]> pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - drivers/firewire/fw-ohci.c
MTD: Add onenand support for omap24xx
[linux-2.6-omap-h63xx.git] / drivers / firewire / fw-ohci.c
index 0088acd7718e291160fdf9aedbd83bcf9b3ac188..0d08bf9b78c21d8f1811a317e5145ae577e4cc98 100644 (file)
@@ -1,6 +1,6 @@
-/*                                             -*- c-basic-offset: 8 -*-
+/*
+ * Driver for OHCI 1394 controllers
  *
- * fw-ohci.c - Driver for OHCI 1394 boards
  * Copyright (C) 2003-2006 Kristian Hoegsberg <krh@bitplanet.net>
  *
  * This program is free software; you can redistribute it and/or modify
@@ -26,6 +26,7 @@
 #include <linux/delay.h>
 #include <linux/poll.h>
 #include <linux/dma-mapping.h>
+#include <linux/mm.h>
 
 #include <asm/uaccess.h>
 #include <asm/semaphore.h>
 #include "fw-transaction.h"
 #include "fw-ohci.h"
 
-#define descriptor_output_more         0
-#define descriptor_output_last         (1 << 12)
-#define descriptor_input_more          (2 << 12)
-#define descriptor_input_last          (3 << 12)
-#define descriptor_status              (1 << 11)
-#define descriptor_key_immediate       (2 << 8)
-#define descriptor_ping                        (1 << 7)
-#define descriptor_yy                  (1 << 6)
-#define descriptor_no_irq              (0 << 4)
-#define descriptor_irq_error           (1 << 4)
-#define descriptor_irq_always          (3 << 4)
-#define descriptor_branch_always       (3 << 2)
-#define descriptor_wait                        (3 << 0)
+#define DESCRIPTOR_OUTPUT_MORE         0
+#define DESCRIPTOR_OUTPUT_LAST         (1 << 12)
+#define DESCRIPTOR_INPUT_MORE          (2 << 12)
+#define DESCRIPTOR_INPUT_LAST          (3 << 12)
+#define DESCRIPTOR_STATUS              (1 << 11)
+#define DESCRIPTOR_KEY_IMMEDIATE       (2 << 8)
+#define DESCRIPTOR_PING                        (1 << 7)
+#define DESCRIPTOR_YY                  (1 << 6)
+#define DESCRIPTOR_NO_IRQ              (0 << 4)
+#define DESCRIPTOR_IRQ_ERROR           (1 << 4)
+#define DESCRIPTOR_IRQ_ALWAYS          (3 << 4)
+#define DESCRIPTOR_BRANCH_ALWAYS       (3 << 2)
+#define DESCRIPTOR_WAIT                        (3 << 0)
 
 struct descriptor {
        __le16 req_count;
@@ -70,10 +71,10 @@ struct db_descriptor {
        __le32 reserved1;
 } __attribute__((aligned(16)));
 
-#define control_set(regs)      (regs)
-#define control_clear(regs)    ((regs) + 4)
-#define command_ptr(regs)      ((regs) + 12)
-#define context_match(regs)    ((regs) + 16)
+#define CONTROL_SET(regs)      (regs)
+#define CONTROL_CLEAR(regs)    ((regs) + 4)
+#define COMMAND_PTR(regs)      ((regs) + 12)
+#define CONTEXT_MATCH(regs)    ((regs) + 16)
 
 struct ar_buffer {
        struct descriptor descriptor;
@@ -96,9 +97,9 @@ typedef int (*descriptor_callback_t)(struct context *ctx,
                                     struct descriptor *d,
                                     struct descriptor *last);
 struct context {
-       struct fw_ohci *ohci;
+       struct fw_ohci *ohci;
        u32 regs;
+
        struct descriptor *buffer;
        dma_addr_t buffer_bus;
        size_t buffer_size;
@@ -109,36 +110,15 @@ struct context {
 
        descriptor_callback_t callback;
 
-       struct tasklet_struct tasklet;
-};
-
-
-struct at_context {
-       struct fw_ohci *ohci;
-       dma_addr_t descriptor_bus;
-       dma_addr_t buffer_bus;
-       struct fw_packet *current_packet;
-
-       struct list_head list;
-
-       struct {
-               struct descriptor more;
-               __le32 header[4];
-               struct descriptor last;
-       } d;
-
-       u32 regs;
-
        struct tasklet_struct tasklet;
 };
 
-#define it_header_sy(v)          ((v) <<  0)
-#define it_header_tcode(v)       ((v) <<  4)
-#define it_header_channel(v)     ((v) <<  8)
-#define it_header_tag(v)         ((v) << 14)
-#define it_header_speed(v)       ((v) << 16)
-#define it_header_data_length(v) ((v) << 16)
+#define IT_HEADER_SY(v)          ((v) <<  0)
+#define IT_HEADER_TCODE(v)       ((v) <<  4)
+#define IT_HEADER_CHANNEL(v)     ((v) <<  8)
+#define IT_HEADER_TAG(v)         ((v) << 14)
+#define IT_HEADER_SPEED(v)       ((v) << 16)
+#define IT_HEADER_DATA_LENGTH(v) ((v) << 16)
 
 struct iso_context {
        struct fw_iso_context base;
@@ -160,9 +140,12 @@ struct fw_ohci {
        int node_id;
        int generation;
        int request_generation;
+       u32 bus_seconds;
 
-       /* Spinlock for accessing fw_ohci data.  Never call out of
-        * this driver with this lock held. */
+       /*
+        * Spinlock for accessing fw_ohci data.  Never call out of
+        * this driver with this lock held.
+        */
        spinlock_t lock;
        u32 self_id_buffer[512];
 
@@ -175,8 +158,8 @@ struct fw_ohci {
 
        struct ar_context ar_request_ctx;
        struct ar_context ar_response_ctx;
-       struct at_context at_request_ctx;
-       struct at_context at_response_ctx;
+       struct context at_request_ctx;
+       struct context at_response_ctx;
 
        u32 it_context_mask;
        struct iso_context *it_context_list;
@@ -212,6 +195,8 @@ static inline struct fw_ohci *fw_ohci(struct fw_card *card)
 #define SELF_ID_BUF_SIZE               0x800
 #define OHCI_TCODE_PHY_PACKET          0x0e
 #define OHCI_VERSION_1_1               0x010010
+#define ISO_BUFFER_SIZE                        (64 * 1024)
+#define AT_BUFFER_SIZE                 4096
 
 static char ohci_driver_name[] = KBUILD_MODNAME;
 
@@ -271,10 +256,10 @@ static int ar_context_add_page(struct ar_context *ctx)
                return -ENOMEM;
        }
 
-       memset(&ab->descriptor, 0, sizeof ab->descriptor);
-       ab->descriptor.control        = cpu_to_le16(descriptor_input_more |
-                                                   descriptor_status |
-                                                   descriptor_branch_always);
+       memset(&ab->descriptor, 0, sizeof(ab->descriptor));
+       ab->descriptor.control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
+                                                   DESCRIPTOR_STATUS |
+                                                   DESCRIPTOR_BRANCH_ALWAYS);
        offset = offsetof(struct ar_buffer, data);
        ab->descriptor.req_count      = cpu_to_le16(PAGE_SIZE - offset);
        ab->descriptor.data_address   = cpu_to_le32(ab_bus + offset);
@@ -283,11 +268,11 @@ static int ar_context_add_page(struct ar_context *ctx)
 
        dma_sync_single_for_device(dev, ab_bus, PAGE_SIZE, DMA_BIDIRECTIONAL);
 
-       ctx->last_buffer->descriptor.branch_address = ab_bus | 1;
+       ctx->last_buffer->descriptor.branch_address = cpu_to_le32(ab_bus | 1);
        ctx->last_buffer->next = ab;
        ctx->last_buffer = ab;
 
-       reg_write(ctx->ohci, control_set(ctx->regs), CONTEXT_WAKE);
+       reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
        flush_writes(ctx->ohci);
 
        return 0;
@@ -346,13 +331,15 @@ static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
        p.timestamp  = status & 0xffff;
        p.generation = ohci->request_generation;
 
-       /* The OHCI bus reset handler synthesizes a phy packet with
+       /*
+        * The OHCI bus reset handler synthesizes a phy packet with
         * the new generation number when a bus reset happens (see
         * section 8.4.2.3).  This helps us determine when a request
         * was received and make sure we send the response in the same
         * generation.  We only need this for requests; for responses
         * we use the unique tlabel for finding the matching
-        * request. */
+        * request.
+        */
 
        if (p.ack + 16 == 0x09)
                ohci->request_generation = (buffer[2] >> 16) & 0xff;
@@ -378,9 +365,11 @@ static void ar_context_tasklet(unsigned long data)
        if (d->res_count == 0) {
                size_t size, rest, offset;
 
-               /* This descriptor is finished and we may have a
+               /*
+                * This descriptor is finished and we may have a
                 * packet split across this and the next buffer. We
-                * reuse the page for reassembling the split packet. */
+                * reuse the page for reassembling the split packet.
+                */
 
                offset = offsetof(struct ar_buffer, data);
                dma_unmap_single(ohci->card.device,
@@ -428,13 +417,23 @@ ar_context_init(struct ar_context *ctx, struct fw_ohci *ohci, u32 regs)
        ctx->current_buffer = ab.next;
        ctx->pointer = ctx->current_buffer->data;
 
