PATCH SB1250 NAPI support

PATCH SB1250 NAPI support

[Tom Rix]

[-- Attachment #1: Type: text/plain, Size: 1491 bytes --]

A while ago I submitted this patch for NAPI support for SB1250 / SB1480.
I have tested it again on the recent (6/3/06) linux-mips kernel.
Tom


Below is the original posting.

This patch adds NAPI support for the Broadcom Sibyte SB1xxx family.  The
changes are limited to adding a new config key SBMAC_NAPI to the drivers/
net/Kconfig and by adding the poll op and interrupt support to drivers/
net/sb1250-mac.c.

This patch also has a fix to drivers/net/sb1250-mac.c, the dma descriptor
table ptr is allocated, aligned and the aligned ptr is freed.  If the ptr
was not already aligned (usually is) then the free would not work of what
was returned by the kmalloc. A variable was added to store the unaligned
pointer so that it could be properly freed.

I have tested this patch on a BCM91250A-SWARM Pass 2 / An.

Mark Mason from Broadcom was very helpful and tested this patch on at
least a 1480.

Tom


On Thu, 25 May 2006 08:35:05 -0500, Martin Michlmayr <tbm@cyrius.com>  
wrote:

> * Tom Rix <trix@specifix.com> [2006-05-25 08:06]:
>> I am busy now but will get to it sometime.
>> It will require retesting with the latest kernel source.
>> Is there a particular maintainer(s) that I should cc on this patch?
>
> jgarzik@pobox.com
>
> He just reminded people that the merge window for 2.6.18 will open
> soon (and that he's collecting patches for that already) so now would
> be an excellent time.
>
> Thanks.



-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/

[-- Attachment #2: mips-sb1250-mac-NAPI-3.patch --]
[-- Type: application/octet-stream, Size: 10804 bytes --]

diff -rup a/drivers/net/Kconfig b/drivers/net/Kconfig
--- a/drivers/net/Kconfig	2006-03-09 04:25:38.000000000 -0600
+++ b/drivers/net/Kconfig	2006-03-09 05:30:48.000000000 -0600
@@ -2008,10 +2008,6 @@ config R8169_NAPI
 
 	  If in doubt, say N.
 
-config NET_SB1250_MAC
-	tristate "SB1250 Ethernet support"
-	depends on SIBYTE_SB1xxx_SOC
-
 config R8169_VLAN
 	bool "VLAN support"
 	depends on R8169 && VLAN_8021Q
@@ -2021,6 +2017,27 @@ config R8169_VLAN
 	  
 	  If in doubt, say Y.
 
+config NET_SB1250_MAC
+	tristate "SB1250 Ethernet support"
+	depends on SIBYTE_SB1xxx_SOC
+
+config SBMAC_NAPI
+	bool "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 SIS190
 	tristate "SiS190/SiS191 gigabit ethernet support"
 	depends on PCI
diff -rup a/drivers/net/sb1250-mac.c b/drivers/net/sb1250-mac.c
--- a/drivers/net/sb1250-mac.c	2006-03-09 04:25:41.000000000 -0600
+++ b/drivers/net/sb1250-mac.c	2006-03-09 05:30:52.000000000 -0600
@@ -209,6 +209,7 @@ typedef struct sbmacdma_s {
 	 * This stuff is for maintenance of the ring
 	 */
 
+	sbdmadscr_t     *sbdma_dscrtable_unaligned; 
 	sbdmadscr_t     *sbdma_dscrtable;	/* base of descriptor table */
 	sbdmadscr_t     *sbdma_dscrtable_end; /* end of descriptor table */
 
@@ -291,8 +292,8 @@ static int sbdma_add_rcvbuffer(sbmacdma_
 static int sbdma_add_txbuffer(sbmacdma_t *d,struct sk_buff *m);
 static void sbdma_emptyring(sbmacdma_t *d);
 static void sbdma_fillring(sbmacdma_t *d);
-static void sbdma_rx_process(struct sbmac_softc *sc,sbmacdma_t *d);
-static void sbdma_tx_process(struct sbmac_softc *sc,sbmacdma_t *d);
+static int sbdma_rx_process(struct sbmac_softc *sc,sbmacdma_t *d, int poll);
+static void sbdma_tx_process(struct sbmac_softc *sc,sbmacdma_t *d, int poll);
 static int sbmac_initctx(struct sbmac_softc *s);
 static void sbmac_channel_start(struct sbmac_softc *s);
 static void sbmac_channel_stop(struct sbmac_softc *s);
@@ -313,6 +314,10 @@ static struct net_device_stats *sbmac_ge
 static void sbmac_set_rx_mode(struct net_device *dev);
 static int sbmac_mii_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
 static int sbmac_close(struct net_device *dev);
+#if defined (CONFIG_SBMAC_NAPI)
+static int sbmac_poll(struct net_device *poll_dev, int *budget);
+#endif
+
 static int sbmac_mii_poll(struct sbmac_softc *s,int noisy);
 static int sbmac_mii_probe(struct net_device *dev);
 
