shell: add perf_status command for SOF telemetry

Add a new SOF shell command exposing the existing telemetry /
performance_monitor counters from the live shell.

  sof perf_status              - print the ipc4_perf_measurements_state
                                 (disabled/stopped/started/paused) and
                                 per-active-core systick counters from
                                 the SOF telemetry slot (count,
                                 last/max time elapsed, avg_kcps,
                                 peak_kcps, peak4k, peak8k)
  sof perf_status reset        - reset_performance_counters()
  sof perf_status start        - enable_performance_counters() and
                                 transition state to STARTED
  sof perf_status stop|pause   - transition state without zeroing
                                 counters (useful for snapshots)

Per-core data is read via telemetry_get_systick_info_ptr() when the
debug slot manager is enabled, otherwise directly from
ADSP_DW->slots[SOF_DW_TELEMETRY_SLOT]. We iterate only cores enabled
in cpu_enabled_cores() so the output matches the active topology and
matches what get_performance_data() would put on the IPC.

New Kconfig CONFIG_SOF_SHELL_PERF_STATUS (default y, depends on
SOF_TELEMETRY).

The full per-component perf_data_item_comp array (potentially
hundreds of items in MW3) is intentionally not dumped here; that will
land separately as 'perf_components' once there's a stream-friendly
iterator. cpu_load is already available via Zephyr's 'kernel
cpu_load' so it is not duplicated.

Signed-off-by: Liam Girdwood <liam.r.girdwood@linux.intel.com>

Author: lrgirdwo

zephyr/Kconfig

@@ -544,6 +544,23 @@ config SOF_SHELL_DBGWIN_DUMP
 	  critical_log, broken-marker and other slots from the shell
 	  without host-side tooling.
 
+config SOF_SHELL_PERF_STATUS
+	bool "Performance / telemetry status command"
+	default y
+	depends on SHELL && SOF_TELEMETRY
+	help
+	  Enables 'sof perf_status [reset|start|stop|pause]' which prints
+	  the current ipc4_perf_measurements_state plus per-active-core
+	  systick counters from the SOF telemetry slot:
+	  count, last/max time elapsed (cycles), avg/peak KCPS and the
+	  4k/8k peak utilization buckets.
+
+	  With one of the optional sub-commands the command transitions
+	  the perf measurement state machine ('start' calls
+	  enable_performance_counters(), 'reset' calls
+	  reset_performance_counters(); 'stop'/'pause' just update the
+	  state) - useful for taking before/after snapshots without IPC.
+
 endmenu # SOF shell commands
 
 config SOF_VREGIONS

zephyr/shell.md

@@ -88,7 +88,7 @@ debug or testing. Items get ticked off as commands land on `topic/shell`.
 | Audio buffers | `buffer_list`, `buffer_info <id>` | **DONE (task 2)** |
 | **Scheduler** | `sched_tasks`, `sched_load`, `task_info <task>` | **DONE (task 3)** |
 | Logging / trace | `log_status`, `mtrace_dump` (snapshot) | **DONE (task 4)** &mdash; runtime per-source `log_level` deferred (needs `CONFIG_LOG_RUNTIME_FILTERING`, see notes) |
-| Telemetry / perf | `perf_status`, `perf_reset`, `cpu_load` | TODO |
+| **Telemetry / perf** | `perf_status`, `perf_status reset`, `perf_status start/stop/pause` | **DONE (task 6)** |
 | Notifications | `notify_subscribers`, `notify_stats` | TODO |
 | **Debug window / mailbox** | `dbgwin_dump <slot>`, `mailbox_hex` | **DONE (task 5)** |
 | Crash / panic | `crash_log`, `crash_clear`, `panic_info`, `bt`, `regs` | TODO |
@@ -114,7 +114,7 @@ debug or testing. Items get ticked off as commands land on `topic/shell`.
 3. **`sched_tasks` / `sched_load`** &mdash; DONE.
 4. **`log_status` / `mtrace_dump`** &mdash; DONE.
 5. **`mailbox_hex` / `dbgwin_dump`** &mdash; DONE (was originally `crash_log`/`bt`; pivoted because SOF panic.c isn't built on Zephyr and `bt` of a running CPU from itself isn't meaningful).
-6. `perf_status` &mdash; wraps SOF telemetry counters.
+6. **`perf_status`** &mdash; DONE.
 7. `dai_list` / `dma_chan_status` &mdash; link/DMA blind spot.
 8. `kctl_get/set` &mdash; today only doable via tplg/IPC.
 
