[PATCH] New PowerPC 4xx on-chip ethernet controller driver

This patch replaces current PowerPC 4xx EMAC driver with
new, re-written from the scratch version. This patch is quite big
(~234K) because there is virtualy 0% of common code between old and
new version.

New driver uses NAPI, it solves stability problems under heavy packet
load and low memory, corrects chip register access and fixes numerous
small bugs I don't even remember now.

This patch has been tested on all supported in 2.6 PPC 4xx boards.
It's been used in production for almost a year now on custom
4xx hardware. PPC32 specific parts are already upstream.

Patch was acked by the current EMAC driver maintainer (Matt Porter). I
will be maintaining this new version.

Signed-off-by: Eugene Surovegin <ebs@ebshome.net>
--

 Kconfig                   |   72
 ibm_emac/Makefile         |   13
 ibm_emac/ibm_emac.h       |  418 +++--
 ibm_emac/ibm_emac_core.c  | 3414 ++++++++++++++++++++++++----------------------
 ibm_emac/ibm_emac_core.h  |  313 ++--
 ibm_emac/ibm_emac_debug.c |  377 ++---
 ibm_emac/ibm_emac_debug.h |   63
 ibm_emac/ibm_emac_mal.c   |  674 +++++----
 ibm_emac/ibm_emac_mal.h   |  336 +++-
 ibm_emac/ibm_emac_phy.c   |  335 ++--
 ibm_emac/ibm_emac_phy.h   |  105 -
 ibm_emac/ibm_emac_rgmii.c |  201 ++
 ibm_emac/ibm_emac_rgmii.h |   68
 ibm_emac/ibm_emac_tah.c   |  111 +
 ibm_emac/ibm_emac_tah.h   |   96 -
 ibm_emac/ibm_emac_zmii.c  |  255 +++
 ibm_emac/ibm_emac_zmii.h  |  114 -
 17 files changed, 4114 insertions(+), 2851 deletions(-)
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>

1 files changed, 397 insertions, 277 deletions

diff --git a/drivers/net/ibm_emac/ibm_emac_mal.c b/drivers/net/ibm_emac/ibm_emac_mal.c
index e59f57f363c..da88d43081c 100644
--- a/drivers/net/ibm_emac/ibm_emac_mal.c
+++ b/drivers/net/ibm_emac/ibm_emac_mal.c
@@ -1,436 +1,565 @@
 /*
- * ibm_ocp_mal.c
+ * drivers/net/ibm_emac/ibm_emac_mal.c
  *
- *      Armin Kuster akuster@mvista.com
- *      Juen, 2002
+ * Memory Access Layer (MAL) support
+ * 
+ * Copyright (c) 2004, 2005 Zultys Technologies.
+ * Eugene Surovegin <eugene.surovegin@zultys.com> or <ebs@ebshome.net>
  *
- * Copyright 2002 MontaVista Softare Inc.
+ * Based on original work by
+ *      Benjamin Herrenschmidt <benh@kernel.crashing.org>,
+ *      David Gibson <hermes@gibson.dropbear.id.au>,
+ *
+ *      Armin Kuster <akuster@mvista.com>
+ *      Copyright 2002 MontaVista Softare Inc.
  *
  * This program is free software; you can redistribute  it and/or modify it
  * under  the terms of  the GNU General  Public License as published by the
  * Free Software Foundation;  either version 2 of the  License, or (at your
  * option) any later version.
+ *
  */
-
 #include <linux/config.h>
 #include <linux/module.h>
 #include <linux/kernel.h>
 #include <linux/errno.h>
 #include <linux/netdevice.h>
 #include <linux/init.h>
+#include <linux/interrupt.h>
 #include <linux/dma-mapping.h>
 
-#include <asm/io.h>
-#include <asm/irq.h>
 #include <asm/ocp.h>
 
+#include "ibm_emac_core.h"
 #include "ibm_emac_mal.h"
+#include "ibm_emac_debug.h"
 
-// Locking: Should we share a lock with the client ? The client could provide
-// a lock pointer (optionally) in the commac structure... I don't think this is
-// really necessary though
-
-/* This lock protects the commac list. On today UP implementations, it's
- * really only used as IRQ protection in mal_{register,unregister}_commac()
- */
-static DEFINE_RWLOCK(mal_list_lock);
-
-int mal_register_commac(struct ibm_ocp_mal *mal, struct mal_commac *commac)
+int __init mal_register_commac(struct ibm_ocp_mal *mal,
+			       struct mal_commac *commac)
 {
 	unsigned long flags;
+	local_irq_save(flags);
 
-	write_lock_irqsave(&mal_list_lock, flags);
+	MAL_DBG("%d: reg(%08x, %08x)" NL, mal->def->index,
+		commac->tx_chan_mask, commac->rx_chan_mask);
 
