[PATCH] bcm43xx-d80211: New DMA engine code

[Michael Buesch <mb@bu3sch.de>]

Hi John,

Please apply this to wireless-dev.
Larry, nothing interresting in here to backport. ;)

--

This is a rewrite of the bcm43xx DMA engine. It adds support
for >1G of memory (for chips that support the extension bits)
and 64-bit DMA (for chips that support it).

Signed-off-by: Michael Buesch <mb@bu3sch.de>

Index: wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx.h
===================================================================
--- wireless-dev.orig/drivers/net/wireless/d80211/bcm43xx/bcm43xx.h	2006-08-16 12:37:02.000000000 +0200
+++ wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx.h	2006-08-16 12:45:34.000000000 +0200
@@ -35,14 +35,18 @@
 #define BCM43xx_PCICFG_ICR		0x94
 
 /* MMIO offsets */
-#define BCM43xx_MMIO_DMA1_REASON	0x20
-#define BCM43xx_MMIO_DMA1_IRQ_MASK	0x24
-#define BCM43xx_MMIO_DMA2_REASON	0x28
-#define BCM43xx_MMIO_DMA2_IRQ_MASK	0x2C
-#define BCM43xx_MMIO_DMA3_REASON	0x30
-#define BCM43xx_MMIO_DMA3_IRQ_MASK	0x34
-#define BCM43xx_MMIO_DMA4_REASON	0x38
-#define BCM43xx_MMIO_DMA4_IRQ_MASK	0x3C
+#define BCM43xx_MMIO_DMA0_REASON	0x20
+#define BCM43xx_MMIO_DMA0_IRQ_MASK	0x24
+#define BCM43xx_MMIO_DMA1_REASON	0x28
+#define BCM43xx_MMIO_DMA1_IRQ_MASK	0x2C
+#define BCM43xx_MMIO_DMA2_REASON	0x30
+#define BCM43xx_MMIO_DMA2_IRQ_MASK	0x34
+#define BCM43xx_MMIO_DMA3_REASON	0x38
+#define BCM43xx_MMIO_DMA3_IRQ_MASK	0x3C
+#define BCM43xx_MMIO_DMA4_REASON	0x40
+#define BCM43xx_MMIO_DMA4_IRQ_MASK	0x44
+#define BCM43xx_MMIO_DMA5_REASON	0x48
+#define BCM43xx_MMIO_DMA5_IRQ_MASK	0x4C
 #define BCM43xx_MMIO_STATUS_BITFIELD	0x120
 #define BCM43xx_MMIO_STATUS2_BITFIELD	0x124
 #define BCM43xx_MMIO_GEN_IRQ_REASON	0x128
@@ -58,14 +62,27 @@
 #define BCM43xx_MMIO_XMITSTAT_1		0x174
 #define BCM43xx_MMIO_REV3PLUS_TSF_LOW	0x180 /* core rev >= 3 only */
 #define BCM43xx_MMIO_REV3PLUS_TSF_HIGH	0x184 /* core rev >= 3 only */
-#define BCM43xx_MMIO_DMA1_BASE		0x200
-#define BCM43xx_MMIO_DMA2_BASE		0x220
-#define BCM43xx_MMIO_DMA3_BASE		0x240
-#define BCM43xx_MMIO_DMA4_BASE		0x260
+
+/* 32-bit DMA */
+#define BCM43xx_MMIO_DMA32_BASE0	0x200
+#define BCM43xx_MMIO_DMA32_BASE1	0x220
+#define BCM43xx_MMIO_DMA32_BASE2	0x240
+#define BCM43xx_MMIO_DMA32_BASE3	0x260
+#define BCM43xx_MMIO_DMA32_BASE4	0x280
+#define BCM43xx_MMIO_DMA32_BASE5	0x2A0
+/* 64-bit DMA */
+#define BCM43xx_MMIO_DMA64_BASE0	0x200
+#define BCM43xx_MMIO_DMA64_BASE1	0x240
+#define BCM43xx_MMIO_DMA64_BASE2	0x280
+#define BCM43xx_MMIO_DMA64_BASE3	0x2C0
+#define BCM43xx_MMIO_DMA64_BASE4	0x300
+#define BCM43xx_MMIO_DMA64_BASE5	0x340
+/* PIO */
 #define BCM43xx_MMIO_PIO1_BASE		0x300
 #define BCM43xx_MMIO_PIO2_BASE		0x310
 #define BCM43xx_MMIO_PIO3_BASE		0x320
 #define BCM43xx_MMIO_PIO4_BASE		0x330
+
 #define BCM43xx_MMIO_PHY_VER		0x3E0
 #define BCM43xx_MMIO_PHY_RADIO		0x3E2
 #define BCM43xx_MMIO_ANTENNA		0x3E8
@@ -234,8 +251,14 @@
 #define BCM43xx_SBTMSTATELOW_FORCE_GATE_CLOCK	0x20000
 
 /* sbtmstatehigh state flags */
-#define BCM43xx_SBTMSTATEHIGH_SERROR		0x1
-#define BCM43xx_SBTMSTATEHIGH_BUSY		0x4
+#define BCM43xx_SBTMSTATEHIGH_SERROR		0x00000001
+#define BCM43xx_SBTMSTATEHIGH_BUSY		0x00000004
+#define BCM43xx_SBTMSTATEHIGH_TIMEOUT		0x00000020
+#define BCM43xx_SBTMSTATEHIGH_COREFLAGS		0x1FFF0000
+#define BCM43xx_SBTMSTATEHIGH_DMA64BIT		0x10000000
+#define BCM43xx_SBTMSTATEHIGH_GATEDCLK		0x20000000
+#define BCM43xx_SBTMSTATEHIGH_BISTFAILED	0x40000000
+#define BCM43xx_SBTMSTATEHIGH_BISTCOMPLETE	0x80000000
 
 /* sbimstate flags */
 #define BCM43xx_SBIMSTATE_IB_ERROR		0x20000
@@ -564,8 +587,11 @@
 	struct bcm43xx_dmaring *tx_ring1;
 	struct bcm43xx_dmaring *tx_ring2;
 	struct bcm43xx_dmaring *tx_ring3;
+	struct bcm43xx_dmaring *tx_ring4;
+	struct bcm43xx_dmaring *tx_ring5;
+
 	struct bcm43xx_dmaring *rx_ring0;
-	struct bcm43xx_dmaring *rx_ring1; /* only available on core.rev < 5 */
+	struct bcm43xx_dmaring *rx_ring3; /* only available on core.rev < 5 */
 };
 
 /* Data structures for PIO transmission, per 80211 core. */
@@ -758,7 +784,7 @@
 
 	/* Reason code of the last interrupt. */
 	u32 irq_reason;
-	u32 dma_reason[4];
+	u32 dma_reason[6];
 	/* saved irq enable/disable state bitfield. */
 	u32 irq_savedstate;
 	/* Link Quality calculation context. */
Index: wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_dma.c
===================================================================
--- wireless-dev.orig/drivers/net/wireless/d80211/bcm43xx/bcm43xx_dma.c	2006-08-16 12:37:02.000000000 +0200
+++ wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_dma.c	2006-08-16 17:23:55.000000000 +0200
@@ -4,7 +4,7 @@
 
   DMA ringbuffer and descriptor allocation/management
 
-  Copyright (c) 2005 Michael Buesch <mbuesch@freenet.de>
+  Copyright (c) 2005, 2006 Michael Buesch <mbuesch@freenet.de>
 
   Some code in this file is derived from the b44.c driver
   Copyright (C) 2002 David S. Miller
@@ -40,6 +40,203 @@
 #include <linux/skbuff.h>
 