@@ -309,3 +309,43 @@ debug or testing. Items get ticked off as commands land on `topic/shell`.
   ACE.
 - `dbgwin_dump` is read-only. We do not implement a write/seize
   command to avoid corrupting host-visible state.
+
+## Task 6 &mdash; `perf_status`
+
+### Commands
+
+| Command | Description |
+|---|---|
+| `sof perf_status` | Print the SOF telemetry performance state (`disabled`/`stopped`/`started`/`paused`) and per-active-core systick counters (`count`, `last_time_elapsed`, `max_time_elapsed`, `avg_kcps`, `peak_kcps`, plus 4k/8k peak utilization). |
+| `sof perf_status reset` | Call `reset_performance_counters()` to zero all counters. |
+| `sof perf_status start` | Call `enable_performance_counters()` and set state to `STARTED`. |
+| `sof perf_status stop` / `pause` | Transition state to `STOPPED` or `PAUSED` (stops sampling without zeroing counters). |
+
+### Implementation
+
+- Reads per-core systick info via
+  `telemetry_get_systick_info_ptr()` (with
+  `CONFIG_INTEL_ADSP_DEBUG_SLOT_MANAGER`) or directly from
+  `ADSP_DW->slots[SOF_DW_TELEMETRY_SLOT]` otherwise.
+- Iterates only cores in `cpu_enabled_cores()` so the output matches
+  the active topology.
+- Uses the existing `perf_meas_get_state()` /
+  `perf_meas_set_state()` /
+  `enable_performance_counters()` /
+  `reset_performance_counters()` API from
+  [src/include/sof/debug/telemetry/performance_monitor.h](src/include/sof/debug/telemetry/performance_monitor.h);
+  no new state added.
+- New Kconfig `CONFIG_SOF_SHELL_PERF_STATUS` (default `y`,
+  depends on `SOF_TELEMETRY`).
+- Shell command in [zephyr/sof_shell.c](zephyr/sof_shell.c).
+
+### Notes / follow-ups
+
+- We deliberately do not dump the full per-component
+  `perf_data_item_comp` array yet: it can grow large
+  (`CONFIG_MEMORY_WIN_3_SIZE` / item size, ~hundreds of items on PTL)
+  and would require a heap allocation. A future `perf_components` /
+  `perf_status -v` could iterate `performance_data_bitmap` and stream
+  one row per occupied slot.
+- Zephyr already provides `kernel cpu_load` and `kernel threads`;
+  `cpu_load` was therefore not duplicated here.

zephyr/sof_shell.c

@@ -2147,6 +2147,95 @@ __cold static int cmd_sof_dbgwin_dump(const struct shell *sh,
 