-	/* Don't let multiple commacs claim the same channel */
+	/* Don't let multiple commacs claim the same channel(s) */
 	if ((mal->tx_chan_mask & commac->tx_chan_mask) ||
 	    (mal->rx_chan_mask & commac->rx_chan_mask)) {
-		write_unlock_irqrestore(&mal_list_lock, flags);
+		local_irq_restore(flags);
+		printk(KERN_WARNING "mal%d: COMMAC channels conflict!\n",
+		       mal->def->index);
 		return -EBUSY;
 	}
 
 	mal->tx_chan_mask |= commac->tx_chan_mask;
 	mal->rx_chan_mask |= commac->rx_chan_mask;
+	list_add(&commac->list, &mal->list);
 
-	list_add(&commac->list, &mal->commac);
-
-	write_unlock_irqrestore(&mal_list_lock, flags);
-
+	local_irq_restore(flags);
 	return 0;
 }
 
-int mal_unregister_commac(struct ibm_ocp_mal *mal, struct mal_commac *commac)
+void __exit mal_unregister_commac(struct ibm_ocp_mal *mal,
+				  struct mal_commac *commac)
 {
 	unsigned long flags;
+	local_irq_save(flags);
 
-	write_lock_irqsave(&mal_list_lock, flags);
+	MAL_DBG("%d: unreg(%08x, %08x)" NL, mal->def->index,
+		commac->tx_chan_mask, commac->rx_chan_mask);
 
 	mal->tx_chan_mask &= ~commac->tx_chan_mask;
 	mal->rx_chan_mask &= ~commac->rx_chan_mask;
-
 	list_del_init(&commac->list);
 
-	write_unlock_irqrestore(&mal_list_lock, flags);
-
-	return 0;
+	local_irq_restore(flags);
 }
 
 int mal_set_rcbs(struct ibm_ocp_mal *mal, int channel, unsigned long size)
 {
-	switch (channel) {
-	case 0:
-		set_mal_dcrn(mal, DCRN_MALRCBS0, size);
-		break;
-#ifdef DCRN_MALRCBS1
-	case 1:
-		set_mal_dcrn(mal, DCRN_MALRCBS1, size);
-		break;
-#endif
-#ifdef DCRN_MALRCBS2
-	case 2:
-		set_mal_dcrn(mal, DCRN_MALRCBS2, size);
-		break;
-#endif
-#ifdef DCRN_MALRCBS3
-	case 3:
-		set_mal_dcrn(mal, DCRN_MALRCBS3, size);
-		break;
-#endif
-	default:
+	struct ocp_func_mal_data *maldata = mal->def->additions;
+	BUG_ON(channel < 0 || channel >= maldata->num_rx_chans ||
+	       size > MAL_MAX_RX_SIZE);
+
+	MAL_DBG("%d: set_rbcs(%d, %lu)" NL, mal->def->index, channel, size);
+
+	if (size & 0xf) {
+		printk(KERN_WARNING
+		       "mal%d: incorrect RX size %lu for the channel %d\n",
+		       mal->def->index, size, channel);
 		return -EINVAL;
 	}
 
+	set_mal_dcrn(mal, MAL_RCBS(channel), size >> 4);
 	return 0;
 }
 
-static irqreturn_t mal_serr(int irq, void *dev_instance, struct pt_regs *regs)
+int mal_tx_bd_offset(struct ibm_ocp_mal *mal, int channel)
 {
-	struct ibm_ocp_mal *mal = dev_instance;
-	unsigned long mal_error;
+	struct ocp_func_mal_data *maldata = mal->def->additions;
+	BUG_ON(channel < 0 || channel >= maldata->num_tx_chans);
+	return channel * NUM_TX_BUFF;
+}
 
-	/*
-	 * This SERR applies to one of the devices on the MAL, here we charge
-	 * it against the first EMAC registered for the MAL.
-	 */
+int mal_rx_bd_offset(struct ibm_ocp_mal *mal, int channel)
+{
+	struct ocp_func_mal_data *maldata = mal->def->additions;
+	BUG_ON(channel < 0 || channel >= maldata->num_rx_chans);
+	return maldata->num_tx_chans * NUM_TX_BUFF + channel * NUM_RX_BUFF;
+}
 
-	mal_error = get_mal_dcrn(mal, DCRN_MALESR);
+void mal_enable_tx_channel(struct ibm_ocp_mal *mal, int channel)
+{
+	local_bh_disable();
+	MAL_DBG("%d: enable_tx(%d)" NL, mal->def->index, channel);
+	set_mal_dcrn(mal, MAL_TXCASR,
+		     get_mal_dcrn(mal, MAL_TXCASR) | MAL_CHAN_MASK(channel));
+	local_bh_enable();
+}
 
-	printk(KERN_ERR "%s: System Error (MALESR=%lx)\n",
-	       "MAL" /* FIXME: get the name right */ , mal_error);
+void mal_disable_tx_channel(struct ibm_ocp_mal *mal, int channel)
+{
+	set_mal_dcrn(mal, MAL_TXCARR, MAL_CHAN_MASK(channel));
+	MAL_DBG("%d: disable_tx(%d)" NL, mal->def->index, channel);
+}
 
