* [PATCH net-next] r8169: improve rtl8169_start_xmit
@ 2020-11-14 20:49 Heiner Kallweit
2020-11-16 15:58 ` Jakub Kicinski
0 siblings, 1 reply; 4+ messages in thread
From: Heiner Kallweit @ 2020-11-14 20:49 UTC (permalink / raw)
To: Jakub Kicinski, David Miller, Realtek linux nic maintainers; +Cc: netdev
Improve the following in rtl8169_start_xmit:
- tp->cur_tx can be accessed in parallel by rtl_tx(), therefore
annotate the race by using WRITE_ONCE
- avoid checking stop_queue a second time by moving the doorbell check
- netif_stop_queue() uses atomic operation set_bit() that includes a
full memory barrier on some platforms, therefore use
smp_mb__after_atomic to avoid overhead
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
---
drivers/net/ethernet/realtek/r8169_main.c | 15 ++++++---------
1 file changed, 6 insertions(+), 9 deletions(-)
diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
index 8910e900e..940fc6590 100644
--- a/drivers/net/ethernet/realtek/r8169_main.c
+++ b/drivers/net/ethernet/realtek/r8169_main.c
@@ -4226,7 +4226,7 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
/* rtl_tx needs to see descriptor changes before updated tp->cur_tx */
smp_wmb();
- tp->cur_tx += frags + 1;
+ WRITE_ONCE(tp->cur_tx, tp->cur_tx + frags + 1);
stop_queue = !rtl_tx_slots_avail(tp, MAX_SKB_FRAGS);
if (unlikely(stop_queue)) {
@@ -4235,13 +4235,6 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
*/
smp_wmb();
netif_stop_queue(dev);
- door_bell = true;
- }
-
- if (door_bell)
- rtl8169_doorbell(tp);
-
- if (unlikely(stop_queue)) {
/* Sync with rtl_tx:
* - publish queue status and cur_tx ring index (write barrier)
* - refresh dirty_tx ring index (read barrier).
@@ -4249,11 +4242,15 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
* status and forget to wake up queue, a racing rtl_tx thread
* can't.
*/
- smp_mb();
+ smp_mb__after_atomic();
if (rtl_tx_slots_avail(tp, MAX_SKB_FRAGS))
netif_start_queue(dev);
+ door_bell = true;
}
+ if (door_bell)
+ rtl8169_doorbell(tp);
+
return NETDEV_TX_OK;
err_dma_1:
--
2.29.2
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH net-next] r8169: improve rtl8169_start_xmit
2020-11-14 20:49 [PATCH net-next] r8169: improve rtl8169_start_xmit Heiner Kallweit
@ 2020-11-16 15:58 ` Jakub Kicinski
0 siblings, 0 replies; 4+ messages in thread
From: Jakub Kicinski @ 2020-11-16 15:58 UTC (permalink / raw)
To: Heiner Kallweit; +Cc: David Miller, Realtek linux nic maintainers, netdev
On Sat, 14 Nov 2020 21:49:53 +0100 Heiner Kallweit wrote:
> Improve the following in rtl8169_start_xmit:
> - tp->cur_tx can be accessed in parallel by rtl_tx(), therefore
> annotate the race by using WRITE_ONCE
> - avoid checking stop_queue a second time by moving the doorbell check
> - netif_stop_queue() uses atomic operation set_bit() that includes a
> full memory barrier on some platforms, therefore use
> smp_mb__after_atomic to avoid overhead
>
> Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Applied, thanks.
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH net-next] r8169: improve rtl8169_start_xmit
2020-02-22 16:02 Heiner Kallweit
@ 2020-02-24 1:18 ` David Miller
0 siblings, 0 replies; 4+ messages in thread
From: David Miller @ 2020-02-24 1:18 UTC (permalink / raw)
To: hkallweit1; +Cc: nic_swsd, netdev
From: Heiner Kallweit <hkallweit1@gmail.com>
Date: Sat, 22 Feb 2020 17:02:51 +0100
> Only call rtl8169_xmit_frags() if the skb is actually fragmented.