 
+/* 32bit DMA ops. */
+static
+struct bcm43xx_dmadesc_generic * op32_idx2desc(struct bcm43xx_dmaring *ring,
+					       int slot,
+					       struct bcm43xx_dmadesc_meta **meta)
+{
+	struct bcm43xx_dmadesc32 *desc;
+
+	*meta = &(ring->meta[slot]);
+	desc = ring->descbase;
+	desc = &(desc[slot]);
+
+	return (struct bcm43xx_dmadesc_generic *)desc;
+}
+
+static void op32_fill_descriptor(struct bcm43xx_dmaring *ring,
+				 struct bcm43xx_dmadesc_generic *desc,
+				 dma_addr_t dmaaddr, u16 bufsize,
+				 int start, int end, int irq)
+{
+	struct bcm43xx_dmadesc32 *descbase = ring->descbase;
+	int slot;
+	u32 ctl;
+	u32 addr;
+	u32 addrext;
+
+	slot = (int)(&(desc->dma32) - descbase);
+	assert(slot >= 0 && slot < ring->nr_slots);
+
+	addr = (u32)(dmaaddr & ~BCM43xx_DMA32_ROUTING);
+	addrext = (u32)(dmaaddr & BCM43xx_DMA32_ROUTING)
+		   >> BCM43xx_DMA32_ROUTING_SHIFT;
+	addr |= ring->routing;
+	ctl = (bufsize - ring->frameoffset)
+	      & BCM43xx_DMA32_DCTL_BYTECNT;
+	if (slot == ring->nr_slots - 1)
+		ctl |= BCM43xx_DMA32_DCTL_DTABLEEND;
+	if (start)
+		ctl |= BCM43xx_DMA32_DCTL_FRAMESTART;
+	if (end)
+		ctl |= BCM43xx_DMA32_DCTL_FRAMEEND;
+	if (irq)
+		ctl |= BCM43xx_DMA32_DCTL_IRQ;
+	ctl |= (addrext << BCM43xx_DMA32_DCTL_ADDREXT_SHIFT)
+	       & BCM43xx_DMA32_DCTL_ADDREXT_MASK;
+
+	desc->dma32.control = cpu_to_le32(ctl);
+	desc->dma32.address = cpu_to_le32(addr);
+}
+
+static void op32_poke_tx(struct bcm43xx_dmaring *ring, int slot)
+{
+	bcm43xx_dma_write(ring, BCM43xx_DMA32_TXINDEX,
+			  (u32)(slot * sizeof(struct bcm43xx_dmadesc32)));
+}
+
+static void op32_tx_suspend(struct bcm43xx_dmaring *ring)
+{
+	bcm43xx_dma_write(ring, BCM43xx_DMA32_TXCTL,
+			  bcm43xx_dma_read(ring, BCM43xx_DMA32_TXCTL)
+			  | BCM43xx_DMA32_TXSUSPEND);
+}
+
+static void op32_tx_resume(struct bcm43xx_dmaring *ring)
+{
+	bcm43xx_dma_write(ring, BCM43xx_DMA32_TXCTL,
+			  bcm43xx_dma_read(ring, BCM43xx_DMA32_TXCTL)
+			  & ~BCM43xx_DMA32_TXSUSPEND);
+}
+
+static int op32_get_current_rxslot(struct bcm43xx_dmaring *ring)
+{
+	u32 val;
+
+	val = bcm43xx_dma_read(ring, BCM43xx_DMA32_RXSTATUS);
+	val &= BCM43xx_DMA32_RXDPTR;
+
+	return (val / sizeof(struct bcm43xx_dmadesc32));
+}
+
+static void op32_set_current_rxslot(struct bcm43xx_dmaring *ring,
+				    int slot)
+{
+	bcm43xx_dma_write(ring, BCM43xx_DMA32_RXINDEX,
+			  (u32)(slot * sizeof(struct bcm43xx_dmadesc32)));
+}
+
+static const struct bcm43xx_dma_ops dma32_ops = {
+	.idx2desc		= op32_idx2desc,
+	.fill_descriptor	= op32_fill_descriptor,
+	.poke_tx		= op32_poke_tx,
+	.tx_suspend		= op32_tx_suspend,
+	.tx_resume		= op32_tx_resume,
+	.get_current_rxslot	= op32_get_current_rxslot,
+	.set_current_rxslot	= op32_set_current_rxslot,
+};
+
+/* 64bit DMA ops. */
+static
+struct bcm43xx_dmadesc_generic * op64_idx2desc(struct bcm43xx_dmaring *ring,
+					       int slot,
+					       struct bcm43xx_dmadesc_meta **meta)
+{
+	struct bcm43xx_dmadesc64 *desc;
+
+	*meta = &(ring->meta[slot]);
+	desc = ring->descbase;
+	desc = &(desc[slot]);
+
+	return (struct bcm43xx_dmadesc_generic *)desc;
+}
+
+static void op64_fill_descriptor(struct bcm43xx_dmaring *ring,
+				 struct bcm43xx_dmadesc_generic *desc,
+				 dma_addr_t dmaaddr, u16 bufsize,
+				 int start, int end, int irq)
+{
+	struct bcm43xx_dmadesc64 *descbase = ring->descbase;
+	int slot;
+	u32 ctl0 = 0, ctl1 = 0;
+	u32 addrlo, addrhi;
+	u32 addrext;
+
+	slot = (int)(&(desc->dma64) - descbase);
+	assert(slot >= 0 && slot < ring->nr_slots);
+
+	addrlo = (u32)(dmaaddr & 0xFFFFFFFF);
+	addrhi = (((u64)dmaaddr >> 32) & ~BCM43xx_DMA64_ROUTING);
+	addrext = (((u64)dmaaddr >> 32) >> BCM43xx_DMA64_ROUTING_SHIFT);
+	addrhi |= ring->routing;
+	if (slot == ring->nr_slots - 1)
+		ctl0 |= BCM43xx_DMA64_DCTL0_DTABLEEND;
+	if (start)
+		ctl0 |= BCM43xx_DMA64_DCTL0_FRAMESTART;
+	if (end)
+		ctl0 |= BCM43xx_DMA64_DCTL0_FRAMEEND;
+	if (irq)
+		ctl0 |= BCM43xx_DMA64_DCTL0_IRQ;
+	ctl1 |= (bufsize - ring->frameoffset)
+		& BCM43xx_DMA64_DCTL1_BYTECNT;
+	ctl1 |= (addrext << BCM43xx_DMA64_DCTL1_ADDREXT_SHIFT)
+		& BCM43xx_DMA64_DCTL1_ADDREXT_MASK;
+
+	desc->dma64.control0 = cpu_to_le32(ctl0);
+	desc->dma64.control1 = cpu_to_le32(ctl1);
+	desc->dma64.address_low = cpu_to_le32(addrlo);
+	desc->dma64.address_high = cpu_to_le32(addrhi);
+}
+
+static void op64_poke_tx(struct bcm43xx_dmaring *ring, int slot)
+{
+	bcm43xx_dma_write(ring, BCM43xx_DMA64_TXINDEX,
+			  (u32)(slot * sizeof(struct bcm43xx_dmadesc64)));
+}
+
+static void op64_tx_suspend(struct bcm43xx_dmaring *ring)
+{
+	bcm43xx_dma_write(ring, BCM43xx_DMA64_TXCTL,
+			  bcm43xx_dma_read(ring, BCM43xx_DMA64_TXCTL)
+			  | BCM43xx_DMA64_TXSUSPEND);
+}
+
+static void op64_tx_resume(struct bcm43xx_dmaring *ring)
+{
+	bcm43xx_dma_write(ring, BCM43xx_DMA64_TXCTL,
+			  bcm43xx_dma_read(ring, BCM43xx_DMA64_TXCTL)
+			  & ~BCM43xx_DMA64_TXSUSPEND);
+}
+
+static int op64_get_current_rxslot(struct bcm43xx_dmaring *ring)
+{
+	u32 val;
+
+	val = bcm43xx_dma_read(ring, BCM43xx_DMA64_RXSTATUS);
+	val &= BCM43xx_DMA64_RXSTATDPTR;
+
+	return (val / sizeof(struct bcm43xx_dmadesc64));
+}
+
+static void op64_set_current_rxslot(struct bcm43xx_dmaring *ring,
+				    int slot)
+{
+	bcm43xx_dma_write(ring, BCM43xx_DMA64_RXINDEX,
+			  (u32)(slot * sizeof(struct bcm43xx_dmadesc64)));
+}
+
+static const struct bcm43xx_dma_ops dma64_ops = {
+	.idx2desc		= op64_idx2desc,
+	.fill_descriptor	= op64_fill_descriptor,
+	.poke_tx		= op64_poke_tx,
+	.tx_suspend		= op64_tx_suspend,
+	.tx_resume		= op64_tx_resume,
+	.get_current_rxslot	= op64_get_current_rxslot,
+	.set_current_rxslot	= op64_set_current_rxslot,
+};
+
+
 static inline int free_slots(struct bcm43xx_dmaring *ring)
 {
 	return (ring->nr_slots - ring->used_slots);
@@ -68,7 +265,7 @@
 	int slot;
 
 	assert(ring->tx);
-	assert(!ring->suspended);
+	assert(!ring->stopped);
 	assert(free_slots(ring) != 0);
 
 	slot = next_slot(ring, ring->current_slot);
@@ -92,6 +289,35 @@
 	ring->used_slots--;
 }
 
+u16 bcm43xx_dmacontroller_base(int dma64bit, int controller_idx)
+{
+	static const u16 map64[] = {
+		BCM43xx_MMIO_DMA64_BASE0,
+		BCM43xx_MMIO_DMA64_BASE1,
+		BCM43xx_MMIO_DMA64_BASE2,
+		BCM43xx_MMIO_DMA64_BASE3,
+		BCM43xx_MMIO_DMA64_BASE4,
+		BCM43xx_MMIO_DMA64_BASE5,
+	};
+	static const u16 map32[] = {
+		BCM43xx_MMIO_DMA32_BASE0,
+		BCM43xx_MMIO_DMA32_BASE1,
+		BCM43xx_MMIO_DMA32_BASE2,
+		BCM43xx_MMIO_DMA32_BASE3,
+		BCM43xx_MMIO_DMA32_BASE4,
+		BCM43xx_MMIO_DMA32_BASE5,
+	};
+
+	if (dma64bit) {
+		assert(controller_idx >= 0 &&
+		       controller_idx < ARRAY_SIZE(map64));
+		return map64[controller_idx];
+	}
+	assert(controller_idx >= 0 &&
+	       controller_idx < ARRAY_SIZE(map32));
+	return map32[controller_idx];
+}
+
 static inline
 dma_addr_t map_descbuffer(struct bcm43xx_dmaring *ring,
 			  unsigned char *buf,
@@ -152,41 +378,30 @@
 				   addr, len, DMA_FROM_DEVICE);
 }
 
-/* Unmap and free a descriptor buffer. */
 static inline
 void free_descriptor_buffer(struct bcm43xx_dmaring *ring,
-			    struct bcm43xx_dmadesc *desc,
 			    struct bcm43xx_dmadesc_meta *meta,
 			    int irq_context)
 {
-	assert(meta->skb);
-	if (irq_context)
-		dev_kfree_skb_irq(meta->skb);
-	else
-		dev_kfree_skb(meta->skb);
-	meta->skb = NULL;
+	if (meta->skb) {
+		if (irq_context)
+			dev_kfree_skb_irq(meta->skb);
+		else
+			dev_kfree_skb(meta->skb);
+	}
 }
 
 static int alloc_ringmemory(struct bcm43xx_dmaring *ring)
 {
 	struct device *dev = &(ring->bcm->pci_dev->dev);
 
-	ring->vbase = dma_alloc_coherent(dev, BCM43xx_DMA_RINGMEMSIZE,
-					 &(ring->dmabase), GFP_KERNEL);
-	if (!ring->vbase) {
+	ring->descbase = dma_alloc_coherent(dev, BCM43xx_DMA_RINGMEMSIZE,
+					    &(ring->dmabase), GFP_KERNEL);
+	if (!ring->descbase) {
 		printk(KERN_ERR PFX "DMA ringmemory allocation failed\n");
 		return -ENOMEM;
 	}
-	if (ring->dmabase + BCM43xx_DMA_RINGMEMSIZE > BCM43xx_DMA_BUSADDRMAX) {
-		printk(KERN_ERR PFX ">>>FATAL ERROR<<<  DMA RINGMEMORY >1G "
-				    "(0x%08x, len: %lu)\n",
-		       (unsigned int)ring->dmabase, BCM43xx_DMA_RINGMEMSIZE);
-		dma_free_coherent(dev, BCM43xx_DMA_RINGMEMSIZE,
-				  ring->vbase, ring->dmabase);
-		return -ENOMEM;
-	}
-	assert(!(ring->dmabase & 0x000003FF));
-	memset(ring->vbase, 0, BCM43xx_DMA_RINGMEMSIZE);
+	memset(ring->descbase, 0, BCM43xx_DMA_RINGMEMSIZE);
 
 	return 0;
 }
@@ -196,26 +411,34 @@
 	struct device *dev = &(ring->bcm->pci_dev->dev);
 
 	dma_free_coherent(dev, BCM43xx_DMA_RINGMEMSIZE,
-			  ring->vbase, ring->dmabase);
+			  ring->descbase, ring->dmabase);
 }
 
 /* Reset the RX DMA channel */
 int bcm43xx_dmacontroller_rx_reset(struct bcm43xx_private *bcm,
-				   u16 mmio_base)
+				   u16 mmio_base, int dma64)
 {
 	int i;
 	u32 value;
+	u16 offset;
 
-	bcm43xx_write32(bcm,
-			mmio_base + BCM43xx_DMA_RX_CONTROL,
-			0x00000000);
+	offset = dma64 ? BCM43xx_DMA64_RXCTL : BCM43xx_DMA32_RXCTL;
+	bcm43xx_write32(bcm, mmio_base + offset, 0);
 	for (i = 0; i < 1000; i++) {
-		value = bcm43xx_read32(bcm,
-				       mmio_base + BCM43xx_DMA_RX_STATUS);
-		value &= BCM43xx_DMA_RXSTAT_STAT_MASK;
-		if (value == BCM43xx_DMA_RXSTAT_STAT_DISABLED) {
-			i = -1;
-			break;
+		offset = dma64 ? BCM43xx_DMA64_RXSTATUS : BCM43xx_DMA32_RXSTATUS;
+		value = bcm43xx_read32(bcm, mmio_base + offset);
+		if (dma64) {
+			value &= BCM43xx_DMA64_RXSTAT;
+			if (value == BCM43xx_DMA64_RXSTAT_DISABLED) {
+				i = -1;
+				break;
+			}
+		} else {
+			value &= BCM43xx_DMA32_RXSTATE;
+			if (value == BCM43xx_DMA32_RXSTAT_DISABLED) {
+				i = -1;
+				break;
+			}
 		}
 		udelay(10);
 	}
@@ -229,31 +452,47 @@
 
