Merge branches 'release', 'asus', 'bugzilla-8573', 'bugzilla-9995', 'bugzilla-10272...
[sfrench/cifs-2.6.git] / drivers / firewire / fw-ohci.c
index 2f307c4df33563217ae7acbb85ae4c89224b866c..996d61f0d4602cf0cd942aea5daf54e08739c1cb 100644 (file)
 #include <asm/page.h>
 #include <asm/system.h>
 
+#ifdef CONFIG_PPC_PMAC
+#include <asm/pmac_feature.h>
+#endif
+
 #include "fw-ohci.h"
 #include "fw-transaction.h"
 
@@ -98,17 +102,48 @@ 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;
 
-       struct descriptor *buffer;
-       dma_addr_t buffer_bus;
-       size_t buffer_size;
-       struct descriptor *head_descriptor;
-       struct descriptor *tail_descriptor;
-       struct descriptor *tail_descriptor_last;
-       struct descriptor *prev_descriptor;
+       /*
+        * 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 in the DMA program.  It contains the branch
+        * address that must be updated upon appending a new descriptor.
+        */
+       struct descriptor *prev;
 
        descriptor_callback_t callback;
 
@@ -125,6 +160,7 @@ struct context {
 struct iso_context {
        struct fw_iso_context base;
        struct context context;
+       int excess_bytes;
        void *header;
        size_t header_length;
 };
@@ -143,6 +179,7 @@ struct fw_ohci {
        int generation;
        int request_generation;
        u32 bus_seconds;
+       bool old_uninorth;
 
        /*
         * Spinlock for accessing fw_ohci data.  Never call out of
@@ -197,8 +234,6 @@ static inline struct fw_ohci *fw_ohci(struct fw_card *card)
 #define SELF_ID_BUF_SIZE               0x800
 #define OHCI_TCODE_PHY_PACKET          0x0e
 #define OHCI_VERSION_1_1               0x010010
-#define ISO_BUFFER_SIZE                        (64 * 1024)
-#define AT_BUFFER_SIZE                 4096
 
 static char ohci_driver_name[] = KBUILD_MODNAME;
 
@@ -246,19 +281,13 @@ static int ar_context_add_page(struct ar_context *ctx)
 {
        struct device *dev = ctx->ohci->card.device;
        struct ar_buffer *ab;
-       dma_addr_t ab_bus;
+       dma_addr_t uninitialized_var(ab_bus);
        size_t offset;
 
-       ab = (struct ar_buffer *) __get_free_page(GFP_ATOMIC);
+       ab = dma_alloc_coherent(dev, PAGE_SIZE, &ab_bus, GFP_ATOMIC);
        if (ab == NULL)
                return -ENOMEM;
 
-       ab_bus = dma_map_single(dev, ab, PAGE_SIZE, DMA_BIDIRECTIONAL);
-       if (dma_mapping_error(ab_bus)) {
-               free_page((unsigned long) ab);
-               return -ENOMEM;
-       }
-
        memset(&ab->descriptor, 0, sizeof(ab->descriptor));
        ab->descriptor.control        = cpu_to_le16(DESCRIPTOR_INPUT_MORE |
                                                    DESCRIPTOR_STATUS |
@@ -269,8 +298,6 @@ static int ar_context_add_page(struct ar_context *ctx)
        ab->descriptor.res_count      = cpu_to_le16(PAGE_SIZE - offset);
        ab->descriptor.branch_address = 0;
 
-       dma_sync_single_for_device(dev, ab_bus, PAGE_SIZE, DMA_BIDIRECTIONAL);
-
        ctx->last_buffer->descriptor.branch_address = cpu_to_le32(ab_bus | 1);
        ctx->last_buffer->next = ab;
        ctx->last_buffer = ab;
@@ -281,15 +308,22 @@ static int ar_context_add_page(struct ar_context *ctx)
        return 0;
 }
 
+#if defined(CONFIG_PPC_PMAC) && defined(CONFIG_PPC32)
+#define cond_le32_to_cpu(v) \
+       (ohci->old_uninorth ? (__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;
 
