Merge git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6 into for-linus

16 files changed, 420 insertions, 93 deletions

diff --git a/arch/powerpc/platforms/cell/interrupt.c b/arch/powerpc/platforms/cell/interrupt.c
index 04f74f9f9ab..5bf7df14602 100644
--- a/arch/powerpc/platforms/cell/interrupt.c
+++ b/arch/powerpc/platforms/cell/interrupt.c
@@ -35,6 +35,7 @@
 #include <linux/percpu.h>
 #include <linux/types.h>
 #include <linux/ioport.h>
+#include <linux/kernel_stat.h>
 
 #include <asm/io.h>
 #include <asm/pgtable.h>
@@ -231,6 +232,54 @@ static int iic_host_match(struct irq_host *h, struct device_node *node)
 				    "IBM,CBEA-Internal-Interrupt-Controller");
 }
 
+extern int noirqdebug;
+
+static void handle_iic_irq(unsigned int irq, struct irq_desc *desc)
+{
+	const unsigned int cpu = smp_processor_id();
+
+	spin_lock(&desc->lock);
+
+	desc->status &= ~(IRQ_REPLAY | IRQ_WAITING);
+
+	/*
+	 * If we're currently running this IRQ, or its disabled,
+	 * we shouldn't process the IRQ. Mark it pending, handle
+	 * the necessary masking and go out
+	 */
+	if (unlikely((desc->status & (IRQ_INPROGRESS | IRQ_DISABLED)) ||
+		    !desc->action)) {
+		desc->status |= IRQ_PENDING;
+		goto out_eoi;
+	}
+
+	kstat_cpu(cpu).irqs[irq]++;
+
+	/* Mark the IRQ currently in progress.*/
+	desc->status |= IRQ_INPROGRESS;
+
+	do {
+		struct irqaction *action = desc->action;
+		irqreturn_t action_ret;
+
+		if (unlikely(!action))
+			goto out_eoi;
+
+		desc->status &= ~IRQ_PENDING;
+		spin_unlock(&desc->lock);
+		action_ret = handle_IRQ_event(irq, action);
+		if (!noirqdebug)
+			note_interrupt(irq, desc, action_ret);
+		spin_lock(&desc->lock);
+
+	} while ((desc->status & (IRQ_PENDING | IRQ_DISABLED)) == IRQ_PENDING);
+
+	desc->status &= ~IRQ_INPROGRESS;
+out_eoi:
+	desc->chip->eoi(irq);
+	spin_unlock(&desc->lock);
+}
+
 static int iic_host_map(struct irq_host *h, unsigned int virq,
 			irq_hw_number_t hw)
 {
@@ -240,10 +289,10 @@ static int iic_host_map(struct irq_host *h, unsigned int virq,
 		break;
 	case IIC_IRQ_TYPE_IOEXC:
 		set_irq_chip_and_handler(virq, &iic_ioexc_chip,
-					 handle_fasteoi_irq);
+					 handle_iic_irq);
 		break;
 	default:
-		set_irq_chip_and_handler(virq, &iic_chip, handle_fasteoi_irq);
+		set_irq_chip_and_handler(virq, &iic_chip, handle_iic_irq);
 	}
 	return 0;
 }
diff --git a/arch/powerpc/platforms/cell/io-workarounds.c b/arch/powerpc/platforms/cell/io-workarounds.c
index 3b84e8be314..b5f84e8f089 100644
--- a/arch/powerpc/platforms/cell/io-workarounds.c
+++ b/arch/powerpc/platforms/cell/io-workarounds.c
@@ -118,7 +118,7 @@ static void iowa_##name at					\
 #undef DEF_PCI_AC_RET
 #undef DEF_PCI_AC_NORET
 
-static struct ppc_pci_io __initdata iowa_pci_io = {
+static const struct ppc_pci_io __devinitconst iowa_pci_io = {
 
 #define DEF_PCI_AC_RET(name, ret, at, al, space, aa)	.name = iowa_##name,
 #define DEF_PCI_AC_NORET(name, at, al, space, aa)	.name = iowa_##name,
@@ -146,7 +146,7 @@ static void __iomem *iowa_ioremap(unsigned long addr, unsigned long size,
 }
 
 /* Regist new bus to support workaround */
-void __init iowa_register_bus(struct pci_controller *phb,
+void __devinit iowa_register_bus(struct pci_controller *phb,
 			struct ppc_pci_io *ops,
 			int (*initfunc)(struct iowa_bus *, void *), void *data)
 {
@@ -173,7 +173,7 @@ void __init iowa_register_bus(struct pci_controller *phb,
 }
 
 /* enable IO workaround */
-void __init io_workaround_init(void)
+void __devinit io_workaround_init(void)
 {
 	static int io_workaround_inited;
 
diff --git a/arch/powerpc/platforms/cell/io-workarounds.h b/arch/powerpc/platforms/cell/io-workarounds.h
index 79d8ed3d510..6efc7782ebf 100644
--- a/arch/powerpc/platforms/cell/io-workarounds.h
+++ b/arch/powerpc/platforms/cell/io-workarounds.h
@@ -31,9 +31,9 @@ struct iowa_bus {
 	void   *private;
 };
 
-void __init io_workaround_init(void);
-void __init iowa_register_bus(struct pci_controller *, struct ppc_pci_io *,
-			      int (*)(struct iowa_bus *, void *), void *);
+void __devinit io_workaround_init(void);
+void __devinit iowa_register_bus(struct pci_controller *, struct ppc_pci_io *,
+				 int (*)(struct iowa_bus *, void *), void *);
 struct iowa_bus *iowa_mem_find_bus(const PCI_IO_ADDR);
 struct iowa_bus *iowa_pio_find_bus(unsigned long);
 
diff --git a/arch/powerpc/platforms/cell/spu_base.c b/arch/powerpc/platforms/cell/spu_base.c
index 6bab44b7716..70c660121ec 100644
--- a/arch/powerpc/platforms/cell/spu_base.c
+++ b/arch/powerpc/platforms/cell/spu_base.c
@@ -141,6 +141,10 @@ static void spu_restart_dma(struct spu *spu)
 
 	if (!test_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags))
 		out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESTART_DMA_COMMAND);
+	else {
+		set_bit(SPU_CONTEXT_FAULT_PENDING, &spu->flags);
+		mb();
+	}
 }
 
 static inline void spu_load_slb(struct spu *spu, int slbe, struct spu_slb *slb)
@@ -226,11 +230,13 @@ static int __spu_trap_data_map(struct spu *spu, unsigned long ea, u64 dsisr)
 		return 0;
 	}
 
