firewire: Add SBP-2 protocol driver for storage devices.
authorKristian Høgsberg <krh@redhat.com>
Wed, 20 Dec 2006 00:58:40 +0000 (19:58 -0500)
committerStefan Richter <stefanr@s5r6.in-berlin.de>
Fri, 9 Mar 2007 21:02:34 +0000 (22:02 +0100)
Signed-off-by: Kristian Høgsberg <krh@redhat.com>
Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
drivers/firewire/Kconfig
drivers/firewire/Makefile
drivers/firewire/fw-sbp2.c [new file with mode: 0644]

index b386334..bfab4b3 100644 (file)
@@ -31,4 +31,16 @@ config FW_OHCI
          To compile this driver as a module, say M here: the
          module will be called fw-ohci.
 
+config FW_SBP2
+       tristate "Support for storage devices (SBP-2 protocol driver)"
+       depends on FW && SCSI
+       help
+         This option enables you to use SBP-2 devices connected to an
+         firewire bus.  SBP-2 devices include storage devices like
+         harddisks and DVD drives, also some other FireWire devices
+         like scanners.
+
+         You should also enable support for disks, CD-ROMs, etc. in the SCSI
+         configuration section.
+
 endmenu
index add3b98..b955c99 100644 (file)
@@ -7,3 +7,4 @@ fw-core-objs := fw-card.o fw-topology.o fw-transaction.o fw-iso.o \
 
 obj-$(CONFIG_FW) += fw-core.o
 obj-$(CONFIG_FW_OHCI) += fw-ohci.o
+obj-$(CONFIG_FW_SBP2) += fw-sbp2.o
\ No newline at end of file
diff --git a/drivers/firewire/fw-sbp2.c b/drivers/firewire/fw-sbp2.c
new file mode 100644 (file)
index 0000000..2756e0c
--- /dev/null
@@ -0,0 +1,1073 @@
+/*                                             -*- c-basic-offset: 8 -*-
+ * fw-sbp2.c -- SBP2 driver (SCSI over IEEE1394)
+ *
+ * Copyright (C) 2005-2006  Kristian Hoegsberg <krh@bitplanet.net>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_dbg.h>
+#include <scsi/scsi_device.h>
+#include <scsi/scsi_host.h>
+
+#include "fw-transaction.h"
+#include "fw-topology.h"
+#include "fw-device.h"
+
+/* I don't know why the SCSI stack doesn't define something like this... */
+typedef void (*scsi_done_fn_t) (struct scsi_cmnd *);
+
+static const char sbp2_driver_name[] = "sbp2";
+
+struct sbp2_device {
+       struct fw_unit *unit;
+       struct fw_address_handler address_handler;
+       struct list_head orb_list;
+       u64 management_agent_address;
+       u64 command_block_agent_address;
+       u32 workarounds;
+       int login_id;
+
+       /* We cache these addresses and only update them once we've
+        * logged in or reconnected to the sbp2 device.  That way, any
+        * IO to the device will automatically fail and get retried if
+        * it happens in a window where the device is not ready to
+        * handle it (e.g. after a bus reset but before we reconnect). */
+       int node_id;
+       int address_high;
+       int generation;
+
+       struct work_struct work;
+       struct Scsi_Host *scsi_host;
+};
+
+#define SBP2_MAX_SG_ELEMENT_LENGTH     0xf000
+#define SBP2_MAX_SECTORS               255     /* Max sectors supported */
+#define SBP2_MAX_CMDS                  8       /* This should be safe */
+
+#define SBP2_ORB_NULL                  0x80000000
+
+#define SBP2_DIRECTION_TO_MEDIA                0x0
+#define SBP2_DIRECTION_FROM_MEDIA      0x1
+
+/* Unit directory keys */
+#define SBP2_COMMAND_SET_SPECIFIER     0x38
+#define SBP2_COMMAND_SET               0x39
+#define SBP2_COMMAND_SET_REVISION      0x3b
+#define SBP2_FIRMWARE_REVISION         0x3c
+
+/* Flags for detected oddities and brokeness */
+#define SBP2_WORKAROUND_128K_MAX_TRANS 0x1
+#define SBP2_WORKAROUND_INQUIRY_36     0x2
+#define SBP2_WORKAROUND_MODE_SENSE_8   0x4
+#define SBP2_WORKAROUND_FIX_CAPACITY   0x8
+#define SBP2_WORKAROUND_OVERRIDE       0x100
+
+/* Management orb opcodes */
+#define SBP2_LOGIN_REQUEST             0x0
+#define SBP2_QUERY_LOGINS_REQUEST      0x1
+#define SBP2_RECONNECT_REQUEST         0x3
+#define SBP2_SET_PASSWORD_REQUEST      0x4
+#define SBP2_LOGOUT_REQUEST            0x7
+#define SBP2_ABORT_TASK_REQUEST                0xb
