Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394:
  firewire: ohci: fix isochronous DMA synchronization
  firewire: ohci: work around selfID junk due to wrong gap count
  firewire: net: Use posted writes
  firewire: use clamp and min3 macros
  firewire: ohci: optimize TSB41BA3D detection
  firewire: ohci: TSB41BA3D support tweaks
  firewire: ohci: Add support for TSB41BA3D phy
  firewire: ohci: Move code from the bus reset tasklet into a workqueue
  firewire: sbp2: fold two functions into one
  firewire: sbp2: move some code to more sensible places
  firewire: sbp2: remove obsolete reference counting

drivers/firewire/core-transaction.c

@@ -1046,8 +1046,8 @@ static void update_split_timeout(struct fw_card *card)
 
 	cycles = card->split_timeout_hi * 8000 + (card->split_timeout_lo >> 19);
 
-	cycles = max(cycles, 800u); /* minimum as per the spec */
-	cycles = min(cycles, 3u * 8000u); /* maximum OHCI timeout */
+	/* minimum per IEEE 1394, maximum which doesn't overflow OHCI */
+	cycles = clamp(cycles, 800u, 3u * 8000u);
 
 	card->split_timeout_cycles = cycles;
 	card->split_timeout_jiffies = DIV_ROUND_UP(cycles * HZ, 8000);

drivers/firewire/net.c

@@ -502,11 +502,7 @@ static struct fwnet_peer *fwnet_peer_find_by_node_id(struct fwnet_device *dev,
 static unsigned fwnet_max_payload(unsigned max_rec, unsigned speed)
 {
 	max_rec = min(max_rec, speed + 8);
-	max_rec = min(max_rec, 0xbU); /* <= 4096 */
-	if (max_rec < 8) {
-		fw_notify("max_rec %x out of range\n", max_rec);
-		max_rec = 8;
-	}
+	max_rec = clamp(max_rec, 8U, 11U); /* 512...4096 */
 
 	return (1 << (max_rec + 1)) - RFC2374_FRAG_HDR_SIZE;
 }
@@ -1125,17 +1121,12 @@ static int fwnet_broadcast_start(struct fwnet_device *dev)
 	unsigned u;
 
 	if (dev->local_fifo == FWNET_NO_FIFO_ADDR) {
-		/* outside OHCI posted write area? */
-		static const struct fw_address_region region = {
-			.start = 0xffff00000000ULL,
-			.end   = CSR_REGISTER_BASE,
-		};
-
 		dev->handler.length = 4096;
 		dev->handler.address_callback = fwnet_receive_packet;
 		dev->handler.callback_data = dev;
 
-		retval = fw_core_add_address_handler(&dev->handler, &region);
+		retval = fw_core_add_address_handler(&dev->handler,
+					&fw_high_memory_region);
 		if (retval < 0)
 			goto failed_initial;
 

drivers/firewire/ohci.c

@@ -42,6 +42,7 @@
 #include <linux/string.h>
 #include <linux/time.h>
 #include <linux/vmalloc.h>
+#include <linux/workqueue.h>
 
 #include <asm/byteorder.h>
 #include <asm/page.h>
@@ -125,6 +126,7 @@ struct context {
 	struct fw_ohci *ohci;
 	u32 regs;
 	int total_allocation;
+	u32 current_bus;
 	bool running;
 	bool flushing;
 
@@ -226,7 +228,7 @@ struct fw_ohci {
 
 	__le32    *self_id_cpu;
 	dma_addr_t self_id_bus;
-	struct tasklet_struct bus_reset_tasklet;
+	struct work_struct bus_reset_work;
 
 	u32 self_id_buffer[512];
 };
@@ -263,6 +265,8 @@ static char ohci_driver_name[] = KBUILD_MODNAME;
 #define PCI_DEVICE_ID_AGERE_FW643	0x5901
 #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_VENDOR_ID_PINNACLE_SYSTEMS	0x11bd
 