@@ -740,7 +745,7 @@ static void sbdma_initctx(sbmacdma_t *d,
 
 	d->sbdma_maxdescr = maxdescr;
 
-	d->sbdma_dscrtable = (sbdmadscr_t *)
+	d->sbdma_dscrtable_unaligned = d->sbdma_dscrtable = (sbdmadscr_t *)
 		kmalloc((d->sbdma_maxdescr+1)*sizeof(sbdmadscr_t), GFP_KERNEL);
 
 	/*
@@ -782,7 +787,6 @@ static void sbdma_initctx(sbmacdma_t *d,
 		d->sbdma_int_timeout = 0;
 	}
 #endif
-
 }
 
 /**********************************************************************
@@ -1150,13 +1154,14 @@ static void sbdma_fillring(sbmacdma_t *d
  *  	   nothing
  ********************************************************************* */
 
-static void sbdma_rx_process(struct sbmac_softc *sc,sbmacdma_t *d)
+static int sbdma_rx_process(struct sbmac_softc *sc,sbmacdma_t *d, int poll)
 {
 	int curidx;
 	int hwidx;
 	sbdmadscr_t *dsc;
 	struct sk_buff *sb;
 	int len;
+	int work_done = 0;
 
 	for (;;) {
 		/*
@@ -1234,8 +1239,15 @@ static void sbdma_rx_process(struct sbma
 						sb->ip_summed = CHECKSUM_NONE;
 					}
 				}
-
+				
+#if defined(CONFIG_SBMAC_NAPI)
+				if (0 == poll)
+					netif_rx(sb);
+				else
+					netif_receive_skb(sb);
+#else
 				netif_rx(sb);
+#endif				
 			}
 		} else {
 			/*
@@ -1252,8 +1264,9 @@ static void sbdma_rx_process(struct sbma
 		 */
 
 		d->sbdma_remptr = SBDMA_NEXTBUF(d,sbdma_remptr);
-
+		work_done++;
 	}
+	return work_done;
 }
 
 
@@ -1275,15 +1288,22 @@ static void sbdma_rx_process(struct sbma
  *  	   nothing
  ********************************************************************* */
 
-static void sbdma_tx_process(struct sbmac_softc *sc,sbmacdma_t *d)
+static void sbdma_tx_process(struct sbmac_softc *sc,sbmacdma_t *d, int poll)
 {
 	int curidx;
 	int hwidx;
 	sbdmadscr_t *dsc;
 	struct sk_buff *sb;
 	unsigned long flags;
+	int packets_handled = 0;
 
 	spin_lock_irqsave(&(sc->sbm_lock), flags);
+	
+	if (d->sbdma_remptr == d->sbdma_addptr)
+		goto end_unlock;
+
+	hwidx = (int) (((__raw_readq(d->sbdma_curdscr) & M_DMA_CURDSCR_ADDR) -
+			d->sbdma_dscrtable_phys) / sizeof(sbdmadscr_t));
 
 	for (;;) {
 		/*
@@ -1298,15 +1318,12 @@ static void sbdma_tx_process(struct sbma
 		 */
 
 		curidx = d->sbdma_remptr - d->sbdma_dscrtable;
-		hwidx = (int) (((__raw_readq(d->sbdma_curdscr) & M_DMA_CURDSCR_ADDR) -
-				d->sbdma_dscrtable_phys) / sizeof(sbdmadscr_t));
 
 		/*
 		 * If they're the same, that means we've processed all
 		 * of the descriptors up to (but not including) the one that
 		 * the hardware is working on right now.
 		 */
-
 		if (curidx == hwidx)
 			break;
 
@@ -1337,6 +1354,7 @@ static void sbdma_tx_process(struct sbma
 
 		d->sbdma_remptr = SBDMA_NEXTBUF(d,sbdma_remptr);
 
