[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <573a7ab0-ebfc-1dc2-6497-95c7411d6ddb@gmail.com>
Date: Sun, 10 Feb 2019 14:50:13 +0100
From: Heiner Kallweit <hkallweit1@...il.com>
To: Realtek linux nic maintainers <nic_swsd@...ltek.com>,
David Miller <davem@...emloft.net>
Cc: Sander Eikelenboom <linux@...elenboom.it>,
"netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: [PATCH net 2/2] Revert "r8169: make use of xmit_more and
__netdev_sent_queue"
This reverts commit 2e6eedb4813e34d8d84ac0eb3afb668966f3f356.
Sander reported a regression [1], therefore let's revert these commits.
Removal of the barriers doesn't seem to contribute to the issue, the
patch just overlaps with the problematic one and reverting
"r8169: make use of xmit_more and __netdev_sent_queue" only wasn't
tested.
[1] https://marc.info/?t=154965066400001&r=1&w=2
Signed-off-by: Heiner Kallweit <hkallweit1@...il.com>
---
drivers/net/ethernet/realtek/r8169.c | 19 ++++++++++---------
1 file changed, 10 insertions(+), 9 deletions(-)
diff --git a/drivers/net/ethernet/realtek/r8169.c b/drivers/net/ethernet/realtek/r8169.c
index bba806ce57d3..6e36b88ca7c9 100644
--- a/drivers/net/ethernet/realtek/r8169.c
+++ b/drivers/net/ethernet/realtek/r8169.c
@@ -6074,7 +6074,6 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
struct device *d = tp_to_dev(tp);
dma_addr_t mapping;
u32 opts[2], len;
- bool stop_queue;
int frags;
if (unlikely(!rtl_tx_slots_avail(tp, skb_shinfo(skb)->nr_frags))) {
@@ -6116,6 +6115,8 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
txd->opts2 = cpu_to_le32(opts[1]);
+ netdev_sent_queue(dev, skb->len);
+
skb_tx_timestamp(skb);
/* Force memory writes to complete before releasing descriptor */
@@ -6128,16 +6129,16 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
tp->cur_tx += frags + 1;
- stop_queue = !rtl_tx_slots_avail(tp, MAX_SKB_FRAGS);
- if (unlikely(stop_queue))
- netif_stop_queue(dev);
+ RTL_W8(tp, TxPoll, NPQ);
- if (__netdev_sent_queue(dev, skb->len, skb->xmit_more)) {
- RTL_W8(tp, TxPoll, NPQ);
- mmiowb();
- }
+ mmiowb();
- if (unlikely(stop_queue)) {
+ if (!rtl_tx_slots_avail(tp, MAX_SKB_FRAGS)) {
+ /* Avoid wrongly optimistic queue wake-up: rtl_tx thread must
+ * not miss a ring update when it notices a stopped queue.
+ */
+ smp_wmb();
+ netif_stop_queue(dev);
/* Sync with rtl_tx:
* - publish queue status and cur_tx ring index (write barrier)
* - refresh dirty_tx ring index (read barrier).
--
2.20.1
Powered by blists - more mailing lists