-       reg_write(ctx->ohci, command_ptr(ctx->regs), ab.descriptor.branch_address);
-       reg_write(ctx->ohci, control_set(ctx->regs), CONTEXT_RUN);
-       flush_writes(ctx->ohci);
-
        return 0;
 }
+
+static void ar_context_run(struct ar_context *ctx)
+{
+       struct ar_buffer *ab = ctx->current_buffer;
+       dma_addr_t ab_bus;
+       size_t offset;
+
+       offset = offsetof(struct ar_buffer, data);
+       ab_bus = ab->descriptor.data_address - offset;
+
+       reg_write(ctx->ohci, COMMAND_PTR(ctx->regs), ab_bus | 1);
+       reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN);
+       flush_writes(ctx->ohci);
+}
+
 static void context_tasklet(unsigned long data)
 {
        struct context *ctx = (struct context *) data;
@@ -452,7 +451,7 @@ static void context_tasklet(unsigned long data)
        while (last->branch_address != 0) {
                address = le32_to_cpu(last->branch_address);
                z = address & 0xf;
-               d = ctx->buffer + (address - ctx->buffer_bus) / sizeof *d;
+               d = ctx->buffer + (address - ctx->buffer_bus) / sizeof(*d);
                last = (z == 2) ? d : d + z - 1;
 
                if (!ctx->callback(ctx, d, last))
@@ -491,14 +490,16 @@ context_init(struct context *ctx, struct fw_ohci *ohci,
        ctx->tail_descriptor      = ctx->buffer;
        ctx->tail_descriptor_last = ctx->buffer;
 
-       /* We put a dummy descriptor in the buffer that has a NULL
+       /*
+        * We put a dummy descriptor in the buffer that has a NULL
         * branch address and looks like it's been sent.  That way we
         * have a descriptor to append DMA programs to.  Also, the
         * ring buffer invariant is that it always has at least one
-        * element so that head == tail means buffer full. */
+        * element so that head == tail means buffer full.
+        */
 
-       memset(ctx->head_descriptor, 0, sizeof *ctx->head_descriptor);
-       ctx->head_descriptor->control = cpu_to_le16(descriptor_output_last);
+       memset(ctx->head_descriptor, 0, sizeof(*ctx->head_descriptor));
+       ctx->head_descriptor->control = cpu_to_le16(DESCRIPTOR_OUTPUT_LAST);
        ctx->head_descriptor->transfer_status = cpu_to_le16(0x8011);
        ctx->head_descriptor++;
 
@@ -522,7 +523,7 @@ context_get_descriptors(struct context *ctx, int z, dma_addr_t *d_bus)
 
        d = ctx->head_descriptor;
        tail = ctx->tail_descriptor;
-       end = ctx->buffer + ctx->buffer_size / sizeof(struct descriptor);
+       end = ctx->buffer + ctx->buffer_size / sizeof(*d);
 
        if (d + z <= tail) {
                goto has_space;
@@ -536,8 +537,8 @@ context_get_descriptors(struct context *ctx, int z, dma_addr_t *d_bus)
        return NULL;
 
  has_space:
-       memset(d, 0, z * sizeof *d);
-       *d_bus = ctx->buffer_bus + (d - ctx->buffer) * sizeof *d;
+       memset(d, 0, z * sizeof(*d));
+       *d_bus = ctx->buffer_bus + (d - ctx->buffer) * sizeof(*d);
 
        return d;
 }
@@ -546,10 +547,10 @@ static void context_run(struct context *ctx, u32 extra)
 {
        struct fw_ohci *ohci = ctx->ohci;
 
-       reg_write(ohci, command_ptr(ctx->regs),
+       reg_write(ohci, COMMAND_PTR(ctx->regs),
                  le32_to_cpu(ctx->tail_descriptor_last->branch_address));
-       reg_write(ohci, control_clear(ctx->regs), ~0);
-       reg_write(ohci, control_set(ctx->regs), CONTEXT_RUN | extra);
+       reg_write(ohci, CONTROL_CLEAR(ctx->regs), ~0);
+       reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN | extra);
        flush_writes(ohci);
 }
 
@@ -558,7 +559,7 @@ static void context_append(struct context *ctx,
 {
        dma_addr_t d_bus;
 
-       d_bus = ctx->buffer_bus + (d - ctx->buffer) * sizeof *d;
+       d_bus = ctx->buffer_bus + (d - ctx->buffer) * sizeof(*d);
 
        ctx->head_descriptor = d + z + extra;
        ctx->prev_descriptor->branch_address = cpu_to_le32(d_bus | z);
@@ -567,7 +568,7 @@ static void context_append(struct context *ctx,
        dma_sync_single_for_device(ctx->ohci->card.device, ctx->buffer_bus,
                                   ctx->buffer_size, DMA_TO_DEVICE);
 
-       reg_write(ctx->ohci, control_set(ctx->regs), CONTEXT_WAKE);
+       reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
        flush_writes(ctx->ohci);
 }
 
@@ -576,11 +577,11 @@ static void context_stop(struct context *ctx)
        u32 reg;
        int i;
 
-       reg_write(ctx->ohci, control_clear(ctx->regs), CONTEXT_RUN);
+       reg_write(ctx->ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
        flush_writes(ctx->ohci);
 
        for (i = 0; i < 10; i++) {
-               reg = reg_read(ctx->ohci, control_set(ctx->regs));
+               reg = reg_read(ctx->ohci, CONTROL_SET(ctx->regs));
                if ((reg & CONTEXT_ACTIVE) == 0)
                        break;
 
@@ -589,217 +590,182 @@ static void context_stop(struct context *ctx)
        }
 }
 
-static void
-do_packet_callbacks(struct fw_ohci *ohci, struct list_head *list)
-{
-       struct fw_packet *p, *next;
-
-       list_for_each_entry_safe(p, next, list, link)
-               p->callback(p, &ohci->card, p->ack);
-}
-
-static void
-complete_transmission(struct fw_packet *packet,
-                     int ack, struct list_head *list)
-{
-       list_move_tail(&packet->link, list);
-       packet->ack = ack;
-}
+struct driver_data {
+       struct fw_packet *packet;
+};
 
-/* This function prepares the first packet in the context queue for
- * transmission.  Must always be called with the ochi->lock held to
- * ensure proper generation handling and locking around packet queue
- * manipulation. */
-static void
-at_context_setup_packet(struct at_context *ctx, struct list_head *list)
+/*
+ * This function apppends a packet to the DMA queue for transmission.
+ * Must always be called with the ochi->lock held to ensure proper
+ * generation handling and locking around packet queue manipulation.
+ */
+static int
+at_context_queue_packet(struct context *ctx, struct fw_packet *packet)
 {
-       struct fw_packet *packet;
        struct fw_ohci *ohci = ctx->ohci;
+       dma_addr_t d_bus, payload_bus;
+       struct driver_data *driver_data;
+       struct descriptor *d, *last;
+       __le32 *header;
        int z, tcode;
+       u32 reg;
 
-       packet = fw_packet(ctx->list.next);
-
-       memset(&ctx->d, 0, sizeof ctx->d);
-       if (packet->payload_length > 0) {
-               packet->payload_bus = dma_map_single(ohci->card.device,
-                                                    packet->payload,
-                                                    packet->payload_length,
-                                                    DMA_TO_DEVICE);
-               if (dma_mapping_error(packet->payload_bus)) {
-                       complete_transmission(packet, RCODE_SEND_ERROR, list);
-                       return;
-               }
-
-               ctx->d.more.control      =
-                       cpu_to_le16(descriptor_output_more |
-                                   descriptor_key_immediate);
-               ctx->d.more.req_count    = cpu_to_le16(packet->header_length);
-               ctx->d.more.res_count    = cpu_to_le16(packet->timestamp);
-               ctx->d.last.control      =
-                       cpu_to_le16(descriptor_output_last |
-                                   descriptor_irq_always |
-                                   descriptor_branch_always);
-               ctx->d.last.req_count    = cpu_to_le16(packet->payload_length);
-               ctx->d.last.data_address = cpu_to_le32(packet->payload_bus);
-               z = 3;
-       } else {
-               ctx->d.more.control   =
-                       cpu_to_le16(descriptor_output_last |
-                                   descriptor_key_immediate |
-                                   descriptor_irq_always |
-                                   descriptor_branch_always);
-               ctx->d.more.req_count = cpu_to_le16(packet->header_length);
-               ctx->d.more.res_count = cpu_to_le16(packet->timestamp);
-               z = 2;
+       d = context_get_descriptors(ctx, 4, &d_bus);
+       if (d == NULL) {
+               packet->ack = RCODE_SEND_ERROR;
+               return -1;
        }
 