-	/* FIXME: decipher error */
-	/* DIXME: distribute to commacs, if possible */
+void mal_enable_rx_channel(struct ibm_ocp_mal *mal, int channel)
+{
+	local_bh_disable();
+	MAL_DBG("%d: enable_rx(%d)" NL, mal->def->index, channel);
+	set_mal_dcrn(mal, MAL_RXCASR,
+		     get_mal_dcrn(mal, MAL_RXCASR) | MAL_CHAN_MASK(channel));
+	local_bh_enable();
+}
 
-	/* Clear the error status register */
-	set_mal_dcrn(mal, DCRN_MALESR, mal_error);
+void mal_disable_rx_channel(struct ibm_ocp_mal *mal, int channel)
+{
+	set_mal_dcrn(mal, MAL_RXCARR, MAL_CHAN_MASK(channel));
+	MAL_DBG("%d: disable_rx(%d)" NL, mal->def->index, channel);
+}
 
-	return IRQ_HANDLED;
+void mal_poll_add(struct ibm_ocp_mal *mal, struct mal_commac *commac)
+{
+	local_bh_disable();
+	MAL_DBG("%d: poll_add(%p)" NL, mal->def->index, commac);
+	list_add_tail(&commac->poll_list, &mal->poll_list);
+	local_bh_enable();
 }
 
-static irqreturn_t mal_txeob(int irq, void *dev_instance, struct pt_regs *regs)
+void mal_poll_del(struct ibm_ocp_mal *mal, struct mal_commac *commac)
+{
+	local_bh_disable();
+	MAL_DBG("%d: poll_del(%p)" NL, mal->def->index, commac);
+	list_del(&commac->poll_list);
+	local_bh_enable();
+}
+
+/* synchronized by mal_poll() */
+static inline void mal_enable_eob_irq(struct ibm_ocp_mal *mal)
+{
+	MAL_DBG2("%d: enable_irq" NL, mal->def->index);
+	set_mal_dcrn(mal, MAL_CFG, get_mal_dcrn(mal, MAL_CFG) | MAL_CFG_EOPIE);
+}
+
+/* synchronized by __LINK_STATE_RX_SCHED bit in ndev->state */
+static inline void mal_disable_eob_irq(struct ibm_ocp_mal *mal)
+{
+	set_mal_dcrn(mal, MAL_CFG, get_mal_dcrn(mal, MAL_CFG) & ~MAL_CFG_EOPIE);
+	MAL_DBG2("%d: disable_irq" NL, mal->def->index);
+}
+
+static irqreturn_t mal_serr(int irq, void *dev_instance, struct pt_regs *regs)
 {
 	struct ibm_ocp_mal *mal = dev_instance;
-	struct list_head *l;
-	unsigned long isr;
+	u32 esr = get_mal_dcrn(mal, MAL_ESR);
 
-	isr = get_mal_dcrn(mal, DCRN_MALTXEOBISR);
-	set_mal_dcrn(mal, DCRN_MALTXEOBISR, isr);
+	/* Clear the error status register */
+	set_mal_dcrn(mal, MAL_ESR, esr);
 
-	read_lock(&mal_list_lock);
-	list_for_each(l, &mal->commac) {
-		struct mal_commac *mc = list_entry(l, struct mal_commac, list);
+	MAL_DBG("%d: SERR %08x" NL, mal->def->index, esr);
 
-		if (isr & mc->tx_chan_mask) {
-			mc->ops->txeob(mc->dev, isr & mc->tx_chan_mask);
+	if (esr & MAL_ESR_EVB) {
+		if (esr & MAL_ESR_DE) {
+			/* We ignore Descriptor error,
+			 * TXDE or RXDE interrupt will be generated anyway.
+			 */
+			return IRQ_HANDLED;
 		}
+
+		if (esr & MAL_ESR_PEIN) {
+			/* PLB error, it's probably buggy hardware or
+			 * incorrect physical address in BD (i.e. bug)
+			 */
+			if (net_ratelimit())
+				printk(KERN_ERR
+				       "mal%d: system error, PLB (ESR = 0x%08x)\n",
+				       mal->def->index, esr);
+			return IRQ_HANDLED;
+		}
+
+		/* OPB error, it's probably buggy hardware or incorrect EBC setup */
+		if (net_ratelimit())
+			printk(KERN_ERR
+			       "mal%d: system error, OPB (ESR = 0x%08x)\n",
+			       mal->def->index, esr);
 	}
-	read_unlock(&mal_list_lock);
+	return IRQ_HANDLED;
+}
+
+static inline void mal_schedule_poll(struct ibm_ocp_mal *mal)
+{
+	if (likely(netif_rx_schedule_prep(&mal->poll_dev))) {
+		MAL_DBG2("%d: schedule_poll" NL, mal->def->index);
+		mal_disable_eob_irq(mal);
+		__netif_rx_schedule(&mal->poll_dev);
+	} else
+		MAL_DBG2("%d: already in poll" NL, mal->def->index);
+}
 
