Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / firewire / ohci.c
diff --git a/kernel/drivers/firewire/ohci.c b/kernel/drivers/firewire/ohci.c
new file mode 100644 (file)
index 0000000..f51d376
--- /dev/null
@@ -0,0 +1,3892 @@
+/*
+ * Driver for OHCI 1394 controllers
+ *
+ * Copyright (C) 2003-2006 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/bitops.h>
+#include <linux/bug.h>
+#include <linux/compiler.h>
+#include <linux/delay.h>
+#include <linux/device.h>
+#include <linux/dma-mapping.h>
+#include <linux/firewire.h>
+#include <linux/firewire-constants.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/pci_ids.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/string.h>
+#include <linux/time.h>
+#include <linux/vmalloc.h>
+#include <linux/workqueue.h>
+
+#include <asm/byteorder.h>
+#include <asm/page.h>
+
+#ifdef CONFIG_PPC_PMAC
+#include <asm/pmac_feature.h>
+#endif
+
+#include "core.h"
+#include "ohci.h"
+
+#define ohci_info(ohci, f, args...)    dev_info(ohci->card.device, f, ##args)
+#define ohci_notice(ohci, f, args...)  dev_notice(ohci->card.device, f, ##args)
+#define ohci_err(ohci, f, args...)     dev_err(ohci->card.device, f, ##args)
+
+#define DESCRIPTOR_OUTPUT_MORE         0
+#define DESCRIPTOR_OUTPUT_LAST         (1 << 12)
+#define DESCRIPTOR_INPUT_MORE          (2 << 12)
+#define DESCRIPTOR_INPUT_LAST          (3 << 12)
+#define DESCRIPTOR_STATUS              (1 << 11)
+#define DESCRIPTOR_KEY_IMMEDIATE       (2 << 8)
+#define DESCRIPTOR_PING                        (1 << 7)
+#define DESCRIPTOR_YY                  (1 << 6)
+#define DESCRIPTOR_NO_IRQ              (0 << 4)
+#define DESCRIPTOR_IRQ_ERROR           (1 << 4)
+#define DESCRIPTOR_IRQ_ALWAYS          (3 << 4)
+#define DESCRIPTOR_BRANCH_ALWAYS       (3 << 2)
+#define DESCRIPTOR_WAIT                        (3 << 0)
+
+#define DESCRIPTOR_CMD                 (0xf << 12)
+
+struct descriptor {
+       __le16 req_count;
+       __le16 control;
+       __le32 data_address;
+       __le32 branch_address;
+       __le16 res_count;
+       __le16 transfer_status;
+} __attribute__((aligned(16)));
+
+#define CONTROL_SET(regs)      (regs)
+#define CONTROL_CLEAR(regs)    ((regs) + 4)
+#define COMMAND_PTR(regs)      ((regs) + 12)
+#define CONTEXT_MATCH(regs)    ((regs) + 16)
+
+#define AR_BUFFER_SIZE (32*1024)
+#define AR_BUFFERS_MIN DIV_ROUND_UP(AR_BUFFER_SIZE, PAGE_SIZE)
+/* we need at least two pages for proper list management */
+#define AR_BUFFERS     (AR_BUFFERS_MIN >= 2 ? AR_BUFFERS_MIN : 2)
+
+#define MAX_ASYNC_PAYLOAD      4096
+#define MAX_AR_PACKET_SIZE     (16 + MAX_ASYNC_PAYLOAD + 4)
+#define AR_WRAPAROUND_PAGES    DIV_ROUND_UP(MAX_AR_PACKET_SIZE, PAGE_SIZE)
+
+struct ar_context {
+       struct fw_ohci *ohci;
+       struct page *pages[AR_BUFFERS];
+       void *buffer;
+       struct descriptor *descriptors;
+       dma_addr_t descriptors_bus;
+       void *pointer;
+       unsigned int last_buffer_index;
+       u32 regs;
+       struct tasklet_struct tasklet;
+};
+
+struct context;
+
+typedef int (*descriptor_callback_t)(struct context *ctx,
+                                    struct descriptor *d,
+                                    struct descriptor *last);
+
+/*
+ * A buffer that contains a block of DMA-able coherent memory used for
+ * storing a portion of a DMA descriptor program.
+ */
+struct descriptor_buffer {
+       struct list_head list;
+       dma_addr_t buffer_bus;
+       size_t buffer_size;
+       size_t used;
+       struct descriptor buffer[0];
+};
+
+struct context {
+       struct fw_ohci *ohci;
+       u32 regs;
+       int total_allocation;
+       u32 current_bus;
+       bool running;
+       bool flushing;
+
+       /*
+        * List of page-sized buffers for storing DMA descriptors.
+        * Head of list contains buffers in use and tail of list contains
+        * free buffers.
+        */
+       struct list_head buffer_list;
+
+       /*
+        * Pointer to a buffer inside buffer_list that contains the tail
+        * end of the current DMA program.
+        */
+       struct descriptor_buffer *buffer_tail;
+
+       /*
+        * The descriptor containing the branch address of the first
+        * descriptor that has not yet been filled by the device.
+        */
+       struct descriptor *last;
+
+       /*
+        * The last descriptor block in the DMA program. It contains the branch
+        * address that must be updated upon appending a new descriptor.
+        */
+       struct descriptor *prev;
+       int prev_z;
+
+       descriptor_callback_t callback;
+
+       struct tasklet_struct tasklet;
+};
+
+#define IT_HEADER_SY(v)          ((v) <<  0)
+#define IT_HEADER_TCODE(v)       ((v) <<  4)
+#define IT_HEADER_CHANNEL(v)     ((v) <<  8)
+#define IT_HEADER_TAG(v)         ((v) << 14)
+#define IT_HEADER_SPEED(v)       ((v) << 16)
+#define IT_HEADER_DATA_LENGTH(v) ((v) << 16)
+
+struct iso_context {
+       struct fw_iso_context base;
+       struct context context;
+       void *header;
+       size_t header_length;
+       unsigned long flushing_completions;
+       u32 mc_buffer_bus;
+       u16 mc_completed;
+       u16 last_timestamp;
+       u8 sync;
+       u8 tags;
+};
+
+#define CONFIG_ROM_SIZE 1024
+
+struct fw_ohci {
+       struct fw_card card;
+
+       __iomem char *registers;
+       int node_id;
+       int generation;
+       int request_generation; /* for timestamping incoming requests */
+       unsigned quirks;
+       unsigned int pri_req_max;
+       u32 bus_time;
+       bool bus_time_running;
+       bool is_root;
+       bool csr_state_setclear_abdicate;
+       int n_ir;
+       int n_it;
+       /*
+        * Spinlock for accessing fw_ohci data.  Never call out of
+        * this driver with this lock held.
+        */
+       spinlock_t lock;
+
+       struct mutex phy_reg_mutex;
+
+       void *misc_buffer;
+       dma_addr_t misc_buffer_bus;
+
+       struct ar_context ar_request_ctx;
+       struct ar_context ar_response_ctx;
+       struct context at_request_ctx;
+       struct context at_response_ctx;
+
+       u32 it_context_support;
+       u32 it_context_mask;     /* unoccupied IT contexts */
+       struct iso_context *it_context_list;
+       u64 ir_context_channels; /* unoccupied channels */
+       u32 ir_context_support;
+       u32 ir_context_mask;     /* unoccupied IR contexts */
+       struct iso_context *ir_context_list;
+       u64 mc_channels; /* channels in use by the multichannel IR context */
+       bool mc_allocated;
+
+       __be32    *config_rom;
+       dma_addr_t config_rom_bus;
+       __be32    *next_config_rom;
+       dma_addr_t next_config_rom_bus;
+       __be32     next_header;
+
+       __le32    *self_id;
+       dma_addr_t self_id_bus;
+       struct work_struct bus_reset_work;
+
+       u32 self_id_buffer[512];
+};
+
+static struct workqueue_struct *selfid_workqueue;
+
+static inline struct fw_ohci *fw_ohci(struct fw_card *card)
+{
+       return container_of(card, struct fw_ohci, card);
+}
+
+#define IT_CONTEXT_CYCLE_MATCH_ENABLE  0x80000000
+#define IR_CONTEXT_BUFFER_FILL         0x80000000
+#define IR_CONTEXT_ISOCH_HEADER                0x40000000
+#define IR_CONTEXT_CYCLE_MATCH_ENABLE  0x20000000
+#define IR_CONTEXT_MULTI_CHANNEL_MODE  0x10000000
+#define IR_CONTEXT_DUAL_BUFFER_MODE    0x08000000
+
+#define CONTEXT_RUN    0x8000
+#define CONTEXT_WAKE   0x1000
+#define CONTEXT_DEAD   0x0800
+#define CONTEXT_ACTIVE 0x0400
+
+#define OHCI1394_MAX_AT_REQ_RETRIES    0xf
+#define OHCI1394_MAX_AT_RESP_RETRIES   0x2
+#define OHCI1394_MAX_PHYS_RESP_RETRIES 0x8
+
+#define OHCI1394_REGISTER_SIZE         0x800
+#define OHCI1394_PCI_HCI_Control       0x40
+#define SELF_ID_BUF_SIZE               0x800
+#define OHCI_TCODE_PHY_PACKET          0x0e
+#define OHCI_VERSION_1_1               0x010010
+
+static char ohci_driver_name[] = KBUILD_MODNAME;
+
+#define PCI_VENDOR_ID_PINNACLE_SYSTEMS 0x11bd
+#define PCI_DEVICE_ID_AGERE_FW643      0x5901
+#define PCI_DEVICE_ID_CREATIVE_SB1394  0x4001
+#define PCI_DEVICE_ID_JMICRON_JMB38X_FW        0x2380
+#define PCI_DEVICE_ID_TI_TSB12LV22     0x8009
+#define PCI_DEVICE_ID_TI_TSB12LV26     0x8020
+#define PCI_DEVICE_ID_TI_TSB82AA2      0x8025
+#define PCI_DEVICE_ID_VIA_VT630X       0x3044
+#define PCI_REV_ID_VIA_VT6306          0x46
+#define PCI_DEVICE_ID_VIA_VT6315       0x3403
+
+#define QUIRK_CYCLE_TIMER              0x1
+#define QUIRK_RESET_PACKET             0x2
+#define QUIRK_BE_HEADERS               0x4
+#define QUIRK_NO_1394A                 0x8
+#define QUIRK_NO_MSI                   0x10
+#define QUIRK_TI_SLLZ059               0x20
+#define QUIRK_IR_WAKE                  0x40
+
+/* In case of multiple matches in ohci_quirks[], only the first one is used. */
+static const struct {
+       unsigned short vendor, device, revision, flags;
+} ohci_quirks[] = {
+       {PCI_VENDOR_ID_AL, PCI_ANY_ID, PCI_ANY_ID,
+               QUIRK_CYCLE_TIMER},
+
+       {PCI_VENDOR_ID_APPLE, PCI_DEVICE_ID_APPLE_UNI_N_FW, PCI_ANY_ID,
+               QUIRK_BE_HEADERS},
+
+       {PCI_VENDOR_ID_ATT, PCI_DEVICE_ID_AGERE_FW643, 6,
+               QUIRK_NO_MSI},
+
+       {PCI_VENDOR_ID_CREATIVE, PCI_DEVICE_ID_CREATIVE_SB1394, PCI_ANY_ID,
+               QUIRK_RESET_PACKET},
+
+       {PCI_VENDOR_ID_JMICRON, PCI_DEVICE_ID_JMICRON_JMB38X_FW, PCI_ANY_ID,
+               QUIRK_NO_MSI},
+
+       {PCI_VENDOR_ID_NEC, PCI_ANY_ID, PCI_ANY_ID,
+               QUIRK_CYCLE_TIMER},
+
+       {PCI_VENDOR_ID_O2, PCI_ANY_ID, PCI_ANY_ID,
+               QUIRK_NO_MSI},
+
+       {PCI_VENDOR_ID_RICOH, PCI_ANY_ID, PCI_ANY_ID,
+               QUIRK_CYCLE_TIMER | QUIRK_NO_MSI},
+
+       {PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB12LV22, PCI_ANY_ID,
+               QUIRK_CYCLE_TIMER | QUIRK_RESET_PACKET | QUIRK_NO_1394A},
+
+       {PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB12LV26, PCI_ANY_ID,
+               QUIRK_RESET_PACKET | QUIRK_TI_SLLZ059},
+
+       {PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_TSB82AA2, PCI_ANY_ID,
+               QUIRK_RESET_PACKET | QUIRK_TI_SLLZ059},
+
+       {PCI_VENDOR_ID_TI, PCI_ANY_ID, PCI_ANY_ID,
+               QUIRK_RESET_PACKET},
+
+       {PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT630X, PCI_REV_ID_VIA_VT6306,
+               QUIRK_CYCLE_TIMER | QUIRK_IR_WAKE},
+
+       {PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT6315, 0,
+               QUIRK_CYCLE_TIMER /* FIXME: necessary? */ | QUIRK_NO_MSI},
+
+       {PCI_VENDOR_ID_VIA, PCI_DEVICE_ID_VIA_VT6315, PCI_ANY_ID,
+               QUIRK_NO_MSI},
+
+       {PCI_VENDOR_ID_VIA, PCI_ANY_ID, PCI_ANY_ID,
+               QUIRK_CYCLE_TIMER | QUIRK_NO_MSI},
+};
+
+/* This overrides anything that was found in ohci_quirks[]. */
+static int param_quirks;
+module_param_named(quirks, param_quirks, int, 0644);
+MODULE_PARM_DESC(quirks, "Chip quirks (default = 0"
+       ", nonatomic cycle timer = "    __stringify(QUIRK_CYCLE_TIMER)
+       ", reset packet generation = "  __stringify(QUIRK_RESET_PACKET)
+       ", AR/selfID endianness = "     __stringify(QUIRK_BE_HEADERS)
+       ", no 1394a enhancements = "    __stringify(QUIRK_NO_1394A)
+       ", disable MSI = "              __stringify(QUIRK_NO_MSI)
+       ", TI SLLZ059 erratum = "       __stringify(QUIRK_TI_SLLZ059)
+       ", IR wake unreliable = "       __stringify(QUIRK_IR_WAKE)
+       ")");
+
+#define OHCI_PARAM_DEBUG_AT_AR         1
+#define OHCI_PARAM_DEBUG_SELFIDS       2
+#define OHCI_PARAM_DEBUG_IRQS          4
+#define OHCI_PARAM_DEBUG_BUSRESETS     8 /* only effective before chip init */
+
+static int param_debug;
+module_param_named(debug, param_debug, int, 0644);
+MODULE_PARM_DESC(debug, "Verbose logging (default = 0"
+       ", AT/AR events = "     __stringify(OHCI_PARAM_DEBUG_AT_AR)
+       ", self-IDs = "         __stringify(OHCI_PARAM_DEBUG_SELFIDS)
+       ", IRQs = "             __stringify(OHCI_PARAM_DEBUG_IRQS)
+       ", busReset events = "  __stringify(OHCI_PARAM_DEBUG_BUSRESETS)
+       ", or a combination, or all = -1)");
+
+static bool param_remote_dma;
+module_param_named(remote_dma, param_remote_dma, bool, 0444);
+MODULE_PARM_DESC(remote_dma, "Enable unfiltered remote DMA (default = N)");
+
+static void log_irqs(struct fw_ohci *ohci, u32 evt)
+{
+       if (likely(!(param_debug &
+                       (OHCI_PARAM_DEBUG_IRQS | OHCI_PARAM_DEBUG_BUSRESETS))))
+               return;
+
+       if (!(param_debug & OHCI_PARAM_DEBUG_IRQS) &&
+           !(evt & OHCI1394_busReset))
+               return;
+
+       ohci_notice(ohci, "IRQ %08x%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n", evt,
+           evt & OHCI1394_selfIDComplete       ? " selfID"             : "",
+           evt & OHCI1394_RQPkt                ? " AR_req"             : "",
+           evt & OHCI1394_RSPkt                ? " AR_resp"            : "",
+           evt & OHCI1394_reqTxComplete        ? " AT_req"             : "",
+           evt & OHCI1394_respTxComplete       ? " AT_resp"            : "",
+           evt & OHCI1394_isochRx              ? " IR"                 : "",
+           evt & OHCI1394_isochTx              ? " IT"                 : "",
+           evt & OHCI1394_postedWriteErr       ? " postedWriteErr"     : "",
+           evt & OHCI1394_cycleTooLong         ? " cycleTooLong"       : "",
+           evt & OHCI1394_cycle64Seconds       ? " cycle64Seconds"     : "",
+           evt & OHCI1394_cycleInconsistent    ? " cycleInconsistent"  : "",
+           evt & OHCI1394_regAccessFail        ? " regAccessFail"      : "",
+           evt & OHCI1394_unrecoverableError   ? " unrecoverableError" : "",
+           evt & OHCI1394_busReset             ? " busReset"           : "",
+           evt & ~(OHCI1394_selfIDComplete | OHCI1394_RQPkt |
+                   OHCI1394_RSPkt | OHCI1394_reqTxComplete |
+                   OHCI1394_respTxComplete | OHCI1394_isochRx |
+                   OHCI1394_isochTx | OHCI1394_postedWriteErr |
+                   OHCI1394_cycleTooLong | OHCI1394_cycle64Seconds |
+                   OHCI1394_cycleInconsistent |
+                   OHCI1394_regAccessFail | OHCI1394_busReset)
+                                               ? " ?"                  : "");
+}
+
+static const char *speed[] = {
+       [0] = "S100", [1] = "S200", [2] = "S400",    [3] = "beta",
+};
+static const char *power[] = {
+       [0] = "+0W",  [1] = "+15W", [2] = "+30W",    [3] = "+45W",
+       [4] = "-3W",  [5] = " ?W",  [6] = "-3..-6W", [7] = "-3..-10W",
+};
+static const char port[] = { '.', '-', 'p', 'c', };
+
+static char _p(u32 *s, int shift)
+{
+       return port[*s >> shift & 3];
+}
+
+static void log_selfids(struct fw_ohci *ohci, int generation, int self_id_count)
+{
+       u32 *s;
+
+       if (likely(!(param_debug & OHCI_PARAM_DEBUG_SELFIDS)))
+               return;
+
+       ohci_notice(ohci, "%d selfIDs, generation %d, local node ID %04x\n",
+                   self_id_count, generation, ohci->node_id);
+
+       for (s = ohci->self_id_buffer; self_id_count--; ++s)
+               if ((*s & 1 << 23) == 0)
+                       ohci_notice(ohci,
+                           "selfID 0: %08x, phy %d [%c%c%c] %s gc=%d %s %s%s%s\n",
+                           *s, *s >> 24 & 63, _p(s, 6), _p(s, 4), _p(s, 2),
+                           speed[*s >> 14 & 3], *s >> 16 & 63,
+                           power[*s >> 8 & 7], *s >> 22 & 1 ? "L" : "",
+                           *s >> 11 & 1 ? "c" : "", *s & 2 ? "i" : "");
+               else
+                       ohci_notice(ohci,
+                           "selfID n: %08x, phy %d [%c%c%c%c%c%c%c%c]\n",
+                           *s, *s >> 24 & 63,
+                           _p(s, 16), _p(s, 14), _p(s, 12), _p(s, 10),
+                           _p(s,  8), _p(s,  6), _p(s,  4), _p(s,  2));
+}
+
+static const char *evts[] = {