-       p.header[0] = le32_to_cpu(buffer[0]);
-       p.header[1] = le32_to_cpu(buffer[1]);
-       p.header[2] = le32_to_cpu(buffer[2]);
+       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) {
@@ -301,7 +335,7 @@ static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
                break;
 
        case TCODE_READ_BLOCK_REQUEST :
-               p.header[3] = le32_to_cpu(buffer[3]);
+               p.header[3] = cond_le32_to_cpu(buffer[3]);
                p.header_length = 16;
                p.payload_length = 0;
                break;
@@ -310,7 +344,7 @@ static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
        case TCODE_READ_BLOCK_RESPONSE:
        case TCODE_LOCK_REQUEST:
        case TCODE_LOCK_RESPONSE:
-               p.header[3] = le32_to_cpu(buffer[3]);
+               p.header[3] = cond_le32_to_cpu(buffer[3]);
                p.header_length = 16;
                p.payload_length = p.header[3] >> 16;
                break;
@@ -327,7 +361,7 @@ static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
 
        /* FIXME: What to do about evt_* errors? */
        length = (p.header_length + p.payload_length + 3) / 4;
-       status = le32_to_cpu(buffer[length]);
+       status = cond_le32_to_cpu(buffer[length]);
 
        p.ack        = ((status >> 16) & 0x1f) - 16;
        p.speed      = (status >> 21) & 0x7;
@@ -345,7 +379,7 @@ static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
         */
 
        if (p.ack + 16 == 0x09)
-               ohci->request_generation = (buffer[2] >> 16) & 0xff;
+               ohci->request_generation = (p.header[2] >> 16) & 0xff;
        else if (ctx == &ohci->ar_request_ctx)
                fw_core_handle_request(&ohci->card, &p);
        else
@@ -367,6 +401,7 @@ static void ar_context_tasklet(unsigned long data)
 
        if (d->res_count == 0) {
                size_t size, rest, offset;
+               dma_addr_t buffer_bus;
 
                /*
                 * This descriptor is finished and we may have a
@@ -375,9 +410,7 @@ static void ar_context_tasklet(unsigned long data)
                 */
 
                offset = offsetof(struct ar_buffer, data);
-               dma_unmap_single(ohci->card.device,
-                       le32_to_cpu(ab->descriptor.data_address) - offset,
-                       PAGE_SIZE, DMA_BIDIRECTIONAL);
+               buffer_bus = le32_to_cpu(ab->descriptor.data_address) - offset;
 
                buffer = ab;
                ab = ab->next;
@@ -393,7 +426,8 @@ static void ar_context_tasklet(unsigned long data)
                while (buffer < end)
                        buffer = handle_ar_packet(ctx, buffer);
 
-               free_page((unsigned long)buffer);
+               dma_free_coherent(ohci->card.device, PAGE_SIZE,
+                                 buffer, buffer_bus);
                ar_context_add_page(ctx);
        } else {
                buffer = ctx->pointer;
@@ -437,74 +471,126 @@ static void ar_context_run(struct ar_context *ctx)
        flush_writes(ctx->ohci);
 }
 
+static struct descriptor *
+find_branch_descriptor(struct descriptor *d, int z)
+{
+       int b, key;
+
+       b   = (le16_to_cpu(d->control) & DESCRIPTOR_BRANCH_ALWAYS) >> 2;
+       key = (le16_to_cpu(d->control) & DESCRIPTOR_KEY_IMMEDIATE) >> 8;
+
+       /* figure out which descriptor the branch address goes in */
+       if (z == 2 && (b == 3 || key == 2))
+               return d;
+       else
+               return d + z - 1;
+}
+
 static void context_tasklet(unsigned long data)
 {
        struct context *ctx = (struct context *) data;
-       struct fw_ohci *ohci = ctx->ohci;
        struct descriptor *d, *last;
        u32 address;
        int z;
+       struct descriptor_buffer *desc;
 