+		packets_handled++;
 	}
 
 	/*
@@ -1344,15 +1362,15 @@ static void sbdma_tx_process(struct sbma
 	 * Other drivers seem to do this when we reach a low
 	 * watermark on the transmit queue.
 	 */
+	
+	if (packets_handled)
+		netif_wake_queue(d->sbdma_eth->sbm_dev);
 
-	netif_wake_queue(d->sbdma_eth->sbm_dev);
-
+ end_unlock:
 	spin_unlock_irqrestore(&(sc->sbm_lock), flags);
 
 }
 
-
-
 /**********************************************************************
  *  SBMAC_INITCTX(s)
  *
@@ -1396,7 +1414,6 @@ static int sbmac_initctx(struct sbmac_so
 	 * Initialize the DMA channels.  Right now, only one per MAC is used
 	 * Note: Only do this _once_, as it allocates memory from the kernel!
 	 */
-
 	sbdma_initctx(&(s->sbm_txdma),s,0,DMA_TX,SBMAC_MAX_TXDESCR);
 	sbdma_initctx(&(s->sbm_rxdma),s,0,DMA_RX,SBMAC_MAX_RXDESCR);
 
@@ -1420,9 +1437,9 @@ static int sbmac_initctx(struct sbmac_so
 
 static void sbdma_uninitctx(struct sbmacdma_s *d)
 {
-	if (d->sbdma_dscrtable) {
-		kfree(d->sbdma_dscrtable);
-		d->sbdma_dscrtable = NULL;
+	if (d->sbdma_dscrtable_unaligned) {
+		kfree(d->sbdma_dscrtable_unaligned);
+		d->sbdma_dscrtable_unaligned = d->sbdma_dscrtable = NULL;
 	}
 
 	if (d->sbdma_ctxtable) {
@@ -1609,12 +1626,12 @@ static void sbmac_channel_start(struct s
 
 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
 	__raw_writeq(M_MAC_RXDMA_EN0 |
-		       M_MAC_TXDMA_EN0, s->sbm_macenable);
+		     M_MAC_TXDMA_EN0, s->sbm_macenable);
 #elif defined(CONFIG_SIBYTE_SB1250) || defined(CONFIG_SIBYTE_BCM112X)
 	__raw_writeq(M_MAC_RXDMA_EN0 |
-		       M_MAC_TXDMA_EN0 |
-		       M_MAC_RX_ENABLE |
-		       M_MAC_TX_ENABLE, s->sbm_macenable);
+		     M_MAC_TXDMA_EN0 |
+		     M_MAC_RX_ENABLE |
+		     M_MAC_TX_ENABLE, s->sbm_macenable);
 #else
 #error invalid SiByte MAC configuation
 #endif
@@ -1624,15 +1641,15 @@ static void sbmac_channel_start(struct s
 	 * Accept any TX interrupt and EOP count/timer RX interrupts on ch 0
 	 */
 	__raw_writeq(((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_TX_CH0) |
-		       ((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_RX_CH0), s->sbm_imr);
+		     ((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_RX_CH0), 
+		     s->sbm_imr);
 #else
 	/*
 	 * Accept any kind of interrupt on TX and RX DMA channel 0
 	 */
 	__raw_writeq((M_MAC_INT_CHANNEL << S_MAC_TX_CH0) |
-		       (M_MAC_INT_CHANNEL << S_MAC_RX_CH0), s->sbm_imr);
+		     (M_MAC_INT_CHANNEL << S_MAC_RX_CH0), s->sbm_imr);
 #endif
-
 	/*
 	 * Enable receiving unicasts and broadcasts
 	 */
@@ -2067,7 +2084,6 @@ static irqreturn_t sbmac_intr(int irq,vo
 		 * 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)
@@ -2079,13 +2095,31 @@ static irqreturn_t sbmac_intr(int irq,vo
 		 * Transmits on channel 0
 		 */
 
+#if defined(CONFIG_SBMAC_NAPI)
 		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);
 		}
 
 		/*
 		 * Receives on channel 0
 		 */
+		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);
+			}
+			else
+			{
+				sbdma_rx_process(sc,&(sc->sbm_rxdma), 0);
+			}
+		}
+#else
+		/* Non NAPI */
+
+		if (isr & (M_MAC_INT_CHANNEL << S_MAC_TX_CH0)) {
+			sbdma_tx_process(sc,&(sc->sbm_txdma), 0);
+		}
 