+       [0x00] = "evt_no_status",       [0x01] = "-reserved-",
+       [0x02] = "evt_long_packet",     [0x03] = "evt_missing_ack",
+       [0x04] = "evt_underrun",        [0x05] = "evt_overrun",
+       [0x06] = "evt_descriptor_read", [0x07] = "evt_data_read",
+       [0x08] = "evt_data_write",      [0x09] = "evt_bus_reset",
+       [0x0a] = "evt_timeout",         [0x0b] = "evt_tcode_err",
+       [0x0c] = "-reserved-",          [0x0d] = "-reserved-",
+       [0x0e] = "evt_unknown",         [0x0f] = "evt_flushed",
+       [0x10] = "-reserved-",          [0x11] = "ack_complete",
+       [0x12] = "ack_pending ",        [0x13] = "-reserved-",
+       [0x14] = "ack_busy_X",          [0x15] = "ack_busy_A",
+       [0x16] = "ack_busy_B",          [0x17] = "-reserved-",
+       [0x18] = "-reserved-",          [0x19] = "-reserved-",
+       [0x1a] = "-reserved-",          [0x1b] = "ack_tardy",
+       [0x1c] = "-reserved-",          [0x1d] = "ack_data_error",
+       [0x1e] = "ack_type_error",      [0x1f] = "-reserved-",
+       [0x20] = "pending/cancelled",
+};
+static const char *tcodes[] = {
+       [0x0] = "QW req",               [0x1] = "BW req",
+       [0x2] = "W resp",               [0x3] = "-reserved-",
+       [0x4] = "QR req",               [0x5] = "BR req",
+       [0x6] = "QR resp",              [0x7] = "BR resp",
+       [0x8] = "cycle start",          [0x9] = "Lk req",
+       [0xa] = "async stream packet",  [0xb] = "Lk resp",
+       [0xc] = "-reserved-",           [0xd] = "-reserved-",
+       [0xe] = "link internal",        [0xf] = "-reserved-",
+};
+
+static void log_ar_at_event(struct fw_ohci *ohci,
+                           char dir, int speed, u32 *header, int evt)
+{
+       int tcode = header[0] >> 4 & 0xf;
+       char specific[12];
+
+       if (likely(!(param_debug & OHCI_PARAM_DEBUG_AT_AR)))
+               return;
+
+       if (unlikely(evt >= ARRAY_SIZE(evts)))
+                       evt = 0x1f;
+
+       if (evt == OHCI1394_evt_bus_reset) {
+               ohci_notice(ohci, "A%c evt_bus_reset, generation %d\n",
+                           dir, (header[2] >> 16) & 0xff);
+               return;
+       }
+
+       switch (tcode) {
+       case 0x0: case 0x6: case 0x8:
+               snprintf(specific, sizeof(specific), " = %08x",
+                        be32_to_cpu((__force __be32)header[3]));
+               break;
+       case 0x1: case 0x5: case 0x7: case 0x9: case 0xb:
+               snprintf(specific, sizeof(specific), " %x,%x",
+                        header[3] >> 16, header[3] & 0xffff);
+               break;
+       default:
+               specific[0] = '\0';
+       }
+
+       switch (tcode) {
+       case 0xa:
+               ohci_notice(ohci, "A%c %s, %s\n",
+                           dir, evts[evt], tcodes[tcode]);
+               break;
+       case 0xe:
+               ohci_notice(ohci, "A%c %s, PHY %08x %08x\n",
+                           dir, evts[evt], header[1], header[2]);
+               break;
+       case 0x0: case 0x1: case 0x4: case 0x5: case 0x9:
+               ohci_notice(ohci,
+                           "A%c spd %x tl %02x, %04x -> %04x, %s, %s, %04x%08x%s\n",
+                           dir, speed, header[0] >> 10 & 0x3f,
+                           header[1] >> 16, header[0] >> 16, evts[evt],
+                           tcodes[tcode], header[1] & 0xffff, header[2], specific);
+               break;
+       default:
+               ohci_notice(ohci,
+                           "A%c spd %x tl %02x, %04x -> %04x, %s, %s%s\n",
+                           dir, speed, header[0] >> 10 & 0x3f,
+                           header[1] >> 16, header[0] >> 16, evts[evt],
+                           tcodes[tcode], specific);
+       }
+}
+
+static inline void reg_write(const struct fw_ohci *ohci, int offset, u32 data)
+{
+       writel(data, ohci->registers + offset);
+}
+
+static inline u32 reg_read(const struct fw_ohci *ohci, int offset)
+{
+       return readl(ohci->registers + offset);
+}
+
+static inline void flush_writes(const struct fw_ohci *ohci)
+{
+       /* Do a dummy read to flush writes. */
+       reg_read(ohci, OHCI1394_Version);
+}
+
+/*
+ * Beware!  read_phy_reg(), write_phy_reg(), update_phy_reg(), and
+ * read_paged_phy_reg() require the caller to hold ohci->phy_reg_mutex.
+ * In other words, only use ohci_read_phy_reg() and ohci_update_phy_reg()
+ * directly.  Exceptions are intrinsically serialized contexts like pci_probe.
+ */
+static int read_phy_reg(struct fw_ohci *ohci, int addr)
+{
+       u32 val;
+       int i;
+
+       reg_write(ohci, OHCI1394_PhyControl, OHCI1394_PhyControl_Read(addr));
+       for (i = 0; i < 3 + 100; i++) {
+               val = reg_read(ohci, OHCI1394_PhyControl);
+               if (!~val)
+                       return -ENODEV; /* Card was ejected. */
+
+               if (val & OHCI1394_PhyControl_ReadDone)
+                       return OHCI1394_PhyControl_ReadData(val);
+
+               /*
+                * Try a few times without waiting.  Sleeping is necessary
+                * only when the link/PHY interface is busy.
+                */
+               if (i >= 3)
+                       msleep(1);
+       }
+       ohci_err(ohci, "failed to read phy reg %d\n", addr);
+       dump_stack();
+
+       return -EBUSY;
+}
+
+static int write_phy_reg(const struct fw_ohci *ohci, int addr, u32 val)
+{
+       int i;
+
+       reg_write(ohci, OHCI1394_PhyControl,
+                 OHCI1394_PhyControl_Write(addr, val));
+       for (i = 0; i < 3 + 100; i++) {
+               val = reg_read(ohci, OHCI1394_PhyControl);
+               if (!~val)
+                       return -ENODEV; /* Card was ejected. */
+
+               if (!(val & OHCI1394_PhyControl_WritePending))
+                       return 0;
+
+               if (i >= 3)
+                       msleep(1);
+       }
+       ohci_err(ohci, "failed to write phy reg %d, val %u\n", addr, val);
+       dump_stack();
+
+       return -EBUSY;
+}
+
+static int update_phy_reg(struct fw_ohci *ohci, int addr,
+                         int clear_bits, int set_bits)
+{
+       int ret = read_phy_reg(ohci, addr);
+       if (ret < 0)
+               return ret;
+
+       /*
+        * The interrupt status bits are cleared by writing a one bit.
+        * Avoid clearing them unless explicitly requested in set_bits.
+        */
+       if (addr == 5)
+               clear_bits |= PHY_INT_STATUS_BITS;
+
+       return write_phy_reg(ohci, addr, (ret & ~clear_bits) | set_bits);
+}
+
+static int read_paged_phy_reg(struct fw_ohci *ohci, int page, int addr)
+{
+       int ret;
+
+       ret = update_phy_reg(ohci, 7, PHY_PAGE_SELECT, page << 5);
+       if (ret < 0)
+               return ret;
+
+       return read_phy_reg(ohci, addr);
+}
+
+static int ohci_read_phy_reg(struct fw_card *card, int addr)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+       int ret;
+
+       mutex_lock(&ohci->phy_reg_mutex);
+       ret = read_phy_reg(ohci, addr);
+       mutex_unlock(&ohci->phy_reg_mutex);
+
+       return ret;
+}
+
+static int ohci_update_phy_reg(struct fw_card *card, int addr,
+                              int clear_bits, int set_bits)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+       int ret;
+
+       mutex_lock(&ohci->phy_reg_mutex);
+       ret = update_phy_reg(ohci, addr, clear_bits, set_bits);
+       mutex_unlock(&ohci->phy_reg_mutex);
+
+       return ret;
+}
+
+static inline dma_addr_t ar_buffer_bus(struct ar_context *ctx, unsigned int i)
+{
+       return page_private(ctx->pages[i]);
+}
+
+static void ar_context_link_page(struct ar_context *ctx, unsigned int index)
+{
+       struct descriptor *d;
+
+       d = &ctx->descriptors[index];
+       d->branch_address  &= cpu_to_le32(~0xf);
+       d->res_count       =  cpu_to_le16(PAGE_SIZE);
+       d->transfer_status =  0;
+
+       wmb(); /* finish init of new descriptors before branch_address update */
+       d = &ctx->descriptors[ctx->last_buffer_index];
+       d->branch_address  |= cpu_to_le32(1);
+
+       ctx->last_buffer_index = index;
+
+       reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
+}
+
+static void ar_context_release(struct ar_context *ctx)
+{
+       unsigned int i;
+
+       vunmap(ctx->buffer);
+
+       for (i = 0; i < AR_BUFFERS; i++)
+               if (ctx->pages[i]) {
+                       dma_unmap_page(ctx->ohci->card.device,
+                                      ar_buffer_bus(ctx, i),
+                                      PAGE_SIZE, DMA_FROM_DEVICE);
+                       __free_page(ctx->pages[i]);
+               }
+}
+
+static void ar_context_abort(struct ar_context *ctx, const char *error_msg)
+{
+       struct fw_ohci *ohci = ctx->ohci;
+
+       if (reg_read(ohci, CONTROL_CLEAR(ctx->regs)) & CONTEXT_RUN) {
+               reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
+               flush_writes(ohci);
+
+               ohci_err(ohci, "AR error: %s; DMA stopped\n", error_msg);
+       }
+       /* FIXME: restart? */
+}
+
+static inline unsigned int ar_next_buffer_index(unsigned int index)
+{
+       return (index + 1) % AR_BUFFERS;
+}
+
+static inline unsigned int ar_first_buffer_index(struct ar_context *ctx)
+{
+       return ar_next_buffer_index(ctx->last_buffer_index);
+}
+
+/*
+ * We search for the buffer that contains the last AR packet DMA data written
+ * by the controller.
+ */
+static unsigned int ar_search_last_active_buffer(struct ar_context *ctx,
+                                                unsigned int *buffer_offset)
+{
+       unsigned int i, next_i, last = ctx->last_buffer_index;
+       __le16 res_count, next_res_count;
+
+       i = ar_first_buffer_index(ctx);
+       res_count = ACCESS_ONCE(ctx->descriptors[i].res_count);
+
+       /* A buffer that is not yet completely filled must be the last one. */
+       while (i != last && res_count == 0) {
+
+               /* Peek at the next descriptor. */
+               next_i = ar_next_buffer_index(i);
+               rmb(); /* read descriptors in order */
+               next_res_count = ACCESS_ONCE(
+                               ctx->descriptors[next_i].res_count);
+               /*
+                * If the next descriptor is still empty, we must stop at this
+                * descriptor.
+                */
+               if (next_res_count == cpu_to_le16(PAGE_SIZE)) {
+                       /*
+                        * The exception is when the DMA data for one packet is
+                        * split over three buffers; in this case, the middle
+                        * buffer's descriptor might be never updated by the
+                        * controller and look still empty, and we have to peek
+                        * at the third one.
+                        */
+                       if (MAX_AR_PACKET_SIZE > PAGE_SIZE && i != last) {
+                               next_i = ar_next_buffer_index(next_i);
+                               rmb();
+                               next_res_count = ACCESS_ONCE(
+                                       ctx->descriptors[next_i].res_count);
+                               if (next_res_count != cpu_to_le16(PAGE_SIZE))
+                                       goto next_buffer_is_active;
+                       }
+
+                       break;
+               }
+
+next_buffer_is_active:
+               i = next_i;
+               res_count = next_res_count;
+       }
+
+       rmb(); /* read res_count before the DMA data */
+
+       *buffer_offset = PAGE_SIZE - le16_to_cpu(res_count);
+       if (*buffer_offset > PAGE_SIZE) {
+               *buffer_offset = 0;
+               ar_context_abort(ctx, "corrupted descriptor");
+       }
+
+       return i;
+}
+
+static void ar_sync_buffers_for_cpu(struct ar_context *ctx,
+                                   unsigned int end_buffer_index,
+                                   unsigned int end_buffer_offset)
+{
+       unsigned int i;
+
+       i = ar_first_buffer_index(ctx);
+       while (i != end_buffer_index) {
+               dma_sync_single_for_cpu(ctx->ohci->card.device,
+                                       ar_buffer_bus(ctx, i),
+                                       PAGE_SIZE, DMA_FROM_DEVICE);
+               i = ar_next_buffer_index(i);
+       }
+       if (end_buffer_offset > 0)
+               dma_sync_single_for_cpu(ctx->ohci->card.device,
+                                       ar_buffer_bus(ctx, i),
+                                       end_buffer_offset, DMA_FROM_DEVICE);
+}
+
+#if defined(CONFIG_PPC_PMAC) && defined(CONFIG_PPC32)
+#define cond_le32_to_cpu(v) \
+       (ohci->quirks & QUIRK_BE_HEADERS ? (__force __u32)(v) : le32_to_cpu(v))
+#else
+#define cond_le32_to_cpu(v) le32_to_cpu(v)
+#endif
+
+static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
+{
+       struct fw_ohci *ohci = ctx->ohci;
+       struct fw_packet p;
+       u32 status, length, tcode;
+       int evt;
+
+       p.header[0] = cond_le32_to_cpu(buffer[0]);
+       p.header[1] = cond_le32_to_cpu(buffer[1]);
+       p.header[2] = cond_le32_to_cpu(buffer[2]);
+
+       tcode = (p.header[0] >> 4) & 0x0f;
+       switch (tcode) {
+       case TCODE_WRITE_QUADLET_REQUEST:
+       case TCODE_READ_QUADLET_RESPONSE:
+               p.header[3] = (__force __u32) buffer[3];
+               p.header_length = 16;
+               p.payload_length = 0;
+               break;
+
+       case TCODE_READ_BLOCK_REQUEST :
+               p.header[3] = cond_le32_to_cpu(buffer[3]);
+               p.header_length = 16;
+               p.payload_length = 0;
+               break;
+
+       case TCODE_WRITE_BLOCK_REQUEST:
+       case TCODE_READ_BLOCK_RESPONSE:
+       case TCODE_LOCK_REQUEST:
+       case TCODE_LOCK_RESPONSE:
+               p.header[3] = cond_le32_to_cpu(buffer[3]);
+               p.header_length = 16;
+               p.payload_length = p.header[3] >> 16;
+               if (p.payload_length > MAX_ASYNC_PAYLOAD) {
+                       ar_context_abort(ctx, "invalid packet length");
+                       return NULL;
+               }
+               break;
+
+       case TCODE_WRITE_RESPONSE:
+       case TCODE_READ_QUADLET_REQUEST:
+       case OHCI_TCODE_PHY_PACKET:
+               p.header_length = 12;
+               p.payload_length = 0;
+               break;
+
+       default:
+               ar_context_abort(ctx, "invalid tcode");
+               return NULL;
+       }
+
+       p.payload = (void *) buffer + p.header_length;
+
+       /* FIXME: What to do about evt_* errors? */
+       length = (p.header_length + p.payload_length + 3) / 4;
+       status = cond_le32_to_cpu(buffer[length]);
+       evt    = (status >> 16) & 0x1f;
+
+       p.ack        = evt - 16;
+       p.speed      = (status >> 21) & 0x7;
+       p.timestamp  = status & 0xffff;
+       p.generation = ohci->request_generation;
+
+       log_ar_at_event(ohci, 'R', p.speed, p.header, evt);
+
+       /*
+        * Several controllers, notably from NEC and VIA, forget to
+        * write ack_complete status at PHY packet reception.
+        */
+       if (evt == OHCI1394_evt_no_status &&
+           (p.header[0] & 0xff) == (OHCI1394_phy_tcode << 4))
+               p.ack = ACK_COMPLETE;
+
+       /*
+        * The OHCI bus reset handler synthesizes a PHY packet with
+        * the new generation number when a bus reset happens (see
+        * section 8.4.2.3).  This helps us determine when a request
+        * was received and make sure we send the response in the same
+        * generation.  We only need this for requests; for responses
+        * we use the unique tlabel for finding the matching
+        * request.
+        *
+        * Alas some chips sometimes emit bus reset packets with a
+        * wrong generation.  We set the correct generation for these
+        * at a slightly incorrect time (in bus_reset_work).