-       dma_sync_single_for_cpu(ohci->card.device, ctx->buffer_bus,
-                               ctx->buffer_size, DMA_TO_DEVICE);
-
-       d    = ctx->tail_descriptor;
-       last = ctx->tail_descriptor_last;
-
+       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;
-               d = ctx->buffer + (address - ctx->buffer_bus) / sizeof(*d);
-               last = (z == 2) ? d : d + z - 1;
+               address &= ~0xf;
+
+               /* 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;
 
-               ctx->tail_descriptor      = d;
-               ctx->tail_descriptor_last = last;
+               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,
-            size_t buffer_size, u32 regs,
-            descriptor_callback_t callback)
+            u32 regs, descriptor_callback_t callback)
 {
        ctx->ohci = ohci;
        ctx->regs = regs;
-       ctx->buffer_size = buffer_size;
-       ctx->buffer = kmalloc(buffer_size, GFP_KERNEL);
-       if (ctx->buffer == NULL)
+       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;
 
-       ctx->buffer_bus =
-               dma_map_single(ohci->card.device, ctx->buffer,
-                              buffer_size, DMA_TO_DEVICE);
-       if (dma_mapping_error(ctx->buffer_bus)) {
-               kfree(ctx->buffer);
-               return -ENOMEM;
-       }
-
-       ctx->head_descriptor      = ctx->buffer;
-       ctx->prev_descriptor      = ctx->buffer;
-       ctx->tail_descriptor      = ctx->buffer;
-       ctx->tail_descriptor_last = ctx->buffer;
-
        /*
         * We put a dummy descriptor in the buffer that has a NULL
         * branch address and looks like it's been sent.  That way we
-        * have a descriptor to append DMA programs to.  Also, the
-        * ring buffer invariant is that it always has at least one
-        * element so that head == tail means buffer full.
+        * have a descriptor to append DMA programs to.
         */
-
-       memset(ctx->head_descriptor, 0, sizeof(*ctx->head_descriptor));
-       ctx->head_descriptor->control = cpu_to_le16(DESCRIPTOR_OUTPUT_LAST);
-       ctx->head_descriptor->transfer_status = cpu_to_le16(0x8011);
-       ctx->head_descriptor++;
+       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;
 
        return 0;
 }
@@ -513,35 +599,42 @@ static void
 context_release(struct context *ctx)
 {
        struct fw_card *card = &ctx->ohci->card;
+       struct descriptor_buffer *desc, *tmp;
 
-       dma_unmap_single(card->device, ctx->buffer_bus,
-                        ctx->buffer_size, DMA_TO_DEVICE);
-       kfree(ctx->buffer);
+       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, *tail, *end;
-
-       d = ctx->head_descriptor;
-       tail = ctx->tail_descriptor;
-       end = ctx->buffer + ctx->buffer_size / sizeof(*d);
-
-       if (d + z <= tail) {
-               goto has_space;
-       } else if (d > tail && d + z <= end) {
-               goto has_space;
-       } else if (d > tail && ctx->buffer + z <= tail) {
-               d = ctx->buffer;
-               goto has_space;
-       }
+       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. */
 
-       return NULL;
+               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;
+       }
 
- has_space:
+       d = desc->buffer + desc->used / sizeof(*d);
        memset(d, 0, z * sizeof(*d));
-       *d_bus = ctx->buffer_bus + (d - ctx->buffer) * sizeof(*d);
+       *d_bus = desc->buffer_bus + desc->used;
 
        return d;
 }
@@ -551,7 +644,7 @@ 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->tail_descriptor_last->branch_address));
+                 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);
        flush_writes(ohci);
@@ -561,15 +654,13 @@ 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;
 
-       d_bus = ctx->buffer_bus + (d - ctx->buffer) * sizeof(*d);
+       d_bus = desc->buffer_bus + (d - desc->buffer) * sizeof(*d);
 
