[Copybara Auto Merge] Merge branch pro into android13-gs-pixel-5.10-udc

edgetpu: usage_stats send metrics v2 requests with v1 fallback
Bug: 271372136
edgetpu: usage stats: sync additional metrics v2 changes
Bug: 271372136 (repeat)
GitOrigin-RevId: af7a954e7f7903ae5c1dcdf6f24f48d9c1ae517c
Change-Id: Ie8f815636eaaf818338a1573668725b37bc6de45

5 files changed, 65 insertions, 21 deletions

diff --git a/drivers/edgetpu/edgetpu-firmware.c b/drivers/edgetpu/edgetpu-firmware.c
index cf9009b..ad27ec9 100644
--- a/drivers/edgetpu/edgetpu-firmware.c
+++ b/drivers/edgetpu/edgetpu-firmware.c
@@ -478,20 +478,21 @@ int edgetpu_firmware_run_locked(struct edgetpu_firmware *et_fw,
 				enum edgetpu_firmware_flags flags)
 {
 	const struct edgetpu_firmware_chip_data *chip_fw = et_fw->p->chip_fw;
+	struct edgetpu_dev *etdev = et_fw->etdev;
 	struct edgetpu_firmware_desc new_fw_desc;
 	int ret;
 	bool is_bl1_run = (flags & FW_BL1);
 
 	edgetpu_firmware_set_loading(et_fw);
 	if (!is_bl1_run)
-		edgetpu_sw_wdt_stop(et_fw->etdev);
+		edgetpu_sw_wdt_stop(etdev);
 
 	memset(&new_fw_desc, 0, sizeof(new_fw_desc));
 	ret = edgetpu_firmware_load_locked(et_fw, &new_fw_desc, name, flags);
 	if (ret)
 		goto out_failed;
 
-	etdev_dbg(et_fw->etdev, "run fw %s flags=%#x", name, flags);
+	etdev_dbg(etdev, "run fw %s flags=%#x", name, flags);
 	if (chip_fw->prepare_run) {
 		/* Note this may recursively call us to run BL1 */
 		ret = chip_fw->prepare_run(et_fw, &new_fw_desc.buf);
@@ -516,13 +517,16 @@ int edgetpu_firmware_run_locked(struct edgetpu_firmware *et_fw,
 
 	/* Don't start wdt if loaded firmware is second stage bootloader. */
 	if (!ret && !is_bl1_run && et_fw->p->fw_info.fw_flavor != FW_FLAVOR_BL1)
-		edgetpu_sw_wdt_start(et_fw->etdev);
+		edgetpu_sw_wdt_start(etdev);
 
 	if (!ret && !is_bl1_run && chip_fw->launch_complete)
 		chip_fw->launch_complete(et_fw);
 	else if (ret && chip_fw->launch_failed)
 		chip_fw->launch_failed(et_fw, ret);
 	edgetpu_firmware_set_state(et_fw, ret);
+	/* If previous firmware was metrics v1-only reset that flag and probe this again. */
+	if (etdev->usage_stats)
+		etdev->usage_stats->use_metrics_v1 = false;
 	return ret;
 
 out_unload_new_fw:
diff --git a/drivers/edgetpu/edgetpu-kci.c b/drivers/edgetpu/edgetpu-kci.c
index 65c16ec..fd52c3c 100644
--- a/drivers/edgetpu/edgetpu-kci.c
+++ b/drivers/edgetpu/edgetpu-kci.c
@@ -934,7 +934,7 @@ int edgetpu_kci_update_usage_locked(struct edgetpu_dev *etdev)
 {
 #define EDGETPU_USAGE_BUFFER_SIZE	4096
 	struct edgetpu_command_element cmd = {
-		.code = KCI_CODE_GET_USAGE,
+		.code = KCI_CODE_GET_USAGE_V2,
 		.dma = {
 			.address = 0,
 			.size = 0,
@@ -954,13 +954,22 @@ int edgetpu_kci_update_usage_locked(struct edgetpu_dev *etdev)
 		return ret;
 	}
 