 #define QUIRK_CYCLE_TIMER		1
@@ -270,6 +274,7 @@ static char ohci_driver_name[] = KBUILD_MODNAME;
 #define QUIRK_BE_HEADERS		4
 #define QUIRK_NO_1394A			8
 #define QUIRK_NO_MSI			16
+#define QUIRK_TI_SLLZ059		32
 
 /* In case of multiple matches in ohci_quirks[], only the first one is used. */
 static const struct {
@@ -299,6 +304,12 @@ static const struct {
 	{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},
 
@@ -315,6 +326,7 @@ MODULE_PARM_DESC(quirks, "Chip quirks (default = 0"
 	", AR/selfID endianess = "	__stringify(QUIRK_BE_HEADERS)
 	", no 1394a enhancements = "	__stringify(QUIRK_NO_1394A)
 	", disable MSI = "		__stringify(QUIRK_NO_MSI)
+	", TI SLLZ059 erratum = "	__stringify(QUIRK_TI_SLLZ059)
 	")");
 
 #define OHCI_PARAM_DEBUG_AT_AR		1
@@ -859,7 +871,7 @@ static __le32 *handle_ar_packet(struct ar_context *ctx, __le32 *buffer)
 	 *
 	 * 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_tasklet).
+	 * at a slightly incorrect time (in bus_reset_work).
 	 */
 	if (evt == OHCI1394_evt_bus_reset) {
 		if (!(ohci->quirks & QUIRK_RESET_PACKET))
@@ -1046,6 +1058,7 @@ static void context_tasklet(unsigned long data)
 		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. */
@@ -1713,9 +1726,94 @@ static u32 update_bus_time(struct fw_ohci *ohci)
 	return ohci->bus_time | cycle_time_seconds;
 }
 
-static void bus_reset_tasklet(unsigned long data)
+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;
+}
+
+/*
+ * 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.
+ * FIXME:  How to determine the selfID.i flag?
+ */
+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)) {
+		fw_notify("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)));
+	}
+
+	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 = (struct fw_ohci *)data;
+	struct fw_ohci *ohci =
+		container_of(work, struct fw_ohci, bus_reset_work);
 	int self_id_count, i, j, reg;
 	int generation, new_generation;
 	unsigned long flags;
@@ -1753,21 +1851,50 @@ static void bus_reset_tasklet(unsigned long data)
 	 * bit extra to get the actual number of self IDs.
 	 */
 	self_id_count = (reg >> 3) & 0xff;
-	if (self_id_count == 0 || self_id_count > 252) {
+
+	if (self_id_count > 252) {
 		fw_notify("inconsistent self IDs\n");
 		return;
 	}
+
 	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_notify("inconsistent self IDs\n");
-			return;
+			/*
+			 * 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 (cond_le32_to_cpu(ohci->self_id_cpu[i])
+							== 0xffff008f) {
+				fw_notify("ignoring spurious self IDs\n");
+				self_id_count = j;
+				break;
+			} else {
+				fw_notify("inconsistent self IDs\n");
+				return;
+			}
 		}
 		ohci->self_id_buffer[j] =
 				cond_le32_to_cpu(ohci->self_id_cpu[i]);
 	}
+
+	if (ohci->quirks & QUIRK_TI_SLLZ059) {
+		self_id_count = find_and_insert_self_id(ohci, self_id_count);
+		if (self_id_count < 0) {
+			fw_notify("could not construct local self ID\n");
+			return;
+		}
+	}
+
+	if (self_id_count == 0) {
+		fw_notify("inconsistent self IDs\n");
+		return;
+	}
 	rmb();
 