+        */
+       if (evt == OHCI1394_evt_bus_reset) {
+               if (!(ohci->quirks & QUIRK_RESET_PACKET))
+                       ohci->request_generation = (p.header[2] >> 16) & 0xff;
+       } else if (ctx == &ohci->ar_request_ctx) {
+               fw_core_handle_request(&ohci->card, &p);
+       } else {
+               fw_core_handle_response(&ohci->card, &p);
+       }
+
+       return buffer + length + 1;
+}
+
+static void *handle_ar_packets(struct ar_context *ctx, void *p, void *end)
+{
+       void *next;
+
+       while (p < end) {
+               next = handle_ar_packet(ctx, p);
+               if (!next)
+                       return p;
+               p = next;
+       }
+
+       return p;
+}
+
+static void ar_recycle_buffers(struct ar_context *ctx, unsigned int end_buffer)
+{
+       unsigned int i;
+
+       i = ar_first_buffer_index(ctx);
+       while (i != end_buffer) {
+               dma_sync_single_for_device(ctx->ohci->card.device,
+                                          ar_buffer_bus(ctx, i),
+                                          PAGE_SIZE, DMA_FROM_DEVICE);
+               ar_context_link_page(ctx, i);
+               i = ar_next_buffer_index(i);
+       }
+}
+
+static void ar_context_tasklet(unsigned long data)
+{
+       struct ar_context *ctx = (struct ar_context *)data;
+       unsigned int end_buffer_index, end_buffer_offset;
+       void *p, *end;
+
+       p = ctx->pointer;
+       if (!p)
+               return;
+
+       end_buffer_index = ar_search_last_active_buffer(ctx,
+                                                       &end_buffer_offset);
+       ar_sync_buffers_for_cpu(ctx, end_buffer_index, end_buffer_offset);
+       end = ctx->buffer + end_buffer_index * PAGE_SIZE + end_buffer_offset;
+
+       if (end_buffer_index < ar_first_buffer_index(ctx)) {
+               /*
+                * The filled part of the overall buffer wraps around; handle
+                * all packets up to the buffer end here.  If the last packet
+                * wraps around, its tail will be visible after the buffer end
+                * because the buffer start pages are mapped there again.
+                */
+               void *buffer_end = ctx->buffer + AR_BUFFERS * PAGE_SIZE;
+               p = handle_ar_packets(ctx, p, buffer_end);
+               if (p < buffer_end)
+                       goto error;
+               /* adjust p to point back into the actual buffer */
+               p -= AR_BUFFERS * PAGE_SIZE;
+       }
+
+       p = handle_ar_packets(ctx, p, end);
+       if (p != end) {
+               if (p > end)
+                       ar_context_abort(ctx, "inconsistent descriptor");
+               goto error;
+       }
+
+       ctx->pointer = p;
+       ar_recycle_buffers(ctx, end_buffer_index);
+
+       return;
+
+error:
+       ctx->pointer = NULL;
+}
+
+static int ar_context_init(struct ar_context *ctx, struct fw_ohci *ohci,
+                          unsigned int descriptors_offset, u32 regs)
+{
+       unsigned int i;
+       dma_addr_t dma_addr;
+       struct page *pages[AR_BUFFERS + AR_WRAPAROUND_PAGES];
+       struct descriptor *d;
+
+       ctx->regs        = regs;
+       ctx->ohci        = ohci;
+       tasklet_init(&ctx->tasklet, ar_context_tasklet, (unsigned long)ctx);
+
+       for (i = 0; i < AR_BUFFERS; i++) {
+               ctx->pages[i] = alloc_page(GFP_KERNEL | GFP_DMA32);
+               if (!ctx->pages[i])
+                       goto out_of_memory;
+               dma_addr = dma_map_page(ohci->card.device, ctx->pages[i],
+                                       0, PAGE_SIZE, DMA_FROM_DEVICE);
+               if (dma_mapping_error(ohci->card.device, dma_addr)) {
+                       __free_page(ctx->pages[i]);
+                       ctx->pages[i] = NULL;
+                       goto out_of_memory;
+               }
+               set_page_private(ctx->pages[i], dma_addr);
+       }
+
+       for (i = 0; i < AR_BUFFERS; i++)
+               pages[i]              = ctx->pages[i];
+       for (i = 0; i < AR_WRAPAROUND_PAGES; i++)
+               pages[AR_BUFFERS + i] = ctx->pages[i];
+       ctx->buffer = vmap(pages, ARRAY_SIZE(pages), VM_MAP, PAGE_KERNEL);
+       if (!ctx->buffer)
+               goto out_of_memory;
+
+       ctx->descriptors     = ohci->misc_buffer     + descriptors_offset;
+       ctx->descriptors_bus = ohci->misc_buffer_bus + descriptors_offset;
+
+       for (i = 0; i < AR_BUFFERS; i++) {
+               d = &ctx->descriptors[i];
+               d->req_count      = cpu_to_le16(PAGE_SIZE);
+               d->control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
+                                               DESCRIPTOR_STATUS |
+                                               DESCRIPTOR_BRANCH_ALWAYS);
+               d->data_address   = cpu_to_le32(ar_buffer_bus(ctx, i));
+               d->branch_address = cpu_to_le32(ctx->descriptors_bus +
+                       ar_next_buffer_index(i) * sizeof(struct descriptor));
+       }
+
+       return 0;
+
+out_of_memory:
+       ar_context_release(ctx);
+
+       return -ENOMEM;
+}
+
+static void ar_context_run(struct ar_context *ctx)
+{
+       unsigned int i;
+
+       for (i = 0; i < AR_BUFFERS; i++)
+               ar_context_link_page(ctx, i);
+
+       ctx->pointer = ctx->buffer;
+
+       reg_write(ctx->ohci, COMMAND_PTR(ctx->regs), ctx->descriptors_bus | 1);
+       reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN);
+}
+
+static struct descriptor *find_branch_descriptor(struct descriptor *d, int z)
+{
+       __le16 branch;
+
+       branch = d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS);
+
+       /* figure out which descriptor the branch address goes in */
+       if (z == 2 && branch == cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
+               return d;
+       else
+               return d + z - 1;
+}
+
+static void context_tasklet(unsigned long data)
+{
+       struct context *ctx = (struct context *) data;
+       struct descriptor *d, *last;
+       u32 address;
+       int z;
+       struct descriptor_buffer *desc;
+
+       desc = list_entry(ctx->buffer_list.next,
+                       struct descriptor_buffer, list);
+       last = ctx->last;
+       while (last->branch_address != 0) {
+               struct descriptor_buffer *old_desc = desc;
+               address = le32_to_cpu(last->branch_address);
+               z = address & 0xf;
+               address &= ~0xf;
+               ctx->current_bus = address;
+
+               /* If the branch address points to a buffer outside of the
+                * current buffer, advance to the next buffer. */
+               if (address < desc->buffer_bus ||
+                               address >= desc->buffer_bus + desc->used)
+                       desc = list_entry(desc->list.next,
+                                       struct descriptor_buffer, list);
+               d = desc->buffer + (address - desc->buffer_bus) / sizeof(*d);
+               last = find_branch_descriptor(d, z);
+
+               if (!ctx->callback(ctx, d, last))
+                       break;
+
+               if (old_desc != desc) {
+                       /* If we've advanced to the next buffer, move the
+                        * previous buffer to the free list. */
+                       unsigned long flags;
+                       old_desc->used = 0;
+                       spin_lock_irqsave(&ctx->ohci->lock, flags);
+                       list_move_tail(&old_desc->list, &ctx->buffer_list);
+                       spin_unlock_irqrestore(&ctx->ohci->lock, flags);
+               }
+               ctx->last = last;
+       }
+}
+
+/*
+ * Allocate a new buffer and add it to the list of free buffers for this
+ * context.  Must be called with ohci->lock held.
+ */
+static int context_add_buffer(struct context *ctx)
+{
+       struct descriptor_buffer *desc;
+       dma_addr_t uninitialized_var(bus_addr);
+       int offset;
+
+       /*
+        * 16MB of descriptors should be far more than enough for any DMA
+        * program.  This will catch run-away userspace or DoS attacks.
+        */
+       if (ctx->total_allocation >= 16*1024*1024)
+               return -ENOMEM;
+
+       desc = dma_alloc_coherent(ctx->ohci->card.device, PAGE_SIZE,
+                       &bus_addr, GFP_ATOMIC);
+       if (!desc)
+               return -ENOMEM;
+
+       offset = (void *)&desc->buffer - (void *)desc;
+       desc->buffer_size = PAGE_SIZE - offset;
+       desc->buffer_bus = bus_addr + offset;
+       desc->used = 0;
+
+       list_add_tail(&desc->list, &ctx->buffer_list);
+       ctx->total_allocation += PAGE_SIZE;
+
+       return 0;
+}
+
+static int context_init(struct context *ctx, struct fw_ohci *ohci,
+                       u32 regs, descriptor_callback_t callback)
+{
+       ctx->ohci = ohci;
+       ctx->regs = regs;
+       ctx->total_allocation = 0;
+
+       INIT_LIST_HEAD(&ctx->buffer_list);
+       if (context_add_buffer(ctx) < 0)
+               return -ENOMEM;
+
+       ctx->buffer_tail = list_entry(ctx->buffer_list.next,
+                       struct descriptor_buffer, list);
+
+       tasklet_init(&ctx->tasklet, context_tasklet, (unsigned long)ctx);
+       ctx->callback = callback;
+
+       /*
+        * We put a dummy descriptor in the buffer that has a NULL
+        * branch address and looks like it's been sent.  That way we
+        * have a descriptor to append DMA programs to.
+        */
+       memset(ctx->buffer_tail->buffer, 0, sizeof(*ctx->buffer_tail->buffer));
+       ctx->buffer_tail->buffer->control = cpu_to_le16(DESCRIPTOR_OUTPUT_LAST);
+       ctx->buffer_tail->buffer->transfer_status = cpu_to_le16(0x8011);
+       ctx->buffer_tail->used += sizeof(*ctx->buffer_tail->buffer);
+       ctx->last = ctx->buffer_tail->buffer;
+       ctx->prev = ctx->buffer_tail->buffer;
+       ctx->prev_z = 1;
+
+       return 0;
+}
+
+static void context_release(struct context *ctx)
+{
+       struct fw_card *card = &ctx->ohci->card;
+       struct descriptor_buffer *desc, *tmp;
+
+       list_for_each_entry_safe(desc, tmp, &ctx->buffer_list, list)
+               dma_free_coherent(card->device, PAGE_SIZE, desc,
+                       desc->buffer_bus -
+                       ((void *)&desc->buffer - (void *)desc));
+}
+
+/* Must be called with ohci->lock held */
+static struct descriptor *context_get_descriptors(struct context *ctx,
+                                                 int z, dma_addr_t *d_bus)
+{
+       struct descriptor *d = NULL;
+       struct descriptor_buffer *desc = ctx->buffer_tail;
+
+       if (z * sizeof(*d) > desc->buffer_size)
+               return NULL;
+
+       if (z * sizeof(*d) > desc->buffer_size - desc->used) {
+               /* No room for the descriptor in this buffer, so advance to the
+                * next one. */
+
+               if (desc->list.next == &ctx->buffer_list) {
+                       /* If there is no free buffer next in the list,
+                        * allocate one. */
+                       if (context_add_buffer(ctx) < 0)
+                               return NULL;
+               }
+               desc = list_entry(desc->list.next,
+                               struct descriptor_buffer, list);
+               ctx->buffer_tail = desc;
+       }
+
+       d = desc->buffer + desc->used / sizeof(*d);
+       memset(d, 0, z * sizeof(*d));
+       *d_bus = desc->buffer_bus + desc->used;
+
+       return d;
+}
+
+static void context_run(struct context *ctx, u32 extra)
+{
+       struct fw_ohci *ohci = ctx->ohci;
+
+       reg_write(ohci, COMMAND_PTR(ctx->regs),
+                 le32_to_cpu(ctx->last->branch_address));
+       reg_write(ohci, CONTROL_CLEAR(ctx->regs), ~0);
+       reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_RUN | extra);
+       ctx->running = true;
+       flush_writes(ohci);
+}
+
+static void context_append(struct context *ctx,
+                          struct descriptor *d, int z, int extra)
+{
+       dma_addr_t d_bus;
+       struct descriptor_buffer *desc = ctx->buffer_tail;
+       struct descriptor *d_branch;
+
+       d_bus = desc->buffer_bus + (d - desc->buffer) * sizeof(*d);
+
+       desc->used += (z + extra) * sizeof(*d);
+
+       wmb(); /* finish init of new descriptors before branch_address update */
+
+       d_branch = find_branch_descriptor(ctx->prev, ctx->prev_z);
+       d_branch->branch_address = cpu_to_le32(d_bus | z);
+
+       /*
+        * VT6306 incorrectly checks only the single descriptor at the
+        * CommandPtr when the wake bit is written, so if it's a
+        * multi-descriptor block starting with an INPUT_MORE, put a copy of
+        * the branch address in the first descriptor.
+        *
+        * Not doing this for transmit contexts since not sure how it interacts
+        * with skip addresses.
+        */
+       if (unlikely(ctx->ohci->quirks & QUIRK_IR_WAKE) &&
+           d_branch != ctx->prev &&
+           (ctx->prev->control & cpu_to_le16(DESCRIPTOR_CMD)) ==
+            cpu_to_le16(DESCRIPTOR_INPUT_MORE)) {
+               ctx->prev->branch_address = cpu_to_le32(d_bus | z);
+       }
+
+       ctx->prev = d;
+       ctx->prev_z = z;
+}
+
+static void context_stop(struct context *ctx)
+{
+       struct fw_ohci *ohci = ctx->ohci;
+       u32 reg;
+       int i;
+
+       reg_write(ohci, CONTROL_CLEAR(ctx->regs), CONTEXT_RUN);
+       ctx->running = false;
+
+       for (i = 0; i < 1000; i++) {
+               reg = reg_read(ohci, CONTROL_SET(ctx->regs));
+               if ((reg & CONTEXT_ACTIVE) == 0)
+                       return;
+
+               if (i)
+                       udelay(10);
+       }
+       ohci_err(ohci, "DMA context still active (0x%08x)\n", reg);
+}
+
+struct driver_data {
+       u8 inline_data[8];
+       struct fw_packet *packet;
+};
+
+/*
+ * This function apppends a packet to the DMA queue for transmission.
+ * Must always be called with the ochi->lock held to ensure proper
+ * generation handling and locking around packet queue manipulation.
+ */
+static int at_context_queue_packet(struct context *ctx,
+                                  struct fw_packet *packet)
+{
+       struct fw_ohci *ohci = ctx->ohci;
+       dma_addr_t d_bus, uninitialized_var(payload_bus);
+       struct driver_data *driver_data;
+       struct descriptor *d, *last;
+       __le32 *header;
+       int z, tcode;
+
+       d = context_get_descriptors(ctx, 4, &d_bus);
+       if (d == NULL) {
+               packet->ack = RCODE_SEND_ERROR;
+               return -1;
+       }
+
+       d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
+       d[0].res_count = cpu_to_le16(packet->timestamp);
+
+       /*
+        * The DMA format for asynchronous link packets is different
+        * from the IEEE1394 layout, so shift the fields around
+        * accordingly.
+        */
+
+       tcode = (packet->header[0] >> 4) & 0x0f;
+       header = (__le32 *) &d[1];
+       switch (tcode) {
+       case TCODE_WRITE_QUADLET_REQUEST:
+       case TCODE_WRITE_BLOCK_REQUEST:
+       case TCODE_WRITE_RESPONSE:
+       case TCODE_READ_QUADLET_REQUEST:
+       case TCODE_READ_BLOCK_REQUEST:
+       case TCODE_READ_QUADLET_RESPONSE:
+       case TCODE_READ_BLOCK_RESPONSE:
+       case TCODE_LOCK_REQUEST:
+       case TCODE_LOCK_RESPONSE:
+               header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
+                                       (packet->speed << 16));
+               header[1] = cpu_to_le32((packet->header[1] & 0xffff) |
+                                       (packet->header[0] & 0xffff0000));
+               header[2] = cpu_to_le32(packet->header[2]);
+
+               if (TCODE_IS_BLOCK_PACKET(tcode))
+                       header[3] = cpu_to_le32(packet->header[3]);
+               else
+                       header[3] = (__force __le32) packet->header[3];
+
+               d[0].req_count = cpu_to_le16(packet->header_length);
+               break;
+
+       case TCODE_LINK_INTERNAL:
+               header[0] = cpu_to_le32((OHCI1394_phy_tcode << 4) |
+                                       (packet->speed << 16));
+               header[1] = cpu_to_le32(packet->header[1]);
+               header[2] = cpu_to_le32(packet->header[2]);
+               d[0].req_count = cpu_to_le16(12);
+
+               if (is_ping_packet(&packet->header[1]))
+                       d[0].control |= cpu_to_le16(DESCRIPTOR_PING);
+               break;
+
+       case TCODE_STREAM_DATA:
+               header[0] = cpu_to_le32((packet->header[0] & 0xffff) |
+                                       (packet->speed << 16));
+               header[1] = cpu_to_le32(packet->header[0] & 0xffff0000);
+               d[0].req_count = cpu_to_le16(8);
+               break;
+
+       default:
+               /* BUG(); */
+               packet->ack = RCODE_SEND_ERROR;
+               return -1;
+       }
+
+       BUILD_BUG_ON(sizeof(struct driver_data) > sizeof(struct descriptor));
+       driver_data = (struct driver_data *) &d[3];
+       driver_data->packet = packet;
+       packet->driver_data = driver_data;
+
+       if (packet->payload_length > 0) {
+               if (packet->payload_length > sizeof(driver_data->inline_data)) {
+                       payload_bus = dma_map_single(ohci->card.device,
+                                                    packet->payload,
+                                                    packet->payload_length,
+                                                    DMA_TO_DEVICE);
+                       if (dma_mapping_error(ohci->card.device, payload_bus)) {
+                               packet->ack = RCODE_SEND_ERROR;
+                               return -1;
+                       }
+                       packet->payload_bus     = payload_bus;
+                       packet->payload_mapped  = true;
+               } else {
+                       memcpy(driver_data->inline_data, packet->payload,
+                              packet->payload_length);
+                       payload_bus = d_bus + 3 * sizeof(*d);
+               }
+
+               d[2].req_count    = cpu_to_le16(packet->payload_length);
+               d[2].data_address = cpu_to_le32(payload_bus);
+               last = &d[2];
+               z = 3;
+       } else {
+               last = &d[0];
+               z = 2;
+       }
+
+       last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
+                                    DESCRIPTOR_IRQ_ALWAYS |
+                                    DESCRIPTOR_BRANCH_ALWAYS);
+
+       /* FIXME: Document how the locking works. */
+       if (ohci->generation != packet->generation) {
+               if (packet->payload_mapped)
+                       dma_unmap_single(ohci->card.device, payload_bus,
+                                        packet->payload_length, DMA_TO_DEVICE);
+               packet->ack = RCODE_GENERATION;
+               return -1;
+       }
+
+       context_append(ctx, d, z, 4 - z);
+
+       if (ctx->running)
+               reg_write(ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
+       else
+               context_run(ctx, 0);
+
+       return 0;
+}
+
+static void at_context_flush(struct context *ctx)
+{
+       tasklet_disable(&ctx->tasklet);
+
+       ctx->flushing = true;
+       context_tasklet((unsigned long)ctx);
+       ctx->flushing = false;
+
+       tasklet_enable(&ctx->tasklet);
+}
+
+static int handle_at_packet(struct context *context,
+                           struct descriptor *d,
+                           struct descriptor *last)
+{
+       struct driver_data *driver_data;
+       struct fw_packet *packet;
+       struct fw_ohci *ohci = context->ohci;
+       int evt;
+
+       if (last->transfer_status == 0 && !context->flushing)
+               /* This descriptor isn't done yet, stop iteration. */
+               return 0;
+
+       driver_data = (struct driver_data *) &d[3];
+       packet = driver_data->packet;
+       if (packet == NULL)
+               /* This packet was cancelled, just continue. */
+               return 1;
+
+       if (packet->payload_mapped)
+               dma_unmap_single(ohci->card.device, packet->payload_bus,
+                                packet->payload_length, DMA_TO_DEVICE);
+
+       evt = le16_to_cpu(last->transfer_status) & 0x1f;
+       packet->timestamp = le16_to_cpu(last->res_count);
+
+       log_ar_at_event(ohci, 'T', packet->speed, packet->header, evt);
+
+       switch (evt) {
+       case OHCI1394_evt_timeout:
+               /* Async response transmit timed out. */
+               packet->ack = RCODE_CANCELLED;
+               break;
+
+       case OHCI1394_evt_flushed:
+               /*
+                * The packet was flushed should give same error as
+                * when we try to use a stale generation count.