-	spu->class_0_pending = 0;
-	spu->dar = ea;
-	spu->dsisr = dsisr;
+	spu->class_1_dar = ea;
+	spu->class_1_dsisr = dsisr;
+
+	spu->stop_callback(spu, 1);
 
-	spu->stop_callback(spu);
+	spu->class_1_dar = 0;
+	spu->class_1_dsisr = 0;
 
 	return 0;
 }
@@ -318,11 +324,15 @@ spu_irq_class_0(int irq, void *data)
 	stat = spu_int_stat_get(spu, 0) & mask;
 
 	spu->class_0_pending |= stat;
-	spu->dsisr = spu_mfc_dsisr_get(spu);
-	spu->dar = spu_mfc_dar_get(spu);
+	spu->class_0_dsisr = spu_mfc_dsisr_get(spu);
+	spu->class_0_dar = spu_mfc_dar_get(spu);
 	spin_unlock(&spu->register_lock);
 
-	spu->stop_callback(spu);
+	spu->stop_callback(spu, 0);
+
+	spu->class_0_pending = 0;
+	spu->class_0_dsisr = 0;
+	spu->class_0_dar = 0;
 
 	spu_int_stat_clear(spu, 0, stat);
 
@@ -363,6 +373,9 @@ spu_irq_class_1(int irq, void *data)
 	if (stat & CLASS1_LS_COMPARE_SUSPEND_ON_PUT_INTR)
 		;
 
+	spu->class_1_dsisr = 0;
+	spu->class_1_dar = 0;
+
 	return stat ? IRQ_HANDLED : IRQ_NONE;
 }
 
@@ -396,10 +409,10 @@ spu_irq_class_2(int irq, void *data)
 		spu->ibox_callback(spu);
 
 	if (stat & CLASS2_SPU_STOP_INTR)
-		spu->stop_callback(spu);
+		spu->stop_callback(spu, 2);
 
 	if (stat & CLASS2_SPU_HALT_INTR)
-		spu->stop_callback(spu);
+		spu->stop_callback(spu, 2);
 
 	if (stat & CLASS2_SPU_DMA_TAG_GROUP_COMPLETE_INTR)
 		spu->mfc_callback(spu);
diff --git a/arch/powerpc/platforms/cell/spu_priv1_mmio.c b/arch/powerpc/platforms/cell/spu_priv1_mmio.c
index 67fa7247b80..906a0a2a9fe 100644
--- a/arch/powerpc/platforms/cell/spu_priv1_mmio.c
+++ b/arch/powerpc/platforms/cell/spu_priv1_mmio.c
@@ -28,6 +28,7 @@
 #include <linux/io.h>
 #include <linux/mutex.h>
 #include <linux/device.h>
+#include <linux/sched.h>
 
 #include <asm/spu.h>
 #include <asm/spu_priv1.h>
@@ -75,8 +76,19 @@ static u64 int_stat_get(struct spu *spu, int class)
 
 static void cpu_affinity_set(struct spu *spu, int cpu)
 {
-	u64 target = iic_get_target_id(cpu);
-	u64 route = target << 48 | target << 32 | target << 16;
+	u64 target;
+	u64 route;
+
+	if (nr_cpus_node(spu->node)) {
+		cpumask_t spumask = node_to_cpumask(spu->node);
+		cpumask_t cpumask = node_to_cpumask(cpu_to_node(cpu));
+
+		if (!cpus_intersects(spumask, cpumask))
+			return;
+	}
+
+	target = iic_get_target_id(cpu);
+	route = target << 48 | target << 32 | target << 16;
 	out_be64(&spu->priv1->int_route_RW, route);
 }
 
diff --git a/arch/powerpc/platforms/cell/spufs/.gitignore b/arch/powerpc/platforms/cell/spufs/.gitignore
new file mode 100644
index 00000000000..a09ee8d84d6
--- /dev/null
+++ b/arch/powerpc/platforms/cell/spufs/.gitignore
@@ -0,0 +1,2 @@
+spu_save_dump.h
+spu_restore_dump.h
diff --git a/arch/powerpc/platforms/cell/spufs/context.c b/arch/powerpc/platforms/cell/spufs/context.c
index 0ad83aeb70b..177735f7931 100644
--- a/arch/powerpc/platforms/cell/spufs/context.c
+++ b/arch/powerpc/platforms/cell/spufs/context.c
@@ -78,6 +78,7 @@ void destroy_spu_context(struct kref *kref)
 {
 	struct spu_context *ctx;
 	ctx = container_of(kref, struct spu_context, kref);
+	spu_context_nospu_trace(destroy_spu_context__enter, ctx);
 	mutex_lock(&ctx->state_mutex);
 	spu_deactivate(ctx);
 	mutex_unlock(&ctx->state_mutex);
@@ -88,6 +89,7 @@ void destroy_spu_context(struct kref *kref)
 		kref_put(ctx->prof_priv_kref, ctx->prof_priv_release);
 	BUG_ON(!list_empty(&ctx->rq));
 	atomic_dec(&nr_spu_contexts);
+	kfree(ctx->switch_log);
 	kfree(ctx);
 }
 
@@ -150,6 +152,8 @@ int spu_acquire_saved(struct spu_context *ctx)
 {
 	int ret;
 