 	/*
@@ -1887,7 +2014,7 @@ static irqreturn_t irq_handler(int irq, void *data)
 	log_irqs(event);
 
 	if (event & OHCI1394_selfIDComplete)
-		tasklet_schedule(&ohci->bus_reset_tasklet);
+		queue_work(fw_workqueue, &ohci->bus_reset_work);
 
 	if (event & OHCI1394_RQPkt)
 		tasklet_schedule(&ohci->ar_request_ctx.tasklet);
@@ -1934,7 +2061,8 @@ static irqreturn_t irq_handler(int irq, void *data)
 		reg_read(ohci, OHCI1394_PostedWriteAddressLo);
 		reg_write(ohci, OHCI1394_IntEventClear,
 			  OHCI1394_postedWriteErr);
-		fw_error("PCI posted write error\n");
+		if (printk_ratelimit())
+			fw_error("PCI posted write error\n");
 	}
 
 	if (unlikely(event & OHCI1394_cycleTooLong)) {
@@ -2048,6 +2176,28 @@ static int configure_1394a_enhancements(struct fw_ohci *ohci)
 	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)
 {
@@ -2085,6 +2235,16 @@ static int ohci_enable(struct fw_card *card,
 		return -EIO;
 	}
 
+	if (ohci->quirks & QUIRK_TI_SLLZ059) {
+		ret = probe_tsb41ba3d(ohci);
+		if (ret < 0)
+			return ret;
+		if (ret)
+			fw_notify("local TSB41BA3D phy\n");
+		else
+			ohci->quirks &= ~QUIRK_TI_SLLZ059;
+	}
+
 	reg_write(ohci, OHCI1394_HCControlClear,
 		  OHCI1394_HCControl_noByteSwapData);
 
@@ -2260,7 +2420,7 @@ static int ohci_set_config_rom(struct fw_card *card,
 	 * 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_tasklet).
+	 * ohci->next_config_rom to NULL (see bus_reset_work).
 	 */
 
 	next_config_rom =
@@ -2539,6 +2699,7 @@ static int handle_ir_packet_per_buffer(struct context *context,
 	struct iso_context *ctx =
 		container_of(context, struct iso_context, context);
 	struct descriptor *pd;
+	u32 buffer_dma;
 	__le32 *ir_header;
 	void *p;
 
@@ -2549,6 +2710,16 @@ static int handle_ir_packet_per_buffer(struct context *context,
 		/* 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);
+	}
+
 	p = last + 1;
 	copy_iso_headers(ctx, p);
 
@@ -2571,11 +2742,19 @@ static int handle_ir_buffer_fill(struct context *context,
 {
 	struct iso_context *ctx =
 		container_of(context, struct iso_context, context);
+	u32 buffer_dma;
 
 	if (!last->transfer_status)
 		/* Descriptor(s) not done yet, stop iteration */
 		return 0;
 
+	buffer_dma = le32_to_cpu(last->data_address);
+	dma_sync_single_range_for_cpu(context->ohci->card.device,
+				      buffer_dma & PAGE_MASK,
+				      buffer_dma & ~PAGE_MASK,
+				      le16_to_cpu(last->req_count),
+				      DMA_FROM_DEVICE);
+
 	if (le16_to_cpu(last->control) & DESCRIPTOR_IRQ_ALWAYS)
 		ctx->base.callback.mc(&ctx->base,
 				      le32_to_cpu(last->data_address) +
@@ -2586,6 +2765,43 @@ static int handle_ir_buffer_fill(struct context *context,
 	return 1;
 }
 
+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)
@@ -2602,6 +2818,8 @@ static int handle_it_packet(struct context *context,
 		/* Descriptor(s) not done yet, stop iteration */
 		return 0;
 
+	sync_it_packet_for_cpu(context, d);
+
 	i = ctx->header_length;
 	if (i + 4 < PAGE_SIZE) {
 		/* Present this value as big-endian to match the receive code */
@@ -2971,6 +3189,10 @@ static int queue_iso_transmit(struct iso_context *ctx,
 		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;
 	}
 
