Re: Looking for testers with these NICs

Re: Looking for testers with these NICs

[Alan Cox]

On Wed, 2002-10-09 at 22:31, Denis Vlasenko wrote:
> depca.c

Already done

> ewrk3.c
> lp486e.c
> ni5010.c
> ni65.c
> smc9194.c
> 
> These drivers currently don't compile. I'm fixing them but
> can't test on a live hardware. Anyone?

You may want to merge the 2.4 changes as you go. For lp486e you may just
have to pray, its an unusual 486 onboard ethernet. I know 3 people who
have one, and one (mine) burned out the psu and died

Re: Looking for testers with these NICs

[Adam Kropelin]

On Wed, Oct 09, 2002 at 07:31:17PM -0200, Denis Vlasenko wrote:
> ewrk3.c

I've got a few of these laying around. Send whatever patches you want tested
and I'll give it a shot.

--Adam

Re: Looking for testers with these NICs

[Martin Josefsson]

On Thu, 2002-10-10 at 00:37, Denis Vlasenko wrote:

> I'd suggest SMP/preempt heavy IO. Is there stress test software for NICs?
> What is pktgen?

A kernelmodule that's capable of generating over a million packets per
second on a fast enough machine. It stresstests NIC's and NIC-drivers a
lot.

-- 
/Martin

Never argue with an idiot. They drag you down to their level, then beat
you with experience.

Re: Looking for testers with these NICs

[Ben Greear]

Denis Vlasenko wrote:
> On 9 October 2002 15:14, Adam Kropelin wrote:
> 
>>On Wed, Oct 09, 2002 at 07:31:17PM -0200, Denis Vlasenko wrote:
>>
>>>ewrk3.c
>>
>>I've got a few of these laying around. Send whatever patches you want
>>tested and I'll give it a shot.
> 
> 
> Please do your best in trying to break it, especially since you say you have
> more than one. Can you plug them all in one box?
> 
> I'd suggest SMP/preempt heavy IO. Is there stress test software for NICs?
> What is pktgen?

pktgen will definately stress the nics.  Try sending to yourself
as well so you can test the pkt receive code..it's normally the weakest.

Ben

> --
> vda
> 
> diff -u --recursive linux-2.5.40org/drivers/net/ewrk3.c linux-2.5.40/drivers/net/ewrk3.c
> --- linux-2.5.40org/drivers/net/ewrk3.c	Tue Oct  1 05:06:58 2002
> +++ linux-2.5.40/drivers/net/ewrk3.c	Thu Oct  3 12:09:46 2002
> @@ -930,6 +930,7 @@
>  	spin_unlock(&lp->hw_lock);
>  }
> 
> +/* Called with lp->hw_lock held */
>  static int ewrk3_rx(struct net_device *dev)
>  {
>  	struct ewrk3_private *lp = (struct ewrk3_private *) dev->priv;
> @@ -1055,8 +1056,9 @@
>  }
> 
>  /*
> -   ** Buffer sent - check for TX buffer errors.
> - */
> +** Buffer sent - check for TX buffer errors.
> +** Called with lp->hw_lock held
> +*/
>  static int ewrk3_tx(struct net_device *dev)
>  {
>  	struct ewrk3_private *lp = (struct ewrk3_private *) dev->priv;
> @@ -1631,6 +1633,7 @@
>  	u_long iobase = dev->base_addr;
>  	int i, j, status = 0;
>  	u_char csr;
> +	unsigned long flags;
>  	union ewrk3_addr {
>  		u_char addr[HASH_TABLE_LEN * ETH_ALEN];
>  		u_short val[(HASH_TABLE_LEN * ETH_ALEN) >> 1];
> @@ -1745,19 +1748,26 @@
>  		}
> 
>  		break;
> -	case EWRK3_GET_STATS:	/* Get the driver statistics */
> -		cli();
> -		ioc->len = sizeof(lp->pktStats);
> -		if (copy_to_user(ioc->data, &lp->pktStats, ioc->len))
> -			status = -EFAULT;
> -		sti();
> +	case EWRK3_GET_STATS: { /* Get the driver statistics */
> +		typeof(lp->pktStats) *tmp_stats =
> +        		kmalloc(sizeof(lp->pktStats), GFP_KERNEL);
> +		if (!tmp_stats) return -ENOMEM;
> +
> +		spin_lock_irqsave(&lp->hw_lock, flags);
> +		memcpy(tmp_stats, &lp->pktStats, sizeof(lp->pktStats));
> +		spin_unlock_irqrestore(&lp->hw_lock, flags);
> 
> +		ioc->len = sizeof(lp->pktStats);
> +		if (copy_to_user(ioc->data, tmp_stats, sizeof(lp->pktStats)))
> +    			status = -EFAULT;
> +		kfree(tmp_stats);
>  		break;
> +	}
>  	case EWRK3_CLR_STATS:	/* Zero out the driver statistics */
>  		if (capable(CAP_NET_ADMIN)) {
> -			cli();
> +			spin_lock_irqsave(&lp->hw_lock, flags);
>  			memset(&lp->pktStats, 0, sizeof(lp->pktStats));
> -			sti();
> +			spin_unlock_irqrestore(&lp->hw_lock,flags);
>  		} else {
>  			status = -EPERM;
>  		}
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
> 


