Skip to contentSkip to conversation

NAPI support for Sibyte MAC

PATCH In review mason 2 participants 3 messages Mar 23, 2007 – Mar 26, 2007

Subsystems (3)
Networking Network drivers MIPS
lore
Share & tools
lore

mason

author
#0 lore raw

PATCH 1/1 NAPI support for Sibyte MAC

[ This is a re-post, but the patch still applies and works fine against the linux-mips.org tip. We'd really like to get this in. -Mark]

This patch completes the NAPI functionality for SB1250 MAC, including making NAPI a kernel option that can be turned on or off and adds the "sbmac_poll" routine.

Signed off by: Mark Mason ([email protected]) Signed off by: Dan Krejsa ([email protected]) Signed off by: Steve Yang ([email protected])

drivers/net/Kconfig +19 -0
Index: linux-2.6.14-cgl/drivers/net/Kconfig
===================================================================
@@ -2031,6 +2031,23 @@
 	tristate "SB1250 Ethernet support"
 	depends on SIBYTE_SB1xxx_SOC
 
+config SBMAC_NAPI
+	bool "SBMAC: Use Rx Polling (NAPI) (EXPERIMENTAL)"
+	depends on NET_SB1250_MAC && EXPERIMENTAL
+	help
+	  NAPI is a new driver API designed to reduce CPU and interrupt load
+	  when the driver is receiving lots of packets from the card. It is
+	  still somewhat experimental and thus not yet enabled by default.
+
+	  If your estimated Rx load is 10kpps or more, or if the card will be
+	  deployed on potentially unfriendly networks (e.g. in a firewall),
+	  then say Y here.
+
+	  See <file:Documentation/networking/NAPI_HOWTO.txt> for more
+	  information.
+
+	  If in doubt, say y.
+
 config R8169_VLAN
 	bool "VLAN support"
 	depends on R8169 && VLAN_8021Q

@@ -2826,3 +2843,5 @@
 	def_bool NETPOLL
 
 endmenu
+
+
drivers/net/sb1250-mac.c +207 -65

532 lines not on this page · raw message

arch/mips/sibyte/bcm1480/irq.c +2 -2
Index: linux-2.6.14-cgl/arch/mips/sibyte/bcm1480/irq.c
===================================================================
@@ -144,11 +144,11 @@
 	unsigned long flags;
 	unsigned int irq_dirty;
 
-	i = first_cpu(mask);
-	if (next_cpu(i, mask) <= NR_CPUS) {
+	if (cpus_weight(mask) != 1) {
 		printk("attempted to set irq affinity for irq %d to multiple CPUs\n", irq);
 		return;
 	}
+	i = first_cpu(mask);
 
 	/* Convert logical CPU to physical CPU */
 	cpu = cpu_logical_map(i);
#1 lore raw
Quoted 8 lines · you

This patch completes the NAPI functionality for SB1250 MAC, including making NAPI a kernel option that can be turned on or off and adds the "sbmac_poll" routine.

Index: linux-2.6.14-cgl/drivers/net/Kconfig
===================================================================

drivers/net/Kconfig +0 -0

[...]

drivers/net/Kconfig line 2153

Quoted 13 lines
drivers/net/Kconfig +7 -1
@@ -2075,12 +2143,52 @@
 		 */
 
 		if (isr & (M_MAC_INT_CHANNEL << S_MAC_TX_CH0)) {
-			sbdma_tx_process(sc,&(sc->sbm_txdma));
+			sbdma_tx_process(sc,&(sc->sbm_txdma), 0);
+#ifdef CONFIG_NETPOLL_TRAP
+		       if (netpoll_trap()) {
+			       if (test_and_clear_bit(__LINK_STATE_XOFF, &dev->state)) 
+				       __netif_schedule(dev);
+		       }
+#endif
 		}

This just doesn't make sense. That option is enabled to *prevent* calls to
__netif_schedule() -- you can't override it that way. (Well, how it works currently, doesn't make much sense either since it totally breaks the TX queue control -- I was going to post a patch).

Quoted 28 lines
drivers/net/Kconfig +25 -1
+	if (isr & (M_MAC_INT_CHANNEL << S_MAC_RX_CH0)) {
+		if (netif_rx_schedule_prep(dev)) {
+			__raw_writeq(0, sc->sbm_imr);
+			__netif_rx_schedule(dev);
+			/* Depend on the exit from poll to reenable intr */
+		}
+		else {
+			/* may leave some packets behind */
+			sbdma_rx_process(sc,&(sc->sbm_rxdma),
+					 SBMAC_MAX_RXDESCR * 2, 0);
+		}
+	}
+#else
+	/* Non NAPI */
+	for (;;) {
+ 
 		/*
-		 * Receives on channel 0
+		 * Read the ISR (this clears the bits in the real
+		 * register, except for counter addr)
 		 */
+		isr = __raw_readq(sc->sbm_isr) & ~M_MAC_COUNTER_ADDR;
+
+		if (isr == 0)
+			break;
+
+		handled = 1;
+

drivers/net/Kconfig line 2190

Quoted 10 lines
drivers/net/Kconfig +10 -0
+		if (isr & (M_MAC_INT_CHANNEL << S_MAC_TX_CH0)) {
+			sbdma_tx_process(sc,&(sc->sbm_txdma),
+					 SBMAC_MAX_RXDESCR * 2);
+#ifdef CONFIG_NETPOLL_TRAP
+		       if (netpoll_trap()) {
+			       if (test_and_clear_bit(__LINK_STATE_XOFF, &dev->state)) 
+				       __netif_schedule(dev);
+		       }
+#endif
+		}

Same here.

WBR, Sergei

author
#2 lore raw

Hello,

Quoted 6 lines

-----Original Message-----
From: Sergei Shtylyov [mailto:[email protected]]
Sent: Saturday, March 24, 2007 12:49 PM
To: Mark E Mason
Cc: [email protected]; [email protected]
Subject: Re: [PATCH] NAPI support for Sibyte MAC

[snip]

Quoted 10 lines

drivers/net/sb1250-mac.c line 2149

Quoted 9 lines
drivers/net/sb1250-mac.c +4 -1
@@ -2075,12 +2143,52 @@
 		 */
 
 		if (isr & (M_MAC_INT_CHANNEL << S_MAC_TX_CH0)) {
-			sbdma_tx_process(sc,&(sc->sbm_txdma));
+			sbdma_tx_process(sc,&(sc->sbm_txdma), 0);
+#ifdef CONFIG_NETPOLL_TRAP
+		       if (netpoll_trap()) {
+			       if

(test_and_clear_bit(__LINK_STATE_XOFF, &dev->state))

Quoted 12 lines

drivers/net/sb1250-mac.c line 2153

Quoted 4 lines
drivers/net/sb1250-mac.c +3 -0
+				       __netif_schedule(dev);
+		       }
+#endif
 		}

This just doesn't make sense. That option is enabled to
*prevent* calls to
__netif_schedule() -- you can't override it that way. (Well, how it works
currently, doesn't make much sense either since it totally breaks the TX queue
control -- I was going to post a patch).

I'll admit - this was just copied over from the non-NAPI version we started with ;-).

Are you saying that the #ifdef test should simply be an #ifndef here?

[I'm working on a revised patch based on feedback from Ralf and others which removes the Kconfig option for NAPI - it'll just be on, all the time.]

Thanks,
Mark