@@ -2995,6 +3217,7 @@ static int queue_iso_packet_per_buffer(struct iso_context *ctx,
 				       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;
@@ -3049,6 +3272,10 @@ static int queue_iso_packet_per_buffer(struct iso_context *ctx,
 			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)
@@ -3108,6 +3335,10 @@ static int queue_iso_buffer_fill(struct iso_context *ctx,
 		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++;
@@ -3239,8 +3470,7 @@ static int __devinit pci_probe(struct pci_dev *dev,
 	spin_lock_init(&ohci->lock);
 	mutex_init(&ohci->phy_reg_mutex);
 
-	tasklet_init(&ohci->bus_reset_tasklet,
-		     bus_reset_tasklet, (unsigned long)ohci);
+	INIT_WORK(&ohci->bus_reset_work, bus_reset_work);
 
 	err = pci_request_region(dev, 0, ohci_driver_name);
 	if (err) {
@@ -3382,6 +3612,7 @@ static void pci_remove(struct pci_dev *dev)
 	ohci = pci_get_drvdata(dev);
 	reg_write(ohci, OHCI1394_IntMaskClear, ~0);
 	flush_writes(ohci);
+	cancel_work_sync(&ohci->bus_reset_work);
 	fw_core_remove_card(&ohci->card);
 
 	/*

drivers/firewire/sbp2.c

@@ -154,12 +154,16 @@ struct sbp2_logical_unit {
 	bool blocked;
 };
 
+static void sbp2_queue_work(struct sbp2_logical_unit *lu, unsigned long delay)
+{
+	queue_delayed_work(fw_workqueue, &lu->work, delay);
+}
+
 /*
  * We create one struct sbp2_target per IEEE 1212 Unit Directory
  * and one struct Scsi_Host per sbp2_target.
  */
 struct sbp2_target {
-	struct kref kref;
 	struct fw_unit *unit;
 	const char *bus_id;
 	struct list_head lu_list;
@@ -772,71 +776,6 @@ static int sbp2_lun2int(u16 lun)
 	return scsilun_to_int(&eight_bytes_lun);
 }
 
-static void sbp2_release_target(struct kref *kref)
-{
-	struct sbp2_target *tgt = container_of(kref, struct sbp2_target, kref);
-	struct sbp2_logical_unit *lu, *next;
-	struct Scsi_Host *shost =
-		container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
-	struct scsi_device *sdev;
-	struct fw_device *device = target_device(tgt);
-
-	/* prevent deadlocks */
-	sbp2_unblock(tgt);
-
-	list_for_each_entry_safe(lu, next, &tgt->lu_list, link) {
-		sdev = scsi_device_lookup(shost, 0, 0, sbp2_lun2int(lu->lun));
-		if (sdev) {
-			scsi_remove_device(sdev);
-			scsi_device_put(sdev);
-		}
-		if (lu->login_id != INVALID_LOGIN_ID) {
-			int generation, node_id;
-			/*
-			 * tgt->node_id may be obsolete here if we failed
-			 * during initial login or after a bus reset where
-			 * the topology changed.
-			 */
-			generation = device->generation;
-			smp_rmb(); /* node_id vs. generation */
-			node_id    = device->node_id;
-			sbp2_send_management_orb(lu, node_id, generation,
-						 SBP2_LOGOUT_REQUEST,
-						 lu->login_id, NULL);
-		}
-		fw_core_remove_address_handler(&lu->address_handler);
-		list_del(&lu->link);
-		kfree(lu);
-	}
-	scsi_remove_host(shost);
-	fw_notify("released %s, target %d:0:0\n", tgt->bus_id, shost->host_no);
-
-	fw_unit_put(tgt->unit);
-	scsi_host_put(shost);
-	fw_device_put(device);
-}
-
-static void sbp2_target_get(struct sbp2_target *tgt)
-{
-	kref_get(&tgt->kref);
-}
-
-static void sbp2_target_put(struct sbp2_target *tgt)
-{
-	kref_put(&tgt->kref, sbp2_release_target);
-}
-
-/*
- * Always get the target's kref when scheduling work on one its units.
- * Each workqueue job is responsible to call sbp2_target_put() upon return.
- */
-static void sbp2_queue_work(struct sbp2_logical_unit *lu, unsigned long delay)
-{
-	sbp2_target_get(lu->tgt);
-	if (!queue_delayed_work(fw_workqueue, &lu->work, delay))
-		sbp2_target_put(lu->tgt);
-}
-
 /*
  * Write retransmit retry values into the BUSY_TIMEOUT register.
  * - The single-phase retry protocol is supported by all SBP-2 devices, but the
@@ -877,7 +816,7 @@ static void sbp2_login(struct work_struct *work)
 	int generation, node_id, local_node_id;
 