+	spu_context_nospu_trace(spu_acquire_saved__enter, ctx);
+
 	ret = spu_acquire(ctx);
 	if (ret)
 		return ret;
diff --git a/arch/powerpc/platforms/cell/spufs/coredump.c b/arch/powerpc/platforms/cell/spufs/coredump.c
index b962c3ab470..af116aadba1 100644
--- a/arch/powerpc/platforms/cell/spufs/coredump.c
+++ b/arch/powerpc/platforms/cell/spufs/coredump.c
@@ -22,6 +22,7 @@
 
 #include <linux/elf.h>
 #include <linux/file.h>
+#include <linux/fdtable.h>
 #include <linux/fs.h>
 #include <linux/list.h>
 #include <linux/module.h>
diff --git a/arch/powerpc/platforms/cell/spufs/fault.c b/arch/powerpc/platforms/cell/spufs/fault.c
index e46d300e21a..f093a581ac7 100644
--- a/arch/powerpc/platforms/cell/spufs/fault.c
+++ b/arch/powerpc/platforms/cell/spufs/fault.c
@@ -83,13 +83,18 @@ int spufs_handle_class0(struct spu_context *ctx)
 		return 0;
 
 	if (stat & CLASS0_DMA_ALIGNMENT_INTR)
-		spufs_handle_event(ctx, ctx->csa.dar, SPE_EVENT_DMA_ALIGNMENT);
+		spufs_handle_event(ctx, ctx->csa.class_0_dar,
+			SPE_EVENT_DMA_ALIGNMENT);
 
 	if (stat & CLASS0_INVALID_DMA_COMMAND_INTR)
-		spufs_handle_event(ctx, ctx->csa.dar, SPE_EVENT_INVALID_DMA);
+		spufs_handle_event(ctx, ctx->csa.class_0_dar,
+			SPE_EVENT_INVALID_DMA);
 
 	if (stat & CLASS0_SPU_ERROR_INTR)
-		spufs_handle_event(ctx, ctx->csa.dar, SPE_EVENT_SPE_ERROR);
+		spufs_handle_event(ctx, ctx->csa.class_0_dar,
+			SPE_EVENT_SPE_ERROR);
+
+	ctx->csa.class_0_pending = 0;
 
 	return -EIO;
 }
@@ -119,8 +124,8 @@ int spufs_handle_class1(struct spu_context *ctx)
 	 * in time, we can still expect to get the same fault
 	 * the immediately after the context restore.
 	 */
-	ea = ctx->csa.dar;
-	dsisr = ctx->csa.dsisr;
+	ea = ctx->csa.class_1_dar;
+	dsisr = ctx->csa.class_1_dsisr;
 
 	if (!(dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED)))
 		return 0;
@@ -158,7 +163,7 @@ int spufs_handle_class1(struct spu_context *ctx)
 	 * time slicing will not preempt the context while the page fault
 	 * handler is running. Context switch code removes mappings.
 	 */
-	ctx->csa.dar = ctx->csa.dsisr = 0;
+	ctx->csa.class_1_dar = ctx->csa.class_1_dsisr = 0;
 
 	/*
 	 * If we handled the fault successfully and are in runnable
diff --git a/arch/powerpc/platforms/cell/spufs/file.c b/arch/powerpc/platforms/cell/spufs/file.c
index 08f44d1971a..c81341ff75b 100644
--- a/arch/powerpc/platforms/cell/spufs/file.c
+++ b/arch/powerpc/platforms/cell/spufs/file.c
@@ -32,6 +32,7 @@
 #include <linux/marker.h>
 
 #include <asm/io.h>
+#include <asm/time.h>
 #include <asm/spu.h>
 #include <asm/spu_info.h>
 #include <asm/uaccess.h>
@@ -2386,6 +2387,171 @@ static const struct file_operations spufs_stat_fops = {
 	.release	= single_release,
 };
 
