firewire: add IPv4 support
Jay Fenlason [Mon, 18 May 2009 17:08:06 +0000 (13:08 -0400)]
Implement IPv4 over IEEE 1394 as per RFC 2734 for the newer firewire
stack.  This feature has only been present in the older ieee1394 stack
via the eth1394 driver.

Still to do:
  - fix ipv4_priv and ipv4_node lifetime logic
  - fix determination of speeds and max payloads
  - fix bus reset handling
  - fix unaligned memory accesses
  - fix coding style
  - further testing/ improvement of fragment reassembly
  - perhaps multicast support

Signed-off-by: Jay Fenlason <fenlason@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de> (rebased, copyright note, changelog)

drivers/firewire/Makefile
drivers/firewire/core-card.c
drivers/firewire/core-iso.c
drivers/firewire/core.h
drivers/firewire/fw-ipv4.c [new file with mode: 0644]
include/linux/firewire.h

index bc3b9bf..31edf30 100644 (file)
@@ -6,7 +6,9 @@ firewire-core-y += core-card.o core-cdev.o core-device.o \
                    core-iso.o core-topology.o core-transaction.o
 firewire-ohci-y += ohci.o
 firewire-sbp2-y += sbp2.o
+firewire-ipv4-y += fw-ipv4.o
 
 obj-$(CONFIG_FIREWIRE) += firewire-core.o
 obj-$(CONFIG_FIREWIRE_OHCI) += firewire-ohci.o
 obj-$(CONFIG_FIREWIRE_SBP2) += firewire-sbp2.o
+obj-$(CONFIG_FIREWIRE_IPV4) += firewire-ipv4.o
index 4c1be64..cdab32b 100644 (file)
@@ -176,6 +176,7 @@ int fw_core_add_descriptor(struct fw_descriptor *desc)
 
        return 0;
 }
+EXPORT_SYMBOL(fw_core_add_descriptor);
 
 void fw_core_remove_descriptor(struct fw_descriptor *desc)
 {
@@ -189,6 +190,7 @@ void fw_core_remove_descriptor(struct fw_descriptor *desc)
 
        mutex_unlock(&card_mutex);
 }
+EXPORT_SYMBOL(fw_core_remove_descriptor);
 
 static void allocate_broadcast_channel(struct fw_card *card, int generation)
 {
@@ -427,6 +429,8 @@ void fw_card_initialize(struct fw_card *card,
        card->local_node = NULL;
 
        INIT_DELAYED_WORK(&card->work, fw_card_bm_work);
+       card->netdev = NULL;
+       INIT_LIST_HEAD(&card->ipv4_nodes);
 }
 EXPORT_SYMBOL(fw_card_initialize);
 
index 28076c8..448ddd7 100644 (file)
@@ -80,6 +80,7 @@ int fw_iso_buffer_init(struct fw_iso_buffer *buffer, struct fw_card *card,
 
        return -ENOMEM;
 }
+EXPORT_SYMBOL(fw_iso_buffer_init);
 
 int fw_iso_buffer_map(struct fw_iso_buffer *buffer, struct vm_area_struct *vma)
 {
@@ -114,6 +115,7 @@ void fw_iso_buffer_destroy(struct fw_iso_buffer *buffer,
        kfree(buffer->pages);
        buffer->pages = NULL;
 }
+EXPORT_SYMBOL(fw_iso_buffer_destroy);
 
 struct fw_iso_context *fw_iso_context_create(struct fw_card *card,
                int type, int channel, int speed, size_t header_size,
@@ -136,6 +138,7 @@ struct fw_iso_context *fw_iso_context_create(struct fw_card *card,
 
        return ctx;
 }
+EXPORT_SYMBOL(fw_iso_context_create);
 
 void fw_iso_context_destroy(struct fw_iso_context *ctx)
 {
@@ -143,12 +146,14 @@ void fw_iso_context_destroy(struct fw_iso_context *ctx)
 
        card->driver->free_iso_context(ctx);
 }
+EXPORT_SYMBOL(fw_iso_context_destroy);
 
 int fw_iso_context_start(struct fw_iso_context *ctx,
                         int cycle, int sync, int tags)
 {
        return ctx->card->driver->start_iso(ctx, cycle, sync, tags);
 }
+EXPORT_SYMBOL(fw_iso_context_start);
 
 int fw_iso_context_queue(struct fw_iso_context *ctx,
                         struct fw_iso_packet *packet,
@@ -159,11 +164,13 @@ int fw_iso_context_queue(struct fw_iso_context *ctx,
 
        return card->driver->queue_iso(ctx, packet, buffer, payload);
 }
+EXPORT_SYMBOL(fw_iso_context_queue);
 
 int fw_iso_context_stop(struct fw_iso_context *ctx)
 {
        return ctx->card->driver->stop_iso(ctx);
 }
+EXPORT_SYMBOL(fw_iso_context_stop);
 
 /*
  * Isochronous bus resource management (channels, bandwidth), client side
index 0a25a7b..c3cfc64 100644 (file)
@@ -1,7 +1,6 @@
 #ifndef _FIREWIRE_CORE_H
 #define _FIREWIRE_CORE_H
 
-#include <linux/dma-mapping.h>
 #include <linux/fs.h>
 #include <linux/list.h>
 #include <linux/idr.h>
@@ -97,17 +96,6 @@ int fw_core_initiate_bus_reset(struct fw_card *card, int short_reset);
 int fw_compute_block_crc(u32 *block);
 void fw_schedule_bm_work(struct fw_card *card, unsigned long delay);
 
-struct fw_descriptor {
-       struct list_head link;
-       size_t length;
-       u32 immediate;
-       u32 key;
-       const u32 *data;
-};
-
-int fw_core_add_descriptor(struct fw_descriptor *desc);
-void fw_core_remove_descriptor(struct fw_descriptor *desc);
-
 
 /* -cdev */
 
@@ -130,77 +118,7 @@ void fw_node_event(struct fw_card *card, struct fw_node *node, int event);
 
 /* -iso */
 
-/*
- * The iso packet format allows for an immediate header/payload part
- * stored in 'header' immediately after the packet info plus an
- * indirect payload part that is pointer to by the 'payload' field.
- * Applications can use one or the other or both to implement simple
- * low-bandwidth streaming (e.g. audio) or more advanced
- * scatter-gather streaming (e.g. assembling video frame automatically).
- */
-struct fw_iso_packet {
-       u16 payload_length;     /* Length of indirect payload. */
-       u32 interrupt:1;        /* Generate interrupt on this packet */
-       u32 skip:1;             /* Set to not send packet at all. */
-       u32 tag:2;
-       u32 sy:4;
-       u32 header_length:8;    /* Length of immediate header. */
-       u32 header[0];
-};
-
-#define FW_ISO_CONTEXT_TRANSMIT        0
-#define FW_ISO_CONTEXT_RECEIVE 1
-
-#define FW_ISO_CONTEXT_MATCH_TAG0       1
-#define FW_ISO_CONTEXT_MATCH_TAG1       2
-#define FW_ISO_CONTEXT_MATCH_TAG2       4
-#define FW_ISO_CONTEXT_MATCH_TAG3       8
-#define FW_ISO_CONTEXT_MATCH_ALL_TAGS  15
-
-/*
- * An iso buffer is just a set of pages mapped for DMA in the
- * specified direction.  Since the pages are to be used for DMA, they
- * are not mapped into the kernel virtual address space.  We store the
- * DMA address in the page private. The helper function
- * fw_iso_buffer_map() will map the pages into a given vma.
- */
-struct fw_iso_buffer {
-       enum dma_data_direction direction;
-       struct page **pages;
-       int page_count;
-};
-
-typedef void (*fw_iso_callback_t)(struct fw_iso_context *context,
-                                 u32 cycle, size_t header_length,
-                                 void *header, void *data);
-
-struct fw_iso_context {
-       struct fw_card *card;
-       int type;
-       int channel;
-       int speed;
-       size_t header_size;
-       fw_iso_callback_t callback;
-       void *callback_data;
-};
-
-int fw_iso_buffer_init(struct fw_iso_buffer *buffer, struct fw_card *card,
-                      int page_count, enum dma_data_direction direction);
 int fw_iso_buffer_map(struct fw_iso_buffer *buffer, struct vm_area_struct *vma);
-void fw_iso_buffer_destroy(struct fw_iso_buffer *buffer, struct fw_card *card);
-
-struct fw_iso_context *fw_iso_context_create(struct fw_card *card,
-               int type, int channel, int speed, size_t header_size,
-               fw_iso_callback_t callback, void *callback_data);
-int fw_iso_context_queue(struct fw_iso_context *ctx,
-                        struct fw_iso_packet *packet,
-                        struct fw_iso_buffer *buffer,
-                        unsigned long payload);
-int fw_iso_context_start(struct fw_iso_context *ctx,
-                        int cycle, int sync, int tags);
-int fw_iso_context_stop(struct fw_iso_context *ctx);
-void fw_iso_context_destroy(struct fw_iso_context *ctx);
-
 void fw_iso_resource_manage(struct fw_card *card, int generation,
                u64 channels_mask, int *channel, int *bandwidth, bool allocate);
 
@@ -285,9 +203,4 @@ void fw_flush_transactions(struct fw_card *card);
 void fw_send_phy_config(struct fw_card *card,
                        int node_id, int generation, int gap_count);
 