-       ctx->head_descriptor = d + z + extra;
-       ctx->prev_descriptor->branch_address = cpu_to_le32(d_bus | z);
-       ctx->prev_descriptor = z == 2 ? d : d + z - 1;
-
-       dma_sync_single_for_device(ctx->ohci->card.device, ctx->buffer_bus,
-                                  ctx->buffer_size, DMA_TO_DEVICE);
+       desc->used += (z + extra) * sizeof(*d);
+       ctx->prev->branch_address = cpu_to_le32(d_bus | z);
+       ctx->prev = find_branch_descriptor(d, z);
 
        reg_write(ctx->ohci, CONTROL_SET(ctx->regs), CONTEXT_WAKE);
        flush_writes(ctx->ohci);
@@ -606,7 +697,7 @@ static int
 at_context_queue_packet(struct context *ctx, struct fw_packet *packet)
 {
        struct fw_ohci *ohci = ctx->ohci;
-       dma_addr_t d_bus, payload_bus;
+       dma_addr_t d_bus, uninitialized_var(payload_bus);
        struct driver_data *driver_data;
        struct descriptor *d, *last;
        __le32 *header;
@@ -655,7 +746,7 @@ at_context_queue_packet(struct context *ctx, struct fw_packet *packet)
        driver_data = (struct driver_data *) &d[3];
        driver_data->packet = packet;
        packet->driver_data = driver_data;
-       
+
        if (packet->payload_length > 0) {
                payload_bus =
                        dma_map_single(ohci->card.device, packet->payload,
@@ -903,7 +994,7 @@ at_context_transmit(struct context *ctx, struct fw_packet *packet)
 
        if (retval < 0)
                packet->callback(packet, &ctx->ohci->card, packet->ack);
-       
+
 }
 
 static void bus_reset_tasklet(unsigned long data)
@@ -935,13 +1026,14 @@ static void bus_reset_tasklet(unsigned long data)
         */
 
        self_id_count = (reg_read(ohci, OHCI1394_SelfIDCount) >> 3) & 0x3ff;
-       generation = (le32_to_cpu(ohci->self_id_cpu[0]) >> 16) & 0xff;
+       generation = (cond_le32_to_cpu(ohci->self_id_cpu[0]) >> 16) & 0xff;
        rmb();
 
        for (i = 1, j = 0; j < self_id_count; i += 2, j++) {
                if (ohci->self_id_cpu[i] != ~ohci->self_id_cpu[i + 1])
                        fw_error("inconsistent self IDs\n");
-               ohci->self_id_buffer[j] = le32_to_cpu(ohci->self_id_cpu[i]);
+               ohci->self_id_buffer[j] =
+                               cond_le32_to_cpu(ohci->self_id_cpu[i]);
        }
        rmb();
 
@@ -984,8 +1076,10 @@ static void bus_reset_tasklet(unsigned long data)
         */
 
        if (ohci->next_config_rom != NULL) {
-               free_rom     = ohci->config_rom;
-               free_rom_bus = ohci->config_rom_bus;
+               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;
@@ -1061,6 +1155,13 @@ static irqreturn_t irq_handler(int irq, void *data)
        if (unlikely(event & OHCI1394_postedWriteErr))
                fw_error("PCI posted write error\n");
 
+       if (unlikely(event & OHCI1394_cycleTooLong)) {
+               if (printk_ratelimit())
+                       fw_notify("isochronous cycle too long\n");
+               reg_write(ohci, OHCI1394_LinkControlSet,
+                         OHCI1394_LinkControl_cycleMaster);
+       }
+
        if (event & OHCI1394_cycle64Seconds) {
                cycle_time = reg_read(ohci, OHCI1394_IsochronousCycleTimer);
                if ((cycle_time & 0x80000000) == 0)
@@ -1134,8 +1235,8 @@ static int ohci_enable(struct fw_card *card, u32 *config_rom, size_t length)
                  OHCI1394_RQPkt | OHCI1394_RSPkt |
                  OHCI1394_reqTxComplete | OHCI1394_respTxComplete |
                  OHCI1394_isochRx | OHCI1394_isochTx |
-                 OHCI1394_postedWriteErr | OHCI1394_cycle64Seconds |
-                 OHCI1394_masterIntEnable);
+                 OHCI1394_postedWriteErr | OHCI1394_cycleTooLong |
+                 OHCI1394_cycle64Seconds | OHCI1394_masterIntEnable);
 