 /* Reset the RX DMA channel */
 int bcm43xx_dmacontroller_tx_reset(struct bcm43xx_private *bcm,
-				   u16 mmio_base)
+				   u16 mmio_base, int dma64)
 {
 	int i;
 	u32 value;
+	u16 offset;
 
 	for (i = 0; i < 1000; i++) {
-		value = bcm43xx_read32(bcm,
-				       mmio_base + BCM43xx_DMA_TX_STATUS);
-		value &= BCM43xx_DMA_TXSTAT_STAT_MASK;
-		if (value == BCM43xx_DMA_TXSTAT_STAT_DISABLED ||
-		    value == BCM43xx_DMA_TXSTAT_STAT_IDLEWAIT ||
-		    value == BCM43xx_DMA_TXSTAT_STAT_STOPPED)
-			break;
+		offset = dma64 ? BCM43xx_DMA64_TXSTATUS : BCM43xx_DMA32_TXSTATUS;
+		value = bcm43xx_read32(bcm, mmio_base + offset);
+		if (dma64) {
+			value &= BCM43xx_DMA64_TXSTAT;
+			if (value == BCM43xx_DMA64_TXSTAT_DISABLED ||
+			    value == BCM43xx_DMA64_TXSTAT_IDLEWAIT ||
+			    value == BCM43xx_DMA64_TXSTAT_STOPPED)
+				break;
+		} else {
+			value &= BCM43xx_DMA32_TXSTATE;
+			if (value == BCM43xx_DMA32_TXSTAT_DISABLED ||
+			    value == BCM43xx_DMA32_TXSTAT_IDLEWAIT ||
+			    value == BCM43xx_DMA32_TXSTAT_STOPPED)
+				break;
+		}
 		udelay(10);
 	}
-	bcm43xx_write32(bcm,
-			mmio_base + BCM43xx_DMA_TX_CONTROL,
-			0x00000000);
+	offset = dma64 ? BCM43xx_DMA64_TXCTL : BCM43xx_DMA32_TXCTL;
+	bcm43xx_write32(bcm, mmio_base + offset, 0);
 	for (i = 0; i < 1000; i++) {
-		value = bcm43xx_read32(bcm,
-				       mmio_base + BCM43xx_DMA_TX_STATUS);
-		value &= BCM43xx_DMA_TXSTAT_STAT_MASK;
-		if (value == BCM43xx_DMA_TXSTAT_STAT_DISABLED) {
-			i = -1;
-			break;
+		offset = dma64 ? BCM43xx_DMA64_TXSTATUS : BCM43xx_DMA32_TXSTATUS;
+		value = bcm43xx_read32(bcm, mmio_base + offset);
+		if (dma64) {
+			value &= BCM43xx_DMA64_TXSTAT;
+			if (value == BCM43xx_DMA64_TXSTAT_DISABLED) {
+				i = -1;
+				break;
+			}
+		} else {
+			value &= BCM43xx_DMA32_TXSTATE;
+			if (value == BCM43xx_DMA32_TXSTAT_DISABLED) {
+				i = -1;
+				break;
+			}
 		}
 		udelay(10);
 	}
@@ -268,18 +507,15 @@
 }
 
 static int setup_rx_descbuffer(struct bcm43xx_dmaring *ring,
-			       struct bcm43xx_dmadesc *desc,
+			       struct bcm43xx_dmadesc_generic *desc,
 			       struct bcm43xx_dmadesc_meta *meta,
 			       gfp_t gfp_flags)
 {
 	struct bcm43xx_rxhdr *rxhdr;
+	struct bcm43xx_hwxmitstatus *xmitstat;
 	dma_addr_t dmaaddr;
-	u32 desc_addr;
-	u32 desc_ctl;
-	const int slot = (int)(desc - ring->vbase);
 	struct sk_buff *skb;
 
-	assert(slot >= 0 && slot < ring->nr_slots);
 	assert(!ring->tx);
 
 	skb = __dev_alloc_skb(ring->rx_buffersize, gfp_flags);
@@ -287,28 +523,18 @@
 		return -ENOMEM;
 	dmaaddr = map_descbuffer(ring, skb->data,
 				 ring->rx_buffersize, 0);
-	if (unlikely(dmaaddr + ring->rx_buffersize > BCM43xx_DMA_BUSADDRMAX)) {
-		unmap_descbuffer(ring, dmaaddr, ring->rx_buffersize, 0);
-		dev_kfree_skb_any(skb);
-		printk(KERN_ERR PFX ">>>FATAL ERROR<<<  DMA RX SKB >1G "
-				    "(0x%08x, len: %u)\n",
-		       (unsigned int)dmaaddr, ring->rx_buffersize);
-		return -ENOMEM;
-	}
 	meta->skb = skb;
 	meta->dmaaddr = dmaaddr;
 	skb->dev = ring->bcm->net_dev;
-	desc_addr = (u32)(dmaaddr + ring->memoffset);
-	desc_ctl = (BCM43xx_DMADTOR_BYTECNT_MASK &
-		    (u32)(ring->rx_buffersize - ring->frameoffset));
-	if (slot == ring->nr_slots - 1)
-		desc_ctl |= BCM43xx_DMADTOR_DTABLEEND;
-	set_desc_addr(desc, desc_addr);
-	set_desc_ctl(desc, desc_ctl);
+
+	ring->ops->fill_descriptor(ring, desc, dmaaddr,
+				   ring->rx_buffersize, 0, 0, 0);
 
 	rxhdr = (struct bcm43xx_rxhdr *)(skb->data);
 	rxhdr->frame_length = 0;
 	rxhdr->flags1 = 0;
+	xmitstat = (struct bcm43xx_hwxmitstatus *)(skb->data);
+	xmitstat->cookie = 0;
 
 	return 0;
 }
@@ -319,17 +545,17 @@
 static int alloc_initial_descbuffers(struct bcm43xx_dmaring *ring)
 {
 	int i, err = -ENOMEM;
-	struct bcm43xx_dmadesc *desc;
+	struct bcm43xx_dmadesc_generic *desc;
 	struct bcm43xx_dmadesc_meta *meta;
 
 	for (i = 0; i < ring->nr_slots; i++) {
-		desc = ring->vbase + i;
-		meta = ring->meta + i;
+		desc = ring->ops->idx2desc(ring, i, &meta);
 
 		err = setup_rx_descbuffer(ring, desc, meta, GFP_KERNEL);
 		if (err)
 			goto err_unwind;
 	}
+	mb();
 	ring->used_slots = ring->nr_slots;
 	err = 0;
 out:
@@ -337,8 +563,7 @@
 
 err_unwind:
 	for (i--; i >= 0; i--) {
-		desc = ring->vbase + i;
-		meta = ring->meta + i;
+		desc = ring->ops->idx2desc(ring, i, &meta);
 
 		unmap_descbuffer(ring, meta->dmaaddr, ring->rx_buffersize, 0);
 		dev_kfree_skb(meta->skb);
@@ -354,27 +579,67 @@
 {
 	int err = 0;
 	u32 value;
+	u32 addrext;
 
 	if (ring->tx) {
-		/* Set Transmit Control register to "transmit enable" */
-		bcm43xx_dma_write(ring, BCM43xx_DMA_TX_CONTROL,
-				  BCM43xx_DMA_TXCTRL_ENABLE);
-		/* Set Transmit Descriptor ring address. */
-		bcm43xx_dma_write(ring, BCM43xx_DMA_TX_DESC_RING,
-				  ring->dmabase + ring->memoffset);
+		if (ring->dma64) {
+			u64 ringbase = (u64)(ring->dmabase);
+
+			addrext = ((ringbase >> 32) >> BCM43xx_DMA64_ROUTING_SHIFT);
+			value = BCM43xx_DMA64_TXENABLE;
+			value |= (addrext << BCM43xx_DMA64_TXADDREXT_SHIFT)
+				& BCM43xx_DMA64_TXADDREXT_MASK;
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_TXCTL, value);
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_TXRINGLO,
+					(ringbase & 0xFFFFFFFF));
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_TXRINGHI,
+					((ringbase >> 32) & ~BCM43xx_DMA64_ROUTING)
+					| ring->routing);
+		} else {
+			u32 ringbase = (u32)(ring->dmabase);
+
+			addrext = (ringbase >> BCM43xx_DMA32_ROUTING_SHIFT);
+			value = BCM43xx_DMA32_TXENABLE;
+			value |= (addrext << BCM43xx_DMA32_TXADDREXT_SHIFT)
+				& BCM43xx_DMA32_TXADDREXT_MASK;
+			bcm43xx_dma_write(ring, BCM43xx_DMA32_TXCTL, value);
+			bcm43xx_dma_write(ring, BCM43xx_DMA32_TXRING,
+					(ringbase & ~BCM43xx_DMA32_ROUTING)
+					| ring->routing);
+		}
 	} else {
 		err = alloc_initial_descbuffers(ring);
 		if (err)
 			goto out;
-		/* Set Receive Control "receive enable" and frame offset */
-		value = (ring->frameoffset << BCM43xx_DMA_RXCTRL_FRAMEOFF_SHIFT);
-		value |= BCM43xx_DMA_RXCTRL_ENABLE;
-		bcm43xx_dma_write(ring, BCM43xx_DMA_RX_CONTROL, value);
-		/* Set Receive Descriptor ring address. */
-		bcm43xx_dma_write(ring, BCM43xx_DMA_RX_DESC_RING,
-				  ring->dmabase + ring->memoffset);
-		/* Init the descriptor pointer. */
-		bcm43xx_dma_write(ring, BCM43xx_DMA_RX_DESC_INDEX, 200);
+		if (ring->dma64) {
+			u64 ringbase = (u64)(ring->dmabase);
+
+			addrext = ((ringbase >> 32) >> BCM43xx_DMA64_ROUTING_SHIFT);
+			value = (ring->frameoffset << BCM43xx_DMA64_RXFROFF_SHIFT);
+			value |= BCM43xx_DMA64_RXENABLE;
+			value |= (addrext << BCM43xx_DMA64_RXADDREXT_SHIFT)
+				& BCM43xx_DMA64_RXADDREXT_MASK;
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_RXCTL, value);
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_RXRINGLO,
+					(ringbase & 0xFFFFFFFF));
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_RXRINGHI,
+					((ringbase >> 32) & ~BCM43xx_DMA64_ROUTING)
+					| ring->routing);
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_RXINDEX, 200);
+		} else {
+			u32 ringbase = (u32)(ring->dmabase);
+
+			addrext = (ringbase >> BCM43xx_DMA32_ROUTING_SHIFT);
+			value = (ring->frameoffset << BCM43xx_DMA32_RXFROFF_SHIFT);
+			value |= BCM43xx_DMA32_RXENABLE;
+			value |= (addrext << BCM43xx_DMA32_RXADDREXT_SHIFT)
+				& BCM43xx_DMA32_RXADDREXT_MASK;
+			bcm43xx_dma_write(ring, BCM43xx_DMA32_RXCTL, value);
+			bcm43xx_dma_write(ring, BCM43xx_DMA32_RXRING,
+					(ringbase & ~BCM43xx_DMA32_ROUTING)
+					| ring->routing);
+			bcm43xx_dma_write(ring, BCM43xx_DMA32_RXINDEX, 200);
+		}
 	}
 
 out:
@@ -385,27 +650,32 @@
 static void dmacontroller_cleanup(struct bcm43xx_dmaring *ring)
 {
 	if (ring->tx) {
-		bcm43xx_dmacontroller_tx_reset(ring->bcm, ring->mmio_base);
-		/* Zero out Transmit Descriptor ring address. */
-		bcm43xx_dma_write(ring, BCM43xx_DMA_TX_DESC_RING, 0);
+		bcm43xx_dmacontroller_tx_reset(ring->bcm, ring->mmio_base, ring->dma64);
+		if (ring->dma64) {
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_TXRINGLO, 0);
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_TXRINGHI, 0);
+		} else
+			bcm43xx_dma_write(ring, BCM43xx_DMA32_TXRING, 0);
 	} else {
-		bcm43xx_dmacontroller_rx_reset(ring->bcm, ring->mmio_base);
-		/* Zero out Receive Descriptor ring address. */
-		bcm43xx_dma_write(ring, BCM43xx_DMA_RX_DESC_RING, 0);
+		bcm43xx_dmacontroller_rx_reset(ring->bcm, ring->mmio_base, ring->dma64);
+		if (ring->dma64) {
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_RXRINGLO, 0);
+			bcm43xx_dma_write(ring, BCM43xx_DMA64_RXRINGHI, 0);
+		} else
+			bcm43xx_dma_write(ring, BCM43xx_DMA32_RXRING, 0);
 	}
 }
 
 static void free_all_descbuffers(struct bcm43xx_dmaring *ring)
 {
-	struct bcm43xx_dmadesc *desc;
+	struct bcm43xx_dmadesc_generic *desc;
 	struct bcm43xx_dmadesc_meta *meta;
 	int i;
 
 	if (!ring->used_slots)
 		return;
 	for (i = 0; i < ring->nr_slots; i++) {
-		desc = ring->vbase + i;
-		meta = ring->meta + i;
+		desc = ring->ops->idx2desc(ring, i, &meta);
 
 		if (!meta->skb) {
 			assert(ring->tx);
@@ -413,59 +683,75 @@
 		}
 		if (ring->tx) {
 			unmap_descbuffer(ring, meta->dmaaddr,
-					 meta->skb->len, 1);
+					meta->skb->len, 1);
 		} else {
 			unmap_descbuffer(ring, meta->dmaaddr,
-					 ring->rx_buffersize, 0);
+					ring->rx_buffersize, 0);
 		}
-		free_descriptor_buffer(ring, desc, meta, 0);
+		free_descriptor_buffer(ring, meta, 0);
 	}
 }
 