+                */
+               packet->ack = RCODE_GENERATION;
+               break;
+
+       case OHCI1394_evt_missing_ack:
+               if (context->flushing)
+                       packet->ack = RCODE_GENERATION;
+               else {
+                       /*
+                        * Using a valid (current) generation count, but the
+                        * node is not on the bus or not sending acks.
+                        */
+                       packet->ack = RCODE_NO_ACK;
+               }
+               break;
+
+       case ACK_COMPLETE + 0x10:
+       case ACK_PENDING + 0x10:
+       case ACK_BUSY_X + 0x10:
+       case ACK_BUSY_A + 0x10:
+       case ACK_BUSY_B + 0x10:
+       case ACK_DATA_ERROR + 0x10:
+       case ACK_TYPE_ERROR + 0x10:
+               packet->ack = evt - 0x10;
+               break;
+
+       case OHCI1394_evt_no_status:
+               if (context->flushing) {
+                       packet->ack = RCODE_GENERATION;
+                       break;
+               }
+               /* fall through */
+
+       default:
+               packet->ack = RCODE_SEND_ERROR;
+               break;
+       }
+
+       packet->callback(packet, &ohci->card, packet->ack);
+
+       return 1;
+}
+
+#define HEADER_GET_DESTINATION(q)      (((q) >> 16) & 0xffff)
+#define HEADER_GET_TCODE(q)            (((q) >> 4) & 0x0f)
+#define HEADER_GET_OFFSET_HIGH(q)      (((q) >> 0) & 0xffff)
+#define HEADER_GET_DATA_LENGTH(q)      (((q) >> 16) & 0xffff)
+#define HEADER_GET_EXTENDED_TCODE(q)   (((q) >> 0) & 0xffff)
+
+static void handle_local_rom(struct fw_ohci *ohci,
+                            struct fw_packet *packet, u32 csr)
+{
+       struct fw_packet response;
+       int tcode, length, i;
+
+       tcode = HEADER_GET_TCODE(packet->header[0]);
+       if (TCODE_IS_BLOCK_PACKET(tcode))
+               length = HEADER_GET_DATA_LENGTH(packet->header[3]);
+       else
+               length = 4;
+
+       i = csr - CSR_CONFIG_ROM;
+       if (i + length > CONFIG_ROM_SIZE) {
+               fw_fill_response(&response, packet->header,
+                                RCODE_ADDRESS_ERROR, NULL, 0);
+       } else if (!TCODE_IS_READ_REQUEST(tcode)) {
+               fw_fill_response(&response, packet->header,
+                                RCODE_TYPE_ERROR, NULL, 0);
+       } else {
+               fw_fill_response(&response, packet->header, RCODE_COMPLETE,
+                                (void *) ohci->config_rom + i, length);
+       }
+
+       fw_core_handle_response(&ohci->card, &response);
+}
+
+static void handle_local_lock(struct fw_ohci *ohci,
+                             struct fw_packet *packet, u32 csr)
+{
+       struct fw_packet response;
+       int tcode, length, ext_tcode, sel, try;
+       __be32 *payload, lock_old;
+       u32 lock_arg, lock_data;
+
+       tcode = HEADER_GET_TCODE(packet->header[0]);
+       length = HEADER_GET_DATA_LENGTH(packet->header[3]);
+       payload = packet->payload;
+       ext_tcode = HEADER_GET_EXTENDED_TCODE(packet->header[3]);
+
+       if (tcode == TCODE_LOCK_REQUEST &&
+           ext_tcode == EXTCODE_COMPARE_SWAP && length == 8) {
+               lock_arg = be32_to_cpu(payload[0]);
+               lock_data = be32_to_cpu(payload[1]);
+       } else if (tcode == TCODE_READ_QUADLET_REQUEST) {
+               lock_arg = 0;
+               lock_data = 0;
+       } else {
+               fw_fill_response(&response, packet->header,
+                                RCODE_TYPE_ERROR, NULL, 0);
+               goto out;
+       }
+
+       sel = (csr - CSR_BUS_MANAGER_ID) / 4;
+       reg_write(ohci, OHCI1394_CSRData, lock_data);
+       reg_write(ohci, OHCI1394_CSRCompareData, lock_arg);
+       reg_write(ohci, OHCI1394_CSRControl, sel);
+
+       for (try = 0; try < 20; try++)
+               if (reg_read(ohci, OHCI1394_CSRControl) & 0x80000000) {
+                       lock_old = cpu_to_be32(reg_read(ohci,
+                                                       OHCI1394_CSRData));
+                       fw_fill_response(&response, packet->header,
+                                        RCODE_COMPLETE,
+                                        &lock_old, sizeof(lock_old));
+                       goto out;
+               }
+
+       ohci_err(ohci, "swap not done (CSR lock timeout)\n");
+       fw_fill_response(&response, packet->header, RCODE_BUSY, NULL, 0);
+
+ out:
+       fw_core_handle_response(&ohci->card, &response);
+}
+
+static void handle_local_request(struct context *ctx, struct fw_packet *packet)
+{
+       u64 offset, csr;
+
+       if (ctx == &ctx->ohci->at_request_ctx) {
+               packet->ack = ACK_PENDING;
+               packet->callback(packet, &ctx->ohci->card, packet->ack);
+       }
+
+       offset =
+               ((unsigned long long)
+                HEADER_GET_OFFSET_HIGH(packet->header[1]) << 32) |
+               packet->header[2];
+       csr = offset - CSR_REGISTER_BASE;
+
+       /* Handle config rom reads. */
+       if (csr >= CSR_CONFIG_ROM && csr < CSR_CONFIG_ROM_END)
+               handle_local_rom(ctx->ohci, packet, csr);
+       else switch (csr) {
+       case CSR_BUS_MANAGER_ID:
+       case CSR_BANDWIDTH_AVAILABLE:
+       case CSR_CHANNELS_AVAILABLE_HI:
+       case CSR_CHANNELS_AVAILABLE_LO:
+               handle_local_lock(ctx->ohci, packet, csr);
+               break;
+       default:
+               if (ctx == &ctx->ohci->at_request_ctx)
+                       fw_core_handle_request(&ctx->ohci->card, packet);
+               else
+                       fw_core_handle_response(&ctx->ohci->card, packet);
+               break;
+       }
+
+       if (ctx == &ctx->ohci->at_response_ctx) {
+               packet->ack = ACK_COMPLETE;
+               packet->callback(packet, &ctx->ohci->card, packet->ack);
+       }
+}
+
+static void at_context_transmit(struct context *ctx, struct fw_packet *packet)
+{
+       unsigned long flags;
+       int ret;
+
+       spin_lock_irqsave(&ctx->ohci->lock, flags);
+
+       if (HEADER_GET_DESTINATION(packet->header[0]) == ctx->ohci->node_id &&
+           ctx->ohci->generation == packet->generation) {
+               spin_unlock_irqrestore(&ctx->ohci->lock, flags);
+               handle_local_request(ctx, packet);
+               return;
+       }
+
+       ret = at_context_queue_packet(ctx, packet);
+       spin_unlock_irqrestore(&ctx->ohci->lock, flags);
+
+       if (ret < 0)
+               packet->callback(packet, &ctx->ohci->card, packet->ack);
+
+}
+
+static void detect_dead_context(struct fw_ohci *ohci,
+                               const char *name, unsigned int regs)
+{
+       u32 ctl;
+
+       ctl = reg_read(ohci, CONTROL_SET(regs));
+       if (ctl & CONTEXT_DEAD)
+               ohci_err(ohci, "DMA context %s has stopped, error code: %s\n",
+                       name, evts[ctl & 0x1f]);
+}
+
+static void handle_dead_contexts(struct fw_ohci *ohci)
+{
+       unsigned int i;
+       char name[8];
+
+       detect_dead_context(ohci, "ATReq", OHCI1394_AsReqTrContextBase);
+       detect_dead_context(ohci, "ATRsp", OHCI1394_AsRspTrContextBase);
+       detect_dead_context(ohci, "ARReq", OHCI1394_AsReqRcvContextBase);
+       detect_dead_context(ohci, "ARRsp", OHCI1394_AsRspRcvContextBase);
+       for (i = 0; i < 32; ++i) {
+               if (!(ohci->it_context_support & (1 << i)))
+                       continue;
+               sprintf(name, "IT%u", i);
+               detect_dead_context(ohci, name, OHCI1394_IsoXmitContextBase(i));
+       }
+       for (i = 0; i < 32; ++i) {
+               if (!(ohci->ir_context_support & (1 << i)))
+                       continue;
+               sprintf(name, "IR%u", i);
+               detect_dead_context(ohci, name, OHCI1394_IsoRcvContextBase(i));
+       }
+       /* TODO: maybe try to flush and restart the dead contexts */
+}
+
+static u32 cycle_timer_ticks(u32 cycle_timer)
+{
+       u32 ticks;
+
+       ticks = cycle_timer & 0xfff;
+       ticks += 3072 * ((cycle_timer >> 12) & 0x1fff);
+       ticks += (3072 * 8000) * (cycle_timer >> 25);
+
+       return ticks;
+}
+
+/*
+ * Some controllers exhibit one or more of the following bugs when updating the
+ * iso cycle timer register:
+ *  - When the lowest six bits are wrapping around to zero, a read that happens
+ *    at the same time will return garbage in the lowest ten bits.
+ *  - When the cycleOffset field wraps around to zero, the cycleCount field is
+ *    not incremented for about 60 ns.
+ *  - Occasionally, the entire register reads zero.
+ *
+ * To catch these, we read the register three times and ensure that the
+ * difference between each two consecutive reads is approximately the same, i.e.
+ * less than twice the other.  Furthermore, any negative difference indicates an
+ * error.  (A PCI read should take at least 20 ticks of the 24.576 MHz timer to
+ * execute, so we have enough precision to compute the ratio of the differences.)
+ */
+static u32 get_cycle_time(struct fw_ohci *ohci)
+{
+       u32 c0, c1, c2;
+       u32 t0, t1, t2;
+       s32 diff01, diff12;
+       int i;
+
+       c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
+
+       if (ohci->quirks & QUIRK_CYCLE_TIMER) {
+               i = 0;
+               c1 = c2;
+               c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
+               do {
+                       c0 = c1;
+                       c1 = c2;
+                       c2 = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
+                       t0 = cycle_timer_ticks(c0);
+                       t1 = cycle_timer_ticks(c1);
+                       t2 = cycle_timer_ticks(c2);
+                       diff01 = t1 - t0;
+                       diff12 = t2 - t1;
+               } while ((diff01 <= 0 || diff12 <= 0 ||
+                         diff01 / diff12 >= 2 || diff12 / diff01 >= 2)
+                        && i++ < 20);
+       }
+
+       return c2;
+}
+
+/*
+ * This function has to be called at least every 64 seconds.  The bus_time
+ * field stores not only the upper 25 bits of the BUS_TIME register but also
+ * the most significant bit of the cycle timer in bit 6 so that we can detect
+ * changes in this bit.
+ */
+static u32 update_bus_time(struct fw_ohci *ohci)
+{
+       u32 cycle_time_seconds = get_cycle_time(ohci) >> 25;
+
+       if (unlikely(!ohci->bus_time_running)) {
+               reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_cycle64Seconds);
+               ohci->bus_time = (lower_32_bits(get_seconds()) & ~0x7f) |
+                                (cycle_time_seconds & 0x40);
+               ohci->bus_time_running = true;
+       }
+
+       if ((ohci->bus_time & 0x40) != (cycle_time_seconds & 0x40))
+               ohci->bus_time += 0x40;
+
+       return ohci->bus_time | cycle_time_seconds;
+}
+
+static int get_status_for_port(struct fw_ohci *ohci, int port_index)
+{
+       int reg;
+
+       mutex_lock(&ohci->phy_reg_mutex);
+       reg = write_phy_reg(ohci, 7, port_index);
+       if (reg >= 0)
+               reg = read_phy_reg(ohci, 8);
+       mutex_unlock(&ohci->phy_reg_mutex);
+       if (reg < 0)
+               return reg;
+
+       switch (reg & 0x0f) {
+       case 0x06:
+               return 2;       /* is child node (connected to parent node) */
+       case 0x0e:
+               return 3;       /* is parent node (connected to child node) */
+       }
+       return 1;               /* not connected */
+}
+
+static int get_self_id_pos(struct fw_ohci *ohci, u32 self_id,
+       int self_id_count)
+{
+       int i;
+       u32 entry;
+
+       for (i = 0; i < self_id_count; i++) {
+               entry = ohci->self_id_buffer[i];
+               if ((self_id & 0xff000000) == (entry & 0xff000000))
+                       return -1;
+               if ((self_id & 0xff000000) < (entry & 0xff000000))
+                       return i;
+       }
+       return i;
+}
+
+static int initiated_reset(struct fw_ohci *ohci)
+{
+       int reg;
+       int ret = 0;
+
+       mutex_lock(&ohci->phy_reg_mutex);
+       reg = write_phy_reg(ohci, 7, 0xe0); /* Select page 7 */
+       if (reg >= 0) {
+               reg = read_phy_reg(ohci, 8);
+               reg |= 0x40;
+               reg = write_phy_reg(ohci, 8, reg); /* set PMODE bit */
+               if (reg >= 0) {
+                       reg = read_phy_reg(ohci, 12); /* read register 12 */
+                       if (reg >= 0) {
+                               if ((reg & 0x08) == 0x08) {
+                                       /* bit 3 indicates "initiated reset" */
+                                       ret = 0x2;
+                               }
+                       }
+               }
+       }
+       mutex_unlock(&ohci->phy_reg_mutex);
+       return ret;
+}
+
+/*
+ * TI TSB82AA2B and TSB12LV26 do not receive the selfID of a locally
+ * attached TSB41BA3D phy; see http://www.ti.com/litv/pdf/sllz059.
+ * Construct the selfID from phy register contents.
+ */
+static int find_and_insert_self_id(struct fw_ohci *ohci, int self_id_count)
+{
+       int reg, i, pos, status;
+       /* link active 1, speed 3, bridge 0, contender 1, more packets 0 */
+       u32 self_id = 0x8040c800;
+
+       reg = reg_read(ohci, OHCI1394_NodeID);
+       if (!(reg & OHCI1394_NodeID_idValid)) {
+               ohci_notice(ohci,
+                           "node ID not valid, new bus reset in progress\n");
+               return -EBUSY;
+       }
+       self_id |= ((reg & 0x3f) << 24); /* phy ID */
+
+       reg = ohci_read_phy_reg(&ohci->card, 4);
+       if (reg < 0)
+               return reg;
+       self_id |= ((reg & 0x07) << 8); /* power class */
+
+       reg = ohci_read_phy_reg(&ohci->card, 1);
+       if (reg < 0)
+               return reg;
+       self_id |= ((reg & 0x3f) << 16); /* gap count */
+
+       for (i = 0; i < 3; i++) {
+               status = get_status_for_port(ohci, i);
+               if (status < 0)
+                       return status;
+               self_id |= ((status & 0x3) << (6 - (i * 2)));
+       }
+
+       self_id |= initiated_reset(ohci);
+
+       pos = get_self_id_pos(ohci, self_id, self_id_count);
+       if (pos >= 0) {
+               memmove(&(ohci->self_id_buffer[pos+1]),
+                       &(ohci->self_id_buffer[pos]),
+                       (self_id_count - pos) * sizeof(*ohci->self_id_buffer));
+               ohci->self_id_buffer[pos] = self_id;
+               self_id_count++;
+       }
+       return self_id_count;
+}
+
+static void bus_reset_work(struct work_struct *work)
+{
+       struct fw_ohci *ohci =
+               container_of(work, struct fw_ohci, bus_reset_work);
+       int self_id_count, generation, new_generation, i, j;
+       u32 reg;
+       void *free_rom = NULL;
+       dma_addr_t free_rom_bus = 0;
+       bool is_new_root;
+
+       reg = reg_read(ohci, OHCI1394_NodeID);
+       if (!(reg & OHCI1394_NodeID_idValid)) {
+               ohci_notice(ohci,
+                           "node ID not valid, new bus reset in progress\n");
+               return;
+       }
+       if ((reg & OHCI1394_NodeID_nodeNumber) == 63) {
+               ohci_notice(ohci, "malconfigured bus\n");
+               return;
+       }
+       ohci->node_id = reg & (OHCI1394_NodeID_busNumber |
+                              OHCI1394_NodeID_nodeNumber);
+
+       is_new_root = (reg & OHCI1394_NodeID_root) != 0;
+       if (!(ohci->is_root && is_new_root))
+               reg_write(ohci, OHCI1394_LinkControlSet,
+                         OHCI1394_LinkControl_cycleMaster);
+       ohci->is_root = is_new_root;
+
+       reg = reg_read(ohci, OHCI1394_SelfIDCount);
+       if (reg & OHCI1394_SelfIDCount_selfIDError) {
+               ohci_notice(ohci, "self ID receive error\n");
+               return;
+       }
+       /*
+        * The count in the SelfIDCount register is the number of
+        * bytes in the self ID receive buffer.  Since we also receive
+        * the inverted quadlets and a header quadlet, we shift one
+        * bit extra to get the actual number of self IDs.