+static inline int spufs_switch_log_used(struct spu_context *ctx)
+{
+	return (ctx->switch_log->head - ctx->switch_log->tail) %
+		SWITCH_LOG_BUFSIZE;
+}
+
+static inline int spufs_switch_log_avail(struct spu_context *ctx)
+{
+	return SWITCH_LOG_BUFSIZE - spufs_switch_log_used(ctx);
+}
+
+static int spufs_switch_log_open(struct inode *inode, struct file *file)
+{
+	struct spu_context *ctx = SPUFS_I(inode)->i_ctx;
+
+	/*
+	 * We (ab-)use the mapping_lock here because it serves the similar
+	 * purpose for synchronizing open/close elsewhere.  Maybe it should
+	 * be renamed eventually.
+	 */
+	mutex_lock(&ctx->mapping_lock);
+	if (ctx->switch_log) {
+		spin_lock(&ctx->switch_log->lock);
+		ctx->switch_log->head = 0;
+		ctx->switch_log->tail = 0;
+		spin_unlock(&ctx->switch_log->lock);
+	} else {
+		/*
+		 * We allocate the switch log data structures on first open.
+		 * They will never be free because we assume a context will
+		 * be traced until it goes away.
+		 */
+		ctx->switch_log = kzalloc(sizeof(struct switch_log) +
+			SWITCH_LOG_BUFSIZE * sizeof(struct switch_log_entry),
+			GFP_KERNEL);
+		if (!ctx->switch_log)
+			goto out;
+		spin_lock_init(&ctx->switch_log->lock);
+		init_waitqueue_head(&ctx->switch_log->wait);
+	}
+	mutex_unlock(&ctx->mapping_lock);
+
+	return 0;
+ out:
+	mutex_unlock(&ctx->mapping_lock);
+	return -ENOMEM;
+}
+
+static int switch_log_sprint(struct spu_context *ctx, char *tbuf, int n)
+{
+	struct switch_log_entry *p;
+
+	p = ctx->switch_log->log + ctx->switch_log->tail % SWITCH_LOG_BUFSIZE;
+
+	return snprintf(tbuf, n, "%u.%09u %d %u %u %llu\n",
+			(unsigned int) p->tstamp.tv_sec,
+			(unsigned int) p->tstamp.tv_nsec,
+			p->spu_id,
+			(unsigned int) p->type,
+			(unsigned int) p->val,
+			(unsigned long long) p->timebase);
+}
+
+static ssize_t spufs_switch_log_read(struct file *file, char __user *buf,
+			     size_t len, loff_t *ppos)
+{
+	struct inode *inode = file->f_path.dentry->d_inode;
+	struct spu_context *ctx = SPUFS_I(inode)->i_ctx;
+	int error = 0, cnt = 0;
+
+	if (!buf || len < 0)
+		return -EINVAL;
+
+	while (cnt < len) {
+		char tbuf[128];
+		int width;
+
+		if (file->f_flags & O_NONBLOCK) {
+			if (spufs_switch_log_used(ctx) <= 0)
+				return cnt ? cnt : -EAGAIN;
+		} else {
+			/* Wait for data in buffer */
+			error = wait_event_interruptible(ctx->switch_log->wait,
+					spufs_switch_log_used(ctx) > 0);
+			if (error)
+				break;
+		}
+
+		spin_lock(&ctx->switch_log->lock);
+		if (ctx->switch_log->head == ctx->switch_log->tail) {
+			/* multiple readers race? */
+			spin_unlock(&ctx->switch_log->lock);
+			continue;
+		}
+
+		width = switch_log_sprint(ctx, tbuf, sizeof(tbuf));
+		if (width < len) {
+			ctx->switch_log->tail =
+				(ctx->switch_log->tail + 1) %
+				 SWITCH_LOG_BUFSIZE;
+		}
+
+		spin_unlock(&ctx->switch_log->lock);
+
+		/*
+		 * If the record is greater than space available return
+		 * partial buffer (so far)
+		 */
+		if (width >= len)
+			break;
+
+		error = copy_to_user(buf + cnt, tbuf, width);
+		if (error)
+			break;
+		cnt += width;
+	}
+
+	return cnt == 0 ? error : cnt;
+}
+
+static unsigned int spufs_switch_log_poll(struct file *file, poll_table *wait)
+{
+	struct inode *inode = file->f_path.dentry->d_inode;
+	struct spu_context *ctx = SPUFS_I(inode)->i_ctx;
+	unsigned int mask = 0;
+
+	poll_wait(file, &ctx->switch_log->wait, wait);
+
+	if (spufs_switch_log_used(ctx) > 0)
+		mask |= POLLIN;
+
+	return mask;
+}
+
+static const struct file_operations spufs_switch_log_fops = {
+	.owner	= THIS_MODULE,
+	.open	= spufs_switch_log_open,
+	.read	= spufs_switch_log_read,
+	.poll	= spufs_switch_log_poll,
+};
+
+void spu_switch_log_notify(struct spu *spu, struct spu_context *ctx,
+		u32 type, u32 val)
+{
+	if (!ctx->switch_log)
+		return;
+
+	spin_lock(&ctx->switch_log->lock);
+	if (spufs_switch_log_avail(ctx) > 1) {
+		struct switch_log_entry *p;
+
+		p = ctx->switch_log->log + ctx->switch_log->head;
+		ktime_get_ts(&p->tstamp);
+		p->timebase = get_tb();
+		p->spu_id = spu ? spu->number : -1;
+		p->type = type;
+		p->val = val;
+
+		ctx->switch_log->head =
+			(ctx->switch_log->head + 1) % SWITCH_LOG_BUFSIZE;
+	}
+	spin_unlock(&ctx->switch_log->lock);
+
+	wake_up(&ctx->switch_log->wait);
+}
 
 struct tree_descr spufs_dir_contents[] = {
 	{ "capabilities", &spufs_caps_fops, 0444, },
@@ -2422,6 +2588,7 @@ struct tree_descr spufs_dir_contents[] = {
 	{ "proxydma_info", &spufs_proxydma_info_fops, 0444, },
 	{ "tid", &spufs_tid_fops, 0444, },
 	{ "stat", &spufs_stat_fops, 0444, },
+	{ "switch_log", &spufs_switch_log_fops, 0444 },
 	{},
 };
 
diff --git a/arch/powerpc/platforms/cell/spufs/inode.c b/arch/powerpc/platforms/cell/spufs/inode.c
index 0c32a05ab06..f407b247185 100644
--- a/arch/powerpc/platforms/cell/spufs/inode.c
+++ b/arch/powerpc/platforms/cell/spufs/inode.c
@@ -23,6 +23,7 @@
 
 #include <linux/file.h>
 #include <linux/fs.h>
+#include <linux/fsnotify.h>
 #include <linux/backing-dev.h>
 #include <linux/init.h>
 #include <linux/ioctl.h>
@@ -223,7 +224,7 @@ static int spufs_dir_close(struct inode *inode, struct file *file)
 	parent = dir->d_parent->d_inode;
 	ctx = SPUFS_I(dir->d_inode)->i_ctx;
 
-	mutex_lock(&parent->i_mutex);
+	mutex_lock_nested(&parent->i_mutex, I_MUTEX_PARENT);
 	ret = spufs_rmdir(parent, dir);
 	mutex_unlock(&parent->i_mutex);
 	WARN_ON(ret);
@@ -618,12 +619,15 @@ long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode,
 	mode &= ~current->fs->umask;
 
 	if (flags & SPU_CREATE_GANG)
-		return spufs_create_gang(nd->path.dentry->d_inode,
+		ret = spufs_create_gang(nd->path.dentry->d_inode,
 					 dentry, nd->path.mnt, mode);
 	else
-		return spufs_create_context(nd->path.dentry->d_inode,
+		ret = spufs_create_context(nd->path.dentry->d_inode,
 					    dentry, nd->path.mnt, flags, mode,
 					    filp);
+	if (ret >= 0)
+		fsnotify_mkdir(nd->path.dentry->d_inode, dentry);
+	return ret;
 
 out_dput:
 	dput(dentry);
diff --git a/arch/powerpc/platforms/cell/spufs/run.c b/arch/powerpc/platforms/cell/spufs/run.c
index 96bf7c2b86f..b7493b86581 100644
--- a/arch/powerpc/platforms/cell/spufs/run.c
+++ b/arch/powerpc/platforms/cell/spufs/run.c
@@ -11,7 +11,7 @@
 #include "spufs.h"
 