-- 
Ben Greear <greearb@candelatech.com>       <Ben_Greear AT excite.com>
President of Candela Technologies Inc      http://www.candelatech.com
ScryMUD:  http://scry.wanfear.com     http://scry.wanfear.com/~greear

Re: Looking for testers with these NICs

[Derek Fawcus]

On Wed, Oct 09, 2002 at 07:31:17PM -0200, Denis Vlasenko wrote:
> ni65.c

I've got some of these knocking about,  but rather than use that driver,
I have a (quite old) patch,  that allows the normal lance driver to be
used.  The patch dates back from 1.3.x,  but I think I may have a more
recent version around.

DF

Re: Looking for testers with these NICs

[Adam Kropelin]

On Wed, Oct 09, 2002 at 08:37:48PM -0200, Denis Vlasenko wrote:
> On 9 October 2002 15:14, Adam Kropelin wrote:
> > On Wed, Oct 09, 2002 at 07:31:17PM -0200, Denis Vlasenko wrote:
> > > ewrk3.c
> >
> > I've got a few of these laying around. Send whatever patches you want
> > tested and I'll give it a shot.
> 
> Please do your best in trying to break it, especially since you say you have
> more than one. Can you plug them all in one box?
> 
> I'd suggest SMP/preempt heavy IO. Is there stress test software for NICs?
> What is pktgen?

These are ISA nics and my SMP box has a lot of other hardware in it right now,
but I'll see what I can do. I haven't tried to do unreasonable things with old
hardware in a while so it should be good for a grin, anyway.

I'll report my findings.

--Adam

Looking for testers with these NICs

[Denis Vlasenko]

I'm looking for people who is brave (or stupid) enough to try 2.5
and who has network card(s) driven by one of the following drivers:

3c505.c
3c515.c
82596.c
at1700.c
depca.c
ewrk3.c
lp486e.c
ni5010.c
ni65.c
smc9194.c

These drivers currently don't compile. I'm fixing them but
can't test on a live hardware. Anyone?
--
vda

Re: Looking for testers with these NICs

[Denis Vlasenko]

On 9 October 2002 15:14, Adam Kropelin wrote:
> On Wed, Oct 09, 2002 at 07:31:17PM -0200, Denis Vlasenko wrote:
> > ewrk3.c
>
> I've got a few of these laying around. Send whatever patches you want
> tested and I'll give it a shot.

Please do your best in trying to break it, especially since you say you have
more than one. Can you plug them all in one box?