 	if (fw_device_is_shutdown(device))
-		goto out;
+		return;
 
 	generation    = device->generation;
 	smp_rmb();    /* node IDs must not be older than generation */
@@ -899,7 +838,7 @@ static void sbp2_login(struct work_struct *work)
 			/* Let any waiting I/O fail from now on. */
 			sbp2_unblock(lu->tgt);
 		}
-		goto out;
+		return;
 	}
 
 	tgt->node_id	  = node_id;
@@ -925,7 +864,8 @@ static void sbp2_login(struct work_struct *work)
 	if (lu->has_sdev) {
 		sbp2_cancel_orbs(lu);
 		sbp2_conditionally_unblock(lu);
-		goto out;
+
+		return;
 	}
 
 	if (lu->tgt->workarounds & SBP2_WORKAROUND_DELAY_INQUIRY)
@@ -957,7 +897,8 @@ static void sbp2_login(struct work_struct *work)
 	lu->has_sdev = true;
 	scsi_device_put(sdev);
 	sbp2_allow_block(lu);
-	goto out;
+
+	return;
 
  out_logout_login:
 	smp_rmb(); /* generation may have changed */
@@ -971,8 +912,57 @@ static void sbp2_login(struct work_struct *work)
 	 * lu->work already.  Reset the work from reconnect to login.
 	 */
 	PREPARE_DELAYED_WORK(&lu->work, sbp2_login);
- out:
-	sbp2_target_put(tgt);
+}
+
+static void sbp2_reconnect(struct work_struct *work)
+{
+	struct sbp2_logical_unit *lu =
+		container_of(work, struct sbp2_logical_unit, work.work);
+	struct sbp2_target *tgt = lu->tgt;
+	struct fw_device *device = target_device(tgt);
+	int generation, node_id, local_node_id;
+
+	if (fw_device_is_shutdown(device))
+		return;
+
+	generation    = device->generation;
+	smp_rmb();    /* node IDs must not be older than generation */
+	node_id       = device->node_id;
+	local_node_id = device->card->node_id;
+
+	if (sbp2_send_management_orb(lu, node_id, generation,
+				     SBP2_RECONNECT_REQUEST,
+				     lu->login_id, NULL) < 0) {
+		/*
+		 * If reconnect was impossible even though we are in the
+		 * current generation, fall back and try to log in again.
+		 *
+		 * We could check for "Function rejected" status, but
+		 * looking at the bus generation as simpler and more general.
+		 */
+		smp_rmb(); /* get current card generation */
+		if (generation == device->card->generation ||
+		    lu->retries++ >= 5) {
+			fw_error("%s: failed to reconnect\n", tgt->bus_id);
+			lu->retries = 0;
+			PREPARE_DELAYED_WORK(&lu->work, sbp2_login);
+		}
+		sbp2_queue_work(lu, DIV_ROUND_UP(HZ, 5));
+
+		return;
+	}
+
+	tgt->node_id      = node_id;
+	tgt->address_high = local_node_id << 16;
+	smp_wmb();	  /* node IDs must not be older than generation */
+	lu->generation	  = generation;
+
+	fw_notify("%s: reconnected to LUN %04x (%d retries)\n",
+		  tgt->bus_id, lu->lun, lu->retries);
+
+	sbp2_agent_reset(lu);
+	sbp2_cancel_orbs(lu);
+	sbp2_conditionally_unblock(lu);
 }
 
