consolidate arch ctxsw locking (non-bitop version)

diff -puN kernel/sched.c~task-running-flag kernel/sched.c
@@ -265,14 +265,71 @@ static DEFINE_PER_CPU(struct runqueue, r
 #define task_rq(p)		cpu_rq(task_cpu(p))
 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
 
-/*
- * Default context-switch locking:
- */
 #ifndef prepare_arch_switch
-# define prepare_arch_switch(rq, next)	do { } while (0)
-# define finish_arch_switch(rq, next)	spin_unlock_irq(&(rq)->lock)
-# define task_running(rq, p)		((rq)->curr == (p))
+# define prepare_arch_switch(next)	do { } while (0)
+#endif
+#ifndef finish_arch_switch
+# define finish_arch_switch(prev)	do { } while (0)
+#endif
+
+#ifndef __ARCH_WANT_UNLOCKED_CTXSW
+static inline int task_running(runqueue_t *rq, task_t *p)
+{
+	return rq->curr == p;
+}
+
+static inline void prepare_lock_switch(runqueue_t *rq, task_t *next)
+{
+}
+
+static inline void finish_lock_switch(runqueue_t *rq, task_t *prev)
+{
+	spin_unlock_irq(&rq->lock);
+}
+
+#else /* __ARCH_WANT_UNLOCKED_CTXSW */
+static inline int task_running(runqueue_t *rq, task_t *p)
+{
+#ifdef CONFIG_SMP
+	return p->oncpu;
+#else
+	return rq->curr == p;
+#endif
+}
+
+static inline void prepare_lock_switch(runqueue_t *rq, task_t *next)
+{
+#ifdef CONFIG_SMP
+	/*
+	 * We can optimise this out completely for !SMP, because the
+	 * SMP rebalancing from interrupt is the only thing that cares
+	 * here.
+	 */
+	next->oncpu = 1;
+#endif
+#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
+	spin_unlock_irq(&rq->lock);
+#else
+	spin_unlock(&rq->lock);
+#endif
+}
+
+static inline void finish_lock_switch(runqueue_t *rq, task_t *prev)
+{
+#ifdef CONFIG_SMP
+	/*
+	 * After ->oncpu is cleared, the task can be moved to a different CPU.
+	 * We must ensure this doesn't happen until the switch is completely
+	 * finished.
+	 */
+	smp_wmb();
+	prev->oncpu = 0;
 #endif
+#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
+	local_irq_enable();
+#endif
+}
+#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
 
 /*
  * task_rq_lock - lock the runqueue a given task resides on and disable

@@ -1006,16 +1063,14 @@ void fastcall sched_fork(task_t *p)
 	p->state = TASK_RUNNING;
 	INIT_LIST_HEAD(&p->run_list);
 	p->array = NULL;
-	spin_lock_init(&p->switch_lock);
+#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
+	p->oncpu = 0;
+#endif
 #ifdef CONFIG_PREEMPT
-	/*
-	 * During context-switch we hold precisely one spinlock, which
-	 * schedule_tail drops. (in the common case it's this_rq()->lock,
-	 * but it also can be p->switch_lock.) So we compensate with a count
-	 * of 1. Also, we want to start with kernel preemption disabled.
-	 */
+	/* Want to start with kernel preemption disabled. */
 	p->thread_info->preempt_count = 1;
 #endif
+
 	/*
 	 * Share the timeslice between parent and child, thus the
 	 * total amount of pending timeslices in the system doesn't change,

@@ -1152,7 +1207,8 @@ static void finish_task_switch(task_t *p
 	 *		Manfred Spraul <[email protected]>
 	 */
 	prev_task_flags = prev->flags;
-	finish_arch_switch(rq, prev);
+	finish_arch_switch(prev);
+	finish_lock_switch(rq, prev);
 	if (mm)
 		mmdrop(mm);
 	if (unlikely(prev_task_flags & PF_DEAD))

@@ -1166,7 +1222,10 @@ static void finish_task_switch(task_t *p
 asmlinkage void schedule_tail(task_t *prev)
 {
 	finish_task_switch(prev);
-
+#ifdef __ARCH_WANT_UNLOCKED_CTXSW
+	/* In this case, finish_task_switch does not reenable preemption */
+	preempt_enable();
+#endif
 	if (current->set_child_tid)
 		put_user(current->pid, current->set_child_tid);
 }

@@ -2464,10 +2523,10 @@ switch_tasks:
 		rq->curr = next;
 		++*switch_count;
 