I'd suggest SMP/preempt heavy IO. Is there stress test software for NICs?
What is pktgen?
--
vda

diff -u --recursive linux-2.5.40org/drivers/net/ewrk3.c linux-2.5.40/drivers/net/ewrk3.c
--- linux-2.5.40org/drivers/net/ewrk3.c	Tue Oct  1 05:06:58 2002
+++ linux-2.5.40/drivers/net/ewrk3.c	Thu Oct  3 12:09:46 2002
@@ -930,6 +930,7 @@
 	spin_unlock(&lp->hw_lock);
 }

+/* Called with lp->hw_lock held */
 static int ewrk3_rx(struct net_device *dev)
 {
 	struct ewrk3_private *lp = (struct ewrk3_private *) dev->priv;
@@ -1055,8 +1056,9 @@
 }

 /*
-   ** Buffer sent - check for TX buffer errors.
- */
+** Buffer sent - check for TX buffer errors.
+** Called with lp->hw_lock held
+*/
 static int ewrk3_tx(struct net_device *dev)
 {
 	struct ewrk3_private *lp = (struct ewrk3_private *) dev->priv;
@@ -1631,6 +1633,7 @@
 	u_long iobase = dev->base_addr;
 	int i, j, status = 0;
 	u_char csr;
+	unsigned long flags;
 	union ewrk3_addr {
 		u_char addr[HASH_TABLE_LEN * ETH_ALEN];
 		u_short val[(HASH_TABLE_LEN * ETH_ALEN) >> 1];
@@ -1745,19 +1748,26 @@
 		}

 		break;
-	case EWRK3_GET_STATS:	/* Get the driver statistics */
-		cli();
-		ioc->len = sizeof(lp->pktStats);
-		if (copy_to_user(ioc->data, &lp->pktStats, ioc->len))
-			status = -EFAULT;
-		sti();
+	case EWRK3_GET_STATS: { /* Get the driver statistics */
+		typeof(lp->pktStats) *tmp_stats =
+        		kmalloc(sizeof(lp->pktStats), GFP_KERNEL);
+		if (!tmp_stats) return -ENOMEM;
+
+		spin_lock_irqsave(&lp->hw_lock, flags);
+		memcpy(tmp_stats, &lp->pktStats, sizeof(lp->pktStats));
+		spin_unlock_irqrestore(&lp->hw_lock, flags);

+		ioc->len = sizeof(lp->pktStats);
+		if (copy_to_user(ioc->data, tmp_stats, sizeof(lp->pktStats)))
+    			status = -EFAULT;
+		kfree(tmp_stats);
 		break;
+	}
 	case EWRK3_CLR_STATS:	/* Zero out the driver statistics */
 		if (capable(CAP_NET_ADMIN)) {
-			cli();
+			spin_lock_irqsave(&lp->hw_lock, flags);
 			memset(&lp->pktStats, 0, sizeof(lp->pktStats));
-			sti();
+			spin_unlock_irqrestore(&lp->hw_lock,flags);
 		} else {
 			status = -EPERM;
 		}

[PATCH] ewrk3: Add support for multiple NICs when modular (WAS: Looking for testers...)

[Adam Kropelin]

Testing is in progress... Three NICs, 2 CPU SMP, preempt enabled. No problems so
far. For my own convenience I wanted support for multiple NICs when the driver
is built as a module, so here's a fairly simple patch to add that capability.
Against 2.5.41 + your cli/sti removal patch.