 static int sbp2_add_logical_unit(struct sbp2_target *tgt, int lun_entry)
@@ -1120,6 +1110,7 @@ static void sbp2_init_workarounds(struct sbp2_target *tgt, u32 model,
 }
 
 static struct scsi_host_template scsi_driver_template;
+static int sbp2_remove(struct device *dev);
 
 static int sbp2_probe(struct device *dev)
 {
@@ -1141,7 +1132,6 @@ static int sbp2_probe(struct device *dev)
 	tgt = (struct sbp2_target *)shost->hostdata;
 	dev_set_drvdata(&unit->device, tgt);
 	tgt->unit = unit;
-	kref_init(&tgt->kref);
 	INIT_LIST_HEAD(&tgt->lu_list);
 	tgt->bus_id = dev_name(&unit->device);
 	tgt->guid = (u64)device->config_rom[3] << 32 | device->config_rom[4];
@@ -1154,9 +1144,6 @@ static int sbp2_probe(struct device *dev)
 	if (scsi_add_host(shost, &unit->device) < 0)
 		goto fail_shost_put;
 
-	fw_device_get(device);
-	fw_unit_get(unit);
-
 	/* implicit directory ID */
 	tgt->directory_id = ((unit->directory - device->config_rom) * 4
 			     + CSR_CONFIG_ROM) & 0xffffff;
@@ -1166,7 +1153,7 @@ static int sbp2_probe(struct device *dev)
 
 	if (sbp2_scan_unit_dir(tgt, unit->directory, &model,
 			       &firmware_revision) < 0)
-		goto fail_tgt_put;
+		goto fail_remove;
 
 	sbp2_clamp_management_orb_timeout(tgt);
 	sbp2_init_workarounds(tgt, model, firmware_revision);
@@ -1177,16 +1164,17 @@ static int sbp2_probe(struct device *dev)
 	 * specifies the max payload size as 2 ^ (max_payload + 2), so
 	 * if we set this to max_speed + 7, we get the right value.
 	 */
-	tgt->max_payload = min(device->max_speed + 7, 10U);
-	tgt->max_payload = min(tgt->max_payload, device->card->max_receive - 1);
+	tgt->max_payload = min3(device->max_speed + 7, 10U,
+				device->card->max_receive - 1);
 
 	/* Do the login in a workqueue so we can easily reschedule retries. */
 	list_for_each_entry(lu, &tgt->lu_list, link)
 		sbp2_queue_work(lu, DIV_ROUND_UP(HZ, 5));
+
 	return 0;
 
- fail_tgt_put:
-	sbp2_target_put(tgt);
+ fail_remove:
+	sbp2_remove(dev);
 	return -ENOMEM;
 
  fail_shost_put:
@@ -1194,71 +1182,6 @@ static int sbp2_probe(struct device *dev)
 	return -ENOMEM;
 }
 