+static irqreturn_t mal_txeob(int irq, void *dev_instance, struct pt_regs *regs)
+{
+	struct ibm_ocp_mal *mal = dev_instance;
+	u32 r = get_mal_dcrn(mal, MAL_TXEOBISR);
+	MAL_DBG2("%d: txeob %08x" NL, mal->def->index, r);
+	mal_schedule_poll(mal);
+	set_mal_dcrn(mal, MAL_TXEOBISR, r);
 	return IRQ_HANDLED;
 }
 
 static irqreturn_t mal_rxeob(int irq, void *dev_instance, struct pt_regs *regs)
 {
 	struct ibm_ocp_mal *mal = dev_instance;
-	struct list_head *l;
-	unsigned long isr;
+	u32 r = get_mal_dcrn(mal, MAL_RXEOBISR);
+	MAL_DBG2("%d: rxeob %08x" NL, mal->def->index, r);
+	mal_schedule_poll(mal);
+	set_mal_dcrn(mal, MAL_RXEOBISR, r);
+	return IRQ_HANDLED;
+}
 
-	isr = get_mal_dcrn(mal, DCRN_MALRXEOBISR);
-	set_mal_dcrn(mal, DCRN_MALRXEOBISR, isr);
+static irqreturn_t mal_txde(int irq, void *dev_instance, struct pt_regs *regs)
+{
+	struct ibm_ocp_mal *mal = dev_instance;
+	u32 deir = get_mal_dcrn(mal, MAL_TXDEIR);
+	set_mal_dcrn(mal, MAL_TXDEIR, deir);
 
-	read_lock(&mal_list_lock);
-	list_for_each(l, &mal->commac) {
-		struct mal_commac *mc = list_entry(l, struct mal_commac, list);
+	MAL_DBG("%d: txde %08x" NL, mal->def->index, deir);
 
-		if (isr & mc->rx_chan_mask) {
-			mc->ops->rxeob(mc->dev, isr & mc->rx_chan_mask);
-		}
-	}
-	read_unlock(&mal_list_lock);
+	if (net_ratelimit())
+		printk(KERN_ERR
+		       "mal%d: TX descriptor error (TXDEIR = 0x%08x)\n",
+		       mal->def->index, deir);
 
 	return IRQ_HANDLED;
 }
 
-static irqreturn_t mal_txde(int irq, void *dev_instance, struct pt_regs *regs)
+static irqreturn_t mal_rxde(int irq, void *dev_instance, struct pt_regs *regs)
 {
 	struct ibm_ocp_mal *mal = dev_instance;
 	struct list_head *l;
-	unsigned long deir;
+	u32 deir = get_mal_dcrn(mal, MAL_RXDEIR);
 
-	deir = get_mal_dcrn(mal, DCRN_MALTXDEIR);
+	MAL_DBG("%d: rxde %08x" NL, mal->def->index, deir);
 
-	/* FIXME: print which MAL correctly */
-	printk(KERN_WARNING "%s: Tx descriptor error (MALTXDEIR=%lx)\n",
-	       "MAL", deir);
-
-	read_lock(&mal_list_lock);
-	list_for_each(l, &mal->commac) {
+	list_for_each(l, &mal->list) {
 		struct mal_commac *mc = list_entry(l, struct mal_commac, list);
-
-		if (deir & mc->tx_chan_mask) {
-			mc->ops->txde(mc->dev, deir & mc->tx_chan_mask);
+		if (deir & mc->rx_chan_mask) {
+			mc->rx_stopped = 1;
+			mc->ops->rxde(mc->dev);
 		}
 	}
-	read_unlock(&mal_list_lock);
+
+	mal_schedule_poll(mal);
+	set_mal_dcrn(mal, MAL_RXDEIR, deir);
 
 	return IRQ_HANDLED;
 }
 
-/*
- * This interrupt should be very rare at best.  This occurs when
- * the hardware has a problem with the receive descriptors.  The manual
- * states that it occurs when the hardware cannot the receive descriptor
- * empty bit is not set.  The recovery mechanism will be to
- * traverse through the descriptors, handle any that are marked to be
- * handled and reinitialize each along the way.  At that point the driver
- * will be restarted.
- */
-static irqreturn_t mal_rxde(int irq, void *dev_instance, struct pt_regs *regs)
+static int mal_poll(struct net_device *ndev, int *budget)
 {
-	struct ibm_ocp_mal *mal = dev_instance;
+	struct ibm_ocp_mal *mal = ndev->priv;
 	struct list_head *l;
-	unsigned long deir;
-
-	deir = get_mal_dcrn(mal, DCRN_MALRXDEIR);
+	int rx_work_limit = min(ndev->quota, *budget), received = 0, done;
+
+	MAL_DBG2("%d: poll(%d) %d ->" NL, mal->def->index, *budget,
+		 rx_work_limit);
+      again:
+	/* Process TX skbs */
+	list_for_each(l, &mal->poll_list) {
+		struct mal_commac *mc =
+		    list_entry(l, struct mal_commac, poll_list);
+		mc->ops->poll_tx(mc->dev);
+	}
 
