[PATCH 0/2] Two more pasemi_mac patches for 2.6.26

Hi Jeff,

The below two patches go on top of the previous patches that were acked
by you but applied to pasemi.git for feed up through powerpc.git.

One is a couple of fixes for various corner cases in the jumbo support
configuration. The second is addition of netpoll support, from Nate
Case.

Review/ack would be appreciated, I'll feed them up the same path as the
other changes.


-Olof

Fix a couple of corner cases around interface up/down when jumbo frames are
configured. Resources weren't always freed and reallocated properly.
    
Signed-off-by: Olof Johansson <redacted>

diff --git a/drivers/net/pasemi_mac.c b/drivers/net/pasemi_mac.c
index c50f0f4..abb1dc4 100644
--- a/drivers/net/pasemi_mac.c
+++ b/drivers/net/pasemi_mac.c

@@ -404,6 +404,7 @@ static void pasemi_mac_free_csring(struct pasemi_mac_csring *csring)
 	pasemi_dma_free_flag(csring->events[1]);
 	pasemi_dma_free_ring(&csring->chan);
 	pasemi_dma_free_chan(&csring->chan);
+	pasemi_dma_free_fun(csring->fun);
 }
 
 static int pasemi_mac_setup_rx_resources(const struct net_device *dev)

@@ -1150,7 +1151,10 @@ static int pasemi_mac_open(struct net_device *dev)
 	if (!mac->tx)
 		goto out_tx_ring;
 
-	if (dev->mtu > 1500) {
+	/* We might already have allocated rings in case mtu was changed
+	 * before interface was brought up.
+	 */
+	if (dev->mtu > 1500 && !mac->num_cs) {
 		pasemi_mac_setup_csrings(mac);
 		if (!mac->num_cs)
 			goto out_tx_ring;

@@ -1388,8 +1392,12 @@ static int pasemi_mac_close(struct net_device *dev)
 	free_irq(mac->tx->chan.irq, mac->tx);
 	free_irq(mac->rx->chan.irq, mac->rx);
 
-	for (i = 0; i < mac->num_cs; i++)
+	for (i = 0; i < mac->num_cs; i++) {
 		pasemi_mac_free_csring(mac->cs[i]);
+		mac->cs[i] = NULL;
+	}
+
+	mac->num_cs = 0;
 
 	/* Free resources */
 	pasemi_mac_free_rx_resources(mac);

Add netpoll support to allow use of netconsole.

Signed-off-by: Nate Case <redacted>
Signed-off-by: Olof Johansson <redacted>

diff --git a/drivers/net/pasemi_mac.c b/drivers/net/pasemi_mac.c
index abb1dc4..6030ffe 100644
--- a/drivers/net/pasemi_mac.c
+++ b/drivers/net/pasemi_mac.c

@@ -1648,6 +1648,26 @@ static int pasemi_mac_poll(struct napi_struct *napi, int budget)
 	return pkts;
 }
 
+#ifdef CONFIG_NET_POLL_CONTROLLER
+/*
+ * Polling 'interrupt' - used by things like netconsole to send skbs
+ * without having to re-enable interrupts. It's not called while
+ * the interrupt routine is executing.
+ */
+static void pasemi_mac_netpoll(struct net_device *dev)
+{
+	const struct pasemi_mac *mac = netdev_priv(dev);
+
+	disable_irq(mac->tx->chan.irq);
+	pasemi_mac_tx_intr(mac->tx->chan.irq, mac->tx);
+	enable_irq(mac->tx->chan.irq);
+
+	disable_irq(mac->rx->chan.irq);
+	pasemi_mac_rx_intr(mac->rx->chan.irq, dev);
+	enable_irq(mac->rx->chan.irq);
+}
+#endif
+
 static int pasemi_mac_change_mtu(struct net_device *dev, int new_mtu)
 {
 	struct pasemi_mac *mac = netdev_priv(dev);

@@ -1807,6 +1827,9 @@ pasemi_mac_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	dev->mtu = PE_DEF_MTU;
 	/* 1500 MTU + ETH_HLEN + VLAN_HLEN + 2 64B cachelines */
 	mac->bufsz = dev->mtu + ETH_HLEN + ETH_FCS_LEN + LOCAL_SKB_ALIGN + 128;
+#ifdef CONFIG_NET_POLL_CONTROLLER
+	dev->poll_controller = pasemi_mac_netpoll;
+#endif
 
 	dev->change_mtu = pasemi_mac_change_mtu;
 	dev->ethtool_ops = &pasemi_mac_ethtool_ops;

Olof Johansson wrote:

Hi Jeff,

The below two patches go on top of the previous patches that were acked
by you but applied to pasemi.git for feed up through powerpc.git.

One is a couple of fixes for various corner cases in the jumbo support
configuration. The second is addition of netpoll support, from Nate
Case.

Review/ack would be appreciated, I'll feed them up the same path as the
other changes.

ACK 1-2, though I would consider closely patch #2, whether you need disable_irq() or spin_lock_irqsave() -- which is best for your specific driver + platform?

Olof Johansson wrote:

Shouldn't this actually be pasemi_mac_rx_intr(mac->rx->chan.irq, mac->rx)?
Thanks,
Valentine.