+#define SBP2_ABORT_TASK_SET            0xc
+#define SBP2_LOGICAL_UNIT_RESET                0xe
+#define SBP2_TARGET_RESET_REQUEST      0xf
+
+/* Offsets for command block agent registers */
+#define SBP2_AGENT_STATE               0x00
+#define SBP2_AGENT_RESET               0x04
+#define SBP2_ORB_POINTER               0x08
+#define SBP2_DOORBELL                  0x10
+#define SBP2_UNSOLICITED_STATUS_ENABLE 0x14
+
+/* Status write response codes */
+#define SBP2_STATUS_REQUEST_COMPLETE   0x0
+#define SBP2_STATUS_TRANSPORT_FAILURE  0x1
+#define SBP2_STATUS_ILLEGAL_REQUEST    0x2
+#define SBP2_STATUS_VENDOR_DEPENDENT   0x3
+
+#define status_get_orb_high(v)         ((v).status & 0xffff)
+#define status_get_sbp_status(v)       (((v).status >> 16) & 0xff)
+#define status_get_len(v)              (((v).status >> 24) & 0x07)
+#define status_get_dead(v)             (((v).status >> 27) & 0x01)
+#define status_get_response(v)         (((v).status >> 28) & 0x03)
+#define status_get_source(v)           (((v).status >> 30) & 0x03)
+#define status_get_orb_low(v)          ((v).orb_low)
+#define status_get_data(v)             ((v).data)
+
+struct sbp2_status {
+       u32 status;
+       u32 orb_low;
+       u8 data[24];
+};
+
+struct sbp2_pointer {
+       u32 high;
+       u32 low;
+};
+
+struct sbp2_orb {
+       struct fw_transaction t;
+       dma_addr_t request_bus;
+       int rcode;
+       struct sbp2_pointer pointer;
+       void (*callback) (struct sbp2_orb * orb, struct sbp2_status * status);
+       struct list_head link;
+};
+
+#define management_orb_lun(v)                  ((v))
+#define management_orb_function(v)             ((v) << 16)
+#define management_orb_reconnect(v)            ((v) << 20)
+#define management_orb_exclusive               ((1) << 28)
+#define management_orb_request_format(v)       ((v) << 29)
+#define management_orb_notify                  ((1) << 31)
+
+#define management_orb_response_length(v)      ((v))
+#define management_orb_password_length(v)      ((v) << 16)
+
+struct sbp2_management_orb {
+       struct sbp2_orb base;
+       struct {
+               struct sbp2_pointer password;
+               struct sbp2_pointer response;
+               u32 misc;
+               u32 length;
+               struct sbp2_pointer status_fifo;
+       } request;
+       __be32 response[4];
+       dma_addr_t response_bus;
+       struct completion done;
+       struct sbp2_status status;
+};
+
+#define login_response_get_login_id(v) ((v).misc & 0xffff)
+#define login_response_get_length(v)   (((v).misc >> 16) & 0xffff)
+
+struct sbp2_login_response {
+       u32 misc;
+       struct sbp2_pointer command_block_agent;
+       u32 reconnect_hold;
+};
+
+#define command_orb_data_size(v)       ((v))
+#define command_orb_page_size(v)       ((v) << 16)
+#define command_orb_page_table_present ((1) << 19)
+#define command_orb_max_payload(v)     ((v) << 20)
+#define command_orb_speed(v)           ((v) << 24)
+#define command_orb_direction(v)       ((v) << 27)
+#define command_orb_request_format(v)  ((v) << 29)
+#define command_orb_notify             ((1) << 31)
+
+struct sbp2_command_orb {
+       struct sbp2_orb base;
+       struct {
+               struct sbp2_pointer next;
+               struct sbp2_pointer data_descriptor;
+               u32 misc;
+               u8 command_block[12];
+       } request;
+       struct scsi_cmnd *cmd;
+       scsi_done_fn_t done;
+       struct fw_unit *unit;
+
+       struct sbp2_pointer page_table[SG_ALL];
+       dma_addr_t page_table_bus;
+       dma_addr_t request_buffer_bus;
+};
+
+/*
+ * List of devices with known bugs.
+ *
+ * The firmware_revision field, masked with 0xffff00, is the best
+ * indicator for the type of bridge chip of a device.  It yields a few
+ * false positives but this did not break correctly behaving devices
+ * so far.  We use ~0 as a wildcard, since the 24 bit values we get
+ * from the config rom can never match that.