 /* interrupt-level stop callback function. */
-void spufs_stop_callback(struct spu *spu)
+void spufs_stop_callback(struct spu *spu, int irq)
 {
 	struct spu_context *ctx = spu->ctx;
 
@@ -24,9 +24,19 @@ void spufs_stop_callback(struct spu *spu)
 	 */
 	if (ctx) {
 		/* Copy exception arguments into module specific structure */
-		ctx->csa.class_0_pending = spu->class_0_pending;
-		ctx->csa.dsisr = spu->dsisr;
-		ctx->csa.dar = spu->dar;
+		switch(irq) {
+		case 0 :
+			ctx->csa.class_0_pending = spu->class_0_pending;
+			ctx->csa.class_0_dsisr = spu->class_0_dsisr;
+			ctx->csa.class_0_dar = spu->class_0_dar;
+			break;
+		case 1 :
+			ctx->csa.class_1_dsisr = spu->class_1_dsisr;
+			ctx->csa.class_1_dar = spu->class_1_dar;
+			break;
+		case 2 :
+			break;
+		}
 
 		/* ensure that the exception status has hit memory before a
 		 * thread waiting on the context's stop queue is woken */
@@ -34,11 +44,6 @@ void spufs_stop_callback(struct spu *spu)
 
 		wake_up_all(&ctx->stop_wq);
 	}
-
-	/* Clear callback arguments from spu structure */
-	spu->class_0_pending = 0;
-	spu->dsisr = 0;
-	spu->dar = 0;
 }
 
 int spu_stopped(struct spu_context *ctx, u32 *stat)
@@ -56,7 +61,11 @@ int spu_stopped(struct spu_context *ctx, u32 *stat)
 	if (!(*stat & SPU_STATUS_RUNNING) && (*stat & stopped))
 		return 1;
 
-	dsisr = ctx->csa.dsisr;
+	dsisr = ctx->csa.class_0_dsisr;
+	if (dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED))
+		return 1;
+
+	dsisr = ctx->csa.class_1_dsisr;
 	if (dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED))
 		return 1;
 
@@ -294,7 +303,7 @@ static int spu_process_callback(struct spu_context *ctx)
 	u32 ls_pointer, npc;
 	void __iomem *ls;
 	long spu_ret;
-	int ret, ret2;
+	int ret;
 
 	/* get syscall block from local store */
 	npc = ctx->ops->npc_read(ctx) & ~3;
@@ -316,11 +325,9 @@ static int spu_process_callback(struct spu_context *ctx)
 		if (spu_ret <= -ERESTARTSYS) {
 			ret = spu_handle_restartsys(ctx, &spu_ret, &npc);
 		}
-		ret2 = spu_acquire(ctx);
+		mutex_lock(&ctx->state_mutex);
 		if (ret == -ERESTARTSYS)
 			return ret;
-		if (ret2)
-			return -EINTR;
 	}
 
 	/* need to re-get the ls, as it may have changed when we released the
@@ -343,13 +350,14 @@ long spufs_run_spu(struct spu_context *ctx, u32 *npc, u32 *event)
 	if (mutex_lock_interruptible(&ctx->run_mutex))
 		return -ERESTARTSYS;
 
-	spu_enable_spu(ctx);
 	ctx->event_return = 0;
 
 	ret = spu_acquire(ctx);
 	if (ret)
 		goto out_unlock;
 
+	spu_enable_spu(ctx);
+
 	spu_update_sched_info(ctx);
 
 	ret = spu_run_init(ctx, npc);
@@ -405,6 +413,8 @@ long spufs_run_spu(struct spu_context *ctx, u32 *npc, u32 *event)
 	ret = spu_run_fini(ctx, npc, &status);
 	spu_yield(ctx);
 
+	spu_switch_log_notify(NULL, ctx, SWITCH_LOG_EXIT, status);
+
 	if ((status & SPU_STATUS_STOPPED_BY_STOP) &&
 	    (((status >> SPU_STOP_STATUS_SHIFT) & 0x3f00) == 0x2100))
 		ctx->stats.libassist++;
diff --git a/arch/powerpc/platforms/cell/spufs/sched.c b/arch/powerpc/platforms/cell/spufs/sched.c
index 00528ef84ad..745dd51ec37 100644
--- a/arch/powerpc/platforms/cell/spufs/sched.c
+++ b/arch/powerpc/platforms/cell/spufs/sched.c
@@ -140,6 +140,9 @@ void __spu_update_sched_info(struct spu_context *ctx)
 	 * if it is timesliced or preempted.
 	 */
 	ctx->cpus_allowed = current->cpus_allowed;
+
+	/* Save the current cpu id for spu interrupt routing. */
+	ctx->last_ran = raw_smp_processor_id();
 }
 
 void spu_update_sched_info(struct spu_context *ctx)
@@ -240,9 +243,9 @@ static void spu_bind_context(struct spu *spu, struct spu_context *ctx)
 	spu->mfc_callback = spufs_mfc_callback;
 	mb();
 	spu_unmap_mappings(ctx);
+	spu_switch_log_notify(spu, ctx, SWITCH_LOG_START, 0);
 	spu_restore(&ctx->csa, spu);
 	spu->timestamp = jiffies;
-	spu_cpu_affinity_set(spu, raw_smp_processor_id());
 	spu_switch_notify(spu, ctx);
 	ctx->state = SPU_STATE_RUNNABLE;
 
