Merge branch 'cell-merge' of git+ssh://master.kernel.org/pub/scm/linux/kernel/git/arnd/cell-2.6

arch/powerpc/platforms/cell/pmu.c

@@ -345,18 +345,12 @@ EXPORT_SYMBOL_GPL(cbe_read_trace_buffer);
  * Enabling/disabling interrupts for the entire performance monitoring unit.
  */
 
-u32 cbe_query_pm_interrupts(u32 cpu)
-{
-	return cbe_read_pm(cpu, pm_status);
-}
-EXPORT_SYMBOL_GPL(cbe_query_pm_interrupts);
-
-u32 cbe_clear_pm_interrupts(u32 cpu)
+u32 cbe_get_and_clear_pm_interrupts(u32 cpu)
 {
 	/* Reading pm_status clears the interrupt bits. */
-	return cbe_query_pm_interrupts(cpu);
+	return cbe_read_pm(cpu, pm_status);
 }
-EXPORT_SYMBOL_GPL(cbe_clear_pm_interrupts);
+EXPORT_SYMBOL_GPL(cbe_get_and_clear_pm_interrupts);
 
 void cbe_enable_pm_interrupts(u32 cpu, u32 thread, u32 mask)
 {
@@ -371,7 +365,7 @@ EXPORT_SYMBOL_GPL(cbe_enable_pm_interrupts);
 
 void cbe_disable_pm_interrupts(u32 cpu)
 {
-	cbe_clear_pm_interrupts(cpu);
+	cbe_get_and_clear_pm_interrupts(cpu);
 	cbe_write_pm(cpu, pm_status, 0);
 }
 EXPORT_SYMBOL_GPL(cbe_disable_pm_interrupts);

arch/powerpc/platforms/cell/spufs/context.c

@@ -42,7 +42,7 @@ struct spu_context *alloc_spu_context(struct spu_gang *gang)
 	}
 	spin_lock_init(&ctx->mmio_lock);
 	kref_init(&ctx->kref);
-	init_rwsem(&ctx->state_sema);
+	mutex_init(&ctx->state_mutex);
 	init_MUTEX(&ctx->run_sema);
 	init_waitqueue_head(&ctx->ibox_wq);
 	init_waitqueue_head(&ctx->wbox_wq);
@@ -53,6 +53,10 @@ struct spu_context *alloc_spu_context(struct spu_gang *gang)
 	ctx->owner = get_task_mm(current);
 	if (gang)
 		spu_gang_add_ctx(gang, ctx);
+	ctx->rt_priority = current->rt_priority;
+	ctx->policy = current->policy;
+	ctx->prio = current->prio;
+	INIT_DELAYED_WORK(&ctx->sched_work, spu_sched_tick);
 	goto out;
 out_free:
 	kfree(ctx);
@@ -65,9 +69,9 @@ void destroy_spu_context(struct kref *kref)
 {
 	struct spu_context *ctx;
 	ctx = container_of(kref, struct spu_context, kref);
-	down_write(&ctx->state_sema);
+	mutex_lock(&ctx->state_mutex);
 	spu_deactivate(ctx);
-	up_write(&ctx->state_sema);
+	mutex_unlock(&ctx->state_mutex);
 	spu_fini_csa(&ctx->csa);
 	if (ctx->gang)
 		spu_gang_remove_ctx(ctx->gang, ctx);
@@ -96,16 +100,6 @@ void spu_forget(struct spu_context *ctx)
 	spu_release(ctx);
 }
 
-void spu_acquire(struct spu_context *ctx)
-{
-	down_read(&ctx->state_sema);
-}
-
-void spu_release(struct spu_context *ctx)
-{
-	up_read(&ctx->state_sema);
-}
-
 void spu_unmap_mappings(struct spu_context *ctx)
 {
 	if (ctx->local_store)
@@ -124,83 +118,84 @@ void spu_unmap_mappings(struct spu_context *ctx)
 		unmap_mapping_range(ctx->psmap, 0, 0x20000, 1);
 }
 
+/**
+ * spu_acquire_exclusive - lock spu contex and protect against userspace access
+ * @ctx:	spu contex to lock
+ *
+ * Note:
+ *	Returns 0 and with the context locked on success
+ *	Returns negative error and with the context _unlocked_ on failure.
+ */
 int spu_acquire_exclusive(struct spu_context *ctx)
 {
-	int ret = 0;
+	int ret = -EINVAL;
 