 		/*
 		 * It's important to test all the bits (or at least the
@@ -2105,8 +2139,9 @@ static irqreturn_t sbmac_intr(int irq,vo
 
 
 		if (isr & (M_MAC_INT_CHANNEL << S_MAC_RX_CH0)) {
-			sbdma_rx_process(sc,&(sc->sbm_rxdma));
+			sbdma_rx_process(sc,&(sc->sbm_rxdma), 0);
 		}
+#endif
 	}
 	return IRQ_RETVAL(handled);
 }
@@ -2405,7 +2440,10 @@ static int sbmac_init(struct net_device 
 	dev->do_ioctl           = sbmac_mii_ioctl;
 	dev->tx_timeout         = sbmac_tx_timeout;
 	dev->watchdog_timeo     = TX_TIMEOUT;
-
+#if defined(CONFIG_SBMAC_NAPI)
+	dev->poll               = sbmac_poll;
+	dev->weight             = 16;
+#endif
 	dev->change_mtu         = sb1250_change_mtu;
 
 	/* This is needed for PASS2 for Rx H/W checksum feature */
@@ -2818,7 +2856,37 @@ static int sbmac_close(struct net_device
 	return 0;
 }
 
+#if defined(CONFIG_SBMAC_NAPI)
+static int sbmac_poll(struct net_device *dev, int *budget)
+{
+	int work_to_do;
+	int work_done;
+	struct sbmac_softc *sc = netdev_priv(dev);
+
+	work_to_do = min(*budget, dev->quota);
+	work_done = sbdma_rx_process(sc,&(sc->sbm_rxdma), 1);
 
+	sbdma_tx_process(sc,&(sc->sbm_txdma), 1);
+
+	*budget -= work_done;
+	dev->quota -= work_done;
+
+	if (work_done < work_to_do) {
+		netif_rx_complete(dev);
+
+#ifdef CONFIG_SBMAC_COALESCE
+		__raw_writeq(((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_TX_CH0) |
+			     ((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_RX_CH0), 
+			     sc->sbm_imr);
+#else
+		__raw_writeq((M_MAC_INT_CHANNEL << S_MAC_TX_CH0) |
+			     (M_MAC_INT_CHANNEL << S_MAC_RX_CH0), sc->sbm_imr);
+#endif
+	}
+
+	return (work_done >= work_to_do);
+}
+#endif
 
 #if defined(SBMAC_ETH0_HWADDR) || defined(SBMAC_ETH1_HWADDR) || defined(SBMAC_ETH2_HWADDR) || defined(SBMAC_ETH3_HWADDR)
 static void

Re: PATCH SB1250 NAPI support

[Martin Michlmayr]

* Tom Rix <trix@specifix.com> [2006-06-04 11:45]:
> A while ago I submitted this patch for NAPI support for SB1250 / SB1480.
> I have tested it again on the recent (6/3/06) linux-mips kernel.

Are there any comments regarding this patch (apart from the fact that
Tom forgot to include a Signed-off-by line).  Can it go in when it's
resent with a Signed-off-by line or does the patch have any other
problems?
-- 
Martin Michlmayr
tbm@cyrius.com

RE: PATCH SB1250 NAPI support

[Mark E Mason]

Hello,

We've done some further work on the bcm1250/bcm1480 MAC driver since
then, primarily in the area of performance improvements, but we've not
quite ready to submit the patch for those changes yet.

In the meantime, Tom's patch fixes all of the really bad problems, and
we'd really appreciate it if it could be merged in.

Thx,
Mark Mason
Broadcom

> -----Original Message-----
> From: Martin Michlmayr [mailto:tbm@cyrius.com] 
> Sent: Thursday, June 29, 2006 1:22 AM
> To: jgarzik@pobox.com; netdev@vger.kernel.org
> Cc: Tom Rix; Mark E Mason
> Subject: Re: PATCH SB1250 NAPI support
> 
> * Tom Rix <trix@specifix.com> [2006-06-04 11:45]:
> > A while ago I submitted this patch for NAPI support for 
> SB1250 / SB1480.
> > I have tested it again on the recent (6/3/06) linux-mips kernel.
> 
> Are there any comments regarding this patch (apart from the fact that
> Tom forgot to include a Signed-off-by line).  Can it go in when it's
> resent with a Signed-off-by line or does the patch have any other
> problems?
> -- 
> Martin Michlmayr
> tbm@cyrius.com
> 
> 

Re: PATCH SB1250 NAPI support

[Francois Romieu]

