Initial import

8 files changed, 1644 insertions, 0 deletions

diff --git a/arch/cris/arch-v32/mach-a3/Kconfig b/arch/cris/arch-v32/mach-a3/Kconfig
new file mode 100644
index 000000000..87547271a
--- /dev/null
+++ b/arch/cris/arch-v32/mach-a3/Kconfig
@@ -0,0 +1,110 @@
+if CRIS_MACH_ARTPEC3
+
+menu "Artpec-3 options"
+       depends on CRIS_MACH_ARTPEC3
+
+config ETRAX_DRAM_VIRTUAL_BASE
+	hex
+	default "c0000000"
+
+config ETRAX_L2CACHE
+       bool
+       default y
+
+config ETRAX_SERIAL_PORTS
+       int
+       default 5
+
+config ETRAX_DDR2_MRS
+	hex "DDR2 MRS"
+	default "0"
+
+config ETRAX_DDR2_TIMING
+	hex "DDR2 SDRAM timing"
+	default "0"
+	help
+	  SDRAM timing parameters.
+
+config ETRAX_DDR2_CONFIG
+	hex "DDR2 config"
+	default "0"
+
+config ETRAX_DDR2_LATENCY
+	hex "DDR2 latency"
+	default "0"
+
+config ETRAX_PIO_CE0_CFG
+       hex "PIO CE0 configuration"
+       default "0"
+
+config ETRAX_PIO_CE1_CFG
+       hex "PIO CE1 configuration"
+       default "0"
+
+config ETRAX_PIO_CE2_CFG
+       hex "PIO CE2 configuration"
+       default "0"
+
+config ETRAX_DEF_GIO_PA_OE
+	hex "GIO_PA_OE"
+	default "00000000"
+	help
+	  Configures the direction of general port A bits.  1 is out, 0 is in.
+	  This is often totally different depending on the product used.
+	  There are some guidelines though - if you know that only LED's are
+	  connected to port PA, then they are usually connected to bits 2-4
+	  and you can therefore use 1c.  On other boards which don't have the
+	  LED's at the general ports, these bits are used for all kinds of
+	  stuff.  If you don't know what to use, it is always safe to put all
+	  as inputs, although floating inputs isn't good.
+
+config ETRAX_DEF_GIO_PA_OUT
+	hex "GIO_PA_OUT"
+	default "00000000"
+	help
+	  Configures the initial data for the general port A bits.  Most
+	  products should use 00 here.
+
+config ETRAX_DEF_GIO_PB_OE
+	hex "GIO_PB_OE"
+	default "000000000"
+	help
+	  Configures the direction of general port B bits.  1 is out, 0 is in.
+	  This is often totally different depending on the product used.
+	  There are some guidelines though - if you know that only LED's are
+	  connected to port PA, then they are usually connected to bits 2-4
+	  and you can therefore use 1c.  On other boards which don't have the
+	  LED's at the general ports, these bits are used for all kinds of
+	  stuff.  If you don't know what to use, it is always safe to put all
+	  as inputs, although floating inputs isn't good.
+
+config ETRAX_DEF_GIO_PB_OUT
+	hex "GIO_PB_OUT"
+	default "000000000"
+	help
+	  Configures the initial data for the general port B bits.  Most
+	  products should use 00000 here.
+
+config ETRAX_DEF_GIO_PC_OE
+	hex "GIO_PC_OE"
+	default "00000"
+	help
+	  Configures the direction of general port C bits.  1 is out, 0 is in.
+	  This is often totally different depending on the product used.
+	  There are some guidelines though - if you know that only LED's are
+	  connected to port PA, then they are usually connected to bits 2-4
+	  and you can therefore use 1c.  On other boards which don't have the
+	  LED's at the general ports, these bits are used for all kinds of
+	  stuff.  If you don't know what to use, it is always safe to put all
+	  as inputs, although floating inputs isn't good.
+
+config ETRAX_DEF_GIO_PC_OUT
+	hex "GIO_PC_OUT"
+	default "00000"
+	help
+	  Configures the initial data for the general port C bits.  Most
+	  products should use 00000 here.
+
+endmenu
+
+endif
diff --git a/arch/cris/arch-v32/mach-a3/Makefile b/arch/cris/arch-v32/mach-a3/Makefile
new file mode 100644
index 000000000..18a227196
--- /dev/null
+++ b/arch/cris/arch-v32/mach-a3/Makefile
@@ -0,0 +1,8 @@
+#
+# Makefile for the linux kernel.
+#
+
+obj-y   := dma.o pinmux.o io.o arbiter.o
+
+clean:
+
diff --git a/arch/cris/arch-v32/mach-a3/arbiter.c b/arch/cris/arch-v32/mach-a3/arbiter.c
new file mode 100644
index 000000000..ab5c421a4
--- /dev/null
+++ b/arch/cris/arch-v32/mach-a3/arbiter.c
@@ -0,0 +1,634 @@
+/*
+ * Memory arbiter functions. Allocates bandwidth through the
+ * arbiter and sets up arbiter breakpoints.
+ *
+ * The algorithm first assigns slots to the clients that has specified
+ * bandwidth (e.g. ethernet) and then the remaining slots are divided
+ * on all the active clients.
+ *
+ * Copyright (c) 2004-2007 Axis Communications AB.
+ *
+ * The artpec-3 has two arbiters. The memory hierarchy looks like this:
+ *
+ *
+ * CPU DMAs
+ *  |   |
+ *  |   |
+ * --------------    ------------------
+ * | foo arbiter|----| Internal memory|
+ * --------------    ------------------
+ *      |
+ * --------------
+ * | L2 cache   |
+ * --------------
+ *             |
+ * h264 etc    |
+ *    |        |
+ *    |        |
+ * --------------
+ * | bar arbiter|
+ * --------------
+ *       |
+ * ---------
+ * | SDRAM |
+ * ---------
+ *
+ */
+
+#include <hwregs/reg_map.h>
+#include <hwregs/reg_rdwr.h>
+#include <hwregs/marb_foo_defs.h>
+#include <hwregs/marb_bar_defs.h>
+#include <arbiter.h>
+#include <hwregs/intr_vect.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/signal.h>
+#include <linux/errno.h>
+#include <linux/spinlock.h>
+#include <asm/io.h>
+#include <asm/irq_regs.h>
+
+#define D(x)
+
+struct crisv32_watch_entry {
+  unsigned long instance;
+  watch_callback *cb;
+  unsigned long start;
+  unsigned long end;
+  int used;
+};
+
+#define NUMBER_OF_BP 4
+#define SDRAM_BANDWIDTH 400000000
+#define INTMEM_BANDWIDTH 400000000
+#define NBR_OF_SLOTS 64
+#define NBR_OF_REGIONS 2
+#define NBR_OF_CLIENTS 15
+#define ARBITERS 2
+#define UNASSIGNED 100
+
+struct arbiter {
+  unsigned long instance;
+  int nbr_regions;
+  int nbr_clients;
+  int requested_slots[NBR_OF_REGIONS][NBR_OF_CLIENTS];
+  int active_clients[NBR_OF_REGIONS][NBR_OF_CLIENTS];
+};
+
+static struct crisv32_watch_entry watches[ARBITERS][NUMBER_OF_BP] =
+{
+  {
+  {regi_marb_foo_bp0},
+  {regi_marb_foo_bp1},
+  {regi_marb_foo_bp2},
+  {regi_marb_foo_bp3}
+  },
+  {
+  {regi_marb_bar_bp0},
+  {regi_marb_bar_bp1},
+  {regi_marb_bar_bp2},
+  {regi_marb_bar_bp3}
+  }
+};
+
+struct arbiter arbiters[ARBITERS] =
+{
+  { /* L2 cache arbiter */
+    .instance = regi_marb_foo,
+    .nbr_regions = 2,
+    .nbr_clients = 15
+  },
+  { /* DDR2 arbiter */
+    .instance = regi_marb_bar,
+    .nbr_regions = 1,
+    .nbr_clients = 9
+  }
+};
+
+static int max_bandwidth[NBR_OF_REGIONS] = {SDRAM_BANDWIDTH, INTMEM_BANDWIDTH};
+
+DEFINE_SPINLOCK(arbiter_lock);
+
+static irqreturn_t
+crisv32_foo_arbiter_irq(int irq, void *dev_id);
+static irqreturn_t
+crisv32_bar_arbiter_irq(int irq, void *dev_id);
+
+/*
+ * "I'm the arbiter, I know the score.