+ */
+static const struct {
+       u32 firmware_revision;
+       u32 model;
+       unsigned workarounds;
+} sbp2_workarounds_table[] = {
+       /* DViCO Momobay CX-1 with TSB42AA9 bridge */ {
+               .firmware_revision      = 0x002800,
+               .model                  = 0x001010,
+               .workarounds            = SBP2_WORKAROUND_INQUIRY_36 |
+                                         SBP2_WORKAROUND_MODE_SENSE_8,
+       },
+       /* Initio bridges, actually only needed for some older ones */ {
+               .firmware_revision      = 0x000200,
+               .model                  = ~0,
+               .workarounds            = SBP2_WORKAROUND_INQUIRY_36,
+       },
+       /* Symbios bridge */ {
+               .firmware_revision      = 0xa0b800,
+               .model                  = ~0,
+               .workarounds            = SBP2_WORKAROUND_128K_MAX_TRANS,
+       },
+       /* There are iPods (2nd gen, 3rd gen) with model_id == 0, but
+        * these iPods do not feature the read_capacity bug according
+        * to one report.  Read_capacity behaviour as well as model_id
+        * could change due to Apple-supplied firmware updates though. */
+       /* iPod 4th generation. */ {
+               .firmware_revision      = 0x0a2700,
+               .model                  = 0x000021,
+               .workarounds            = SBP2_WORKAROUND_FIX_CAPACITY,
+       },
+       /* iPod mini */ {
+               .firmware_revision      = 0x0a2700,
+               .model                  = 0x000023,
+               .workarounds            = SBP2_WORKAROUND_FIX_CAPACITY,
+       },
+       /* iPod Photo */ {
+               .firmware_revision      = 0x0a2700,
+               .model                  = 0x00007e,
+               .workarounds            = SBP2_WORKAROUND_FIX_CAPACITY,
+       }
+};
+
+static void
+sbp2_status_write(struct fw_card *card, struct fw_request *request,
+                 int tcode, int destination, int source,
+                 int generation, int speed,
+                 unsigned long long offset,
+                 void *payload, size_t length, void *callback_data)
+{
+       struct sbp2_device *sd = callback_data;
+       struct sbp2_orb *orb;
+       struct sbp2_status status;
+       size_t header_size;
+       unsigned long flags;
+
+       if (tcode != TCODE_WRITE_BLOCK_REQUEST ||
+           length == 0 || length > sizeof status) {
+               fw_send_response(card, request, RCODE_TYPE_ERROR);
+               return;
+       }
+
+       header_size = min(length, 2 * sizeof(u32));
+       fw_memcpy_from_be32(&status, payload, header_size);
+       if (length > header_size)
+               memcpy(status.data, payload + 8, length - header_size);
+       if (status_get_source(status) == 2 || status_get_source(status) == 3) {
+               fw_notify("non-orb related status write, not handled\n");
+               fw_send_response(card, request, RCODE_COMPLETE);
+               return;
+       }
+
+       /* Lookup the orb corresponding to this status write. */
+       spin_lock_irqsave(&card->lock, flags);
+       list_for_each_entry(orb, &sd->orb_list, link) {
+               if (status_get_orb_high(status) == 0 &&
+                   status_get_orb_low(status) == orb->request_bus) {
+                       list_del(&orb->link);
+                       break;
+               }
+       }
+       spin_unlock_irqrestore(&card->lock, flags);
+
+       if (&orb->link != &sd->orb_list)
+               orb->callback(orb, &status);
+       else
+               fw_error("status write for unknown orb\n");
+
+       fw_send_response(card, request, RCODE_COMPLETE);
+}
+
+static void
+complete_transaction(struct fw_card *card, int rcode,
+                    void *payload, size_t length, void *data)
+{
+       struct sbp2_orb *orb = data;
+       unsigned long flags;
+
+       orb->rcode = rcode;
+       if (rcode != RCODE_COMPLETE) {
+               spin_lock_irqsave(&card->lock, flags);
+               list_del(&orb->link);
+               spin_unlock_irqrestore(&card->lock, flags);
+               orb->callback(orb, NULL);
+       }
+}
+
+static void
+sbp2_send_orb(struct sbp2_orb *orb, struct fw_unit *unit,
+             int node_id, int generation, u64 offset)
+{
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd = unit->device.driver_data;
+       unsigned long flags;
+
+       orb->pointer.high = 0;
+       orb->pointer.low = orb->request_bus;
+       fw_memcpy_to_be32(&orb->pointer, &orb->pointer, sizeof orb->pointer);
+
+       spin_lock_irqsave(&device->card->lock, flags);
+       list_add_tail(&orb->link, &sd->orb_list);
+       spin_unlock_irqrestore(&device->card->lock, flags);
+
+       fw_send_request(device->card, &orb->t, TCODE_WRITE_BLOCK_REQUEST,
+                       node_id | LOCAL_BUS, generation,
+                       device->node->max_speed, offset,
+                       &orb->pointer, sizeof orb->pointer,
+                       complete_transaction, orb);
+}
+
+static void sbp2_cancel_orbs(struct fw_unit *unit)
+{
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd = unit->device.driver_data;
+       struct sbp2_orb *orb, *next;
+       struct list_head list;
+       unsigned long flags;
+