-       /* The DMA format for asyncronous link packets is different
+       d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
+       d[0].res_count = cpu_to_le16(packet->timestamp);
+
+       /*
+        * The DMA format for asyncronous link packets is different
         * from the IEEE1394 layout, so shift the fields around
         * accordingly.  If header_length is 8, it's a PHY packet, to
-        * which we need to prepend an extra quadlet. */
+        * which we need to prepend an extra quadlet.
+        */
+
+       header = (__le32 *) &d[1];
        if (packet->header_length > 8) {
-               ctx->d.header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
-                                              (packet->speed << 16));
-               ctx->d.header[1] = cpu_to_le32((packet->header[1] & 0xffff) |
-                                              (packet->header[0] & 0xffff0000));
-               ctx->d.header[2] = cpu_to_le32(packet->header[2]);
+               header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
+                                       (packet->speed << 16));
+               header[1] = cpu_to_le32((packet->header[1] & 0xffff) |
+                                       (packet->header[0] & 0xffff0000));
+               header[2] = cpu_to_le32(packet->header[2]);
 
                tcode = (packet->header[0] >> 4) & 0x0f;
                if (TCODE_IS_BLOCK_PACKET(tcode))
-                       ctx->d.header[3] = cpu_to_le32(packet->header[3]);
+                       header[3] = cpu_to_le32(packet->header[3]);
                else
-                       ctx->d.header[3] = packet->header[3];
+                       header[3] = (__force __le32) packet->header[3];
+
+               d[0].req_count = cpu_to_le16(packet->header_length);
        } else {
-               ctx->d.header[0] =
-                       cpu_to_le32((OHCI1394_phy_tcode << 4) |
-                                   (packet->speed << 16));
-               ctx->d.header[1] = cpu_to_le32(packet->header[0]);
-               ctx->d.header[2] = cpu_to_le32(packet->header[1]);
-               ctx->d.more.req_count = cpu_to_le16(12);
+               header[0] = cpu_to_le32((OHCI1394_phy_tcode << 4) |
+                                       (packet->speed << 16));
+               header[1] = cpu_to_le32(packet->header[0]);
+               header[2] = cpu_to_le32(packet->header[1]);
+               d[0].req_count = cpu_to_le16(12);
        }
 
-       /* FIXME: Document how the locking works. */
-       if (ohci->generation == packet->generation) {
-               reg_write(ctx->ohci, command_ptr(ctx->regs),
-                         ctx->descriptor_bus | z);
-               reg_write(ctx->ohci, control_set(ctx->regs),
-                         CONTEXT_RUN | CONTEXT_WAKE);
-               ctx->current_packet = packet;
+       driver_data = (struct driver_data *) &d[3];
+       driver_data->packet = packet;
+       packet->driver_data = driver_data;
+       
+       if (packet->payload_length > 0) {
+               payload_bus =
+                       dma_map_single(ohci->card.device, packet->payload,
+                                      packet->payload_length, DMA_TO_DEVICE);
+               if (dma_mapping_error(payload_bus)) {
+                       packet->ack = RCODE_SEND_ERROR;
+                       return -1;
+               }
+
+               d[2].req_count    = cpu_to_le16(packet->payload_length);
+               d[2].data_address = cpu_to_le32(payload_bus);
+               last = &d[2];
+               z = 3;
        } else {
-               /* We dont return error codes from this function; all
-                * transmission errors are reported through the
-                * callback. */
-               complete_transmission(packet, RCODE_GENERATION, list);
+               last = &d[0];
+               z = 2;
        }
-}
 
-static void at_context_stop(struct at_context *ctx)
-{
-       u32 reg;
+       last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
+                                    DESCRIPTOR_IRQ_ALWAYS |
+                                    DESCRIPTOR_BRANCH_ALWAYS);
+
+       /* FIXME: Document how the locking works. */
+       if (ohci->generation != packet->generation) {
+               packet->ack = RCODE_GENERATION;
+               return -1;
+       }
+
+       context_append(ctx, d, z, 4 - z);
 
-       reg_write(ctx->ohci, control_clear(ctx->regs), CONTEXT_RUN);
+       /* If the context isn't already running, start it up. */
+       reg = reg_read(ctx->ohci, CONTROL_SET(ctx->regs));
+       if ((reg & CONTEXT_RUN) == 0)
+               context_run(ctx, 0);
 
-       reg = reg_read(ctx->ohci, control_set(ctx->regs));
-       if (reg & CONTEXT_ACTIVE)
-               fw_notify("Tried to stop context, but it is still active "
-                         "(0x%08x).\n", reg);
+       return 0;
 }
 
-static void at_context_tasklet(unsigned long data)
+static int handle_at_packet(struct context *context,
+                           struct descriptor *d,
+                           struct descriptor *last)
 {
-       struct at_context *ctx = (struct at_context *)data;
-       struct fw_ohci *ohci = ctx->ohci;
+       struct driver_data *driver_data;
        struct fw_packet *packet;
-       LIST_HEAD(list);
-       unsigned long flags;
+       struct fw_ohci *ohci = context->ohci;
+       dma_addr_t payload_bus;
        int evt;
 
-       spin_lock_irqsave(&ohci->lock, flags);
-
-       packet = fw_packet(ctx->list.next);
-
-       at_context_stop(ctx);
+       if (last->transfer_status == 0)
+               /* This descriptor isn't done yet, stop iteration. */
+               return 0;
 
-       /* If the head of the list isn't the packet that just got
-        * transmitted, the packet got cancelled before we finished
-        * transmitting it. */
-       if (ctx->current_packet != packet)
-               goto skip_to_next;
+       driver_data = (struct driver_data *) &d[3];
+       packet = driver_data->packet;
+       if (packet == NULL)
+               /* This packet was cancelled, just continue. */
+               return 1;
 
-       if (packet->payload_length > 0) {
-               dma_unmap_single(ohci->card.device, packet->payload_bus,
+       payload_bus = le32_to_cpu(last->data_address);
+       if (payload_bus != 0)
+               dma_unmap_single(ohci->card.device, payload_bus,
                                 packet->payload_length, DMA_TO_DEVICE);
-               evt = le16_to_cpu(ctx->d.last.transfer_status) & 0x1f;
-               packet->timestamp = le16_to_cpu(ctx->d.last.res_count);
-       }
-       else {
-               evt = le16_to_cpu(ctx->d.more.transfer_status) & 0x1f;
-               packet->timestamp = le16_to_cpu(ctx->d.more.res_count);
-       }
 
-       if (evt < 16) {
-               switch (evt) {
-               case OHCI1394_evt_timeout:
-                       /* Async response transmit timed out. */
-                       complete_transmission(packet, RCODE_CANCELLED, &list);
-                       break;
+       evt = le16_to_cpu(last->transfer_status) & 0x1f;
+       packet->timestamp = le16_to_cpu(last->res_count);
 
-               case OHCI1394_evt_flushed:
-                       /* The packet was flushed should give same
-                        * error as when we try to use a stale
-                        * generation count. */
-                       complete_transmission(packet,
-                                             RCODE_GENERATION, &list);
-                       break;
-
-               case OHCI1394_evt_missing_ack:
-                       /* Using a valid (current) generation count,
-                        * but the node is not on the bus or not
-                        * sending acks. */
-                       complete_transmission(packet, RCODE_NO_ACK, &list);
-                       break;
-
-               default:
-                       complete_transmission(packet, RCODE_SEND_ERROR, &list);
-                       break;
-               }
-       } else
-               complete_transmission(packet, evt - 16, &list);
-
- skip_to_next:
-       /* If more packets are queued, set up the next one. */
-       if (!list_empty(&ctx->list))
-               at_context_setup_packet(ctx, &list);
-
-       spin_unlock_irqrestore(&ohci->lock, flags);
+       switch (evt) {
+       case OHCI1394_evt_timeout:
+               /* Async response transmit timed out. */
+               packet->ack = RCODE_CANCELLED;
+               break;
 
-       do_packet_callbacks(ohci, &list);
-}
+       case OHCI1394_evt_flushed:
+               /*
+                * The packet was flushed should give same error as
+                * when we try to use a stale generation count.
+                */
+               packet->ack = RCODE_GENERATION;
+               break;
 
-static int
-at_context_init(struct at_context *ctx, struct fw_ohci *ohci, u32 regs)
-{
-       INIT_LIST_HEAD(&ctx->list);
+       case OHCI1394_evt_missing_ack:
+               /*
+                * Using a valid (current) generation count, but the
+                * node is not on the bus or not sending acks.
+                */
+               packet->ack = RCODE_NO_ACK;
+               break;
 
-       ctx->descriptor_bus =
-               dma_map_single(ohci->card.device, &ctx->d,
-                              sizeof ctx->d, DMA_TO_DEVICE);
-       if (dma_mapping_error(ctx->descriptor_bus))
-               return -ENOMEM;
+       case ACK_COMPLETE + 0x10:
+       case ACK_PENDING + 0x10:
+       case ACK_BUSY_X + 0x10:
+       case ACK_BUSY_A + 0x10:
+       case ACK_BUSY_B + 0x10:
+       case ACK_DATA_ERROR + 0x10:
+       case ACK_TYPE_ERROR + 0x10:
+               packet->ack = evt - 0x10;
+               break;
 
-       ctx->regs = regs;
-       ctx->ohci = ohci;
+       default:
+               packet->ack = RCODE_SEND_ERROR;
+               break;
+       }
 
-       tasklet_init(&ctx->tasklet, at_context_tasklet, (unsigned long)ctx);
+       packet->callback(packet, &ohci->card, packet->ack);
 