        /* Activate link_on bit and contender bit in our self ID packets.*/
        if (ohci_update_phy_reg(card, 4, 0,
@@ -1161,19 +1262,30 @@ static int ohci_enable(struct fw_card *card, u32 *config_rom, size_t length)
         * the right values in the bus reset tasklet.
         */
 
-       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;
+       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;
 
-       memset(ohci->next_config_rom, 0, CONFIG_ROM_SIZE);
-       fw_memcpy_to_be32(ohci->next_config_rom, config_rom, length * 4);
+               memset(ohci->next_config_rom, 0, CONFIG_ROM_SIZE);
+               fw_memcpy_to_be32(ohci->next_config_rom, config_rom, length * 4);
+       } 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 = config_rom[0];
+       ohci->next_header = be32_to_cpu(ohci->next_config_rom[0]);
        ohci->next_config_rom[0] = 0;
        reg_write(ohci, OHCI1394_ConfigROMhdr, 0);
-       reg_write(ohci, OHCI1394_BusOptions, config_rom[2]);
+       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);
@@ -1209,7 +1321,7 @@ ohci_set_config_rom(struct fw_card *card, u32 *config_rom, size_t length)
        unsigned long flags;
        int retval = -EBUSY;
        __be32 *next_config_rom;
-       dma_addr_t next_config_rom_bus;
+       dma_addr_t uninitialized_var(next_config_rom_bus);
 
        ohci = fw_ohci(card);
 
@@ -1380,9 +1492,13 @@ static int handle_ir_dualbuffer_packet(struct context *context,
        void *p, *end;
        int i;
 
-       if (db->first_res_count > 0 && db->second_res_count > 0)
-               /* This descriptor isn't done yet, stop iteration. */
-               return 0;
+       if (db->first_res_count != 0 && db->second_res_count != 0) {
+               if (ctx->excess_bytes <= le16_to_cpu(db->second_req_count)) {
+                       /* This descriptor isn't done yet, stop iteration. */
+                       return 0;
+               }
+               ctx->excess_bytes -= le16_to_cpu(db->second_req_count);
+       }
 
        header_length = le16_to_cpu(db->first_req_count) -
                le16_to_cpu(db->first_res_count);
@@ -1401,11 +1517,15 @@ static int handle_ir_dualbuffer_packet(struct context *context,
                *(u32 *) (ctx->header + i) = __swab32(*(u32 *) (p + 4));
                memcpy(ctx->header + i + 4, p + 8, ctx->base.header_size - 4);
                i += ctx->base.header_size;
+               ctx->excess_bytes +=
+                       (le32_to_cpu(*(__le32 *)(p + 4)) >> 16) & 0xffff;
                p += ctx->base.header_size + 4;
        }
-
        ctx->header_length = i;
 
+       ctx->excess_bytes -= le16_to_cpu(db->second_req_count) -
+               le16_to_cpu(db->second_res_count);
+
        if (le16_to_cpu(db->control) & DESCRIPTOR_IRQ_ALWAYS) {
                ir_header = (__le32 *) (db + 1);
                ctx->base.callback(&ctx->base,
@@ -1418,6 +1538,53 @@ static int handle_ir_dualbuffer_packet(struct context *context,
        return 1;
 }
 
