performance measurements: add extended performance data get ipc

Implement extended global performance data get ipc which extracts
performance data from MW3

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

Author: tobonex

src/audio/base_fw.c

@@ -428,6 +428,24 @@ int set_perf_meas_state(const char *data)
 	return IPC4_SUCCESS;
 }
 
+static int extended_global_perf_data_get(uint32_t *data_off_size, char *data)
+{
+#ifdef CONFIG_SOF_TELEMETRY_PERFORMANCE_MEASUREMENTS
+	int ret;
+	struct extended_global_perf_data *perf_data = (struct extended_global_perf_data *)data;
+
+	ret = get_extended_performance_data(perf_data);
+	if (ret < 0)
+		return IPC4_ERROR_INVALID_PARAM;
+	*data_off_size = sizeof(*perf_data)
+			+ perf_data->perf_item_count * sizeof(*perf_data->perf_items);
+
+	return IPC4_SUCCESS;
+#else
+	return IPC4_UNAVAILABLE;
+#endif
+}
+
 static int global_perf_data_get(uint32_t *data_off_size, char *data)
 {
 #ifdef CONFIG_SOF_TELEMETRY_PERFORMANCE_MEASUREMENTS
@@ -485,6 +503,8 @@ static int basefw_get_large_config(struct comp_dev *dev,
 		return basefw_modules_info_get(data_offset, data);
 	case IPC4_LIBRARIES_INFO_GET:
 		return basefw_libraries_info_get(data_offset, data);
+	case IPC4_EXTENDED_GLOBAL_PERF_DATA:
+		return extended_global_perf_data_get(data_offset, data);
 	case IPC4_GLOBAL_PERF_DATA:
 		return global_perf_data_get(data_offset, data);
 	/* TODO: add more support */

src/debug/telemetry/telemetry.c

@@ -238,6 +238,73 @@ void perf_meas_set_state(enum ipc4_perf_measurements_state_set state)
 	perf_measurements_state = state;
 }
 
+int get_extended_performance_data(struct extended_global_perf_data * const ext_global_perf_data)
+{
+	if (!ext_global_perf_data) {
+		tr_err(&ipc_tr, "IPC data is NULL");
+		return -EINVAL;
+	}
+
+	size_t slots_count;
+	size_t slot_idx = 0;
+	uint64_t total_dsp_cycles[CONFIG_MAX_CORE_COUNT];
+
+	/* TODO
+	 * Setting temporary values here.
+	 * Replace this with actual total dsp cycles info once it is available.
+	 */
+	for (size_t core_id = 0; core_id < CONFIG_MAX_CORE_COUNT; ++core_id)
+		total_dsp_cycles[core_id] = 1;
+
+	/* Fill one performance record per core with total dsp cycles */
+	for (size_t core_id = 0; core_id < CONFIG_MAX_CORE_COUNT; ++core_id) {
+		if (!(cpu_enabled_cores() & BIT(core_id)))
+			continue;
+
+		memset(&ext_global_perf_data->perf_items[slot_idx], 0,
+		       sizeof(struct ext_perf_data_item));
+
+		ext_global_perf_data->perf_items[slot_idx].resource_id = core_id;
+		ext_global_perf_data->perf_items[slot_idx].module_total_dsp_cycles_consumed =
+			total_dsp_cycles[core_id];
+		++slot_idx;
+	}
+
+	slots_count = perf_bitmap_get_occupied(&performance_data_bitmap) + slot_idx;
+	ext_global_perf_data->perf_item_count = slots_count;
+
+	/* fill the rest of the IPC records with data from
+	 * components registered in MW3 for performance measurement
+	 */
+	for (size_t idx = 0; idx < perf_bitmap_get_size(&performance_data_bitmap) &&
+	     slot_idx < slots_count; ++idx) {
+		if (perf_bitmap_is_bit_clear(&performance_data_bitmap, idx))
+			continue;
+
+		ext_global_perf_data->perf_items[slot_idx].resource_id =
+			perf_data[idx].item.resource_id;
+		ext_global_perf_data->perf_items[slot_idx].power_mode =
+			perf_data[idx].item.power_mode;
+		ext_global_perf_data->perf_items[slot_idx].is_removed =
+			perf_data[idx].item.is_removed;
+		ext_global_perf_data->perf_items[slot_idx].module_total_dsp_iterations =
+			perf_data[idx].total_iteration_count;
+		ext_global_perf_data->perf_items[slot_idx].module_total_dsp_cycles_consumed =
+			perf_data[idx].total_cycles_consumed;
+		ext_global_perf_data->perf_items[slot_idx].module_peak_dsp_cycles =
+			perf_data[idx].item.peak_kcps * 1000;
+		ext_global_perf_data->perf_items[slot_idx].module_peak_restricted_cycles =
+			perf_data[idx].restricted_peak_cycles;
+		ext_global_perf_data->perf_items[slot_idx].module_total_restricted_cycles_consumed =
+			perf_data[idx].restricted_total_cycles;
+		ext_global_perf_data->perf_items[slot_idx].module_total_restricted_iterations =
+			perf_data[idx].restricted_total_iterations;
+		ext_global_perf_data->perf_items[slot_idx].rsvd = 0;
+		++slot_idx;
+	}
+	return 0;
+}
+
 #ifdef CONFIG_SOF_TELEMETRY_PERFORMANCE_MEASUREMENTS
 static void telemetry_perf_queue_append(struct telemetry_perf_queue *q, size_t element)
 {

src/include/ipc4/base_fw.h

@@ -745,6 +745,39 @@ struct perf_data_item {
 	uint32_t avg_kcps;
 } __packed __aligned(4);
 
+struct ext_perf_data_item {
+	/* ID of the FW component */
+	uint32_t resource_id;
+	/* 0 - D0, 1 - D0i3. */
+	uint32_t power_mode : 1;
+	uint32_t rsvd : 30;
+	/* the component still exists (0) or has been already deleted (1) */
+	uint32_t is_removed : 1;
+	/*  peak number of DSP cycles used by a module */
+	uint32_t module_peak_dsp_cycles;
+	/* total number of DSP used since by a module start of measurements, */
+	uint64_t module_total_dsp_cycles_consumed;
+	/*  how many times a module was executed */
+	uint64_t module_total_dsp_iterations;
+	/*  peak number of <restricted> cycles used by a module.
+	 *  It is measured against DSP wall clock
+	 */
+	uint32_t module_peak_restricted_cycles;
+	/* total number of <restricted> cycles used by a module since start of
+	 * measurements. It is measured against DSP wall clock
+	 */
+	uint64_t module_total_restricted_cycles_consumed;
+	/*  how many times a module invoke <restricted> */
+	uint64_t module_total_restricted_iterations;
+} __packed __aligned(4);
+
+struct extended_global_perf_data {
+	/*  Specifies number of items in perf_items array. */
+	uint32_t perf_item_count;
+	/*  Array of extended global performance measurements. */
+	struct ext_perf_data_item perf_items[0];
+} __packed __aligned(4);
+
 struct perf_data_item_comp {
 	struct perf_data_item item;
 	/* Total iteration count of module instance */

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

@@ -137,4 +137,12 @@ enum ipc4_perf_measurements_state_set perf_meas_get_state(void);
  */
 int get_performance_data(struct global_perf_data * const global_perf_data);
 
+/**
+ * Get extended global performance data entries.
+ *
+ * @param[out] ext_global_perf_data Struct to be filled with data
+ * @return 0 if succeeded, error code otherwise.
+ */
+int get_extended_performance_data(struct extended_global_perf_data * const ext_global_perf_data);
+
 #endif /*__SOF_TELEMETRY_H__ */