+       INIT_LIST_HEAD(&list);
+       spin_lock_irqsave(&device->card->lock, flags);
+       list_splice_init(&sd->orb_list, &list);
+       spin_unlock_irqrestore(&device->card->lock, flags);
+
+       list_for_each_entry_safe(orb, next, &list, link) {
+               orb->rcode = RCODE_CANCELLED;
+               orb->callback(orb, NULL);
+       }
+}
+
+static void
+complete_management_orb(struct sbp2_orb *base_orb, struct sbp2_status *status)
+{
+       struct sbp2_management_orb *orb =
+           (struct sbp2_management_orb *)base_orb;
+
+       if (status)
+               memcpy(&orb->status, status, sizeof *status);
+       complete(&orb->done);
+}
+
+static int
+sbp2_send_management_orb(struct fw_unit *unit, int node_id, int generation,
+                        int function, int lun, void *response)
+{
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd = unit->device.driver_data;
+       struct sbp2_management_orb *orb;
+       unsigned long timeout;
+       int retval = -ENOMEM;
+
+       orb = kzalloc(sizeof *orb, GFP_ATOMIC);
+       if (orb == NULL)
+               return -ENOMEM;
+
+       /* The sbp2 device is going to send a block read request to
+        * read out the request from host memory, so map it for
+        * dma. */
+       orb->base.request_bus =
+               dma_map_single(device->card->device, &orb->request,
+                              sizeof orb->request, DMA_TO_DEVICE);
+       if (orb->base.request_bus == 0)
+               goto out;
+
+       orb->response_bus =
+               dma_map_single(device->card->device, &orb->response,
+                              sizeof orb->response, DMA_FROM_DEVICE);
+       if (orb->response_bus == 0)
+               goto out;
+
+       orb->request.response.high    = 0;
+       orb->request.response.low     = orb->response_bus;
+
+       orb->request.misc =
+               management_orb_notify |
+               management_orb_function(function) |
+               management_orb_lun(lun);
+       orb->request.length =
+               management_orb_response_length(sizeof orb->response);
+
+       orb->request.status_fifo.high = sd->address_handler.offset >> 32;
+       orb->request.status_fifo.low  = sd->address_handler.offset;
+
+       /* FIXME: Yeah, ok this isn't elegant, we hardwire exclusive
+        * login and 1 second reconnect time.  The reconnect setting
+        * is probably fine, but the exclusive login should be an
+        * option. */
+       if (function == SBP2_LOGIN_REQUEST) {
+               orb->request.misc |=
+                       management_orb_exclusive |
+                       management_orb_reconnect(0);
+       }
+
+       fw_memcpy_to_be32(&orb->request, &orb->request, sizeof orb->request);
+
+       init_completion(&orb->done);
+       orb->base.callback = complete_management_orb;
+       sbp2_send_orb(&orb->base, unit,
+                     node_id, generation, sd->management_agent_address);
+
+       timeout = wait_for_completion_timeout(&orb->done, 10 * HZ);
+
+       /* FIXME: Handle bus reset race here. */
+
+       retval = -EIO;
+       if (orb->base.rcode != RCODE_COMPLETE) {
+               fw_error("management write failed, rcode 0x%02x\n",
+                        orb->base.rcode);
+               goto out;
+       }
+
+       if (timeout == 0) {
+               fw_error("orb reply timed out, rcode=0x%02x\n",
+                        orb->base.rcode);
+               goto out;
+       }
+
+       if (status_get_response(orb->status) != 0 ||
+           status_get_sbp_status(orb->status) != 0) {
+               fw_error("error status: %d:%d\n",
+                        status_get_response(orb->status),
+                        status_get_sbp_status(orb->status));
+               goto out;
+       }
+
+       retval = 0;
+ out:
+       dma_unmap_single(device->card->device, orb->base.request_bus,
+                        sizeof orb->request, DMA_TO_DEVICE);
+       dma_unmap_single(device->card->device, orb->response_bus,
+                        sizeof orb->response, DMA_FROM_DEVICE);
+
+       if (response)
+               fw_memcpy_from_be32(response,
+                                   orb->response, sizeof orb->response);
+       kfree(orb);
+
+       return retval;
+}
+
+static void
+complete_agent_reset_write(struct fw_card *card, int rcode,
+                          void *payload, size_t length, void *data)
+{
+       struct fw_transaction *t = data;
+
+       fw_notify("agent reset write rcode=%d\n", rcode);
+       kfree(t);
+}
+
+static int sbp2_agent_reset(struct fw_unit *unit)
+{
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd = unit->device.driver_data;
+       struct fw_transaction *t;
+       static u32 zero;
+
+       t = kzalloc(sizeof *t, GFP_ATOMIC);
+       if (t == NULL)
+               return -ENOMEM;
+
+       fw_send_request(device->card, t, TCODE_WRITE_QUADLET_REQUEST,
+                       sd->node_id | LOCAL_BUS, sd->generation, SCODE_400,
+                       sd->command_block_agent_address + SBP2_AGENT_RESET,
+                       &zero, sizeof zero, complete_agent_reset_write, t);
+
+       return 0;
+}
+
+static int add_scsi_devices(struct fw_unit *unit);
+static void remove_scsi_devices(struct fw_unit *unit);
+
+static int sbp2_probe(struct device *dev)