+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;
+       __le32 *ir_header;
+       void *p;
+       int i;
+
+       for (pd = d; pd <= last; pd++) {
+               if (pd->transfer_status)
+                       break;
+       }
+       if (pd > last)
+               /* Descriptor(s) not done yet, stop iteration */
+               return 0;
+
+       i   = ctx->header_length;
+       p   = last + 1;
+
+       if (ctx->base.header_size > 0 &&
+                       i + ctx->base.header_size <= PAGE_SIZE) {
+               /*
+                * The iso header is byteswapped to little endian by
+                * the controller, but the remaining header quadlets
+                * are big endian.  We want to present all the headers
+                * as big endian, so we have to swap the first quadlet.
+                */
+               *(u32 *) (ctx->header + i) = __swab32(*(u32 *) (p + 4));
+               memcpy(ctx->header + i + 4, p + 8, ctx->base.header_size - 4);
+               ctx->header_length += ctx->base.header_size;
+       }
+
+       if (le16_to_cpu(last->control) & DESCRIPTOR_IRQ_ALWAYS) {
+               ir_header = (__le32 *) p;
+               ctx->base.callback(&ctx->base,
+                                  le32_to_cpu(ir_header[0]) & 0xffff,
+                                  ctx->header_length, ctx->header,
+                                  ctx->base.callback_data);
+               ctx->header_length = 0;
+       }
+
+       return 1;
+}
+
 static int handle_it_packet(struct context *context,
                            struct descriptor *d,
                            struct descriptor *last)
@@ -1453,14 +1620,12 @@ ohci_allocate_iso_context(struct fw_card *card, int type, size_t header_size)
        } else {
                mask = &ohci->ir_context_mask;
                list = ohci->ir_context_list;
-               callback = handle_ir_dualbuffer_packet;
+               if (ohci->version >= OHCI_VERSION_1_1)
+                       callback = handle_ir_dualbuffer_packet;
+               else
+                       callback = handle_ir_packet_per_buffer;
        }
 
-       /* FIXME: We need a fallback for pre 1.1 OHCI. */
-       if (callback == handle_ir_dualbuffer_packet &&
-           ohci->version < OHCI_VERSION_1_1)
-               return ERR_PTR(-EINVAL);
-
        spin_lock_irqsave(&ohci->lock, flags);
        index = ffs(*mask) - 1;
        if (index >= 0)
@@ -1482,8 +1647,7 @@ ohci_allocate_iso_context(struct fw_card *card, int type, size_t header_size)
        if (ctx->header == NULL)
                goto out;
 
-       retval = context_init(&ctx->context, ohci, ISO_BUFFER_SIZE,
-                             regs, callback);
+       retval = context_init(&ctx->context, ohci, regs, callback);
        if (retval < 0)
                goto out_with_header;
 
@@ -1519,7 +1683,9 @@ static int ohci_start_iso(struct fw_iso_context *base,
                context_run(&ctx->context, match);
        } else {
                index = ctx - ohci->ir_context_list;
-               control = IR_CONTEXT_DUAL_BUFFER_MODE | IR_CONTEXT_ISOCH_HEADER;
+               control = IR_CONTEXT_ISOCH_HEADER;
+               if (ohci->version >= OHCI_VERSION_1_1)
+                       control |= IR_CONTEXT_DUAL_BUFFER_MODE;
                match = (tags << 28) | (sync << 8) | ctx->base.channel;
                if (cycle >= 0) {
                        match |= (cycle & 0x07fff) << 12;
@@ -1696,19 +1862,6 @@ ohci_queue_iso_receive_dualbuffer(struct fw_iso_context *base,
         * packet, retransmit or terminate..
         */
 
-       if (packet->skip) {
-               d = context_get_descriptors(&ctx->context, 2, &d_bus);
-               if (d == NULL)
-                       return -ENOMEM;
-
-               db = (struct db_descriptor *) d;
-               db->control = cpu_to_le16(DESCRIPTOR_STATUS |
-                                         DESCRIPTOR_BRANCH_ALWAYS |
-                                         DESCRIPTOR_WAIT);
-               db->first_size = cpu_to_le16(ctx->base.header_size + 4);
-               context_append(&ctx->context, d, 2, 0);
-       }
-
        p = packet;
        z = 2;
 
@@ -1725,7 +1878,6 @@ ohci_queue_iso_receive_dualbuffer(struct fw_iso_context *base,
        offset   = payload & ~PAGE_MASK;
        rest     = p->payload_length;
 