Tom Rix <trix@specifix.com> :
[...]
diff -rup a/drivers/net/sb1250-mac.c b/drivers/net/sb1250-mac.c
--- a/drivers/net/sb1250-mac.c	2006-03-09 04:25:41.000000000 -0600
+++ b/drivers/net/sb1250-mac.c	2006-03-09 05:30:52.000000000 -0600
[...]
@@ -2079,13 +2095,31 @@ static irqreturn_t sbmac_intr(int irq,vo
 		 * Transmits on channel 0
 		 */
 
+#if defined(CONFIG_SBMAC_NAPI)
 		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);
 		}
 
 		/*
 		 * Receives on channel 0
 		 */
+		if (isr & (M_MAC_INT_CHANNEL << S_MAC_RX_CH0)) {
+			if (netif_rx_schedule_prep(dev))
+			{

An irq could appear here. I am skeptical that it is safe to write
the irq mask register so late.

One should probably consider a break in the enclosing "for" loop too.


+				__raw_writeq(0, sc->sbm_imr);
+				__netif_rx_schedule(dev);
+			}
+			else
+			{

} else {, please.

-- 
Ueimor

Re: PATCH SB1250 NAPI support

[Tom Rix]

[-- Attachment #1: Type: text/plain, Size: 1394 bytes --]

Here is a revised patch that does the interupt masking earlier.  Since  
resetting the mask is done now in 3 places, I created a macro to handle  
it.  This patch has been tested on the linux-mips kernel from 6/29/06 on a  
sb1250.
Tom

On Thu, 29 Jun 2006 15:01:07 -0500, Francois Romieu <romieu@fr.zoreil.com>  
wrote:

> Tom Rix <trix@specifix.com> :
> [...]
> diff -rup a/drivers/net/sb1250-mac.c b/drivers/net/sb1250-mac.c
> --- a/drivers/net/sb1250-mac.c	2006-03-09 04:25:41.000000000 -0600
> +++ b/drivers/net/sb1250-mac.c	2006-03-09 05:30:52.000000000 -0600
> [...]
> @@ -2079,13 +2095,31 @@ static irqreturn_t sbmac_intr(int irq,vo
>  		 * Transmits on channel 0
>  		 */
> +#if defined(CONFIG_SBMAC_NAPI)
>  		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);
>  		}
> 		/*
>  		 * Receives on channel 0
>  		 */
> +		if (isr & (M_MAC_INT_CHANNEL << S_MAC_RX_CH0)) {
> +			if (netif_rx_schedule_prep(dev))
> +			{
>
> An irq could appear here. I am skeptical that it is safe to write
> the irq mask register so late.
>
> One should probably consider a break in the enclosing "for" loop too.
>
>
> +				__raw_writeq(0, sc->sbm_imr);
> +				__netif_rx_schedule(dev);
> +			}
> +			else
> +			{
>
> } else {, please.
>



-- 
Using Opera's revolutionary e-mail client: http://www.opera.com/mail/

[-- Attachment #2: mips-sb1250-mac-NAPI-5.patch --]
[-- Type: application/octet-stream, Size: 11138 bytes --]

--- a/drivers/net/Kconfig	2006-07-01 06:29:11.000000000 -0500
+++ b/drivers/net/Kconfig	2006-07-01 06:28:58.000000000 -0500
@@ -2013,10 +2013,6 @@ config R8169_NAPI
 
 	  If in doubt, say N.
 
-config NET_SB1250_MAC
-	tristate "SB1250 Ethernet support"
-	depends on SIBYTE_SB1xxx_SOC
-
 config R8169_VLAN
 	bool "VLAN support"
 	depends on R8169 && VLAN_8021Q
@@ -2026,6 +2022,27 @@ config R8169_VLAN
 	  
 	  If in doubt, say Y.
 
+config NET_SB1250_MAC
+	tristate "SB1250 Ethernet support"
+	depends on SIBYTE_SB1xxx_SOC
+
+config SBMAC_NAPI
+	bool "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 SIS190
 	tristate "SiS190/SiS191 gigabit ethernet support"
 	depends on PCI
--- a/drivers/net/sb1250-mac.c	2006-07-01 06:29:27.000000000 -0500
+++ b/drivers/net/sb1250-mac.c	2006-07-01 06:29:18.000000000 -0500
@@ -164,6 +164,25 @@ typedef enum { sbmac_state_uninit, sbmac
 #define ENET_PACKET_SIZE	1518
 /*#define ENET_PACKET_SIZE	9216 */
 