-static inline int fw_stream_packet_destination_id(int tag, int channel, int sy)
-{
-       return tag << 14 | channel << 8 | sy;
-}
-
 #endif /* _FIREWIRE_CORE_H */
diff --git a/drivers/firewire/fw-ipv4.c b/drivers/firewire/fw-ipv4.c
new file mode 100644 (file)
index 0000000..4de6dbb
--- /dev/null
@@ -0,0 +1,1819 @@
+/*
+ * IPv4 over IEEE 1394, per RFC 2734
+ *
+ * Copyright (C) 2009 Jay Fenlason <fenlason@redhat.com>
+ *
+ * based on eth1394 by Ben Collins et al
+ */
+
+#include <linux/device.h>
+#include <linux/ethtool.h>
+#include <linux/firewire.h>
+#include <linux/firewire-constants.h>
+#include <linux/highmem.h>
+#include <linux/in.h>
+#include <linux/ip.h>
+#include <linux/mod_devicetable.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/netdevice.h>
+#include <linux/skbuff.h>
+
+#include <asm/unaligned.h>
+#include <net/arp.h>
+
+/* Things to potentially make runtime cofigurable */
+/* must be at least as large as our maximum receive size */
+#define FIFO_SIZE 4096
+/* Network timeout in glibbles */
+#define IPV4_TIMEOUT       100000
+
+/* Runitme configurable paramaters */
+static int ipv4_mpd = 25;
+static int ipv4_max_xmt = 0;
+/* 16k for receiving arp and broadcast packets.  Enough? */
+static int ipv4_iso_page_count = 4;
+
+MODULE_AUTHOR("Jay Fenlason (fenlason@redhat.com)");
+MODULE_DESCRIPTION("Firewire IPv4 Driver (IPv4-over-IEEE1394 as per RFC 2734)");
+MODULE_LICENSE("GPL");
+MODULE_DEVICE_TABLE(ieee1394, ipv4_id_table);
+module_param_named(max_partial_datagrams, ipv4_mpd, int, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(max_partial_datagrams, "Maximum number of received"
+ " incomplete fragmented datagrams (default = 25).");
+
+/* Max xmt is useful for forcing fragmentation, which makes testing easier. */
+module_param_named(max_transmit, ipv4_max_xmt, int, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(max_transmit, "Maximum datagram size to transmit"
+ " (larger datagrams will be fragmented) (default = 0 (use hardware defaults).");
+
+/* iso page count controls how many pages will be used for receiving broadcast packets. */
+module_param_named(iso_pages, ipv4_iso_page_count, int, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(iso_pages, "Number of pages to use for receiving broadcast packets"
+ " (default = 4).");
+
+/* uncomment this line to do debugging */
+#define fw_debug(s, args...) printk(KERN_DEBUG KBUILD_MODNAME ": " s, ## args)
+
+/* comment out these lines to do debugging. */
+/* #undef fw_debug */
+/* #define fw_debug(s...) */
+/* #define print_hex_dump(l...) */
+
+/* Define a fake hardware header format for the networking core.  Note that
+ * header size cannot exceed 16 bytes as that is the size of the header cache.
+ * Also, we do not need the source address in the header so we omit it and
+ * keep the header to under 16 bytes */
+#define IPV4_ALEN (8)
+/* This must equal sizeof(struct ipv4_ether_hdr) */
+#define IPV4_HLEN (10)
+
+/* FIXME: what's a good size for this? */
+#define INVALID_FIFO_ADDR (u64)~0ULL
+
+/* Things specified by standards */
+#define BROADCAST_CHANNEL 31
+
+#define S100_BUFFER_SIZE 512
+#define MAX_BUFFER_SIZE 4096
+
+#define IPV4_GASP_SPECIFIER_ID 0x00005EU
+#define IPV4_GASP_VERSION      0x00000001U
+
+#define IPV4_GASP_OVERHEAD (2 * sizeof(u32)) /* for GASP header */
+
+#define IPV4_UNFRAG_HDR_SIZE   sizeof(u32)
+#define IPV4_FRAG_HDR_SIZE     (2 * sizeof(u32))
+#define IPV4_FRAG_OVERHEAD     sizeof(u32)
+
+#define ALL_NODES (0xffc0 | 0x003f)
+
+#define IPV4_HDR_UNFRAG                0       /* unfragmented         */
+#define IPV4_HDR_FIRSTFRAG     1       /* first fragment       */
+#define IPV4_HDR_LASTFRAG      2       /* last fragment        */
+#define IPV4_HDR_INTFRAG       3       /* interior fragment    */
+
+/* Our arp packet (ARPHRD_IEEE1394) */
+/* FIXME: note that this is probably bogus on weird-endian machines */
+struct ipv4_arp {
+       u16 hw_type;            /* 0x0018       */
+       u16 proto_type;         /* 0x0806       */
+       u8 hw_addr_len;         /* 16           */
+       u8 ip_addr_len;         /* 4            */
+       u16 opcode;             /* ARP Opcode   */
+       /* Above is exactly the same format as struct arphdr */
+
+       u64 s_uniq_id;          /* Sender's 64bit EUI                   */
+       u8 max_rec;             /* Sender's max packet size             */
+       u8 sspd;                /* Sender's max speed                   */
+       u16 fifo_hi;            /* hi 16bits of sender's FIFO addr      */
+       u32 fifo_lo;            /* lo 32bits of sender's FIFO addr      */
+       u32 sip;                /* Sender's IP Address                  */
+       u32 tip;                /* IP Address of requested hw addr      */
+} __attribute__((packed));
+
+struct ipv4_ether_hdr {
+       unsigned char   h_dest[IPV4_ALEN];      /* destination address */
+       unsigned short  h_proto;                /* packet type ID field */
+}  __attribute__((packed));
+
+static inline struct ipv4_ether_hdr *ipv4_ether_hdr(const struct sk_buff *skb)
+{
+       return (struct ipv4_ether_hdr *)skb_mac_header(skb);
+}
+
+enum ipv4_tx_type {
+       IPV4_UNKNOWN = 0,
+       IPV4_GASP = 1,
+       IPV4_WRREQ = 2,
+};
+
+enum ipv4_broadcast_state {
+       IPV4_BROADCAST_ERROR,
+       IPV4_BROADCAST_RUNNING,
+       IPV4_BROADCAST_STOPPED,
+};
+
+#define ipv4_get_hdr_lf(h)             (((h)->w0&0xC0000000)>>30)
+#define ipv4_get_hdr_ether_type(h)     (((h)->w0&0x0000FFFF)    )
+#define ipv4_get_hdr_dg_size(h)                (((h)->w0&0x0FFF0000)>>16)
+#define ipv4_get_hdr_fg_off(h)         (((h)->w0&0x00000FFF)    )
+#define ipv4_get_hdr_dgl(h)            (((h)->w1&0xFFFF0000)>>16)
+
+#define ipv4_set_hdr_lf(lf)            (( lf)<<30)
+#define ipv4_set_hdr_ether_type(et)    (( et)    )
+#define ipv4_set_hdr_dg_size(dgs)      ((dgs)<<16)
+#define ipv4_set_hdr_fg_off(fgo)       ((fgo)    )
+
+#define ipv4_set_hdr_dgl(dgl)          ((dgl)<<16)
+
+struct ipv4_hdr {
+       u32 w0;
+       u32 w1;
+};
+
+static inline void ipv4_make_uf_hdr( struct ipv4_hdr *hdr, unsigned ether_type) {
+       hdr->w0 = ipv4_set_hdr_lf(IPV4_HDR_UNFRAG)
+                  |ipv4_set_hdr_ether_type(ether_type);
+       fw_debug ( "Setting unfragmented header %p to %x\n", hdr, hdr->w0 );
+}
+
+static inline void ipv4_make_ff_hdr ( struct ipv4_hdr *hdr, unsigned ether_type, unsigned dg_size, unsigned dgl ) {
+       hdr->w0 = ipv4_set_hdr_lf(IPV4_HDR_FIRSTFRAG)
+                  |ipv4_set_hdr_dg_size(dg_size)
+                  |ipv4_set_hdr_ether_type(ether_type);
+       hdr->w1 = ipv4_set_hdr_dgl(dgl);
+       fw_debug ( "Setting fragmented header %p to first_frag %x,%x (et %x, dgs %x, dgl %x)\n", hdr, hdr->w0, hdr->w1,
+ ether_type, dg_size, dgl );
+}
+
+static inline void ipv4_make_sf_hdr ( struct ipv4_hdr *hdr, unsigned lf, unsigned dg_size, unsigned fg_off, unsigned dgl) {
+       hdr->w0 = ipv4_set_hdr_lf(lf)
+                |ipv4_set_hdr_dg_size(dg_size)
+                |ipv4_set_hdr_fg_off(fg_off);
+       hdr->w1 = ipv4_set_hdr_dgl(dgl);
+       fw_debug ( "Setting fragmented header %p to %x,%x (lf %x, dgs %x, fo %x dgl %x)\n",
+ hdr, hdr->w0, hdr->w1,
+ lf, dg_size, fg_off, dgl );
+}
+
+/* End of IP1394 headers */
+
+/* Fragment types */
+#define ETH1394_HDR_LF_UF      0       /* unfragmented         */
+#define ETH1394_HDR_LF_FF      1       /* first fragment       */
+#define ETH1394_HDR_LF_LF      2       /* last fragment        */
+#define ETH1394_HDR_LF_IF      3       /* interior fragment    */
+
+#define IP1394_HW_ADDR_LEN     16      /* As per RFC           */
+
+/* This list keeps track of what parts of the datagram have been filled in */
+struct ipv4_fragment_info {
+        struct list_head fragment_info;
+       u16 offset;
+       u16 len;
+};
+
+struct ipv4_partial_datagram {
+       struct list_head pdg_list;
+       struct list_head fragment_info;
+       struct sk_buff *skb;
+       /* FIXME Why not use skb->data? */
+       char *pbuf;
+       u16 datagram_label;
+       u16 ether_type;
+       u16 datagram_size;
+};
+
+/*
+ * We keep one of these for each IPv4 capable device attached to a fw_card.
+ * The list of them is stored in the fw_card structure rather than in the
+ * ipv4_priv because the remote IPv4 nodes may be probed before the card is,
+ * so we need a place to store them before the ipv4_priv structure is
+ * allocated.
+ */
+struct ipv4_node {
+       struct list_head ipv4_nodes;
+       /* guid of the remote node */
+       u64 guid;
+       /* FIFO address to transmit datagrams to, or INVALID_FIFO_ADDR */
+       u64 fifo;
+
+       spinlock_t pdg_lock;    /* partial datagram lock                */
+       /* List of partial datagrams received from this node */
+       struct list_head pdg_list;
+       /* Number of entries in pdg_list at the moment */
+       unsigned pdg_size;
+
+       /* max payload to transmit to this remote node */
+       /* This already includes the IPV4_FRAG_HDR_SIZE overhead */
+       u16 max_payload;
+       /* outgoing datagram label */
+       u16 datagram_label;
+       /* Current node_id of the remote node */
+       u16 nodeid;
+       /* current generation of the remote node */
+       u8 generation;
+       /* max speed that this node can receive at */
+       u8 xmt_speed;
+};
+
+struct ipv4_priv {
+       spinlock_t lock;
+
+       enum ipv4_broadcast_state broadcast_state;
+       struct fw_iso_context *broadcast_rcv_context;
+       struct fw_iso_buffer broadcast_rcv_buffer;
+       void **broadcast_rcv_buffer_ptrs;
+       unsigned broadcast_rcv_next_ptr;
+       unsigned num_broadcast_rcv_ptrs;
+       unsigned rcv_buffer_size;
+       /*
+        * This value is the maximum unfragmented datagram size that can be
+        * sent by the hardware.  It already has the GASP overhead and the
+        * unfragmented datagram header overhead calculated into it.
+        */
+       unsigned broadcast_xmt_max_payload;
+       u16 broadcast_xmt_datagramlabel;
+
+       /*
+        * The csr address that remote nodes must send datagrams to for us to
+        * receive them.
+        */
+       struct fw_address_handler handler;
+       u64 local_fifo;
+
+       /* Wake up to xmt        */
+        /* struct work_struct wake;*/
+       /* List of packets to be sent */
+       struct list_head packet_list;
+       /*
+        * List of packets that were broadcasted.  When we get an ISO interrupt
+        * one of them has been sent
+        */
+       struct list_head broadcasted_list;
+       /* List of packets that have been sent but not yet acked */
+       struct list_head sent_list;
+
+       struct fw_card *card;
+};
+
+/* This is our task struct. It's used for the packet complete callback.  */
+struct ipv4_packet_task {
+       /*
+        * ptask can actually be on priv->packet_list, priv->broadcasted_list,
+        * or priv->sent_list depending on its current state.
+        */
+       struct list_head packet_list;
+       struct fw_transaction transaction;
+       struct ipv4_hdr hdr;
+       struct sk_buff *skb;
+       struct ipv4_priv *priv;
+       enum ipv4_tx_type tx_type;
+       int outstanding_pkts;
+       unsigned max_payload;
+       u64 fifo_addr;
+       u16 dest_node;
+       u8 generation;
+       u8 speed;
+};
+
+static struct kmem_cache *ipv4_packet_task_cache;
+
+static const char ipv4_driver_name[] = "firewire-ipv4";
+
+static const struct ieee1394_device_id ipv4_id_table[] = {
+       {
+               .match_flags  = IEEE1394_MATCH_SPECIFIER_ID |
+                               IEEE1394_MATCH_VERSION,
+               .specifier_id = IPV4_GASP_SPECIFIER_ID,
+               .version      = IPV4_GASP_VERSION,
+       },
+       { }
+};
+
+static u32 ipv4_unit_directory_data[] = {
+       0x00040000,                                     /* unit directory */
+       0x12000000 | IPV4_GASP_SPECIFIER_ID,    /* specifier ID */
+       0x81000003,                                     /* text descriptor */
+       0x13000000 | IPV4_GASP_VERSION,         /* version */
+       0x81000005,                                     /* text descriptor */
+
+       0x00030000,                                     /* Three quadlets */
+       0x00000000,                                     /* Text */