+        */
+       self_id_count = (reg >> 3) & 0xff;
+
+       if (self_id_count > 252) {
+               ohci_notice(ohci, "bad selfIDSize (%08x)\n", reg);
+               return;
+       }
+
+       generation = (cond_le32_to_cpu(ohci->self_id[0]) >> 16) & 0xff;
+       rmb();
+
+       for (i = 1, j = 0; j < self_id_count; i += 2, j++) {
+               u32 id  = cond_le32_to_cpu(ohci->self_id[i]);
+               u32 id2 = cond_le32_to_cpu(ohci->self_id[i + 1]);
+
+               if (id != ~id2) {
+                       /*
+                        * If the invalid data looks like a cycle start packet,
+                        * it's likely to be the result of the cycle master
+                        * having a wrong gap count.  In this case, the self IDs
+                        * so far are valid and should be processed so that the
+                        * bus manager can then correct the gap count.
+                        */
+                       if (id == 0xffff008f) {
+                               ohci_notice(ohci, "ignoring spurious self IDs\n");
+                               self_id_count = j;
+                               break;
+                       }
+
+                       ohci_notice(ohci, "bad self ID %d/%d (%08x != ~%08x)\n",
+                                   j, self_id_count, id, id2);
+                       return;
+               }
+               ohci->self_id_buffer[j] = id;
+       }
+
+       if (ohci->quirks & QUIRK_TI_SLLZ059) {
+               self_id_count = find_and_insert_self_id(ohci, self_id_count);
+               if (self_id_count < 0) {
+                       ohci_notice(ohci,
+                                   "could not construct local self ID\n");
+                       return;
+               }
+       }
+
+       if (self_id_count == 0) {
+               ohci_notice(ohci, "no self IDs\n");
+               return;
+       }
+       rmb();
+
+       /*
+        * Check the consistency of the self IDs we just read.  The
+        * problem we face is that a new bus reset can start while we
+        * read out the self IDs from the DMA buffer. If this happens,
+        * the DMA buffer will be overwritten with new self IDs and we
+        * will read out inconsistent data.  The OHCI specification
+        * (section 11.2) recommends a technique similar to
+        * linux/seqlock.h, where we remember the generation of the
+        * self IDs in the buffer before reading them out and compare
+        * it to the current generation after reading them out.  If
+        * the two generations match we know we have a consistent set
+        * of self IDs.
+        */
+
+       new_generation = (reg_read(ohci, OHCI1394_SelfIDCount) >> 16) & 0xff;
+       if (new_generation != generation) {
+               ohci_notice(ohci, "new bus reset, discarding self ids\n");
+               return;
+       }
+
+       /* FIXME: Document how the locking works. */
+       spin_lock_irq(&ohci->lock);
+
+       ohci->generation = -1; /* prevent AT packet queueing */
+       context_stop(&ohci->at_request_ctx);
+       context_stop(&ohci->at_response_ctx);
+
+       spin_unlock_irq(&ohci->lock);
+
+       /*
+        * Per OHCI 1.2 draft, clause 7.2.3.3, hardware may leave unsent
+        * packets in the AT queues and software needs to drain them.
+        * Some OHCI 1.1 controllers (JMicron) apparently require this too.
+        */
+       at_context_flush(&ohci->at_request_ctx);
+       at_context_flush(&ohci->at_response_ctx);
+
+       spin_lock_irq(&ohci->lock);
+
+       ohci->generation = generation;
+       reg_write(ohci, OHCI1394_IntEventClear, OHCI1394_busReset);
+
+       if (ohci->quirks & QUIRK_RESET_PACKET)
+               ohci->request_generation = generation;
+
+       /*
+        * This next bit is unrelated to the AT context stuff but we
+        * have to do it under the spinlock also.  If a new config rom
+        * was set up before this reset, the old one is now no longer
+        * in use and we can free it. Update the config rom pointers
+        * to point to the current config rom and clear the
+        * next_config_rom pointer so a new update can take place.
+        */
+
+       if (ohci->next_config_rom != NULL) {
+               if (ohci->next_config_rom != ohci->config_rom) {
+                       free_rom      = ohci->config_rom;
+                       free_rom_bus  = ohci->config_rom_bus;
+               }
+               ohci->config_rom      = ohci->next_config_rom;
+               ohci->config_rom_bus  = ohci->next_config_rom_bus;
+               ohci->next_config_rom = NULL;
+
+               /*
+                * Restore config_rom image and manually update
+                * config_rom registers.  Writing the header quadlet
+                * will indicate that the config rom is ready, so we
+                * do that last.
+                */
+               reg_write(ohci, OHCI1394_BusOptions,
+                         be32_to_cpu(ohci->config_rom[2]));
+               ohci->config_rom[0] = ohci->next_header;
+               reg_write(ohci, OHCI1394_ConfigROMhdr,
+                         be32_to_cpu(ohci->next_header));
+       }
+
+       if (param_remote_dma) {
+               reg_write(ohci, OHCI1394_PhyReqFilterHiSet, ~0);
+               reg_write(ohci, OHCI1394_PhyReqFilterLoSet, ~0);
+       }
+
+       spin_unlock_irq(&ohci->lock);
+
+       if (free_rom)
+               dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
+                                 free_rom, free_rom_bus);
+
+       log_selfids(ohci, generation, self_id_count);
+
+       fw_core_handle_bus_reset(&ohci->card, ohci->node_id, generation,
+                                self_id_count, ohci->self_id_buffer,
+                                ohci->csr_state_setclear_abdicate);
+       ohci->csr_state_setclear_abdicate = false;
+}
+
+static irqreturn_t irq_handler(int irq, void *data)
+{
+       struct fw_ohci *ohci = data;
+       u32 event, iso_event;
+       int i;
+
+       event = reg_read(ohci, OHCI1394_IntEventClear);
+
+       if (!event || !~event)
+               return IRQ_NONE;
+
+       /*
+        * busReset and postedWriteErr must not be cleared yet
+        * (OHCI 1.1 clauses 7.2.3.2 and 13.2.8.1)
+        */
+       reg_write(ohci, OHCI1394_IntEventClear,
+                 event & ~(OHCI1394_busReset | OHCI1394_postedWriteErr));
+       log_irqs(ohci, event);
+
+       if (event & OHCI1394_selfIDComplete)
+               queue_work(selfid_workqueue, &ohci->bus_reset_work);
+
+       if (event & OHCI1394_RQPkt)
+               tasklet_schedule(&ohci->ar_request_ctx.tasklet);
+
+       if (event & OHCI1394_RSPkt)
+               tasklet_schedule(&ohci->ar_response_ctx.tasklet);
+
+       if (event & OHCI1394_reqTxComplete)
+               tasklet_schedule(&ohci->at_request_ctx.tasklet);
+
+       if (event & OHCI1394_respTxComplete)
+               tasklet_schedule(&ohci->at_response_ctx.tasklet);
+
+       if (event & OHCI1394_isochRx) {
+               iso_event = reg_read(ohci, OHCI1394_IsoRecvIntEventClear);
+               reg_write(ohci, OHCI1394_IsoRecvIntEventClear, iso_event);
+
+               while (iso_event) {
+                       i = ffs(iso_event) - 1;
+                       tasklet_schedule(
+                               &ohci->ir_context_list[i].context.tasklet);
+                       iso_event &= ~(1 << i);
+               }
+       }
+
+       if (event & OHCI1394_isochTx) {
+               iso_event = reg_read(ohci, OHCI1394_IsoXmitIntEventClear);
+               reg_write(ohci, OHCI1394_IsoXmitIntEventClear, iso_event);
+
+               while (iso_event) {
+                       i = ffs(iso_event) - 1;
+                       tasklet_schedule(
+                               &ohci->it_context_list[i].context.tasklet);
+                       iso_event &= ~(1 << i);
+               }
+       }
+
+       if (unlikely(event & OHCI1394_regAccessFail))
+               ohci_err(ohci, "register access failure\n");
+
+       if (unlikely(event & OHCI1394_postedWriteErr)) {
+               reg_read(ohci, OHCI1394_PostedWriteAddressHi);
+               reg_read(ohci, OHCI1394_PostedWriteAddressLo);
+               reg_write(ohci, OHCI1394_IntEventClear,
+                         OHCI1394_postedWriteErr);
+               if (printk_ratelimit())
+                       ohci_err(ohci, "PCI posted write error\n");
+       }
+
+       if (unlikely(event & OHCI1394_cycleTooLong)) {
+               if (printk_ratelimit())
+                       ohci_notice(ohci, "isochronous cycle too long\n");
+               reg_write(ohci, OHCI1394_LinkControlSet,
+                         OHCI1394_LinkControl_cycleMaster);
+       }
+
+       if (unlikely(event & OHCI1394_cycleInconsistent)) {
+               /*
+                * We need to clear this event bit in order to make
+                * cycleMatch isochronous I/O work.  In theory we should
+                * stop active cycleMatch iso contexts now and restart
+                * them at least two cycles later.  (FIXME?)
+                */
+               if (printk_ratelimit())
+                       ohci_notice(ohci, "isochronous cycle inconsistent\n");
+       }
+
+       if (unlikely(event & OHCI1394_unrecoverableError))
+               handle_dead_contexts(ohci);
+
+       if (event & OHCI1394_cycle64Seconds) {
+               spin_lock(&ohci->lock);
+               update_bus_time(ohci);
+               spin_unlock(&ohci->lock);
+       } else
+               flush_writes(ohci);
+
+       return IRQ_HANDLED;
+}
+
+static int software_reset(struct fw_ohci *ohci)
+{
+       u32 val;
+       int i;
+
+       reg_write(ohci, OHCI1394_HCControlSet, OHCI1394_HCControl_softReset);
+       for (i = 0; i < 500; i++) {
+               val = reg_read(ohci, OHCI1394_HCControlSet);
+               if (!~val)
+                       return -ENODEV; /* Card was ejected. */
+
+               if (!(val & OHCI1394_HCControl_softReset))
+                       return 0;
+
+               msleep(1);
+       }
+
+       return -EBUSY;
+}
+
+static void copy_config_rom(__be32 *dest, const __be32 *src, size_t length)
+{
+       size_t size = length * 4;
+
+       memcpy(dest, src, size);
+       if (size < CONFIG_ROM_SIZE)
+               memset(&dest[length], 0, CONFIG_ROM_SIZE - size);
+}
+
+static int configure_1394a_enhancements(struct fw_ohci *ohci)
+{
+       bool enable_1394a;
+       int ret, clear, set, offset;
+
+       /* Check if the driver should configure link and PHY. */
+       if (!(reg_read(ohci, OHCI1394_HCControlSet) &
+             OHCI1394_HCControl_programPhyEnable))
+               return 0;
+
+       /* Paranoia: check whether the PHY supports 1394a, too. */
+       enable_1394a = false;
+       ret = read_phy_reg(ohci, 2);
+       if (ret < 0)
+               return ret;
+       if ((ret & PHY_EXTENDED_REGISTERS) == PHY_EXTENDED_REGISTERS) {
+               ret = read_paged_phy_reg(ohci, 1, 8);
+               if (ret < 0)
+                       return ret;
+               if (ret >= 1)
+                       enable_1394a = true;
+       }
+
+       if (ohci->quirks & QUIRK_NO_1394A)
+               enable_1394a = false;
+
+       /* Configure PHY and link consistently. */
+       if (enable_1394a) {
+               clear = 0;
+               set = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
+       } else {
+               clear = PHY_ENABLE_ACCEL | PHY_ENABLE_MULTI;
+               set = 0;
+       }
+       ret = update_phy_reg(ohci, 5, clear, set);
+       if (ret < 0)
+               return ret;
+
+       if (enable_1394a)
+               offset = OHCI1394_HCControlSet;
+       else
+               offset = OHCI1394_HCControlClear;
+       reg_write(ohci, offset, OHCI1394_HCControl_aPhyEnhanceEnable);
+
+       /* Clean up: configuration has been taken care of. */
+       reg_write(ohci, OHCI1394_HCControlClear,
+                 OHCI1394_HCControl_programPhyEnable);
+
+       return 0;
+}
+
+static int probe_tsb41ba3d(struct fw_ohci *ohci)
+{
+       /* TI vendor ID = 0x080028, TSB41BA3D product ID = 0x833005 (sic) */
+       static const u8 id[] = { 0x08, 0x00, 0x28, 0x83, 0x30, 0x05, };
+       int reg, i;
+
+       reg = read_phy_reg(ohci, 2);
+       if (reg < 0)
+               return reg;
+       if ((reg & PHY_EXTENDED_REGISTERS) != PHY_EXTENDED_REGISTERS)
+               return 0;
+
+       for (i = ARRAY_SIZE(id) - 1; i >= 0; i--) {
+               reg = read_paged_phy_reg(ohci, 1, i + 10);
+               if (reg < 0)
+                       return reg;
+               if (reg != id[i])
+                       return 0;
+       }
+       return 1;
+}
+
+static int ohci_enable(struct fw_card *card,
+                      const __be32 *config_rom, size_t length)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+       u32 lps, version, irqs;
+       int i, ret;
+
+       if (software_reset(ohci)) {
+               ohci_err(ohci, "failed to reset ohci card\n");
+               return -EBUSY;
+       }
+
+       /*
+        * Now enable LPS, which we need in order to start accessing
+        * most of the registers.  In fact, on some cards (ALI M5251),
+        * accessing registers in the SClk domain without LPS enabled
+        * will lock up the machine.  Wait 50msec to make sure we have
+        * full link enabled.  However, with some cards (well, at least
+        * a JMicron PCIe card), we have to try again sometimes.
+        *
+        * TI TSB82AA2 + TSB81BA3(A) cards signal LPS enabled early but
+        * cannot actually use the phy at that time.  These need tens of
+        * millisecods pause between LPS write and first phy access too.
+        */
+
+       reg_write(ohci, OHCI1394_HCControlSet,
+                 OHCI1394_HCControl_LPS |
+                 OHCI1394_HCControl_postedWriteEnable);
+       flush_writes(ohci);
+
+       for (lps = 0, i = 0; !lps && i < 3; i++) {
+               msleep(50);
+               lps = reg_read(ohci, OHCI1394_HCControlSet) &
+                     OHCI1394_HCControl_LPS;
+       }
+
+       if (!lps) {
+               ohci_err(ohci, "failed to set Link Power Status\n");
+               return -EIO;
+       }
+
+       if (ohci->quirks & QUIRK_TI_SLLZ059) {
+               ret = probe_tsb41ba3d(ohci);
+               if (ret < 0)
+                       return ret;
+               if (ret)
+                       ohci_notice(ohci, "local TSB41BA3D phy\n");
+               else
+                       ohci->quirks &= ~QUIRK_TI_SLLZ059;
+       }
+
+       reg_write(ohci, OHCI1394_HCControlClear,
+                 OHCI1394_HCControl_noByteSwapData);
+
+       reg_write(ohci, OHCI1394_SelfIDBuffer, ohci->self_id_bus);
+       reg_write(ohci, OHCI1394_LinkControlSet,
+                 OHCI1394_LinkControl_cycleTimerEnable |
+                 OHCI1394_LinkControl_cycleMaster);
+
+       reg_write(ohci, OHCI1394_ATRetries,
+                 OHCI1394_MAX_AT_REQ_RETRIES |
+                 (OHCI1394_MAX_AT_RESP_RETRIES << 4) |
+                 (OHCI1394_MAX_PHYS_RESP_RETRIES << 8) |
+                 (200 << 16));
+
+       ohci->bus_time_running = false;
+
+       for (i = 0; i < 32; i++)
+               if (ohci->ir_context_support & (1 << i))
+                       reg_write(ohci, OHCI1394_IsoRcvContextControlClear(i),
+                                 IR_CONTEXT_MULTI_CHANNEL_MODE);
+
+       version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
+       if (version >= OHCI_VERSION_1_1) {
+               reg_write(ohci, OHCI1394_InitialChannelsAvailableHi,
+                         0xfffffffe);
+               card->broadcast_channel_auto_allocated = true;
+       }
+
+       /* Get implemented bits of the priority arbitration request counter. */
+       reg_write(ohci, OHCI1394_FairnessControl, 0x3f);
+       ohci->pri_req_max = reg_read(ohci, OHCI1394_FairnessControl) & 0x3f;
+       reg_write(ohci, OHCI1394_FairnessControl, 0);
+       card->priority_budget_implemented = ohci->pri_req_max != 0;
+
+       reg_write(ohci, OHCI1394_PhyUpperBound, FW_MAX_PHYSICAL_RANGE >> 16);
+       reg_write(ohci, OHCI1394_IntEventClear, ~0);
+       reg_write(ohci, OHCI1394_IntMaskClear, ~0);
+
+       ret = configure_1394a_enhancements(ohci);
+       if (ret < 0)
+               return ret;
+
+       /* Activate link_on bit and contender bit in our self ID packets.*/
+       ret = ohci_update_phy_reg(card, 4, 0, PHY_LINK_ACTIVE | PHY_CONTENDER);
+       if (ret < 0)
+               return ret;
+
+       /*
+        * When the link is not yet enabled, the atomic config rom
+        * update mechanism described below in ohci_set_config_rom()
+        * is not active.  We have to update ConfigRomHeader and
+        * BusOptions manually, and the write to ConfigROMmap takes
+        * effect immediately.  We tie this to the enabling of the
+        * link, so we have a valid config rom before enabling - the
+        * OHCI requires that ConfigROMhdr and BusOptions have valid
+        * values before enabling.