-	/*
-	 * This really is needed.  This case encountered in stress testing.
+	/* Process RX skbs.
+	 * We _might_ need something more smart here to enforce polling fairness.
 	 */
-	if (deir == 0)
-		return IRQ_HANDLED;
-
-	/* FIXME: print which MAL correctly */
-	printk(KERN_WARNING "%s: Rx descriptor error (MALRXDEIR=%lx)\n",
-	       "MAL", deir);
-
-	read_lock(&mal_list_lock);
-	list_for_each(l, &mal->commac) {
-		struct mal_commac *mc = list_entry(l, struct mal_commac, list);
+	list_for_each(l, &mal->poll_list) {
+		struct mal_commac *mc =
+		    list_entry(l, struct mal_commac, poll_list);
+		int n = mc->ops->poll_rx(mc->dev, rx_work_limit);
+		if (n) {
+			received += n;
+			rx_work_limit -= n;
+			if (rx_work_limit <= 0) {
+				done = 0;
+				goto more_work;	// XXX What if this is the last one ?
+			}
+		}
+	}
 
-		if (deir & mc->rx_chan_mask) {
-			mc->ops->rxde(mc->dev, deir & mc->rx_chan_mask);
+	/* We need to disable IRQs to protect from RXDE IRQ here */
+	local_irq_disable();
+	__netif_rx_complete(ndev);
+	mal_enable_eob_irq(mal);
+	local_irq_enable();
+
+	done = 1;
+
+	/* Check for "rotting" packet(s) */
+	list_for_each(l, &mal->poll_list) {
+		struct mal_commac *mc =
+		    list_entry(l, struct mal_commac, poll_list);
+		if (unlikely(mc->ops->peek_rx(mc->dev) || mc->rx_stopped)) {
+			MAL_DBG2("%d: rotting packet" NL, mal->def->index);
+			if (netif_rx_reschedule(ndev, received))
+				mal_disable_eob_irq(mal);
+			else
+				MAL_DBG2("%d: already in poll list" NL,
+					 mal->def->index);
+
+			if (rx_work_limit > 0)
+				goto again;
+			else
+				goto more_work;
 		}
+		mc->ops->poll_tx(mc->dev);
 	}
-	read_unlock(&mal_list_lock);
 
-	return IRQ_HANDLED;
+      more_work:
+	ndev->quota -= received;
+	*budget -= received;
+
+	MAL_DBG2("%d: poll() %d <- %d" NL, mal->def->index, *budget,
+		 done ? 0 : 1);
+	return done ? 0 : 1;
+}
+
+static void mal_reset(struct ibm_ocp_mal *mal)
+{
+	int n = 10;
+	MAL_DBG("%d: reset" NL, mal->def->index);
+
+	set_mal_dcrn(mal, MAL_CFG, MAL_CFG_SR);
+
+	/* Wait for reset to complete (1 system clock) */
+	while ((get_mal_dcrn(mal, MAL_CFG) & MAL_CFG_SR) && n)
+		--n;
+
+	if (unlikely(!n))
+		printk(KERN_ERR "mal%d: reset timeout\n", mal->def->index);
+}
+
+int mal_get_regs_len(struct ibm_ocp_mal *mal)
+{
+	return sizeof(struct emac_ethtool_regs_subhdr) +
+	    sizeof(struct ibm_mal_regs);
+}
+
+void *mal_dump_regs(struct ibm_ocp_mal *mal, void *buf)
+{
+	struct emac_ethtool_regs_subhdr *hdr = buf;
+	struct ibm_mal_regs *regs = (struct ibm_mal_regs *)(hdr + 1);
+	struct ocp_func_mal_data *maldata = mal->def->additions;
+	int i;
+
+	hdr->version = MAL_VERSION;
+	hdr->index = mal->def->index;
+
+	regs->tx_count = maldata->num_tx_chans;
+	regs->rx_count = maldata->num_rx_chans;
+
+	regs->cfg = get_mal_dcrn(mal, MAL_CFG);
+	regs->esr = get_mal_dcrn(mal, MAL_ESR);
+	regs->ier = get_mal_dcrn(mal, MAL_IER);
+	regs->tx_casr = get_mal_dcrn(mal, MAL_TXCASR);
+	regs->tx_carr = get_mal_dcrn(mal, MAL_TXCARR);
+	regs->tx_eobisr = get_mal_dcrn(mal, MAL_TXEOBISR);
+	regs->tx_deir = get_mal_dcrn(mal, MAL_TXDEIR);
+	regs->rx_casr = get_mal_dcrn(mal, MAL_RXCASR);
+	regs->rx_carr = get_mal_dcrn(mal, MAL_RXCARR);
+	regs->rx_eobisr = get_mal_dcrn(mal, MAL_RXEOBISR);
+	regs->rx_deir = get_mal_dcrn(mal, MAL_RXDEIR);
+
+	for (i = 0; i < regs->tx_count; ++i)
+		regs->tx_ctpr[i] = get_mal_dcrn(mal, MAL_TXCTPR(i));
+
+	for (i = 0; i < regs->rx_count; ++i) {
+		regs->rx_ctpr[i] = get_mal_dcrn(mal, MAL_RXCTPR(i));
+		regs->rcbs[i] = get_mal_dcrn(mal, MAL_RCBS(i));
+	}
+	return regs + 1;
 }
 
 static int __init mal_probe(struct ocp_device *ocpdev)
 {
-	struct ibm_ocp_mal *mal = NULL;
+	struct ibm_ocp_mal *mal;
 	struct ocp_func_mal_data *maldata;
-	int err = 0;
+	int err = 0, i, bd_size;
+
+	MAL_DBG("%d: probe" NL, ocpdev->def->index);
 