--Adam

--- linux-2.5.41/drivers/net/ewrk3.c	Wed Oct  9 21:29:15 2002
+++ linux-2.5.41-fix2/drivers/net/ewrk3.c	Wed Oct  9 21:40:23 2002
@@ -76,6 +76,7 @@
    kernel with the ewrk3 configuration turned off and reboot.
    5) insmod ewrk3.o
    [Alan Cox: Changed this so you can insmod ewrk3.o irq=x io=y]
+   [Adam Kropelin: now accepts irq=x1,x2 io=y1,y2 for multiple cards]
    6) run the net startup bits for your new eth?? interface manually
    (usually /etc/rc.inet[12] at boot time).
    7) enjoy!
@@ -130,10 +131,12 @@
    Add new multicasting code.
    0.41    20-Jan-96   Fix IRQ set up problem reported by
    <kenneth@bbs.sas.ntu.ac.sg>.
-   0.42    22-Apr-96      Fix alloc_device() bug <jari@markkus2.fimr.fi>
-   0.43    16-Aug-96      Update alloc_device() to conform to de4x5.c
-   0.44    08-Nov-01      use library crc32 functions <Matt_Domsch@dell.com>
-   0.45	   19-Jul-02	  fix unaligned access on alpha <martin@bruli.net>
+   0.42    22-Apr-96   Fix alloc_device() bug <jari@markkus2.fimr.fi>
+   0.43    16-Aug-96   Update alloc_device() to conform to de4x5.c
+   0.44    08-Nov-01   use library crc32 functions <Matt_Domsch@dell.com>
+   0.45    19-Jul-02   fix unaligned access on alpha <martin@bruli.net>
+   0.46    10-Oct-02   cli/sti removal <VDA@port.imtp.ilyichevsk.odessa.ua>
+   Multiple NIC support when module <akropel1@rochester.rr.com>
 
    =========================================================================
  */
@@ -167,7 +170,7 @@
 #include "ewrk3.h"
 
 static char version[] __initdata =
-"ewrk3.c:v0.43a 2001/02/04 davies@maniac.ultranet.com\n";
+"ewrk3.c:v0.46 2002/10/09 davies@maniac.ultranet.com\n";
 
 #ifdef EWRK3_DEBUG
 static int ewrk3_debug = EWRK3_DEBUG;
@@ -196,6 +199,7 @@
 #define EWRK3_IOP_INC 0x20	/* I/O address increment */
 #define EWRK3_TOTAL_SIZE 0x20	/* required I/O address length */
 
+/* If you change this, remember to also change MODULE_PARM array limits */
 #ifndef MAX_NUM_EWRK3S
 #define MAX_NUM_EWRK3S 21
 #endif
@@ -1853,35 +1857,62 @@
 }
 
 #ifdef MODULE
-static struct net_device thisEthwrk;
-static int io = 0x300;		/* <--- EDIT THESE LINES FOR YOUR CONFIGURATION */
-static int irq = 5;		/* or use the insmod io= irq= options           */
-
-MODULE_PARM(io, "i");
-MODULE_PARM(irq, "i");
-MODULE_PARM_DESC(io, "EtherWORKS 3 I/O base address");
-MODULE_PARM_DESC(irq, "EtherWORKS 3 IRQ number");
+static struct net_device *ewrk3_devs[MAX_NUM_EWRK3S];
+static int ndevs;
+static int io[MAX_NUM_EWRK3S+1] = { 0x300, 0, };		/* <--- EDIT THESE LINES FOR YOUR CONFIGURATION */
+static int irq[MAX_NUM_EWRK3S+1] = { 5, 0, };		/* or use the insmod io= irq= options           */
+
+/* '21' below should really be 'MAX_NUM_EWRK3S' */
+MODULE_PARM(io, "0-21i");
+MODULE_PARM(irq, "0-21i");
+MODULE_PARM_DESC(io, "EtherWORKS 3 I/O base address(es)");
+MODULE_PARM_DESC(irq, "EtherWORKS 3 IRQ number(s)");
 