-       /* FIXME: OHCI 1.0 doesn't support dual buffer receive */
        /* FIXME: make packet-per-buffer/dual-buffer a context option */
        while (rest > 0) {
                d = context_get_descriptors(&ctx->context,
@@ -1737,11 +1889,18 @@ ohci_queue_iso_receive_dualbuffer(struct fw_iso_context *base,
                db->control = cpu_to_le16(DESCRIPTOR_STATUS |
                                          DESCRIPTOR_BRANCH_ALWAYS);
                db->first_size = cpu_to_le16(ctx->base.header_size + 4);
-               db->first_req_count = cpu_to_le16(header_size);
+               if (p->skip && rest == p->payload_length) {
+                       db->control |= cpu_to_le16(DESCRIPTOR_WAIT);
+                       db->first_req_count = db->first_size;
+               } else {
+                       db->first_req_count = cpu_to_le16(header_size);
+               }
                db->first_res_count = db->first_req_count;
                db->first_buffer = cpu_to_le32(d_bus + sizeof(*db));
 
-               if (offset + rest < PAGE_SIZE)
+               if (p->skip && rest == p->payload_length)
+                       length = 4;
+               else if (offset + rest < PAGE_SIZE)
                        length = rest;
                else
                        length = PAGE_SIZE - offset;
@@ -1757,7 +1916,86 @@ ohci_queue_iso_receive_dualbuffer(struct fw_iso_context *base,
                context_append(&ctx->context, d, z, header_z);
                offset = (offset + length) & ~PAGE_MASK;
                rest -= length;
-               page++;
+               if (offset == 0)
+                       page++;
+       }
+
+       return 0;
+}
+
+static int
+ohci_queue_iso_receive_packet_per_buffer(struct fw_iso_context *base,
+                                        struct fw_iso_packet *packet,
+                                        struct fw_iso_buffer *buffer,
+                                        unsigned long payload)
+{
+       struct iso_context *ctx = container_of(base, struct iso_context, base);
+       struct descriptor *d = NULL, *pd = NULL;
+       struct fw_iso_packet *p = packet;
+       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 status word in the
+        * buffer too, so we need 4 extra bytes per packet.
+        */
+       packet_count = p->header_length / ctx->base.header_size;
+       header_size  = ctx->base.header_size + 4;
+
+       /* 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 = p->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 (p->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;
+               for (j = 1; j < z; j++) {
+                       pd = d + j;
+                       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);
+
+                       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 (p->interrupt && i == packet_count - 1)
+                       pd->control |= cpu_to_le16(DESCRIPTOR_IRQ_ALWAYS);
+
+               context_append(&ctx->context, d, z, header_z);
        }
 
        return 0;
@@ -1770,15 +2008,22 @@ ohci_queue_iso(struct fw_iso_context *base,
               unsigned long payload)
 {
        struct iso_context *ctx = container_of(base, struct iso_context, base);
+       unsigned long flags;
+       int retval;
 
+       spin_lock_irqsave(&ctx->context.ohci->lock, flags);
        if (base->type == FW_ISO_CONTEXT_TRANSMIT)
-               return ohci_queue_iso_transmit(base, packet, buffer, payload);
+               retval = ohci_queue_iso_transmit(base, packet, buffer, payload);
        else if (ctx->context.ohci->version >= OHCI_VERSION_1_1)
-               return ohci_queue_iso_receive_dualbuffer(base, packet,
+               retval = ohci_queue_iso_receive_dualbuffer(base, packet,
                                                         buffer, payload);
        else
-               /* FIXME: Implement fallback for OHCI 1.0 controllers. */
-               return -EINVAL;
+               retval = ohci_queue_iso_receive_packet_per_buffer(base, packet,
+                                                               buffer,
+                                                               payload);
+       spin_unlock_irqrestore(&ctx->context.ohci->lock, flags);
+
+       return retval;
 }
 
 static const struct fw_card_driver ohci_driver = {
@@ -1808,6 +2053,18 @@ pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
        int err;
        size_t size;
 