+ *  From square one I'll be watching all 64."
+ * (memory arbiter slots, that is)
+ *
+ *  Or in other words:
+ * Program the memory arbiter slots for "region" according to what's
+ * in requested_slots[] and active_clients[], while minimizing
+ * latency. A caller may pass a non-zero positive amount for
+ * "unused_slots", which must then be the unallocated, remaining
+ * number of slots, free to hand out to any client.
+ */
+
+static void crisv32_arbiter_config(int arbiter, int region, int unused_slots)
+{
+	int slot;
+	int client;
+	int interval = 0;
+
+	/*
+	 * This vector corresponds to the hardware arbiter slots (see
+	 * the hardware documentation for semantics). We initialize
+	 * each slot with a suitable sentinel value outside the valid
+	 * range {0 .. NBR_OF_CLIENTS - 1} and replace them with
+	 * client indexes. Then it's fed to the hardware.
+	 */
+	s8 val[NBR_OF_SLOTS];
+
+	for (slot = 0; slot < NBR_OF_SLOTS; slot++)
+	    val[slot] = -1;
+
+	for (client = 0; client < arbiters[arbiter].nbr_clients; client++) {
+	    int pos;
+	    /* Allocate the requested non-zero number of slots, but
+	     * also give clients with zero-requests one slot each
+	     * while stocks last. We do the latter here, in client
+	     * order. This makes sure zero-request clients are the
+	     * first to get to any spare slots, else those slots
+	     * could, when bandwidth is allocated close to the limit,
+	     * all be allocated to low-index non-zero-request clients
+	     * in the default-fill loop below. Another positive but
+	     * secondary effect is a somewhat better spread of the
+	     * zero-bandwidth clients in the vector, avoiding some of
+	     * the latency that could otherwise be caused by the
+	     * partitioning of non-zero-bandwidth clients at low
+	     * indexes and zero-bandwidth clients at high
+	     * indexes. (Note that this spreading can only affect the
+	     * unallocated bandwidth.)  All the above only matters for
+	     * memory-intensive situations, of course.
+	     */
+	    if (!arbiters[arbiter].requested_slots[region][client]) {
+		/*
+		 * Skip inactive clients. Also skip zero-slot
+		 * allocations in this pass when there are no known
+		 * free slots.
+		 */
+		if (!arbiters[arbiter].active_clients[region][client] ||
+				unused_slots <= 0)
+			continue;
+
+		unused_slots--;
+
+		/* Only allocate one slot for this client. */
+		interval = NBR_OF_SLOTS;
+	    } else
+		interval = NBR_OF_SLOTS /
+			arbiters[arbiter].requested_slots[region][client];
+
+	    pos = 0;
+	    while (pos < NBR_OF_SLOTS) {
+		if (val[pos] >= 0)
+		   pos++;
+		else {
+			val[pos] = client;
+			pos += interval;
+		}
+	    }
+	}
+
+	client = 0;
+	for (slot = 0; slot < NBR_OF_SLOTS; slot++) {
+		/*
+		 * Allocate remaining slots in round-robin
+		 * client-number order for active clients. For this
+		 * pass, we ignore requested bandwidth and previous
+		 * allocations.
+		 */
+		if (val[slot] < 0) {
+			int first = client;
+			while (!arbiters[arbiter].active_clients[region][client]) {
+				client = (client + 1) %
+					arbiters[arbiter].nbr_clients;
+				if (client == first)
+				   break;
+			}
+			val[slot] = client;
+			client = (client + 1) % arbiters[arbiter].nbr_clients;
+		}
+		if (arbiter == 0) {
+			if (region == EXT_REGION)
+				REG_WR_INT_VECT(marb_foo, regi_marb_foo,
+					rw_l2_slots, slot, val[slot]);
+			else if (region == INT_REGION)
+				REG_WR_INT_VECT(marb_foo, regi_marb_foo,
+					rw_intm_slots, slot, val[slot]);
+		} else {
+			REG_WR_INT_VECT(marb_bar, regi_marb_bar,
+				rw_ddr2_slots, slot, val[slot]);
+		}
+	}
+}
+
+extern char _stext, _etext;
+
+static void crisv32_arbiter_init(void)
+{
+	static int initialized;
+
+	if (initialized)
+		return;
+
+	initialized = 1;
+
+	/*
+	 * CPU caches are always set to active, but with zero
+	 * bandwidth allocated. It should be ok to allocate zero
+	 * bandwidth for the caches, because DMA for other channels
+	 * will supposedly finish, once their programmed amount is
+	 * done, and then the caches will get access according to the
+	 * "fixed scheme" for unclaimed slots. Though, if for some
+	 * use-case somewhere, there's a maximum CPU latency for
+	 * e.g. some interrupt, we have to start allocating specific
+	 * bandwidth for the CPU caches too.
+	 */
+	arbiters[0].active_clients[EXT_REGION][11] = 1;
+	arbiters[0].active_clients[EXT_REGION][12] = 1;
+	crisv32_arbiter_config(0, EXT_REGION, 0);
+	crisv32_arbiter_config(0, INT_REGION, 0);
+	crisv32_arbiter_config(1, EXT_REGION, 0);
+
+	if (request_irq(MEMARB_FOO_INTR_VECT, crisv32_foo_arbiter_irq,
+			0, "arbiter", NULL))
+		printk(KERN_ERR "Couldn't allocate arbiter IRQ\n");
+
+	if (request_irq(MEMARB_BAR_INTR_VECT, crisv32_bar_arbiter_irq,
+			0, "arbiter", NULL))
+		printk(KERN_ERR "Couldn't allocate arbiter IRQ\n");
+
+#ifndef CONFIG_ETRAX_KGDB
+	/* Global watch for writes to kernel text segment. */
+	crisv32_arbiter_watch(virt_to_phys(&_stext), &_etext - &_stext,
+		MARB_CLIENTS(arbiter_all_clients, arbiter_bar_all_clients),
+			      arbiter_all_write, NULL);
+#endif
+
+	/* Set up max burst sizes by default */
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_h264_rd_burst, 3);
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_h264_wr_burst, 3);
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_ccd_burst, 3);
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_vin_wr_burst, 3);
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_vin_rd_burst, 3);
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_sclr_rd_burst, 3);
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_vout_burst, 3);
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_sclr_fifo_burst, 3);
+	REG_WR_INT(marb_bar, regi_marb_bar, rw_l2cache_burst, 3);
+}
+
+int crisv32_arbiter_allocate_bandwidth(int client, int region,
+				      unsigned long bandwidth)
+{
+	int i;
+	int total_assigned = 0;
+	int total_clients = 0;
+	int req;
+	int arbiter = 0;
+
+	crisv32_arbiter_init();
+
+	if (client & 0xffff0000) {
+		arbiter = 1;
+		client >>= 16;
+	}
+
+	for (i = 0; i < arbiters[arbiter].nbr_clients; i++) {
+		total_assigned += arbiters[arbiter].requested_slots[region][i];
+		total_clients += arbiters[arbiter].active_clients[region][i];
+	}
+
+	/* Avoid division by 0 for 0-bandwidth requests. */
+	req = bandwidth == 0
+		? 0 : NBR_OF_SLOTS / (max_bandwidth[region] / bandwidth);
+
+	/*
+	 * We make sure that there are enough slots only for non-zero
+	 * requests. Requesting 0 bandwidth *may* allocate slots,
+	 * though if all bandwidth is allocated, such a client won't
+	 * get any and will have to rely on getting memory access
+	 * according to the fixed scheme that's the default when one
+	 * of the slot-allocated clients doesn't claim their slot.
+	 */
+	if (total_assigned + req > NBR_OF_SLOTS)
+	   return -ENOMEM;
+
+	arbiters[arbiter].active_clients[region][client] = 1;
+	arbiters[arbiter].requested_slots[region][client] = req;
+	crisv32_arbiter_config(arbiter, region, NBR_OF_SLOTS - total_assigned);
+
+	/* Propagate allocation from foo to bar */
+	if (arbiter == 0)
+		crisv32_arbiter_allocate_bandwidth(8 << 16,
+			EXT_REGION, bandwidth);
+	return 0;
+}
+
+/*
+ * Main entry for bandwidth deallocation.
