net: ethernet: mtk_eth_soc: log DMA hangs discarded by the forced-reset gate

Commit 049c428 ("net: ethernet: mtk_eth_soc: support forced reset
control") gates the hang-triggered SER behind eth->reset.force, which
defaults to off.  mtk_hw_reset_monitor_work() still evaluates
mtk_hw_check_dma_hang() every second, but when the gate is closed a
positive detection is discarded without a trace: the DMA engine stays
in whatever state tripped three consecutive hang checks, and nothing in
the log records that it happened.

Upstream schedules the recovery unconditionally here, so the gate is a
deliberate semantic deviation; keep it, but emit a ratelimited error
when a detection is dropped so that degraded-DMA states can be
correlated with traffic symptoms from the log alone.

No functional change when no hang is detected or when forced reset is
enabled.

Signed-off-by: Mihai Ordean <research@mihaiordean.com>

Author: meehien

drivers/net/ethernet/mediatek/mtk_eth_soc.c

@@ -4648,8 +4648,13 @@ static void mtk_hw_reset_monitor_work(struct work_struct *work)
 		goto out;
 
 	/* DMA stuck checks */
-	if (mtk_hw_check_dma_hang(eth) && atomic_read(&eth->reset.force))
-		schedule_work(&eth->pending_work);
+	if (mtk_hw_check_dma_hang(eth)) {
+		if (atomic_read(&eth->reset.force))
+			schedule_work(&eth->pending_work);
+		else
+			dev_err_ratelimited(eth->dev,
+					    "DMA hang detected, automatic recovery is disabled (echo 2 > /sys/kernel/debug/mtketh/reset to enable)\n");
+	}
 
 out:
 	schedule_delayed_work(&eth->reset.monitor_work,