-	maldata = (struct ocp_func_mal_data *)ocpdev->def->additions;
+	maldata = ocpdev->def->additions;
 	if (maldata == NULL) {
-		printk(KERN_ERR "mal%d: Missing additional datas !\n",
+		printk(KERN_ERR "mal%d: missing additional data!\n",
 		       ocpdev->def->index);
 		return -ENODEV;
 	}
 
-	mal = kmalloc(sizeof(struct ibm_ocp_mal), GFP_KERNEL);
-	if (mal == NULL) {
+	mal = kzalloc(sizeof(struct ibm_ocp_mal), GFP_KERNEL);
+	if (!mal) {
 		printk(KERN_ERR
-		       "mal%d: Out of memory allocating MAL structure !\n",
+		       "mal%d: out of memory allocating MAL structure!\n",
 		       ocpdev->def->index);
 		return -ENOMEM;
 	}
-	memset(mal, 0, sizeof(*mal));
-
-	switch (ocpdev->def->index) {
-	case 0:
-		mal->dcrbase = DCRN_MAL_BASE;
-		break;
-#ifdef DCRN_MAL1_BASE
-	case 1:
-		mal->dcrbase = DCRN_MAL1_BASE;
-		break;
-#endif
-	default:
-		BUG();
-	}
-
-	/**************************/
+	mal->dcrbase = maldata->dcr_base;
+	mal->def = ocpdev->def;
 
-	INIT_LIST_HEAD(&mal->commac);
+	INIT_LIST_HEAD(&mal->poll_list);
+	set_bit(__LINK_STATE_START, &mal->poll_dev.state);
+	mal->poll_dev.weight = CONFIG_IBM_EMAC_POLL_WEIGHT;
+	mal->poll_dev.poll = mal_poll;
+	mal->poll_dev.priv = mal;
+	atomic_set(&mal->poll_dev.refcnt, 1);
 
-	set_mal_dcrn(mal, DCRN_MALRXCARR, 0xFFFFFFFF);
-	set_mal_dcrn(mal, DCRN_MALTXCARR, 0xFFFFFFFF);
+	INIT_LIST_HEAD(&mal->list);
 
-	set_mal_dcrn(mal, DCRN_MALCR, MALCR_MMSR);	/* 384 */
-	/* FIXME: Add delay */
+	/* Load power-on reset defaults */
+	mal_reset(mal);
 
 	/* Set the MAL configuration register */
-	set_mal_dcrn(mal, DCRN_MALCR,
-		     MALCR_PLBB | MALCR_OPBBL | MALCR_LEA |
-		     MALCR_PLBLT_DEFAULT);
-
-	/* It would be nice to allocate buffers separately for each
-	 * channel, but we can't because the channels share the upper
-	 * 13 bits of address lines.  Each channels buffer must also
-	 * be 4k aligned, so we allocate 4k for each channel.  This is
-	 * inefficient FIXME: do better, if possible */
-	mal->tx_virt_addr = dma_alloc_coherent(&ocpdev->dev,
-					       MAL_DT_ALIGN *
-					       maldata->num_tx_chans,
-					       &mal->tx_phys_addr, GFP_KERNEL);
-	if (mal->tx_virt_addr == NULL) {
+	set_mal_dcrn(mal, MAL_CFG, MAL_CFG_DEFAULT | MAL_CFG_PLBB |
+		     MAL_CFG_OPBBL | MAL_CFG_LEA);
+
+	mal_enable_eob_irq(mal);
+
+	/* Allocate space for BD rings */
+	BUG_ON(maldata->num_tx_chans <= 0 || maldata->num_tx_chans > 32);
+	BUG_ON(maldata->num_rx_chans <= 0 || maldata->num_rx_chans > 32);
+	bd_size = sizeof(struct mal_descriptor) *
+	    (NUM_TX_BUFF * maldata->num_tx_chans +
+	     NUM_RX_BUFF * maldata->num_rx_chans);
+	mal->bd_virt =
+	    dma_alloc_coherent(&ocpdev->dev, bd_size, &mal->bd_dma, GFP_KERNEL);
+
+	if (!mal->bd_virt) {
 		printk(KERN_ERR
-		       "mal%d: Out of memory allocating MAL descriptors !\n",
-		       ocpdev->def->index);
+		       "mal%d: out of memory allocating RX/TX descriptors!\n",
+		       mal->def->index);
 		err = -ENOMEM;
 		goto fail;
 	}
+	memset(mal->bd_virt, 0, bd_size);
 