+ *
+ * Strictly speaking, for a somewhat constant set of clients where
+ * each client gets a constant bandwidth and is just enabled or
+ * disabled (somewhat dynamically), no action is necessary here to
+ * avoid starvation for non-zero-allocation clients, as the allocated
+ * slots will just be unused. However, handing out those unused slots
+ * to active clients avoids needless latency if the "fixed scheme"
+ * would give unclaimed slots to an eager low-index client.
+ */
+
+void crisv32_arbiter_deallocate_bandwidth(int client, int region)
+{
+	int i;
+	int total_assigned = 0;
+	int arbiter = 0;
+
+	if (client & 0xffff0000)
+		arbiter = 1;
+
+	arbiters[arbiter].requested_slots[region][client] = 0;
+	arbiters[arbiter].active_clients[region][client] = 0;
+
+	for (i = 0; i < arbiters[arbiter].nbr_clients; i++)
+		total_assigned += arbiters[arbiter].requested_slots[region][i];
+
+	crisv32_arbiter_config(arbiter, region, NBR_OF_SLOTS - total_assigned);
+}
+
+int crisv32_arbiter_watch(unsigned long start, unsigned long size,
+			  unsigned long clients, unsigned long accesses,
+			  watch_callback *cb)
+{
+	int i;
+	int arbiter;
+	int used[2];
+	int ret = 0;
+
+	crisv32_arbiter_init();
+
+	if (start > 0x80000000) {
+		printk(KERN_ERR "Arbiter: %lX doesn't look like a "
+			"physical address", start);
+		return -EFAULT;
+	}
+
+	spin_lock(&arbiter_lock);
+
+	if (clients & 0xffff)
+		used[0] = 1;
+	if (clients & 0xffff0000)
+		used[1] = 1;
+
+	for (arbiter = 0; arbiter < ARBITERS; arbiter++) {
+		if (!used[arbiter])
+			continue;
+
+		for (i = 0; i < NUMBER_OF_BP; i++) {
+			if (!watches[arbiter][i].used) {
+				unsigned intr_mask;
+				if (arbiter)
+					intr_mask = REG_RD_INT(marb_bar,
+						regi_marb_bar, rw_intr_mask);
+				else
+					intr_mask = REG_RD_INT(marb_foo,
+						regi_marb_foo, rw_intr_mask);
+
+				watches[arbiter][i].used = 1;
+				watches[arbiter][i].start = start;
+				watches[arbiter][i].end = start + size;
+				watches[arbiter][i].cb = cb;
+
+				ret |= (i + 1) << (arbiter + 8);
+				if (arbiter) {
+					REG_WR_INT(marb_bar_bp,
+						watches[arbiter][i].instance,
+						rw_first_addr,
+						watches[arbiter][i].start);
+					REG_WR_INT(marb_bar_bp,
+						watches[arbiter][i].instance,
+						rw_last_addr,
+						watches[arbiter][i].end);
+					REG_WR_INT(marb_bar_bp,
+						watches[arbiter][i].instance,
+						rw_op, accesses);
+					REG_WR_INT(marb_bar_bp,
+						watches[arbiter][i].instance,
+						rw_clients,
+						clients & 0xffff);
+				} else {
+					REG_WR_INT(marb_foo_bp,
+						watches[arbiter][i].instance,
+						rw_first_addr,
+						watches[arbiter][i].start);
+					REG_WR_INT(marb_foo_bp,
+						watches[arbiter][i].instance,
+						rw_last_addr,
+						watches[arbiter][i].end);
+					REG_WR_INT(marb_foo_bp,
+						watches[arbiter][i].instance,
+						rw_op, accesses);
+					REG_WR_INT(marb_foo_bp,
+						watches[arbiter][i].instance,
+						rw_clients, clients >> 16);
+				}
+
+				if (i == 0)
+					intr_mask |= 1;
+				else if (i == 1)
+					intr_mask |= 2;
+				else if (i == 2)
+					intr_mask |= 4;
+				else if (i == 3)
+					intr_mask |= 8;
+
+				if (arbiter)
+					REG_WR_INT(marb_bar, regi_marb_bar,
+						rw_intr_mask, intr_mask);
+				else
+					REG_WR_INT(marb_foo, regi_marb_foo,
+						rw_intr_mask, intr_mask);
+
+				spin_unlock(&arbiter_lock);
+
+				break;
+			}
+		}
+	}
+	spin_unlock(&arbiter_lock);
+	if (ret)
+		return ret;
+	else
+		return -ENOMEM;
+}
+
+int crisv32_arbiter_unwatch(int id)
+{
+	int arbiter;
+	int intr_mask;
+
+	crisv32_arbiter_init();
+
+	spin_lock(&arbiter_lock);
+
+	for (arbiter = 0; arbiter < ARBITERS; arbiter++) {
+		int id2;
+
+		if (arbiter)
+			intr_mask = REG_RD_INT(marb_bar, regi_marb_bar,
+				rw_intr_mask);
+		else
+			intr_mask = REG_RD_INT(marb_foo, regi_marb_foo,
+				rw_intr_mask);
+
+		id2 = (id & (0xff << (arbiter + 8))) >> (arbiter + 8);
+		if (id2 == 0)
+			continue;
+		id2--;
+		if ((id2 >= NUMBER_OF_BP) || (!watches[arbiter][id2].used)) {
+			spin_unlock(&arbiter_lock);
+			return -EINVAL;
+		}
+
+		memset(&watches[arbiter][id2], 0,
+			sizeof(struct crisv32_watch_entry));
+
+		if (id2 == 0)
+			intr_mask &= ~1;
+		else if (id2 == 1)
+			intr_mask &= ~2;
+		else if (id2 == 2)
+			intr_mask &= ~4;
+		else if (id2 == 3)
+			intr_mask &= ~8;
+
+		if (arbiter)
+			REG_WR_INT(marb_bar, regi_marb_bar, rw_intr_mask,
+				intr_mask);
+		else
+			REG_WR_INT(marb_foo, regi_marb_foo, rw_intr_mask,
+				intr_mask);
+	}
+
+	spin_unlock(&arbiter_lock);
+	return 0;
+}
+
+extern void show_registers(struct pt_regs *regs);
+
+
+static irqreturn_t
+crisv32_foo_arbiter_irq(int irq, void *dev_id)
+{
+	reg_marb_foo_r_masked_intr masked_intr =
+		REG_RD(marb_foo, regi_marb_foo, r_masked_intr);
+	reg_marb_foo_bp_r_brk_clients r_clients;
+	reg_marb_foo_bp_r_brk_addr r_addr;
+	reg_marb_foo_bp_r_brk_op r_op;
+	reg_marb_foo_bp_r_brk_first_client r_first;
+	reg_marb_foo_bp_r_brk_size r_size;
+	reg_marb_foo_bp_rw_ack ack = {0};
+	reg_marb_foo_rw_ack_intr ack_intr = {
+		.bp0 = 1, .bp1 = 1, .bp2 = 1, .bp3 = 1
+	};
+	struct crisv32_watch_entry *watch;
+	unsigned arbiter = (unsigned)dev_id;
+
+	masked_intr = REG_RD(marb_foo, regi_marb_foo, r_masked_intr);
+
+	if (masked_intr.bp0)
+		watch = &watches[arbiter][0];
+	else if (masked_intr.bp1)
+		watch = &watches[arbiter][1];
+	else if (masked_intr.bp2)
+		watch = &watches[arbiter][2];
+	else if (masked_intr.bp3)
+		watch = &watches[arbiter][3];
+	else
+		return IRQ_NONE;
+
+	/* Retrieve all useful information and print it. */
+	r_clients = REG_RD(marb_foo_bp, watch->instance, r_brk_clients);