-	down_write(&ctx->state_sema);
-	/* ctx is about to be freed, can't acquire any more */
-	if (!ctx->owner) {
-		ret = -EINVAL;
-		goto out;
-	}
+	spu_acquire(ctx);
+	/*
+	 * Context is about to be freed, so we can't acquire it anymore.
+	 */
+	if (!ctx->owner)
+		goto out_unlock;
 
 	if (ctx->state == SPU_STATE_SAVED) {
 		ret = spu_activate(ctx, 0);
 		if (ret)
-			goto out;
-		ctx->state = SPU_STATE_RUNNABLE;
+			goto out_unlock;
 	} else {
-		/* We need to exclude userspace access to the context. */
+		/*
+		 * We need to exclude userspace access to the context.
+		 *
+		 * To protect against memory access we invalidate all ptes
+		 * and make sure the pagefault handlers block on the mutex.
+		 */
 		spu_unmap_mappings(ctx);
 	}
 
-out:
-	if (ret)
-		up_write(&ctx->state_sema);
+	return 0;
+
+ out_unlock:
+	spu_release(ctx);
 	return ret;
 }
 
-int spu_acquire_runnable(struct spu_context *ctx)
+/**
+ * spu_acquire_runnable - lock spu contex and make sure it is in runnable state
+ * @ctx:	spu contex to lock
+ *
+ * Note:
+ *	Returns 0 and with the context locked on success
+ *	Returns negative error and with the context _unlocked_ on failure.
+ */
+int spu_acquire_runnable(struct spu_context *ctx, unsigned long flags)
 {
-	int ret = 0;
-
-	down_read(&ctx->state_sema);
-	if (ctx->state == SPU_STATE_RUNNABLE) {
-		ctx->spu->prio = current->prio;
-		return 0;
-	}
-	up_read(&ctx->state_sema);
-
-	down_write(&ctx->state_sema);
-	/* ctx is about to be freed, can't acquire any more */
-	if (!ctx->owner) {
-		ret = -EINVAL;
-		goto out;
-	}
+	int ret = -EINVAL;
 
+	spu_acquire(ctx);
 	if (ctx->state == SPU_STATE_SAVED) {
-		ret = spu_activate(ctx, 0);
+		/*
+		 * Context is about to be freed, so we can't acquire it anymore.
+		 */
+		if (!ctx->owner)
+			goto out_unlock;
+		ret = spu_activate(ctx, flags);
 		if (ret)
-			goto out;
-		ctx->state = SPU_STATE_RUNNABLE;
+			goto out_unlock;
 	}
 
-	downgrade_write(&ctx->state_sema);
-	/* On success, we return holding the lock */
-
-	return ret;
-out:
-	/* Release here, to simplify calling code. */
-	up_write(&ctx->state_sema);
+	return 0;
 
+ out_unlock:
+	spu_release(ctx);
 	return ret;
 }
 
+/**
+ * spu_acquire_saved - lock spu contex and make sure it is in saved state
+ * @ctx:	spu contex to lock
+ */
 void spu_acquire_saved(struct spu_context *ctx)
 {
-	down_read(&ctx->state_sema);
-
-	if (ctx->state == SPU_STATE_SAVED)
-		return;
-
-	up_read(&ctx->state_sema);
-	down_write(&ctx->state_sema);
-
-	if (ctx->state == SPU_STATE_RUNNABLE) {
+	spu_acquire(ctx);
+	if (ctx->state != SPU_STATE_SAVED)
 		spu_deactivate(ctx);
-		ctx->state = SPU_STATE_SAVED;
-	}
-
-	downgrade_write(&ctx->state_sema);
 }

arch/powerpc/platforms/cell/spufs/file.c

@@ -103,6 +103,9 @@ static unsigned long spufs_mem_mmap_nopfn(struct vm_area_struct *vma,
 
 	offset += vma->vm_pgoff << PAGE_SHIFT;
 
+	if (offset >= LS_SIZE)
+		return NOPFN_SIGBUS;
+
 	spu_acquire(ctx);
 
 	if (ctx->state == SPU_STATE_SAVED) {
@@ -164,7 +167,7 @@ static unsigned long spufs_ps_nopfn(struct vm_area_struct *vma,
 	/* error here usually means a signal.. we might want to test
 	 * the error code more precisely though
 	 */
-	ret = spu_acquire_runnable(ctx);
+	ret = spu_acquire_runnable(ctx, 0);
 	if (ret)
 		return NOPFN_REFAULT;
 
@@ -1306,7 +1309,7 @@ static ssize_t spufs_mfc_write(struct file *file, const char __user *buffer,
 	if (ret)
 		goto out;
 