+       0x00000000,                                     /* Language 0 */
+       0x49414e41,                                     /* I A N A */
+       0x00030000,                                     /* Three quadlets */
+       0x00000000,                                     /* Text */
+       0x00000000,                                     /* Language 0 */
+       0x49507634,                                     /* I P v 4 */
+};
+
+static struct fw_descriptor ipv4_unit_directory = {
+       .length = ARRAY_SIZE(ipv4_unit_directory_data),
+       .key = 0xd1000000,
+       .data = ipv4_unit_directory_data
+};
+
+static int ipv4_send_packet(struct ipv4_packet_task *ptask );
+
+/* ------------------------------------------------------------------ */
+/******************************************
+ * HW Header net device functions
+ ******************************************/
+  /* These functions have been adapted from net/ethernet/eth.c */
+
+/* Create a fake MAC header for an arbitrary protocol layer.
+ * saddr=NULL means use device source address
+ * daddr=NULL means leave destination address (eg unresolved arp). */
+
+static int ipv4_header ( struct sk_buff *skb, struct net_device *dev,
+                      unsigned short type, const void *daddr,
+                      const void *saddr, unsigned len) {
+       struct ipv4_ether_hdr *eth;
+
+       eth = (struct ipv4_ether_hdr *)skb_push(skb, sizeof(*eth));
+       eth->h_proto = htons(type);
+
+       if (dev->flags & (IFF_LOOPBACK | IFF_NOARP)) {
+               memset(eth->h_dest, 0, dev->addr_len);
+               return dev->hard_header_len;
+       }
+
+       if (daddr) {
+               memcpy(eth->h_dest, daddr, dev->addr_len);
+               return dev->hard_header_len;
+       }
+
+       return -dev->hard_header_len;
+}
+
+/* Rebuild the faked MAC header. This is called after an ARP
+ * (or in future other address resolution) has completed on this
+ * sk_buff. We now let ARP fill in the other fields.
+ *
+ * This routine CANNOT use cached dst->neigh!
+ * Really, it is used only when dst->neigh is wrong.
+ */
+
+static int ipv4_rebuild_header(struct sk_buff *skb)
+{
+       struct ipv4_ether_hdr *eth;
+
+       eth = (struct ipv4_ether_hdr *)skb->data;
+       if (eth->h_proto == htons(ETH_P_IP))
+               return arp_find((unsigned char *)&eth->h_dest, skb);
+
+       fw_notify ( "%s: unable to resolve type %04x addresses\n",
+                  skb->dev->name,ntohs(eth->h_proto) );
+       return 0;
+}
+
+static int ipv4_header_cache(const struct neighbour *neigh, struct hh_cache *hh) {
+       unsigned short type = hh->hh_type;
+       struct net_device *dev;
+       struct ipv4_ether_hdr *eth;
+
+       if (type == htons(ETH_P_802_3))
+               return -1;
+       dev = neigh->dev;
+       eth = (struct ipv4_ether_hdr *)((u8 *)hh->hh_data + 16 - sizeof(*eth));
+       eth->h_proto = type;
+       memcpy(eth->h_dest, neigh->ha, dev->addr_len);
+
+       hh->hh_len = IPV4_HLEN;
+       return 0;
+}
+
+/* Called by Address Resolution module to notify changes in address. */
+static void ipv4_header_cache_update(struct hh_cache *hh, const struct net_device *dev, const unsigned char * haddr ) {
+       memcpy((u8 *)hh->hh_data + 16 - IPV4_HLEN, haddr, dev->addr_len);
+}
+
+static int ipv4_header_parse(const struct sk_buff *skb, unsigned char *haddr) {
+       memcpy(haddr, skb->dev->dev_addr, IPV4_ALEN);
+       return IPV4_ALEN;
+}
+
+static const struct header_ops ipv4_header_ops = {
+       .create         = ipv4_header,
+       .rebuild        = ipv4_rebuild_header,
+       .cache          = ipv4_header_cache,
+       .cache_update   = ipv4_header_cache_update,
+       .parse          = ipv4_header_parse,
+};
+
+/* ------------------------------------------------------------------ */
+
+/* FIXME: is this correct for all cases? */
+static bool ipv4_frag_overlap(struct ipv4_partial_datagram *pd, unsigned offset, unsigned len)
+{
+        struct ipv4_fragment_info *fi;
+       unsigned end = offset + len;
+
+       list_for_each_entry(fi, &pd->fragment_info, fragment_info) {
+               if (offset < fi->offset + fi->len && end > fi->offset) {
+                       fw_debug ( "frag_overlap pd %p fi %p (%x@%x) with %x@%x\n", pd, fi, fi->len, fi->offset, len, offset );
+                       return true;
+               }
+       }
+       fw_debug ( "frag_overlap %p does not overlap with %x@%x\n", pd, len, offset );
+       return false;
+}
+
+/* Assumes that new fragment does not overlap any existing fragments */
+static struct ipv4_fragment_info *ipv4_frag_new ( struct ipv4_partial_datagram *pd, unsigned offset, unsigned len ) {
+       struct ipv4_fragment_info *fi, *fi2, *new;
+       struct list_head *list;
+
+       fw_debug ( "frag_new pd %p %x@%x\n", pd, len, offset );
+       list = &pd->fragment_info;
+       list_for_each_entry(fi, &pd->fragment_info, fragment_info) {
+               if (fi->offset + fi->len == offset) {
+                       /* The new fragment can be tacked on to the end */
+                       /* Did the new fragment plug a hole? */
+                       fi2 = list_entry(fi->fragment_info.next, struct ipv4_fragment_info, fragment_info);
+                       if (fi->offset + fi->len == fi2->offset) {
+                               fw_debug ( "pd %p: hole filling %p (%x@%x) and %p(%x@%x): now %x@%x\n", pd, fi, fi->len, fi->offset,
+                               fi2, fi2->len, fi2->offset, fi->len + len + fi2->len, fi->offset );
+                               /* glue fragments together */
+                               fi->len += len + fi2->len;
+                               list_del(&fi2->fragment_info);
+                               kfree(fi2);
+                       } else {
+                               fw_debug ( "pd %p: extending %p from %x@%x to %x@%x\n", pd, fi, fi->len, fi->offset, fi->len+len, fi->offset );
+                               fi->len += len;
+                       }
+                       return fi;
+               }
+               if (offset + len == fi->offset) {
+                       /* The new fragment can be tacked on to the beginning */
+                       /* Did the new fragment plug a hole? */
+                       fi2 = list_entry(fi->fragment_info.prev, struct ipv4_fragment_info, fragment_info);
+                       if (fi2->offset + fi2->len == fi->offset) {
+                               /* glue fragments together */
+                               fw_debug ( "pd %p: extending %p and merging with %p from %x@%x to %x@%x\n",
+ pd, fi2, fi, fi2->len, fi2->offset, fi2->len + fi->len + len, fi2->offset );
+                               fi2->len += fi->len + len;
+                               list_del(&fi->fragment_info);
+                               kfree(fi);
+                               return fi2;
+                       }
+                       fw_debug ( "pd %p: extending %p from %x@%x to %x@%x\n", pd, fi, fi->len, fi->offset, offset, fi->len + len );
+                       fi->offset = offset;
+                       fi->len += len;
+                       return fi;
+               }
+               if (offset > fi->offset + fi->len) {
+                       list = &fi->fragment_info;
+                       break;
+               }
+               if (offset + len < fi->offset) {
+                       list = fi->fragment_info.prev;
+                       break;
+               }
+       }
+
+       new = kmalloc(sizeof(*new), GFP_ATOMIC);
+       if (!new) {
+               fw_error ( "out of memory in fragment handling!\n" );
+               return NULL;
+       }
+
+       new->offset = offset;
+       new->len = len;
+       list_add(&new->fragment_info, list);
+       fw_debug ( "pd %p: new frag %p %x@%x\n", pd, new, new->len, new->offset );
+       list_for_each_entry( fi, &pd->fragment_info, fragment_info )
+               fw_debug ( "fi %p %x@%x\n", fi, fi->len, fi->offset );
+       return new;
+}
+
+/* ------------------------------------------------------------------ */
+
+static struct ipv4_partial_datagram *ipv4_pd_new(struct net_device *netdev,
+ struct ipv4_node *node, u16 datagram_label, unsigned dg_size, u32 *frag_buf,
+ unsigned frag_off, unsigned frag_len) {
+       struct ipv4_partial_datagram *new;
+       struct ipv4_fragment_info *fi;
+
+       new = kmalloc(sizeof(*new), GFP_ATOMIC);
+       if (!new)
+               goto fail;
+       INIT_LIST_HEAD(&new->fragment_info);
+       fi = ipv4_frag_new ( new, frag_off, frag_len);
+       if ( fi == NULL )
+               goto fail_w_new;
+       new->datagram_label = datagram_label;
+       new->datagram_size = dg_size;
+       new->skb = dev_alloc_skb(dg_size + netdev->hard_header_len + 15);
+       if ( new->skb == NULL )
+               goto fail_w_fi;
+       skb_reserve(new->skb, (netdev->hard_header_len + 15) & ~15);
+       new->pbuf = skb_put(new->skb, dg_size);
+       memcpy(new->pbuf + frag_off, frag_buf, frag_len);
+       list_add_tail(&new->pdg_list, &node->pdg_list);
+       fw_debug ( "pd_new: new pd %p { dgl %u, dg_size %u, skb %p, pbuf %p } on node %p\n",
+ new, new->datagram_label, new->datagram_size, new->skb, new->pbuf, node );
+       return new;
+
+fail_w_fi:
+       kfree(fi);
+fail_w_new:
+       kfree(new);
+fail:
+       fw_error("ipv4_pd_new: no memory\n");
+       return NULL;
+}
+
+static struct ipv4_partial_datagram *ipv4_pd_find(struct ipv4_node *node, u16 datagram_label) {
+       struct ipv4_partial_datagram *pd;
+
+       list_for_each_entry(pd, &node->pdg_list, pdg_list) {
+               if ( pd->datagram_label == datagram_label ) {
+                       fw_debug ( "pd_find(node %p, label %u): pd %p\n", node, datagram_label, pd );
+                       return pd;
+               }
+       }
+       fw_debug ( "pd_find(node %p, label %u) no entry\n", node, datagram_label );
+       return NULL;
+}
+
+
+static void ipv4_pd_delete ( struct ipv4_partial_datagram *old ) {
+       struct ipv4_fragment_info *fi, *n;
+
+       fw_debug ( "pd_delete %p\n", old );
+       list_for_each_entry_safe(fi, n, &old->fragment_info, fragment_info) {
+               fw_debug ( "Freeing fi %p\n", fi );
+               kfree(fi);
+       }
+       list_del(&old->pdg_list);
+       dev_kfree_skb_any(old->skb);
+       kfree(old);
+}
+
+static bool ipv4_pd_update ( struct ipv4_node *node, struct ipv4_partial_datagram *pd,
+ u32 *frag_buf, unsigned frag_off, unsigned frag_len) {
+       fw_debug ( "pd_update node %p, pd %p, frag_buf %p, %x@%x\n", node, pd, frag_buf, frag_len, frag_off );
+       if ( ipv4_frag_new ( pd, frag_off, frag_len ) == NULL)
+               return false;
+       memcpy(pd->pbuf + frag_off, frag_buf, frag_len);
+
+       /*
+        * Move list entry to beginnig of list so that oldest partial
+        * datagrams percolate to the end of the list
+        */
+       list_move_tail(&pd->pdg_list, &node->pdg_list);
+       fw_debug ( "New pd list:\n" );
+       list_for_each_entry ( pd, &node->pdg_list, pdg_list ) {
+               fw_debug ( "pd %p\n", pd );
+       }
+       return true;
+}
+
+static bool ipv4_pd_is_complete ( struct ipv4_partial_datagram *pd ) {
+       struct ipv4_fragment_info *fi;
+       bool ret;
+
+       fi = list_entry(pd->fragment_info.next, struct ipv4_fragment_info, fragment_info);
+
+       ret = (fi->len == pd->datagram_size);
+       fw_debug ( "pd_is_complete (pd %p, dgs %x): fi %p (%x@%x) %s\n", pd, pd->datagram_size, fi, fi->len, fi->offset, ret ? "yes" : "no" );
+       return ret;
+}
+
+/* ------------------------------------------------------------------ */
+
+static int ipv4_node_new ( struct fw_card *card, struct fw_device *device ) {
+       struct ipv4_node *node;
+
+       node = kmalloc ( sizeof(*node), GFP_KERNEL );
+       if ( ! node ) {
+               fw_error ( "allocate new node failed\n" );
+               return -ENOMEM;
+       }
+       node->guid = (u64)device->config_rom[3] << 32 | device->config_rom[4];
+       node->fifo = INVALID_FIFO_ADDR;
+       INIT_LIST_HEAD(&node->pdg_list);
+       spin_lock_init(&node->pdg_lock);
+       node->pdg_size = 0;
+       node->generation = device->generation;
+       rmb();
+       node->nodeid = device->node_id;