+	/* TODO(b/271372136): remove v1 when v1 firmware no longer in use. */
+retry_v1:
+	if (etdev->usage_stats && etdev->usage_stats->use_metrics_v1)
+		cmd.code = KCI_CODE_GET_USAGE_V1;
 	cmd.dma.address = mem.tpu_addr;
 	cmd.dma.size = EDGETPU_USAGE_BUFFER_SIZE;
 	memset(mem.vaddr, 0, sizeof(struct edgetpu_usage_header));
 	ret = edgetpu_kci_send_cmd_return_resp(etdev->kci, &cmd, &resp);
 
-	if (ret == KCI_ERROR_UNIMPLEMENTED || ret == KCI_ERROR_UNAVAILABLE)
+	if (ret == KCI_ERROR_UNIMPLEMENTED || ret == KCI_ERROR_UNAVAILABLE) {
+		if (etdev->usage_stats && !etdev->usage_stats->use_metrics_v1) {
+			etdev->usage_stats->use_metrics_v1 = true;
+			goto retry_v1;
+		}
 		etdev_dbg(etdev, "firmware does not report usage\n");
+	}
 	else if (ret == KCI_ERROR_OK)
 		edgetpu_usage_stats_process_buffer(etdev, mem.vaddr);
 	else if (ret != -ETIMEDOUT)
diff --git a/drivers/edgetpu/edgetpu-kci.h b/drivers/edgetpu/edgetpu-kci.h
index b32b097..4d2f4b0 100644
--- a/drivers/edgetpu/edgetpu-kci.h
+++ b/drivers/edgetpu/edgetpu-kci.h
@@ -112,11 +112,14 @@ enum edgetpu_kci_code {
 	KCI_CODE_OPEN_DEVICE = 9,
 	KCI_CODE_CLOSE_DEVICE = 10,
 	KCI_CODE_FIRMWARE_INFO = 11,
-	KCI_CODE_GET_USAGE = 12,
+	/* TODO(b/271372136): remove v1 when v1 firmware no longer in use. */
+	KCI_CODE_GET_USAGE_V1 = 12,
 	KCI_CODE_NOTIFY_THROTTLING = 13,
 	KCI_CODE_BLOCK_BUS_SPEED_CONTROL = 14,
 	/* 15..18 not implemented in this branch */
 	KCI_CODE_FIRMWARE_TRACING_LEVEL = 19,
+	/* 20 not implemented in this branch */
+	KCI_CODE_GET_USAGE_V2 = 21,
 
 	KCI_CODE_RKCI_ACK = 256,
 };
diff --git a/drivers/edgetpu/edgetpu-usage-stats.c b/drivers/edgetpu/edgetpu-usage-stats.c
index e7b224c..60751dd 100644
--- a/drivers/edgetpu/edgetpu-usage-stats.c
+++ b/drivers/edgetpu/edgetpu-usage-stats.c
@@ -241,23 +241,44 @@ out:
 
 void edgetpu_usage_stats_process_buffer(struct edgetpu_dev *etdev, void *buf)
 {
-	struct edgetpu_usage_header_v1 *header = buf;
-	struct edgetpu_usage_metric *metric =
-		(struct edgetpu_usage_metric *)(header + 1);
+	struct edgetpu_usage_stats *ustats = etdev->usage_stats;
+	struct edgetpu_usage_metric *metric;
+	uint metric_size;
+	uint num_metrics;
+	uint version;
 	int i;
 
-	etdev_dbg(etdev, "%s: n=%u sz=%u", __func__,
-		  header->num_metrics, header->metric_size);
-	if (header->metric_size < EDGETPU_USAGE_METRIC_SIZE_V1) {
+	if (!ustats)
+		return;
+
+	/* TODO(b/271372136): remove v1 when v1 firmware no longer in use. */
+	if (ustats->use_metrics_v1) {
+		struct edgetpu_usage_header_v1 *header = buf;
+
+		metric_size = header->metric_size;
+		num_metrics = header->num_metrics;
+		version = 1;
+		metric = (struct edgetpu_usage_metric *)(header + 1);
+	} else {
+		struct edgetpu_usage_header *header = buf;
+
+		metric_size = header->metric_size;
+		num_metrics = header->num_metrics;
+		version = header->version;
+		metric = (struct edgetpu_usage_metric *)((char *)header + header->header_bytes);
+	}
+
+	etdev_dbg(etdev, "%s: v=%u n=%u sz=%u", __func__, version, num_metrics, metric_size);
+	if (metric_size < EDGETPU_USAGE_METRIC_SIZE_V1) {
 		etdev_warn_once(etdev, "fw metric size %u less than minimum %u",
-				header->metric_size, EDGETPU_USAGE_METRIC_SIZE_V1);
+				metric_size, EDGETPU_USAGE_METRIC_SIZE_V1);
 		return;
 	}
 