+	r_addr = REG_RD(marb_foo_bp, watch->instance, r_brk_addr);
+	r_op = REG_RD(marb_foo_bp, watch->instance, r_brk_op);
+	r_first = REG_RD(marb_foo_bp, watch->instance, r_brk_first_client);
+	r_size = REG_RD(marb_foo_bp, watch->instance, r_brk_size);
+
+	printk(KERN_DEBUG "Arbiter IRQ\n");
+	printk(KERN_DEBUG "Clients %X addr %X op %X first %X size %X\n",
+	       REG_TYPE_CONV(int, reg_marb_foo_bp_r_brk_clients, r_clients),
+	       REG_TYPE_CONV(int, reg_marb_foo_bp_r_brk_addr, r_addr),
+	       REG_TYPE_CONV(int, reg_marb_foo_bp_r_brk_op, r_op),
+	       REG_TYPE_CONV(int, reg_marb_foo_bp_r_brk_first_client, r_first),
+	       REG_TYPE_CONV(int, reg_marb_foo_bp_r_brk_size, r_size));
+
+	REG_WR(marb_foo_bp, watch->instance, rw_ack, ack);
+	REG_WR(marb_foo, regi_marb_foo, rw_ack_intr, ack_intr);
+
+	printk(KERN_DEBUG "IRQ occurred at %X\n", (unsigned)get_irq_regs());
+
+	if (watch->cb)
+		watch->cb();
+
+	return IRQ_HANDLED;
+}
+
+static irqreturn_t
+crisv32_bar_arbiter_irq(int irq, void *dev_id)
+{
+	reg_marb_bar_r_masked_intr masked_intr =
+		REG_RD(marb_bar, regi_marb_bar, r_masked_intr);
+	reg_marb_bar_bp_r_brk_clients r_clients;
+	reg_marb_bar_bp_r_brk_addr r_addr;
+	reg_marb_bar_bp_r_brk_op r_op;
+	reg_marb_bar_bp_r_brk_first_client r_first;
+	reg_marb_bar_bp_r_brk_size r_size;
+	reg_marb_bar_bp_rw_ack ack = {0};
+	reg_marb_bar_rw_ack_intr ack_intr = {
+		.bp0 = 1, .bp1 = 1, .bp2 = 1, .bp3 = 1
+	};
+	struct crisv32_watch_entry *watch;
+	unsigned arbiter = (unsigned)dev_id;
+
+	masked_intr = REG_RD(marb_bar, regi_marb_bar, r_masked_intr);
+
+	if (masked_intr.bp0)
+		watch = &watches[arbiter][0];
+	else if (masked_intr.bp1)
+		watch = &watches[arbiter][1];
+	else if (masked_intr.bp2)
+		watch = &watches[arbiter][2];
+	else if (masked_intr.bp3)
+		watch = &watches[arbiter][3];
+	else
+		return IRQ_NONE;
+
+	/* Retrieve all useful information and print it. */
+	r_clients = REG_RD(marb_bar_bp, watch->instance, r_brk_clients);
+	r_addr = REG_RD(marb_bar_bp, watch->instance, r_brk_addr);
+	r_op = REG_RD(marb_bar_bp, watch->instance, r_brk_op);
+	r_first = REG_RD(marb_bar_bp, watch->instance, r_brk_first_client);
+	r_size = REG_RD(marb_bar_bp, watch->instance, r_brk_size);
+
+	printk(KERN_DEBUG "Arbiter IRQ\n");
+	printk(KERN_DEBUG "Clients %X addr %X op %X first %X size %X\n",
+	       REG_TYPE_CONV(int, reg_marb_bar_bp_r_brk_clients, r_clients),
+	       REG_TYPE_CONV(int, reg_marb_bar_bp_r_brk_addr, r_addr),
+	       REG_TYPE_CONV(int, reg_marb_bar_bp_r_brk_op, r_op),
+	       REG_TYPE_CONV(int, reg_marb_bar_bp_r_brk_first_client, r_first),
+	       REG_TYPE_CONV(int, reg_marb_bar_bp_r_brk_size, r_size));
+
+	REG_WR(marb_bar_bp, watch->instance, rw_ack, ack);
+	REG_WR(marb_bar, regi_marb_bar, rw_ack_intr, ack_intr);
+
+	printk(KERN_DEBUG "IRQ occurred at %X\n", (unsigned)get_irq_regs()->erp);
+
+	if (watch->cb)
+		watch->cb();
+
+	return IRQ_HANDLED;
+}
+
diff --git a/arch/cris/arch-v32/mach-a3/dma.c b/arch/cris/arch-v32/mach-a3/dma.c
new file mode 100644
index 000000000..47c64bf40
--- /dev/null
+++ b/arch/cris/arch-v32/mach-a3/dma.c
@@ -0,0 +1,184 @@
+/* Wrapper for DMA channel allocator that starts clocks etc */
+
+#include <linux/kernel.h>
+#include <linux/spinlock.h>
+#include <mach/dma.h>
+#include <hwregs/reg_map.h>
+#include <hwregs/reg_rdwr.h>
+#include <hwregs/marb_defs.h>
+#include <hwregs/clkgen_defs.h>
+#include <hwregs/strmux_defs.h>
+#include <linux/errno.h>
+#include <arbiter.h>
+
+static char used_dma_channels[MAX_DMA_CHANNELS];
+static const char *used_dma_channels_users[MAX_DMA_CHANNELS];
+
+static DEFINE_SPINLOCK(dma_lock);
+
+int crisv32_request_dma(unsigned int dmanr, const char *device_id,
+	unsigned options, unsigned int bandwidth, enum dma_owner owner)
+{
+	unsigned long flags;
+	reg_clkgen_rw_clk_ctrl clk_ctrl;
+	reg_strmux_rw_cfg strmux_cfg;
+
+	if (crisv32_arbiter_allocate_bandwidth(dmanr,
+			options & DMA_INT_MEM ? INT_REGION : EXT_REGION,
+			bandwidth))
+		return -ENOMEM;
+
+	spin_lock_irqsave(&dma_lock, flags);
+
+	if (used_dma_channels[dmanr]) {
+		spin_unlock_irqrestore(&dma_lock, flags);
+		if (options & DMA_VERBOSE_ON_ERROR)
+			printk(KERN_ERR "Failed to request DMA %i for %s, "
+				"already allocated by %s\n",
+				dmanr,
+				device_id,
+				used_dma_channels_users[dmanr]);
+
+		if (options & DMA_PANIC_ON_ERROR)
+			panic("request_dma error!");
+		spin_unlock_irqrestore(&dma_lock, flags);
+		return -EBUSY;
+	}
+	clk_ctrl = REG_RD(clkgen, regi_clkgen, rw_clk_ctrl);
+	strmux_cfg = REG_RD(strmux, regi_strmux, rw_cfg);
+
+	switch (dmanr) {
+	case 0:
+	case 1:
+		clk_ctrl.dma0_1_eth = 1;
+		break;
+	case 2:
+	case 3:
+		clk_ctrl.dma2_3_strcop = 1;
+		break;
+	case 4:
+	case 5:
+		clk_ctrl.dma4_5_iop = 1;
+		break;
+	case 6:
+	case 7:
+		clk_ctrl.sser_ser_dma6_7 = 1;
+		break;
+	case 9:
+	case 11:
+		clk_ctrl.dma9_11 = 1;
+		break;
+#if MAX_DMA_CHANNELS-1 != 11
+#error Check dma.c
+#endif
+	default:
+		spin_unlock_irqrestore(&dma_lock, flags);
+		if (options & DMA_VERBOSE_ON_ERROR)
+			printk(KERN_ERR "Failed to request DMA %i for %s, "
+				"only 0-%i valid)\n",
+				dmanr, device_id, MAX_DMA_CHANNELS-1);
+
+		if (options & DMA_PANIC_ON_ERROR)
+			panic("request_dma error!");
+		return -EINVAL;
+	}
+
+	switch (owner) {
+	case dma_eth:
+		if (dmanr == 0)
+			strmux_cfg.dma0 = regk_strmux_eth;
+		else if (dmanr == 1)
+			strmux_cfg.dma1 = regk_strmux_eth;
+		else
+			panic("Invalid DMA channel for eth\n");
+		break;
+	case dma_ser0:
+		if (dmanr == 0)
+			strmux_cfg.dma0 = regk_strmux_ser0;
+		else if (dmanr == 1)
+			strmux_cfg.dma1 = regk_strmux_ser0;
+		else
+			panic("Invalid DMA channel for ser0\n");