 /* Main initialization function. */
 static
 struct bcm43xx_dmaring * bcm43xx_setup_dmaring(struct bcm43xx_private *bcm,
-					       u16 dma_controller_base,
-					       int nr_descriptor_slots,
-					       int tx)
+					       int controller_index,
+					       int for_tx,
+					       int dma64)
 {
 	struct bcm43xx_dmaring *ring;
 	int err;
+	int nr_slots;
 
 	ring = kzalloc(sizeof(*ring), GFP_KERNEL);
 	if (!ring)
 		goto out;
 
-	ring->meta = kzalloc(sizeof(*ring->meta) * nr_descriptor_slots,
+	nr_slots = BCM43xx_RXRING_SLOTS;
+	if (for_tx)
+		nr_slots = BCM43xx_TXRING_SLOTS;
+
+	ring->meta = kcalloc(nr_slots, sizeof(struct bcm43xx_dmadesc_meta),
 			     GFP_KERNEL);
 	if (!ring->meta)
 		goto err_kfree_ring;
+	if (for_tx) {
+		ring->txhdr_cache = kcalloc(nr_slots,
+					    sizeof(struct bcm43xx_txhdr),
+					    GFP_KERNEL);
+		if (!ring->txhdr_cache)
+			goto err_kfree_meta;
+	}
 
-	ring->memoffset = BCM43xx_DMA_DMABUSADDROFFSET;
+	ring->routing = BCM43xx_DMA32_CLIENTTRANS;
+	if (dma64)
+		ring->routing = BCM43xx_DMA64_CLIENTTRANS;
 #ifdef CONFIG_BCM947XX
 	if (bcm->pci_dev->bus->number == 0)
-		ring->memoffset = 0;
+		ring->routing = dma64 ? BCM43xx_DMA64_NOTRANS : BCM43xx_DMA32_NOTRANS;
 #endif
 
 	ring->bcm = bcm;
-	ring->nr_slots = nr_descriptor_slots;
-	ring->mmio_base = dma_controller_base;
-	if (tx) {
+	ring->nr_slots = nr_slots;
+	ring->mmio_base = bcm43xx_dmacontroller_base(dma64, controller_index);
+	ring->index = controller_index;
+	ring->dma64 = !!dma64;
+	if (dma64)
+		ring->ops = &dma64_ops;
+	else
+		ring->ops = &dma32_ops;
+	if (for_tx) {
 		ring->tx = 1;
 		ring->current_slot = -1;
 	} else {
-		switch (dma_controller_base) {
-		case BCM43xx_MMIO_DMA1_BASE:
-			ring->rx_buffersize = BCM43xx_DMA1_RXBUFFERSIZE;
-			ring->frameoffset = BCM43xx_DMA1_RX_FRAMEOFFSET;
-			break;
-		case BCM43xx_MMIO_DMA4_BASE:
-			ring->rx_buffersize = BCM43xx_DMA4_RXBUFFERSIZE;
-			ring->frameoffset = BCM43xx_DMA4_RX_FRAMEOFFSET;
-			break;
-		default:
+		if (ring->index == 0) {
+			ring->rx_buffersize = BCM43xx_DMA0_RX_BUFFERSIZE;
+			ring->frameoffset = BCM43xx_DMA0_RX_FRAMEOFFSET;
+		} else if (ring->index == 3) {
+			ring->rx_buffersize = BCM43xx_DMA3_RX_BUFFERSIZE;
+			ring->frameoffset = BCM43xx_DMA3_RX_FRAMEOFFSET;
+		} else
 			assert(0);
-		}
 	}
 
 	err = alloc_ringmemory(ring);
@@ -494,7 +780,8 @@
 	if (!ring)
 		return;
 
-	dprintk(KERN_INFO PFX "DMA 0x%04x (%s) max used slots: %d/%d\n",
+	dprintk(KERN_INFO PFX "DMA-%s 0x%04X (%s) max used slots: %d/%d\n",
+		(ring->dma64) ? "64" : "32",
 		ring->mmio_base,
 		(ring->tx) ? "TX" : "RX",
 		ring->max_used_slots, ring->nr_slots);
@@ -505,6 +792,7 @@
 	free_all_descbuffers(ring);
 	free_ringmemory(ring);
 
+	kfree(ring->txhdr_cache);
 	kfree(ring->meta);
 	kfree(ring);
 }
@@ -517,10 +805,15 @@
 		return;
 	dma = bcm43xx_current_dma(bcm);
 
-	bcm43xx_destroy_dmaring(dma->rx_ring1);
-	dma->rx_ring1 = NULL;
+	bcm43xx_destroy_dmaring(dma->rx_ring3);
+	dma->rx_ring3 = NULL;
 	bcm43xx_destroy_dmaring(dma->rx_ring0);
 	dma->rx_ring0 = NULL;
+
+	bcm43xx_destroy_dmaring(dma->tx_ring5);
+	dma->tx_ring5 = NULL;
+	bcm43xx_destroy_dmaring(dma->tx_ring4);
+	dma->tx_ring4 = NULL;
 	bcm43xx_destroy_dmaring(dma->tx_ring3);
 	dma->tx_ring3 = NULL;
 	bcm43xx_destroy_dmaring(dma->tx_ring2);
@@ -536,48 +829,59 @@
 	struct bcm43xx_dma *dma = bcm43xx_current_dma(bcm);
 	struct bcm43xx_dmaring *ring;
 	int err = -ENOMEM;
+	int dma64 = 0;
+	u32 sbtmstatehi;
+
+	sbtmstatehi = bcm43xx_read32(bcm, BCM43xx_CIR_SBTMSTATEHIGH);
+	if (sbtmstatehi & BCM43xx_SBTMSTATEHIGH_DMA64BIT)
+		dma64 = 1;
 
 	/* setup TX DMA channels. */
-	ring = bcm43xx_setup_dmaring(bcm, BCM43xx_MMIO_DMA1_BASE,
-				     BCM43xx_TXRING_SLOTS, 1);
+	ring = bcm43xx_setup_dmaring(bcm, 0, 1, dma64);
 	if (!ring)
 		goto out;
 	dma->tx_ring0 = ring;
 
-	ring = bcm43xx_setup_dmaring(bcm, BCM43xx_MMIO_DMA2_BASE,
-				     BCM43xx_TXRING_SLOTS, 1);
+	ring = bcm43xx_setup_dmaring(bcm, 1, 1, dma64);
 	if (!ring)
 		goto err_destroy_tx0;
 	dma->tx_ring1 = ring;
 
-	ring = bcm43xx_setup_dmaring(bcm, BCM43xx_MMIO_DMA3_BASE,
-				     BCM43xx_TXRING_SLOTS, 1);
+	ring = bcm43xx_setup_dmaring(bcm, 2, 1, dma64);
 	if (!ring)
 		goto err_destroy_tx1;
 	dma->tx_ring2 = ring;
 
-	ring = bcm43xx_setup_dmaring(bcm, BCM43xx_MMIO_DMA4_BASE,
-				     BCM43xx_TXRING_SLOTS, 1);
+	ring = bcm43xx_setup_dmaring(bcm, 3, 1, dma64);
 	if (!ring)
 		goto err_destroy_tx2;
 	dma->tx_ring3 = ring;
 
-	/* setup RX DMA channels. */
-	ring = bcm43xx_setup_dmaring(bcm, BCM43xx_MMIO_DMA1_BASE,
-				     BCM43xx_RXRING_SLOTS, 0);
+	ring = bcm43xx_setup_dmaring(bcm, 4, 1, dma64);
 	if (!ring)
 		goto err_destroy_tx3;
+	dma->tx_ring4 = ring;
+
+	ring = bcm43xx_setup_dmaring(bcm, 5, 1, dma64);
+	if (!ring)
+		goto err_destroy_tx4;
+	dma->tx_ring5 = ring;
+
+	/* setup RX DMA channels. */
+	ring = bcm43xx_setup_dmaring(bcm, 0, 0, dma64);
+	if (!ring)
+		goto err_destroy_tx5;
 	dma->rx_ring0 = ring;
 
 	if (bcm->current_core->rev < 5) {
-		ring = bcm43xx_setup_dmaring(bcm, BCM43xx_MMIO_DMA4_BASE,
-					     BCM43xx_RXRING_SLOTS, 0);
+		ring = bcm43xx_setup_dmaring(bcm, 3, 0, dma64);
 		if (!ring)
 			goto err_destroy_rx0;
-		dma->rx_ring1 = ring;
+		dma->rx_ring3 = ring;
 	}
 
-	dprintk(KERN_INFO PFX "DMA initialized\n");
+	dprintk(KERN_INFO PFX "%s DMA initialized\n",
+			dma64 ? "64-bit" : "32-bit");
 	err = 0;
 out:
 	return err;
@@ -585,6 +889,12 @@
 err_destroy_rx0:
 	bcm43xx_destroy_dmaring(dma->rx_ring0);
 	dma->rx_ring0 = NULL;
+err_destroy_tx5:
+	bcm43xx_destroy_dmaring(dma->tx_ring5);
+	dma->tx_ring5 = NULL;
+err_destroy_tx4:
+	bcm43xx_destroy_dmaring(dma->tx_ring4);
+	dma->tx_ring4 = NULL;
 err_destroy_tx3:
 	bcm43xx_destroy_dmaring(dma->tx_ring3);
 	dma->tx_ring3 = NULL;
@@ -604,7 +914,7 @@
 static u16 generate_cookie(struct bcm43xx_dmaring *ring,
 			   int slot)
 {
-	u16 cookie = 0xF000;
+	u16 cookie = 0x1000;
 
 	/* Use the upper 4 bits of the cookie as
 	 * DMA controller ID and store the slot number
@@ -612,21 +922,25 @@
 	 * Note that the cookie must never be 0, as this
 	 * is a special value used in RX path.
 	 */
-	switch (ring->mmio_base) {
-	default:
-		assert(0);
-	case BCM43xx_MMIO_DMA1_BASE:
+	switch (ring->index) {
+	case 0:
 		cookie = 0xA000;
 		break;
-	case BCM43xx_MMIO_DMA2_BASE:
+	case 1:
 		cookie = 0xB000;
 		break;
-	case BCM43xx_MMIO_DMA3_BASE:
+	case 2:
 		cookie = 0xC000;
 		break;
-	case BCM43xx_MMIO_DMA4_BASE:
+	case 3:
 		cookie = 0xD000;
 		break;
+	case 4:
+		cookie = 0xE000;
+		break;
+	case 5:
+		cookie = 0xF000;
+		break;
 	}
 	assert(((u16)slot & 0xF000) == 0x0000);
 	cookie |= (u16)slot;
@@ -655,6 +969,12 @@
 	case 0xD000:
 		ring = dma->tx_ring3;
 		break;
+	case 0xE000:
+		ring = dma->tx_ring4;
+		break;
+	case 0xF000:
+		ring = dma->tx_ring5;
+		break;
 	default:
 		assert(0);
 	}
@@ -664,103 +984,50 @@
 	return ring;
 }
 
-static void dmacontroller_poke_tx(struct bcm43xx_dmaring *ring,
-				  int slot)
-{
-	/* Everything is ready to start. Buffers are DMA mapped and
-	 * associated with slots.
-	 * "slot" is the last slot of the new frame we want to transmit.
-	 * Close your seat belts now, please.
-	 */
-	wmb();
-	slot = next_slot(ring, slot);
-	bcm43xx_dma_write(ring, BCM43xx_DMA_TX_DESC_INDEX,
-			  (u32)(slot * sizeof(struct bcm43xx_dmadesc)));
-}
-
-static int dma_tx_fragment(struct bcm43xx_dmaring *ring,
-			   struct sk_buff *skb,
-			   struct ieee80211_tx_control *ctl)
+static void dma_tx_fragment(struct bcm43xx_dmaring *ring,
+			    struct sk_buff *skb,
+			    struct ieee80211_tx_control *ctl)
 {
-	struct sk_buff *hdr_skb;
+	const struct bcm43xx_dma_ops *ops = ring->ops;
+	struct bcm43xx_txhdr *header;
 	int slot;
-	struct bcm43xx_dmadesc *desc;
+	struct bcm43xx_dmadesc_generic *desc;
 	struct bcm43xx_dmadesc_meta *meta;
-	u32 desc_ctl;
-	u32 desc_addr;
 
