * const ext_g
 	return IPC4_SUCCESS;
 }
 
-//==============================================================================================
+enum ipc4_perf_measurements_state_set perf_meas_get_state(void)
+{
+	return perf_measurements_state;
+}
+
+void perf_meas_set_state(enum ipc4_perf_measurements_state_set state)
+{
+	perf_measurements_state = state;
+}
+
+void disable_performance_counters(void)
+{
+	for (size_t idx = 0; idx < perf_bitmap_get_size(&performance_data_bitmap); ++idx) {
+		if (perf_bitmap_is_bit_clear(&performance_data_bitmap, idx))
+			continue;
+		if (perf_data_[idx].item.is_removed)
+			perf_data_free(&perf_data_[idx]);
+	}
+}
+
+int enable_performance_counters(void)
+{
+	//struct sof_man_fw_header header;
+	struct sof_man_module *man_module;
+	struct comp_dev *dev;
+	uint32_t comp_id;
+	const struct sof_man_fw_desc *desc;
+
+	if (perf_measurements_state != IPC4_PERF_MEASUREMENTS_DISABLED)
+		return -EINVAL;
+
+	for (int lib_id = 0; lib_id < LIB_MANAGER_MAX_LIBS; ++lib_id) {
+		if (lib_id == 0) {
+			desc = basefw_vendor_get_manifest();
+		} else {
+#if CONFIG_LIBRARY_MANAGER
+			desc = (struct sof_man_fw_desc *)lib_manager_get_library_manifest(lib_id);
+#else
+			desc = NULL;
+#endif
+		}
+		if (!desc)
+			continue;
+
+		for (int mod_id = 0 ; mod_id < desc->header.num_module_entries; mod_id++) {
+			man_module =
+				(struct sof_man_module *)(desc + SOF_MAN_MODULE_OFFSET(mod_id));
+
+			for (int inst_id = 0; inst_id < man_module->instance_max_count; inst_id++) {
+				comp_id = IPC4_COMP_ID(mod_id, inst_id);
+				dev = ipc4_get_comp_dev(comp_id);
+
+				if (dev)
+					comp_init_performance_data(dev);
+			}
+		}
+	}
+
+	/* TODO clear totaldspcycles here once implemented */
+	return IPC4_SUCCESS;
+}
+
+int reset_performance_counters(void)
+{
+	if (perf_measurements_state == IPC4_PERF_MEASUREMENTS_DISABLED)
+		return -EINVAL;
+
+	struct telemetry_wnd_data *wnd_data =
+			(struct telemetry_wnd_data *)ADSP_DW->slots[SOF_DW_TELEMETRY_SLOT];
+	struct system_tick_info *systick_info =
+			(struct system_tick_info *)wnd_data->system_tick_info;
+
+	for (size_t core_id = 0; core_id < CONFIG_MAX_CORE_COUNT; ++core_id) {
+		if (!(cpu_enabled_cores() & BIT(core_id)))
+			continue;
+		systick_info[core_id].peak_utilization = 0;
+	}
+	for (size_t idx = 0; idx < perf_bitmap_get_size(&performance_data_bitmap); ++idx) {
+		if (!perf_bitmap_is_bit_clear(&performance_data_bitmap, idx))
+			perf_data_item_comp_reset(&perf_data_[idx]);
+	}
+	/* TODO clear totaldspcycles here once implemented */
+
+	return IPC4_SUCCESS;
+}
 
 /* Systic variables, one set per core */
 static int telemetry_systick_counter[CONFIG_MAX_CORE_COUNT];

src/include/sof/debug/telemetry/telemetry.h

@@ -106,4 +106,11 @@ int get_performance_data(struct global_perf_data * const global_perf_data);
 
 int get_extended_performance_data(struct extended_global_perf_data * const ext_global_perf_data);
 
+int reset_performance_counters(void);
+int enable_performance_counters(void);
+void disable_performance_counters(void);
+
+void perf_meas_set_state(enum ipc4_perf_measurements_state_set state);
+enum ipc4_perf_measurements_state_set perf_meas_get_state(void);
+
 #endif /*__SOF_TELEMETRY_H__ */

src/ipc/ipc4/helper.c

@@ -163,7 +163,8 @@ struct comp_dev *comp_new_ipc4(struct ipc4_module_init_instance *module_init)
 	/* this can be null, just no performance measurements in this case */
 	if (dev->perf_data.perf_data_item) {
 		dev->perf_data.perf_data_item->item.resource_id = comp_id;
-		comp_init_performance_data(dev);
+		if (perf_meas_get_state() != IPC4_PERF_MEASUREMENTS_DISABLED)
+			comp_init_performance_data(dev);
 	}
 
 	ipc4_add_comp_dev(dev);