+		break;
+	case dma_ser3:
+		if (dmanr == 2)
+			strmux_cfg.dma2 = regk_strmux_ser3;
+		else if (dmanr == 3)
+			strmux_cfg.dma3 = regk_strmux_ser3;
+		else
+			panic("Invalid DMA channel for ser3\n");
+		break;
+	case dma_strp:
+		if (dmanr == 2)
+			strmux_cfg.dma2 = regk_strmux_strcop;
+		else if (dmanr == 3)
+			strmux_cfg.dma3 = regk_strmux_strcop;
+		else
+			panic("Invalid DMA channel for strp\n");
+		break;
+	case dma_ser1:
+		if (dmanr == 4)
+			strmux_cfg.dma4 = regk_strmux_ser1;
+		else if (dmanr == 5)
+			strmux_cfg.dma5 = regk_strmux_ser1;
+		else
+			panic("Invalid DMA channel for ser1\n");
+		break;
+	case dma_iop:
+		if (dmanr == 4)
+			strmux_cfg.dma4 = regk_strmux_iop;
+		else if (dmanr == 5)
+			strmux_cfg.dma5 = regk_strmux_iop;
+		else
+			panic("Invalid DMA channel for iop\n");
+		break;
+	case dma_ser2:
+		if (dmanr == 6)
+			strmux_cfg.dma6 = regk_strmux_ser2;
+		else if (dmanr == 7)
+			strmux_cfg.dma7 = regk_strmux_ser2;
+		else
+			panic("Invalid DMA channel for ser2\n");
+		break;
+	case dma_sser:
+		if (dmanr == 6)
+			strmux_cfg.dma6 = regk_strmux_sser;
+		else if (dmanr == 7)
+			strmux_cfg.dma7 = regk_strmux_sser;
+		else
+			panic("Invalid DMA channel for sser\n");
+		break;
+	case dma_ser4:
+		if (dmanr == 9)
+			strmux_cfg.dma9 = regk_strmux_ser4;
+		else
+			panic("Invalid DMA channel for ser4\n");
+		break;
+	case dma_jpeg:
+		if (dmanr == 9)
+			strmux_cfg.dma9 = regk_strmux_jpeg;
+		else
+			panic("Invalid DMA channel for JPEG\n");
+		break;
+	case dma_h264:
+		if (dmanr == 11)
+			strmux_cfg.dma11 = regk_strmux_h264;
+		else
+			panic("Invalid DMA channel for H264\n");
+		break;
+	}
+
+	used_dma_channels[dmanr] = 1;
+	used_dma_channels_users[dmanr] = device_id;
+	REG_WR(clkgen, regi_clkgen, rw_clk_ctrl, clk_ctrl);
+	REG_WR(strmux, regi_strmux, rw_cfg, strmux_cfg);
+	spin_unlock_irqrestore(&dma_lock, flags);
+	return 0;
+}
+
+void crisv32_free_dma(unsigned int dmanr)
+{
+	spin_lock(&dma_lock);
+	used_dma_channels[dmanr] = 0;
+	spin_unlock(&dma_lock);
+}
diff --git a/arch/cris/arch-v32/mach-a3/dram_init.S b/arch/cris/arch-v32/mach-a3/dram_init.S
new file mode 100644
index 000000000..ec8648be3
--- /dev/null
+++ b/arch/cris/arch-v32/mach-a3/dram_init.S
@@ -0,0 +1,118 @@
+/*
+ * DDR SDRAM initialization - alter with care
+ * This file is intended to be included from other assembler files
+ *
+ * Note: This file may not modify r8 or r9 because they are used to
+ * carry information from the decompresser to the kernel
+ *
+ * Copyright (C) 2005-2007 Axis Communications AB
+ *
+ * Authors:  Mikael Starvik <starvik@axis.com>
+ */
+
+/* Just to be certain the config file is included, we include it here
+ * explicitely instead of depending on it being included in the file that
+ * uses this code.
+ */
+
+#include <hwregs/asm/reg_map_asm.h>
+#include <hwregs/asm/ddr2_defs_asm.h>
+
+	;; WARNING! The registers r8 and r9 are used as parameters carrying
+	;; information from the decompressor (if the kernel was compressed).
+	;; They should not be used in the code below.
+
+	;; Refer to ddr2 MDS for initialization sequence
+
+	; 2. Wait 200us
+	move.d   10000, $r2
+1:	bne      1b
+	subq     1, $r2
+
+	; Start clock
+	move.d   REG_ADDR(ddr2, regi_ddr2_ctrl, rw_phy_cfg), $r0
+	move.d   REG_STATE(ddr2, rw_phy_cfg, en, yes), $r1
+	move.d   $r1, [$r0]
+
+	; 2. Wait 200us
+	move.d   10000, $r2
+1:	bne      1b
+	subq     1, $r2
+
+	; Reset phy and start calibration
+	move.d   REG_ADDR(ddr2, regi_ddr2_ctrl, rw_phy_ctrl), $r0
+	move.d   REG_STATE(ddr2, rw_phy_ctrl, rst, yes) | \
+		 REG_STATE(ddr2, rw_phy_ctrl, cal_rst, yes), $r1
+	move.d   $r1, [$r0]
+	move.d	 REG_STATE(ddr2, rw_phy_ctrl, cal_start, yes), $r1
+	move.d   $r1, [$r0]
+
+	; 2. Wait 200us
+	move.d   10000, $r2
+1:	bne      1b
+	subq     1, $r2
+
+	; Issue commands
+	move.d   REG_ADDR(ddr2, regi_ddr2_ctrl, rw_ctrl), $r0
+	move.d   sdram_commands_start, $r2
+command_loop:
+	movu.b  [$r2+], $r1
+	movu.w  [$r2+], $r3
+do_cmd:
+	lslq     16, $r1
+	or.d     $r3, $r1
+	move.d   $r1, [$r0]
+	; 2. Wait 200us
+	move.d   10000, $r4
+1:	bne      1b
+	subq     1, $r4
+	cmp.d    sdram_commands_end, $r2
+	blo      command_loop
+	nop
+
+	; Set timing
+	move.d   REG_ADDR(ddr2, regi_ddr2_ctrl, rw_timing), $r0
+	move.d   CONFIG_ETRAX_DDR2_TIMING, $r1
+	move.d   $r1, [$r0]
+
+	; Set latency
+	move.d   REG_ADDR(ddr2, regi_ddr2_ctrl, rw_latency), $r0
+	move.d   CONFIG_ETRAX_DDR2_LATENCY, $r1
+	move.d   $r1, [$r0]
+
+	; Set configuration
+	move.d   REG_ADDR(ddr2, regi_ddr2_ctrl, rw_cfg), $r0
+	move.d   CONFIG_ETRAX_DDR2_CONFIG, $r1
+	move.d   $r1, [$r0]
+
+	ba after_sdram_commands
+	nop
+
+sdram_commands_start:
+	.byte regk_ddr2_deselect
+	.word 0
+	.byte regk_ddr2_pre
+	.word regk_ddr2_pre_all
+	.byte regk_ddr2_emrs2
+	.word 0
+	.byte regk_ddr2_emrs3
+	.word 0
+	.byte regk_ddr2_emrs
+	.word regk_ddr2_dll_en
+	.byte regk_ddr2_mrs
+	.word regk_ddr2_dll_rst
+	.byte regk_ddr2_pre
+	.word regk_ddr2_pre_all
+	.byte regk_ddr2_ref
+	.word 0
+	.byte regk_ddr2_ref
+	.word 0
+	.byte regk_ddr2_mrs
+	.word CONFIG_ETRAX_DDR2_MRS & 0xffff
+	.byte regk_ddr2_emrs
+	.word regk_ddr2_ocd_default | regk_ddr2_dll_en
+	.byte regk_ddr2_emrs
+	.word regk_ddr2_ocd_exit | regk_ddr2_dll_en | (CONFIG_ETRAX_DDR2_MRS >> 16)
+sdram_commands_end:
+	.align 1
+after_sdram_commands:
diff --git a/arch/cris/arch-v32/mach-a3/hw_settings.S b/arch/cris/arch-v32/mach-a3/hw_settings.S
new file mode 100644
index 000000000..0145725a1
--- /dev/null
+++ b/arch/cris/arch-v32/mach-a3/hw_settings.S
@@ -0,0 +1,53 @@
+/*
+ * This table is used by some tools to extract hardware parameters.