-       return 0;
+       return 1;
 }
 
-#define header_get_destination(q)      (((q) >> 16) & 0xffff)
-#define header_get_tcode(q)            (((q) >> 4) & 0x0f)
-#define header_get_offset_high(q)      (((q) >> 0) & 0xffff)
-#define header_get_data_length(q)      (((q) >> 16) & 0xffff)
-#define header_get_extended_tcode(q)   (((q) >> 0) & 0xffff)
+#define HEADER_GET_DESTINATION(q)      (((q) >> 16) & 0xffff)
+#define HEADER_GET_TCODE(q)            (((q) >> 4) & 0x0f)
+#define HEADER_GET_OFFSET_HIGH(q)      (((q) >> 0) & 0xffff)
+#define HEADER_GET_DATA_LENGTH(q)      (((q) >> 16) & 0xffff)
+#define HEADER_GET_EXTENDED_TCODE(q)   (((q) >> 0) & 0xffff)
 
 static void
 handle_local_rom(struct fw_ohci *ohci, struct fw_packet *packet, u32 csr)
@@ -807,9 +773,9 @@ handle_local_rom(struct fw_ohci *ohci, struct fw_packet *packet, u32 csr)
        struct fw_packet response;
        int tcode, length, i;
 
-       tcode = header_get_tcode(packet->header[0]);
+       tcode = HEADER_GET_TCODE(packet->header[0]);
        if (TCODE_IS_BLOCK_PACKET(tcode))
-               length = header_get_data_length(packet->header[3]);
+               length = HEADER_GET_DATA_LENGTH(packet->header[3]);
        else
                length = 4;
 
@@ -836,10 +802,10 @@ handle_local_lock(struct fw_ohci *ohci, struct fw_packet *packet, u32 csr)
        __be32 *payload, lock_old;
        u32 lock_arg, lock_data;
 
-       tcode = header_get_tcode(packet->header[0]);
-       length = header_get_data_length(packet->header[3]);
+       tcode = HEADER_GET_TCODE(packet->header[0]);
+       length = HEADER_GET_DATA_LENGTH(packet->header[3]);
        payload = packet->payload;
-       ext_tcode = header_get_extended_tcode(packet->header[3]);
+       ext_tcode = HEADER_GET_EXTENDED_TCODE(packet->header[3]);
 
        if (tcode == TCODE_LOCK_REQUEST &&
            ext_tcode == EXTCODE_COMPARE_SWAP && length == 8) {
@@ -865,23 +831,25 @@ handle_local_lock(struct fw_ohci *ohci, struct fw_packet *packet, u32 csr)
                fw_notify("swap not done yet\n");
 
        fw_fill_response(&response, packet->header,
-                        RCODE_COMPLETE, &lock_old, sizeof lock_old);
+                        RCODE_COMPLETE, &lock_old, sizeof(lock_old));
  out:
        fw_core_handle_response(&ohci->card, &response);
 }
 
 static void
-handle_local_request(struct at_context *ctx, struct fw_packet *packet)
+handle_local_request(struct context *ctx, struct fw_packet *packet)
 {
        u64 offset;
        u32 csr;
 
-       packet->ack = ACK_PENDING;
-       packet->callback(packet, &ctx->ohci->card, packet->ack);
+       if (ctx == &ctx->ohci->at_request_ctx) {
+               packet->ack = ACK_PENDING;
+               packet->callback(packet, &ctx->ohci->card, packet->ack);
+       }
 
        offset =
                ((unsigned long long)
-                header_get_offset_high(packet->header[1]) << 32) |
+                HEADER_GET_OFFSET_HIGH(packet->header[1]) << 32) |
                packet->header[2];
        csr = offset - CSR_REGISTER_BASE;
 
@@ -902,30 +870,34 @@ handle_local_request(struct at_context *ctx, struct fw_packet *packet)
                        fw_core_handle_response(&ctx->ohci->card, packet);
                break;
        }
+
+       if (ctx == &ctx->ohci->at_response_ctx) {
+               packet->ack = ACK_COMPLETE;
+               packet->callback(packet, &ctx->ohci->card, packet->ack);
+       }
 }
 
 static void
-at_context_transmit(struct at_context *ctx, struct fw_packet *packet)
+at_context_transmit(struct context *ctx, struct fw_packet *packet)
 {
-       LIST_HEAD(list);
        unsigned long flags;
+       int retval;
 
        spin_lock_irqsave(&ctx->ohci->lock, flags);
 
-       if (header_get_destination(packet->header[0]) == ctx->ohci->node_id &&
+       if (HEADER_GET_DESTINATION(packet->header[0]) == ctx->ohci->node_id &&
            ctx->ohci->generation == packet->generation) {
                spin_unlock_irqrestore(&ctx->ohci->lock, flags);
                handle_local_request(ctx, packet);
                return;
        }
 
-       list_add_tail(&packet->link, &ctx->list);
-       if (ctx->list.next == &packet->link)
-               at_context_setup_packet(ctx, &list);
-
+       retval = at_context_queue_packet(ctx, packet);
        spin_unlock_irqrestore(&ctx->ohci->lock, flags);
 
-       do_packet_callbacks(ctx->ohci, &list);
+       if (retval < 0)
+               packet->callback(packet, &ctx->ohci->card, packet->ack);
+       
 }
 
 static void bus_reset_tasklet(unsigned long data)
@@ -942,10 +914,12 @@ static void bus_reset_tasklet(unsigned long data)
        }
        ohci->node_id = reg & 0xffff;
 
-       /* The count in the SelfIDCount register is the number of
+       /*
+        * The count in the SelfIDCount register is the number of
         * bytes in the self ID receive buffer.  Since we also receive
         * the inverted quadlets and a header quadlet, we shift one
-        * bit extra to get the actual number of self IDs. */
+        * bit extra to get the actual number of self IDs.
+        */
 
        self_id_count = (reg_read(ohci, OHCI1394_SelfIDCount) >> 3) & 0x3ff;
        generation = (le32_to_cpu(ohci->self_id_cpu[0]) >> 16) & 0xff;
@@ -956,7 +930,8 @@ static void bus_reset_tasklet(unsigned long data)
                ohci->self_id_buffer[j] = le32_to_cpu(ohci->self_id_cpu[i]);
        }
 
-       /* Check the consistency of the self IDs we just read.  The
+       /*
+        * Check the consistency of the self IDs we just read.  The
         * problem we face is that a new bus reset can start while we
         * read out the self IDs from the DMA buffer. If this happens,
         * the DMA buffer will be overwritten with new self IDs and we
@@ -966,7 +941,8 @@ static void bus_reset_tasklet(unsigned long data)
         * self IDs in the buffer before reading them out and compare
         * it to the current generation after reading them out.  If
         * the two generations match we know we have a consistent set
-        * of self IDs. */
+        * of self IDs.
+        */
 
        new_generation = (reg_read(ohci, OHCI1394_SelfIDCount) >> 16) & 0xff;
        if (new_generation != generation) {
@@ -979,16 +955,18 @@ static void bus_reset_tasklet(unsigned long data)
        spin_lock_irqsave(&ohci->lock, flags);
 
        ohci->generation = generation;
-       at_context_stop(&ohci->at_request_ctx);
-       at_context_stop(&ohci->at_response_ctx);
+       context_stop(&ohci->at_request_ctx);
+       context_stop(&ohci->at_response_ctx);
        reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);
 
-       /* This next bit is unrelated to the AT context stuff but we
+       /*
+        * This next bit is unrelated to the AT context stuff but we
         * have to do it under the spinlock also.  If a new config rom
         * was set up before this reset, the old one is now no longer
         * in use and we can free it. Update the config rom pointers
         * to point to the current config rom and clear the
-        * next_config_rom pointer so a new udpate can take place. */
+        * next_config_rom pointer so a new udpate can take place.
+        */
 
        if (ohci->next_config_rom != NULL) {
                dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
@@ -997,10 +975,12 @@ static void bus_reset_tasklet(unsigned long data)
                ohci->config_rom_bus  = ohci->next_config_rom_bus;
                ohci->next_config_rom = NULL;
 
-               /* Restore config_rom image and manually update
+               /*
+                * Restore config_rom image and manually update
                 * config_rom registers.  Writing the header quadlet
                 * will indicate that the config rom is ready, so we
-                * do that last. */
+                * do that last.
+                */
                reg_write(ohci, OHCI1394_BusOptions,
                          be32_to_cpu(ohci->config_rom[2]));
                ohci->config_rom[0] = cpu_to_be32(ohci->next_header);
@@ -1016,7 +996,7 @@ static void bus_reset_tasklet(unsigned long data)
 static irqreturn_t irq_handler(int irq, void *data)
 {
        struct fw_ohci *ohci = data;
-       u32 event, iso_event;
+       u32 event, iso_event, cycle_time;
        int i;
 
        event = reg_read(ohci, OHCI1394_IntEventClear);
@@ -1059,15 +1039,89 @@ static irqreturn_t irq_handler(int irq, void *data)
                iso_event &= ~(1 << i);
        }
 
+       if (event & OHCI1394_cycle64Seconds) {
+               cycle_time = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
+               if ((cycle_time & 0x80000000) == 0)
+                       ohci->bus_seconds++;
+       }
+
        return IRQ_HANDLED;
 }
 
+static int software_reset(struct fw_ohci *ohci)
+{
+       int i;
+
+       reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);
+
+       for (i = 0; i < OHCI_LOOP_COUNT; i++) {
+               if ((reg_read(ohci, OHCI1394_HCControlSet) &
+                    OHCI1394_HCControl_softReset) == 0)
+                       return 0;
+               msleep(1);
+       }
+
+       return -EBUSY;
+}
+
 static int ohci_enable(struct fw_card *card, u32 *config_rom, size_t length)
 {
        struct fw_ohci *ohci = fw_ohci(card);
        struct pci_dev *dev = to_pci_dev(card->device);
 