+        *
+        * However, when the ConfigROMmap is written, some controllers
+        * always read back quadlets 0 and 2 from the config rom to
+        * the ConfigRomHeader and BusOptions registers on bus reset.
+        * They shouldn't do that in this initial case where the link
+        * isn't enabled.  This means we have to use the same
+        * workaround here, setting the bus header to 0 and then write
+        * the right values in the bus reset tasklet.
+        */
+
+       if (config_rom) {
+               ohci->next_config_rom =
+                       dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
+                                          &ohci->next_config_rom_bus,
+                                          GFP_KERNEL);
+               if (ohci->next_config_rom == NULL)
+                       return -ENOMEM;
+
+               copy_config_rom(ohci->next_config_rom, config_rom, length);
+       } else {
+               /*
+                * In the suspend case, config_rom is NULL, which
+                * means that we just reuse the old config rom.
+                */
+               ohci->next_config_rom = ohci->config_rom;
+               ohci->next_config_rom_bus = ohci->config_rom_bus;
+       }
+
+       ohci->next_header = ohci->next_config_rom[0];
+       ohci->next_config_rom[0] = 0;
+       reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
+       reg_write(ohci, OHCI1394_BusOptions,
+                 be32_to_cpu(ohci->next_config_rom[2]));
+       reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
+
+       reg_write(ohci, OHCI1394_AsReqFilterHiSet, 0x80000000);
+
+       irqs =  OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
+               OHCI1394_RQPkt | OHCI1394_RSPkt |
+               OHCI1394_isochTx | OHCI1394_isochRx |
+               OHCI1394_postedWriteErr |
+               OHCI1394_selfIDComplete |
+               OHCI1394_regAccessFail |
+               OHCI1394_cycleInconsistent |
+               OHCI1394_unrecoverableError |
+               OHCI1394_cycleTooLong |
+               OHCI1394_masterIntEnable;
+       if (param_debug & OHCI_PARAM_DEBUG_BUSRESETS)
+               irqs |= OHCI1394_busReset;
+       reg_write(ohci, OHCI1394_IntMaskSet, irqs);
+
+       reg_write(ohci, OHCI1394_HCControlSet,
+                 OHCI1394_HCControl_linkEnable |
+                 OHCI1394_HCControl_BIBimageValid);
+
+       reg_write(ohci, OHCI1394_LinkControlSet,
+                 OHCI1394_LinkControl_rcvSelfID |
+                 OHCI1394_LinkControl_rcvPhyPkt);
+
+       ar_context_run(&ohci->ar_request_ctx);
+       ar_context_run(&ohci->ar_response_ctx);
+
+       flush_writes(ohci);
+
+       /* We are ready to go, reset bus to finish initialization. */
+       fw_schedule_bus_reset(&ohci->card, false, true);
+
+       return 0;
+}
+
+static int ohci_set_config_rom(struct fw_card *card,
+                              const __be32 *config_rom, size_t length)
+{
+       struct fw_ohci *ohci;
+       __be32 *next_config_rom;
+       dma_addr_t uninitialized_var(next_config_rom_bus);
+
+       ohci = fw_ohci(card);
+
+       /*
+        * When the OHCI controller is enabled, the config rom update
+        * mechanism is a bit tricky, but easy enough to use.  See
+        * section 5.5.6 in the OHCI specification.
+        *
+        * The OHCI controller caches the new config rom address in a
+        * shadow register (ConfigROMmapNext) and needs a bus reset
+        * for the changes to take place.  When the bus reset is
+        * detected, the controller loads the new values for the
+        * ConfigRomHeader and BusOptions registers from the specified
+        * config rom and loads ConfigROMmap from the ConfigROMmapNext
+        * shadow register. All automatically and atomically.
+        *
+        * Now, there's a twist to this story.  The automatic load of
+        * ConfigRomHeader and BusOptions doesn't honor the
+        * noByteSwapData bit, so with a be32 config rom, the
+        * controller will load be32 values in to these registers
+        * during the atomic update, even on litte endian
+        * architectures.  The workaround we use is to put a 0 in the
+        * header quadlet; 0 is endian agnostic and means that the
+        * config rom isn't ready yet.  In the bus reset tasklet we
+        * then set up the real values for the two registers.
+        *
+        * We use ohci->lock to avoid racing with the code that sets
+        * ohci->next_config_rom to NULL (see bus_reset_work).
+        */
+
+       next_config_rom =
+               dma_alloc_coherent(ohci->card.device, CONFIG_ROM_SIZE,
+                                  &next_config_rom_bus, GFP_KERNEL);
+       if (next_config_rom == NULL)
+               return -ENOMEM;
+
+       spin_lock_irq(&ohci->lock);
+
+       /*
+        * If there is not an already pending config_rom update,
+        * push our new allocation into the ohci->next_config_rom
+        * and then mark the local variable as null so that we
+        * won't deallocate the new buffer.
+        *
+        * OTOH, if there is a pending config_rom update, just
+        * use that buffer with the new config_rom data, and
+        * let this routine free the unused DMA allocation.
+        */
+
+       if (ohci->next_config_rom == NULL) {
+               ohci->next_config_rom = next_config_rom;
+               ohci->next_config_rom_bus = next_config_rom_bus;
+               next_config_rom = NULL;
+       }
+
+       copy_config_rom(ohci->next_config_rom, config_rom, length);
+
+       ohci->next_header = config_rom[0];
+       ohci->next_config_rom[0] = 0;
+
+       reg_write(ohci, OHCI1394_ConfigROMmap, ohci->next_config_rom_bus);
+
+       spin_unlock_irq(&ohci->lock);
+
+       /* If we didn't use the DMA allocation, delete it. */
+       if (next_config_rom != NULL)
+               dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
+                                 next_config_rom, next_config_rom_bus);
+
+       /*
+        * Now initiate a bus reset to have the changes take
+        * effect. We clean up the old config rom memory and DMA
+        * mappings in the bus reset tasklet, since the OHCI
+        * controller could need to access it before the bus reset
+        * takes effect.
+        */
+
+       fw_schedule_bus_reset(&ohci->card, true, true);
+
+       return 0;
+}
+
+static void ohci_send_request(struct fw_card *card, struct fw_packet *packet)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+
+       at_context_transmit(&ohci->at_request_ctx, packet);
+}
+
+static void ohci_send_response(struct fw_card *card, struct fw_packet *packet)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+
+       at_context_transmit(&ohci->at_response_ctx, packet);
+}
+
+static int ohci_cancel_packet(struct fw_card *card, struct fw_packet *packet)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+       struct context *ctx = &ohci->at_request_ctx;
+       struct driver_data *driver_data = packet->driver_data;
+       int ret = -ENOENT;
+
+       tasklet_disable(&ctx->tasklet);
+
+       if (packet->ack != 0)
+               goto out;
+
+       if (packet->payload_mapped)
+               dma_unmap_single(ohci->card.device, packet->payload_bus,
+                                packet->payload_length, DMA_TO_DEVICE);
+
+       log_ar_at_event(ohci, 'T', packet->speed, packet->header, 0x20);
+       driver_data->packet = NULL;
+       packet->ack = RCODE_CANCELLED;
+       packet->callback(packet, &ohci->card, packet->ack);
+       ret = 0;
+ out:
+       tasklet_enable(&ctx->tasklet);
+
+       return ret;
+}
+
+static int ohci_enable_phys_dma(struct fw_card *card,
+                               int node_id, int generation)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+       unsigned long flags;
+       int n, ret = 0;
+
+       if (param_remote_dma)
+               return 0;
+
+       /*
+        * FIXME:  Make sure this bitmask is cleared when we clear the busReset
+        * interrupt bit.  Clear physReqResourceAllBuses on bus reset.
+        */
+
+       spin_lock_irqsave(&ohci->lock, flags);
+
+       if (ohci->generation != generation) {
+               ret = -ESTALE;
+               goto out;
+       }
+
+       /*
+        * Note, if the node ID contains a non-local bus ID, physical DMA is
+        * enabled for _all_ nodes on remote buses.
+        */
+
+       n = (node_id & 0xffc0) == LOCAL_BUS ? node_id & 0x3f : 63;
+       if (n < 32)
+               reg_write(ohci, OHCI1394_PhyReqFilterLoSet, 1 << n);
+       else
+               reg_write(ohci, OHCI1394_PhyReqFilterHiSet, 1 << (n - 32));
+
+       flush_writes(ohci);
+ out:
+       spin_unlock_irqrestore(&ohci->lock, flags);
+
+       return ret;
+}
+
+static u32 ohci_read_csr(struct fw_card *card, int csr_offset)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+       unsigned long flags;
+       u32 value;
+
+       switch (csr_offset) {
+       case CSR_STATE_CLEAR:
+       case CSR_STATE_SET:
+               if (ohci->is_root &&
+                   (reg_read(ohci, OHCI1394_LinkControlSet) &
+                    OHCI1394_LinkControl_cycleMaster))
+                       value = CSR_STATE_BIT_CMSTR;
+               else
+                       value = 0;
+               if (ohci->csr_state_setclear_abdicate)
+                       value |= CSR_STATE_BIT_ABDICATE;
+
+               return value;
+
+       case CSR_NODE_IDS:
+               return reg_read(ohci, OHCI1394_NodeID) << 16;
+
+       case CSR_CYCLE_TIME:
+               return get_cycle_time(ohci);
+
+       case CSR_BUS_TIME:
+               /*
+                * We might be called just after the cycle timer has wrapped
+                * around but just before the cycle64Seconds handler, so we
+                * better check here, too, if the bus time needs to be updated.
+                */
+               spin_lock_irqsave(&ohci->lock, flags);
+               value = update_bus_time(ohci);
+               spin_unlock_irqrestore(&ohci->lock, flags);
+               return value;
+
+       case CSR_BUSY_TIMEOUT:
+               value = reg_read(ohci, OHCI1394_ATRetries);
+               return (value >> 4) & 0x0ffff00f;
+
+       case CSR_PRIORITY_BUDGET:
+               return (reg_read(ohci, OHCI1394_FairnessControl) & 0x3f) |
+                       (ohci->pri_req_max << 8);
+
+       default:
+               WARN_ON(1);
+               return 0;
+       }
+}
+
+static void ohci_write_csr(struct fw_card *card, int csr_offset, u32 value)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+       unsigned long flags;
+
+       switch (csr_offset) {
+       case CSR_STATE_CLEAR:
+               if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
+                       reg_write(ohci, OHCI1394_LinkControlClear,
+                                 OHCI1394_LinkControl_cycleMaster);
+                       flush_writes(ohci);
+               }
+               if (value & CSR_STATE_BIT_ABDICATE)
+                       ohci->csr_state_setclear_abdicate = false;
+               break;
+
+       case CSR_STATE_SET:
+               if ((value & CSR_STATE_BIT_CMSTR) && ohci->is_root) {
+                       reg_write(ohci, OHCI1394_LinkControlSet,
+                                 OHCI1394_LinkControl_cycleMaster);
+                       flush_writes(ohci);
+               }
+               if (value & CSR_STATE_BIT_ABDICATE)
+                       ohci->csr_state_setclear_abdicate = true;
+               break;
+
+       case CSR_NODE_IDS:
+               reg_write(ohci, OHCI1394_NodeID, value >> 16);
+               flush_writes(ohci);
+               break;
+
+       case CSR_CYCLE_TIME:
+               reg_write(ohci, OHCI1394_IsochronousCycleTimer, value);
+               reg_write(ohci, OHCI1394_IntEventSet,
+                         OHCI1394_cycleInconsistent);
+               flush_writes(ohci);
+               break;
+
+       case CSR_BUS_TIME:
+               spin_lock_irqsave(&ohci->lock, flags);
+               ohci->bus_time = (update_bus_time(ohci) & 0x40) |
+                                (value & ~0x7f);
+               spin_unlock_irqrestore(&ohci->lock, flags);
+               break;
+
+       case CSR_BUSY_TIMEOUT:
+               value = (value & 0xf) | ((value & 0xf) << 4) |
+                       ((value & 0xf) << 8) | ((value & 0x0ffff000) << 4);
+               reg_write(ohci, OHCI1394_ATRetries, value);
+               flush_writes(ohci);
+               break;
+
+       case CSR_PRIORITY_BUDGET:
+               reg_write(ohci, OHCI1394_FairnessControl, value & 0x3f);
+               flush_writes(ohci);
+               break;
+
+       default:
+               WARN_ON(1);
+               break;
+       }
+}
+
+static void flush_iso_completions(struct iso_context *ctx)
+{
+       ctx->base.callback.sc(&ctx->base, ctx->last_timestamp,
+                             ctx->header_length, ctx->header,
+                             ctx->base.callback_data);
+       ctx->header_length = 0;
+}
+
+static void copy_iso_headers(struct iso_context *ctx, const u32 *dma_hdr)
+{
+       u32 *ctx_hdr;
+
+       if (ctx->header_length + ctx->base.header_size > PAGE_SIZE) {
+               if (ctx->base.drop_overflow_headers)
+                       return;
+               flush_iso_completions(ctx);
+       }
+
+       ctx_hdr = ctx->header + ctx->header_length;
+       ctx->last_timestamp = (u16)le32_to_cpu((__force __le32)dma_hdr[0]);
+
+       /*
+        * The two iso header quadlets are byteswapped to little
+        * endian by the controller, but we want to present them
+        * as big endian for consistency with the bus endianness.
+        */
+       if (ctx->base.header_size > 0)
+               ctx_hdr[0] = swab32(dma_hdr[1]); /* iso packet header */
+       if (ctx->base.header_size > 4)
+               ctx_hdr[1] = swab32(dma_hdr[0]); /* timestamp */
+       if (ctx->base.header_size > 8)
+               memcpy(&ctx_hdr[2], &dma_hdr[2], ctx->base.header_size - 8);
+       ctx->header_length += ctx->base.header_size;
+}
+
+static int handle_ir_packet_per_buffer(struct context *context,
+                                      struct descriptor *d,
+                                      struct descriptor *last)
+{
+       struct iso_context *ctx =
+               container_of(context, struct iso_context, context);
+       struct descriptor *pd;
+       u32 buffer_dma;
+
+       for (pd = d; pd <= last; pd++)
+               if (pd->transfer_status)
+                       break;
+       if (pd > last)
+               /* Descriptor(s) not done yet, stop iteration */
+               return 0;
+
+       while (!(d->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))) {
+               d++;
+               buffer_dma = le32_to_cpu(d->data_address);
+               dma_sync_single_range_for_cpu(context->ohci->card.device,
+                                             buffer_dma & PAGE_MASK,
+                                             buffer_dma & ~PAGE_MASK,
+                                             le16_to_cpu(d->req_count),
+                                             DMA_FROM_DEVICE);
+       }
+
+       copy_iso_headers(ctx, (u32 *) (last + 1));
+
+       if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
+               flush_iso_completions(ctx);
+
+       return 1;
+}
+
+/* d == last because each descriptor block is only a single descriptor. */
+static int handle_ir_buffer_fill(struct context *context,
+                                struct descriptor *d,
+                                struct descriptor *last)
+{
+       struct iso_context *ctx =
+               container_of(context, struct iso_context, context);
+       unsigned int req_count, res_count, completed;
+       u32 buffer_dma;
+
+       req_count = le16_to_cpu(last->req_count);
+       res_count = le16_to_cpu(ACCESS_ONCE(last->res_count));
+       completed = req_count - res_count;
+       buffer_dma = le32_to_cpu(last->data_address);
+
+       if (completed > 0) {
+               ctx->mc_buffer_bus = buffer_dma;
+               ctx->mc_completed = completed;
+       }
+
+       if (res_count != 0)
+               /* Descriptor(s) not done yet, stop iteration */
+               return 0;
+
+       dma_sync_single_range_for_cpu(context->ohci->card.device,
+                                     buffer_dma & PAGE_MASK,
+                                     buffer_dma & ~PAGE_MASK,
+                                     completed, DMA_FROM_DEVICE);
+
+       if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS)) {
+               ctx->base.callback.mc(&ctx->base,
+                                     buffer_dma + completed,
+                                     ctx->base.callback_data);
+               ctx->mc_completed = 0;
+       }
+
+       return 1;
+}
+
+static void flush_ir_buffer_fill(struct iso_context *ctx)
+{
+       dma_sync_single_range_for_cpu(ctx->context.ohci->card.device,
+                                     ctx->mc_buffer_bus & PAGE_MASK,
+                                     ctx->mc_buffer_bus & ~PAGE_MASK,
+                                     ctx->mc_completed, DMA_FROM_DEVICE);
+
+       ctx->base.callback.mc(&ctx->base,
+                             ctx->mc_buffer_bus + ctx->mc_completed,
+                             ctx->base.callback_data);
+       ctx->mc_completed = 0;
+}
+
+static inline void sync_it_packet_for_cpu(struct context *context,
+                                         struct descriptor *pd)
+{
+       __le16 control;
+       u32 buffer_dma;
+
+       /* only packets beginning with OUTPUT_MORE* have data buffers */
+       if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
+               return;
+
+       /* skip over the OUTPUT_MORE_IMMEDIATE descriptor */
+       pd += 2;
+
+       /*
+        * If the packet has a header, the first OUTPUT_MORE/LAST descriptor's
+        * data buffer is in the context program's coherent page and must not
+        * be synced.