+#define SLOTS_PER_PACKET  2
 	assert(skb_shinfo(skb)->nr_frags == 0);
 
-	hdr_skb = dev_alloc_skb(sizeof(struct bcm43xx_txhdr));
-	if (unlikely(!hdr_skb))
-		return -ENOMEM;
-	skb_put(hdr_skb, sizeof(struct bcm43xx_txhdr));
-
+	/* Get a slot for the header. */
 	slot = request_slot(ring);
-	desc = ring->vbase + slot;
-	meta = ring->meta + slot;
-
-	bcm43xx_generate_txhdr(ring->bcm,
-			       (struct bcm43xx_txhdr *)hdr_skb->data,
-			       skb->data, skb->len,
-			       ctl->first_fragment,
-			       generate_cookie(ring, slot),
-			       ctl);
-
-	meta->skb = hdr_skb;
-	meta->dmaaddr = map_descbuffer(ring, hdr_skb->data, hdr_skb->len, 1);
-	if (unlikely(meta->dmaaddr + hdr_skb->len > BCM43xx_DMA_BUSADDRMAX)) {
-		return_slot(ring, slot);
-		dev_kfree_skb_irq(hdr_skb);
-		printk(KERN_ERR PFX ">>>FATAL ERROR<<<  DMA TX SKB >1G "
-				    "(0x%08x, len: %u)\n",
-		       (unsigned int)meta->dmaaddr, hdr_skb->len);
-		return -ENOMEM;
-	}
+	desc = ops->idx2desc(ring, slot, &meta);
+	memset(meta, 0, sizeof(*meta));
 
-	desc_addr = (u32)(meta->dmaaddr + ring->memoffset);
-	desc_ctl = BCM43xx_DMADTOR_FRAMESTART |
-		   (BCM43xx_DMADTOR_BYTECNT_MASK & (u32)(hdr_skb->len));
-	if (slot == ring->nr_slots - 1)
-		desc_ctl |= BCM43xx_DMADTOR_DTABLEEND;
-	set_desc_ctl(desc, desc_ctl);
-	set_desc_addr(desc, desc_addr);
+	header = &(ring->txhdr_cache[slot]);
+	bcm43xx_generate_txhdr(ring->bcm, header,
+			       skb->data, skb->len, ctl,
+			       generate_cookie(ring, slot));
+
+	meta->dmaaddr = map_descbuffer(ring, (unsigned char *)header,
+				       sizeof(*header), 1);
+	ops->fill_descriptor(ring, desc, meta->dmaaddr,
+			     sizeof(*header), 1, 0, 0);
 
+	/* Get a slot for the payload. */
 	slot = request_slot(ring);
-	desc = ring->vbase + slot;
-	meta = ring->meta + slot;
+	desc = ops->idx2desc(ring, slot, &meta);
+	memset(meta, 0, sizeof(*meta));
 
-	/* We inspect the txstatus on the FRAMESTART descriptor later
-	 * on xmit-status IRQ.
-	 */
-	meta->must_xmit_txstat = 1;
-	memset(&meta->txstat, 0, sizeof(meta->txstat));
 	memcpy(&meta->txstat.control, ctl, sizeof(*ctl));
-
 	meta->skb = skb;
 	meta->dmaaddr = map_descbuffer(ring, skb->data, skb->len, 1);
-	if (unlikely(meta->dmaaddr + skb->len > BCM43xx_DMA_BUSADDRMAX)) {
-		return_slot(ring, prev_slot(ring, slot));
-		return_slot(ring, slot);
-		dev_kfree_skb_irq(hdr_skb);
-		printk(KERN_ERR PFX ">>>FATAL ERROR<<<  DMA TX SKB >1G "
-				    "(0x%08x, len: %u)\n",
-		       (unsigned int)meta->dmaaddr, skb->len);
-		return -ENOMEM;
-	}
+	meta->is_last_fragment = 1;
 
-	desc_addr = (u32)(meta->dmaaddr + ring->memoffset);
-	desc_ctl = (BCM43xx_DMADTOR_BYTECNT_MASK & (u32)(skb->len));
-	if (slot == ring->nr_slots - 1)
-		desc_ctl |= BCM43xx_DMADTOR_DTABLEEND;
+	ops->fill_descriptor(ring, desc, meta->dmaaddr,
+			     skb->len, 0, 1, 1);
 
-	desc_ctl |= BCM43xx_DMADTOR_FRAMEEND | BCM43xx_DMADTOR_COMPIRQ;
-	set_desc_ctl(desc, desc_ctl);
-	set_desc_addr(desc, desc_addr);
 	/* Now transfer the whole frame. */
-	dmacontroller_poke_tx(ring, slot);
-
-	return 0;
+	wmb();
+	ops->poke_tx(ring, next_slot(ring, slot));
 }
 
 int bcm43xx_dma_tx(struct bcm43xx_private *bcm,
@@ -768,75 +1035,76 @@
 		   struct ieee80211_tx_control *ctl)
 {
 	struct bcm43xx_dmaring *ring = bcm43xx_current_dma(bcm)->tx_ring1;
-	int err;
 
 	assert(ring->tx);
-
-#define SLOTS_PER_PACKET  2
 	if (unlikely(free_slots(ring) < SLOTS_PER_PACKET)) {
-		/* This should never trigger, as the ieee80211 stack
-		 * recognizes if the device queue is full and does
-		 * not send data anymore.
+		/* This should never trigger, as we call
+		 * ieee80211_stop_queue() when it's full.
 		 */
-		ieee80211_stop_queue(bcm->net_dev, 0);
-		printk(KERN_ERR PFX "DMA queue overflow\n");
+		printkl(KERN_ERR PFX "DMA queue overflow\n");
 		return NETDEV_TX_BUSY;
 	}
 
-	err = dma_tx_fragment(ring, skb, ctl);
-	if (likely(!err))
-		ring->nr_tx_packets++;
-	if (free_slots(ring) < SLOTS_PER_PACKET)
+	dma_tx_fragment(ring, skb, ctl);
+	ring->nr_tx_packets++;
+	if (free_slots(ring) < SLOTS_PER_PACKET) {
+		/* FIXME: we currently only have one queue */
 		ieee80211_stop_queue(bcm->net_dev, 0);
+		ring->stopped = 1;
+	}
 
-	return err;
+	return 0;
 }
 
 void bcm43xx_dma_handle_xmitstatus(struct bcm43xx_private *bcm,
 				   struct bcm43xx_xmitstatus *status)
 {
+	const struct bcm43xx_dma_ops *ops;
 	struct bcm43xx_dmaring *ring;
-	struct bcm43xx_dmadesc *desc;
+	struct bcm43xx_dmadesc_generic *desc;
 	struct bcm43xx_dmadesc_meta *meta;
-	int is_last_fragment;
 	int slot;
 
 	ring = parse_cookie(bcm, status->cookie, &slot);
 	assert(ring);
 	assert(ring->tx);
-	assert(get_desc_ctl(ring->vbase + slot) & BCM43xx_DMADTOR_FRAMESTART);
+	ops = ring->ops;
 	while (1) {
 		assert(slot >= 0 && slot < ring->nr_slots);
-		desc = ring->vbase + slot;
-		meta = ring->meta + slot;
+		desc = ops->idx2desc(ring, slot, &meta);
 
-		is_last_fragment = !!(get_desc_ctl(desc) & BCM43xx_DMADTOR_FRAMEEND);
 		unmap_descbuffer(ring, meta->dmaaddr, meta->skb->len, 1);
-
-		if (meta->must_xmit_txstat) {
-			meta->must_xmit_txstat = 0;
+		if (meta->is_last_fragment) {
 			/* Call back to inform the ieee80211 subsystem about the
 			 * status of the transmission.
 			 * Some fields of txstat are already filled in dma_tx().
 			 */
 			meta->txstat.ack = !!(status->flags & BCM43xx_TXSTAT_FLAG_ACK);
 			meta->txstat.retry_count = status->cnt2 - 1;
-			//FIXME: Fill in more information?
 			ieee80211_tx_status_irqsafe(bcm->net_dev, meta->skb, &(meta->txstat));
-			meta->skb = NULL;
-		} else
-			free_descriptor_buffer(ring, desc, meta, 1);
+			/* skb is freed by ieee80211_tx_status_irqsafe() */
+		} else {
+			/* No need to call free_descriptor_buffer here, as
+			 * this is only the txhdr, which is not allocated.
+			 */
+			assert(meta->skb == NULL);
+		}
 		/* Everything belonging to the slot is unmapped
 		 * and freed, so we can return it.
 		 */
 		return_slot(ring, slot);
 
-		if (is_last_fragment)
+		if (meta->is_last_fragment)
 			break;
 		slot = next_slot(ring, slot);
 	}
 	bcm->stats.last_tx = jiffies;
-	ieee80211_wake_queue(bcm->net_dev, 0);
+	if (ring->stopped) {
+		assert(free_slots(ring) >= SLOTS_PER_PACKET);
+		/* FIXME: we currently only have one queue */
+		ieee80211_wake_queue(bcm->net_dev, 0);
+		ring->stopped = 0;
+	}
 }
 
 void bcm43xx_dma_get_tx_stats(struct bcm43xx_private *bcm,
@@ -856,7 +1124,8 @@
 static void dma_rx(struct bcm43xx_dmaring *ring,
 		   int *slot)
 {
-	struct bcm43xx_dmadesc *desc;
+	const struct bcm43xx_dma_ops *ops = ring->ops;
+	struct bcm43xx_dmadesc_generic *desc;
 	struct bcm43xx_dmadesc_meta *meta;
 	struct bcm43xx_rxhdr *rxhdr;
 	struct sk_buff *skb;
@@ -864,13 +1133,12 @@
 	int err;
 	dma_addr_t dmaaddr;
 
-	desc = ring->vbase + *slot;
-	meta = ring->meta + *slot;
+	desc = ops->idx2desc(ring, *slot, &meta);
 
 	sync_descbuffer_for_cpu(ring, meta->dmaaddr, ring->rx_buffersize);
 	skb = meta->skb;
 
-	if (ring->mmio_base == BCM43xx_MMIO_DMA4_BASE) {
+	if (ring->index == 3) {
 		/* We received an xmit status. */
 		struct bcm43xx_hwxmitstatus *hw = (struct bcm43xx_hwxmitstatus *)skb->data;
 		struct bcm43xx_xmitstatus stat;
@@ -926,8 +1194,7 @@
 		s32 tmp = len;
 
 		while (1) {
-			desc = ring->vbase + *slot;
-			meta = ring->meta + *slot;
+			desc = ops->idx2desc(ring, *slot, &meta);
 			/* recycle the descriptor buffer. */
 			sync_descbuffer_for_device(ring, meta->dmaaddr,
 						   ring->rx_buffersize);
@@ -938,8 +1205,8 @@
 				break;
 		}
 		printkl(KERN_ERR PFX "DMA RX buffer too small "
-				     "(len: %u, buffer: %u, nr-dropped: %d)\n",
-		        len, ring->rx_buffersize, cnt);
+			"(len: %u, buffer: %u, nr-dropped: %d)\n",
+			len, ring->rx_buffersize, cnt);
 		goto drop;
 	}
 
@@ -963,17 +1230,14 @@
 
 void bcm43xx_dma_rx(struct bcm43xx_dmaring *ring)
 {
-	u32 status;
-	u16 descptr;
+	const struct bcm43xx_dma_ops *ops = ring->ops;
 	int slot, current_slot;
 #ifdef CONFIG_BCM43XX_D80211_DEBUG
 	int used_slots = 0;
 #endif
 
 	assert(!ring->tx);
-	status = bcm43xx_dma_read(ring, BCM43xx_DMA_RX_STATUS);
-	descptr = (status & BCM43xx_DMA_RXSTAT_DPTR_MASK);
-	current_slot = descptr / sizeof(struct bcm43xx_dmadesc);
+	current_slot = ops->get_current_rxslot(ring);
 	assert(current_slot >= 0 && current_slot < ring->nr_slots);
 
 	slot = ring->current_slot;
@@ -984,8 +1248,7 @@
 			ring->max_used_slots = used_slots;
 #endif
 	}
-	bcm43xx_dma_write(ring, BCM43xx_DMA_RX_DESC_INDEX,
-			  (u32)(slot * sizeof(struct bcm43xx_dmadesc)));
+	ops->set_current_rxslot(ring, slot);
 	ring->current_slot = slot;
 }
 