-	if (header->metric_size > sizeof(struct edgetpu_usage_metric))
+	if (metric_size > sizeof(struct edgetpu_usage_metric))
 		etdev_dbg(etdev, "fw metrics are later version with unknown fields");
 
-	for (i = 0; i < header->num_metrics; i++) {
+	for (i = 0; i < num_metrics; i++) {
 		switch (metric->type) {
 		case EDGETPU_METRIC_TYPE_TPU_USAGE:
 			edgetpu_usage_add(etdev, &metric->tpu_usage);
@@ -287,7 +308,7 @@ void edgetpu_usage_stats_process_buffer(struct edgetpu_dev *etdev, void *buf)
 			break;
 		}
 
-		metric = (struct edgetpu_usage_metric *)((char *)metric + header->metric_size);
+		metric = (struct edgetpu_usage_metric *)((char *)metric + metric_size);
 	}
 }
 
@@ -915,6 +936,7 @@ static struct attribute *usage_stats_dev_attrs[] = {
 static const struct attribute_group usage_stats_attr_group = {
 	.attrs = usage_stats_dev_attrs,
 };
+
 void edgetpu_usage_stats_init(struct edgetpu_dev *etdev)
 {
 	struct edgetpu_usage_stats *ustats;
diff --git a/drivers/edgetpu/edgetpu-usage-stats.h b/drivers/edgetpu/edgetpu-usage-stats.h
index 9d93122..ee908e1 100644
--- a/drivers/edgetpu/edgetpu-usage-stats.h
+++ b/drivers/edgetpu/edgetpu-usage-stats.h
@@ -11,7 +11,7 @@
 #include <linux/mutex.h>
 
 /* The highest version of usage metrics handled by this driver. */
-#define EDGETPU_USAGE_METRIC_VERSION	1
+#define EDGETPU_USAGE_METRIC_VERSION	2
 
 /*
  * Size in bytes of usage metric v1.
@@ -66,9 +66,12 @@ struct tpu_usage {
 enum edgetpu_usage_component {
 	/* The device as a whole */
 	EDGETPU_USAGE_COMPONENT_DEVICE = 0,
-	/* Just the TPU core */
+	/* Just the TPU core (scalar core and tiles) */
 	EDGETPU_USAGE_COMPONENT_TPU = 1,
-	EDGETPU_USAGE_COMPONENT_COUNT = 2, /* number of components above */
+	/* Control core (ARM Cortex-R52 CPU) */
+	EDGETPU_USAGE_COMPONENT_CONTROLCORE = 2,
+
+	EDGETPU_USAGE_COMPONENT_COUNT = 3, /* number of components above */
 };
 
 /*
@@ -169,10 +172,11 @@ struct __packed edgetpu_usage_max_watermark {
 /* An enum to identify the tracked firmware threads. */
 /* Must be kept in sync with firmware enum class UsageTrackerThreadId. */
 enum edgetpu_usage_threadid {
-	/* Individual thread IDs are not tracked. */
+	/* Individual thread IDs do not have identifiers assigned. */
+	/* Thread ID 14, used for other IP, is not used for TPU */
 
 	/* Number of task identifiers. */
-	EDGETPU_FW_THREAD_COUNT = 12,
+	EDGETPU_FW_THREAD_COUNT = 14,
 };
 
 /* Statistics related to a single thread in firmware. */
@@ -216,6 +220,8 @@ struct edgetpu_usage_metric {
 #define UID_HASH_BITS 3
 
 struct edgetpu_usage_stats {
+	/* if true the current firmware only implements metrics V1 */
+	bool use_metrics_v1;
 	DECLARE_HASHTABLE(uid_hash_table, UID_HASH_BITS);
 	/* component utilization values reported by firmware */
 	int32_t component_utilization[EDGETPU_USAGE_COMPONENT_COUNT];