[RFC net-next 1/5] r8169: recheck the TX descriptor after a vector-map interrupt
From: Vlad Banea
Date: Wed Nov 11 2026 - 17:16:08 EST
With the vector-map interrupt model the TX interrupt can arrive before
the descriptor write-back is visible. rtl_tx() then stops at a
descriptor that still looks owned, and the end of a burst stays
unreaped. Once the BQL limit drops below the stranded bytes, the queue
stops and the watchdog resets the chip.
The legacy handler reads the interrupt status register before the poll
runs; the vector-map handler only writes registers. Do as Realtek's
driver does: if the first pending descriptor still looks owned, read a
register and look again.
Fixes: 677f8f8c4c5f ("r8169: add support for new interrupt mapping")
Signed-off-by: Vlad Banea <montrealtango@xxxxxxxxx>
---
drivers/net/ethernet/realtek/r8169_main.c | 16 ++++++++++++++--
1 file changed, 14 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
index cd4ebc654dd0..6bd137358d1f 100644
--- a/drivers/net/ethernet/realtek/r8169_main.c
+++ b/drivers/net/ethernet/realtek/r8169_main.c
@@ -5228,6 +5228,7 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
int budget)
{
unsigned int dirty_tx, bytes_compl = 0, pkts_compl = 0;
+ bool recheck = tp->irq_nvecs > 1;
struct sk_buff *skb;
dirty_tx = tp->dirty_tx;
@@ -5237,8 +5238,19 @@ static void rtl_tx(struct net_device *dev, struct rtl8169_private *tp,
u32 status;
status = le32_to_cpu(READ_ONCE(tp->TxDescArray[entry].opts1));
- if (status & DescOwn)
- break;
+ if (status & DescOwn) {
+ /* The vector-map TX interrupt can arrive before the
+ * descriptor write-back is visible. Read a register and
+ * look again, as Realtek's driver does.
+ */
+ if (!recheck)
+ break;
+ recheck = false;
+ RTL_R8(tp, IMR_CLEAR_VEC_MAP_REG);
+ status = le32_to_cpu(READ_ONCE(tp->TxDescArray[entry].opts1));
+ if (status & DescOwn)
+ break;
+ }
skb = tp->tx_skb[entry].skb;
rtl8169_unmap_tx_skb(tp, entry);
--
2.53.0