+        */
+       if ((le32_to_cpu(pd->data_address) & PAGE_MASK) ==
+           (context->current_bus          & PAGE_MASK)) {
+               if (pd->control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS))
+                       return;
+               pd++;
+       }
+
+       do {
+               buffer_dma = le32_to_cpu(pd->data_address);
+               dma_sync_single_range_for_cpu(context->ohci->card.device,
+                                             buffer_dma & PAGE_MASK,
+                                             buffer_dma & ~PAGE_MASK,
+                                             le16_to_cpu(pd->req_count),
+                                             DMA_TO_DEVICE);
+               control = pd->control;
+               pd++;
+       } while (!(control & cpu_to_le16(DESCRIPTOR_BRANCH_ALWAYS)));
+}
+
+static int handle_it_packet(struct context *context,
+                           struct descriptor *d,
+                           struct descriptor *last)
+{
+       struct iso_context *ctx =
+               container_of(context, struct iso_context, context);
+       struct descriptor *pd;
+       __be32 *ctx_hdr;
+
+       for (pd = d; pd <= last; pd++)
+               if (pd->transfer_status)
+                       break;
+       if (pd > last)
+               /* Descriptor(s) not done yet, stop iteration */
+               return 0;
+
+       sync_it_packet_for_cpu(context, d);
+
+       if (ctx->header_length + 4 > PAGE_SIZE) {
+               if (ctx->base.drop_overflow_headers)
+                       return 1;
+               flush_iso_completions(ctx);
+       }
+
+       ctx_hdr = ctx->header + ctx->header_length;
+       ctx->last_timestamp = le16_to_cpu(last->res_count);
+       /* Present this value as big-endian to match the receive code */
+       *ctx_hdr = cpu_to_be32((le16_to_cpu(pd->transfer_status) << 16) |
+                              le16_to_cpu(pd->res_count));
+       ctx->header_length += 4;
+
+       if (last->control & cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS))
+               flush_iso_completions(ctx);
+
+       return 1;
+}
+
+static void set_multichannel_mask(struct fw_ohci *ohci, u64 channels)
+{
+       u32 hi = channels >> 32, lo = channels;
+
+       reg_write(ohci, OHCI1394_IRMultiChanMaskHiClear, ~hi);
+       reg_write(ohci, OHCI1394_IRMultiChanMaskLoClear, ~lo);
+       reg_write(ohci, OHCI1394_IRMultiChanMaskHiSet, hi);
+       reg_write(ohci, OHCI1394_IRMultiChanMaskLoSet, lo);
+       mmiowb();
+       ohci->mc_channels = channels;
+}
+
+static struct fw_iso_context *ohci_allocate_iso_context(struct fw_card *card,
+                               int type, int channel, size_t header_size)
+{
+       struct fw_ohci *ohci = fw_ohci(card);
+       struct iso_context *uninitialized_var(ctx);
+       descriptor_callback_t uninitialized_var(callback);
+       u64 *uninitialized_var(channels);
+       u32 *uninitialized_var(mask), uninitialized_var(regs);
+       int index, ret = -EBUSY;
+
+       spin_lock_irq(&ohci->lock);
+
+       switch (type) {
+       case FW_ISO_CONTEXT_TRANSMIT:
+               mask     = &ohci->it_context_mask;
+               callback = handle_it_packet;
+               index    = ffs(*mask) - 1;
+               if (index >= 0) {
+                       *mask &= ~(1 << index);
+                       regs = OHCI1394_IsoXmitContextBase(index);
+                       ctx  = &ohci->it_context_list[index];
+               }
+               break;
+
+       case FW_ISO_CONTEXT_RECEIVE:
+               channels = &ohci->ir_context_channels;
+               mask     = &ohci->ir_context_mask;
+               callback = handle_ir_packet_per_buffer;
+               index    = *channels & 1ULL << channel ? ffs(*mask) - 1 : -1;
+               if (index >= 0) {
+                       *channels &= ~(1ULL << channel);
+                       *mask     &= ~(1 << index);
+                       regs = OHCI1394_IsoRcvContextBase(index);
+                       ctx  = &ohci->ir_context_list[index];
+               }
+               break;
+
+       case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
+               mask     = &ohci->ir_context_mask;
+               callback = handle_ir_buffer_fill;
+               index    = !ohci->mc_allocated ? ffs(*mask) - 1 : -1;
+               if (index >= 0) {
+                       ohci->mc_allocated = true;
+                       *mask &= ~(1 << index);
+                       regs = OHCI1394_IsoRcvContextBase(index);
+                       ctx  = &ohci->ir_context_list[index];
+               }
+               break;
+
+       default:
+               index = -1;
+               ret = -ENOSYS;
+       }
+
+       spin_unlock_irq(&ohci->lock);
+
+       if (index < 0)
+               return ERR_PTR(ret);
+
+       memset(ctx, 0, sizeof(*ctx));
+       ctx->header_length = 0;
+       ctx->header = (void *) __get_free_page(GFP_KERNEL);
+       if (ctx->header == NULL) {
+               ret = -ENOMEM;
+               goto out;
+       }
+       ret = context_init(&ctx->context, ohci, regs, callback);
+       if (ret < 0)
+               goto out_with_header;
+
+       if (type == FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL) {
+               set_multichannel_mask(ohci, 0);
+               ctx->mc_completed = 0;
+       }
+
+       return &ctx->base;
+
+ out_with_header:
+       free_page((unsigned long)ctx->header);
+ out:
+       spin_lock_irq(&ohci->lock);
+
+       switch (type) {
+       case FW_ISO_CONTEXT_RECEIVE:
+               *channels |= 1ULL << channel;
+               break;
+
+       case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
+               ohci->mc_allocated = false;
+               break;
+       }
+       *mask |= 1 << index;
+
+       spin_unlock_irq(&ohci->lock);
+
+       return ERR_PTR(ret);
+}
+
+static int ohci_start_iso(struct fw_iso_context *base,
+                         s32 cycle, u32 sync, u32 tags)
+{
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       struct fw_ohci *ohci = ctx->context.ohci;
+       u32 control = IR_CONTEXT_ISOCH_HEADER, match;
+       int index;
+
+       /* the controller cannot start without any queued packets */
+       if (ctx->context.last->branch_address == 0)
+               return -ENODATA;
+
+       switch (ctx->base.type) {
+       case FW_ISO_CONTEXT_TRANSMIT:
+               index = ctx - ohci->it_context_list;
+               match = 0;
+               if (cycle >= 0)
+                       match = IT_CONTEXT_CYCLE_MATCH_ENABLE |
+                               (cycle & 0x7fff) << 16;
+
+               reg_write(ohci, OHCI1394_IsoXmitIntEventClear, 1 << index);
+               reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, 1 << index);
+               context_run(&ctx->context, match);
+               break;
+
+       case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
+               control |= IR_CONTEXT_BUFFER_FILL|IR_CONTEXT_MULTI_CHANNEL_MODE;
+               /* fall through */
+       case FW_ISO_CONTEXT_RECEIVE:
+               index = ctx - ohci->ir_context_list;
+               match = (tags << 28) | (sync << 8) | ctx->base.channel;
+               if (cycle >= 0) {
+                       match |= (cycle & 0x07fff) << 12;
+                       control |= IR_CONTEXT_CYCLE_MATCH_ENABLE;
+               }
+
+               reg_write(ohci, OHCI1394_IsoRecvIntEventClear, 1 << index);
+               reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, 1 << index);
+               reg_write(ohci, CONTEXT_MATCH(ctx->context.regs), match);
+               context_run(&ctx->context, control);
+
+               ctx->sync = sync;
+               ctx->tags = tags;
+
+               break;
+       }
+
+       return 0;
+}
+
+static int ohci_stop_iso(struct fw_iso_context *base)
+{
+       struct fw_ohci *ohci = fw_ohci(base->card);
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       int index;
+
+       switch (ctx->base.type) {
+       case FW_ISO_CONTEXT_TRANSMIT:
+               index = ctx - ohci->it_context_list;
+               reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, 1 << index);
+               break;
+
+       case FW_ISO_CONTEXT_RECEIVE:
+       case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
+               index = ctx - ohci->ir_context_list;
+               reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, 1 << index);
+               break;
+       }
+       flush_writes(ohci);
+       context_stop(&ctx->context);
+       tasklet_kill(&ctx->context.tasklet);
+
+       return 0;
+}
+
+static void ohci_free_iso_context(struct fw_iso_context *base)
+{
+       struct fw_ohci *ohci = fw_ohci(base->card);
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       unsigned long flags;
+       int index;
+
+       ohci_stop_iso(base);
+       context_release(&ctx->context);
+       free_page((unsigned long)ctx->header);
+
+       spin_lock_irqsave(&ohci->lock, flags);
+
+       switch (base->type) {
+       case FW_ISO_CONTEXT_TRANSMIT:
+               index = ctx - ohci->it_context_list;
+               ohci->it_context_mask |= 1 << index;
+               break;
+
+       case FW_ISO_CONTEXT_RECEIVE:
+               index = ctx - ohci->ir_context_list;
+               ohci->ir_context_mask |= 1 << index;
+               ohci->ir_context_channels |= 1ULL << base->channel;
+               break;
+
+       case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
+               index = ctx - ohci->ir_context_list;
+               ohci->ir_context_mask |= 1 << index;
+               ohci->ir_context_channels |= ohci->mc_channels;
+               ohci->mc_channels = 0;
+               ohci->mc_allocated = false;
+               break;
+       }
+
+       spin_unlock_irqrestore(&ohci->lock, flags);
+}
+
+static int ohci_set_iso_channels(struct fw_iso_context *base, u64 *channels)
+{
+       struct fw_ohci *ohci = fw_ohci(base->card);
+       unsigned long flags;
+       int ret;
+
+       switch (base->type) {
+       case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
+
+               spin_lock_irqsave(&ohci->lock, flags);
+
+               /* Don't allow multichannel to grab other contexts' channels. */
+               if (~ohci->ir_context_channels & ~ohci->mc_channels & *channels) {
+                       *channels = ohci->ir_context_channels;
+                       ret = -EBUSY;
+               } else {
+                       set_multichannel_mask(ohci, *channels);
+                       ret = 0;
+               }
+
+               spin_unlock_irqrestore(&ohci->lock, flags);
+
+               break;
+       default:
+               ret = -EINVAL;
+       }
+
+       return ret;
+}
+
+#ifdef CONFIG_PM
+static void ohci_resume_iso_dma(struct fw_ohci *ohci)
+{
+       int i;
+       struct iso_context *ctx;
+
+       for (i = 0 ; i < ohci->n_ir ; i++) {
+               ctx = &ohci->ir_context_list[i];
+               if (ctx->context.running)
+                       ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
+       }
+
+       for (i = 0 ; i < ohci->n_it ; i++) {
+               ctx = &ohci->it_context_list[i];
+               if (ctx->context.running)
+                       ohci_start_iso(&ctx->base, 0, ctx->sync, ctx->tags);
+       }
+}
+#endif
+
+static int queue_iso_transmit(struct iso_context *ctx,
+                             struct fw_iso_packet *packet,
+                             struct fw_iso_buffer *buffer,
+                             unsigned long payload)
+{
+       struct descriptor *d, *last, *pd;
+       struct fw_iso_packet *p;
+       __le32 *header;
+       dma_addr_t d_bus, page_bus;
+       u32 z, header_z, payload_z, irq;
+       u32 payload_index, payload_end_index, next_page_index;
+       int page, end_page, i, length, offset;
+
+       p = packet;
+       payload_index = payload;
+
+       if (p->skip)
+               z = 1;
+       else
+               z = 2;
+       if (p->header_length > 0)
+               z++;
+
+       /* Determine the first page the payload isn't contained in. */
+       end_page = PAGE_ALIGN(payload_index + p->payload_length) >> PAGE_SHIFT;
+       if (p->payload_length > 0)
+               payload_z = end_page - (payload_index >> PAGE_SHIFT);
+       else
+               payload_z = 0;
+
+       z += payload_z;
+
+       /* Get header size in number of descriptors. */
+       header_z = DIV_ROUND_UP(p->header_length, sizeof(*d));
+
+       d = context_get_descriptors(&ctx->context, z + header_z, &d_bus);
+       if (d == NULL)
+               return -ENOMEM;
+
+       if (!p->skip) {
+               d[0].control   = cpu_to_le16(DESCRIPTOR_KEY_IMMEDIATE);
+               d[0].req_count = cpu_to_le16(8);
+               /*
+                * Link the skip address to this descriptor itself.  This causes
+                * a context to skip a cycle whenever lost cycles or FIFO
+                * overruns occur, without dropping the data.  The application
+                * should then decide whether this is an error condition or not.
+                * FIXME:  Make the context's cycle-lost behaviour configurable?
+                */
+               d[0].branch_address = cpu_to_le32(d_bus | z);
+
+               header = (__le32 *) &d[1];
+               header[0] = cpu_to_le32(IT_HEADER_SY(p->sy) |
+                                       IT_HEADER_TAG(p->tag) |
+                                       IT_HEADER_TCODE(TCODE_STREAM_DATA) |
+                                       IT_HEADER_CHANNEL(ctx->base.channel) |
+                                       IT_HEADER_SPEED(ctx->base.speed));
+               header[1] =
+                       cpu_to_le32(IT_HEADER_DATA_LENGTH(p->header_length +
+                                                         p->payload_length));
+       }
+
+       if (p->header_length > 0) {
+               d[2].req_count    = cpu_to_le16(p->header_length);
+               d[2].data_address = cpu_to_le32(d_bus + z * sizeof(*d));
+               memcpy(&d[z], p->header, p->header_length);
+       }
+
+       pd = d + z - payload_z;
+       payload_end_index = payload_index + p->payload_length;
+       for (i = 0; i < payload_z; i++) {
+               page               = payload_index >> PAGE_SHIFT;
+               offset             = payload_index & ~PAGE_MASK;
+               next_page_index    = (page + 1) << PAGE_SHIFT;
+               length             =
+                       min(next_page_index, payload_end_index) - payload_index;
+               pd[i].req_count    = cpu_to_le16(length);
+
+               page_bus = page_private(buffer->pages[page]);
+               pd[i].data_address = cpu_to_le32(page_bus + offset);
+
+               dma_sync_single_range_for_device(ctx->context.ohci->card.device,
+                                                page_bus, offset, length,
+                                                DMA_TO_DEVICE);
+
+               payload_index += length;
+       }
+
+       if (p->interrupt)
+               irq = DESCRIPTOR_IRQ_ALWAYS;
+       else
+               irq = DESCRIPTOR_NO_IRQ;
+
+       last = z == 2 ? d : d + z - 1;
+       last->control |= cpu_to_le16(DESCRIPTOR_OUTPUT_LAST |
+                                    DESCRIPTOR_STATUS |
+                                    DESCRIPTOR_BRANCH_ALWAYS |
+                                    irq);
+
+       context_append(&ctx->context, d, z, header_z);
+
+       return 0;
+}
+
+static int queue_iso_packet_per_buffer(struct iso_context *ctx,
+                                      struct fw_iso_packet *packet,
+                                      struct fw_iso_buffer *buffer,
+                                      unsigned long payload)
+{
+       struct device *device = ctx->context.ohci->card.device;
+       struct descriptor *d, *pd;
+       dma_addr_t d_bus, page_bus;
+       u32 z, header_z, rest;
+       int i, j, length;
+       int page, offset, packet_count, header_size, payload_per_buffer;
+
+       /*
+        * The OHCI controller puts the isochronous header and trailer in the
+        * buffer, so we need at least 8 bytes.