-	/* God, oh, god, I hate DCRs */
-	set_mal_dcrn(mal, DCRN_MALTXCTP0R, mal->tx_phys_addr);
-#ifdef DCRN_MALTXCTP1R
-	if (maldata->num_tx_chans > 1)
-		set_mal_dcrn(mal, DCRN_MALTXCTP1R,
-			     mal->tx_phys_addr + MAL_DT_ALIGN);
-#endif				/* DCRN_MALTXCTP1R */
-#ifdef DCRN_MALTXCTP2R
-	if (maldata->num_tx_chans > 2)
-		set_mal_dcrn(mal, DCRN_MALTXCTP2R,
-			     mal->tx_phys_addr + 2 * MAL_DT_ALIGN);
-#endif				/* DCRN_MALTXCTP2R */
-#ifdef DCRN_MALTXCTP3R
-	if (maldata->num_tx_chans > 3)
-		set_mal_dcrn(mal, DCRN_MALTXCTP3R,
-			     mal->tx_phys_addr + 3 * MAL_DT_ALIGN);
-#endif				/* DCRN_MALTXCTP3R */
-#ifdef DCRN_MALTXCTP4R
-	if (maldata->num_tx_chans > 4)
-		set_mal_dcrn(mal, DCRN_MALTXCTP4R,
-			     mal->tx_phys_addr + 4 * MAL_DT_ALIGN);
-#endif				/* DCRN_MALTXCTP4R */
-#ifdef DCRN_MALTXCTP5R
-	if (maldata->num_tx_chans > 5)
-		set_mal_dcrn(mal, DCRN_MALTXCTP5R,
-			     mal->tx_phys_addr + 5 * MAL_DT_ALIGN);
-#endif				/* DCRN_MALTXCTP5R */
-#ifdef DCRN_MALTXCTP6R
-	if (maldata->num_tx_chans > 6)
-		set_mal_dcrn(mal, DCRN_MALTXCTP6R,
-			     mal->tx_phys_addr + 6 * MAL_DT_ALIGN);
-#endif				/* DCRN_MALTXCTP6R */
-#ifdef DCRN_MALTXCTP7R
-	if (maldata->num_tx_chans > 7)
-		set_mal_dcrn(mal, DCRN_MALTXCTP7R,
-			     mal->tx_phys_addr + 7 * MAL_DT_ALIGN);
-#endif				/* DCRN_MALTXCTP7R */
-
-	mal->rx_virt_addr = dma_alloc_coherent(&ocpdev->dev,
-					       MAL_DT_ALIGN *
-					       maldata->num_rx_chans,
-					       &mal->rx_phys_addr, GFP_KERNEL);
-
-	set_mal_dcrn(mal, DCRN_MALRXCTP0R, mal->rx_phys_addr);
-#ifdef DCRN_MALRXCTP1R
-	if (maldata->num_rx_chans > 1)
-		set_mal_dcrn(mal, DCRN_MALRXCTP1R,
-			     mal->rx_phys_addr + MAL_DT_ALIGN);
-#endif				/* DCRN_MALRXCTP1R */
-#ifdef DCRN_MALRXCTP2R
-	if (maldata->num_rx_chans > 2)
-		set_mal_dcrn(mal, DCRN_MALRXCTP2R,
-			     mal->rx_phys_addr + 2 * MAL_DT_ALIGN);
-#endif				/* DCRN_MALRXCTP2R */
-#ifdef DCRN_MALRXCTP3R
-	if (maldata->num_rx_chans > 3)
-		set_mal_dcrn(mal, DCRN_MALRXCTP3R,
-			     mal->rx_phys_addr + 3 * MAL_DT_ALIGN);
-#endif				/* DCRN_MALRXCTP3R */
+	for (i = 0; i < maldata->num_tx_chans; ++i)
+		set_mal_dcrn(mal, MAL_TXCTPR(i), mal->bd_dma +
+			     sizeof(struct mal_descriptor) *
+			     mal_tx_bd_offset(mal, i));
+
+	for (i = 0; i < maldata->num_rx_chans; ++i)
+		set_mal_dcrn(mal, MAL_RXCTPR(i), mal->bd_dma +
+			     sizeof(struct mal_descriptor) *
+			     mal_rx_bd_offset(mal, i));
 
 	err = request_irq(maldata->serr_irq, mal_serr, 0, "MAL SERR", mal);
 	if (err)
-		goto fail;
-	err = request_irq(maldata->txde_irq, mal_txde, 0, "MAL TX DE ", mal);
+		goto fail2;
+	err = request_irq(maldata->txde_irq, mal_txde, 0, "MAL TX DE", mal);
 	if (err)
-		goto fail;
+		goto fail3;
 	err = request_irq(maldata->txeob_irq, mal_txeob, 0, "MAL TX EOB", mal);
 	if (err)
-		goto fail;
+		goto fail4;
 	err = request_irq(maldata->rxde_irq, mal_rxde, 0, "MAL RX DE", mal);
 	if (err)
-		goto fail;
+		goto fail5;
 	err = request_irq(maldata->rxeob_irq, mal_rxeob, 0, "MAL RX EOB", mal);
 	if (err)
-		goto fail;
+		goto fail6;
 