@@ -993,16 +1256,12 @@
 {
 	assert(ring->tx);
 	bcm43xx_power_saving_ctl_bits(ring->bcm, -1, 1);
-	bcm43xx_dma_write(ring, BCM43xx_DMA_TX_CONTROL,
-			  bcm43xx_dma_read(ring, BCM43xx_DMA_TX_CONTROL)
-			  | BCM43xx_DMA_TXCTRL_SUSPEND);
+	ring->ops->tx_suspend(ring);
 }
 
 void bcm43xx_dma_tx_resume(struct bcm43xx_dmaring *ring)
 {
 	assert(ring->tx);
-	bcm43xx_dma_write(ring, BCM43xx_DMA_TX_CONTROL,
-			  bcm43xx_dma_read(ring, BCM43xx_DMA_TX_CONTROL)
-			  & ~BCM43xx_DMA_TXCTRL_SUSPEND);
+	ring->ops->tx_resume(ring);
 	bcm43xx_power_saving_ctl_bits(ring->bcm, -1, -1);
 }
Index: wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_dma.h
===================================================================
--- wireless-dev.orig/drivers/net/wireless/d80211/bcm43xx/bcm43xx_dma.h	2006-08-15 00:11:47.000000000 +0200
+++ wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_dma.h	2006-08-16 17:34:10.000000000 +0200
@@ -16,63 +16,179 @@
 #define BCM43xx_DMAIRQ_NONFATALMASK	(1 << 13)
 #define BCM43xx_DMAIRQ_RX_DONE		(1 << 16)
 
-/* DMA controller register offsets. (relative to BCM43xx_DMA#_BASE) */
-#define BCM43xx_DMA_TX_CONTROL		0x00
-#define BCM43xx_DMA_TX_DESC_RING	0x04
-#define BCM43xx_DMA_TX_DESC_INDEX	0x08
-#define BCM43xx_DMA_TX_STATUS		0x0c
-#define BCM43xx_DMA_RX_CONTROL		0x10
-#define BCM43xx_DMA_RX_DESC_RING	0x14
-#define BCM43xx_DMA_RX_DESC_INDEX	0x18
-#define BCM43xx_DMA_RX_STATUS		0x1c
-
-/* DMA controller channel control word values. */
-#define BCM43xx_DMA_TXCTRL_ENABLE		(1 << 0)
-#define BCM43xx_DMA_TXCTRL_SUSPEND		(1 << 1)
-#define BCM43xx_DMA_TXCTRL_LOOPBACK		(1 << 2)
-#define BCM43xx_DMA_TXCTRL_FLUSH		(1 << 4)
-#define BCM43xx_DMA_RXCTRL_ENABLE		(1 << 0)
-#define BCM43xx_DMA_RXCTRL_FRAMEOFF_MASK	0x000000fe
-#define BCM43xx_DMA_RXCTRL_FRAMEOFF_SHIFT	1
-#define BCM43xx_DMA_RXCTRL_PIO			(1 << 8)
-/* DMA controller channel status word values. */
-#define BCM43xx_DMA_TXSTAT_DPTR_MASK		0x00000fff
-#define BCM43xx_DMA_TXSTAT_STAT_MASK		0x0000f000
-#define BCM43xx_DMA_TXSTAT_STAT_DISABLED	0x00000000
-#define BCM43xx_DMA_TXSTAT_STAT_ACTIVE		0x00001000
-#define BCM43xx_DMA_TXSTAT_STAT_IDLEWAIT	0x00002000
-#define BCM43xx_DMA_TXSTAT_STAT_STOPPED		0x00003000
-#define BCM43xx_DMA_TXSTAT_STAT_SUSP		0x00004000
-#define BCM43xx_DMA_TXSTAT_ERROR_MASK		0x000f0000
-#define BCM43xx_DMA_TXSTAT_FLUSHED		(1 << 20)
-#define BCM43xx_DMA_RXSTAT_DPTR_MASK		0x00000fff
-#define BCM43xx_DMA_RXSTAT_STAT_MASK		0x0000f000
-#define BCM43xx_DMA_RXSTAT_STAT_DISABLED	0x00000000
-#define BCM43xx_DMA_RXSTAT_STAT_ACTIVE		0x00001000
-#define BCM43xx_DMA_RXSTAT_STAT_IDLEWAIT	0x00002000
-#define BCM43xx_DMA_RXSTAT_STAT_RESERVED	0x00003000
-#define BCM43xx_DMA_RXSTAT_STAT_ERRORS		0x00004000
-#define BCM43xx_DMA_RXSTAT_ERROR_MASK		0x000f0000
-
-/* DMA descriptor control field values. */
-#define BCM43xx_DMADTOR_BYTECNT_MASK		0x00001fff
-#define BCM43xx_DMADTOR_DTABLEEND		(1 << 28) /* End of descriptor table */
-#define BCM43xx_DMADTOR_COMPIRQ			(1 << 29) /* IRQ on completion request */
-#define BCM43xx_DMADTOR_FRAMEEND		(1 << 30)
-#define BCM43xx_DMADTOR_FRAMESTART		(1 << 31)
+
+/*** 32-bit DMA Engine. ***/
+
+/* 32-bit DMA controller registers. */
+#define BCM43xx_DMA32_TXCTL				0x00
+#define		BCM43xx_DMA32_TXENABLE			0x00000001
+#define		BCM43xx_DMA32_TXSUSPEND			0x00000002
+#define		BCM43xx_DMA32_TXLOOPBACK		0x00000004
+#define		BCM43xx_DMA32_TXFLUSH			0x00000010
+#define		BCM43xx_DMA32_TXADDREXT_MASK		0x00030000
+#define		BCM43xx_DMA32_TXADDREXT_SHIFT		16
+#define BCM43xx_DMA32_TXRING				0x04
+#define BCM43xx_DMA32_TXINDEX				0x08
+#define BCM43xx_DMA32_TXSTATUS				0x0C
+#define		BCM43xx_DMA32_TXDPTR			0x00000FFF
+#define		BCM43xx_DMA32_TXSTATE			0x0000F000
+#define			BCM43xx_DMA32_TXSTAT_DISABLED	0x00000000
+#define			BCM43xx_DMA32_TXSTAT_ACTIVE	0x00001000
+#define			BCM43xx_DMA32_TXSTAT_IDLEWAIT	0x00002000
+#define			BCM43xx_DMA32_TXSTAT_STOPPED	0x00003000
+#define			BCM43xx_DMA32_TXSTAT_SUSP	0x00004000
+#define		BCM43xx_DMA32_TXERROR			0x000F0000
+#define			BCM43xx_DMA32_TXERR_NOERR	0x00000000
+#define			BCM43xx_DMA32_TXERR_PROT	0x00010000
+#define			BCM43xx_DMA32_TXERR_UNDERRUN	0x00020000
+#define			BCM43xx_DMA32_TXERR_BUFREAD	0x00030000
+#define			BCM43xx_DMA32_TXERR_DESCREAD	0x00040000
+#define		BCM43xx_DMA32_TXACTIVE			0xFFF00000
+#define BCM43xx_DMA32_RXCTL				0x10
+#define		BCM43xx_DMA32_RXENABLE			0x00000001
+#define		BCM43xx_DMA32_RXFROFF_MASK		0x000000FE
+#define		BCM43xx_DMA32_RXFROFF_SHIFT		1
+#define		BCM43xx_DMA32_RXDIRECTFIFO		0x00000100
+#define		BCM43xx_DMA32_RXADDREXT_MASK		0x00030000
+#define		BCM43xx_DMA32_RXADDREXT_SHIFT		16
+#define BCM43xx_DMA32_RXRING				0x14
+#define BCM43xx_DMA32_RXINDEX				0x18
+#define BCM43xx_DMA32_RXSTATUS				0x1C
+#define		BCM43xx_DMA32_RXDPTR			0x00000FFF
+#define		BCM43xx_DMA32_RXSTATE			0x0000F000
+#define			BCM43xx_DMA32_RXSTAT_DISABLED	0x00000000
+#define			BCM43xx_DMA32_RXSTAT_ACTIVE	0x00001000
+#define			BCM43xx_DMA32_RXSTAT_IDLEWAIT	0x00002000
+#define			BCM43xx_DMA32_RXSTAT_STOPPED	0x00003000
+#define		BCM43xx_DMA32_RXERROR			0x000F0000
+#define			BCM43xx_DMA32_RXERR_NOERR	0x00000000
+#define			BCM43xx_DMA32_RXERR_PROT	0x00010000
+#define			BCM43xx_DMA32_RXERR_OVERFLOW	0x00020000
+#define			BCM43xx_DMA32_RXERR_BUFWRITE	0x00030000
+#define			BCM43xx_DMA32_RXERR_DESCREAD	0x00040000
+#define		BCM43xx_DMA32_RXACTIVE			0xFFF00000
+
+/* 32-bit DMA descriptor. */
+struct bcm43xx_dmadesc32 {
+	__le32 control;
+	__le32 address;
+} __attribute__((__packed__));
+#define BCM43xx_DMA32_DCTL_BYTECNT		0x00001FFF
+#define BCM43xx_DMA32_DCTL_ADDREXT_MASK		0x00030000
+#define BCM43xx_DMA32_DCTL_ADDREXT_SHIFT	16
+#define BCM43xx_DMA32_DCTL_DTABLEEND		0x10000000
+#define BCM43xx_DMA32_DCTL_IRQ			0x20000000
+#define BCM43xx_DMA32_DCTL_FRAMEEND		0x40000000
+#define BCM43xx_DMA32_DCTL_FRAMESTART		0x80000000
+
+/* Address field Routing value. */
+#define BCM43xx_DMA32_ROUTING			0xC0000000
+#define BCM43xx_DMA32_ROUTING_SHIFT		30
+#define		BCM43xx_DMA32_NOTRANS		0x00000000
+#define		BCM43xx_DMA32_CLIENTTRANS	0x40000000
+
+
+
+/*** 64-bit DMA Engine. ***/
+
+/* 64-bit DMA controller registers. */
+#define BCM43xx_DMA64_TXCTL				0x00
+#define		BCM43xx_DMA64_TXENABLE			0x00000001
+#define		BCM43xx_DMA64_TXSUSPEND			0x00000002
+#define		BCM43xx_DMA64_TXLOOPBACK		0x00000004
+#define		BCM43xx_DMA64_TXFLUSH			0x00000010
+#define		BCM43xx_DMA64_TXADDREXT_MASK		0x00030000
+#define		BCM43xx_DMA64_TXADDREXT_SHIFT		16
+#define BCM43xx_DMA64_TXINDEX				0x04
+#define BCM43xx_DMA64_TXRINGLO				0x08
+#define BCM43xx_DMA64_TXRINGHI				0x0C
+#define BCM43xx_DMA64_TXSTATUS				0x10
+#define		BCM43xx_DMA64_TXSTATDPTR		0x00001FFF
+#define		BCM43xx_DMA64_TXSTAT			0xF0000000
+#define			BCM43xx_DMA64_TXSTAT_DISABLED	0x00000000
+#define			BCM43xx_DMA64_TXSTAT_ACTIVE	0x10000000
+#define			BCM43xx_DMA64_TXSTAT_IDLEWAIT	0x20000000
+#define			BCM43xx_DMA64_TXSTAT_STOPPED	0x30000000
+#define			BCM43xx_DMA64_TXSTAT_SUSP	0x40000000
+#define BCM43xx_DMA64_TXERROR				0x14
+#define		BCM43xx_DMA64_TXERRDPTR			0x0001FFFF
+#define		BCM43xx_DMA64_TXERR			0xF0000000
+#define			BCM43xx_DMA64_TXERR_NOERR	0x00000000
+#define			BCM43xx_DMA64_TXERR_PROT	0x10000000
+#define			BCM43xx_DMA64_TXERR_UNDERRUN	0x20000000
+#define			BCM43xx_DMA64_TXERR_TRANSFER	0x30000000
+#define			BCM43xx_DMA64_TXERR_DESCREAD	0x40000000
+#define			BCM43xx_DMA64_TXERR_CORE	0x50000000
+#define BCM43xx_DMA64_RXCTL				0x20
+#define		BCM43xx_DMA64_RXENABLE			0x00000001
+#define		BCM43xx_DMA64_RXFROFF_MASK		0x000000FE
+#define		BCM43xx_DMA64_RXFROFF_SHIFT		1
+#define		BCM43xx_DMA64_RXDIRECTFIFO		0x00000100
+#define		BCM43xx_DMA64_RXADDREXT_MASK		0x00030000
+#define		BCM43xx_DMA64_RXADDREXT_SHIFT		16
+#define BCM43xx_DMA64_RXINDEX				0x24
+#define BCM43xx_DMA64_RXRINGLO				0x28
+#define BCM43xx_DMA64_RXRINGHI				0x2C
+#define BCM43xx_DMA64_RXSTATUS				0x30
+#define		BCM43xx_DMA64_RXSTATDPTR		0x00001FFF
+#define		BCM43xx_DMA64_RXSTAT			0xF0000000
+#define			BCM43xx_DMA64_RXSTAT_DISABLED	0x00000000
+#define			BCM43xx_DMA64_RXSTAT_ACTIVE	0x10000000
+#define			BCM43xx_DMA64_RXSTAT_IDLEWAIT	0x20000000
+#define			BCM43xx_DMA64_RXSTAT_STOPPED	0x30000000
+#define			BCM43xx_DMA64_RXSTAT_SUSP	0x40000000
+#define BCM43xx_DMA64_RXERROR				0x34
+#define		BCM43xx_DMA64_RXERRDPTR			0x0001FFFF
+#define		BCM43xx_DMA64_RXERR			0xF0000000
+#define			BCM43xx_DMA64_RXERR_NOERR	0x00000000
+#define			BCM43xx_DMA64_RXERR_PROT	0x10000000
+#define			BCM43xx_DMA64_RXERR_UNDERRUN	0x20000000
+#define			BCM43xx_DMA64_RXERR_TRANSFER	0x30000000
+#define			BCM43xx_DMA64_RXERR_DESCREAD	0x40000000
+#define			BCM43xx_DMA64_RXERR_CORE	0x50000000
+
+/* 64-bit DMA descriptor. */
+struct bcm43xx_dmadesc64 {
+	__le32 control0;
+	__le32 control1;
+	__le32 address_low;
+	__le32 address_high;
+} __attribute__((__packed__));
+#define BCM43xx_DMA64_DCTL0_DTABLEEND		0x10000000
+#define BCM43xx_DMA64_DCTL0_IRQ			0x20000000
+#define BCM43xx_DMA64_DCTL0_FRAMEEND		0x40000000
+#define BCM43xx_DMA64_DCTL0_FRAMESTART		0x80000000
+#define BCM43xx_DMA64_DCTL1_BYTECNT		0x00001FFF
+#define BCM43xx_DMA64_DCTL1_ADDREXT_MASK	0x00030000
+#define BCM43xx_DMA64_DCTL1_ADDREXT_SHIFT	16
+
+/* Address field Routing value. */
+#define BCM43xx_DMA64_ROUTING			0xC0000000
+#define BCM43xx_DMA64_ROUTING_SHIFT		30
+#define		BCM43xx_DMA64_NOTRANS		0x00000000
+#define		BCM43xx_DMA64_CLIENTTRANS	0x80000000
+
+
+
+struct bcm43xx_dmadesc_generic {
+	union {
+		struct bcm43xx_dmadesc32 dma32;
+		struct bcm43xx_dmadesc64 dma64;
+	} __attribute__((__packed__));
+} __attribute__((__packed__));
+
 