+ * The table should be included in the kernel and the decompressor.
+ * Don't forget to update the tools if you change this table.
+ *
+ * Copyright (C) 2001-2007 Axis Communications AB
+ *
+ * Authors:  Mikael Starvik <starvik@axis.com>
+ */
+
+#include <hwregs/asm/reg_map_asm.h>
+#include <hwregs/asm/ddr2_defs_asm.h>
+#include <hwregs/asm/gio_defs_asm.h>
+
+	.ascii "HW_PARAM_MAGIC" ; Magic number
+	.dword 0xc0004000	; Kernel start address
+
+	; Debug port
+#ifdef CONFIG_ETRAX_DEBUG_PORT0
+	.dword 0
+#elif defined(CONFIG_ETRAX_DEBUG_PORT1)
+	.dword 1
+#elif defined(CONFIG_ETRAX_DEBUG_PORT2)
+	.dword 2
+#elif defined(CONFIG_ETRAX_DEBUG_PORT3)
+	.dword 3
+#else
+	.dword 4 ; No debug
+#endif
+
+	; Register values
+	.dword REG_ADDR(ddr2, regi_ddr2_ctrl, rw_cfg)
+	.dword CONFIG_ETRAX_DDR2_CONFIG
+	.dword REG_ADDR(ddr2, regi_ddr2_ctrl, rw_latency)
+	.dword CONFIG_ETRAX_DDR2_LATENCY
+	.dword REG_ADDR(ddr2, regi_ddr2_ctrl, rw_timing)
+	.dword CONFIG_ETRAX_DDR2_TIMING
+	.dword CONFIG_ETRAX_DDR2_MRS
+
+	.dword REG_ADDR(gio, regi_gio, rw_pa_dout)
+	.dword CONFIG_ETRAX_DEF_GIO_PA_OUT
+	.dword REG_ADDR(gio, regi_gio, rw_pa_oe)
+	.dword CONFIG_ETRAX_DEF_GIO_PA_OE
+	.dword REG_ADDR(gio, regi_gio, rw_pb_dout)
+	.dword CONFIG_ETRAX_DEF_GIO_PB_OUT
+	.dword REG_ADDR(gio, regi_gio, rw_pb_oe)
+	.dword CONFIG_ETRAX_DEF_GIO_PB_OE
+	.dword REG_ADDR(gio, regi_gio, rw_pc_dout)
+	.dword CONFIG_ETRAX_DEF_GIO_PC_OUT
+	.dword REG_ADDR(gio, regi_gio, rw_pc_oe)
+	.dword CONFIG_ETRAX_DEF_GIO_PC_OE
+
+	.dword 0 ; No more register values
diff --git a/arch/cris/arch-v32/mach-a3/io.c b/arch/cris/arch-v32/mach-a3/io.c
new file mode 100644
index 000000000..090ceb99e
--- /dev/null
+++ b/arch/cris/arch-v32/mach-a3/io.c
@@ -0,0 +1,149 @@
+/*
+ * Helper functions for I/O pins.
+ *
+ * Copyright (c) 2005-2007 Axis Communications AB.
+ */
+
+#include <linux/types.h>
+#include <linux/errno.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/ctype.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <asm/io.h>
+#include <mach/pinmux.h>
+#include <hwregs/gio_defs.h>
+
+struct crisv32_ioport crisv32_ioports[] = {
+	{
+		(unsigned long *)REG_ADDR(gio, regi_gio, rw_pa_oe),
+		(unsigned long *)REG_ADDR(gio, regi_gio, rw_pa_dout),
+		(unsigned long *)REG_ADDR(gio, regi_gio, r_pa_din),
+		32
+	},
+	{
+		(unsigned long *)REG_ADDR(gio, regi_gio, rw_pb_oe),
+		(unsigned long *)REG_ADDR(gio, regi_gio, rw_pb_dout),
+		(unsigned long *)REG_ADDR(gio, regi_gio, r_pb_din),
+		32
+	},
+	{
+		(unsigned long *)REG_ADDR(gio, regi_gio, rw_pc_oe),
+		(unsigned long *)REG_ADDR(gio, regi_gio, rw_pc_dout),
+		(unsigned long *)REG_ADDR(gio, regi_gio, r_pc_din),
+		16
+	},
+};
+
+#define NBR_OF_PORTS ARRAY_SIZE(crisv32_ioports)
+
+struct crisv32_iopin crisv32_led_net0_green;
+struct crisv32_iopin crisv32_led_net0_red;
+struct crisv32_iopin crisv32_led2_green;
+struct crisv32_iopin crisv32_led2_red;
+struct crisv32_iopin crisv32_led3_green;
+struct crisv32_iopin crisv32_led3_red;
+
+/* Dummy port used when green LED and red LED is on the same bit */
+static unsigned long io_dummy;
+static struct crisv32_ioport dummy_port = {
+	&io_dummy,
+	&io_dummy,
+	&io_dummy,
+	32
+};
+static struct crisv32_iopin dummy_led = {
+	&dummy_port,
+	0
+};
+
+static int __init crisv32_io_init(void)
+{
+	int ret = 0;
+
+	u32 i;
+
+	/* Locks *should* be dynamically initialized. */
+	for (i = 0; i < ARRAY_SIZE(crisv32_ioports); i++)
+		spin_lock_init(&crisv32_ioports[i].lock);
+	spin_lock_init(&dummy_port.lock);
+
+	/* Initialize LEDs */
+#if (defined(CONFIG_ETRAX_NBR_LED_GRP_ONE) || defined(CONFIG_ETRAX_NBR_LED_GRP_TWO))
+	ret += crisv32_io_get_name(&crisv32_led_net0_green,
+		CONFIG_ETRAX_LED_G_NET0);
+	crisv32_io_set_dir(&crisv32_led_net0_green, crisv32_io_dir_out);
+	if (strcmp(CONFIG_ETRAX_LED_G_NET0, CONFIG_ETRAX_LED_R_NET0)) {
+		ret += crisv32_io_get_name(&crisv32_led_net0_red,
+			CONFIG_ETRAX_LED_R_NET0);
+		crisv32_io_set_dir(&crisv32_led_net0_red, crisv32_io_dir_out);
+	} else
+		crisv32_led_net0_red = dummy_led;
+#endif
+
+	ret += crisv32_io_get_name(&crisv32_led2_green, CONFIG_ETRAX_V32_LED2G);
+	ret += crisv32_io_get_name(&crisv32_led2_red, CONFIG_ETRAX_V32_LED2R);
+	ret += crisv32_io_get_name(&crisv32_led3_green, CONFIG_ETRAX_V32_LED3G);
+	ret += crisv32_io_get_name(&crisv32_led3_red, CONFIG_ETRAX_V32_LED3R);
+
+	crisv32_io_set_dir(&crisv32_led2_green, crisv32_io_dir_out);
+	crisv32_io_set_dir(&crisv32_led2_red, crisv32_io_dir_out);
+	crisv32_io_set_dir(&crisv32_led3_green, crisv32_io_dir_out);
+	crisv32_io_set_dir(&crisv32_led3_red, crisv32_io_dir_out);
+
+	return ret;
+}
+
+__initcall(crisv32_io_init);
+
+int crisv32_io_get(struct crisv32_iopin *iopin,
+	unsigned int port, unsigned int pin)
+{
+	if (port > NBR_OF_PORTS)
+		return -EINVAL;
+	if (port > crisv32_ioports[port].pin_count)
+		return -EINVAL;
+
+	iopin->bit = 1 << pin;
+	iopin->port = &crisv32_ioports[port];
+
+	if (crisv32_pinmux_alloc(port, pin, pin, pinmux_gpio))
+		return -EIO;
+
+	return 0;
+}
+
+int crisv32_io_get_name(struct crisv32_iopin *iopin, const char *name)
+{
+	int port;
+	int pin;
+
+	if (toupper(*name) == 'P')
+		name++;
+
+	if (toupper(*name) < 'A' || toupper(*name) > 'E')
+		return -EINVAL;
+
+	port = toupper(*name) - 'A';
+	name++;
+	pin = simple_strtoul(name, NULL, 10);
+
+	if (pin < 0 || pin > crisv32_ioports[port].pin_count)
+		return -EINVAL;
+
+	iopin->bit = 1 << pin;
+	iopin->port = &crisv32_ioports[port];
+
+	if (crisv32_pinmux_alloc(port, pin, pin, pinmux_gpio))
+		return -EIO;
+
+	return 0;
+}
+
+#ifdef CONFIG_PCI
+/* PCI I/O access stuff */
+struct cris_io_operations *cris_iops = NULL;
+EXPORT_SYMBOL(cris_iops);
+#endif
+
diff --git a/arch/cris/arch-v32/mach-a3/pinmux.c b/arch/cris/arch-v32/mach-a3/pinmux.c
new file mode 100644
index 000000000..591f77526
--- /dev/null
+++ b/arch/cris/arch-v32/mach-a3/pinmux.c
@@ -0,0 +1,388 @@
+/*
+ * Allocator for I/O pins. All pins are allocated to GPIO at bootup.