-       /* When the link is not yet enabled, the atomic config rom
+       if (software_reset(ohci)) {
+               fw_error("Failed to reset ohci card.\n");
+               return -EBUSY;
+       }
+
+       /*
+        * Now enable LPS, which we need in order to start accessing
+        * most of the registers.  In fact, on some cards (ALI M5251),
+        * accessing registers in the SClk domain without LPS enabled
+        * will lock up the machine.  Wait 50msec to make sure we have
+        * full link enabled.
+        */
+       reg_write(ohci, OHCI1394_HCControlSet,
+                 OHCI1394_HCControl_LPS |
+                 OHCI1394_HCControl_postedWriteEnable);
+       flush_writes(ohci);
+       msleep(50);
+
+       reg_write(ohci, OHCI1394_HCControlClear,
+                 OHCI1394_HCControl_noByteSwapData);
+
+       reg_write(ohci, OHCI1394_LinkControlSet,
+                 OHCI1394_LinkControl_rcvSelfID |
+                 OHCI1394_LinkControl_cycleTimerEnable |
+                 OHCI1394_LinkControl_cycleMaster);
+
+       reg_write(ohci, OHCI1394_ATRetries,
+                 OHCI1394_MAX_AT_REQ_RETRIES |
+                 (OHCI1394_MAX_AT_RESP_RETRIES << 4) |
+                 (OHCI1394_MAX_PHYS_RESP_RETRIES << 8));
+
+       ar_context_run(&ohci->ar_request_ctx);
+       ar_context_run(&ohci->ar_response_ctx);
+
+       reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->self_id_bus);
+       reg_write(ohci, OHCI1394_PhyUpperBound, 0x00010000);
+       reg_write(ohci, OHCI1394_IntEventClear, ~0);
+       reg_write(ohci, OHCI1394_IntMaskClear, ~0);
+       reg_write(ohci, OHCI1394_IntMaskSet,
+                 OHCI1394_selfIDComplete |
+                 OHCI1394_RQPkt | OHCI1394_RSPkt |
+                 OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
+                 OHCI1394_isochRx | OHCI1394_isochTx |
+                 OHCI1394_masterIntEnable |
+                 OHCI1394_cycle64Seconds);
+
+       /* Activate link_on bit and contender bit in our self ID packets.*/
+       if (ohci_update_phy_reg(card, 4, 0,
+                               PHY_LINK_ACTIVE | PHY_CONTENDER) < 0)
+               return -EIO;
+
+       /*
+        * When the link is not yet enabled, the atomic config rom
         * update mechanism described below in ohci_set_config_rom()
         * is not active.  We have to update ConfigRomHeader and
         * BusOptions manually, and the write to ConfigROMmap takes
@@ -1103,7 +1157,7 @@ static int ohci_enable(struct fw_card *card, u32 *config_rom, size_t length)
        reg_write(ohci, OHCI1394_AsReqFilterHiSet, 0x80000000);
 
        if (request_irq(dev->irq, irq_handler,
-                       SA_SHIRQ, ohci_driver_name, ohci)) {
+                       IRQF_SHARED, ohci_driver_name, ohci)) {
                fw_error("Failed to allocate shared interrupt %d.\n",
                         dev->irq);
                dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
@@ -1116,8 +1170,10 @@ static int ohci_enable(struct fw_card *card, u32 *config_rom, size_t length)
                  OHCI1394_HCControl_BIBimageValid);
        flush_writes(ohci);
 
-       /* We are ready to go, initiate bus reset to finish the
-        * initialization. */
+       /*
+        * We are ready to go, initiate bus reset to finish the
+        * initialization.
+        */
 
        fw_core_initiate_bus_reset(&ohci->card, 1);
 
@@ -1135,7 +1191,8 @@ ohci_set_config_rom(struct fw_card *card, u32 *config_rom, size_t length)
 
        ohci = fw_ohci(card);
 
-       /* When the OHCI controller is enabled, the config rom update
+       /*
+        * When the OHCI controller is enabled, the config rom update
         * mechanism is a bit tricky, but easy enough to use.  See
         * section 5.5.6 in the OHCI specification.
         *
@@ -1190,11 +1247,13 @@ ohci_set_config_rom(struct fw_card *card, u32 *config_rom, size_t length)
 
        spin_unlock_irqrestore(&ohci->lock, flags);
 
-       /* Now initiate a bus reset to have the changes take
+       /*
+        * Now initiate a bus reset to have the changes take
         * effect. We clean up the old config rom memory and DMA
         * mappings in the bus reset tasklet, since the OHCI
         * controller could need to access it before the bus reset
-        * takes effect. */
+        * takes effect.
+        */
        if (retval == 0)
                fw_core_initiate_bus_reset(&ohci->card, 1);
 
@@ -1218,24 +1277,24 @@ static void ohci_send_response(struct fw_card *card, struct fw_packet *packet)
 static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
 {
        struct fw_ohci *ohci = fw_ohci(card);
-       LIST_HEAD(list);
-       unsigned long flags;
+       struct context *ctx = &ohci->at_request_ctx;
+       struct driver_data *driver_data = packet->driver_data;
+       int retval = -ENOENT;
 
-       spin_lock_irqsave(&ohci->lock, flags);
+       tasklet_disable(&ctx->tasklet);
 
-       if (packet->ack == 0) {
-               fw_notify("cancelling packet %p (header[0]=%08x)\n",
-                         packet, packet->header[0]);
+       if (packet->ack != 0)
+               goto out;
 
-               complete_transmission(packet, RCODE_CANCELLED, &list);
-       }
+       driver_data->packet = NULL;
+       packet->ack = RCODE_CANCELLED;
+       packet->callback(packet, &ohci->card, packet->ack);
+       retval = 0;
 
-       spin_unlock_irqrestore(&ohci->lock, flags);
-
-       do_packet_callbacks(ohci, &list);
+ out:
+       tasklet_enable(&ctx->tasklet);
 
-       /* Return success if we actually cancelled something. */
-       return list_empty(&list) ? -ENOENT : 0;
+       return retval;
 }
 
 static int
@@ -1245,8 +1304,10 @@ ohci_enable_phys_dma(struct fw_card *card, int node_id, int generation)
        unsigned long flags;
        int n, retval = 0;
 
-       /* FIXME:  Make sure this bitmask is cleared when we clear the busReset
-        * interrupt bit.  Clear physReqResourceAllBuses on bus reset. */
+       /*
+        * FIXME:  Make sure this bitmask is cleared when we clear the busReset
+        * interrupt bit.  Clear physReqResourceAllBuses on bus reset.
+        */
 
        spin_lock_irqsave(&ohci->lock, flags);
 
@@ -1255,8 +1316,10 @@ ohci_enable_phys_dma(struct fw_card *card, int node_id, int generation)
                goto out;
        }
 
-       /* NOTE, if the node ID contains a non-local bus ID, physical DMA is
-        * enabled for _all_ nodes on remote buses. */
+       /*
+        * Note, if the node ID contains a non-local bus ID, physical DMA is
+        * enabled for _all_ nodes on remote buses.
+        */
 
        n = (node_id & 0xffc0) == LOCAL_BUS ? node_id & 0x3f : 63;
        if (n < 32)
@@ -1269,23 +1332,18 @@ ohci_enable_phys_dma(struct fw_card *card, int node_id, int generation)
        spin_unlock_irqrestore(&ohci->lock, flags);
        return retval;
 }
-static int handle_ir_bufferfill_packet(struct context *context,
-                                      struct descriptor *d,
-                                      struct descriptor *last)
+
+static u64
+ohci_get_bus_time(struct fw_card *card)
 {
-       struct iso_context *ctx =
-               container_of(context, struct iso_context, context);
+       struct fw_ohci *ohci = fw_ohci(card);
+       u32 cycle_time;
+       u64 bus_time;
 
-       if (d->res_count > 0)
-               return 0;
+       cycle_time = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
+       bus_time = ((u64) ohci->bus_seconds << 32) | cycle_time;
 
-       if (le16_to_cpu(last->control) & descriptor_irq_always)
-               ctx->base.callback(&ctx->base,
-                                  le16_to_cpu(last->res_count),
-                                  0, NULL, ctx->base.callback_data);
-
-       return 1;
+       return bus_time;
 }
 
 static int handle_ir_dualbuffer_packet(struct context *context,
@@ -1295,19 +1353,41 @@ static int handle_ir_dualbuffer_packet(struct context *context,
        struct iso_context *ctx =
                container_of(context, struct iso_context, context);
        struct db_descriptor *db = (struct db_descriptor *) d;
+       __le32 *ir_header;
        size_t header_length;
+       void *p, *end;
+       int i;
 
        if (db->first_res_count > 0 && db->second_res_count > 0)
                /* This descriptor isn't done yet, stop iteration. */
                return 0;
 
-       header_length = db->first_req_count - db->first_res_count;
-       if (ctx->header_length + header_length <= PAGE_SIZE)
-               memcpy(ctx->header + ctx->header_length, db + 1, header_length);
-       ctx->header_length += header_length;
+       header_length = le16_to_cpu(db->first_req_count) -
+               le16_to_cpu(db->first_res_count);
+
+       i = ctx->header_length;
+       p = db + 1;
+       end = p + header_length;
+       while (p < end && i + ctx->base.header_size <= PAGE_SIZE) {
+               /*
+                * The iso header is byteswapped to little endian by
+                * the controller, but the remaining header quadlets
+                * are big endian.  We want to present all the headers
+                * as big endian, so we have to swap the first
+                * quadlet.
+                */
+               *(u32 *) (ctx->header + i) = __swab32(*(u32 *) (p + 4));
+               memcpy(ctx->header + i + 4, p + 8, ctx->base.header_size - 4);
+               i += ctx->base.header_size;
+               p += ctx->base.header_size + 4;
+       }
 