 /* Misc DMA constants */
 #define BCM43xx_DMA_RINGMEMSIZE		PAGE_SIZE
-#define BCM43xx_DMA_BUSADDRMAX		0x3FFFFFFF
-#define BCM43xx_DMA_DMABUSADDROFFSET	(1 << 30)
-#define BCM43xx_DMA1_RX_FRAMEOFFSET	30
-#define BCM43xx_DMA4_RX_FRAMEOFFSET	0
+#define BCM43xx_DMA0_RX_FRAMEOFFSET	30
+#define BCM43xx_DMA3_RX_FRAMEOFFSET	0
+
 
 /* DMA engine tuning knobs */
-#define BCM43xx_TXRING_SLOTS		512
+#define BCM43xx_TXRING_SLOTS		128
 #define BCM43xx_RXRING_SLOTS		64
-#define BCM43xx_DMA1_RXBUFFERSIZE	(2304 + 100)
-#define BCM43xx_DMA4_RXBUFFERSIZE	16
+#define BCM43xx_DMA0_RX_BUFFERSIZE	(2304 + 100)
+#define BCM43xx_DMA3_RX_BUFFERSIZE	16
 
 
 
@@ -84,37 +200,50 @@
 struct bcm43xx_xmitstatus;
 
 
-struct bcm43xx_dmadesc {
-	__le32 _control;
-	__le32 _address;
-} __attribute__((__packed__));
-
-/* Macros to access the bcm43xx_dmadesc struct */
-#define get_desc_ctl(desc)		le32_to_cpu((desc)->_control)
-#define set_desc_ctl(desc, ctl)		do { (desc)->_control = cpu_to_le32(ctl); } while (0)
-#define get_desc_addr(desc)		le32_to_cpu((desc)->_address)
-#define set_desc_addr(desc, addr)	do { (desc)->_address = cpu_to_le32(addr); } while (0)
-
 struct bcm43xx_dmadesc_meta {
 	/* The kernel DMA-able buffer. */
 	struct sk_buff *skb;
 	/* DMA base bus-address of the descriptor buffer. */
 	dma_addr_t dmaaddr;
 	/* ieee80211 TX status. Only used once per 802.11 frag. */
-	u8 must_xmit_txstat:1;
+	u8 is_last_fragment;
 	struct ieee80211_tx_status txstat;
 };
 
+struct bcm43xx_dmaring;
+
+/* Lowlevel DMA operations that differ between 32bit and 64bit DMA. */
+struct bcm43xx_dma_ops {
+	struct bcm43xx_dmadesc_generic * (*idx2desc)(struct bcm43xx_dmaring *ring,
+						     int slot,
+						     struct bcm43xx_dmadesc_meta **meta);
+	void (*fill_descriptor)(struct bcm43xx_dmaring *ring,
+				struct bcm43xx_dmadesc_generic *desc,
+				dma_addr_t dmaaddr, u16 bufsize,
+				int start, int end, int irq);
+	void (*poke_tx)(struct bcm43xx_dmaring *ring, int slot);
+	void (*tx_suspend)(struct bcm43xx_dmaring *ring);
+	void (*tx_resume)(struct bcm43xx_dmaring *ring);
+	int (*get_current_rxslot)(struct bcm43xx_dmaring *ring);
+	void (*set_current_rxslot)(struct bcm43xx_dmaring *ring, int slot);
+};
+
 struct bcm43xx_dmaring {
-	struct bcm43xx_private *bcm;
+	/* Lowlevel DMA ops. */
+	const struct bcm43xx_dma_ops *ops;
 	/* Kernel virtual base address of the ring memory. */
-	struct bcm43xx_dmadesc *vbase;
-	/* DMA memory offset */
-	dma_addr_t memoffset;
-	/* (Unadjusted) DMA base bus-address of the ring memory. */
-	dma_addr_t dmabase;
+	void *descbase;
 	/* Meta data about all descriptors. */
 	struct bcm43xx_dmadesc_meta *meta;
+	/* Cache of struct bcm43xx_txhdr for each slot.
+	 * This is to avoid an allocation on each TX.
+	 * This is NULL for an RX ring.
+	 */
+	struct bcm43xx_txhdr *txhdr_cache;
+	/* DMA Routing value. */
+	u32 routing;
+	/* (Unadjusted) DMA base bus-address of the ring memory. */
+	dma_addr_t dmabase;
 	/* Number of descriptor slots in the ring. */
 	int nr_slots;
 	/* Number of used descriptor slots. */
@@ -127,12 +256,17 @@
 	u32 frameoffset;
 	/* Descriptor buffer size. */
 	u16 rx_buffersize;
-	/* The MMIO base register of the DMA controller, this
-	 * ring is posted to.
-	 */
+	/* The MMIO base register of the DMA controller. */
 	u16 mmio_base;
-	u8 tx:1,	/* TRUE, if this is a TX ring. */
-	   suspended:1;	/* TRUE, if transfers are suspended on this ring. */
+	/* DMA controller index number (0-5). */
+	int index;
+	/* Boolean. Is this a TX ring? */
+	u8 tx;
+	/* Boolean. 64bit DMA if true, 32bit DMA otherwise. */
+	u8 dma64;
+	/* Boolean. Is this ring stopped at ieee80211 level? */
+	u8 stopped;
+	struct bcm43xx_private *bcm;
 #ifdef CONFIG_BCM43XX_D80211_DEBUG
 	/* Maximum number of used slots. */
 	int max_used_slots;
@@ -159,9 +293,13 @@
 void bcm43xx_dma_free(struct bcm43xx_private *bcm);
 
 int bcm43xx_dmacontroller_rx_reset(struct bcm43xx_private *bcm,
-				   u16 dmacontroller_mmio_base);
+				   u16 dmacontroller_mmio_base,
+				   int dma64);
 int bcm43xx_dmacontroller_tx_reset(struct bcm43xx_private *bcm,
-				   u16 dmacontroller_mmio_base);
+				   u16 dmacontroller_mmio_base,
+				   int dma64);
+
+u16 bcm43xx_dmacontroller_base(int dma64bit, int dmacontroller_idx);
 
 void bcm43xx_dma_tx_suspend(struct bcm43xx_dmaring *ring);
 void bcm43xx_dma_tx_resume(struct bcm43xx_dmaring *ring);
@@ -177,7 +315,6 @@
 
 void bcm43xx_dma_rx(struct bcm43xx_dmaring *ring);
 
-
 #else /* CONFIG_BCM43XX_D80211_DMA */
 
 
@@ -192,13 +329,15 @@
 }
 static inline
 int bcm43xx_dmacontroller_rx_reset(struct bcm43xx_private *bcm,
-				   u16 dmacontroller_mmio_base)
+				   u16 dmacontroller_mmio_base,
+				   int dma64)
 {
 	return 0;
 }
 static inline
 int bcm43xx_dmacontroller_tx_reset(struct bcm43xx_private *bcm,
-				   u16 dmacontroller_mmio_base)
+				   u16 dmacontroller_mmio_base,
+				   int dma64)
 {
 	return 0;
 }
Index: wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_main.c
===================================================================
--- wireless-dev.orig/drivers/net/wireless/d80211/bcm43xx/bcm43xx_main.c	2006-08-16 12:37:02.000000000 +0200
+++ wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_main.c	2006-08-16 17:46:14.000000000 +0200
@@ -1337,6 +1337,7 @@
 	if ((bcm43xx_core_enabled(bcm)) &&
 	    !bcm43xx_using_pio(bcm)) {
 //FIXME: Do we _really_ want #ifndef CONFIG_BCM947XX here?
+#if 0
 #ifndef CONFIG_BCM947XX
 		/* reset all used DMA controllers. */
 		bcm43xx_dmacontroller_tx_reset(bcm, BCM43xx_MMIO_DMA1_BASE);
@@ -1347,6 +1348,7 @@
 		if (bcm->current_core->rev < 5)
 			bcm43xx_dmacontroller_rx_reset(bcm, BCM43xx_MMIO_DMA4_BASE);
 #endif
+#endif
 	}
 	if (bcm43xx_status(bcm) == BCM43xx_STAT_SHUTTINGDOWN) {
 		bcm43xx_write32(bcm, BCM43xx_MMIO_STATUS_BITFIELD,
@@ -1846,8 +1848,9 @@
 static void bcm43xx_interrupt_tasklet(struct bcm43xx_private *bcm)
 {
 	u32 reason;
-	u32 dma_reason[4];
-	int activity = 0;
+	u32 dma_reason[ARRAY_SIZE(bcm->dma_reason)];
+	u32 merged_dma_reason = 0;
+	int i, activity = 0;
 	unsigned long flags;
 