-	spu_acquire_runnable(ctx);
+	spu_acquire_runnable(ctx, 0);
 	if (file->f_flags & O_NONBLOCK) {
 		ret = ctx->ops->send_mfc_command(ctx, &cmd);
 	} else {

arch/powerpc/platforms/cell/spufs/run.c

@@ -133,7 +133,7 @@ out_drop_priv:
 	spu_mfc_sr1_set(ctx->spu, sr1);
 
 out_unlock:
-	spu_release_exclusive(ctx);
+	spu_release(ctx);
 out:
 	return ret;
 }
@@ -143,7 +143,7 @@ static inline int spu_run_init(struct spu_context *ctx, u32 * npc)
 	int ret;
 	unsigned long runcntl = SPU_RUNCNTL_RUNNABLE;
 
-	ret = spu_acquire_runnable(ctx);
+	ret = spu_acquire_runnable(ctx, SPU_ACTIVATE_NOWAKE);
 	if (ret)
 		return ret;
 
@@ -155,7 +155,7 @@ static inline int spu_run_init(struct spu_context *ctx, u32 * npc)
 			spu_release(ctx);
 			ret = spu_setup_isolated(ctx);
 			if (!ret)
-				ret = spu_acquire_runnable(ctx);
+				ret = spu_acquire_runnable(ctx, SPU_ACTIVATE_NOWAKE);
 		}
 
 		/* if userspace has set the runcntrl register (eg, to issue an
@@ -164,8 +164,10 @@ static inline int spu_run_init(struct spu_context *ctx, u32 * npc)
 			(SPU_RUNCNTL_RUNNABLE | SPU_RUNCNTL_ISOLATE);
 		if (runcntl == 0)
 			runcntl = SPU_RUNCNTL_RUNNABLE;
-	} else
+	} else {
+		spu_start_tick(ctx);
 		ctx->ops->npc_write(ctx, *npc);
+	}
 
 	ctx->ops->runcntl_write(ctx, runcntl);
 	return ret;
@@ -176,6 +178,7 @@ static inline int spu_run_fini(struct spu_context *ctx, u32 * npc,
 {
 	int ret = 0;
 
+	spu_stop_tick(ctx);
 	*status = ctx->ops->status_read(ctx);
 	*npc = ctx->ops->npc_read(ctx);
 	spu_release(ctx);
@@ -329,8 +332,10 @@ long spufs_run_spu(struct file *file, struct spu_context *ctx,
 		}
 		if (unlikely(ctx->state != SPU_STATE_RUNNABLE)) {
 			ret = spu_reacquire_runnable(ctx, npc, &status);
-			if (ret)
+			if (ret) {
+				spu_stop_tick(ctx);
 				goto out2;
+			}
 			continue;
 		}
 		ret = spu_process_events(ctx);
@@ -361,4 +366,3 @@ out:
 	up(&ctx->run_sema);
 	return ret;
 }
-

arch/powerpc/platforms/cell/spufs/spufs.h

@@ -23,7 +23,7 @@
 #define SPUFS_H
 
 #include <linux/kref.h>
-#include <linux/rwsem.h>
+#include <linux/mutex.h>
 #include <linux/spinlock.h>
 #include <linux/fs.h>
 
@@ -37,11 +37,13 @@ enum {
 };
 
 struct spu_context_ops;
-
-#define SPU_CONTEXT_PREEMPT          0UL
-
 struct spu_gang;
 
+/* ctx->sched_flags */
+enum {
+	SPU_SCHED_WAKE = 0,
+};
+
 struct spu_context {
 	struct spu *spu;		  /* pointer to a physical SPU */
 	struct spu_state csa;		  /* SPU context save area. */
@@ -56,7 +58,7 @@ struct spu_context {
 	u64 object_id;		   /* user space pointer for oprofile */
 
 	enum { SPU_STATE_RUNNABLE, SPU_STATE_SAVED } state;
-	struct rw_semaphore state_sema;
+	struct mutex state_mutex;
 	struct semaphore run_sema;
 
 	struct mm_struct *owner;
@@ -77,6 +79,14 @@ struct spu_context {
 
 	struct list_head gang_list;
 	struct spu_gang *gang;
+
+	/* scheduler fields */
+ 	struct list_head rq;
+	struct delayed_work sched_work;
+	unsigned long sched_flags;
+	unsigned long rt_priority;
+	int policy;
+	int prio;
 };
 
 struct spu_gang {
@@ -161,6 +171,16 @@ void spu_gang_remove_ctx(struct spu_gang *gang, struct spu_context *ctx);
 void spu_gang_add_ctx(struct spu_gang *gang, struct spu_context *ctx);
 