-       if (le16_to_cpu(db->control) & descriptor_irq_always) {
-               ctx->base.callback(&ctx->base, 0,
+       ctx->header_length = i;
+
+       if (le16_to_cpu(db->control) & DESCRIPTOR_IRQ_ALWAYS) {
+               ir_header = (__le32 *) (db + 1);
+               ctx->base.callback(&ctx->base,
+                                  le32_to_cpu(ir_header[0]) & 0xffff,
                                   ctx->header_length, ctx->header,
                                   ctx->base.callback_data);
                ctx->header_length = 0;
@@ -1316,20 +1396,18 @@ static int handle_ir_dualbuffer_packet(struct context *context,
        return 1;
 }
 
-#define ISO_BUFFER_SIZE (64 * 1024)
-
 static int handle_it_packet(struct context *context,
                            struct descriptor *d,
                            struct descriptor *last)
 {
        struct iso_context *ctx =
                container_of(context, struct iso_context, context);
+
        if (last->transfer_status == 0)
                /* This descriptor isn't done yet, stop iteration. */
                return 0;
 
-       if (le16_to_cpu(last->control) & descriptor_irq_always)
+       if (le16_to_cpu(last->control) & DESCRIPTOR_IRQ_ALWAYS)
                ctx->base.callback(&ctx->base, le16_to_cpu(last->res_count),
                                   0, NULL, ctx->base.callback_data);
 
@@ -1351,14 +1429,12 @@ ohci_allocate_iso_context(struct fw_card *card, int type, size_t header_size)
                list = ohci->it_context_list;
                callback = handle_it_packet;
        } else {
-               mask = &ohci->ir_context_mask;
-               list = ohci->ir_context_list;
-               if (header_size > 0)
-                       callback = handle_ir_dualbuffer_packet;
-               else
-                       callback = handle_ir_bufferfill_packet;
+               mask = &ohci->ir_context_mask;
+               list = ohci->ir_context_list;
+               callback = handle_ir_dualbuffer_packet;
        }
 
+       /* FIXME: We need a fallback for pre 1.1 OHCI. */
        if (callback == handle_ir_dualbuffer_packet &&
            ohci->version < OHCI_VERSION_1_1)
                return ERR_PTR(-EINVAL);
@@ -1372,13 +1448,13 @@ ohci_allocate_iso_context(struct fw_card *card, int type, size_t header_size)
        if (index < 0)
                return ERR_PTR(-EBUSY);
 
-       if (type == FW_ISO_CONTEXT_TRANSMIT)
-               regs = OHCI1394_IsoXmitContextBase(index);
-       else
-               regs = OHCI1394_IsoRcvContextBase(index);
+       if (type == FW_ISO_CONTEXT_TRANSMIT)
+               regs = OHCI1394_IsoXmitContextBase(index);
+       else
+               regs = OHCI1394_IsoRcvContextBase(index);
+
        ctx = &list[index];
-       memset(ctx, 0, sizeof *ctx);
+       memset(ctx, 0, sizeof(*ctx));
        ctx->header_length = 0;
        ctx->header = (void *) __get_free_page(GFP_KERNEL);
        if (ctx->header == NULL)
@@ -1401,34 +1477,37 @@ ohci_allocate_iso_context(struct fw_card *card, int type, size_t header_size)
        return ERR_PTR(retval);
 }
 
-static int ohci_start_iso(struct fw_iso_context *base, s32 cycle)
+static int ohci_start_iso(struct fw_iso_context *base,
+                         s32 cycle, u32 sync, u32 tags)
 {
-       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
        struct fw_ohci *ohci = ctx->context.ohci;
-       u32 cycle_match = 0, mode;
+       u32 control, match;
        int index;
 
        if (ctx->base.type == FW_ISO_CONTEXT_TRANSMIT) {
                index = ctx - ohci->it_context_list;
-               if (cycle > 0)
-                       cycle_match = IT_CONTEXT_CYCLE_MATCH_ENABLE |
+               match = 0;
+               if (cycle >= 0)
+                       match = IT_CONTEXT_CYCLE_MATCH_ENABLE |
                                (cycle & 0x7fff) << 16;
 
                reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 1 << index);
                reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, 1 << index);
-               context_run(&ctx->context, cycle_match);
+               context_run(&ctx->context, match);
        } else {
                index = ctx - ohci->ir_context_list;
+               control = IR_CONTEXT_DUAL_BUFFER_MODE | IR_CONTEXT_ISOCH_HEADER;
+               match = (tags << 28) | (sync << 8) | ctx->base.channel;
+               if (cycle >= 0) {
+                       match |= (cycle & 0x07fff) << 12;
+                       control |= IR_CONTEXT_CYCLE_MATCH_ENABLE;
+               }
 
-               if (ctx->base.header_size > 0)
-                       mode = IR_CONTEXT_DUAL_BUFFER_MODE;
-               else
-                       mode = IR_CONTEXT_BUFFER_FILL;
                reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 1 << index);
                reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, 1 << index);
-               reg_write(ohci, context_match(ctx->context.regs),
-                         0xf0000000 | ctx->base.channel);
-               context_run(&ctx->context, mode);
+               reg_write(ohci, CONTEXT_MATCH(ctx->context.regs), match);
+               context_run(&ctx->context, control);
        }
 
        return 0;
@@ -1437,7 +1516,7 @@ static int ohci_start_iso(struct fw_iso_context *base, s32 cycle)
 static int ohci_stop_iso(struct fw_iso_context *base)
 {
        struct fw_ohci *ohci = fw_ohci(base->card);
-       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
        int index;
 
        if (ctx->base.type == FW_ISO_CONTEXT_TRANSMIT) {
@@ -1456,7 +1535,7 @@ static int ohci_stop_iso(struct fw_iso_context *base)
 static void ohci_free_iso_context(struct fw_iso_context *base)
 {
        struct fw_ohci *ohci = fw_ohci(base->card);
-       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
        unsigned long flags;
        int index;
 
@@ -1483,7 +1562,7 @@ ohci_queue_iso_transmit(struct fw_iso_context *base,
                        struct fw_iso_buffer *buffer,
                        unsigned long payload)
 {
-       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
        struct descriptor *d, *last, *pd;
        struct fw_iso_packet *p;
        __le32 *header;
@@ -1492,8 +1571,10 @@ ohci_queue_iso_transmit(struct fw_iso_context *base,
        u32 payload_index, payload_end_index, next_page_index;
        int page, end_page, i, length, offset;
 
-       /* FIXME: Cycle lost behavior should be configurable: lose
-        * packet, retransmit or terminate.. */
+       /*
+        * FIXME: Cycle lost behavior should be configurable: lose
+        * packet, retransmit or terminate..
+        */
 
        p = packet;
        payload_index = payload;
@@ -1515,30 +1596,30 @@ ohci_queue_iso_transmit(struct fw_iso_context *base,
        z += payload_z;
 
        /* Get header size in number of descriptors. */
-       header_z = DIV_ROUND_UP(p->header_length, sizeof *d);
+       header_z = DIV_ROUND_UP(p->header_length, sizeof(*d));
 
        d = context_get_descriptors(&ctx->context, z + header_z, &d_bus);
        if (d == NULL)
                return -ENOMEM;
 
        if (!p->skip) {
-               d[0].control   = cpu_to_le16(descriptor_key_immediate);
+               d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
                d[0].req_count = cpu_to_le16(8);
 
                header = (__le32 *) &d[1];
-               header[0] = cpu_to_le32(it_header_sy(p->sy) |
-                                       it_header_tag(p->tag) |
-                                       it_header_tcode(TCODE_STREAM_DATA) |
-                                       it_header_channel(ctx->base.channel) |
-                                       it_header_speed(ctx->base.speed));
+               header[0] = cpu_to_le32(IT_HEADER_SY(p->sy) |
+                                       IT_HEADER_TAG(p->tag) |
+                                       IT_HEADER_TCODE(TCODE_STREAM_DATA) |
+                                       IT_HEADER_CHANNEL(ctx->base.channel) |
+                                       IT_HEADER_SPEED(ctx->base.speed));
                header[1] =
-                       cpu_to_le32(it_header_data_length(p->header_length +
+                       cpu_to_le32(IT_HEADER_DATA_LENGTH(p->header_length +
                                                          p->payload_length));
        }
 
        if (p->header_length > 0) {
                d[2].req_count    = cpu_to_le16(p->header_length);
-               d[2].data_address = cpu_to_le32(d_bus + z * sizeof *d);
+               d[2].data_address = cpu_to_le32(d_bus + z * sizeof(*d));
                memcpy(&d[z], p->header, p->header_length);
        }
 