+ * Unassigned pins and GPIO pins can be allocated to a fixed interface
+ * or the I/O processor instead.
+ *
+ * Copyright (c) 2005-2007 Axis Communications AB.
+ */
+
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/spinlock.h>
+#include <hwregs/reg_map.h>
+#include <hwregs/reg_rdwr.h>
+#include <pinmux.h>
+#include <hwregs/pinmux_defs.h>
+#include <hwregs/clkgen_defs.h>
+
+#undef DEBUG
+
+#define PINS 80
+#define PORT_PINS 32
+#define PORTS 3
+
+static char pins[PINS];
+static DEFINE_SPINLOCK(pinmux_lock);
+
+static void crisv32_pinmux_set(int port);
+
+int
+crisv32_pinmux_init(void)
+{
+	static int initialized;
+
+	if (!initialized) {
+		initialized = 1;
+		REG_WR_INT(pinmux, regi_pinmux, rw_hwprot, 0);
+		crisv32_pinmux_alloc(PORT_A, 0, 31, pinmux_gpio);
+		crisv32_pinmux_alloc(PORT_B, 0, 31, pinmux_gpio);
+		crisv32_pinmux_alloc(PORT_C, 0, 15, pinmux_gpio);
+	}
+
+	return 0;
+}
+
+int
+crisv32_pinmux_alloc(int port, int first_pin, int last_pin, enum pin_mode mode)
+{
+	int i;
+	unsigned long flags;
+
+	crisv32_pinmux_init();
+
+	if (port >= PORTS)
+		return -EINVAL;
+
+	spin_lock_irqsave(&pinmux_lock, flags);
+
+	for (i = first_pin; i <= last_pin; i++) {
+		if ((pins[port * PORT_PINS + i] != pinmux_none) &&
+		    (pins[port * PORT_PINS + i] != pinmux_gpio) &&
+		    (pins[port * PORT_PINS + i] != mode)) {
+			spin_unlock_irqrestore(&pinmux_lock, flags);
+#ifdef DEBUG
+			panic("Pinmux alloc failed!\n");
+#endif
+			return -EPERM;
+		}
+	}
+
+	for (i = first_pin; i <= last_pin; i++)
+		pins[port * PORT_PINS + i] = mode;
+
+	crisv32_pinmux_set(port);
+
+	spin_unlock_irqrestore(&pinmux_lock, flags);
+
+	return 0;
+}
+
+int
+crisv32_pinmux_alloc_fixed(enum fixed_function function)
+{
+	int ret = -EINVAL;
+	char saved[sizeof pins];
+	unsigned long flags;
+	reg_pinmux_rw_hwprot hwprot;
+	reg_clkgen_rw_clk_ctrl clk_ctrl;
+
+	spin_lock_irqsave(&pinmux_lock, flags);
+
+	/* Save internal data for recovery */
+	memcpy(saved, pins, sizeof pins);
+
+	crisv32_pinmux_init(); /* must be done before we read rw_hwprot */
+
+	hwprot = REG_RD(pinmux, regi_pinmux, rw_hwprot);
+	clk_ctrl = REG_RD(clkgen, regi_clkgen, rw_clk_ctrl);
+
+	switch (function) {
+	case pinmux_eth:
+		clk_ctrl.eth = regk_clkgen_yes;
+		clk_ctrl.dma0_1_eth = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_B, 8, 23, pinmux_fixed);
+		ret |= crisv32_pinmux_alloc(PORT_B, 24, 25, pinmux_fixed);
+		hwprot.eth = hwprot.eth_mdio = regk_pinmux_yes;
+		break;
+	case pinmux_geth:
+		ret = crisv32_pinmux_alloc(PORT_B, 0, 7, pinmux_fixed);
+		hwprot.geth = regk_pinmux_yes;
+		break;
+	case pinmux_tg_cmos:
+		clk_ctrl.ccd_tg_100 = clk_ctrl.ccd_tg_200 = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_B, 27, 29, pinmux_fixed);
+		hwprot.tg_clk = regk_pinmux_yes;
+		break;
+	case pinmux_tg_ccd:
+		clk_ctrl.ccd_tg_100 = clk_ctrl.ccd_tg_200 = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_B, 27, 31, pinmux_fixed);
+		ret |= crisv32_pinmux_alloc(PORT_C, 0, 15, pinmux_fixed);
+		hwprot.tg = hwprot.tg_clk = regk_pinmux_yes;
+		break;
+	case pinmux_vout:
+		clk_ctrl.strdma0_2_video = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_A, 8, 18, pinmux_fixed);
+		hwprot.vout = hwprot.vout_sync = regk_pinmux_yes;
+		break;
+	case pinmux_ser1:
+		clk_ctrl.sser_ser_dma6_7 = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_A, 24, 25, pinmux_fixed);
+		hwprot.ser1 = regk_pinmux_yes;
+		break;
+	case pinmux_ser2:
+		clk_ctrl.sser_ser_dma6_7 = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_A, 26, 27, pinmux_fixed);
+		hwprot.ser2 = regk_pinmux_yes;
+		break;
+	case pinmux_ser3:
+		clk_ctrl.sser_ser_dma6_7 = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_A, 28, 29, pinmux_fixed);
+		hwprot.ser3 = regk_pinmux_yes;
+		break;
+	case pinmux_ser4:
+		clk_ctrl.sser_ser_dma6_7 = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_A, 30, 31, pinmux_fixed);
+		hwprot.ser4 = regk_pinmux_yes;
+		break;
+	case pinmux_sser:
+		clk_ctrl.sser_ser_dma6_7 = regk_clkgen_yes;
+		ret = crisv32_pinmux_alloc(PORT_A, 19, 23, pinmux_fixed);
+		hwprot.sser = regk_pinmux_yes;
+		break;
+	case pinmux_pio:
+		hwprot.pio = regk_pinmux_yes;
+		ret = 0;
+		break;
+	case pinmux_pwm0:
+		ret = crisv32_pinmux_alloc(PORT_A, 30, 30, pinmux_fixed);
+		hwprot.pwm0 = regk_pinmux_yes;
+		break;
+	case pinmux_pwm1:
+		ret = crisv32_pinmux_alloc(PORT_A, 31, 31, pinmux_fixed);
+		hwprot.pwm1 = regk_pinmux_yes;
+		break;
+	case pinmux_pwm2:
+		ret = crisv32_pinmux_alloc(PORT_B, 26, 26, pinmux_fixed);
+		hwprot.pwm2 = regk_pinmux_yes;
+		break;
+	case pinmux_i2c0:
+		ret = crisv32_pinmux_alloc(PORT_A, 0, 1, pinmux_fixed);
+		hwprot.i2c0 = regk_pinmux_yes;
+		break;
+	case pinmux_i2c1:
+		ret = crisv32_pinmux_alloc(PORT_A, 2, 3, pinmux_fixed);
+		hwprot.i2c1 = regk_pinmux_yes;
+		break;
+	case pinmux_i2c1_3wire:
+		ret = crisv32_pinmux_alloc(PORT_A, 2, 3, pinmux_fixed);
+		ret |= crisv32_pinmux_alloc(PORT_A, 7, 7, pinmux_fixed);
+		hwprot.i2c1 = hwprot.i2c1_sen = regk_pinmux_yes;
+		break;
+	case pinmux_i2c1_sda1:
+		ret = crisv32_pinmux_alloc(PORT_A, 2, 4, pinmux_fixed);
+		hwprot.i2c1 = hwprot.i2c1_sda1 = regk_pinmux_yes;