-static int sbp2_remove(struct device *dev)
-{
-	struct fw_unit *unit = fw_unit(dev);
-	struct sbp2_target *tgt = dev_get_drvdata(&unit->device);
-	struct sbp2_logical_unit *lu;
-
-	list_for_each_entry(lu, &tgt->lu_list, link)
-		cancel_delayed_work_sync(&lu->work);
-
-	sbp2_target_put(tgt);
-	return 0;
-}
-
-static void sbp2_reconnect(struct work_struct *work)
-{
-	struct sbp2_logical_unit *lu =
-		container_of(work, struct sbp2_logical_unit, work.work);
-	struct sbp2_target *tgt = lu->tgt;
-	struct fw_device *device = target_device(tgt);
-	int generation, node_id, local_node_id;
-
-	if (fw_device_is_shutdown(device))
-		goto out;
-
-	generation    = device->generation;
-	smp_rmb();    /* node IDs must not be older than generation */
-	node_id       = device->node_id;
-	local_node_id = device->card->node_id;
-
-	if (sbp2_send_management_orb(lu, node_id, generation,
-				     SBP2_RECONNECT_REQUEST,
-				     lu->login_id, NULL) < 0) {
-		/*
-		 * If reconnect was impossible even though we are in the
-		 * current generation, fall back and try to log in again.
-		 *
-		 * We could check for "Function rejected" status, but
-		 * looking at the bus generation as simpler and more general.
-		 */
-		smp_rmb(); /* get current card generation */
-		if (generation == device->card->generation ||
-		    lu->retries++ >= 5) {
-			fw_error("%s: failed to reconnect\n", tgt->bus_id);
-			lu->retries = 0;
-			PREPARE_DELAYED_WORK(&lu->work, sbp2_login);
-		}
-		sbp2_queue_work(lu, DIV_ROUND_UP(HZ, 5));
-		goto out;
-	}
-
-	tgt->node_id      = node_id;
-	tgt->address_high = local_node_id << 16;
-	smp_wmb();	  /* node IDs must not be older than generation */
-	lu->generation	  = generation;
-
-	fw_notify("%s: reconnected to LUN %04x (%d retries)\n",
-		  tgt->bus_id, lu->lun, lu->retries);
-
-	sbp2_agent_reset(lu);
-	sbp2_cancel_orbs(lu);
-	sbp2_conditionally_unblock(lu);
- out:
-	sbp2_target_put(tgt);
-}
-
 static void sbp2_update(struct fw_unit *unit)
 {
 	struct sbp2_target *tgt = dev_get_drvdata(&unit->device);
@@ -1277,6 +1200,51 @@ static void sbp2_update(struct fw_unit *unit)
 	}
 }
 
+static int sbp2_remove(struct device *dev)
+{
+	struct fw_unit *unit = fw_unit(dev);
+	struct fw_device *device = fw_parent_device(unit);
+	struct sbp2_target *tgt = dev_get_drvdata(&unit->device);
+	struct sbp2_logical_unit *lu, *next;
+	struct Scsi_Host *shost =
+		container_of((void *)tgt, struct Scsi_Host, hostdata[0]);
+	struct scsi_device *sdev;
+
+	/* prevent deadlocks */
+	sbp2_unblock(tgt);
+
+	list_for_each_entry_safe(lu, next, &tgt->lu_list, link) {
+		cancel_delayed_work_sync(&lu->work);
+		sdev = scsi_device_lookup(shost, 0, 0, sbp2_lun2int(lu->lun));
+		if (sdev) {
+			scsi_remove_device(sdev);
+			scsi_device_put(sdev);
+		}
+		if (lu->login_id != INVALID_LOGIN_ID) {
+			int generation, node_id;
+			/*
+			 * tgt->node_id may be obsolete here if we failed
+			 * during initial login or after a bus reset where
+			 * the topology changed.
+			 */
+			generation = device->generation;
+			smp_rmb(); /* node_id vs. generation */
+			node_id    = device->node_id;
+			sbp2_send_management_orb(lu, node_id, generation,
+						 SBP2_LOGOUT_REQUEST,
+						 lu->login_id, NULL);
+		}
+		fw_core_remove_address_handler(&lu->address_handler);
+		list_del(&lu->link);
+		kfree(lu);
+	}
+	scsi_remove_host(shost);
+	fw_notify("released %s, target %d:0:0\n", tgt->bus_id, shost->host_no);
+
+	scsi_host_put(shost);
+	return 0;
+}
+
 #define SBP2_UNIT_SPEC_ID_ENTRY	0x0000609e
 #define SBP2_SW_VERSION_ENTRY	0x00010483