> This avoid a small overhead for non-fragmented skb's, and it allows
> to simplify rtl8169_xmit_frags() a little.
>
> Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Applied, thanks Heiner.
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH net-next] r8169: improve rtl8169_start_xmit
@ 2020-02-22 16:02 Heiner Kallweit
2020-02-24 1:18 ` David Miller
0 siblings, 1 reply; 4+ messages in thread
From: Heiner Kallweit @ 2020-02-22 16:02 UTC (permalink / raw)
To: Realtek linux nic maintainers, David Miller; +Cc: netdev
Only call rtl8169_xmit_frags() if the skb is actually fragmented.
This avoid a small overhead for non-fragmented skb's, and it allows
to simplify rtl8169_xmit_frags() a little.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
---
drivers/net/ethernet/realtek/r8169_main.c | 23 ++++++++++-------------
1 file changed, 10 insertions(+), 13 deletions(-)
diff --git a/drivers/net/ethernet/realtek/r8169_main.c b/drivers/net/ethernet/realtek/r8169_main.c
index cc4b6fd60..f081007a2 100644
--- a/drivers/net/ethernet/realtek/r8169_main.c
+++ b/drivers/net/ethernet/realtek/r8169_main.c
@@ -4087,12 +4087,10 @@ static int rtl8169_xmit_frags(struct rtl8169_private *tp, struct sk_buff *skb,
tp->tx_skb[entry].len = len;
}
- if (cur_frag) {
- tp->tx_skb[entry].skb = skb;
- txd->opts1 |= cpu_to_le32(LastFrag);
- }
+ tp->tx_skb[entry].skb = skb;
+ txd->opts1 |= cpu_to_le32(LastFrag);
- return cur_frag;
+ return 0;
err_out:
rtl8169_tx_clear_range(tp, tp->cur_tx + 1, cur_frag);
@@ -4217,6 +4215,7 @@ static void rtl8169_doorbell(struct rtl8169_private *tp)
static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
struct net_device *dev)
{
+ unsigned int frags = skb_shinfo(skb)->nr_frags;
struct rtl8169_private *tp = netdev_priv(dev);
unsigned int entry = tp->cur_tx % NUM_TX_DESC;
struct TxDesc *txd = tp->TxDescArray + entry;
@@ -4225,9 +4224,8 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
u32 opts[2], len;
bool stop_queue;
bool door_bell;
- int frags;
- if (unlikely(!rtl_tx_slots_avail(tp, skb_shinfo(skb)->nr_frags))) {
+ if (unlikely(!rtl_tx_slots_avail(tp, frags))) {
netif_err(tp, drv, dev, "BUG! Tx Ring full when queue awake!\n");
goto err_stop_0;
}
@@ -4256,14 +4254,13 @@ static netdev_tx_t rtl8169_start_xmit(struct sk_buff *skb,
tp->tx_skb[entry].len = len;
txd->addr = cpu_to_le64(mapping);
- frags = rtl8169_xmit_frags(tp, skb, opts);
- if (frags < 0)
- goto err_dma_1;
- else if (frags)
- opts[0] |= FirstFrag;
- else {
+ if (!frags) {
opts[0] |= FirstFrag | LastFrag;
tp->tx_skb[entry].skb = skb;
+ } else {
+ if (rtl8169_xmit_frags(tp, skb, opts))
+ goto err_dma_1;
+ opts[0] |= FirstFrag;
}
txd->opts2 = cpu_to_le32(opts[1]);
--
2.25.1
^ permalink raw reply related [flat|nested] 4+ messages in thread
end of thread, other threads:[~2020-11-16 15:58 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-11-14 20:49 [PATCH net-next] r8169: improve rtl8169_start_xmit Heiner Kallweit
2020-11-16 15:58 ` Jakub Kicinski
-- strict thread matches above, loose matches on Subject: below --
2020-02-22 16:02 Heiner Kallweit
2020-02-24 1:18 ` David Miller
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.