+        /* FIXME what should it really be? */
+       node->max_payload = S100_BUFFER_SIZE - IPV4_UNFRAG_HDR_SIZE;
+       node->datagram_label = 0U;
+       node->xmt_speed = device->max_speed;
+       list_add_tail ( &node->ipv4_nodes, &card->ipv4_nodes );
+       fw_debug ( "node_new: %p { guid %016llx, generation %u, nodeid %x, max_payload %x, xmt_speed %x } added\n",
+ node, (unsigned long long)node->guid, node->generation, node->nodeid, node->max_payload, node->xmt_speed );
+       return 0;
+}
+
+static struct ipv4_node *ipv4_node_find_by_guid(struct ipv4_priv *priv, u64 guid) {
+       struct ipv4_node *node;
+       unsigned long flags;
+
+       spin_lock_irqsave(&priv->lock, flags);
+       list_for_each_entry(node, &priv->card->ipv4_nodes, ipv4_nodes)
+               if (node->guid == guid) {
+                       /* FIXME: lock the node first? */
+                       spin_unlock_irqrestore ( &priv->lock, flags );
+                       fw_debug ( "node_find_by_guid (%016llx) found %p\n", (unsigned long long)guid, node );
+                       return node;
+               }
+
+       spin_unlock_irqrestore ( &priv->lock, flags );
+       fw_debug ( "node_find_by_guid (%016llx) not found\n", (unsigned long long)guid );
+       return NULL;
+}
+
+static struct ipv4_node *ipv4_node_find_by_nodeid(struct ipv4_priv *priv, u16 nodeid) {
+       struct ipv4_node *node;
+       unsigned long flags;
+
+       spin_lock_irqsave(&priv->lock, flags);
+       list_for_each_entry(node, &priv->card->ipv4_nodes, ipv4_nodes)
+               if (node->nodeid == nodeid) {
+                       /* FIXME: lock the node first? */
+                       spin_unlock_irqrestore ( &priv->lock, flags );
+                       fw_debug ( "node_find_by_nodeid (%x) found %p\n", nodeid, node );
+                       return node;
+               }
+       fw_debug ( "node_find_by_nodeid (%x) not found\n", nodeid );
+       spin_unlock_irqrestore ( &priv->lock, flags );
+       return NULL;
+}
+
+/* This is only complicated because we can't assume priv exists */
+static void ipv4_node_delete ( struct fw_card *card, struct fw_device *device ) {
+       struct net_device *netdev;
+       struct ipv4_priv *priv;
+       struct ipv4_node *node;
+       u64 guid;
+       unsigned long flags;
+       struct ipv4_partial_datagram *pd, *pd_next;
+
+       guid = (u64)device->config_rom[3] << 32 | device->config_rom[4];
+       netdev = card->netdev;
+       if ( netdev )
+               priv = netdev_priv ( netdev );
+       else
+               priv = NULL;
+       if ( priv )
+               spin_lock_irqsave ( &priv->lock, flags );
+       list_for_each_entry( node, &card->ipv4_nodes, ipv4_nodes ) {
+               if ( node->guid == guid ) {
+                       list_del ( &node->ipv4_nodes );
+                       list_for_each_entry_safe( pd, pd_next, &node->pdg_list, pdg_list )
+                               ipv4_pd_delete ( pd );
+                       break;
+               }
+       }
+       if ( priv )
+               spin_unlock_irqrestore ( &priv->lock, flags );
+}
+
+/* ------------------------------------------------------------------ */
+
+
+static int ipv4_finish_incoming_packet ( struct net_device *netdev,
+ struct sk_buff *skb, u16 source_node_id, bool is_broadcast, u16 ether_type ) {
+       struct ipv4_priv *priv;
+       static u64 broadcast_hw = ~0ULL;
+       int status;
+       u64 guid;
+
+       fw_debug ( "ipv4_finish_incoming_packet(%p, %p, %x, %s, %x\n",
+ netdev, skb, source_node_id, is_broadcast ? "true" : "false", ether_type );
+       priv = netdev_priv(netdev);
+       /* Write metadata, and then pass to the receive level */
+       skb->dev = netdev;
+       skb->ip_summed = CHECKSUM_UNNECESSARY;  /* don't check it */
+
+       /*
+        * Parse the encapsulation header. This actually does the job of
+        * converting to an ethernet frame header, as well as arp
+        * conversion if needed. ARP conversion is easier in this
+        * direction, since we are using ethernet as our backend.
+        */
+       /*
+        * If this is an ARP packet, convert it. First, we want to make
+        * use of some of the fields, since they tell us a little bit
+        * about the sending machine.
+        */
+       if (ether_type == ETH_P_ARP) {
+               struct ipv4_arp *arp1394;
+               struct arphdr *arp;
+               unsigned char *arp_ptr;
+               u64 fifo_addr;
+               u8 max_rec;
+               u8 sspd;
+               u16 max_payload;
+               struct ipv4_node *node;
+               static const u16 ipv4_speed_to_max_payload[] = {
+                       /* S100, S200, S400, S800, S1600, S3200 */
+                           512, 1024, 2048, 4096,  4096,  4096
+               };
+
+               /* fw_debug ( "ARP packet\n" ); */
+               arp1394 = (struct ipv4_arp *)skb->data;
+               arp = (struct arphdr *)skb->data;
+               arp_ptr = (unsigned char *)(arp + 1);
+               fifo_addr = (u64)ntohs(arp1394->fifo_hi) << 32 |
+ ntohl(arp1394->fifo_lo);
+               max_rec = priv->card->max_receive;
+               if ( arp1394->max_rec < max_rec )
+                       max_rec = arp1394->max_rec;
+               sspd = arp1394->sspd;
+               /*
+                * Sanity check. MacOSX seems to be sending us 131 in this
+                * field (atleast on my Panther G5). Not sure why.
+                */
+               if (sspd > 5 ) {
+                       fw_notify ( "sspd %x out of range\n", sspd );
+                       sspd = 0;
+               }
+
+               max_payload = min(ipv4_speed_to_max_payload[sspd],
+ (u16)(1 << (max_rec + 1))) - IPV4_UNFRAG_HDR_SIZE;
+
+               guid = be64_to_cpu(get_unaligned(&arp1394->s_uniq_id));
+               node = ipv4_node_find_by_guid(priv, guid);
+               if (!node) {
+                       fw_notify ( "No node for ARP packet from %llx\n", guid );
+                       goto failed_proto;
+               }
+               if ( node->nodeid != source_node_id || node->generation != priv->card->generation ) {
+                       fw_notify ( "Internal error: node->nodeid (%x) != soucre_node_id (%x) or node->generation (%x) != priv->card->generation(%x)\n",
+ node->nodeid, source_node_id, node->generation, priv->card->generation );
+                       node->nodeid = source_node_id;
+                       node->generation = priv->card->generation;
+               }
+
+               /* FIXME: for debugging */
+               if ( sspd > SCODE_400 )
+                       sspd = SCODE_400;
+               /* Update our speed/payload/fifo_offset table */
+               /*
+                * FIXME: this does not handle cases where two high-speed endpoints must use a slower speed because of
+                * a lower speed hub between them.  We need to look at the actual topology map here.
+                */
+               fw_debug ( "Setting node %p fifo %llx (was %llx), max_payload %x (was %x), speed %x (was %x)\n",
+ node, fifo_addr, node->fifo, max_payload, node->max_payload, sspd, node->xmt_speed );
+               node->fifo =    fifo_addr;
+               node->max_payload = max_payload;
+               /*
+                * Only allow speeds to go down from their initial value.
+                * Otherwise a local node that can only do S400 or slower may
+                * be told to transmit at S800 to a faster remote node.
+                */
+               if ( node->xmt_speed > sspd )
+                       node->xmt_speed = sspd;
+
+               /*
+                * Now that we're done with the 1394 specific stuff, we'll
+                * need to alter some of the data.  Believe it or not, all
+                * that needs to be done is sender_IP_address needs to be
+                * moved, the destination hardware address get stuffed
+                * in and the hardware address length set to 8.
+                *
+                * IMPORTANT: The code below overwrites 1394 specific data
+                * needed above so keep the munging of the data for the
+                * higher level IP stack last.
+                */
+
+               arp->ar_hln = 8;
+               arp_ptr += arp->ar_hln;         /* skip over sender unique id */
+               *(u32 *)arp_ptr = arp1394->sip; /* move sender IP addr */
+               arp_ptr += arp->ar_pln;         /* skip over sender IP addr */
+
+               if (arp->ar_op == htons(ARPOP_REQUEST))
+                       memset(arp_ptr, 0, sizeof(u64));
+               else
+                       memcpy(arp_ptr, netdev->dev_addr, sizeof(u64));
+       }
+
+       /* Now add the ethernet header. */
+       guid = cpu_to_be64(priv->card->guid);
+       if (dev_hard_header(skb, netdev, ether_type, is_broadcast ? &broadcast_hw : &guid, NULL,
+ skb->len) >= 0) {
+               struct ipv4_ether_hdr *eth;
+               u16 *rawp;
+               __be16 protocol;
+
+               skb_reset_mac_header(skb);
+               skb_pull(skb, sizeof(*eth));
+               eth = ipv4_ether_hdr(skb);
+               if (*eth->h_dest & 1) {
+                       if (memcmp(eth->h_dest, netdev->broadcast, netdev->addr_len) == 0) {
+                               fw_debug ( "Broadcast\n" );
+                               skb->pkt_type = PACKET_BROADCAST;
+                       }
+#if 0
+                       else
+                               skb->pkt_type = PACKET_MULTICAST;
+#endif
+               } else {
+                       if (memcmp(eth->h_dest, netdev->dev_addr, netdev->addr_len)) {
+                               u64 a1, a2;
+
+                               memcpy ( &a1, eth->h_dest, sizeof(u64));
+                               memcpy ( &a2, netdev->dev_addr, sizeof(u64));
+                               fw_debug ( "Otherhost %llx %llx %x\n", a1, a2, netdev->addr_len );
+                               skb->pkt_type = PACKET_OTHERHOST;
+                       }
+               }
+               if (ntohs(eth->h_proto) >= 1536) {
+                       fw_debug ( " proto %x %x\n", eth->h_proto, ntohs(eth->h_proto) );
+                       protocol = eth->h_proto;
+               } else {
+                       rawp = (u16 *)skb->data;
+                       if (*rawp == 0xFFFF) {
+                               fw_debug ( "proto 802_3\n" );
+                               protocol = htons(ETH_P_802_3);
+                       } else {
+                               fw_debug ( "proto 802_2\n" );
+                               protocol = htons(ETH_P_802_2);
+                       }
+               }
+               skb->protocol = protocol;
+       }
+       status = netif_rx(skb);
+       if ( status == NET_RX_DROP) {
+               netdev->stats.rx_errors++;
+               netdev->stats.rx_dropped++;
+       } else {
+               netdev->stats.rx_packets++;
+               netdev->stats.rx_bytes += skb->len;
+       }
+       if (netif_queue_stopped(netdev))
+               netif_wake_queue(netdev);
+       return 0;
+
+ failed_proto:
+       netdev->stats.rx_errors++;
+       netdev->stats.rx_dropped++;
+       dev_kfree_skb_any(skb);
+       if (netif_queue_stopped(netdev))
+               netif_wake_queue(netdev);
+       netdev->last_rx = jiffies;
+       return 0;
+}
+
+/* ------------------------------------------------------------------ */
+
+static int ipv4_incoming_packet ( struct ipv4_priv *priv, u32 *buf, int len, u16 source_node_id, bool is_broadcast ) {
+       struct sk_buff *skb;
+       struct net_device *netdev;
+       struct ipv4_hdr hdr;
+       unsigned lf;
+       unsigned long flags;
+       struct ipv4_node *node;
+       struct ipv4_partial_datagram *pd;
+       int fg_off;
+       int dg_size;
+       u16 datagram_label;
+       int retval;
+       u16 ether_type;
+
+       fw_debug ( "ipv4_incoming_packet(%p, %p, %d, %x, %s)\n", priv, buf, len, source_node_id, is_broadcast ? "true" : "false" );
+       netdev = priv->card->netdev;
+
+       hdr.w0 = ntohl(buf[0]);
+       lf = ipv4_get_hdr_lf(&hdr);
+       if ( lf == IPV4_HDR_UNFRAG ) {
+               /*
+                * An unfragmented datagram has been received by the ieee1394
