[2/4] pollcontroller patch for 2.6.0-test10-bk25-netdrvr-exp1

[2/4] pollcontroller patch for 2.6.0-test10-bk25-netdrvr-exp1

[Prasanna S Panchamukhi]

Hi Jeff,

Below is the pollcontroller patch for e1000.
This patch can be applied over 2.6.0-test9-bk25-netdrvr-exp1.patch

diff -urNp linux-netdrvr/drivers/net/e1000/e1000_main.c linux-2.6.0-test10/drivers/net/e1000/e1000_main.c
--- linux-netdrvr/drivers/net/e1000/e1000_main.c	2003-12-04 11:18:03.000000000 +0530
+++ linux-2.6.0-test10/drivers/net/e1000/e1000_main.c	2003-12-04 11:43:35.412199928 +0530
@@ -338,6 +338,15 @@ e1000_reset(struct e1000_adapter *adapte
 	e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
 }
 
+#ifdef CONFIG_NET_POLL_CONTROLLER
+static void e1000_poll(struct net_device *dev)
+{
+	disable_irq(dev->irq);
+	e1000_intr(dev->irq, dev, NULL);
+	enable_irq(dev->irq);
+}
+#endif
+
 /**
  * e1000_probe - Device Initialization Routine
  * @pdev: PCI device information struct
@@ -439,6 +448,9 @@ e1000_probe(struct pci_dev *pdev,
 
 	adapter->bd_number = cards_found;
 
+#ifdef CONFIG_NET_POLL_CONTROLLER
+	netdev->poll_controller = &e1000_poll;
+#endif
 	/* setup the private structure */
 
 	if((err = e1000_sw_init(adapter)))
-- 
Thanks & Regards
Prasanna S Panchamukhi
Linux Technology Center
India Software Labs, IBM Bangalore
Ph: 91-80-5044632

Re: [2/4] pollcontroller patch for 2.6.0-test10-bk25-netdrvr-exp1

[Jeff Garzik]

I'm applying smc-ultra and tlan patches, but holding off on e100 and e1000.

For both, as Scott mentioned, they don't look tested under NAPI.  For 
e100 specifically, there is a spiffy new e100 that would need to be 
re-diffed and tested against.

	Jeff

RE: [2/4] pollcontroller patch for 2.6.0-test10-bk25-netdrvr-exp1

[Feldman, Scott]

> Below is the pollcontroller patch for e1000.
> This patch can be applied over 2.6.0-test9-bk25-netdrvr-exp1.patch

Did you test this w/ and w/o CONFIG_E1000_NAPI?

-scott

RE: [2/4] pollcontroller patch for 2.6.0-test10-bk25-netdrvr-exp1

[Feldman, Scott]

> For both, as Scott mentioned, they don't look tested under NAPI.  For 
> e100 specifically, there is a spiffy new e100 that would need to be 
> re-diffed and tested against.

Specifically, we need to care about insuring netif_rx is called in the
poll_controller callback, rather than netif_receive_skb.  If you use
netif_receive_skb (NAPI mode), the netdump conversation is one-sided.
tg3.c looks broken in this regard in the 2.6-exp BK tree, BTW.

Additionally, if you have VLANs enabled on the interface, we might need
to care about netif_hwaccell_[rx|receive_skb].  Something tells me all
of the netif_* receive funcs need to handle the netdump conversation.
That would be easier than special casing poll_controller in each driver.
Is someone working on this?

We're working on e100/e1000 patches for the current scheme, so we'll
post these when we've tested the various NAPI/VLAN combinations.

-scott

RE: [2/4] pollcontroller patch for 2.6.0-test10-bk25-netdrvr-exp1

[Feldman, Scott]

> Additionally, if you have VLANs enabled on the interface, we 
> might need to care about netif_hwaccell_[rx|receive_skb].  
> Something tells me all of the netif_* receive funcs need to 
> handle the netdump conversation. That would be easier than 
> special casing poll_controller in each driver. Is someone 
> working on this?

Ok, nevermind, we looked closer at lkcd/netpoll patches and it looks
like this is addressed.

-scott