+		break;
+	case pinmux_i2c1_sda2:
+		ret = crisv32_pinmux_alloc(PORT_A, 2, 3, pinmux_fixed);
+		ret |= crisv32_pinmux_alloc(PORT_A, 5, 5, pinmux_fixed);
+		hwprot.i2c1 = hwprot.i2c1_sda2 = regk_pinmux_yes;
+		break;
+	case pinmux_i2c1_sda3:
+		ret = crisv32_pinmux_alloc(PORT_A, 2, 3, pinmux_fixed);
+		ret |= crisv32_pinmux_alloc(PORT_A, 6, 6, pinmux_fixed);
+		hwprot.i2c1 = hwprot.i2c1_sda3 = regk_pinmux_yes;
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+
+	if (!ret) {
+		REG_WR(pinmux, regi_pinmux, rw_hwprot, hwprot);
+		REG_WR(clkgen, regi_clkgen, rw_clk_ctrl, clk_ctrl);
+	} else
+		memcpy(pins, saved, sizeof pins);
+
+  spin_unlock_irqrestore(&pinmux_lock, flags);
+
+  return ret;
+}
+
+void
+crisv32_pinmux_set(int port)
+{
+	int i;
+	int gpio_val = 0;
+	int iop_val = 0;
+	int pin = port * PORT_PINS;
+
+	for (i = 0; (i < PORT_PINS) && (pin < PINS); i++, pin++) {
+		if (pins[pin] == pinmux_gpio)
+			gpio_val |= (1 << i);
+		else if (pins[pin] == pinmux_iop)
+			iop_val |= (1 << i);
+	}
+
+	REG_WRITE(int, regi_pinmux + REG_RD_ADDR_pinmux_rw_gio_pa + 4 * port,
+		gpio_val);
+	REG_WRITE(int, regi_pinmux + REG_RD_ADDR_pinmux_rw_iop_pa + 4 * port,
+		iop_val);
+
+#ifdef DEBUG
+       crisv32_pinmux_dump();
+#endif
+}
+
+int
+crisv32_pinmux_dealloc(int port, int first_pin, int last_pin)
+{
+	int i;
+	unsigned long flags;
+
+	crisv32_pinmux_init();
+
+	if (port > PORTS || port < 0)
+		return -EINVAL;
+
+	spin_lock_irqsave(&pinmux_lock, flags);
+
+	for (i = first_pin; i <= last_pin; i++)
+		pins[port * PORT_PINS + i] = pinmux_none;
+
+	crisv32_pinmux_set(port);
+	spin_unlock_irqrestore(&pinmux_lock, flags);
+
+	return 0;
+}
+
+int
+crisv32_pinmux_dealloc_fixed(enum fixed_function function)
+{
+	int ret = -EINVAL;
+	char saved[sizeof pins];
+	unsigned long flags;
+	reg_pinmux_rw_hwprot hwprot;
+
+	spin_lock_irqsave(&pinmux_lock, flags);
+
+	/* Save internal data for recovery */
+	memcpy(saved, pins, sizeof pins);
+
+	crisv32_pinmux_init(); /* must be done before we read rw_hwprot */
+
+	hwprot = REG_RD(pinmux, regi_pinmux, rw_hwprot);
+
+	switch (function) {
+	case pinmux_eth:
+		ret = crisv32_pinmux_dealloc(PORT_B, 8, 23);
+		ret |= crisv32_pinmux_dealloc(PORT_B, 24, 25);
+		ret |= crisv32_pinmux_dealloc(PORT_B, 0, 7);
+		hwprot.eth = hwprot.eth_mdio = hwprot.geth = regk_pinmux_no;
+		break;
+	case pinmux_tg_cmos:
+		ret = crisv32_pinmux_dealloc(PORT_B, 27, 29);
+		hwprot.tg_clk = regk_pinmux_no;
+		break;
+	case pinmux_tg_ccd:
+		ret = crisv32_pinmux_dealloc(PORT_B, 27, 31);
+		ret |= crisv32_pinmux_dealloc(PORT_C, 0, 15);
+		hwprot.tg = hwprot.tg_clk = regk_pinmux_no;
+		break;
+	case pinmux_vout:
+		ret = crisv32_pinmux_dealloc(PORT_A, 8, 18);
+		hwprot.vout = hwprot.vout_sync = regk_pinmux_no;
+		break;
+	case pinmux_ser1:
+		ret = crisv32_pinmux_dealloc(PORT_A, 24, 25);
+		hwprot.ser1 = regk_pinmux_no;
+		break;
+	case pinmux_ser2:
+		ret = crisv32_pinmux_dealloc(PORT_A, 26, 27);
+		hwprot.ser2 = regk_pinmux_no;
+		break;
+	case pinmux_ser3:
+		ret = crisv32_pinmux_dealloc(PORT_A, 28, 29);
+		hwprot.ser3 = regk_pinmux_no;
+		break;
+	case pinmux_ser4:
+		ret = crisv32_pinmux_dealloc(PORT_A, 30, 31);
+		hwprot.ser4 = regk_pinmux_no;
+		break;
+	case pinmux_sser:
+		ret = crisv32_pinmux_dealloc(PORT_A, 19, 23);
+		hwprot.sser = regk_pinmux_no;
+		break;
+	case pinmux_pwm0:
+		ret = crisv32_pinmux_dealloc(PORT_A, 30, 30);
+		hwprot.pwm0 = regk_pinmux_no;
+		break;
+	case pinmux_pwm1:
+		ret = crisv32_pinmux_dealloc(PORT_A, 31, 31);
+		hwprot.pwm1 = regk_pinmux_no;
+		break;
+	case pinmux_pwm2:
+		ret = crisv32_pinmux_dealloc(PORT_B, 26, 26);
+		hwprot.pwm2 = regk_pinmux_no;
+		break;
+	case pinmux_i2c0:
+		ret = crisv32_pinmux_dealloc(PORT_A, 0, 1);
+		hwprot.i2c0 = regk_pinmux_no;
+		break;
+	case pinmux_i2c1:
+		ret = crisv32_pinmux_dealloc(PORT_A, 2, 3);
+		hwprot.i2c1 = regk_pinmux_no;
+		break;
+	case pinmux_i2c1_3wire:
+		ret = crisv32_pinmux_dealloc(PORT_A, 2, 3);
+		ret |= crisv32_pinmux_dealloc(PORT_A, 7, 7);
+		hwprot.i2c1 = hwprot.i2c1_sen = regk_pinmux_no;
+		break;
+	case pinmux_i2c1_sda1:
+		ret = crisv32_pinmux_dealloc(PORT_A, 2, 4);
+		hwprot.i2c1_sda1 = regk_pinmux_no;
+		break;
+	case pinmux_i2c1_sda2:
+		ret = crisv32_pinmux_dealloc(PORT_A, 2, 3);
+		ret |= crisv32_pinmux_dealloc(PORT_A, 5, 5);
+		hwprot.i2c1_sda2 = regk_pinmux_no;
+		break;
+	case pinmux_i2c1_sda3:
+		ret = crisv32_pinmux_dealloc(PORT_A, 2, 3);
+		ret |= crisv32_pinmux_dealloc(PORT_A, 6, 6);
+		hwprot.i2c1_sda3 = regk_pinmux_no;
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+
+	if (!ret)
+		REG_WR(pinmux, regi_pinmux, rw_hwprot, hwprot);
+	else
+		memcpy(pins, saved, sizeof pins);
+
+  spin_unlock_irqrestore(&pinmux_lock, flags);
+
+  return ret;
+}
+
+void
+crisv32_pinmux_dump(void)
+{
+	int i, j;
+	int pin = 0;
+
+	crisv32_pinmux_init();
+
+	for (i = 0; i < PORTS; i++) {
+		pin++;
+		printk(KERN_DEBUG "Port %c\n", 'A'+i);
+		for (j = 0; (j < PORT_PINS) && (pin < PINS); j++, pin++)
+			printk(KERN_DEBUG
+				"  Pin %d = %d\n", j, pins[i * PORT_PINS + j]);
+	}
+}
+
+__initcall(crisv32_pinmux_init);