+                * bus. Build an skbuff around it so we can pass it to the
+                * high level network layer.
+                */
+               ether_type = ipv4_get_hdr_ether_type(&hdr);
+               fw_debug ( "header w0 = %x, lf = %x, ether_type = %x\n", hdr.w0, lf, ether_type );
+               buf++;
+               len -= IPV4_UNFRAG_HDR_SIZE;
+
+               skb = dev_alloc_skb(len + netdev->hard_header_len + 15);
+               if (unlikely(!skb)) {
+                       fw_error ( "Out of memory for incoming packet\n");
+                       netdev->stats.rx_dropped++;
+                       return -1;
+               }
+               skb_reserve(skb, (netdev->hard_header_len + 15) & ~15);
+               memcpy(skb_put(skb, len), buf, len );
+               return ipv4_finish_incoming_packet(netdev, skb, source_node_id, is_broadcast, ether_type );
+       }
+       /* A datagram fragment has been received, now the fun begins. */
+       hdr.w1 = ntohl(buf[1]);
+       buf +=2;
+       len -= IPV4_FRAG_HDR_SIZE;
+       if ( lf ==IPV4_HDR_FIRSTFRAG ) {
+               ether_type = ipv4_get_hdr_ether_type(&hdr);
+               fg_off = 0;
+       } else {
+               fg_off = ipv4_get_hdr_fg_off(&hdr);
+               ether_type = 0; /* Shut up compiler! */
+       }
+       datagram_label = ipv4_get_hdr_dgl(&hdr);
+       dg_size = ipv4_get_hdr_dg_size(&hdr); /* ??? + 1 */
+       fw_debug ( "fragmented: %x.%x = lf %x, ether_type %x, fg_off %x, dgl %x, dg_size %x\n", hdr.w0, hdr.w1, lf, ether_type, fg_off, datagram_label, dg_size );
+       node = ipv4_node_find_by_nodeid ( priv, source_node_id);
+       spin_lock_irqsave(&node->pdg_lock, flags);
+       pd = ipv4_pd_find( node, datagram_label );
+       if (pd == NULL) {
+               while ( node->pdg_size >= ipv4_mpd ) {
+                       /* remove the oldest */
+                       ipv4_pd_delete ( list_first_entry(&node->pdg_list, struct ipv4_partial_datagram, pdg_list) );
+                       node->pdg_size--;
+               }
+               pd = ipv4_pd_new ( netdev, node, datagram_label, dg_size,
+ buf, fg_off, len);
+               if ( pd == NULL) {
+                       retval = -ENOMEM;
+                       goto bad_proto;
+               }
+               node->pdg_size++;
+       } else {
+               if (ipv4_frag_overlap(pd, fg_off, len) || pd->datagram_size != dg_size) {
+                       /*
+                        * Differing datagram sizes or overlapping fragments,
+                        * Either way the remote machine is playing silly buggers
+                        * with us: obliterate the old datagram and start a new one.
+                        */
+                       ipv4_pd_delete ( pd );
+                       pd = ipv4_pd_new ( netdev, node, datagram_label,
+ dg_size, buf, fg_off, len);
+                       if ( pd == NULL ) {
+                               retval = -ENOMEM;
+                               node->pdg_size--;
+                               goto bad_proto;
+                       }
+               } else {
+                       bool worked;
+
+                       worked = ipv4_pd_update ( node, pd,
+ buf, fg_off, len );
+                       if ( ! worked ) {
+                               /*
+                                * Couldn't save off fragment anyway
+                                * so might as well obliterate the
+                                * datagram now.
+                                */
+                               ipv4_pd_delete ( pd );
+                               node->pdg_size--;
+                               goto bad_proto;
+                       }
+               }
+       } /* new datagram or add to existing one */
+
+       if ( lf == IPV4_HDR_FIRSTFRAG )
+               pd->ether_type = ether_type;
+       if ( ipv4_pd_is_complete ( pd ) ) {
+               ether_type = pd->ether_type;
+               node->pdg_size--;
+               skb = skb_get(pd->skb);
+               ipv4_pd_delete ( pd );
+               spin_unlock_irqrestore(&node->pdg_lock, flags);
+               return ipv4_finish_incoming_packet ( netdev, skb, source_node_id, false, ether_type );
+       }
+       /*
+        * Datagram is not complete, we're done for the
+        * moment.
+        */
+       spin_unlock_irqrestore(&node->pdg_lock, flags);
+       return 0;
+
+ bad_proto:
+       spin_unlock_irqrestore(&node->pdg_lock, flags);
+       if (netif_queue_stopped(netdev))
+               netif_wake_queue(netdev);
+       return 0;
+}
+
+static void ipv4_receive_packet ( struct fw_card *card, struct fw_request *r,
+ int tcode, int destination, int source, int generation, int speed,
+ unsigned long long offset, void *payload, size_t length, void *callback_data ) {
+       struct ipv4_priv *priv;
+       int status;
+
+       fw_debug ( "ipv4_receive_packet(%p,%p,%x,%x,%x,%x,%x,%llx,%p,%lx,%p)\n",
+ card, r, tcode, destination, source, generation, speed, offset, payload,
+ (unsigned long)length, callback_data);
+       print_hex_dump ( KERN_DEBUG, "header: ", DUMP_PREFIX_OFFSET, 32, 1, payload, length, false );
+       priv = callback_data;
+       if (   tcode != TCODE_WRITE_BLOCK_REQUEST
+           || destination != card->node_id
+           || generation != card->generation
+           || offset != priv->handler.offset ) {
+               fw_send_response(card, r, RCODE_CONFLICT_ERROR);
+               fw_debug("Conflict error card node_id=%x, card generation=%x, local offset %llx\n",
+ card->node_id, card->generation, (unsigned long long)priv->handler.offset );
+               return;
+       }
+       status = ipv4_incoming_packet ( priv, payload, length, source, false );
+       if ( status != 0 ) {
+               fw_error ( "Incoming packet failure\n" );
+               fw_send_response ( card, r, RCODE_CONFLICT_ERROR );
+               return;
+       }
+       fw_send_response ( card, r, RCODE_COMPLETE );
+}
+
+static void ipv4_receive_broadcast(struct fw_iso_context *context, u32 cycle,
+ size_t header_length, void *header, void *data) {
+       struct ipv4_priv *priv;
+       struct fw_iso_packet packet;
+       struct fw_card *card;
+       u16 *hdr_ptr;
+       u32 *buf_ptr;
+       int retval;
+       u32 length;
+       u16 source_node_id;
+       u32 specifier_id;
+       u32 ver;
+       unsigned long offset;
+       unsigned long flags;
+
+       fw_debug ( "ipv4_receive_broadcast ( context=%p, cycle=%x, header_length=%lx, header=%p, data=%p )\n", context, cycle, (unsigned long)header_length, header, data );
+       print_hex_dump ( KERN_DEBUG, "header: ", DUMP_PREFIX_OFFSET, 32, 1, header, header_length, false );
+       priv = data;
+       card = priv->card;
+       hdr_ptr = header;
+       length = ntohs(hdr_ptr[0]);
+       spin_lock_irqsave(&priv->lock,flags);
+       offset = priv->rcv_buffer_size * priv->broadcast_rcv_next_ptr;
+       buf_ptr = priv->broadcast_rcv_buffer_ptrs[priv->broadcast_rcv_next_ptr++];
+       if ( priv->broadcast_rcv_next_ptr == priv->num_broadcast_rcv_ptrs )
+               priv->broadcast_rcv_next_ptr = 0;
+       spin_unlock_irqrestore(&priv->lock,flags);
+       fw_debug ( "length %u at %p\n", length, buf_ptr );
+       print_hex_dump ( KERN_DEBUG, "buffer: ", DUMP_PREFIX_OFFSET, 32, 1, buf_ptr, length, false );
+
+       specifier_id =    (be32_to_cpu(buf_ptr[0]) & 0xffff) << 8
+                       | (be32_to_cpu(buf_ptr[1]) & 0xff000000) >> 24;
+       ver = be32_to_cpu(buf_ptr[1]) & 0xFFFFFF;
+       source_node_id = be32_to_cpu(buf_ptr[0]) >> 16;
+       /* fw_debug ( "source %x SpecID %x ver %x\n", source_node_id, specifier_id, ver ); */
+       if ( specifier_id == IPV4_GASP_SPECIFIER_ID && ver == IPV4_GASP_VERSION ) {
+               buf_ptr += 2;
+               length -= IPV4_GASP_OVERHEAD;
+               ipv4_incoming_packet(priv, buf_ptr, length, source_node_id, true);
+       } else
+               fw_debug ( "Ignoring packet: not GASP\n" );
+       packet.payload_length = priv->rcv_buffer_size;
+       packet.interrupt = 1;
+       packet.skip = 0;
+       packet.tag = 3;
+       packet.sy = 0;
+       packet.header_length = IPV4_GASP_OVERHEAD;
+       spin_lock_irqsave(&priv->lock,flags);
+       retval = fw_iso_context_queue ( priv->broadcast_rcv_context, &packet,
+ &priv->broadcast_rcv_buffer, offset );
+       spin_unlock_irqrestore(&priv->lock,flags);
+       if ( retval < 0 )
+               fw_error ( "requeue failed\n" );
+}
+
+static void debug_ptask ( struct ipv4_packet_task *ptask ) {
+       static const char *tx_types[] = { "Unknown", "GASP", "Write" };
+
+       fw_debug ( "packet %p { hdr { w0 %x w1 %x }, skb %p, priv %p,"
+ " tx_type %s, outstanding_pkts %d, max_payload %x, fifo %llx,"
+ " speed %x, dest_node %x, generation %x }\n",
+ ptask, ptask->hdr.w0, ptask->hdr.w1, ptask->skb, ptask->priv,
+ ptask->tx_type > IPV4_WRREQ ? "Invalid" : tx_types[ptask->tx_type],
+ ptask->outstanding_pkts,  ptask->max_payload,
+ ptask->fifo_addr, ptask->speed, ptask->dest_node, ptask->generation );
+       print_hex_dump ( KERN_DEBUG, "packet :", DUMP_PREFIX_OFFSET, 32, 1,
+ ptask->skb->data, ptask->skb->len, false );
+}
+
+static void ipv4_transmit_packet_done ( struct ipv4_packet_task *ptask ) {
+       struct ipv4_priv *priv;
+       unsigned long flags;
+
+       priv = ptask->priv;
+       spin_lock_irqsave ( &priv->lock, flags );
+       list_del ( &ptask->packet_list );
+       spin_unlock_irqrestore ( &priv->lock, flags );
+       ptask->outstanding_pkts--;
+       if ( ptask->outstanding_pkts > 0 ) {
+               u16 dg_size;
+               u16 fg_off;
+               u16 datagram_label;
+               u16 lf;
+               struct sk_buff *skb;
+
+               /* Update the ptask to point to the next fragment and send it */
+               lf = ipv4_get_hdr_lf(&ptask->hdr);
+               switch (lf) {
+               case IPV4_HDR_LASTFRAG:
+               case IPV4_HDR_UNFRAG:
+               default:
+                       fw_error ( "Outstanding packet %x lf %x, header %x,%x\n", ptask->outstanding_pkts, lf, ptask->hdr.w0, ptask->hdr.w1 );
+                       BUG();
+
+               case IPV4_HDR_FIRSTFRAG:
+                       /* Set frag type here for future interior fragments */
+                       dg_size = ipv4_get_hdr_dg_size(&ptask->hdr);
+                       fg_off = ptask->max_payload - IPV4_FRAG_HDR_SIZE;
+                       datagram_label = ipv4_get_hdr_dgl(&ptask->hdr);
+                       break;
+
+               case IPV4_HDR_INTFRAG:
+                       dg_size = ipv4_get_hdr_dg_size(&ptask->hdr);
+                       fg_off = ipv4_get_hdr_fg_off(&ptask->hdr) + ptask->max_payload - IPV4_FRAG_HDR_SIZE;
+                       datagram_label = ipv4_get_hdr_dgl(&ptask->hdr);
+                       break;
+               }
+               skb = ptask->skb;
+               skb_pull ( skb, ptask->max_payload );
+               if ( ptask->outstanding_pkts > 1 ) {
+                       ipv4_make_sf_hdr ( &ptask->hdr,
+  IPV4_HDR_INTFRAG, dg_size, fg_off, datagram_label );
+               } else {
+                       ipv4_make_sf_hdr ( &ptask->hdr,
+  IPV4_HDR_LASTFRAG, dg_size, fg_off, datagram_label );
+                       ptask->max_payload = skb->len + IPV4_FRAG_HDR_SIZE;
+
+               }
+               ipv4_send_packet ( ptask );
+       } else {
+               dev_kfree_skb_any ( ptask->skb );
+               kmem_cache_free( ipv4_packet_task_cache, ptask );
+       }
+}
+
+static void ipv4_write_complete ( struct fw_card *card, int rcode,
+ void *payload, size_t length, void *data ) {
+       struct ipv4_packet_task *ptask;
+
+       ptask = data;
+       fw_debug ( "ipv4_write_complete ( %p, %x, %p, %lx, %p )\n",
+ card, rcode, payload, (unsigned long)length, data );
+       debug_ptask ( ptask );
+
+       if ( rcode == RCODE_COMPLETE ) {
+               ipv4_transmit_packet_done ( ptask );
+       } else {
+               fw_error ( "ipv4_write_complete: failed: %x\n", rcode );
+               /* ??? error recovery */
+       }
+}
+
+static int ipv4_send_packet ( struct ipv4_packet_task *ptask ) {