+{
+       struct fw_unit *unit = fw_unit(dev);
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd;
+       struct fw_csr_iterator ci;
+       int i, key, value, lun, retval;
+       int node_id, generation, local_node_id;
+       struct sbp2_login_response response;
+       u32 model, firmware_revision;
+
+       sd = kzalloc(sizeof *sd, GFP_KERNEL);
+       if (sd == NULL)
+               return -ENOMEM;
+
+       unit->device.driver_data = sd;
+       sd->unit = unit;
+       INIT_LIST_HEAD(&sd->orb_list);
+
+       sd->address_handler.length = 0x100;
+       sd->address_handler.address_callback = sbp2_status_write;
+       sd->address_handler.callback_data = sd;
+
+       if (fw_core_add_address_handler(&sd->address_handler,
+                                       &fw_high_memory_region) < 0) {
+               kfree(sd);
+               return -EBUSY;
+       }
+
+       if (fw_device_enable_phys_dma(device) < 0) {
+               fw_core_remove_address_handler(&sd->address_handler);
+               kfree(sd);
+               return -EBUSY;
+       }
+
+       /* Scan unit directory to get management agent address,
+        * firmware revison and model.  Initialize firmware_revision
+        * and model to values that wont match anything in our table. */
+       firmware_revision = 0xff000000;
+       model = 0xff000000;
+       fw_csr_iterator_init(&ci, unit->directory);
+       while (fw_csr_iterator_next(&ci, &key, &value)) {
+               switch (key) {
+               case CSR_DEPENDENT_INFO | CSR_OFFSET:
+                       sd->management_agent_address =
+                               0xfffff0000000ULL + 4 * value;
+                       break;
+               case SBP2_FIRMWARE_REVISION:
+                       firmware_revision = value;
+                       break;
+               case CSR_MODEL:
+                       model = value;
+                       break;
+               }
+       }
+
+       for (i = 0; i < ARRAY_SIZE(sbp2_workarounds_table); i++) {
+               if (sbp2_workarounds_table[i].firmware_revision !=
+                   (firmware_revision & 0xffffff00))
+                       continue;
+               if (sbp2_workarounds_table[i].model != model &&
+                   sbp2_workarounds_table[i].model != ~0)
+                       continue;
+               sd->workarounds |= sbp2_workarounds_table[i].workarounds;
+               break;
+       }
+
+       if (sd->workarounds)
+               fw_notify("Workarounds for node %s: 0x%x "
+                         "(firmware_revision 0x%06x, model_id 0x%06x)\n",
+                         unit->device.bus_id,
+                         sd->workarounds, firmware_revision, model);
+
+       /* FIXME: Make this work for multi-lun devices. */
+       lun = 0;
+
+       generation    = device->card->generation;
+       node_id       = device->node->node_id;
+       local_node_id = device->card->local_node->node_id;
+
+       /* FIXME: We should probably do this from a keventd callback
+        * and handle retries by rescheduling the work. */
+       if (sbp2_send_management_orb(unit, node_id, generation,
+                                    SBP2_LOGIN_REQUEST, lun, &response) < 0) {
+               fw_core_remove_address_handler(&sd->address_handler);
+               kfree(sd);
+               return -EBUSY;
+       }
+
+       sd->generation   = generation;
+       sd->node_id      = node_id;
+       sd->address_high = (LOCAL_BUS | local_node_id) << 16;
+
+       /* Get command block agent offset and login id. */
+       sd->command_block_agent_address =
+               ((u64) response.command_block_agent.high << 32) |
+               response.command_block_agent.low;
+       sd->login_id = login_response_get_login_id(response);
+
+       fw_notify("logged in to sbp2 unit %s\n", unit->device.bus_id);
+       fw_notify(" - management_agent_address: 0x%012llx\n",
+                 (unsigned long long) sd->management_agent_address);
+       fw_notify(" - command_block_agent_address: 0x%012llx\n",
+                 (unsigned long long) sd->command_block_agent_address);
+       fw_notify(" - status write address: 0x%012llx\n",
+                 (unsigned long long) sd->address_handler.offset);
+
+#if 0
+       /* FIXME: The linux1394 sbp2 does this last step. */
+       sbp2_set_busy_timeout(scsi_id);
+#endif
+
+       sbp2_agent_reset(unit);
+
+       retval = add_scsi_devices(unit);
+       if (retval < 0) {
+               sbp2_send_management_orb(unit, sd->node_id, sd->generation,
+                                        SBP2_LOGOUT_REQUEST, sd->login_id,
+                                        NULL);
+               fw_core_remove_address_handler(&sd->address_handler);
+               kfree(sd);
+               return retval;
+       }
+
+       return 0;
+}
+
+static int sbp2_remove(struct device *dev)
+{
+       struct fw_unit *unit = fw_unit(dev);
+       struct sbp2_device *sd = unit->device.driver_data;
+
+       sbp2_send_management_orb(unit, sd->node_id, sd->generation,
+                                SBP2_LOGOUT_REQUEST, sd->login_id, NULL);
+
+       remove_scsi_devices(unit);
+
+       fw_core_remove_address_handler(&sd->address_handler);
+       kfree(sd);
+
