Merge branch 'feat/improve_raw_tp' into 'feature/esp_as_mcu_host'

feat/improve_raw_tp Improve SDIO raw throughput by doing only block transfers

See merge request app-frameworks/esp_hosted!485

Author: mantriyogesh

Kconfig

@@ -359,7 +359,7 @@ ESP32XX_SPI_CLK_FREQ_RANGE_MAX := 40
 
 			choice ESP_SDIO_RX_OPTIMIZATION
 				bool "SDIO Receive Optimization"
-				default ESP_SDIO_OPTIMIZATION_RX_NONE
+				default ESP_SDIO_OPTIMIZATION_RX_STREAMING_MODE
 
 				config ESP_SDIO_OPTIMIZATION_RX_NONE
 					bool "No optimization"

host/drivers/transport/sdio/sdio_drv.c

@@ -363,8 +363,20 @@ static void sdio_write_task(void const* pvParameters)
 		do {
 			len_to_send = data_left;
 
+#if H_SDIO_TX_BLOCK_ONLY_XFER
+			/* Extend the transfer length to do block only transfers.
+			 * This is safe as slave only reads up to data_left, which
+			 * is not changed here. Rest of data is discarded by
+			 * slave.
+			 */
+			uint32_t block_send_len = ((len_to_send + ESP_BLOCK_SIZE - 1) / ESP_BLOCK_SIZE) * ESP_BLOCK_SIZE;
+
+			ret = g_h.funcs->_h_sdio_write_block(ESP_SLAVE_CMD53_END_ADDR - data_left,
+				pos, block_send_len, ACQUIRE_LOCK);
+#else
 			ret = g_h.funcs->_h_sdio_write_block(ESP_SLAVE_CMD53_END_ADDR - data_left,
 				pos, len_to_send, ACQUIRE_LOCK);
+#endif
 			if (ret) {
 				ESP_LOGE(TAG, "%s: %d: Failed to send data: %d %ld %ld", __func__,
 					retries, ret, len_to_send, data_left);
@@ -535,6 +547,11 @@ static esp_err_t sdio_push_data_to_queue(uint8_t * buf, uint32_t buf_len)
 // return a buffer big enough to contain the data
 static uint8_t * sdio_rx_get_buffer(uint32_t len)
 {
+#if H_SDIO_RX_BLOCK_ONLY_XFER
+	// we need to allocate enough memory to hold the padded data
+	len = ((len + ESP_BLOCK_SIZE - 1) / ESP_BLOCK_SIZE) * ESP_BLOCK_SIZE;
+#endif
+
 	// (re)allocate a buffer big enough to contain the data stream
 	if (len > recv_buf_size) {
 		if (recv_buf) {
@@ -703,9 +720,20 @@ static void sdio_read_task(void const* pvParameters)
 		do {
 			len_to_read = data_left;
 
+#if H_SDIO_RX_BLOCK_ONLY_XFER
+			/* Extend the transfer length to do block only transfers.
+			 * This is safe as slave will pad data with 0, which we
+			 * will ignore.
+			 */
+			uint32_t block_read_len = ((len_to_read + ESP_BLOCK_SIZE - 1) / ESP_BLOCK_SIZE) * ESP_BLOCK_SIZE;
+			ret = g_h.funcs->_h_sdio_read_block(
+					ESP_SLAVE_CMD53_END_ADDR - data_left,
+					pos, block_read_len, ACQUIRE_LOCK);
+#else
 			ret = g_h.funcs->_h_sdio_read_block(
 					ESP_SLAVE_CMD53_END_ADDR - data_left,
 					pos, len_to_read, ACQUIRE_LOCK);
+#endif
 			if (ret) {
 				ESP_LOGE(TAG, "%s: Failed to read data - %d %ld %ld",
 					__func__, ret, len_to_read, data_left);

host/port/esp_hosted_config.h

@@ -142,6 +142,18 @@ so feel free to change these if needed.
 #define H_SDIO_TX_LEN_TO_TRANSFER(x) ((x + 3) & (~3))
 #define H_SDIO_RX_LEN_TO_TRANSFER(x) ((x + 3) & (~3))
 