 /* context management */
+static inline void spu_acquire(struct spu_context *ctx)
+{
+	mutex_lock(&ctx->state_mutex);
+}
+
+static inline void spu_release(struct spu_context *ctx)
+{
+	mutex_unlock(&ctx->state_mutex);
+}
+
 struct spu_context * alloc_spu_context(struct spu_gang *gang);
 void destroy_spu_context(struct kref *kref);
 struct spu_context * get_spu_context(struct spu_context *ctx);
@@ -168,20 +188,18 @@ int put_spu_context(struct spu_context *ctx);
 void spu_unmap_mappings(struct spu_context *ctx);
 
 void spu_forget(struct spu_context *ctx);
-void spu_acquire(struct spu_context *ctx);
-void spu_release(struct spu_context *ctx);
-int spu_acquire_runnable(struct spu_context *ctx);
+int spu_acquire_runnable(struct spu_context *ctx, unsigned long flags);
 void spu_acquire_saved(struct spu_context *ctx);
 int spu_acquire_exclusive(struct spu_context *ctx);
-
-static inline void spu_release_exclusive(struct spu_context *ctx)
-{
-	up_write(&ctx->state_sema);
-}
-
-int spu_activate(struct spu_context *ctx, u64 flags);
+enum {
+	SPU_ACTIVATE_NOWAKE = 1,
+};
+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_start_tick(struct spu_context *ctx);
+void spu_stop_tick(struct spu_context *ctx);
+void spu_sched_tick(struct work_struct *work);
 int __init spu_sched_init(void);
 void __exit spu_sched_exit(void);
 

arch/powerpc/xmon/xmon.c

@@ -2811,7 +2811,6 @@ static void dump_spu_fields(struct spu *spu)
 	DUMP_FIELD(spu, "0x%lx", irqs[2]);
 	DUMP_FIELD(spu, "0x%x", slb_replace);
 	DUMP_FIELD(spu, "%d", pid);
-	DUMP_FIELD(spu, "%d", prio);
 	DUMP_FIELD(spu, "0x%p", mm);
 	DUMP_FIELD(spu, "0x%p", ctx);
 	DUMP_FIELD(spu, "0x%p", rq);

include/asm-powerpc/cell-pmu.h

@@ -53,6 +53,11 @@
 #define CBE_PM_CTR_POLARITY                0x01000000
 #define CBE_PM_CTR_COUNT_CYCLES            0x00800000
 #define CBE_PM_CTR_ENABLE                  0x00400000
+#define PM07_CTR_INPUT_MUX(x)              (((x) & 0x3F) << 26)
+#define PM07_CTR_INPUT_CONTROL(x)          (((x) & 1) << 25)
+#define PM07_CTR_POLARITY(x)               (((x) & 1) << 24)
+#define PM07_CTR_COUNT_CYCLES(x)           (((x) & 1) << 23)
+#define PM07_CTR_ENABLE(x)                 (((x) & 1) << 22)
 
 /* Macros for the pm_status register. */
 #define CBE_PM_CTR_OVERFLOW_INTR(ctr)      (1 << (31 - ((ctr) & 7)))
@@ -89,8 +94,7 @@ extern void cbe_read_trace_buffer(u32 cpu, u64 *buf);
 
 extern void cbe_enable_pm_interrupts(u32 cpu, u32 thread, u32 mask);
 extern void cbe_disable_pm_interrupts(u32 cpu);
-extern u32  cbe_query_pm_interrupts(u32 cpu);
-extern u32  cbe_clear_pm_interrupts(u32 cpu);
+extern u32  cbe_get_and_clear_pm_interrupts(u32 cpu);
 extern void cbe_sync_irq(int node);
 
 /* Utility functions, macros */
@@ -103,11 +107,4 @@ extern u32 cbe_get_hw_thread_id(int cpu);
 #define CBE_COUNT_PROBLEM_MODE          2
 #define CBE_COUNT_ALL_MODES             3
 
-/* Macros for the pm07_control registers. */
-#define PM07_CTR_INPUT_MUX(x)                    (((x) & 0x3F) << 26)
-#define PM07_CTR_INPUT_CONTROL(x)                (((x) & 1) << 25)
-#define PM07_CTR_POLARITY(x)                     (((x) & 1) << 24)
-#define PM07_CTR_COUNT_CYCLES(x)                 (((x) & 1) << 23)
-#define PM07_CTR_ENABLE(x)                       (((x) & 1) << 22)
-
 #endif /* __ASM_CELL_PMU_H__ */

include/asm-powerpc/spu.h

@@ -129,7 +129,6 @@ struct spu {
 	struct spu_runqueue *rq;
 	unsigned long long timestamp;
 	pid_t pid;
-	int prio;
 	int class_0_pending;
 	spinlock_t register_lock;