 int init_module(void)
 {
-	thisEthwrk.base_addr = io;
-	thisEthwrk.irq = irq;
-	thisEthwrk.init = ewrk3_probe;
-	if (register_netdev(&thisEthwrk) != 0)
-		return -EIO;
-	return 0;
+	int i=0;
+
+	while( io[i] && irq[i] ) {
+		ewrk3_devs[ndevs] = kmalloc(sizeof(struct net_device), GFP_KERNEL);
+		if (!ewrk3_devs[ndevs])
+			goto error; 
+		memset(ewrk3_devs[ndevs], 0, sizeof(struct net_device));
+		ewrk3_devs[ndevs]->base_addr = io[i];
+		ewrk3_devs[ndevs]->irq = irq[i];
+		ewrk3_devs[ndevs]->init = ewrk3_probe;
+
+		if (register_netdev(ewrk3_devs[ndevs]) == 0)
+			ndevs++;
+		else
+			kfree(ewrk3_devs[ndevs]);
+
+		i++;
+	}
+
+	return ndevs ? 0 : -EIO;
+
+error:
+	cleanup_module();
+	return -ENOMEM;
 }
 
 void cleanup_module(void)
 {
-	unregister_netdev(&thisEthwrk);
-	if (thisEthwrk.priv) {
-		kfree(thisEthwrk.priv);
-		thisEthwrk.priv = NULL;
-	}
-	thisEthwrk.irq = 0;
+	int i;
 
-	release_region(thisEthwrk.base_addr, EWRK3_TOTAL_SIZE);
+	for( i=0; i<ndevs; i++ )
+	{
+		unregister_netdev(ewrk3_devs[i]);
+		if (ewrk3_devs[i]->priv) {
+			kfree(ewrk3_devs[i]->priv);
+			ewrk3_devs[i]->priv = NULL;
+		}
+		ewrk3_devs[i]->irq = 0;
+
+		release_region(ewrk3_devs[i]->base_addr, EWRK3_TOTAL_SIZE);
+		kfree(ewrk3_devs[i]);
+		ewrk3_devs[i] = NULL;
+	}
 }
 #endif				/* MODULE */
 MODULE_LICENSE("GPL");

--- linux-2.5.41/Documentation/networking/ewrk3.txt	Mon Mar 18 15:37:08 2002
+++ linux-2.5.41-fix2/Documentation/networking/ewrk3.txt	Wed Oct  9 21:44:24 2002
@@ -24,6 +24,7 @@
     kernel with the ewrk3 configuration turned off and reboot.
     5) insmod ewrk3.o
           [Alan Cox: Changed this so you can insmod ewrk3.o irq=x io=y]
+          [Adam Kropelin: Multiple cards now supported by irq=x1,x2 io=y1,y2]
     6) run the net startup bits for your new eth?? interface manually 
     (usually /etc/rc.inet[12] at boot time). 
     7) enjoy!

Re: Looking for testers with these NICs

[Denis Vlasenko]

On 9 October 2002 17:05, Derek Fawcus wrote:
> On Wed, Oct 09, 2002 at 07:31:17PM -0200, Denis Vlasenko wrote:
> > ni65.c
>
> I've got some of these knocking about,  but rather than use that
> driver, I have a (quite old) patch,  that allows the normal lance
> driver to be used.  The patch dates back from 1.3.x,  but I think I
> may have a more recent version around.

Well, if you want, give this a spin. Can you test it on
SMP or preempt kernel?

diff -u --recursive linux-2.5.40org/drivers/net/ni65.c linux-2.5.40/drivers/net/ni65.c
--- linux-2.5.40org/drivers/net/ni65.c	Thu Oct  3 12:08:00 2002
+++ linux-2.5.40/drivers/net/ni65.c	Wed Oct  9 10:35:18 2002
@@ -176,6 +176,9 @@
 #define writedatareg(val) { writereg(val,CSR0); }
 #endif