+/* Do Block Mode only transfers
+ *
+ * When enabled, SDIO only uses block mode transfers for higher
+ * throughput. Data lengths are padded to multiples of ESP_BLOCK_SIZE.
+ *
+ * This is safe for the SDIO slave:
+ * - for Host Tx: slave will ignore extra data sent by Host
+ * - for Host Rx: slave will send extra 0 data, ignored by Host
+ */
+#define H_SDIO_TX_BLOCK_ONLY_XFER (1)
+#define H_SDIO_RX_BLOCK_ONLY_XFER (1)
+
 // workarounds for some SDIO transfer errors that may occur
 #if 0
 /* Below workarounds could be enabled for non-ESP MCUs to test first

slave/main/Kconfig.projbuild

@@ -240,7 +240,7 @@ menu "Example Configuration"
 
 		config ESP_SDIO_STREAMING_MODE
 			bool "Enable SDIO Streaming Mode"
-			default n
+			default y
 			help
 				Enable Streaming Mode. Host to receive queued data from slave
 				as one stream instead of individual packets. This can improve
@@ -545,7 +545,7 @@ menu "Example Configuration"
 		config ESP_RAW_TP_REPORT_INTERVAL
 			depends on ESP_RAW_THROUGHPUT_TRANSPORT
 			int "RawTP: periodic duration to report stats accumulated"
-			default 1
+			default 10
 
 			config ESP_PKT_STATS
 				bool "Transport level packet stats"

slave/main/app_main.c

@@ -926,7 +926,7 @@ void app_main()
 
 	while (1) {
 		MEM_DUMP("mem_dump");
-		sleep(2);
+		sleep(10);
 	}
 
 }

slave/main/stats.c

@@ -196,14 +196,14 @@ static void stats_timer_func(void* arg)
 	double actual_bandwidth_tx = 0;
 	int32_t div = 1024;
 
-	actual_bandwidth_tx = (test_raw_tp_tx_len*8);
-	actual_bandwidth_rx = (test_raw_tp_rx_len*8);
+	actual_bandwidth_tx = (test_raw_tp_tx_len*8)/TEST_RAW_TP__TIMEOUT;
+	actual_bandwidth_rx = (test_raw_tp_rx_len*8)/TEST_RAW_TP__TIMEOUT;
 #if ESP_IDF_VERSION >= ESP_IDF_VERSION_VAL(5, 0, 0)
-	ESP_LOGI(TAG,"%lu-%lu sec       Rx: %.2f Tx: %.2f kbps", cur, cur + 1, actual_bandwidth_rx/div, actual_bandwidth_tx/div);
+	ESP_LOGI(TAG,"%lu-%lu sec       Rx: %.2f Tx: %.2f kbps", cur, cur + TEST_RAW_TP__TIMEOUT, actual_bandwidth_rx/div, actual_bandwidth_tx/div);
 #else
-	ESP_LOGI(TAG,"%u-%u sec       Rx: %.2f Tx: %.2f kbps", cur, cur + 1, actual_bandwidth_rx/div, actual_bandwidth_tx/div);
+	ESP_LOGI(TAG,"%u-%u sec       Rx: %.2f Tx: %.2f kbps", cur, cur + TEST_RAW_TP__TIMEOUT, actual_bandwidth_rx/div, actual_bandwidth_tx/div);
 #endif
-	cur++;
+	cur += TEST_RAW_TP__TIMEOUT;
 	test_raw_tp_rx_len = test_raw_tp_tx_len = 0;
 #endif
 #if ESP_PKT_STATS
@@ -239,7 +239,6 @@ static void start_timer_to_display_stats(int periodic_time_sec)
 void process_test_capabilities(uint8_t capabilities)
 {
 	ESP_LOGD(TAG, "capabilites: %d", capabilities);
-	start_timer_to_display_stats(TEST_RAW_TP__TIMEOUT);
 	if ((capabilities & ESP_TEST_RAW_TP__ESP_TO_HOST) ||
 		(capabilities & ESP_TEST_RAW_TP__BIDIRECTIONAL)) {
 		assert(xTaskCreate(raw_tp_tx_task , "raw_tp_tx_task",
@@ -257,7 +256,7 @@ void create_debugging_tasks(void)
 				CONFIG_ESP_DEFAULT_TASK_PRIO, NULL) == pdTRUE);
 #endif
 
-#if ESP_PKT_STATS
+#if TEST_RAW_TP || ESP_PKT_STATS
 	start_timer_to_display_stats(ESP_PKT_STATS_REPORT_INTERVAL);
 #endif
 }