+
+#ifdef CONFIG_SBMAC_COALESCE
+/*
+ * Accept any TX interrupt and EOP count/timer RX interrupts on ch 0
+ */
+#define SET_INTR_MASK(S) \
+__raw_writeq(((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_TX_CH0) | \
+	     ((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_RX_CH0), \
+	     (S)->sbm_imr);
+#else
+/*
+ * Accept any kind of interrupt on TX and RX DMA channel 0
+ */
+#define SET_INTR_MASK(S) \
+__raw_writeq((M_MAC_INT_CHANNEL << S_MAC_TX_CH0) | \
+             (M_MAC_INT_CHANNEL << S_MAC_RX_CH0), (S)->sbm_imr);
+#endif
+
+
 /**********************************************************************
  *  DMA Descriptor structure
  ********************************************************************* */
@@ -205,6 +224,7 @@ typedef struct sbmacdma_s {
 	 * This stuff is for maintenance of the ring
 	 */
 
+	sbdmadscr_t     *sbdma_dscrtable_unaligned; 
 	sbdmadscr_t     *sbdma_dscrtable;	/* base of descriptor table */
 	sbdmadscr_t     *sbdma_dscrtable_end; /* end of descriptor table */
 
@@ -287,8 +307,8 @@ static int sbdma_add_rcvbuffer(sbmacdma_
 static int sbdma_add_txbuffer(sbmacdma_t *d,struct sk_buff *m);
 static void sbdma_emptyring(sbmacdma_t *d);
 static void sbdma_fillring(sbmacdma_t *d);
-static void sbdma_rx_process(struct sbmac_softc *sc,sbmacdma_t *d);
-static void sbdma_tx_process(struct sbmac_softc *sc,sbmacdma_t *d);
+static int sbdma_rx_process(struct sbmac_softc *sc,sbmacdma_t *d, int poll);
+static void sbdma_tx_process(struct sbmac_softc *sc,sbmacdma_t *d, int poll);
 static int sbmac_initctx(struct sbmac_softc *s);
 static void sbmac_channel_start(struct sbmac_softc *s);
 static void sbmac_channel_stop(struct sbmac_softc *s);
@@ -309,6 +329,10 @@ static struct net_device_stats *sbmac_ge
 static void sbmac_set_rx_mode(struct net_device *dev);
 static int sbmac_mii_ioctl(struct net_device *dev, struct ifreq *rq, int cmd);
 static int sbmac_close(struct net_device *dev);
+#if defined (CONFIG_SBMAC_NAPI)
+static int sbmac_poll(struct net_device *poll_dev, int *budget);
+#endif
+
 static int sbmac_mii_poll(struct sbmac_softc *s,int noisy);
 static int sbmac_mii_probe(struct net_device *dev);
 
@@ -736,7 +760,7 @@ static void sbdma_initctx(sbmacdma_t *d,
 
 	d->sbdma_maxdescr = maxdescr;
 
-	d->sbdma_dscrtable = (sbdmadscr_t *)
+	d->sbdma_dscrtable_unaligned = d->sbdma_dscrtable = (sbdmadscr_t *)
 		kmalloc((d->sbdma_maxdescr+1)*sizeof(sbdmadscr_t), GFP_KERNEL);
 
 	/*
@@ -778,7 +802,6 @@ static void sbdma_initctx(sbmacdma_t *d,
 		d->sbdma_int_timeout = 0;
 	}
 #endif
-
 }
 
 /**********************************************************************
@@ -1146,13 +1169,14 @@ static void sbdma_fillring(sbmacdma_t *d
  *  	   nothing
  ********************************************************************* */
 
-static void sbdma_rx_process(struct sbmac_softc *sc,sbmacdma_t *d)
+static int sbdma_rx_process(struct sbmac_softc *sc,sbmacdma_t *d, int poll)
 {
 	int curidx;
 	int hwidx;
 	sbdmadscr_t *dsc;
 	struct sk_buff *sb;
 	int len;
+	int work_done = 0;
 
 	for (;;) {
 		/*
@@ -1230,8 +1254,15 @@ static void sbdma_rx_process(struct sbma
 						sb->ip_summed = CHECKSUM_NONE;
 					}
 				}
-
+				
+#if defined(CONFIG_SBMAC_NAPI)
+				if (0 == poll)
+					netif_rx(sb);
+				else
+					netif_receive_skb(sb);
+#else
 				netif_rx(sb);
+#endif				
 			}
 		} else {
 			/*
@@ -1248,8 +1279,9 @@ static void sbdma_rx_process(struct sbma
 		 */
 
 		d->sbdma_remptr = SBDMA_NEXTBUF(d,sbdma_remptr);
-
+		work_done++;
 	}
+	return work_done;
 }
 