@@ -419,6 +422,7 @@ static void spu_unbind_context(struct spu *spu, struct spu_context *ctx)
 	spu_switch_notify(spu, NULL);
 	spu_unmap_mappings(ctx);
 	spu_save(&ctx->csa, spu);
+	spu_switch_log_notify(spu, ctx, SWITCH_LOG_STOP, 0);
 	spu->timestamp = jiffies;
 	ctx->state = SPU_STATE_SAVED;
 	spu->ibox_callback = NULL;
@@ -591,7 +595,7 @@ static struct spu *find_victim(struct spu_context *ctx)
 	struct spu *spu;
 	int node, n;
 
-	spu_context_nospu_trace(spu_find_vitim__enter, ctx);
+	spu_context_nospu_trace(spu_find_victim__enter, ctx);
 
 	/*
 	 * Look for a possible preemption candidate on the local node first.
@@ -655,7 +659,8 @@ static struct spu *find_victim(struct spu_context *ctx)
 
 			victim->stats.invol_ctx_switch++;
 			spu->stats.invol_ctx_switch++;
-			spu_add_to_rq(victim);
+			if (test_bit(SPU_SCHED_SPU_RUN, &victim->sched_flags))
+				spu_add_to_rq(victim);
 
 			mutex_unlock(&victim->state_mutex);
 
@@ -1063,10 +1068,9 @@ int __init spu_sched_init(void)
 
 	mod_timer(&spuloadavg_timer, 0);
 
-	entry = create_proc_entry("spu_loadavg", 0, NULL);
+	entry = proc_create("spu_loadavg", 0, NULL, &spu_loadavg_fops);
 	if (!entry)
 		goto out_stop_kthread;
-	entry->proc_fops = &spu_loadavg_fops;
 
 	pr_debug("spusched: tick: %d, min ticks: %d, default ticks: %d\n",
 			SPUSCHED_TICK, MIN_SPU_TIMESLICE, DEF_SPU_TIMESLICE);
diff --git a/arch/powerpc/platforms/cell/spufs/spufs.h b/arch/powerpc/platforms/cell/spufs/spufs.h
index cdc515182f8..454c277c145 100644
--- a/arch/powerpc/platforms/cell/spufs/spufs.h
+++ b/arch/powerpc/platforms/cell/spufs/spufs.h
@@ -47,6 +47,30 @@ enum {
 	SPU_SCHED_SPU_RUN,	/* context is within spu_run */
 };
 
+enum {
+	SWITCH_LOG_BUFSIZE = 4096,
+};
+
+enum {
+	SWITCH_LOG_START,
+	SWITCH_LOG_STOP,
+	SWITCH_LOG_EXIT,
+};
+
+struct switch_log {
+	spinlock_t		lock;
+	wait_queue_head_t	wait;
+	unsigned long		head;
+	unsigned long		tail;
+	struct switch_log_entry {
+		struct timespec	tstamp;
+		s32		spu_id;
+		u32		type;
+		u32		val;
+		u64		timebase;
+	} log[];
+};
+
 struct spu_context {
 	struct spu *spu;		  /* pointer to a physical SPU */
 	struct spu_state csa;		  /* SPU context save area. */
@@ -97,6 +121,7 @@ struct spu_context {
 	cpumask_t cpus_allowed;
 	int policy;
 	int prio;
+	int last_ran;
 
 	/* statistics */
 	struct {
@@ -116,6 +141,9 @@ struct spu_context {
 		unsigned long long libassist;
 	} stats;
 
+	/* context switch log */
+	struct switch_log *switch_log;
+
 	struct list_head aff_list;
 	int aff_head;
 	int aff_offset;
@@ -256,6 +284,8 @@ int spu_activate(struct spu_context *ctx, unsigned long flags);
 void spu_deactivate(struct spu_context *ctx);
 void spu_yield(struct spu_context *ctx);
 void spu_switch_notify(struct spu *spu, struct spu_context *ctx);
+void spu_switch_log_notify(struct spu *spu, struct spu_context *ctx,
+		u32 type, u32 val);
 void spu_set_timeslice(struct spu_context *ctx);
 void spu_update_sched_info(struct spu_context *ctx);
 void __spu_update_sched_info(struct spu_context *ctx);
@@ -302,7 +332,7 @@ size_t spu_ibox_read(struct spu_context *ctx, u32 *data);
 /* irq callback funcs. */
 void spufs_ibox_callback(struct spu *spu);
 void spufs_wbox_callback(struct spu *spu);
-void spufs_stop_callback(struct spu *spu);
+void spufs_stop_callback(struct spu *spu, int irq);
 void spufs_mfc_callback(struct spu *spu);
 void spufs_dma_callback(struct spu *spu, int type);
 
@@ -330,8 +360,8 @@ extern void spuctx_switch_state(struct spu_context *ctx,
 		enum spu_utilization_state new_state);
 
 #define spu_context_trace(name, ctx, spu) \
-	trace_mark(name, "%p %p", ctx, spu);
+	trace_mark(name, "ctx %p spu %p", ctx, spu);
 #define spu_context_nospu_trace(name, ctx) \
-	trace_mark(name, "%p", ctx);
+	trace_mark(name, "ctx %p", ctx);
 