+       fw_notify("removed sbp2 unit %s\n", dev->bus_id);
+
+       return 0;
+}
+
+static void sbp2_reconnect(struct work_struct *work)
+{
+       struct sbp2_device *sd = container_of(work, struct sbp2_device, work);
+       struct fw_unit *unit = sd->unit;
+       struct fw_device *device = fw_device(unit->device.parent);
+       int generation, node_id, local_node_id;
+
+       fw_notify("in sbp2_reconnect, reconnecting to unit %s\n",
+                 unit->device.bus_id);
+
+       generation    = device->card->generation;
+       node_id       = device->node->node_id;
+       local_node_id = device->card->local_node->node_id;
+
+       sbp2_send_management_orb(unit, node_id, generation,
+                                SBP2_RECONNECT_REQUEST, sd->login_id, NULL);
+
+       /* FIXME: handle reconnect failures. */
+
+       sbp2_cancel_orbs(unit);
+
+       sd->generation   = generation;
+       sd->node_id      = node_id;
+       sd->address_high = (LOCAL_BUS | local_node_id) << 16;
+}
+
+static void sbp2_update(struct fw_unit *unit)
+{
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd = unit->device.driver_data;
+
+       fw_device_enable_phys_dma(device);
+
+       INIT_WORK(&sd->work, sbp2_reconnect);
+       schedule_work(&sd->work);
+}
+
+#define SBP2_UNIT_SPEC_ID_ENTRY        0x0000609e
+#define SBP2_SW_VERSION_ENTRY  0x00010483
+
+static struct fw_device_id sbp2_id_table[] = {
+       {
+               .match_flags  = FW_MATCH_SPECIFIER_ID | FW_MATCH_VERSION,
+               .specifier_id = SBP2_UNIT_SPEC_ID_ENTRY,
+               .version      = SBP2_SW_VERSION_ENTRY
+       },
+       { }
+};
+
+static struct fw_driver sbp2_driver = {
+       .driver   = {
+               .owner  = THIS_MODULE,
+               .name   = sbp2_driver_name,
+               .bus    = &fw_bus_type,
+               .probe  = sbp2_probe,
+               .remove = sbp2_remove,
+       },
+       .update   = sbp2_update,
+       .id_table = sbp2_id_table,
+};
+
+static unsigned int sbp2_status_to_sense_data(u8 * sbp2_status, u8 * sense_data)
+{
+       sense_data[0] = 0x70;
+       sense_data[1] = 0x0;
+       sense_data[2] = sbp2_status[1];
+       sense_data[3] = sbp2_status[4];
+       sense_data[4] = sbp2_status[5];
+       sense_data[5] = sbp2_status[6];
+       sense_data[6] = sbp2_status[7];
+       sense_data[7] = 10;
+       sense_data[8] = sbp2_status[8];
+       sense_data[9] = sbp2_status[9];
+       sense_data[10] = sbp2_status[10];
+       sense_data[11] = sbp2_status[11];
+       sense_data[12] = sbp2_status[2];
+       sense_data[13] = sbp2_status[3];
+       sense_data[14] = sbp2_status[12];
+       sense_data[15] = sbp2_status[13];
+
+       switch (sbp2_status[0] & 0x3f) {
+       case SAM_STAT_GOOD:
+               return DID_OK;
+
+       case SAM_STAT_CHECK_CONDITION:
+               /* return CHECK_CONDITION << 1 | DID_OK << 16; */
+               return DID_OK;
+
+       case SAM_STAT_BUSY:
+               return DID_BUS_BUSY;
+
+       case SAM_STAT_CONDITION_MET:
+       case SAM_STAT_RESERVATION_CONFLICT:
+       case SAM_STAT_COMMAND_TERMINATED:
+       default:
+               return DID_ERROR;
+       }
+}
+
+static void
+complete_command_orb(struct sbp2_orb *base_orb, struct sbp2_status *status)
+{
+       struct sbp2_command_orb *orb = (struct sbp2_command_orb *)base_orb;
+       struct fw_unit *unit = orb->unit;
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct scatterlist *sg;
+       int result;
+
+       if (status != NULL) {
+               if (status_get_dead(*status)) {
+                       fw_notify("agent died, issuing agent reset\n");
+                       sbp2_agent_reset(unit);
+               }
+
+               switch (status_get_response(*status)) {
+               case SBP2_STATUS_REQUEST_COMPLETE:
+                       result = DID_OK;
+                       break;
+               case SBP2_STATUS_TRANSPORT_FAILURE:
+                       result = DID_BUS_BUSY;
+                       break;
+               case SBP2_STATUS_ILLEGAL_REQUEST:
+               case SBP2_STATUS_VENDOR_DEPENDENT:
+               default:
+                       result = DID_ERROR;
+                       break;
+               }
+
+               if (result == DID_OK && status_get_len(*status) > 1)
+                       result = sbp2_status_to_sense_data(status_get_data(*status),
+                                                          orb->cmd->sense_buffer);
+       } else {
+               /* If the orb completes with status == NULL, something
+                * went wrong, typically a bus reset happened mid-orb
+                * or when sending the write (less likely). */
+               fw_notify("no command orb status, rcode=%d\n",
+                         orb->base.rcode);
+               result = DID_ERROR;
+       }
+
+       dma_unmap_single(device->card->device, orb->base.request_bus,
+                        sizeof orb->request, DMA_TO_DEVICE);
+
+       if (orb->cmd->use_sg > 0) {
+               sg = (struct scatterlist *)orb->cmd->request_buffer;
+               dma_unmap_sg(device->card->device, sg, orb->cmd->use_sg,
+                            orb->cmd->sc_data_direction);
+       }
+
+       if (orb->page_table_bus != 0)
