NRC7292 Analysis Blog

Code analysis and research documentation for NRC7292 HaLow driver

NRC7292 iperf TCP/UDP TX Flow Analysis

NRC7292 iperf TCP/UDP TX Flow Analysis

Overview

This post provides a comprehensive analysis of the NRC7292 HaLow driver’s TX (transmission) path when handling iperf TCP and UDP traffic. We’ll examine the complete data flow from application-level iperf commands down to the CSPI hardware interface, tracking each function call and transmission completion mechanism.

iperf Test Commands

# TCP transmission test
iperf3 -c 192.168.1.100 -t 30 -i 1

# UDP transmission test  
iperf3 -c 192.168.1.100 -u -b 10M -t 30 -i 1

TCP Data Transmission Flow

The following sequence diagram shows the complete TCP transmission flow through the NRC7292 driver:

TCP Transmission Flow Steps:

  1. iperf3 Application → write() system call
  2. Network Stack → TCP segmentation and IP header addition → dev_queue_xmit()
  3. mac80211 → ieee80211_tx() call to NRC driver
  4. NRC Driver → nrc_mac_tx() [nrc-trx.c:75] - TX entry point
  5. NRC Driver → setup_ba_session() [nrc-trx.c:229] - AMPDU BA session setup
  6. NRC Driver → TX handler chain execution
  7. NRC Driver → tx_h_put_qos_control() [nrc-trx.c:579] - Convert to QoS data
  8. Credit Check → If sufficient:
    • nrc_xmit_frame() [hif.c:711]
    • nrc_hif_enqueue() [hif.c:98]
    • nrc_hif_work() [hif.c:177]
    • nrc_hif_xmit() [nrc-hif-cspi.c:403]
  9. HIF Layer → C-SPI command transmission to firmware
  10. Firmware → IEEE 802.11ah frame transmission to hardware
  11. Hardware → TX completion interrupt back to firmware
  12. Firmware → WIM Credit Report back to HIF
  13. HIF Layer → nrc_kick_txq() [nrc-mac80211.c:587]
  14. NRC Driver → nrc_tx_tasklet() [nrc-mac80211.c:545] - Process pending packets

UDP Data Transmission Flow

UDP transmission follows a similar path but with optimizations for high-throughput scenarios:

UDP Transmission Flow Steps:

  1. iperf3 Application → sendto() system call
  2. Network Stack → UDP header and IP header addition → dev_queue_xmit()
  3. mac80211 → ieee80211_tx() call to NRC driver
  4. NRC Driver → nrc_mac_tx() [nrc-trx.c:75]
  5. NRC Driver → tx_h_put_qos_control() [nrc-trx.c:579] - Convert UDP to QoS UDP
  6. High-speed Loop → Continuous transmission while credit sufficient:
    • nrc_xmit_frame() [hif.c:711]
    • C-SPI burst transmission to firmware
    • Continuous wireless frame transmission
  7. Batch Completion → After multiple frames:
    • Hardware sends batch TX completion interrupt
    • Firmware sends WIM Credit Report (batch)
    • nrc_kick_txq() to process waiting packets

Key Differences: TCP vs UDP

Packet Size Analysis

TCP Characteristics:

  • Maximum Segment Size: 1460 bytes (MTU 1500 - IP 20 - TCP 20)
  • Total frame size: ~1498 bytes (+ 802.11 headers)
  • Credit consumption: 6 credits (DIV_ROUND_UP(1498, 256))

UDP Characteristics:

  • Payload size: 1472 bytes (MTU 1500 - IP 20 - UDP 8)
  • Total frame size: ~1510 bytes (+ 802.11 headers)
  • Credit consumption: 6 credits (DIV_ROUND_UP(1510, 256))

QoS Processing

Both TCP and UDP data frames undergo the same QoS conversion:

// tx_h_put_qos_control() function
if (ieee80211_is_data(fc) && !ieee80211_is_data_qos(fc)) {
    hdr->frame_control |= cpu_to_le16(IEEE80211_STYPE_QOS_DATA);
    *((u16 *)(skb->data + 24)) = 0;  // TID=0 (Best Effort)
}

AMPDU Block ACK Sessions

TCP Benefits:

  • Order guarantee requirements make AMPDU highly effective
  • Built-in retransmission complements Block ACK mechanism
  • High aggregation efficiency for continuous streams

UDP Benefits:

  • No ordering requirements allow for optimized processing
  • Higher throughput potential due to no retransmission overhead
  • Efficient for real-time applications

Transmission Completion Mechanism

The driver employs a multi-stage completion confirmation system:

1. Hardware Level (C-SPI)

// nrc_hif_xmit() function
ret = nrc_spi_write(hdev, skb->data, skb->len);
if (ret < 0) {
    nrc_hif_free_skb(nw, skb);  // Credit rollback
    return ret;
}

2. Firmware Level (WIM Credit Report)

// nrc_wim_update_tx_credit() [wim.c:695]
static int nrc_wim_update_tx_credit(struct nrc *nw, struct wim *wim)
{
    struct wim_credit_report *r = (void *)(wim + 1);
    
    // Update credits for each AC
    for (ac = 0; ac < (IEEE80211_NUM_ACS*3); ac++)
        atomic_set(&nw->tx_credit[ac], r->v.ac[ac]);
    
    nrc_kick_txq(nw);  // Resume pending packet processing
    return 0;
}

3. Completion Steps

  1. SPI Transmission Complete: Hardware queue arrival
  2. Firmware Processing Complete: Frame generation done
  3. Wireless Transmission Complete: Air interface transmission
  4. Credit Return: Buffer space release for new transmissions

Performance Optimization Elements

Credit-Based Flow Control

  • BE (Best Effort): 40 credits (largest allocation for data)
  • VI (Video): 8 credits
  • VO (Voice): 8 credits
  • BK (Background): 4 credits

Batch Processing

// nrc_hif_work() - Sequential queue processing
for (i = ARRAY_SIZE(hdev->queue)-1; i >= 0; i--) {
    for (;;) {
        skb = skb_dequeue(&hdev->queue[i]);
        if (!skb) break;
        ret = nrc_hif_xmit(hdev, skb);
    }
}

AMPDU Aggregation

Automatic Block ACK session setup enhances throughput:

// setup_ba_session()
ret = ieee80211_start_tx_ba_session(peer_sta, tid, 0);

Performance Monitoring

Real-time Statistics

# Monitor credit status
cat /proc/nrc/stats
# TX Credit: AC0=4, AC1=35, AC2=8, AC3=8
# TX Pending: AC0=0, AC1=5, AC2=0, AC3=0

Debug Information

// Per-frame debugging
nrc_dbg(NRC_DBG_TX, "TX: Data frame to %pM, len=%d, AC=%d", 
        hdr->addr1, skb->len, ac);

Conclusion

The NRC7292 HaLow driver’s TX path demonstrates sophisticated handling of both TCP and UDP traffic through iperf scenarios. Key architectural strengths include:

  1. Unified QoS Processing: Both protocols benefit from automatic QoS conversion
  2. Credit-Based Flow Control: Prevents buffer overflow while maintaining high throughput
  3. AMPDU Optimization: Automatic aggregation improves efficiency for both protocols
  4. Multi-Stage Completion: Reliable transmission confirmation through hardware and firmware layers
  5. Batch Processing: Optimized for high-performance scenarios like iperf testing

This design effectively supports IEEE 802.11ah HaLow’s low-power, long-range characteristics while delivering the performance needed for throughput measurement tools and real-world applications.

For detailed function-level analysis and complete source code references, see the full TX path documentation in our analysis repository.