@@ -1559,14 +1640,14 @@ ohci_queue_iso_transmit(struct fw_iso_context *base,
        }
 
        if (p->interrupt)
-               irq = descriptor_irq_always;
+               irq = DESCRIPTOR_IRQ_ALWAYS;
        else
-               irq = descriptor_no_irq;
+               irq = DESCRIPTOR_NO_IRQ;
 
        last = z == 2 ? d : d + z - 1;
-       last->control |= cpu_to_le16(descriptor_output_last |
-                                    descriptor_status |
-                                    descriptor_branch_always |
+       last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
+                                    DESCRIPTOR_STATUS |
+                                    DESCRIPTOR_BRANCH_ALWAYS |
                                     irq);
 
        context_append(&ctx->context, d, z, header_z);
@@ -1586,22 +1667,43 @@ ohci_queue_iso_receive_dualbuffer(struct fw_iso_context *base,
        struct fw_iso_packet *p;
        dma_addr_t d_bus, page_bus;
        u32 z, header_z, length, rest;
-       int page, offset;
-       /* FIXME: Cycle lost behavior should be configurable: lose
-        * packet, retransmit or terminate.. */
+       int page, offset, packet_count, header_size;
+
+       /*
+        * FIXME: Cycle lost behavior should be configurable: lose
+        * packet, retransmit or terminate..
+        */
+
+       if (packet->skip) {
+               d = context_get_descriptors(&ctx->context, 2, &d_bus);
+               if (d == NULL)
+                       return -ENOMEM;
+
+               db = (struct db_descriptor *) d;
+               db->control = cpu_to_le16(DESCRIPTOR_STATUS |
+                                         DESCRIPTOR_BRANCH_ALWAYS |
+                                         DESCRIPTOR_WAIT);
+               db->first_size = cpu_to_le16(ctx->base.header_size + 4);
+               context_append(&ctx->context, d, 2, 0);
+       }
 
        p = packet;
        z = 2;
 
+       /*
+        * The OHCI controller puts the status word in the header
+        * buffer too, so we need 4 extra bytes per packet.
+        */
+       packet_count = p->header_length / ctx->base.header_size;
+       header_size = packet_count * (ctx->base.header_size + 4);
+
        /* Get header size in number of descriptors. */
-       header_z = DIV_ROUND_UP(p->header_length, sizeof *d);
+       header_z = DIV_ROUND_UP(header_size, sizeof(*d));
        page     = payload >> PAGE_SHIFT;
        offset   = payload & ~PAGE_MASK;
        rest     = p->payload_length;
 
        /* FIXME: OHCI 1.0 doesn't support dual buffer receive */
-       /* FIXME: handle descriptor_wait */
        /* FIXME: make packet-per-buffer/dual-buffer a context option */
        while (rest > 0) {
                d = context_get_descriptors(&ctx->context,
@@ -1610,13 +1712,13 @@ ohci_queue_iso_receive_dualbuffer(struct fw_iso_context *base,
                        return -ENOMEM;
 
                db = (struct db_descriptor *) d;
-               db->control = cpu_to_le16(descriptor_status |
-                                         descriptor_branch_always);
-               db->first_size = cpu_to_le16(ctx->base.header_size);
-               db->first_req_count = cpu_to_le16(p->header_length);
+               db->control = cpu_to_le16(DESCRIPTOR_STATUS |
+                                         DESCRIPTOR_BRANCH_ALWAYS);
+               db->first_size = cpu_to_le16(ctx->base.header_size + 4);
+               db->first_req_count = cpu_to_le16(header_size);
                db->first_res_count = db->first_req_count;
-               db->first_buffer = cpu_to_le32(d_bus + sizeof *db);
-               
+               db->first_buffer = cpu_to_le32(d_bus + sizeof(*db));
+
                if (offset + rest < PAGE_SIZE)
                        length = rest;
                else
@@ -1628,7 +1730,7 @@ ohci_queue_iso_receive_dualbuffer(struct fw_iso_context *base,
                db->second_buffer = cpu_to_le32(page_bus + offset);
 
                if (p->interrupt && length == rest)
-                       db->control |= cpu_to_le16(descriptor_irq_always);
+                       db->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
 
                context_append(&ctx->context, d, z, header_z);
                offset = (offset + length) & ~PAGE_MASK;
@@ -1639,54 +1741,6 @@ ohci_queue_iso_receive_dualbuffer(struct fw_iso_context *base,
        return 0;
 }
 
-static int
-ohci_queue_iso_receive_bufferfill(struct fw_iso_context *base,
-                                 struct fw_iso_packet *packet,
-                                 struct fw_iso_buffer *buffer,
-                                 unsigned long payload)
-{
-       struct iso_context *ctx = container_of(base, struct iso_context, base);
-       struct descriptor *d = NULL;
-       dma_addr_t d_bus, page_bus;
-       u32 length, rest;
-       int page, offset;
-       page   = payload >> PAGE_SHIFT;
-       offset = payload & ~PAGE_MASK;
-       rest   = packet->payload_length;
-
-       while (rest > 0) {
-               d = context_get_descriptors(&ctx->context, 1, &d_bus);
-               if (d == NULL)
-                       return -ENOMEM;
-
-               d->control = cpu_to_le16(descriptor_input_more |
-                                        descriptor_status |
-                                        descriptor_branch_always);
-               
-               if (offset + rest < PAGE_SIZE)
-                       length = rest;
-               else
-                       length = PAGE_SIZE - offset;
-
-               page_bus = page_private(buffer->pages[page]);
-               d->data_address = cpu_to_le32(page_bus + offset);
-               d->req_count = cpu_to_le16(length);
-               d->res_count = cpu_to_le16(length);
-
-               if (packet->interrupt && length == rest)
-                       d->control |= cpu_to_le16(descriptor_irq_always);
-
-               context_append(&ctx->context, d, 1, 0);
-
-               offset = (offset + length) & ~PAGE_MASK;
-               rest -= length;
-               page++;
-       }
-
-       return 0;
-}
-
 static int
 ohci_queue_iso(struct fw_iso_context *base,
               struct fw_iso_packet *packet,
@@ -1697,9 +1751,6 @@ ohci_queue_iso(struct fw_iso_context *base,
 
        if (base->type == FW_ISO_CONTEXT_TRANSMIT)
                return ohci_queue_iso_transmit(base, packet, buffer, payload);
-       else if (base->header_size == 0)
-               return ohci_queue_iso_receive_bufferfill(base, packet,
-                                                        buffer, payload);
        else if (ctx->context.ohci->version >= OHCI_VERSION_1_1)
                return ohci_queue_iso_receive_dualbuffer(base, packet,
                                                         buffer, payload);
@@ -1717,6 +1768,7 @@ static const struct fw_card_driver ohci_driver = {
        .send_response          = ohci_send_response,
        .cancel_packet          = ohci_cancel_packet,
        .enable_phys_dma        = ohci_enable_phys_dma,
+       .get_bus_time           = ohci_get_bus_time,
 
        .allocate_iso_context   = ohci_allocate_iso_context,
        .free_iso_context       = ohci_free_iso_context,
@@ -1725,65 +1777,16 @@ static const struct fw_card_driver ohci_driver = {
        .stop_iso               = ohci_stop_iso,
 };
 
-static int software_reset(struct fw_ohci *ohci)
-{
-       int i;
-
-       reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);
-
-       for (i = 0; i < OHCI_LOOP_COUNT; i++) {
-               if ((reg_read(ohci, OHCI1394_HCControlSet) &
-                    OHCI1394_HCControl_softReset) == 0)
-                       return 0;
-               msleep(1);
-       }
-
-       return -EBUSY;
-}
-
-/* ---------- pci subsystem interface ---------- */
-
-enum {
-       CLEANUP_SELF_ID,
-       CLEANUP_REGISTERS,
-       CLEANUP_IOMEM,
-       CLEANUP_DISABLE,
-       CLEANUP_PUT_CARD,
-};
-
-static int cleanup(struct fw_ohci *ohci, int stage, int code)
-{
-       struct pci_dev *dev = to_pci_dev(ohci->card.device);
-
-       switch (stage) {
-       case CLEANUP_SELF_ID:
-               dma_free_coherent(ohci->card.device, SELF_ID_BUF_SIZE,
-                                 ohci->self_id_cpu, ohci->self_id_bus);
-       case CLEANUP_REGISTERS:
-               kfree(ohci->it_context_list);
-               kfree(ohci->ir_context_list);
-               pci_iounmap(dev, ohci->registers);
-       case CLEANUP_IOMEM:
-               pci_release_region(dev, 0);
-       case CLEANUP_DISABLE:
-               pci_disable_device(dev);
-       case CLEANUP_PUT_CARD:
-               fw_card_put(&ohci->card);
-       }
-
-       return code;
-}
-
 static int __devinit
 pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
 {
        struct fw_ohci *ohci;
        u32 bus_options, max_receive, link_speed;
        u64 guid;
-       int error_code;
+       int err;
        size_t size;
 
-       ohci = kzalloc(sizeof *ohci, GFP_KERNEL);
+       ohci = kzalloc(sizeof(*ohci), GFP_KERNEL);
        if (ohci == NULL) {
                fw_error("Could not malloc fw_ohci data.\n");
                return -ENOMEM;
@@ -1791,9 +1794,10 @@ pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
 
        fw_card_initialize(&ohci->card, &ohci_driver, &dev->dev);
 
-       if (pci_enable_device(dev)) {
+       err = pci_enable_device(dev);
+       if (err) {
                fw_error("Failed to enable OHCI hardware.\n");
-               return cleanup(ohci, CLEANUP_PUT_CARD, -ENODEV);
+               goto fail_put_card;
        }
 
        pci_set_master(dev);
@@ -1805,57 +1809,30 @@ pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
        tasklet_init(&ohci->bus_reset_tasklet,
                     bus_reset_tasklet, (unsigned long)ohci);
 