+               dma_unmap_single(device->card->device, orb->page_table_bus,
+                                sizeof orb->page_table_bus, DMA_TO_DEVICE);
+
+       if (orb->request_buffer_bus != 0)
+               dma_unmap_single(device->card->device, orb->request_buffer_bus,
+                                sizeof orb->request_buffer_bus,
+                                DMA_FROM_DEVICE);
+
+       orb->cmd->result = result << 16;
+       orb->done(orb->cmd);
+
+       kfree(orb);
+}
+
+static void sbp2_command_orb_map_scatterlist(struct sbp2_command_orb *orb)
+{
+       struct fw_unit *unit =
+               (struct fw_unit *)orb->cmd->device->host->hostdata[0];
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd = unit->device.driver_data;
+       struct scatterlist *sg;
+       int sg_len, l, i, j, count;
+       size_t size;
+       dma_addr_t sg_addr;
+
+       sg = (struct scatterlist *)orb->cmd->request_buffer;
+       count = dma_map_sg(device->card->device, sg, orb->cmd->use_sg,
+                          orb->cmd->sc_data_direction);
+
+       /* Handle the special case where there is only one element in
+        * the scatter list by converting it to an immediate block
+        * request. This is also a workaround for broken devices such
+        * as the second generation iPod which doesn't support page
+        * tables. */
+       if (count == 1 && sg_dma_len(sg) < SBP2_MAX_SG_ELEMENT_LENGTH) {
+               orb->request.data_descriptor.high = sd->address_high;
+               orb->request.data_descriptor.low  = sg_dma_address(sg);
+               orb->request.misc |=
+                       command_orb_data_size(sg_dma_len(sg));
+               return;
+       }
+
+       /* Convert the scatterlist to an sbp2 page table.  If any
+        * scatterlist entries are too big for sbp2 we split the as we go. */
+       for (i = 0, j = 0; i < count; i++) {
+               sg_len = sg_dma_len(sg + i);
+               sg_addr = sg_dma_address(sg + i);
+               while (sg_len) {
+                       l = min(sg_len, SBP2_MAX_SG_ELEMENT_LENGTH);
+                       orb->page_table[j].low = sg_addr;
+                       orb->page_table[j].high = (l << 16);
+                       sg_addr += l;
+                       sg_len -= l;
+                       j++;
+               }
+       }
+
+       size = sizeof orb->page_table[0] * j;
+
+       /* The data_descriptor pointer is the one case where we need
+        * to fill in the node ID part of the address.  All other
+        * pointers assume that the data referenced reside on the
+        * initiator (i.e. us), but data_descriptor can refer to data
+        * on other nodes so we need to put our ID in descriptor.high. */
+
+       orb->page_table_bus =
+               dma_map_single(device->card->device, orb->page_table,
+                              size, DMA_TO_DEVICE);
+       orb->request.data_descriptor.high = sd->address_high;
+       orb->request.data_descriptor.low  = orb->page_table_bus;
+       orb->request.misc |=
+               command_orb_page_table_present |
+               command_orb_data_size(j);
+
+       fw_memcpy_to_be32(orb->page_table, orb->page_table, size);
+}
+
+static void sbp2_command_orb_map_buffer(struct sbp2_command_orb *orb)
+{
+       struct fw_unit *unit =
+               (struct fw_unit *)orb->cmd->device->host->hostdata[0];
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd = unit->device.driver_data;
+
+       /* As for map_scatterlist, we need to fill in the high bits of
+        * the data_descriptor pointer. */
+
+       orb->request_buffer_bus =
+               dma_map_single(device->card->device,
+                              orb->cmd->request_buffer,
+                              orb->cmd->request_bufflen,
+                              orb->cmd->sc_data_direction);
+       orb->request.data_descriptor.high = sd->address_high;
+       orb->request.data_descriptor.low  = orb->request_buffer_bus;
+       orb->request.misc |=
+               command_orb_data_size(orb->cmd->request_bufflen);
+}
+
+/* SCSI stack integration */
+
+static int sbp2_scsi_queuecommand(struct scsi_cmnd *cmd, scsi_done_fn_t done)
+{
+       struct fw_unit *unit = (struct fw_unit *)cmd->device->host->hostdata[0];
+       struct fw_device *device = fw_device(unit->device.parent);
+       struct sbp2_device *sd = unit->device.driver_data;
+       struct sbp2_command_orb *orb;
+
+       /* Bidirectional commands are not yet implemented, and unknown
+        * transfer direction not handled. */
+       if (cmd->sc_data_direction == DMA_BIDIRECTIONAL) {
+               fw_error("Cannot handle DMA_BIDIRECTIONAL - rejecting command");
+               cmd->result = DID_ERROR << 16;
+               done(cmd);
+               return 0;
+       }
+
+       orb = kzalloc(sizeof *orb, GFP_ATOMIC);
+       if (orb == NULL) {
+               fw_notify("failed to alloc orb\n");
+               cmd->result = DID_NO_CONNECT << 16;
+               done(cmd);
+               return 0;
+       }
+
+       orb->base.request_bus =
+               dma_map_single(device->card->device, &orb->request,
+                              sizeof orb->request, DMA_TO_DEVICE);
+
+       orb->unit = unit;
+       orb->done = done;
+       orb->cmd  = cmd;
+