 
@@ -1271,15 +1303,22 @@ static void sbdma_rx_process(struct sbma
  *  	   nothing
  ********************************************************************* */
 
-static void sbdma_tx_process(struct sbmac_softc *sc,sbmacdma_t *d)
+static void sbdma_tx_process(struct sbmac_softc *sc,sbmacdma_t *d, int poll)
 {
 	int curidx;
 	int hwidx;
 	sbdmadscr_t *dsc;
 	struct sk_buff *sb;
 	unsigned long flags;
+	int packets_handled = 0;
 
 	spin_lock_irqsave(&(sc->sbm_lock), flags);
+	
+	if (d->sbdma_remptr == d->sbdma_addptr)
+		goto end_unlock;
+
+	hwidx = (int) (((__raw_readq(d->sbdma_curdscr) & M_DMA_CURDSCR_ADDR) -
+			d->sbdma_dscrtable_phys) / sizeof(sbdmadscr_t));
 
 	for (;;) {
 		/*
@@ -1294,15 +1333,12 @@ static void sbdma_tx_process(struct sbma
 		 */
 
 		curidx = d->sbdma_remptr - d->sbdma_dscrtable;
-		hwidx = (int) (((__raw_readq(d->sbdma_curdscr) & M_DMA_CURDSCR_ADDR) -
-				d->sbdma_dscrtable_phys) / sizeof(sbdmadscr_t));
 
 		/*
 		 * If they're the same, that means we've processed all
 		 * of the descriptors up to (but not including) the one that
 		 * the hardware is working on right now.
 		 */
-
 		if (curidx == hwidx)
 			break;
 
@@ -1333,6 +1369,7 @@ static void sbdma_tx_process(struct sbma
 
 		d->sbdma_remptr = SBDMA_NEXTBUF(d,sbdma_remptr);
 
+		packets_handled++;
 	}
 
 	/*
@@ -1340,15 +1377,15 @@ static void sbdma_tx_process(struct sbma
 	 * Other drivers seem to do this when we reach a low
 	 * watermark on the transmit queue.
 	 */
+	
+	if (packets_handled)
+		netif_wake_queue(d->sbdma_eth->sbm_dev);
 
-	netif_wake_queue(d->sbdma_eth->sbm_dev);
-
+ end_unlock:
 	spin_unlock_irqrestore(&(sc->sbm_lock), flags);
 
 }
 
-
-
 /**********************************************************************
  *  SBMAC_INITCTX(s)
  *
@@ -1392,7 +1429,6 @@ static int sbmac_initctx(struct sbmac_so
 	 * Initialize the DMA channels.  Right now, only one per MAC is used
 	 * Note: Only do this _once_, as it allocates memory from the kernel!
 	 */
-
 	sbdma_initctx(&(s->sbm_txdma),s,0,DMA_TX,SBMAC_MAX_TXDESCR);
 	sbdma_initctx(&(s->sbm_rxdma),s,0,DMA_RX,SBMAC_MAX_RXDESCR);
 
@@ -1416,9 +1452,9 @@ static int sbmac_initctx(struct sbmac_so
 
 static void sbdma_uninitctx(struct sbmacdma_s *d)
 {
-	if (d->sbdma_dscrtable) {
-		kfree(d->sbdma_dscrtable);
-		d->sbdma_dscrtable = NULL;
+	if (d->sbdma_dscrtable_unaligned) {
+		kfree(d->sbdma_dscrtable_unaligned);
+		d->sbdma_dscrtable_unaligned = d->sbdma_dscrtable = NULL;
 	}
 
 	if (d->sbdma_ctxtable) {
@@ -1605,29 +1641,16 @@ static void sbmac_channel_start(struct s
 