-       if (pci_request_region(dev, 0, ohci_driver_name)) {
+       err = pci_request_region(dev, 0, ohci_driver_name);
+       if (err) {
                fw_error("MMIO resource unavailable\n");
-               return cleanup(ohci, CLEANUP_DISABLE, -EBUSY);
+               goto fail_disable;
        }
 
        ohci->registers = pci_iomap(dev, 0, OHCI1394_REGISTER_SIZE);
        if (ohci->registers == NULL) {
                fw_error("Failed to remap registers\n");
-               return cleanup(ohci, CLEANUP_IOMEM, -ENXIO);
-       }
-
-       if (software_reset(ohci)) {
-               fw_error("Failed to reset ohci card.\n");
-               return cleanup(ohci, CLEANUP_REGISTERS, -EBUSY);
+               err = -ENXIO;
+               goto fail_iomem;
        }
 
-       /* Now enable LPS, which we need in order to start accessing
-        * most of the registers.  In fact, on some cards (ALI M5251),
-        * accessing registers in the SClk domain without LPS enabled
-        * will lock up the machine.  Wait 50msec to make sure we have
-        * full link enabled.  */
-       reg_write(ohci, OHCI1394_HCControlSet,
-                 OHCI1394_HCControl_LPS |
-                 OHCI1394_HCControl_postedWriteEnable);
-       flush_writes(ohci);
-       msleep(50);
-
-       reg_write(ohci, OHCI1394_HCControlClear,
-                 OHCI1394_HCControl_noByteSwapData);
-
-       reg_write(ohci, OHCI1394_LinkControlSet,
-                 OHCI1394_LinkControl_rcvSelfID |
-                 OHCI1394_LinkControl_cycleTimerEnable |
-                 OHCI1394_LinkControl_cycleMaster);
-
        ar_context_init(&ohci->ar_request_ctx, ohci,
                        OHCI1394_AsReqRcvContextControlSet);
 
        ar_context_init(&ohci->ar_response_ctx, ohci,
                        OHCI1394_AsRspRcvContextControlSet);
 
-       at_context_init(&ohci->at_request_ctx, ohci,
-                       OHCI1394_AsReqTrContextControlSet);
+       context_init(&ohci->at_request_ctx, ohci, AT_BUFFER_SIZE,
+                    OHCI1394_AsReqTrContextControlSet, handle_at_packet);
 
-       at_context_init(&ohci->at_response_ctx, ohci,
-                       OHCI1394_AsRspTrContextControlSet);
-
-       reg_write(ohci, OHCI1394_ATRetries,
-                 OHCI1394_MAX_AT_REQ_RETRIES |
-                 (OHCI1394_MAX_AT_RESP_RETRIES << 4) |
-                 (OHCI1394_MAX_PHYS_RESP_RETRIES << 8));
+       context_init(&ohci->at_response_ctx, ohci, AT_BUFFER_SIZE,
+                    OHCI1394_AsRspTrContextControlSet, handle_at_packet);
 
        reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
        ohci->it_context_mask = reg_read(ohci, OHCI1394_IsoRecvIntMaskSet);
@@ -1871,7 +1848,8 @@ pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
 
        if (ohci->it_context_list == NULL || ohci->ir_context_list == NULL) {
                fw_error("Out of memory for it/ir contexts.\n");
-               return cleanup(ohci, CLEANUP_REGISTERS, -ENOMEM);
+               err = -ENOMEM;
+               goto fail_registers;
        }
 
        /* self-id dma buffer allocation */
@@ -1881,35 +1859,41 @@ pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
                                               GFP_KERNEL);
        if (ohci->self_id_cpu == NULL) {
                fw_error("Out of memory for self ID buffer.\n");
-               return cleanup(ohci, CLEANUP_REGISTERS, -ENOMEM);
+               err = -ENOMEM;
+               goto fail_registers;
        }
 
-       reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->self_id_bus);
-       reg_write(ohci, OHCI1394_PhyUpperBound, 0x00010000);
-       reg_write(ohci, OHCI1394_IntEventClear, ~0);
-       reg_write(ohci, OHCI1394_IntMaskClear, ~0);
-       reg_write(ohci, OHCI1394_IntMaskSet,
-                 OHCI1394_selfIDComplete |
-                 OHCI1394_RQPkt | OHCI1394_RSPkt |
-                 OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
-                 OHCI1394_isochRx | OHCI1394_isochTx |
-                 OHCI1394_masterIntEnable);
-
        bus_options = reg_read(ohci, OHCI1394_BusOptions);
        max_receive = (bus_options >> 12) & 0xf;
        link_speed = bus_options & 0x7;
        guid = ((u64) reg_read(ohci, OHCI1394_GUIDHi) << 32) |
                reg_read(ohci, OHCI1394_GUIDLo);
 
-       error_code = fw_card_add(&ohci->card, max_receive, link_speed, guid);
-       if (error_code < 0)
-               return cleanup(ohci, CLEANUP_SELF_ID, error_code);
+       err = fw_card_add(&ohci->card, max_receive, link_speed, guid);
+       if (err < 0)
+               goto fail_self_id;
 
        ohci->version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
        fw_notify("Added fw-ohci device %s, OHCI version %x.%x\n",
                  dev->dev.bus_id, ohci->version >> 16, ohci->version & 0xff);
 
        return 0;
+
+ fail_self_id:
+       dma_free_coherent(ohci->card.device, SELF_ID_BUF_SIZE,
+                         ohci->self_id_cpu, ohci->self_id_bus);
+ fail_registers:
+       kfree(ohci->it_context_list);
+       kfree(ohci->ir_context_list);
+       pci_iounmap(dev, ohci->registers);
+ fail_iomem:
+       pci_release_region(dev, 0);
+ fail_disable:
+       pci_disable_device(dev);
+ fail_put_card:
+       fw_card_put(&ohci->card);
+
+       return err;
 }
 
 static void pci_remove(struct pci_dev *dev)
@@ -1917,19 +1901,68 @@ static void pci_remove(struct pci_dev *dev)
        struct fw_ohci *ohci;
 
        ohci = pci_get_drvdata(dev);
-       reg_write(ohci, OHCI1394_IntMaskClear, OHCI1394_masterIntEnable);
+       reg_write(ohci, OHCI1394_IntMaskClear, ~0);
+       flush_writes(ohci);
        fw_core_remove_card(&ohci->card);
 
-       /* FIXME: Fail all pending packets here, now that the upper
-        * layers can't queue any more. */
+       /*
+        * FIXME: Fail all pending packets here, now that the upper
+        * layers can't queue any more.
+        */
 
        software_reset(ohci);
        free_irq(dev->irq, ohci);
-       cleanup(ohci, CLEANUP_SELF_ID, 0);
+       dma_free_coherent(ohci->card.device, SELF_ID_BUF_SIZE,
+                         ohci->self_id_cpu, ohci->self_id_bus);
+       kfree(ohci->it_context_list);
+       kfree(ohci->ir_context_list);
+       pci_iounmap(dev, ohci->registers);
+       pci_release_region(dev, 0);
+       pci_disable_device(dev);
+       fw_card_put(&ohci->card);
 
        fw_notify("Removed fw-ohci device.\n");
 }
 
+#ifdef CONFIG_PM
+static int pci_suspend(struct pci_dev *pdev, pm_message_t state)
+{
+       struct fw_ohci *ohci = pci_get_drvdata(pdev);
+       int err;
+
+       software_reset(ohci);
+       free_irq(pdev->irq, ohci);
+       err = pci_save_state(pdev);
+       if (err) {
+               fw_error("pci_save_state failed with %d", err);
+               return err;
+       }
+       err = pci_set_power_state(pdev, pci_choose_state(pdev, state));
+       if (err) {
+               fw_error("pci_set_power_state failed with %d", err);
+               return err;
+       }
+
+       return 0;
+}
+
+static int pci_resume(struct pci_dev *pdev)
+{
+       struct fw_ohci *ohci = pci_get_drvdata(pdev);
+       int err;
+
+       pci_set_power_state(pdev, PCI_D0);
+       pci_restore_state(pdev);
+       err = pci_enable_device(pdev);
+       if (err) {
+               fw_error("pci_enable_device failed with %d", err);
+               return err;
+       }
+
+       return ohci_enable(&ohci->card, ohci->config_rom, CONFIG_ROM_SIZE);
+}
+#endif
+
 static struct pci_device_id pci_table[] = {
        { PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_FIREWIRE_OHCI, ~0) },
        { }
@@ -1942,12 +1975,21 @@ static struct pci_driver fw_ohci_pci_driver = {
        .id_table       = pci_table,
        .probe          = pci_probe,
        .remove         = pci_remove,
+#ifdef CONFIG_PM
+       .resume         = pci_resume,
+       .suspend        = pci_suspend,
+#endif
 };
 
 MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
 MODULE_DESCRIPTION("Driver for PCI OHCI IEEE1394 controllers");
 MODULE_LICENSE("GPL");
 
+/* Provide a module alias so root-on-sbp2 initrds don't break. */
+#ifndef CONFIG_IEEE1394_OHCI1394_MODULE
+MODULE_ALIAS("ohci1394");
+#endif
+
 static int __init fw_ohci_init(void)
 {
        return pci_register_driver(&fw_ohci_pci_driver);