 #endif /* CONFIG_SOF_SHELL_DBGWIN_DUMP */
 
+#if CONFIG_SOF_SHELL_PERF_STATUS
+
+#include <sof/debug/telemetry/telemetry.h>
+#include <sof/debug/telemetry/performance_monitor.h>
+#include <ipc4/base_fw.h>
+
+static const char *perf_state_str(enum ipc4_perf_measurements_state_set s)
+{
+	switch (s) {
+	case IPC4_PERF_MEASUREMENTS_DISABLED:	return "disabled";
+	case IPC4_PERF_MEASUREMENTS_STOPPED:	return "stopped";
+	case IPC4_PERF_MEASUREMENTS_STARTED:	return "started";
+	case IPC4_PERF_MEASUREMENTS_PAUSED:	return "paused";
+	default:				return "?";
+	}
+}
+
+__cold static int cmd_sof_perf_status(const struct shell *sh,
+				      size_t argc, char *argv[])
+{
+	struct system_tick_info *systick;
+	int core_id, ret;
+
+	if (argc > 1) {
+		if (!strcmp(argv[1], "reset")) {
+			ret = reset_performance_counters();
+			shell_print(sh, "perf: reset_performance_counters() = %d", ret);
+			return ret;
+		}
+		if (!strcmp(argv[1], "start")) {
+			ret = enable_performance_counters();
+			if (!ret)
+				perf_meas_set_state(IPC4_PERF_MEASUREMENTS_STARTED);
+			shell_print(sh, "perf: enable_performance_counters() = %d", ret);
+			return ret;
+		}
+		if (!strcmp(argv[1], "stop")) {
+			perf_meas_set_state(IPC4_PERF_MEASUREMENTS_STOPPED);
+			shell_print(sh, "perf: stopped");
+			return 0;
+		}
+		if (!strcmp(argv[1], "pause")) {
+			perf_meas_set_state(IPC4_PERF_MEASUREMENTS_PAUSED);
+			shell_print(sh, "perf: paused");
+			return 0;
+		}
+		shell_print(sh, "Usage: sof perf_status [reset|start|stop|pause]");
+		return -EINVAL;
+	}
+
+	shell_print(sh, "Performance measurements: %s",
+		    perf_state_str(perf_meas_get_state()));
+
+#ifdef CONFIG_INTEL_ADSP_DEBUG_SLOT_MANAGER
+	systick = telemetry_get_systick_info_ptr();
+	if (!systick) {
+		shell_print(sh, "telemetry slot not allocated");
+		return 0;
+	}
+#else
+	{
+		struct telemetry_wnd_data *wnd =
+			(struct telemetry_wnd_data *)
+			ADSP_DW->slots[SOF_DW_TELEMETRY_SLOT];
+		systick = (struct system_tick_info *)wnd->system_tick_info;
+	}
+#endif
+
+	shell_print(sh, "Per-core systick (count, last_us_cyc, max_us_cyc, avg_kcps, peak_kcps):");
+	for (core_id = 0; core_id < CONFIG_MAX_CORE_COUNT; core_id++) {
+		if (!(cpu_enabled_cores() & BIT(core_id)))
+			continue;
+		shell_print(sh,
+			    "  core %u: count=%u last=%u max=%u avg_kcps=%u peak_kcps=%u peak4k=%u peak8k=%u",
+			    core_id,
+			    systick[core_id].count,
+			    systick[core_id].last_time_elapsed,
+			    systick[core_id].max_time_elapsed,
+			    systick[core_id].avg_utilization,
+			    systick[core_id].peak_utilization,
+			    systick[core_id].peak_utilization_4k,
+			    systick[core_id].peak_utilization_8k);
+	}
+
+	return 0;
+}
+
+#endif /* CONFIG_SOF_SHELL_PERF_STATUS */
+
 SHELL_STATIC_SUBCMD_SET_CREATE(sof_commands,
 	SHELL_CMD(test_inject_sched_gap, NULL,
 		  "Inject a gap to audio scheduling\n",
@@ -2347,6 +2436,13 @@ SHELL_STATIC_SUBCMD_SET_CREATE(sof_commands,
 		  cmd_sof_dbgwin_dump, 1, 2),
 #endif
 
+#if CONFIG_SOF_SHELL_PERF_STATUS
+	SHELL_CMD_ARG(perf_status, NULL,
+		  "Show telemetry perf state and per-core systick;"
+		  " optional [reset|start|stop|pause]\n",
+		  cmd_sof_perf_status, 1, 1),
+#endif
+
 	SHELL_SUBCMD_SET_END
 );