 #ifdef CONFIG_BCM43XX_D80211_DEBUG
@@ -1859,10 +1862,10 @@
 
 	spin_lock_irqsave(&bcm->irq_lock, flags);
 	reason = bcm->irq_reason;
-	dma_reason[0] = bcm->dma_reason[0];
-	dma_reason[1] = bcm->dma_reason[1];
-	dma_reason[2] = bcm->dma_reason[2];
-	dma_reason[3] = bcm->dma_reason[3];
+	for (i = 0; i < ARRAY_SIZE(dma_reason); i++) {
+		dma_reason[i] = bcm->dma_reason[i];
+		merged_dma_reason |= dma_reason[i];
+	}
 
 	if (unlikely(reason & BCM43xx_IRQ_XMIT_ERROR)) {
 		/* TX error. We get this when Template Ram is written in wrong endianess
@@ -1873,27 +1876,28 @@
 		printkl(KERN_ERR PFX "FATAL ERROR: BCM43xx_IRQ_XMIT_ERROR\n");
 		bcmirq_handled(BCM43xx_IRQ_XMIT_ERROR);
 	}
-	if (unlikely((dma_reason[0] & BCM43xx_DMAIRQ_FATALMASK) |
-		     (dma_reason[1] & BCM43xx_DMAIRQ_FATALMASK) |
-		     (dma_reason[2] & BCM43xx_DMAIRQ_FATALMASK) |
-		     (dma_reason[3] & BCM43xx_DMAIRQ_FATALMASK))) {
-		printkl(KERN_ERR PFX "FATAL ERROR: Fatal DMA error: "
-				     "0x%08X, 0x%08X, 0x%08X, 0x%08X\n",
-		        dma_reason[0], dma_reason[1],
-			dma_reason[2], dma_reason[3]);
-		bcm43xx_controller_restart(bcm, "DMA error");
-		mmiowb();
-		spin_unlock_irqrestore(&bcm->irq_lock, flags);
-		return;
-	}
-	if (unlikely((dma_reason[0] & BCM43xx_DMAIRQ_NONFATALMASK) |
-		     (dma_reason[1] & BCM43xx_DMAIRQ_NONFATALMASK) |
-		     (dma_reason[2] & BCM43xx_DMAIRQ_NONFATALMASK) |
-		     (dma_reason[3] & BCM43xx_DMAIRQ_NONFATALMASK))) {
-		printkl(KERN_ERR PFX "DMA error: "
-				     "0x%08X, 0x%08X, 0x%08X, 0x%08X\n",
-		        dma_reason[0], dma_reason[1],
-			dma_reason[2], dma_reason[3]);
+	if (unlikely(merged_dma_reason & (BCM43xx_DMAIRQ_FATALMASK |
+					  BCM43xx_DMAIRQ_NONFATALMASK))) {
+		if (merged_dma_reason & BCM43xx_DMAIRQ_FATALMASK) {
+			printkl(KERN_ERR PFX "FATAL ERROR: Fatal DMA error: "
+					     "0x%08X, 0x%08X, 0x%08X, "
+					     "0x%08X, 0x%08X, 0x%08X\n",
+			        dma_reason[0], dma_reason[1],
+				dma_reason[2], dma_reason[3],
+				dma_reason[4], dma_reason[5]);
+			bcm43xx_controller_restart(bcm, "DMA error");
+			mmiowb();
+			spin_unlock_irqrestore(&bcm->irq_lock, flags);
+			return;
+		}
+		if (merged_dma_reason & BCM43xx_DMAIRQ_NONFATALMASK) {
+			printkl(KERN_ERR PFX "DMA error: "
+					     "0x%08X, 0x%08X, 0x%08X, "
+					     "0x%08X, 0x%08X, 0x%08X\n",
+			        dma_reason[0], dma_reason[1],
+				dma_reason[2], dma_reason[3],
+				dma_reason[4], dma_reason[5]);
+		}
 	}
 
 	if (reason & BCM43xx_IRQ_PS) {
@@ -1928,8 +1932,6 @@
 	}
 
 	/* Check the DMA reason registers for received data. */
-	assert(!(dma_reason[1] & BCM43xx_DMAIRQ_RX_DONE));
-	assert(!(dma_reason[2] & BCM43xx_DMAIRQ_RX_DONE));
 	if (dma_reason[0] & BCM43xx_DMAIRQ_RX_DONE) {
 		if (bcm43xx_using_pio(bcm))
 			bcm43xx_pio_rx(bcm43xx_current_pio(bcm)->queue0);
@@ -1937,13 +1939,17 @@
 			bcm43xx_dma_rx(bcm43xx_current_dma(bcm)->rx_ring0);
 		/* We intentionally don't set "activity" to 1, here. */
 	}
+	assert(!(dma_reason[1] & BCM43xx_DMAIRQ_RX_DONE));
+	assert(!(dma_reason[2] & BCM43xx_DMAIRQ_RX_DONE));
 	if (dma_reason[3] & BCM43xx_DMAIRQ_RX_DONE) {
 		if (bcm43xx_using_pio(bcm))
 			bcm43xx_pio_rx(bcm43xx_current_pio(bcm)->queue3);
 		else
-			bcm43xx_dma_rx(bcm43xx_current_dma(bcm)->rx_ring1);
+			bcm43xx_dma_rx(bcm43xx_current_dma(bcm)->rx_ring3);
 		activity = 1;
 	}
+	assert(!(dma_reason[4] & BCM43xx_DMAIRQ_RX_DONE));
+	assert(!(dma_reason[5] & BCM43xx_DMAIRQ_RX_DONE));
 	bcmirq_handled(BCM43xx_IRQ_RX);
 
 	if (reason & BCM43xx_IRQ_XMIT_STATUS) {
@@ -2000,14 +2006,18 @@
 
 	bcm43xx_write32(bcm, BCM43xx_MMIO_GEN_IRQ_REASON, reason);
 
-	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA1_REASON,
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA0_REASON,
 			bcm->dma_reason[0]);
-	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA2_REASON,
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA1_REASON,
 			bcm->dma_reason[1]);
-	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA3_REASON,
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA2_REASON,
 			bcm->dma_reason[2]);
-	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA4_REASON,
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA3_REASON,
 			bcm->dma_reason[3]);
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA4_REASON,
+			bcm->dma_reason[4]);
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA5_REASON,
+			bcm->dma_reason[5]);
 }
 
 /* Interrupt handler top-half */
@@ -2035,14 +2045,18 @@
 	if (!reason)
 		goto out;
 
-	bcm->dma_reason[0] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA1_REASON)
-			     & 0x0001dc00;
-	bcm->dma_reason[1] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA2_REASON)
-			     & 0x0000dc00;
-	bcm->dma_reason[2] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA3_REASON)
-			     & 0x0000dc00;
-	bcm->dma_reason[3] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA4_REASON)
-			     & 0x0001dc00;
+	bcm->dma_reason[0] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA0_REASON)
+			     & 0x0001DC00;
+	bcm->dma_reason[1] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA1_REASON)
+			     & 0x0000DC00;
+	bcm->dma_reason[2] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA2_REASON)
+			     & 0x0000DC00;
+	bcm->dma_reason[3] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA3_REASON)
+			     & 0x0001DC00;
+	bcm->dma_reason[4] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA4_REASON)
+			     & 0x0000DC00;
+	bcm->dma_reason[5] = bcm43xx_read32(bcm, BCM43xx_MMIO_DMA5_REASON)
+			     & 0x0000DC00;
 
 	bcm43xx_interrupt_ack(bcm, reason);
 	/* disable all IRQs. They are enabled again in the bottom half. */
@@ -2640,10 +2654,12 @@
 		bcm43xx_write32(bcm, 0x018C, 0x02000000);
 	}
 	bcm43xx_write32(bcm, BCM43xx_MMIO_GEN_IRQ_REASON, 0x00004000);
-	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA1_IRQ_MASK, 0x0001DC00);
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA0_IRQ_MASK, 0x0001DC00);
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA1_IRQ_MASK, 0x0000DC00);
 	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA2_IRQ_MASK, 0x0000DC00);
-	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA3_IRQ_MASK, 0x0000DC00);
-	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA4_IRQ_MASK, 0x0001DC00);
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA3_IRQ_MASK, 0x0001DC00);
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA4_IRQ_MASK, 0x0000DC00);
+	bcm43xx_write32(bcm, BCM43xx_MMIO_DMA5_IRQ_MASK, 0x0000DC00);
 
 	value32 = bcm43xx_read32(bcm, BCM43xx_CIR_SBTMSTATELOW);
 	value32 |= 0x00100000;
Index: wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_pio.c
===================================================================
--- wireless-dev.orig/drivers/net/wireless/d80211/bcm43xx/bcm43xx_pio.c	2006-08-16 12:37:02.000000000 +0200
+++ wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_pio.c	2006-08-16 16:04:43.000000000 +0200
@@ -186,9 +186,8 @@
 	assert(skb_shinfo(skb)->nr_frags == 0);
 	bcm43xx_generate_txhdr(queue->bcm,
 			       &txhdr, skb->data, skb->len,
-			       packet->txstat.control.first_fragment,
-			       generate_cookie(queue, packet),
-			       &packet->txstat.control);
+			       &packet->txstat.control,
+			       generate_cookie(queue, packet));
 
 	tx_start(queue);
 	octets = skb->len + sizeof(txhdr);
Index: wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_xmit.c
===================================================================
--- wireless-dev.orig/drivers/net/wireless/d80211/bcm43xx/bcm43xx_xmit.c	2006-08-09 16:52:43.000000000 +0200
+++ wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_xmit.c	2006-08-16 16:04:02.000000000 +0200
@@ -275,9 +275,8 @@
 			    struct bcm43xx_txhdr *txhdr,
 			    const unsigned char *fragment_data,
 			    const unsigned int fragment_len,
-			    const int is_first_fragment,
-			    const u16 cookie,
-			    struct ieee80211_tx_control *txctl)
+			    struct ieee80211_tx_control *txctl,
+			    const u16 cookie)
 {
 	const struct bcm43xx_phyinfo *phy = bcm43xx_current_phy(bcm);
 	const struct ieee80211_hdr *wireless_header = (const struct ieee80211_hdr *)fragment_data;
@@ -347,7 +346,7 @@
 		flags |= 0x10; // FIXME: unknown meaning.
 	if (fallback_ofdm_modulation)
 		flags |= BCM43xx_TXHDRFLAG_FALLBACKOFDM;
-	if (is_first_fragment)
+	if (txctl->first_fragment)
 		flags |= BCM43xx_TXHDRFLAG_FIRSTFRAGMENT;
 
 	/* Set WSEC/RATE field */
Index: wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_xmit.h
===================================================================
--- wireless-dev.orig/drivers/net/wireless/d80211/bcm43xx/bcm43xx_xmit.h	2006-08-09 16:52:43.000000000 +0200
+++ wireless-dev/drivers/net/wireless/d80211/bcm43xx/bcm43xx_xmit.h	2006-08-16 16:01:38.000000000 +0200
@@ -82,9 +82,8 @@
 			    struct bcm43xx_txhdr *txhdr,
 			    const unsigned char *fragment_data,
 			    const unsigned int fragment_len,
-			    const int is_first_fragment,
-			    const u16 cookie,
-			    struct ieee80211_tx_control *txctl);
+			    struct ieee80211_tx_control *txctl,
+			    const u16 cookie);
 
 /* RX header as received from the hardware. */
 struct bcm43xx_rxhdr {


-- 
Greetings Michael.