+        */
+       packet_count = packet->header_length / ctx->base.header_size;
+       header_size  = max(ctx->base.header_size, (size_t)8);
+
+       /* Get header size in number of descriptors. */
+       header_z = DIV_ROUND_UP(header_size, sizeof(*d));
+       page     = payload >> PAGE_SHIFT;
+       offset   = payload & ~PAGE_MASK;
+       payload_per_buffer = packet->payload_length / packet_count;
+
+       for (i = 0; i < packet_count; i++) {
+               /* d points to the header descriptor */
+               z = DIV_ROUND_UP(payload_per_buffer + offset, PAGE_SIZE) + 1;
+               d = context_get_descriptors(&ctx->context,
+                               z + header_z, &d_bus);
+               if (d == NULL)
+                       return -ENOMEM;
+
+               d->control      = cpu_to_le16(DESCRIPTOR_STATUS |
+                                             DESCRIPTOR_INPUT_MORE);
+               if (packet->skip && i == 0)
+                       d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
+               d->req_count    = cpu_to_le16(header_size);
+               d->res_count    = d->req_count;
+               d->transfer_status = 0;
+               d->data_address = cpu_to_le32(d_bus + (z * sizeof(*d)));
+
+               rest = payload_per_buffer;
+               pd = d;
+               for (j = 1; j < z; j++) {
+                       pd++;
+                       pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
+                                                 DESCRIPTOR_INPUT_MORE);
+
+                       if (offset + rest < PAGE_SIZE)
+                               length = rest;
+                       else
+                               length = PAGE_SIZE - offset;
+                       pd->req_count = cpu_to_le16(length);
+                       pd->res_count = pd->req_count;
+                       pd->transfer_status = 0;
+
+                       page_bus = page_private(buffer->pages[page]);
+                       pd->data_address = cpu_to_le32(page_bus + offset);
+
+                       dma_sync_single_range_for_device(device, page_bus,
+                                                        offset, length,
+                                                        DMA_FROM_DEVICE);
+
+                       offset = (offset + length) & ~PAGE_MASK;
+                       rest -= length;
+                       if (offset == 0)
+                               page++;
+               }
+               pd->control = cpu_to_le16(DESCRIPTOR_STATUS |
+                                         DESCRIPTOR_INPUT_LAST |
+                                         DESCRIPTOR_BRANCH_ALWAYS);
+               if (packet->interrupt && i == packet_count - 1)
+                       pd->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
+
+               context_append(&ctx->context, d, z, header_z);
+       }
+
+       return 0;
+}
+
+static int queue_iso_buffer_fill(struct iso_context *ctx,
+                                struct fw_iso_packet *packet,
+                                struct fw_iso_buffer *buffer,
+                                unsigned long payload)
+{
+       struct descriptor *d;
+       dma_addr_t d_bus, page_bus;
+       int page, offset, rest, z, i, length;
+
+       page   = payload >> PAGE_SHIFT;
+       offset = payload & ~PAGE_MASK;
+       rest   = packet->payload_length;
+
+       /* We need one descriptor for each page in the buffer. */
+       z = DIV_ROUND_UP(offset + rest, PAGE_SIZE);
+
+       if (WARN_ON(offset & 3 || rest & 3 || page + z > buffer->page_count))
+               return -EFAULT;
+
+       for (i = 0; i < z; i++) {
+               d = context_get_descriptors(&ctx->context, 1, &d_bus);
+               if (d == NULL)
+                       return -ENOMEM;
+
+               d->control = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
+                                        DESCRIPTOR_BRANCH_ALWAYS);
+               if (packet->skip && i == 0)
+                       d->control |= cpu_to_le16(DESCRIPTOR_WAIT);
+               if (packet->interrupt && i == z - 1)
+                       d->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
+
+               if (offset + rest < PAGE_SIZE)
+                       length = rest;
+               else
+                       length = PAGE_SIZE - offset;
+               d->req_count = cpu_to_le16(length);
+               d->res_count = d->req_count;
+               d->transfer_status = 0;
+
+               page_bus = page_private(buffer->pages[page]);
+               d->data_address = cpu_to_le32(page_bus + offset);
+
+               dma_sync_single_range_for_device(ctx->context.ohci->card.device,
+                                                page_bus, offset, length,
+                                                DMA_FROM_DEVICE);
+
+               rest -= length;
+               offset = 0;
+               page++;
+
+               context_append(&ctx->context, d, 1, 0);
+       }
+
+       return 0;
+}
+
+static int ohci_queue_iso(struct fw_iso_context *base,
+                         struct fw_iso_packet *packet,
+                         struct fw_iso_buffer *buffer,
+                         unsigned long payload)
+{
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       unsigned long flags;
+       int ret = -ENOSYS;
+
+       spin_lock_irqsave(&ctx->context.ohci->lock, flags);
+       switch (base->type) {
+       case FW_ISO_CONTEXT_TRANSMIT:
+               ret = queue_iso_transmit(ctx, packet, buffer, payload);
+               break;
+       case FW_ISO_CONTEXT_RECEIVE:
+               ret = queue_iso_packet_per_buffer(ctx, packet, buffer, payload);
+               break;
+       case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
+               ret = queue_iso_buffer_fill(ctx, packet, buffer, payload);
+               break;
+       }
+       spin_unlock_irqrestore(&ctx->context.ohci->lock, flags);
+
+       return ret;
+}
+
+static void ohci_flush_queue_iso(struct fw_iso_context *base)
+{
+       struct context *ctx =
+                       &container_of(base, struct iso_context, base)->context;
+
+       reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
+}
+
+static int ohci_flush_iso_completions(struct fw_iso_context *base)
+{
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       int ret = 0;
+
+       tasklet_disable(&ctx->context.tasklet);
+
+       if (!test_and_set_bit_lock(0, &ctx->flushing_completions)) {
+               context_tasklet((unsigned long)&ctx->context);
+
+               switch (base->type) {
+               case FW_ISO_CONTEXT_TRANSMIT:
+               case FW_ISO_CONTEXT_RECEIVE:
+                       if (ctx->header_length != 0)
+                               flush_iso_completions(ctx);
+                       break;
+               case FW_ISO_CONTEXT_RECEIVE_MULTICHANNEL:
+                       if (ctx->mc_completed != 0)
+                               flush_ir_buffer_fill(ctx);
+                       break;
+               default:
+                       ret = -ENOSYS;
+               }
+
+               clear_bit_unlock(0, &ctx->flushing_completions);
+               smp_mb__after_atomic();
+       }
+
+       tasklet_enable(&ctx->context.tasklet);
+
+       return ret;
+}
+
+static const struct fw_card_driver ohci_driver = {
+       .enable                 = ohci_enable,
+       .read_phy_reg           = ohci_read_phy_reg,
+       .update_phy_reg         = ohci_update_phy_reg,
+       .set_config_rom         = ohci_set_config_rom,
+       .send_request           = ohci_send_request,
+       .send_response          = ohci_send_response,
+       .cancel_packet          = ohci_cancel_packet,
+       .enable_phys_dma        = ohci_enable_phys_dma,
+       .read_csr               = ohci_read_csr,
+       .write_csr              = ohci_write_csr,
+
+       .allocate_iso_context   = ohci_allocate_iso_context,
+       .free_iso_context       = ohci_free_iso_context,
+       .set_iso_channels       = ohci_set_iso_channels,
+       .queue_iso              = ohci_queue_iso,
+       .flush_queue_iso        = ohci_flush_queue_iso,
+       .flush_iso_completions  = ohci_flush_iso_completions,
+       .start_iso              = ohci_start_iso,
+       .stop_iso               = ohci_stop_iso,
+};
+
+#ifdef CONFIG_PPC_PMAC
+static void pmac_ohci_on(struct pci_dev *dev)
+{
+       if (machine_is(powermac)) {
+               struct device_node *ofn = pci_device_to_OF_node(dev);
+
+               if (ofn) {
+                       pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 1);
+                       pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
+               }
+       }
+}
+
+static void pmac_ohci_off(struct pci_dev *dev)
+{
+       if (machine_is(powermac)) {
+               struct device_node *ofn = pci_device_to_OF_node(dev);
+
+               if (ofn) {
+                       pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 0);
+                       pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, ofn, 0, 0);
+               }
+       }
+}
+#else
+static inline void pmac_ohci_on(struct pci_dev *dev) {}
+static inline void pmac_ohci_off(struct pci_dev *dev) {}
+#endif /* CONFIG_PPC_PMAC */
+
+static int pci_probe(struct pci_dev *dev,
+                              const struct pci_device_id *ent)
+{
+       struct fw_ohci *ohci;
+       u32 bus_options, max_receive, link_speed, version;
+       u64 guid;
+       int i, err;
+       size_t size;
+
+       if (dev->vendor == PCI_VENDOR_ID_PINNACLE_SYSTEMS) {
+               dev_err(&dev->dev, "Pinnacle MovieBoard is not yet supported\n");
+               return -ENOSYS;
+       }
+
+       ohci = kzalloc(sizeof(*ohci), GFP_KERNEL);
+       if (ohci == NULL) {
+               err = -ENOMEM;
+               goto fail;
+       }
+
+       fw_card_initialize(&ohci->card, &ohci_driver, &dev->dev);
+
+       pmac_ohci_on(dev);
+
+       err = pci_enable_device(dev);
+       if (err) {
+               dev_err(&dev->dev, "failed to enable OHCI hardware\n");
+               goto fail_free;
+       }
+
+       pci_set_master(dev);
+       pci_write_config_dword(dev, OHCI1394_PCI_HCI_Control, 0);
+       pci_set_drvdata(dev, ohci);
+
+       spin_lock_init(&ohci->lock);
+       mutex_init(&ohci->phy_reg_mutex);
+
+       INIT_WORK(&ohci->bus_reset_work, bus_reset_work);
+
+       if (!(pci_resource_flags(dev, 0) & IORESOURCE_MEM) ||
+           pci_resource_len(dev, 0) < OHCI1394_REGISTER_SIZE) {
+               ohci_err(ohci, "invalid MMIO resource\n");
+               err = -ENXIO;
+               goto fail_disable;
+       }
+
+       err = pci_request_region(dev, 0, ohci_driver_name);
+       if (err) {
+               ohci_err(ohci, "MMIO resource unavailable\n");
+               goto fail_disable;
+       }
+
+       ohci->registers = pci_iomap(dev, 0, OHCI1394_REGISTER_SIZE);
+       if (ohci->registers == NULL) {
+               ohci_err(ohci, "failed to remap registers\n");
+               err = -ENXIO;
+               goto fail_iomem;
+       }
+
+       for (i = 0; i < ARRAY_SIZE(ohci_quirks); i++)
+               if ((ohci_quirks[i].vendor == dev->vendor) &&
+                   (ohci_quirks[i].device == (unsigned short)PCI_ANY_ID ||
+                    ohci_quirks[i].device == dev->device) &&
+                   (ohci_quirks[i].revision == (unsigned short)PCI_ANY_ID ||
+                    ohci_quirks[i].revision >= dev->revision)) {
+                       ohci->quirks = ohci_quirks[i].flags;
+                       break;
+               }
+       if (param_quirks)
+               ohci->quirks = param_quirks;
+
+       /*
+        * Because dma_alloc_coherent() allocates at least one page,
+        * we save space by using a common buffer for the AR request/
+        * response descriptors and the self IDs buffer.
+        */
+       BUILD_BUG_ON(AR_BUFFERS * sizeof(struct descriptor) > PAGE_SIZE/4);
+       BUILD_BUG_ON(SELF_ID_BUF_SIZE > PAGE_SIZE/2);
+       ohci->misc_buffer = dma_alloc_coherent(ohci->card.device,
+                                              PAGE_SIZE,
+                                              &ohci->misc_buffer_bus,
+                                              GFP_KERNEL);
+       if (!ohci->misc_buffer) {
+               err = -ENOMEM;
+               goto fail_iounmap;
+       }
+
+       err = ar_context_init(&ohci->ar_request_ctx, ohci, 0,
+                             OHCI1394_AsReqRcvContextControlSet);
+       if (err < 0)
+               goto fail_misc_buf;
+
+       err = ar_context_init(&ohci->ar_response_ctx, ohci, PAGE_SIZE/4,
+                             OHCI1394_AsRspRcvContextControlSet);
+       if (err < 0)
+               goto fail_arreq_ctx;
+
+       err = context_init(&ohci->at_request_ctx, ohci,
+                          OHCI1394_AsReqTrContextControlSet, handle_at_packet);
+       if (err < 0)
+               goto fail_arrsp_ctx;
+
+       err = context_init(&ohci->at_response_ctx, ohci,
+                          OHCI1394_AsRspTrContextControlSet, handle_at_packet);
+       if (err < 0)
+               goto fail_atreq_ctx;
+
+       reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
+       ohci->ir_context_channels = ~0ULL;
+       ohci->ir_context_support = reg_read(ohci, OHCI1394_IsoRecvIntMaskSet);
+       reg_write(ohci, OHCI1394_IsoRecvIntMaskClear, ~0);
+       ohci->ir_context_mask = ohci->ir_context_support;
+       ohci->n_ir = hweight32(ohci->ir_context_mask);
+       size = sizeof(struct iso_context) * ohci->n_ir;
+       ohci->ir_context_list = kzalloc(size, GFP_KERNEL);
+
+       reg_write(ohci, OHCI1394_IsoXmitIntMaskSet, ~0);
+       ohci->it_context_support = reg_read(ohci, OHCI1394_IsoXmitIntMaskSet);
+       reg_write(ohci, OHCI1394_IsoXmitIntMaskClear, ~0);
+       ohci->it_context_mask = ohci->it_context_support;
+       ohci->n_it = hweight32(ohci->it_context_mask);
+       size = sizeof(struct iso_context) * ohci->n_it;
+       ohci->it_context_list = kzalloc(size, GFP_KERNEL);
+
+       if (ohci->it_context_list == NULL || ohci->ir_context_list == NULL) {
+               err = -ENOMEM;
+               goto fail_contexts;
+       }
+
+       ohci->self_id     = ohci->misc_buffer     + PAGE_SIZE/2;
+       ohci->self_id_bus = ohci->misc_buffer_bus + PAGE_SIZE/2;
+
+       bus_options = reg_read(ohci, OHCI1394_BusOptions);
+       max_receive = (bus_options >> 12) & 0xf;
+       link_speed = bus_options & 0x7;
+       guid = ((u64) reg_read(ohci, OHCI1394_GUIDHi) << 32) |
+               reg_read(ohci, OHCI1394_GUIDLo);
+
+       if (!(ohci->quirks & QUIRK_NO_MSI))
+               pci_enable_msi(dev);
+       if (request_irq(dev->irq, irq_handler,
+                       pci_dev_msi_enabled(dev) ? 0 : IRQF_SHARED,
+                       ohci_driver_name, ohci)) {
+               ohci_err(ohci, "failed to allocate interrupt %d\n", dev->irq);
+               err = -EIO;
+               goto fail_msi;
+       }
+
+       err = fw_card_add(&ohci->card, max_receive, link_speed, guid);
+       if (err)
+               goto fail_irq;
+
+       version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
+       ohci_notice(ohci,
+                   "added OHCI v%x.%x device as card %d, "
+                   "%d IR + %d IT contexts, quirks 0x%x%s\n",
+                   version >> 16, version & 0xff, ohci->card.index,
+                   ohci->n_ir, ohci->n_it, ohci->quirks,
+                   reg_read(ohci, OHCI1394_PhyUpperBound) ?
+                       ", physUB" : "");
+
+       return 0;
+
+ fail_irq:
+       free_irq(dev->irq, ohci);
+ fail_msi:
+       pci_disable_msi(dev);
+ fail_contexts:
+       kfree(ohci->ir_context_list);
+       kfree(ohci->it_context_list);
+       context_release(&ohci->at_response_ctx);
+ fail_atreq_ctx:
+       context_release(&ohci->at_request_ctx);
+ fail_arrsp_ctx:
+       ar_context_release(&ohci->ar_response_ctx);
+ fail_arreq_ctx:
+       ar_context_release(&ohci->ar_request_ctx);
+ fail_misc_buf:
+       dma_free_coherent(ohci->card.device, PAGE_SIZE,
+                         ohci->misc_buffer, ohci->misc_buffer_bus);
+ fail_iounmap:
+       pci_iounmap(dev, ohci->registers);
+ fail_iomem:
+       pci_release_region(dev, 0);
+ fail_disable:
+       pci_disable_device(dev);
+ fail_free:
+       kfree(ohci);
+       pmac_ohci_off(dev);
+ fail:
+       return err;
+}
+
+static void pci_remove(struct pci_dev *dev)
+{
+       struct fw_ohci *ohci = pci_get_drvdata(dev);
+
+       /*
+        * If the removal is happening from the suspend state, LPS won't be
+        * enabled and host registers (eg., IntMaskClear) won't be accessible.
+        */
+       if (reg_read(ohci, OHCI1394_HCControlSet) & OHCI1394_HCControl_LPS) {
+               reg_write(ohci, OHCI1394_IntMaskClear, ~0);
+               flush_writes(ohci);
+       }
+       cancel_work_sync(&ohci->bus_reset_work);
+       fw_core_remove_card(&ohci->card);
+
+       /*
+        * FIXME: Fail all pending packets here, now that the upper
+        * layers can't queue any more.
+        */
+
+       software_reset(ohci);
+       free_irq(dev->irq, ohci);
+
+       if (ohci->next_config_rom && ohci->next_config_rom != ohci->config_rom)
+               dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
+                                 ohci->next_config_rom, ohci->next_config_rom_bus);
+       if (ohci->config_rom)
+               dma_free_coherent(ohci->card.device, CONFIG_ROM_SIZE,
+                                 ohci->config_rom, ohci->config_rom_bus);
+       ar_context_release(&ohci->ar_request_ctx);
+       ar_context_release(&ohci->ar_response_ctx);
+       dma_free_coherent(ohci->card.device, PAGE_SIZE,
+                         ohci->misc_buffer, ohci->misc_buffer_bus);
+       context_release(&ohci->at_request_ctx);
+       context_release(&ohci->at_response_ctx);
+       kfree(ohci->it_context_list);
+       kfree(ohci->ir_context_list);
+       pci_disable_msi(dev);
+       pci_iounmap(dev, ohci->registers);
+       pci_release_region(dev, 0);
+       pci_disable_device(dev);
+       kfree(ohci);
+       pmac_ohci_off(dev);
+
+       dev_notice(&dev->dev, "removed fw-ohci device\n");
+}
+
+#ifdef CONFIG_PM
+static int pci_suspend(struct pci_dev *dev, pm_message_t state)
+{
+       struct fw_ohci *ohci = pci_get_drvdata(dev);
+       int err;
+
+       software_reset(ohci);
+       err = pci_save_state(dev);
+       if (err) {
+               ohci_err(ohci, "pci_save_state failed\n");
+               return err;
+       }
+       err = pci_set_power_state(dev, pci_choose_state(dev, state));
+       if (err)
+               ohci_err(ohci, "pci_set_power_state failed with %d\n", err);
+       pmac_ohci_off(dev);
+
+       return 0;
+}
+
+static int pci_resume(struct pci_dev *dev)
+{
+       struct fw_ohci *ohci = pci_get_drvdata(dev);
+       int err;
+
+       pmac_ohci_on(dev);
+       pci_set_power_state(dev, PCI_D0);
+       pci_restore_state(dev);
+       err = pci_enable_device(dev);
+       if (err) {
+               ohci_err(ohci, "pci_enable_device failed\n");
+               return err;
+       }
+
+       /* Some systems don't setup GUID register on resume from ram  */
+       if (!reg_read(ohci, OHCI1394_GUIDLo) &&
+                                       !reg_read(ohci, OHCI1394_GUIDHi)) {
+               reg_write(ohci, OHCI1394_GUIDLo, (u32)ohci->card.guid);
+               reg_write(ohci, OHCI1394_GUIDHi, (u32)(ohci->card.guid >> 32));
+       }
+
+       err = ohci_enable(&ohci->card, NULL, 0);
+       if (err)
+               return err;
+
+       ohci_resume_iso_dma(ohci);
+
+       return 0;
+}
+#endif
+
+static const struct pci_device_id pci_table[] = {
+       { PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_FIREWIRE_OHCI, ~0) },
+       { }
+};
+
+MODULE_DEVICE_TABLE(pci, pci_table);
+
+static struct pci_driver fw_ohci_pci_driver = {
+       .name           = ohci_driver_name,
+       .id_table       = pci_table,
+       .probe          = pci_probe,
+       .remove         = pci_remove,
+#ifdef CONFIG_PM
+       .resume         = pci_resume,
+       .suspend        = pci_suspend,
+#endif
+};
+
+static int __init fw_ohci_init(void)
+{
+       selfid_workqueue = alloc_workqueue(KBUILD_MODNAME, WQ_MEM_RECLAIM, 0);
+       if (!selfid_workqueue)
+               return -ENOMEM;
+
+       return pci_register_driver(&fw_ohci_pci_driver);
+}
+
+static void __exit fw_ohci_cleanup(void)
+{
+       pci_unregister_driver(&fw_ohci_pci_driver);
+       destroy_workqueue(selfid_workqueue);
+}
+
+module_init(fw_ohci_init);
+module_exit(fw_ohci_cleanup);
+
+MODULE_AUTHOR("Kristian Hoegsberg <krh@bitplanet.net>");
+MODULE_DESCRIPTION("Driver for PCI OHCI IEEE1394 controllers");
+MODULE_LICENSE("GPL");
+
+/* Provide a module alias so root-on-sbp2 initrds don't break. */
+MODULE_ALIAS("ohci1394");