 #if defined(CONFIG_SIBYTE_BCM1x55) || defined(CONFIG_SIBYTE_BCM1x80)
 	__raw_writeq(M_MAC_RXDMA_EN0 |
-		       M_MAC_TXDMA_EN0, s->sbm_macenable);
+		     M_MAC_TXDMA_EN0, s->sbm_macenable);
 #elif defined(CONFIG_SIBYTE_SB1250) || defined(CONFIG_SIBYTE_BCM112X)
 	__raw_writeq(M_MAC_RXDMA_EN0 |
-		       M_MAC_TXDMA_EN0 |
-		       M_MAC_RX_ENABLE |
-		       M_MAC_TX_ENABLE, s->sbm_macenable);
+		     M_MAC_TXDMA_EN0 |
+		     M_MAC_RX_ENABLE |
+		     M_MAC_TX_ENABLE, s->sbm_macenable);
 #else
 #error invalid SiByte MAC configuation
 #endif
-
-#ifdef CONFIG_SBMAC_COALESCE
-	/*
-	 * Accept any TX interrupt and EOP count/timer RX interrupts on ch 0
-	 */
-	__raw_writeq(((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_TX_CH0) |
-		       ((M_MAC_INT_EOP_COUNT | M_MAC_INT_EOP_TIMER) << S_MAC_RX_CH0), s->sbm_imr);
-#else
-	/*
-	 * Accept any kind of interrupt on TX and RX DMA channel 0
-	 */
-	__raw_writeq((M_MAC_INT_CHANNEL << S_MAC_TX_CH0) |
-		       (M_MAC_INT_CHANNEL << S_MAC_RX_CH0), s->sbm_imr);
-#endif
+	SET_INTR_MASK(s);
 
 	/*
 	 * Enable receiving unicasts and broadcasts
@@ -2063,7 +2086,6 @@ static irqreturn_t sbmac_intr(int irq,vo
 		 * 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)
@@ -2075,13 +2097,33 @@ static irqreturn_t sbmac_intr(int irq,vo
 		 * Transmits on channel 0
 		 */
 
+#if defined(CONFIG_SBMAC_NAPI)
+		if (isr & (M_MAC_INT_CHANNEL << S_MAC_RX_CH0)) {
+			__raw_writeq(0, sc->sbm_imr);
+		}
+
 		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);
 		}
 
 		/*
 		 * Receives on channel 0
 		 */
+		if (isr & (M_MAC_INT_CHANNEL << S_MAC_RX_CH0)) {
+			if (netif_rx_schedule_prep(dev))
+			{
+				__netif_rx_schedule(dev);
+			} else {
+				SET_INTR_MASK(sc);
+				sbdma_rx_process(sc,&(sc->sbm_rxdma), 0);
+			}
+		}
+#else
+		/* Non NAPI */
+
+		if (isr & (M_MAC_INT_CHANNEL << S_MAC_TX_CH0)) {
+			sbdma_tx_process(sc,&(sc->sbm_txdma), 0);
+		}
 
 		/*
 		 * It's important to test all the bits (or at least the
@@ -2101,8 +2143,9 @@ static irqreturn_t sbmac_intr(int irq,vo
 
 
 		if (isr & (M_MAC_INT_CHANNEL << S_MAC_RX_CH0)) {
-			sbdma_rx_process(sc,&(sc->sbm_rxdma));
+			sbdma_rx_process(sc,&(sc->sbm_rxdma), 0);
 		}
+#endif
 	}
 	return IRQ_RETVAL(handled);
 }
@@ -2401,7 +2444,10 @@ static int sbmac_init(struct net_device 
 	dev->do_ioctl           = sbmac_mii_ioctl;
 	dev->tx_timeout         = sbmac_tx_timeout;
 	dev->watchdog_timeo     = TX_TIMEOUT;
-
+#if defined(CONFIG_SBMAC_NAPI)
+	dev->poll               = sbmac_poll;
+	dev->weight             = 16;
+#endif
 	dev->change_mtu         = sb1250_change_mtu;
 
 	/* This is needed for PASS2 for Rx H/W checksum feature */
@@ -2809,7 +2855,29 @@ static int sbmac_close(struct net_device
 	return 0;
 }
 
+#if defined(CONFIG_SBMAC_NAPI)
+static int sbmac_poll(struct net_device *dev, int *budget)
+{
+	int work_to_do;
+	int work_done;
+	struct sbmac_softc *sc = netdev_priv(dev);
+
+	work_to_do = min(*budget, dev->quota);
+	work_done = sbdma_rx_process(sc,&(sc->sbm_rxdma), 1);
 
+	sbdma_tx_process(sc,&(sc->sbm_txdma), 1);
+
+	*budget -= work_done;
+	dev->quota -= work_done;
+
+	if (work_done < work_to_do) {
+		netif_rx_complete(dev);
+		SET_INTR_MASK(sc);
+	}
+
+	return (work_done >= work_to_do);
+}
+#endif
 
 #if defined(SBMAC_ETH0_HWADDR) || defined(SBMAC_ETH1_HWADDR) || defined(SBMAC_ETH2_HWADDR) || defined(SBMAC_ETH3_HWADDR)
 static void