+       struct ipv4_priv *priv;
+       unsigned tx_len;
+       struct ipv4_hdr *bufhdr;
+       unsigned long flags;
+       struct net_device *netdev;
+#if 0 /* stefanr */
+       int retval;
+#endif
+
+       fw_debug ( "ipv4_send_packet\n" );
+       debug_ptask ( ptask );
+       priv = ptask->priv;
+       tx_len = ptask->max_payload;
+       switch (ipv4_get_hdr_lf(&ptask->hdr)) {
+       case IPV4_HDR_UNFRAG:
+               bufhdr = (struct ipv4_hdr *)skb_push(ptask->skb, IPV4_UNFRAG_HDR_SIZE);
+               bufhdr->w0 = htonl(ptask->hdr.w0);
+               break;
+
+       case IPV4_HDR_FIRSTFRAG:
+       case IPV4_HDR_INTFRAG:
+       case IPV4_HDR_LASTFRAG:
+               bufhdr = (struct ipv4_hdr *)skb_push(ptask->skb, IPV4_FRAG_HDR_SIZE);
+               bufhdr->w0 = htonl(ptask->hdr.w0);
+               bufhdr->w1 = htonl(ptask->hdr.w1);
+               break;
+
+       default:
+               BUG();
+       }
+       if ( ptask->tx_type == IPV4_GASP ) {
+               u32 *packets;
+               int generation;
+               int nodeid;
+
+               /* ptask->generation may not have been set yet */
+               generation = priv->card->generation;
+               smp_rmb();
+               nodeid = priv->card->node_id;
+               packets = (u32 *)skb_push(ptask->skb, sizeof(u32)*2);
+               packets[0] = htonl(nodeid << 16 | (IPV4_GASP_SPECIFIER_ID>>8));
+               packets[1] = htonl((IPV4_GASP_SPECIFIER_ID & 0xFF) << 24 | IPV4_GASP_VERSION);
+               fw_send_request ( priv->card, &ptask->transaction, TCODE_STREAM_DATA,
+ fw_stream_packet_destination_id(3, BROADCAST_CHANNEL, 0),
+ generation, SCODE_100, 0ULL, ptask->skb->data, tx_len + 8, ipv4_write_complete, ptask );
+               spin_lock_irqsave(&priv->lock,flags);
+               list_add_tail ( &ptask->packet_list, &priv->broadcasted_list );
+               spin_unlock_irqrestore(&priv->lock,flags);
+#if 0 /* stefanr */
+               return retval;
+#else
+               return 0;
+#endif
+       }
+       fw_debug("send_request (%p, %p, WRITE_BLOCK, %x, %x, %x, %llx, %p, %d, %p, %p\n",
+ priv->card, &ptask->transaction, ptask->dest_node, ptask->generation,
+ ptask->speed, (unsigned long long)ptask->fifo_addr, ptask->skb->data, tx_len,
+ ipv4_write_complete, ptask );
+       fw_send_request ( priv->card, &ptask->transaction,
+ TCODE_WRITE_BLOCK_REQUEST, ptask->dest_node, ptask->generation, ptask->speed,
+ ptask->fifo_addr, ptask->skb->data, tx_len, ipv4_write_complete, ptask );
+       spin_lock_irqsave(&priv->lock,flags);
+       list_add_tail ( &ptask->packet_list, &priv->sent_list );
+       spin_unlock_irqrestore(&priv->lock,flags);
+       netdev = priv->card->netdev;
+       netdev->trans_start = jiffies;
+       return 0;
+}
+
+static int ipv4_broadcast_start ( struct ipv4_priv *priv ) {
+       struct fw_iso_context *context;
+       int retval;
+       unsigned num_packets;
+       unsigned max_receive;
+       struct fw_iso_packet packet;
+       unsigned long offset;
+       unsigned u;
+       /* unsigned transmit_speed; */
+
+#if 0 /* stefanr */
+       if ( priv->card->broadcast_channel != (BROADCAST_CHANNEL_VALID|BROADCAST_CHANNEL_INITIAL)) {
+               fw_notify ( "Invalid broadcast channel %x\n", priv->card->broadcast_channel );
+               /* FIXME: try again later? */
+               /* return -EINVAL; */
+       }
+#endif
+       if ( priv->local_fifo == INVALID_FIFO_ADDR ) {
+               struct fw_address_region region;
+
+               priv->handler.length = FIFO_SIZE;
+               priv->handler.address_callback = ipv4_receive_packet;
+               priv->handler.callback_data = priv;
+               /* FIXME: this is OHCI, but what about others? */
+               region.start = 0xffff00000000ULL;
+               region.end =   0xfffffffffffcULL;
+
+               retval = fw_core_add_address_handler ( &priv->handler, &region );
+               if ( retval < 0 )
+                       goto failed_initial;
+               priv->local_fifo = priv->handler.offset;
+       }
+
+       /*
+        * FIXME: rawiso limits us to PAGE_SIZE.  This only matters if we ever have
+        * a machine with PAGE_SIZE < 4096
+        */
+       max_receive = 1U << (priv->card->max_receive + 1);
+       num_packets = ( ipv4_iso_page_count * PAGE_SIZE ) / max_receive;
+       if ( ! priv->broadcast_rcv_context ) {
+               void **ptrptr;
+
+               context = fw_iso_context_create ( priv->card,
+ FW_ISO_CONTEXT_RECEIVE, BROADCAST_CHANNEL,
+ priv->card->link_speed, 8, ipv4_receive_broadcast, priv );
+               if (IS_ERR(context)) {
+                       retval = PTR_ERR(context);
+                       goto failed_context_create;
+               }
+               retval = fw_iso_buffer_init ( &priv->broadcast_rcv_buffer,
+ priv->card, ipv4_iso_page_count, DMA_FROM_DEVICE );
+               if ( retval < 0 )
+                       goto failed_buffer_init;
+               ptrptr = kmalloc ( sizeof(void*)*num_packets, GFP_KERNEL );
+               if ( ! ptrptr ) {
+                       retval = -ENOMEM;
+                       goto failed_ptrs_alloc;
+               }
+               priv->broadcast_rcv_buffer_ptrs = ptrptr;
+               for ( u = 0; u < ipv4_iso_page_count; u++ ) {
+                       void *ptr;
+                       unsigned v;
+
+                       ptr = kmap ( priv->broadcast_rcv_buffer.pages[u] );
+                       for ( v = 0; v < num_packets / ipv4_iso_page_count; v++ )
+                               *ptrptr++ = (void *)((char *)ptr + v * max_receive);
+               }
+               priv->broadcast_rcv_context = context;
+       } else
+               context = priv->broadcast_rcv_context;
+
+       packet.payload_length = max_receive;
+       packet.interrupt = 1;
+       packet.skip = 0;
+       packet.tag = 3;
+       packet.sy = 0;
+       packet.header_length = IPV4_GASP_OVERHEAD;
+       offset = 0;
+       for ( u = 0; u < num_packets; u++ ) {
+               retval = fw_iso_context_queue ( context, &packet,
+ &priv->broadcast_rcv_buffer, offset );
+               if ( retval < 0 )
+                       goto failed_rcv_queue;
+               offset += max_receive;
+       }
+       priv->num_broadcast_rcv_ptrs = num_packets;
+       priv->rcv_buffer_size = max_receive;
+       priv->broadcast_rcv_next_ptr = 0U;
+       retval = fw_iso_context_start ( context, -1, 0, FW_ISO_CONTEXT_MATCH_ALL_TAGS ); /* ??? sync */
+       if ( retval < 0 )
+               goto failed_rcv_queue;
+       /* FIXME: adjust this when we know the max receive speeds of all other IP nodes on the bus. */
+       /* since we only xmt at S100 ??? */
+       priv->broadcast_xmt_max_payload = S100_BUFFER_SIZE - IPV4_GASP_OVERHEAD - IPV4_UNFRAG_HDR_SIZE;
+       priv->broadcast_state = IPV4_BROADCAST_RUNNING;
+       return 0;
+
+ failed_rcv_queue:
+       kfree ( priv->broadcast_rcv_buffer_ptrs );
+       priv->broadcast_rcv_buffer_ptrs = NULL;
+ failed_ptrs_alloc:
+       fw_iso_buffer_destroy ( &priv->broadcast_rcv_buffer, priv->card );
+ failed_buffer_init:
+       fw_iso_context_destroy ( context );
+       priv->broadcast_rcv_context = NULL;
+ failed_context_create:
+       fw_core_remove_address_handler ( &priv->handler );
+ failed_initial:
+       priv->local_fifo = INVALID_FIFO_ADDR;
+       return retval;
+}
+
+/* This is called after an "ifup" */
+static int ipv4_open(struct net_device *dev) {
+       struct ipv4_priv *priv;
+       int ret;
+
+       priv = netdev_priv(dev);
+       if (priv->broadcast_state == IPV4_BROADCAST_ERROR) {
+               ret = ipv4_broadcast_start ( priv );
+               if (ret)
+                       return ret;
+       }
+       netif_start_queue(dev);
+       return 0;
+}
+
+/* This is called after an "ifdown" */
+static int ipv4_stop(struct net_device *netdev)
+{
+       /* flush priv->wake */
+       /* flush_scheduled_work(); */
+
+       netif_stop_queue(netdev);
+       return 0;
+}
+
+/* Transmit a packet (called by kernel) */
+static int ipv4_tx(struct sk_buff *skb, struct net_device *netdev)
+{
+       struct ipv4_ether_hdr hdr_buf;
+       struct ipv4_priv *priv = netdev_priv(netdev);
+       __be16 proto;
+       u16 dest_node;
+       enum ipv4_tx_type tx_type;
+       unsigned max_payload;
+       u16 dg_size;
+       u16 *datagram_label_ptr;
+       struct ipv4_packet_task *ptask;
+       struct ipv4_node *node = NULL;
+
+       ptask = kmem_cache_alloc(ipv4_packet_task_cache, GFP_ATOMIC);
+       if (ptask == NULL)
+               goto fail;
+
+       skb = skb_share_check(skb, GFP_ATOMIC);
+       if (!skb)
+               goto fail;
+
+       /*
+        * Get rid of the fake ipv4 header, but first make a copy.
+        * We might need to rebuild the header on tx failure.
+        */
+       memcpy(&hdr_buf, skb->data, sizeof(hdr_buf));
+       skb_pull(skb, sizeof(hdr_buf));
+
+       proto = hdr_buf.h_proto;
+       dg_size = skb->len;
+
+       /*
+        * Set the transmission type for the packet.  ARP packets and IP
+        * broadcast packets are sent via GASP.
+        */
+       if (   memcmp(hdr_buf.h_dest, netdev->broadcast, IPV4_ALEN) == 0
+           || proto == htons(ETH_P_ARP)
+           || (   proto == htons(ETH_P_IP)
+               && IN_MULTICAST(ntohl(ip_hdr(skb)->daddr)) ) ) {
+               /* fw_debug ( "transmitting arp or multicast packet\n" );*/
+               tx_type = IPV4_GASP;
+               dest_node = ALL_NODES;
+               max_payload = priv->broadcast_xmt_max_payload;
+               /* BUG_ON(max_payload < S100_BUFFER_SIZE - IPV4_GASP_OVERHEAD); */
+               datagram_label_ptr = &priv->broadcast_xmt_datagramlabel;
+               ptask->fifo_addr = INVALID_FIFO_ADDR;
+               ptask->generation = 0U;
+               ptask->dest_node = 0U;
+               ptask->speed = 0;
+       } else {
+               __be64 guid = get_unaligned((u64 *)hdr_buf.h_dest);
+               u8 generation;
+
+               node = ipv4_node_find_by_guid(priv, be64_to_cpu(guid));
+               if (!node) {
+                       fw_debug ( "Normal packet but no node\n" );
+                       goto fail;
+               }
+
+               if (node->fifo == INVALID_FIFO_ADDR) {
+                       fw_debug ( "Normal packet but no fifo addr\n" );
+                       goto fail;
+               }
+
+               /* fw_debug ( "Transmitting normal packet to %x at %llxx\n", node->nodeid, node->fifo ); */
+               generation = node->generation;
+               dest_node = node->nodeid;
+               max_payload = node->max_payload;
+               /* BUG_ON(max_payload < S100_BUFFER_SIZE - IPV4_FRAG_HDR_SIZE); */
+
+               datagram_label_ptr = &node->datagram_label;
+               tx_type = IPV4_WRREQ;
+               ptask->fifo_addr = node->fifo;
+               ptask->generation = generation;
+               ptask->dest_node = dest_node;
+               ptask->speed = node->xmt_speed;
+       }
+
+       /* If this is an ARP packet, convert it */
+       if (proto == htons(ETH_P_ARP)) {
+               /* Convert a standard ARP packet to 1394 ARP. The first 8 bytes (the entire
+                * arphdr) is the same format as the ip1394 header, so they overlap.  The rest
+                * needs to be munged a bit.  The remainder of the arphdr is formatted based
+                * on hwaddr len and ipaddr len.  We know what they'll be, so it's easy to
+                * judge.
+                *
+                * Now that the EUI is used for the hardware address all we need to do to make
+                * this work for 1394 is to insert 2 quadlets that contain max_rec size,
+                * speed, and unicast FIFO address information between the sender_unique_id
+                * and the IP addresses.
+                */
+               struct arphdr *arp = (struct arphdr *)skb->data;
+               unsigned char *arp_ptr = (unsigned char *)(arp + 1);
+               struct ipv4_arp *arp1394 = (struct ipv4_arp *)skb->data;
+               u32 ipaddr;
+
+               ipaddr = *(u32*)(arp_ptr + IPV4_ALEN);
+               arp1394->hw_addr_len    = 16;