 #endif
diff --git a/arch/powerpc/platforms/cell/spufs/sputrace.c b/arch/powerpc/platforms/cell/spufs/sputrace.c
index 79aa773f3c9..53202422ba7 100644
--- a/arch/powerpc/platforms/cell/spufs/sputrace.c
+++ b/arch/powerpc/platforms/cell/spufs/sputrace.c
@@ -171,24 +171,24 @@ static void spu_context_nospu_event(void *probe_private, void *call_data,
 }
 
 struct spu_probe spu_probes[] = {
-	{ "spu_bind_context__enter", "%p %p", spu_context_event },
-	{ "spu_unbind_context__enter", "%p %p", spu_context_event },
-	{ "spu_get_idle__enter", "%p", spu_context_nospu_event },
-	{ "spu_get_idle__found", "%p %p", spu_context_event },
-	{ "spu_get_idle__not_found", "%p", spu_context_nospu_event },
-	{ "spu_find_victim__enter", "%p", spu_context_nospu_event },
-	{ "spusched_tick__preempt", "%p %p", spu_context_event },
-	{ "spusched_tick__newslice", "%p", spu_context_nospu_event },
-	{ "spu_yield__enter", "%p", spu_context_nospu_event },
-	{ "spu_deactivate__enter", "%p", spu_context_nospu_event },
-	{ "__spu_deactivate__unload", "%p %p", spu_context_event },
-	{ "spufs_ps_nopfn__enter", "%p", spu_context_nospu_event },
-	{ "spufs_ps_nopfn__sleep", "%p", spu_context_nospu_event },
-	{ "spufs_ps_nopfn__wake", "%p %p", spu_context_event },
-	{ "spufs_ps_nopfn__insert", "%p %p", spu_context_event },
-	{ "spu_acquire_saved__enter", "%p", spu_context_nospu_event },
-	{ "destroy_spu_context__enter", "%p", spu_context_nospu_event },
-	{ "spufs_stop_callback__enter", "%p %p", spu_context_event },
+	{ "spu_bind_context__enter", "ctx %p spu %p", spu_context_event },
+	{ "spu_unbind_context__enter", "ctx %p spu %p", spu_context_event },
+	{ "spu_get_idle__enter", "ctx %p", spu_context_nospu_event },
+	{ "spu_get_idle__found", "ctx %p spu %p", spu_context_event },
+	{ "spu_get_idle__not_found", "ctx %p", spu_context_nospu_event },
+	{ "spu_find_victim__enter", "ctx %p", spu_context_nospu_event },
+	{ "spusched_tick__preempt", "ctx %p spu %p", spu_context_event },
+	{ "spusched_tick__newslice", "ctx %p", spu_context_nospu_event },
+	{ "spu_yield__enter", "ctx %p", spu_context_nospu_event },
+	{ "spu_deactivate__enter", "ctx %p", spu_context_nospu_event },
+	{ "__spu_deactivate__unload", "ctx %p spu %p", spu_context_event },
+	{ "spufs_ps_nopfn__enter", "ctx %p", spu_context_nospu_event },
+	{ "spufs_ps_nopfn__sleep", "ctx %p", spu_context_nospu_event },
+	{ "spufs_ps_nopfn__wake", "ctx %p spu %p", spu_context_event },
+	{ "spufs_ps_nopfn__insert", "ctx %p spu %p", spu_context_event },
+	{ "spu_acquire_saved__enter", "ctx %p", spu_context_nospu_event },
+	{ "destroy_spu_context__enter", "ctx %p", spu_context_nospu_event },
+	{ "spufs_stop_callback__enter", "ctx %p spu %p", spu_context_event },
 };
 
 static int __init sputrace_init(void)
@@ -201,10 +201,9 @@ static int __init sputrace_init(void)
 	if (!sputrace_log)
 		goto out;
 
-	entry = create_proc_entry("sputrace", S_IRUSR, NULL);
+	entry = proc_create("sputrace", S_IRUSR, NULL, &sputrace_fops);
 	if (!entry)
 		goto out_free_log;
-	entry->proc_fops = &sputrace_fops;
 
 	for (i = 0; i < ARRAY_SIZE(spu_probes); i++) {
 		struct spu_probe *p = &spu_probes[i];
diff --git a/arch/powerpc/platforms/cell/spufs/switch.c b/arch/powerpc/platforms/cell/spufs/switch.c
index d2a1249d36d..3df9a36eb2f 100644
--- a/arch/powerpc/platforms/cell/spufs/switch.c
+++ b/arch/powerpc/platforms/cell/spufs/switch.c
@@ -132,6 +132,14 @@ static inline void disable_interrupts(struct spu_state *csa, struct spu *spu)
 	spu_int_mask_set(spu, 2, 0ul);
 	eieio();
 	spin_unlock_irq(&spu->register_lock);
+
+	/*
+	 * This flag needs to be set before calling synchronize_irq so
+	 * that the update will be visible to the relevant handlers
+	 * via a simple load.
+	 */
+	set_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
+	clear_bit(SPU_CONTEXT_FAULT_PENDING, &spu->flags);
 	synchronize_irq(spu->irqs[0]);
 	synchronize_irq(spu->irqs[1]);
 	synchronize_irq(spu->irqs[2]);
@@ -166,9 +174,8 @@ static inline void set_switch_pending(struct spu_state *csa, struct spu *spu)
 	/* Save, Step 7:
 	 * Restore, Step 5:
 	 *     Set a software context switch pending flag.
+	 *     Done above in Step 3 - disable_interrupts().
 	 */
-	set_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
-	mb();
 }
 
 static inline void save_mfc_cntl(struct spu_state *csa, struct spu *spu)
@@ -186,20 +193,21 @@ static inline void save_mfc_cntl(struct spu_state *csa, struct spu *spu)
 				 MFC_CNTL_SUSPEND_COMPLETE);
 		/* fall through */
 	case MFC_CNTL_SUSPEND_COMPLETE:
-		if (csa) {
+		if (csa)
 			csa->priv2.mfc_control_RW =
-				MFC_CNTL_SUSPEND_MASK |
+				in_be64(&priv2->mfc_control_RW) |
 				MFC_CNTL_SUSPEND_DMA_QUEUE;
-		}
 		break;
 	case MFC_CNTL_NORMAL_DMA_QUEUE_OPERATION:
 		out_be64(&priv2->mfc_control_RW, MFC_CNTL_SUSPEND_DMA_QUEUE);
 		POLL_WHILE_FALSE((in_be64(&priv2->mfc_control_RW) &
 				  MFC_CNTL_SUSPEND_DMA_STATUS_MASK) ==
 				 MFC_CNTL_SUSPEND_COMPLETE);
-		if (csa) {
-			csa->priv2.mfc_control_RW = 0;
-		}
+		if (csa)
+			csa->priv2.mfc_control_RW =
+				in_be64(&priv2->mfc_control_RW) &
+				~MFC_CNTL_SUSPEND_DMA_QUEUE &
+				~MFC_CNTL_SUSPEND_MASK;
 		break;
 	}
 }
@@ -249,16 +257,21 @@ static inline void save_spu_status(struct spu_state *csa, struct spu *spu)
 	}
 }
 