_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-dev

On Wed, Mar 26, 2008 at 03:41:01PM +0300, Valentine Barshak wrote:

quoted

+	pasemi_mac_tx_intr(mac->tx->chan.irq, mac->tx);
+	enable_irq(mac->tx->chan.irq);
+
+	disable_irq(mac->rx->chan.irq);
+	pasemi_mac_rx_intr(mac->rx->chan.irq, dev);

Shouldn't this actually be pasemi_mac_rx_intr(mac->rx->chan.irq, mac->rx)?

Yeah, it should. It used to take the netdev pointer instead. And the
void * argument means I never got a compiler warning from it.

Thanks!

-Olof

On Wed, Mar 26, 2008 at 12:12:30AM -0400, Jeff Garzik wrote:

Olof Johansson wrote:

quoted

Hi Jeff,

The below two patches go on top of the previous patches that were acked
by you but applied to pasemi.git for feed up through powerpc.git.

One is a couple of fixes for various corner cases in the jumbo support
configuration. The second is addition of netpoll support, from Nate
Case.

Review/ack would be appreciated, I'll feed them up the same path as the
other changes.

ACK 1-2, though I would consider closely patch #2, whether you need 
disable_irq() or spin_lock_irqsave() -- which is best for your specific 
driver + platform?

I'm tempted to leave the disable/enable_irq() in there. It's certainly
heavier than doing cpu-side irq masking, but adding a spinlock to the
common interrupt path seems worse to me. I didn't think we considered
the polling to be a hot path anyway...


-Olof

Olof Johansson wrote:

On Wed, Mar 26, 2008 at 12:12:30AM -0400, Jeff Garzik wrote:

quoted

Olof Johansson wrote:

quoted

Hi Jeff,

The below two patches go on top of the previous patches that were acked
by you but applied to pasemi.git for feed up through powerpc.git.

One is a couple of fixes for various corner cases in the jumbo support
configuration. The second is addition of netpoll support, from Nate
Case.

Review/ack would be appreciated, I'll feed them up the same path as the
other changes.

ACK 1-2, though I would consider closely patch #2, whether you need disable_irq() or spin_lock_irqsave() -- which is best for your specific driver + platform?

I'm tempted to leave the disable/enable_irq() in there. It's certainly
heavier than doing cpu-side irq masking, but adding a spinlock to the
common interrupt path seems worse to me. I didn't think we considered
the polling to be a hot path anyway...

Definitely not a hot path :)

Add netpoll support to allow use of netconsole.
    
Signed-off-by: Nate Case <redacted>
Signed-off-by: Olof Johansson <redacted>

---
Changed the interrupt handler arguments as found by Valentine. Will push
through powerpc.git with Jeff's ACKs (that were already given).


-Olof

diff --git a/drivers/net/pasemi_mac.c b/drivers/net/pasemi_mac.c
index abb1dc4..965f2e4 100644
--- a/drivers/net/pasemi_mac.c
+++ b/drivers/net/pasemi_mac.c

@@ -1648,6 +1648,26 @@ static int pasemi_mac_poll(struct napi_struct *napi, int budget)
 	return pkts;
 }
 
+#ifdef CONFIG_NET_POLL_CONTROLLER
+/*
+ * Polling 'interrupt' - used by things like netconsole to send skbs
+ * without having to re-enable interrupts. It's not called while
+ * the interrupt routine is executing.
+ */
+static void pasemi_mac_netpoll(struct net_device *dev)
+{
+	const struct pasemi_mac *mac = netdev_priv(dev);
+
+	disable_irq(mac->tx->chan.irq);
+	pasemi_mac_tx_intr(mac->tx->chan.irq, mac->tx);
+	enable_irq(mac->tx->chan.irq);
+
+	disable_irq(mac->rx->chan.irq);
+	pasemi_mac_rx_intr(mac->rx->chan.irq, mac->rx);
+	enable_irq(mac->rx->chan.irq);
+}
+#endif
+
 static int pasemi_mac_change_mtu(struct net_device *dev, int new_mtu)
 {
 	struct pasemi_mac *mac = netdev_priv(dev);

@@ -1807,6 +1827,9 @@ pasemi_mac_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 	dev->mtu = PE_DEF_MTU;
 	/* 1500 MTU + ETH_HLEN + VLAN_HLEN + 2 64B cachelines */
 	mac->bufsz = dev->mtu + ETH_HLEN + ETH_FCS_LEN + LOCAL_SKB_ALIGN + 128;
+#ifdef CONFIG_NET_POLL_CONTROLLER
+	dev->poll_controller = pasemi_mac_netpoll;
+#endif
 
 	dev->change_mtu = pasemi_mac_change_mtu;
 	dev->ethtool_ops = &pasemi_mac_ethtool_ops;

Olof Johansson wrote:

Add netpoll support to allow use of netconsole.
    
Signed-off-by: Nate Case <redacted>
Signed-off-by: Olof Johansson <redacted>

---
Changed the interrupt handler arguments as found by Valentine. Will push
through powerpc.git with Jeff's ACKs (that were already given).


-Olof

ACK