+               arp1394->max_rec        = priv->card->max_receive;
+               arp1394->sspd           = priv->card->link_speed;
+               arp1394->fifo_hi        = htons(priv->local_fifo >> 32);
+               arp1394->fifo_lo        = htonl(priv->local_fifo & 0xFFFFFFFF);
+               arp1394->sip            = ipaddr;
+       }
+       if ( ipv4_max_xmt && max_payload > ipv4_max_xmt )
+               max_payload = ipv4_max_xmt;
+
+       ptask->hdr.w0 = 0;
+       ptask->hdr.w1 = 0;
+       ptask->skb = skb;
+       ptask->priv = priv;
+        ptask->tx_type = tx_type;
+       /* Does it all fit in one packet? */
+       if ( dg_size <= max_payload ) {
+               ipv4_make_uf_hdr(&ptask->hdr, be16_to_cpu(proto));
+               ptask->outstanding_pkts = 1;
+               max_payload = dg_size + IPV4_UNFRAG_HDR_SIZE;
+       } else {
+               u16 datagram_label;
+
+               max_payload -= IPV4_FRAG_OVERHEAD;
+               datagram_label = (*datagram_label_ptr)++;
+               ipv4_make_ff_hdr(&ptask->hdr, be16_to_cpu(proto), dg_size, datagram_label );
+               ptask->outstanding_pkts = DIV_ROUND_UP(dg_size, max_payload);
+               max_payload += IPV4_FRAG_HDR_SIZE;
+       }
+       ptask->max_payload = max_payload;
+       ipv4_send_packet ( ptask );
+       return NETDEV_TX_OK;
+
+ fail:
+       if (ptask)
+               kmem_cache_free(ipv4_packet_task_cache, ptask);
+
+       if (skb != NULL)
+               dev_kfree_skb(skb);
+
+       netdev->stats.tx_dropped++;
+       netdev->stats.tx_errors++;
+
+       /*
+        * FIXME: According to a patch from 2003-02-26, "returning non-zero
+        * causes serious problems" here, allegedly.  Before that patch,
+        * -ERRNO was returned which is not appropriate under Linux 2.6.
+        * Perhaps more needs to be done?  Stop the queue in serious
+        * conditions and restart it elsewhere?
+        */
+       return NETDEV_TX_OK;
+}
+
+/*
+ * FIXME: What to do if we timeout? I think a host reset is probably in order,
+ * so that's what we do. Should we increment the stat counters too?
+ */
+static void ipv4_tx_timeout(struct net_device *dev) {
+       struct ipv4_priv *priv;
+
+       priv = netdev_priv(dev);
+       fw_error ( "%s: Timeout, resetting host\n", dev->name );
+#if 0 /* stefanr */
+       fw_core_initiate_bus_reset ( priv->card, 1 );
+#endif
+}
+
+static int ipv4_change_mtu ( struct net_device *dev, int new_mtu ) {
+#if 0
+       int max_mtu;
+       struct ipv4_priv *priv;
+#endif
+
+       if (new_mtu < 68)
+               return -EINVAL;
+
+#if 0
+       priv = netdev_priv(dev);
+       /* This is not actually true because we can fragment packets at the firewire layer */
+       max_mtu = (1 << (priv->card->max_receive + 1))
+                               - sizeof(struct ipv4_hdr) - IPV4_GASP_OVERHEAD;
+       if (new_mtu > max_mtu) {
+               fw_notify ( "%s: Local node constrains MTU to %d\n", dev->name, max_mtu);
+               return -ERANGE;
+       }
+#endif
+       dev->mtu = new_mtu;
+       return 0;
+}
+
+static void ipv4_get_drvinfo(struct net_device *dev,
+struct ethtool_drvinfo *info) {
+       strcpy(info->driver, ipv4_driver_name);
+       strcpy(info->bus_info, "ieee1394"); /* FIXME provide more detail? */
+}
+
+static struct ethtool_ops ipv4_ethtool_ops = {
+       .get_drvinfo = ipv4_get_drvinfo,
+};
+
+static const struct net_device_ops ipv4_netdev_ops = {
+       .ndo_open       = ipv4_open,
+       .ndo_stop       = ipv4_stop,
+       .ndo_start_xmit = ipv4_tx,
+       .ndo_tx_timeout = ipv4_tx_timeout,
+       .ndo_change_mtu = ipv4_change_mtu,
+};
+
+static void ipv4_init_dev ( struct net_device *dev ) {
+       dev->header_ops         = &ipv4_header_ops;
+       dev->netdev_ops         = &ipv4_netdev_ops;
+       SET_ETHTOOL_OPS(dev, &ipv4_ethtool_ops);
+
+       dev->watchdog_timeo     = IPV4_TIMEOUT;
+       dev->flags              = IFF_BROADCAST | IFF_MULTICAST;
+       dev->features           = NETIF_F_HIGHDMA;
+       dev->addr_len           = IPV4_ALEN;
+       dev->hard_header_len    = IPV4_HLEN;
+       dev->type               = ARPHRD_IEEE1394;
+
+       /* FIXME: This value was copied from ether_setup(). Is it too much? */
+       dev->tx_queue_len       = 1000;
+}
+
+static int ipv4_probe ( struct device *dev ) {
+       struct fw_unit * unit;
+       struct fw_device *device;
+       struct fw_card *card;
+       struct net_device *netdev;
+       struct ipv4_priv *priv;
+       unsigned max_mtu;
+       __be64 guid;
+
+       fw_debug("ipv4 Probing\n" );
+       unit = fw_unit ( dev );
+       device = fw_device ( unit->device.parent );
+       card = device->card;
+
+       if ( ! device->is_local ) {
+               int added;
+
+               fw_debug ( "Non-local, adding remote node entry\n" );
+               added = ipv4_node_new ( card, device );
+               return added;
+       }
+       fw_debug("ipv4 Local: adding netdev\n" );
+       netdev = alloc_netdev ( sizeof(*priv), "fw-ipv4-%d", ipv4_init_dev );
+       if ( netdev == NULL) {
+               fw_error( "Out of memory\n");
+               goto out;
+       }
+
+       SET_NETDEV_DEV(netdev, card->device);
+       priv = netdev_priv(netdev);
+
+       spin_lock_init(&priv->lock);
+       priv->broadcast_state = IPV4_BROADCAST_ERROR;
+       priv->broadcast_rcv_context = NULL;
+       priv->broadcast_xmt_max_payload = 0;
+       priv->broadcast_xmt_datagramlabel = 0;
+
+       priv->local_fifo = INVALID_FIFO_ADDR;
+
+       /* INIT_WORK(&priv->wake, ipv4_handle_queue);*/
+       INIT_LIST_HEAD(&priv->packet_list);
+       INIT_LIST_HEAD(&priv->broadcasted_list);
+       INIT_LIST_HEAD(&priv->sent_list );
+
+       priv->card = card;
+
+       /*
+        * Use the RFC 2734 default 1500 octets or the maximum payload
+        * as initial MTU
+        */
+       max_mtu = (1 << (card->max_receive + 1))
+                 - sizeof(struct ipv4_hdr) - IPV4_GASP_OVERHEAD;
+       netdev->mtu = min(1500U, max_mtu);
+
+       /* Set our hardware address while we're at it */
+       guid = cpu_to_be64(card->guid);
+       memcpy(netdev->dev_addr, &guid, sizeof(u64));
+       memset(netdev->broadcast, 0xff, sizeof(u64));
+       if ( register_netdev ( netdev ) ) {
+               fw_error ( "Cannot register the driver\n");
+               goto out;
+       }
+
+       fw_notify ( "%s: IPv4 over Firewire on device %016llx\n",
+ netdev->name, card->guid );
+       card->netdev = netdev;
+
+       return 0 /* ipv4_new_node ( ud ) */;
+ out:
+       if ( netdev )
+               free_netdev ( netdev );
+       return -ENOENT;
+}
+
+
+static int ipv4_remove ( struct device *dev ) {
+       struct fw_unit * unit;
+       struct fw_device *device;
+       struct fw_card *card;
+       struct net_device *netdev;
+       struct ipv4_priv *priv;
+       struct ipv4_node *node;
+       struct ipv4_partial_datagram *pd, *pd_next;
+       struct ipv4_packet_task *ptask, *pt_next;
+
+       fw_debug("ipv4 Removing\n" );
+       unit = fw_unit ( dev );
+       device = fw_device ( unit->device.parent );
+       card = device->card;
+
+       if ( ! device->is_local ) {
+               fw_debug ( "Node %x is non-local, removing remote node entry\n", device->node_id );
+               ipv4_node_delete ( card, device );
+               return 0;
+       }
+       netdev = card->netdev;
+       if ( netdev ) {
+               fw_debug ( "Node %x is local: deleting netdev\n", device->node_id );
+               priv = netdev_priv ( netdev );
+               unregister_netdev ( netdev );
+               fw_debug ( "unregistered\n" );
+               if ( priv->local_fifo != INVALID_FIFO_ADDR )
+                       fw_core_remove_address_handler ( &priv->handler );
+               fw_debug ( "address handler gone\n" );
+               if ( priv->broadcast_rcv_context ) {
+                       fw_iso_context_stop ( priv->broadcast_rcv_context );
+                       fw_iso_buffer_destroy ( &priv->broadcast_rcv_buffer, priv->card );
+                       fw_iso_context_destroy ( priv->broadcast_rcv_context );
+                       fw_debug ( "rcv stopped\n" );
+               }
+               list_for_each_entry_safe( ptask, pt_next, &priv->packet_list, packet_list ) {
+                       dev_kfree_skb_any ( ptask->skb );
+                       kmem_cache_free( ipv4_packet_task_cache, ptask );
+               }
+               list_for_each_entry_safe( ptask, pt_next, &priv->broadcasted_list, packet_list ) {
+                       dev_kfree_skb_any ( ptask->skb );
+                       kmem_cache_free( ipv4_packet_task_cache, ptask );
+               }
+               list_for_each_entry_safe( ptask, pt_next, &priv->sent_list, packet_list ) {
+                       dev_kfree_skb_any ( ptask->skb );
+                       kmem_cache_free( ipv4_packet_task_cache, ptask );
+               }
+               fw_debug ( "lists emptied\n" );
+               list_for_each_entry( node, &card->ipv4_nodes, ipv4_nodes ) {
+                       if ( node->pdg_size ) {
+                               list_for_each_entry_safe( pd, pd_next, &node->pdg_list, pdg_list )
+                                       ipv4_pd_delete ( pd );
+                               node->pdg_size = 0;
+                       }
+                       node->fifo = INVALID_FIFO_ADDR;
+               }
+               fw_debug ( "nodes cleaned up\n" );
+               free_netdev ( netdev );
+               card->netdev = NULL;
+               fw_debug ( "done\n" );
+       }
+       return 0;
+}
+
+static void ipv4_update ( struct fw_unit *unit ) {
+       struct fw_device *device;
+       struct fw_card *card;
+
+       fw_debug ( "ipv4_update unit %p\n", unit );
+       device = fw_device ( unit->device.parent );
+       card = device->card;
+       if ( ! device->is_local ) {
+               struct ipv4_node *node;
+               u64 guid;
+               struct net_device *netdev;
+               struct ipv4_priv *priv;
+
+               netdev = card->netdev;
+               if ( netdev ) {
+                       priv = netdev_priv ( netdev );
+                       guid = (u64)device->config_rom[3] << 32 | device->config_rom[4];
+                       node = ipv4_node_find_by_guid ( priv, guid );
+                       if ( ! node ) {
+                               fw_error ( "ipv4_update: no node for device %llx\n", guid );
+                               return;
+                       }
+                       fw_debug ( "Non-local, updating remote node entry for guid %llx old generation %x, old nodeid %x\n", guid, node->generation, node->nodeid );
+                       node->generation = device->generation;
+                       rmb();
+                       node->nodeid = device->node_id;
+                       fw_debug ( "New generation %x, new nodeid %x\n", node->generation, node->nodeid );
+               } else
+                       fw_error ( "nonlocal, but no netdev?  How can that be?\n" );
+       } else {
+               /* FIXME: What do we need to do on bus reset? */
+               fw_debug ( "Local, doing nothing\n" );
+       }
+}
+
+static struct fw_driver ipv4_driver = {
+       .driver = {
+               .owner = THIS_MODULE,
+               .name = ipv4_driver_name,
+               .bus = &fw_bus_type,
+               .probe = ipv4_probe,
+               .remove = ipv4_remove,
+       },
+       .update = ipv4_update,
+       .id_table = ipv4_id_table,
+};
+
+static int __init ipv4_init ( void ) {
+       int added;
+
+       added = fw_core_add_descriptor ( &ipv4_unit_directory );
+       if ( added < 0 )
+               fw_error ( "Failed to add descriptor" );
+       ipv4_packet_task_cache = kmem_cache_create("packet_task",
+ sizeof(struct ipv4_packet_task), 0, 0, NULL);
+       fw_debug("Adding ipv4 module\n" );
+       return driver_register ( &ipv4_driver.driver );
+}
+
+static void __exit ipv4_cleanup ( void ) {
+       fw_core_remove_descriptor ( &ipv4_unit_directory );
+       fw_debug("Removing ipv4 module\n" );
+       driver_unregister ( &ipv4_driver.driver );
+}
+
+module_init(ipv4_init);
+module_exit(ipv4_cleanup);
index e584b72..d44f47d 100644 (file)
@@ -3,6 +3,7 @@
 