-		prepare_arch_switch(rq, next);
+		prepare_lock_switch(rq, next);
+		prepare_arch_switch(next);
 		prev = context_switch(rq, prev, next);
 		barrier();
-
 		finish_task_switch(prev);
 	} else
 		spin_unlock_irq(&rq->lock);

@@ -3429,6 +3488,9 @@ void __devinit init_idle(task_t *idle, i
 	double_rq_lock(idle_rq, rq);
 
 	idle_rq->curr = idle_rq->idle = idle;
+#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
+	idle->oncpu = 1;
+#endif
 	deactivate_task(idle, rq);
 	idle->array = NULL;
 	idle->prio = MAX_PRIO;

@@ -4124,6 +4186,9 @@ void __init sched_init(void)
 	rq = this_rq();
 	rq->curr = current;
 	rq->idle = current;
+#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
+	current->oncpu = 1;
+#endif
 	set_task_cpu(current, smp_processor_id());
 	wake_up_forked_process(current);
 

diff -puN include/linux/sched.h~task-running-flag include/linux/sched.h
@@ -315,6 +315,11 @@ struct signal_struct {
 
 #define rt_task(p)		((p)->prio < MAX_RT_PRIO)
 
+/* Context switch must be unlocked if interrupts are to be enabled */
+#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
+# define __ARCH_WANT_UNLOCKED_CTXSW
+#endif
+
 /*
  * Some day this will be a full-fledged user tracking system..
  */

@@ -406,6 +411,9 @@ struct task_struct {
 
 	int lock_depth;		/* Lock depth */
 
+#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
+	int oncpu;
+#endif
 	int prio, static_prio;
 	struct list_head run_list;
 	prio_array_t *array;

@@ -508,8 +516,6 @@ struct task_struct {
 	spinlock_t alloc_lock;
 /* Protection of proc_dentry: nesting proc_lock, dcache_lock, write_lock_irq(&tasklist_lock); */
 	spinlock_t proc_lock;
-/* context-switch lock */
-	spinlock_t switch_lock;
 
 /* journalling filesystem info */
 	void *journal_info;

diff -puN include/linux/init_task.h~task-running-flag include/linux/init_task.h
@@ -110,7 +110,6 @@ extern struct group_info init_groups;
 	.blocked	= {{0}},					\
 	.alloc_lock	= SPIN_LOCK_UNLOCKED,				\
 	.proc_lock	= SPIN_LOCK_UNLOCKED,				\
-	.switch_lock	= SPIN_LOCK_UNLOCKED,				\
 	.journal_info	= NULL,						\
 }
 

diff -puN include/asm-ia64/system.h~task-running-flag include/asm-ia64/system.h
@@ -183,8 +183,6 @@ do {								\
 
 #ifdef __KERNEL__
 
-#define prepare_to_switch()    do { } while(0)
-
 #ifdef CONFIG_IA32_SUPPORT
 # define IS_IA32_PROCESS(regs)	(ia64_psr(regs)->is != 0)
 #else

@@ -274,13 +272,7 @@ extern void ia64_load_extra (struct task
  * of that CPU which will not be released, because there we wait for the
  * tasklist_lock to become available.
  */
-#define prepare_arch_switch(rq, next)		\
-do {						\
-	spin_lock(&(next)->switch_lock);	\
-	spin_unlock(&(rq)->lock);		\
-} while (0)
-#define finish_arch_switch(rq, prev)	spin_unlock_irq(&(prev)->switch_lock)
-#define task_running(rq, p) 		((rq)->curr == (p) || spin_is_locked(&(p)->switch_lock))
+#define __ARCH_WANT_UNLOCKED_CTXSW
 
 #define ia64_platform_is(x) (strcmp(x, platform_name) == 0)
 

diff -puN include/asm-mips/system.h~task-running-flag include/asm-mips/system.h
@@ -491,15 +491,9 @@ static __inline__ int con_is_present(voi
 }
 
 /*
- * Taken from include/asm-ia64/system.h; prevents deadlock on SMP
+ * See include/asm-ia64/system.h; prevents deadlock on SMP
  * systems.
  */
-#define prepare_arch_switch(rq, next)		\
-do {						\
-	spin_lock(&(next)->switch_lock);	\
-	spin_unlock(&(rq)->lock);		\
-} while (0)
-#define finish_arch_switch(rq, prev)	spin_unlock_irq(&(prev)->switch_lock)
-#define task_running(rq, p) 		((rq)->curr == (p) || spin_is_locked(&(p)->switch_lock))
+#define __ARCH_WANT_UNLOCKED_CTXSW
 