+#ifdef CONFIG_PPC_PMAC
+       /* Necessary on some machines if fw-ohci was loaded/ unloaded before */
+       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);
+               }
+       }
+#endif /* CONFIG_PPC_PMAC */
+
        ohci = kzalloc(sizeof(*ohci), GFP_KERNEL);
        if (ohci == NULL) {
                fw_error("Could not malloc fw_ohci data.\n");
@@ -1826,6 +2083,10 @@ pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
        pci_write_config_dword(dev, OHCI1394_PCI_HCI_Control, 0);
        pci_set_drvdata(dev, ohci);
 
+#if defined(CONFIG_PPC_PMAC) && defined(CONFIG_PPC32)
+       ohci->old_uninorth = dev->vendor == PCI_VENDOR_ID_APPLE &&
+                            dev->device == PCI_DEVICE_ID_APPLE_UNI_N_FW;
+#endif
        spin_lock_init(&ohci->lock);
 
        tasklet_init(&ohci->bus_reset_tasklet,
@@ -1850,10 +2111,10 @@ pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
        ar_context_init(&ohci->ar_response_ctx, ohci,
                        OHCI1394_AsRspRcvContextControlSet);
 
-       context_init(&ohci->at_request_ctx, ohci, AT_BUFFER_SIZE,
+       context_init(&ohci->at_request_ctx, ohci,
                     OHCI1394_AsReqTrContextControlSet, handle_at_packet);
 
-       context_init(&ohci->at_response_ctx, ohci, AT_BUFFER_SIZE,
+       context_init(&ohci->at_response_ctx, ohci,
                     OHCI1394_AsRspTrContextControlSet, handle_at_packet);
 
        reg_write(ohci, OHCI1394_IsoRecvIntMaskSet, ~0);
@@ -1898,7 +2159,6 @@ pci_probe(struct pci_dev *dev, const struct pci_device_id *ent)
        ohci->version = reg_read(ohci, OHCI1394_Version) & 0x00ff00ff;
        fw_notify("Added fw-ohci device %s, OHCI version %x.%x\n",
                  dev->dev.bus_id, ohci->version >> 16, ohci->version & 0xff);
-
        return 0;
 
  fail_self_id:
@@ -1943,6 +2203,19 @@ static void pci_remove(struct pci_dev *dev)
        pci_disable_device(dev);
        fw_card_put(&ohci->card);
 
+#ifdef CONFIG_PPC_PMAC
+       /* On UniNorth, power down the cable and turn off the chip clock
+        * to save power on laptops */
+       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);
+               }
+       }
+#endif /* CONFIG_PPC_PMAC */
+
        fw_notify("Removed fw-ohci device.\n");
 }
 
@@ -1963,6 +2236,16 @@ static int pci_suspend(struct pci_dev *pdev, pm_message_t state)
        if (err)
                fw_error("pci_set_power_state failed with %d\n", err);
 
+/* PowerMac suspend code comes last */
+#ifdef CONFIG_PPC_PMAC
+       if (machine_is(powermac)) {
+               struct device_node *ofn = pci_device_to_OF_node(pdev);
+
+               if (ofn)
+                       pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 0);
+       }
+#endif /* CONFIG_PPC_PMAC */
+
        return 0;
 }
 
@@ -1971,6 +2254,16 @@ static int pci_resume(struct pci_dev *pdev)
        struct fw_ohci *ohci = pci_get_drvdata(pdev);
        int err;
 
+/* PowerMac resume code comes first */
+#ifdef CONFIG_PPC_PMAC
+       if (machine_is(powermac)) {
+               struct device_node *ofn = pci_device_to_OF_node(pdev);
+
+               if (ofn)
+                       pmac_call_feature(PMAC_FTR_1394_ENABLE, ofn, 0, 1);
+       }
+#endif /* CONFIG_PPC_PMAC */
+
        pci_set_power_state(pdev, PCI_D0);
        pci_restore_state(pdev);
        err = pci_enable_device(pdev);
@@ -1979,7 +2272,7 @@ static int pci_resume(struct pci_dev *pdev)
                return err;
        }
 
-       return ohci_enable(&ohci->card, ohci->config_rom, CONFIG_ROM_SIZE);
+       return ohci_enable(&ohci->card, NULL, 0);
 }
 #endif