 #include <linux/completion.h>
 #include <linux/device.h>
+#include <linux/dma-mapping.h>
 #include <linux/kernel.h>
 #include <linux/kref.h>
 #include <linux/list.h>
@@ -130,6 +131,13 @@ struct fw_card {
        bool broadcast_channel_allocated;
        u32 broadcast_channel;
        u32 topology_map[(CSR_TOPOLOGY_MAP_END - CSR_TOPOLOGY_MAP) / 4];
+       /* Only non-NULL if firewire-ipv4 is active on this card. */
+       void *netdev;
+       /*
+        * The nodes get probed before the card, so we need a place to store
+        * them independent of card->netdev
+        */
+       struct list_head ipv4_nodes;
 };
 
 static inline struct fw_card *fw_card_get(struct fw_card *card)
@@ -355,4 +363,90 @@ int fw_run_transaction(struct fw_card *card, int tcode, int destination_id,
                       int generation, int speed, unsigned long long offset,
                       void *payload, size_t length);
 
+static inline int fw_stream_packet_destination_id(int tag, int channel, int sy)
+{
+       return tag << 14 | channel << 8 | sy;
+}
+
+struct fw_descriptor {
+       struct list_head link;
+       size_t length;
+       u32 immediate;
+       u32 key;
+       const u32 *data;
+};
+
+int fw_core_add_descriptor(struct fw_descriptor *desc);
+void fw_core_remove_descriptor(struct fw_descriptor *desc);
+
+/*
+ * The iso packet format allows for an immediate header/payload part
+ * stored in 'header' immediately after the packet info plus an
+ * indirect payload part that is pointer to by the 'payload' field.
+ * Applications can use one or the other or both to implement simple
+ * low-bandwidth streaming (e.g. audio) or more advanced
+ * scatter-gather streaming (e.g. assembling video frame automatically).
+ */
+struct fw_iso_packet {
+       u16 payload_length;     /* Length of indirect payload. */
+       u32 interrupt:1;        /* Generate interrupt on this packet */
+       u32 skip:1;             /* Set to not send packet at all. */
+       u32 tag:2;
+       u32 sy:4;
+       u32 header_length:8;    /* Length of immediate header. */
+       u32 header[0];
+};
+
+#define FW_ISO_CONTEXT_TRANSMIT        0
+#define FW_ISO_CONTEXT_RECEIVE 1
+
+#define FW_ISO_CONTEXT_MATCH_TAG0       1
+#define FW_ISO_CONTEXT_MATCH_TAG1       2
+#define FW_ISO_CONTEXT_MATCH_TAG2       4
+#define FW_ISO_CONTEXT_MATCH_TAG3       8
+#define FW_ISO_CONTEXT_MATCH_ALL_TAGS  15
+
+/*
+ * An iso buffer is just a set of pages mapped for DMA in the
+ * specified direction.  Since the pages are to be used for DMA, they
+ * are not mapped into the kernel virtual address space.  We store the
+ * DMA address in the page private. The helper function
+ * fw_iso_buffer_map() will map the pages into a given vma.
+ */
+struct fw_iso_buffer {
+       enum dma_data_direction direction;
+       struct page **pages;
+       int page_count;
+};
+
+int fw_iso_buffer_init(struct fw_iso_buffer *buffer, struct fw_card *card,
+                      int page_count, enum dma_data_direction direction);
+void fw_iso_buffer_destroy(struct fw_iso_buffer *buffer, struct fw_card *card);
+
+struct fw_iso_context;
+typedef void (*fw_iso_callback_t)(struct fw_iso_context *context,
+                                 u32 cycle, size_t header_length,
+                                 void *header, void *data);
+struct fw_iso_context {
+       struct fw_card *card;
+       int type;
+       int channel;
+       int speed;
+       size_t header_size;
+       fw_iso_callback_t callback;
+       void *callback_data;
+};
+
+struct fw_iso_context *fw_iso_context_create(struct fw_card *card,
+               int type, int channel, int speed, size_t header_size,
+               fw_iso_callback_t callback, void *callback_data);
+int fw_iso_context_queue(struct fw_iso_context *ctx,
+                        struct fw_iso_packet *packet,
+                        struct fw_iso_buffer *buffer,
+                        unsigned long payload);
+int fw_iso_context_start(struct fw_iso_context *ctx,
+                        int cycle, int sync, int tags);
+int fw_iso_context_stop(struct fw_iso_context *ctx);
+void fw_iso_context_destroy(struct fw_iso_context *ctx);
+
 #endif /* _LINUX_FIREWIRE_H */