 #endif /* _ASM_SYSTEM_H */

diff -puN include/asm-s390/system.h~task-running-flag include/asm-s390/system.h
@@ -103,11 +103,8 @@ static inline void restore_access_regs(u
 	prev = __switch_to(prev,next);					     \
 } while (0)
 
-#define prepare_arch_switch(rq, next)	do { } while(0)
-#define task_running(rq, p)		((rq)->curr == (p))
-#define finish_arch_switch(rq, prev) do {				     \
+#define finish_arch_switch(prev) do {					     \
 	set_fs(current->thread.mm_segment);				     \
-	spin_unlock_irq(&(rq)->lock);					     \
 } while (0)
 
 #define nop() __asm__ __volatile__ ("nop")

diff -puN include/asm-sparc/system.h~task-running-flag include/asm-sparc/system.h
@@ -101,7 +101,7 @@ extern void fpsave(unsigned long *fpregs
  * SWITCH_ENTER and SWITH_DO_LAZY_FPU do not work yet (e.g. SMP does not work)
  * XXX WTF is the above comment? Found in late teen 2.4.x.
  */
-#define prepare_arch_switch(rq, next) do { \
+#define prepare_arch_switch(next) do { \
 	__asm__ __volatile__( \
 	".globl\tflush_patch_switch\nflush_patch_switch:\n\t" \
 	"save %sp, -0x40, %sp; save %sp, -0x40, %sp; save %sp, -0x40, %sp\n\t" \

@@ -109,8 +109,6 @@ extern void fpsave(unsigned long *fpregs
 	"save %sp, -0x40, %sp\n\t" \
 	"restore; restore; restore; restore; restore; restore; restore"); \
 } while(0)
-#define finish_arch_switch(rq, next)	spin_unlock_irq(&(rq)->lock)
-#define task_running(rq, p)		((rq)->curr == (p))
 
 	/* Much care has gone into this code, do not touch it.
 	 *

diff -puN include/asm-sparc64/system.h~task-running-flag include/asm-sparc64/system.h
@@ -139,19 +139,13 @@ extern void __flushw_user(void);
 #define flush_user_windows flushw_user
 #define flush_register_windows flushw_all
 
-#define prepare_arch_switch(rq, next)		\
-do {	spin_lock(&(next)->switch_lock);	\
-	spin_unlock(&(rq)->lock);		\
+/* Don't hold the runqueue lock over context switch */
+#define __ARCH_WANT_UNLOCKED_CTXSW
+#define prepare_arch_switch(next)		\
+do {						\
 	flushw_all();				\
 } while (0)
 
-#define finish_arch_switch(rq, prev)		\
-do {	spin_unlock_irq(&(prev)->switch_lock);	\
-} while (0)
-
-#define task_running(rq, p) \
-	((rq)->curr == (p) || spin_is_locked(&(p)->switch_lock))
-
 	/* See what happens when you design the chip correctly?
 	 *
 	 * We tell gcc we clobber all non-fixed-usage registers except

diff -puN include/asm-arm/system.h~task-running-flag include/asm-arm/system.h
@@ -137,34 +137,12 @@ extern unsigned int user_debug;
 #define set_wmb(var, value) do { var = value; wmb(); } while (0)
 #define nop() __asm__ __volatile__("mov\tr0,r0\t@ nop\n\t");
 
-#ifdef CONFIG_SMP
 /*
- * Define our own context switch locking.  This allows us to enable
- * interrupts over the context switch, otherwise we end up with high
- * interrupt latency.  The real problem area is switch_mm() which may
- * do a full cache flush.
+ * switch_mm() may do a full cache flush over the context switch,
+ * so enable interrupts over the context switch to avoid high
+ * latency.
  */
-#define prepare_arch_switch(rq,next)					\
-do {									\
-	spin_lock(&(next)->switch_lock);				\
-	spin_unlock_irq(&(rq)->lock);					\
-} while (0)
-
-#define finish_arch_switch(rq,prev)					\
-	spin_unlock(&(prev)->switch_lock)
-
-#define task_running(rq,p)						\
-	((rq)->curr == (p) || spin_is_locked(&(p)->switch_lock))
-#else
-/*
- * Our UP-case is more simple, but we assume knowledge of how
- * spin_unlock_irq() and friends are implemented.  This avoids
- * us needlessly decrementing and incrementing the preempt count.
- */
-#define prepare_arch_switch(rq,next)	local_irq_enable()
-#define finish_arch_switch(rq,prev)	spin_unlock(&(rq)->lock)
-#define task_running(rq,p)		((rq)->curr == (p))
-#endif
+#define __ARCH_WANT_INTERRUPTS_ON_CTXSW
 
 /*
  * switch_to(prev, next) should switch from task `prev' to `next'