realtek: RTL93xx: Make use of correct ring size counters

The receive path of the RTL93xx SoCs is currently discarding packets
in software. Analysis gives the following explanation:

- RX ring size registers are setup with the full software ring size
- When packets are received the packet counter registers are increased
- After RX processing the counter registers are changed the wrong way
- From then SOC is allowed to receive more packets than software allows
- Overflow interrupts are fired
- As a reaction to that the software drops packets

Change the processing as follows:

- Setup ring size registers with a headroom of 2 buffers
- Decrease the counter registers with the real work done

With this change no more overflow interrupts occur because the SoC
disables the queues before they can overflow or hit a buffer that is
still owned by the CPU.

Benchmark from single stream iperf3 run, with server process running
on ZyXEL XGS1210 (RTL930x).

iperf3 run before

-----------------------------------------------------------
Server listening on 5201 (test #1)
-----------------------------------------------------------
Accepted connection from 192.168.2.86, port 54412
[  5] local 192.168.2.71 port 5201 connected to 192.168.2.86 port 54418
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-1.00   sec   384 KBytes  3.14 Mbits/sec
[  5]   1.00-2.00   sec  0.00 Bytes  0.00 bits/sec
[  5]   2.00-3.00   sec  0.00 Bytes  0.00 bits/sec
[  5]   3.00-4.01   sec  5.12 MBytes  42.8 Mbits/sec
[  5]   4.01-5.00   sec  11.4 MBytes  95.8 Mbits/sec
[  5]   5.00-6.00   sec  0.00 Bytes  0.00 bits/sec
[  5]   6.00-7.00   sec  0.00 Bytes  0.00 bits/sec
[  5]   7.00-8.00   sec  0.00 Bytes  0.00 bits/sec
[  5]   8.00-9.00   sec  0.00 Bytes  0.00 bits/sec
[  5]   9.00-10.00  sec  0.00 Bytes  0.00 bits/sec

iperf3 run after

-----------------------------------------------------------
Server listening on 5201 (test #1)
-----------------------------------------------------------
Accepted connection from 192.168.2.86, port 55228
[  5] local 192.168.2.71 port 5201 connected to 192.168.2.86 port 55232
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-1.00   sec  22.8 MBytes   191 Mbits/sec
[  5]   1.00-2.01   sec  25.4 MBytes   211 Mbits/sec
[  5]   2.01-3.00   sec  25.4 MBytes   215 Mbits/sec
[  5]   3.00-4.01   sec  26.5 MBytes   220 Mbits/sec
[  5]   4.01-5.00   sec  26.2 MBytes   222 Mbits/sec
[  5]   5.00-6.00   sec  26.9 MBytes   225 Mbits/sec
[  5]   6.00-7.00   sec  27.0 MBytes   226 Mbits/sec
[  5]   7.00-8.01   sec  26.9 MBytes   224 Mbits/sec
[  5]   8.01-9.00   sec  26.5 MBytes   223 Mbits/sec
[  5]   9.00-10.00  sec  26.8 MBytes   225 Mbits/sec
[  5]  10.00-10.02  sec   640 KBytes   224 Mbits/sec

Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Link: https://github.com/openwrt/openwrt/pull/19960
Signed-off-by: Robert Marko <robimarko@gmail.com>

diff --git a/target/linux/realtek/files-6.12/drivers/net/ethernet/rtl838x_eth.c b/target/linux/realtek/files-6.12/drivers/net/ethernet/rtl838x_eth.c
index 44f0fe5357a5b34f77d4019b5752ab11ecef4b9b..8b30ebcd8585577334be0581aaed3a1c52d5013c 100644 (file)
--- a/target/linux/realtek/files-6.12/drivers/net/ethernet/rtl838x_eth.c
+++ b/target/linux/realtek/files-6.12/drivers/net/ethernet/rtl838x_eth.c
@@ -240,25 +240,18 @@ static void rtl839x_update_cntr(int r, int released)
 
 static void rtl930x_update_cntr(int r, int released)
 {
+       u32 reg = rtl930x_dma_if_rx_ring_cntr(r);
        int pos = (r % 3) * 10;
-       u32 reg = RTL930X_DMA_IF_RX_RING_CNTR + ((r / 3) << 2);
-       u32 v = sw_r32(reg);
 
-       v = (v >> pos) & 0x3ff;
-       pr_debug("RX: Work done %d, old value: %d, pos %d, reg %04x\n", released, v, pos, reg);
-       sw_w32_mask(0x3ff << pos, released << pos, reg);
-       sw_w32(v, reg);
+       sw_w32(released << pos, reg);
 }
 
 static void rtl931x_update_cntr(int r, int released)
 {
+       u32 reg = rtl931x_dma_if_rx_ring_cntr(r);
        int pos = (r % 3) * 10;
-       u32 reg = RTL931X_DMA_IF_RX_RING_CNTR + ((r / 3) << 2);
-       u32 v = sw_r32(reg);
 
-       v = (v >> pos) & 0x3ff;
-       sw_w32_mask(0x3ff << pos, released << pos, reg);
-       sw_w32(v, reg);
+       sw_w32(released << pos, reg);
 }
 
 struct dsa_tag {
@@ -766,10 +759,11 @@ static void rtl93xx_hw_en_rxtx(struct rtl838x_eth_priv *priv)
        sw_w32((DEFAULT_MTU << 16) | RX_TRUNCATE_EN_93XX, priv->r->dma_if_ctrl);
 
        for (int i = 0; i < priv->rxrings; i++) {
+               int cnt = min(priv->rxringlen - 2, 0x3ff);
                int pos = (i % 3) * 10;
                u32 v;
 
-               sw_w32_mask(0x3ff << pos, priv->rxringlen << pos, priv->r->dma_if_rx_ring_size(i));
+               sw_w32_mask(0x3ff << pos, cnt << pos, priv->r->dma_if_rx_ring_size(i));
 
                /* Some SoCs have issues with missing underflow protection */
                v = (sw_r32(priv->r->dma_if_rx_ring_cntr(i)) >> pos) & 0x3ff;
@@ -1329,7 +1323,7 @@ static int rtl838x_hw_receive(struct net_device *dev, int r, int budget)
        netif_receive_skb_list(&rx_list);
 
        /* Update counters */
-       priv->r->update_cntr(r, 0);
+       priv->r->update_cntr(r, work_done);
 
        spin_unlock_irqrestore(&priv->lock, flags);