+       orb->request.next.high   = SBP2_ORB_NULL;
+       orb->request.next.low    = 0x0;
+       /* At speed 100 we can do 512 bytes per packet, at speed 200,
+        * 1024 bytes per packet etc.  The SBP-2 max_payload field
+        * specifies the max payload size as 2 ^ (max_payload + 2), so
+        * if we set this to max_speed + 7, we get the right value. */
+       orb->request.misc =
+               command_orb_max_payload(device->node->max_speed + 7) |
+               command_orb_speed(device->node->max_speed) |
+               command_orb_notify;
+
+       if (cmd->sc_data_direction == DMA_FROM_DEVICE)
+               orb->request.misc |=
+                       command_orb_direction(SBP2_DIRECTION_FROM_MEDIA);
+       else if (cmd->sc_data_direction == DMA_TO_DEVICE)
+               orb->request.misc |=
+                       command_orb_direction(SBP2_DIRECTION_TO_MEDIA);
+
+       if (cmd->use_sg) {
+               sbp2_command_orb_map_scatterlist(orb);
+       } else if (cmd->request_bufflen > SBP2_MAX_SG_ELEMENT_LENGTH) {
+               /* FIXME: Need to split this into a sg list... but
+                * could we get the scsi or blk layer to do that by
+                * reporting our max supported block size? */
+               fw_error("command > 64k\n");
+               cmd->result = DID_ERROR << 16;
+               done(cmd);
+               return 0;
+       } else if (cmd->request_bufflen > 0) {
+               sbp2_command_orb_map_buffer(orb);
+       }
+
+       fw_memcpy_to_be32(&orb->request, &orb->request, sizeof orb->request);
+
+       memset(orb->request.command_block,
+              0, sizeof orb->request.command_block);
+       memcpy(orb->request.command_block, cmd->cmnd, COMMAND_SIZE(*cmd->cmnd));
+
+       orb->base.callback = complete_command_orb;
+
+       sbp2_send_orb(&orb->base, unit, sd->node_id, sd->generation,
+                     sd->command_block_agent_address + SBP2_ORB_POINTER);
+
+       return 0;
+}
+
+static int sbp2_scsi_slave_configure(struct scsi_device *sdev)
+{
+       struct fw_unit *unit = (struct fw_unit *)sdev->host->hostdata[0];
+       struct sbp2_device *sd = unit->device.driver_data;
+
+       if (sdev->type == TYPE_DISK &&
+           sd->workarounds & SBP2_WORKAROUND_MODE_SENSE_8)
+               sdev->skip_ms_page_8 = 1;
+       if (sd->workarounds & SBP2_WORKAROUND_FIX_CAPACITY) {
+               fw_notify("setting fix_capacity for %s\n", unit->device.bus_id);
+               sdev->fix_capacity = 1;
+       }
+
+       return 0;
+}
+
+/*
+ * Called by scsi stack when something has really gone wrong.  Usually
+ * called when a command has timed-out for some reason.
+ */
+static int sbp2_scsi_abort(struct scsi_cmnd *cmd)
+{
+       struct fw_unit *unit = (struct fw_unit *)cmd->device->host->hostdata[0];
+
+       fw_notify("sbp2_scsi_abort\n");
+
+       sbp2_cancel_orbs(unit);
+
+       return SUCCESS;
+}
+
+static struct scsi_host_template scsi_driver_template = {
+       .module                 = THIS_MODULE,
+       .name                   = "SBP-2 IEEE-1394",
+       .proc_name              = (char *)sbp2_driver_name,
+       .queuecommand           = sbp2_scsi_queuecommand,
+       .slave_configure        = sbp2_scsi_slave_configure,
+       .eh_abort_handler       = sbp2_scsi_abort,
+       .this_id                = -1,
+       .sg_tablesize           = SG_ALL,
+       .use_clustering         = ENABLE_CLUSTERING,
+       .cmd_per_lun            = 1,    /* SBP2_MAX_CMDS, */
+       .can_queue              = 1,            /* SBP2_MAX_CMDS, */
+       .emulated               = 1,
+};
+
+static int add_scsi_devices(struct fw_unit *unit)
+{
+       struct sbp2_device *sd = unit->device.driver_data;
+       int retval, lun;
+
+       sd->scsi_host = scsi_host_alloc(&scsi_driver_template,
+                                       sizeof(unsigned long));
+       if (sd->scsi_host == NULL) {
+               fw_error("failed to register scsi host\n");
+               return -1;
+       }
+
+       sd->scsi_host->hostdata[0] = (unsigned long)unit;
+       retval = scsi_add_host(sd->scsi_host, &unit->device);
+       if (retval < 0) {
+               fw_error("failed to add scsi host\n");
+               scsi_host_put(sd->scsi_host);
+               return retval;
+       }
+
+       /* FIXME: Loop over luns here. */
+       lun = 0;
+       retval = scsi_add_device(sd->scsi_host, 0, 0, lun);
+       if (retval < 0) {
+               fw_error("failed to add scsi device\n");
+               scsi_remove_host(sd->scsi_host);
+               scsi_host_put(sd->scsi_host);
+               return retval;
+       }
+
+       return 0;
+}
+
+static void remove_scsi_devices(struct fw_unit *unit)
+{
+       struct sbp2_device *sd = unit->device.driver_data;
+
+       scsi_remove_host(sd->scsi_host);
+       scsi_host_put(sd->scsi_host);
+}
+
+MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
+MODULE_DESCRIPTION("SCSI over IEEE1394");
+MODULE_LICENSE("GPL");
+MODULE_DEVICE_TABLE(ieee1394, sbp2_id_table);
+
+static int __init sbp2_init(void)
+{
+       return driver_register(&sbp2_driver.driver);
+}
+
+static void __exit sbp2_cleanup(void)
+{
+       driver_unregister(&sbp2_driver.driver);
+}
+
+module_init(sbp2_init);
+module_exit(sbp2_cleanup);