-	set_mal_dcrn(mal, DCRN_MALIER,
-		     MALIER_DE | MALIER_NE | MALIER_TE |
-		     MALIER_OPBE | MALIER_PLBE);
+	/* Enable all MAL SERR interrupt sources */
+	set_mal_dcrn(mal, MAL_IER, MAL_IER_EVENTS);
 
-	/* Advertise me to the rest of the world */
+	/* Advertise this instance to the rest of the world */
 	ocp_set_drvdata(ocpdev, mal);
 
-	printk(KERN_INFO "mal%d: Initialized, %d tx channels, %d rx channels\n",
-	       ocpdev->def->index, maldata->num_tx_chans,
-	       maldata->num_rx_chans);
+	mal_dbg_register(mal->def->index, mal);
 
+	printk(KERN_INFO "mal%d: initialized, %d TX channels, %d RX channels\n",
+	       mal->def->index, maldata->num_tx_chans, maldata->num_rx_chans);
 	return 0;
 
+      fail6:
+	free_irq(maldata->rxde_irq, mal);
+      fail5:
+	free_irq(maldata->txeob_irq, mal);
+      fail4:
+	free_irq(maldata->txde_irq, mal);
+      fail3:
+	free_irq(maldata->serr_irq, mal);
+      fail2:
+	dma_free_coherent(&ocpdev->dev, bd_size, mal->bd_virt, mal->bd_dma);
       fail:
-	/* FIXME: dispose requested IRQs ! */
-	if (err && mal)
-		kfree(mal);
+	kfree(mal);
 	return err;
 }
 
 static void __exit mal_remove(struct ocp_device *ocpdev)
 {
 	struct ibm_ocp_mal *mal = ocp_get_drvdata(ocpdev);
-	struct ocp_func_mal_data *maldata = ocpdev->def->additions;
+	struct ocp_func_mal_data *maldata = mal->def->additions;
+
+	MAL_DBG("%d: remove" NL, mal->def->index);
 
-	BUG_ON(!maldata);
+	/* Syncronize with scheduled polling, 
+	   stolen from net/core/dev.c:dev_close() 
+	 */
+	clear_bit(__LINK_STATE_START, &mal->poll_dev.state);
+	netif_poll_disable(&mal->poll_dev);
+
+	if (!list_empty(&mal->list)) {
+		/* This is *very* bad */
+		printk(KERN_EMERG
+		       "mal%d: commac list is not empty on remove!\n",
+		       mal->def->index);
+	}
 
 	ocp_set_drvdata(ocpdev, NULL);
 
-	/* FIXME: shut down the MAL, deal with dependency with emac */
 	free_irq(maldata->serr_irq, mal);
 	free_irq(maldata->txde_irq, mal);
 	free_irq(maldata->txeob_irq, mal);
 	free_irq(maldata->rxde_irq, mal);
 	free_irq(maldata->rxeob_irq, mal);
 
-	if (mal->tx_virt_addr)
-		dma_free_coherent(&ocpdev->dev,
-				  MAL_DT_ALIGN * maldata->num_tx_chans,
-				  mal->tx_virt_addr, mal->tx_phys_addr);
+	mal_reset(mal);
 
-	if (mal->rx_virt_addr)
-		dma_free_coherent(&ocpdev->dev,
-				  MAL_DT_ALIGN * maldata->num_rx_chans,
-				  mal->rx_virt_addr, mal->rx_phys_addr);
+	mal_dbg_register(mal->def->index, NULL);
+
+	dma_free_coherent(&ocpdev->dev,
+			  sizeof(struct mal_descriptor) *
+			  (NUM_TX_BUFF * maldata->num_tx_chans +
+			   NUM_RX_BUFF * maldata->num_rx_chans), mal->bd_virt,
+			  mal->bd_dma);
 
 	kfree(mal);
 }
 
 /* Structure for a device driver */
 static struct ocp_device_id mal_ids[] = {
-	{.vendor = OCP_ANY_ID,.function = OCP_FUNC_MAL},
-	{.vendor = OCP_VENDOR_INVALID}
+	{ .vendor = OCP_VENDOR_IBM, .function = OCP_FUNC_MAL },
+	{ .vendor = OCP_VENDOR_INVALID}
 };
 
 static struct ocp_driver mal_driver = {
@@ -441,23 +570,14 @@ static struct ocp_driver mal_driver = {
 	.remove = mal_remove,
 };
 
-static int __init init_mals(void)
+int __init mal_init(void)
 {
-	int rc;
-
-	rc = ocp_register_driver(&mal_driver);
-	if (rc < 0) {
-		ocp_unregister_driver(&mal_driver);
-		return -ENODEV;
-	}
-
-	return 0;
+	MAL_DBG(": init" NL);
+	return ocp_register_driver(&mal_driver);
 }
 
-static void __exit exit_mals(void)
+void __exit mal_exit(void)
 {
+	MAL_DBG(": exit" NL);
 	ocp_unregister_driver(&mal_driver);
 }
-
-module_init(init_mals);
-module_exit(exit_mals);