-static inline void save_mfc_decr(struct spu_state *csa, struct spu *spu)
+static inline void save_mfc_stopped_status(struct spu_state *csa,
+		struct spu *spu)
 {
 	struct spu_priv2 __iomem *priv2 = spu->priv2;
+	const u64 mask = MFC_CNTL_DECREMENTER_RUNNING |
+			MFC_CNTL_DMA_QUEUES_EMPTY;
 
 	/* Save, Step 12:
 	 *     Read MFC_CNTL[Ds].  Update saved copy of
 	 *     CSA.MFC_CNTL[Ds].
+	 *
+	 * update: do the same with MFC_CNTL[Q].
 	 */
-	csa->priv2.mfc_control_RW |=
-		in_be64(&priv2->mfc_control_RW) & MFC_CNTL_DECREMENTER_RUNNING;
+	csa->priv2.mfc_control_RW &= ~mask;
+	csa->priv2.mfc_control_RW |= in_be64(&priv2->mfc_control_RW) & mask;
 }
 
 static inline void halt_mfc_decr(struct spu_state *csa, struct spu *spu)
@@ -462,7 +475,9 @@ static inline void purge_mfc_queue(struct spu_state *csa, struct spu *spu)
 	 * Restore, Step 14.
 	 *     Write MFC_CNTL[Pc]=1 (purge queue).
 	 */
-	out_be64(&priv2->mfc_control_RW, MFC_CNTL_PURGE_DMA_REQUEST);
+	out_be64(&priv2->mfc_control_RW,
+			MFC_CNTL_PURGE_DMA_REQUEST |
+			MFC_CNTL_SUSPEND_MASK);
 	eieio();
 }
 
@@ -725,10 +740,14 @@ static inline void set_switch_active(struct spu_state *csa, struct spu *spu)
 	/* Save, Step 48:
 	 * Restore, Step 23.
 	 *     Change the software context switch pending flag
-	 *     to context switch active.
+	 *     to context switch active.  This implementation does
+	 *     not uses a switch active flag.
 	 *
-	 *     This implementation does not uses a switch active flag.
+	 * Now that we have saved the mfc in the csa, we can add in the
+	 * restart command if an exception occurred.
 	 */
+	if (test_bit(SPU_CONTEXT_FAULT_PENDING, &spu->flags))
+		csa->priv2.mfc_control_RW |= MFC_CNTL_RESTART_DMA_COMMAND;
 	clear_bit(SPU_CONTEXT_SWITCH_PENDING, &spu->flags);
 	mb();
 }
@@ -1690,6 +1709,13 @@ static inline void restore_mfc_sr1(struct spu_state *csa, struct spu *spu)
 	eieio();
 }
 
+static inline void set_int_route(struct spu_state *csa, struct spu *spu)
+{
+	struct spu_context *ctx = spu->ctx;
+
+	spu_cpu_affinity_set(spu, ctx->last_ran);
+}
+
 static inline void restore_other_spu_access(struct spu_state *csa,
 					    struct spu *spu)
 {
@@ -1721,15 +1747,15 @@ static inline void restore_mfc_cntl(struct spu_state *csa, struct spu *spu)
 	 */
 	out_be64(&priv2->mfc_control_RW, csa->priv2.mfc_control_RW);
 	eieio();
+
 	/*
-	 * FIXME: this is to restart a DMA that we were processing
-	 *        before the save. better remember the fault information
-	 *        in the csa instead.
+	 * The queue is put back into the same state that was evident prior to
+	 * the context switch. The suspend flag is added to the saved state in
+	 * the csa, if the operational state was suspending or suspended. In
+	 * this case, the code that suspended the mfc is responsible for
+	 * continuing it. Note that SPE faults do not change the operational
+	 * state of the spu.
 	 */
-	if ((csa->priv2.mfc_control_RW & MFC_CNTL_SUSPEND_DMA_QUEUE_MASK)) {
-		out_be64(&priv2->mfc_control_RW, MFC_CNTL_RESTART_DMA_COMMAND);
-		eieio();
-	}
 }
 
 static inline void enable_user_access(struct spu_state *csa, struct spu *spu)
@@ -1788,7 +1814,7 @@ static int quiece_spu(struct spu_state *prev, struct spu *spu)
 	save_spu_runcntl(prev, spu);	        /* Step 9. */
 	save_mfc_sr1(prev, spu);	        /* Step 10. */
 	save_spu_status(prev, spu);	        /* Step 11. */
-	save_mfc_decr(prev, spu);	        /* Step 12. */
+	save_mfc_stopped_status(prev, spu);     /* Step 12. */
 	halt_mfc_decr(prev, spu);	        /* Step 13. */
 	save_timebase(prev, spu);		/* Step 14. */
 	remove_other_spu_access(prev, spu);	/* Step 15. */
@@ -2000,6 +2026,7 @@ static void restore_csa(struct spu_state *next, struct spu *spu)
 	check_ppuint_mb_stat(next, spu);	/* Step 67. */
 	spu_invalidate_slbs(spu);		/* Modified Step 68. */
 	restore_mfc_sr1(next, spu);	        /* Step 69. */
+	set_int_route(next, spu);		/* NEW      */
 	restore_other_spu_access(next, spu);	/* Step 70. */
 	restore_spu_runcntl(next, spu);	        /* Step 71. */
 	restore_mfc_cntl(next, spu);	        /* Step 72. */