+/* Not to be confused with priv->lock */
+static spinlock_t irq_lock = SPIN_LOCK_UNLOCKED;
+
 static unsigned char ni_vendor[] = { 0x02,0x07,0x01 };

 static struct card {
@@ -409,7 +412,7 @@
 		p->features = 0x0;
 	}

-	if(test_bit(0,&cards[i].config)) {
+	if(test_bit(0,(unsigned long*)(&cards[i].config))) {
 		dev->irq = irqtab[(inw(ioaddr+L_CONFIG)>>2)&3];
 		dev->dma = dmatab[inw(ioaddr+L_CONFIG)&3];
 		printk("IRQ %d (from card), DMA %d (from card).\n",dev->irq,dev->dma);
@@ -420,7 +423,7 @@
 			int dma_channels = ((inb(DMA1_STAT_REG) >> 4) & 0x0f) | (inb(DMA2_STAT_REG) & 0xf0);
 			for(i=1;i<5;i++) {
 				int dma = dmatab[i];
-				if(test_bit(dma,&dma_channels) || request_dma(dma,"ni6510"))
+				if(test_bit(dma,(unsigned long*)&dma_channels) || request_dma(dma,"ni6510"))
 					continue;

 				flags=claim_dma_lock();
@@ -1118,8 +1121,7 @@
 							 (skb->len > T_BUF_SIZE) ? T_BUF_SIZE : skb->len);
 			dev_kfree_skb (skb);

-			save_flags(flags);
-			cli();
+			spin_lock_irqsave(&irq_lock, flags);

 			tmdp = p->tmdhead + p->tmdnum;
 			tmdp->u.buffer = (u32) isa_virt_to_bus(p->tmdbounce[p->tmdbouncenum]);
@@ -1128,8 +1130,7 @@
 #ifdef XMT_VIA_SKB
 		}
 		else {
-			save_flags(flags);
-			cli();
+			spin_lock_irqsave(&irq_lock, flags);

 			tmdp = p->tmdhead + p->tmdnum;
 			tmdp->u.buffer = (u32) isa_virt_to_bus(skb->data);
@@ -1150,7 +1151,7 @@
 		p->lock = 0;
 		dev->trans_start = jiffies;

-		restore_flags(flags);
+		spin_unlock_irqrestore(&irq_lock, flags);
 	}
 
 	return 0;

Re: Looking for testers with these NICs

[Denis Vlasenko]

On 9 October 2002 15:57, Martin Josefsson wrote:
> On Thu, 2002-10-10 at 00:37, Denis Vlasenko wrote:
> > I'd suggest SMP/preempt heavy IO. Is there stress test software for
> > NICs? What is pktgen?
>
> A kernelmodule that's capable of generating over a million packets
> per second on a fast enough machine. It stresstests NIC's and
> NIC-drivers a lot.

Indeed, I see it in 2.4, but where it has gone from 2.5?!
--
vda

Re: Looking for testers with these NICs

[Adam Kropelin]

On Wed, Oct 09, 2002 at 08:37:48PM -0200, Denis Vlasenko wrote:
> On 9 October 2002 15:14, Adam Kropelin wrote:
> > On Wed, Oct 09, 2002 at 07:31:17PM -0200, Denis Vlasenko wrote:
> > > ewrk3.c
> >
> > I've got a few of these laying around. Send whatever patches you want
> > tested and I'll give it a shot.
> 
> Please do your best in trying to break it, especially since you say you have
> more than one. Can you plug them all in one box?
> 
> I'd suggest SMP/preempt heavy IO. Is there stress test software for NICs?

I've finished beating the heck out of this driver. Over 12 hours of pounding
simultaneously on three NICs in a 2x SMP box running with preempt enabled and
not a single oops, BUG(), or deadlock. I'd say the driver is pretty solid at
this point; vda's locking patches seem to be safe. 

As a sidenote, the max throughput I was able to achieve across three cards was
about 1.4 MBytes/sec. A single card could do about 800 KBytes sec; 2 together
got to 1.2 MBytes/sec. Heavy CPU utilization the whole way, of course, since
these cards do not use DMA. 

--Adam