bitkeeper revision 1.596.1.3 (3fb3b41eWUoRU0H8A0jEX5roXjxKkA)

Many files:
  Greatly simplified Xen softirqs. They are now only executed in outermost Xen activation; they are never called within an irq context.

14 files changed, 54 insertions, 318 deletions

diff --git a/xen/arch/i386/apic.c b/xen/arch/i386/apic.c
index 0f4b9a3538..8a3a6b5cf8 100644
--- a/xen/arch/i386/apic.c
+++ b/xen/arch/i386/apic.c
@@ -711,58 +711,18 @@ int reprogram_ac_timer(s_time_t timeout)
     return 1;
 }
 
-/*
- * Local timer interrupt handler.
- * Here the programmable, accurate timers are executed.
- */
-inline void smp_local_timer_interrupt(struct pt_regs *regs)
-{
-    do_ac_timer();
-}
-
-/*
- * Local APIC timer interrupt. This is the most natural way for doing
- * local interrupts, but local timer interrupts can be emulated by
- * broadcast interrupts too. [in case the hw doesnt support APIC timers]
- *
- * [ if a single-CPU system runs an SMP kernel then we call the local
- *   interrupt as well. Thus we cannot inline the local irq ... ]
- */
 unsigned int apic_timer_irqs [NR_CPUS];
 
 void smp_apic_timer_interrupt(struct pt_regs * regs)
 {
     int cpu = smp_processor_id();
-#ifndef NDEBUG
-    u32 cc_start, cc_end;
-    rdtscl(cc_start);
-#endif
 
-    /*
-     * the NMI deadlock-detector uses this.
-     */
-    apic_timer_irqs[cpu]++;
-
-    /*
-     * NOTE! We'd better ACK the irq immediately, because timer handling can 
-     * be slow. XXX is this save?
-     */
     ack_APIC_irq();
 
-    /* call the local handler */
-    irq_enter(cpu, 0);
+    apic_timer_irqs[cpu]++;
     perfc_incrc(apic_timer);
-    smp_local_timer_interrupt(regs);
-    irq_exit(cpu, 0);
 
-    if (softirq_pending(cpu))
-        do_softirq();
-
-#ifndef NDEBUG
-    rdtscl(cc_end);
-    if ( (cc_end - cc_start) > (cpu_khz * 100) )
-        printk("APIC Long ISR on CPU=%02d %08X -> %08X\n",cpu,cc_start,cc_end);
-#endif
+    __cpu_raise_softirq(cpu, AC_TIMER_SOFTIRQ);
 }
 
 /*
diff --git a/xen/arch/i386/irq.c b/xen/arch/i386/irq.c
index 504c7261d4..2793eba3d7 100644
--- a/xen/arch/i386/irq.c
+++ b/xen/arch/i386/irq.c
@@ -535,9 +535,6 @@ asmlinkage unsigned int do_IRQ(struct pt_regs regs)
     desc->handler->end(irq);
     spin_unlock(&desc->lock);
 
-    if (softirq_pending(cpu))
-        do_softirq();
-
     rdtscl(cc_end);
 
     if ( !action || (!(action->flags & SA_NOPROFILE)) )
diff --git a/xen/arch/i386/smp.c b/xen/arch/i386/smp.c
index baf88b5fba..b1dfe64d4f 100644
--- a/xen/arch/i386/smp.c
+++ b/xen/arch/i386/smp.c
@@ -261,17 +261,11 @@ static volatile unsigned long flush_cpumask;
 static spinlock_t tlbstate_lock = SPIN_LOCK_UNLOCKED;
 #define FLUSH_ALL	0xffffffff
 
-asmlinkage void smp_invalidate_interrupt (void)
+asmlinkage void smp_invalidate_interrupt(void)
 {
-    unsigned long cpu = smp_processor_id();
-
-    if (!test_bit(cpu, &flush_cpumask))
-        return;
-
-    local_flush_tlb();
-
     ack_APIC_irq();
-    clear_bit(cpu, &flush_cpumask);
+    if (test_and_clear_bit(smp_processor_id(), &flush_cpumask))
+        local_flush_tlb();
 }
 
 void flush_tlb_others(unsigned long cpumask)
diff --git a/xen/arch/i386/time.c b/xen/arch/i386/time.c
index ddacbca63d..1328f31fd4 100644
--- a/xen/arch/i386/time.c
+++ b/xen/arch/i386/time.c
@@ -96,7 +96,7 @@ static void timer_interrupt(int irq, void *dev_id, struct pt_regs *regs)
 
     /* Rough hack to allow accurate timers to sort-of-work with no APIC. */
     if ( do_timer_lists_from_pit )
-        do_ac_timer();
+        __cpu_raise_softirq(smp_processor_id(), AC_TIMER_SOFTIRQ);
 }
 
 static struct irqaction irq0  = { timer_interrupt, SA_INTERRUPT, 0,
diff --git a/xen/common/ac_timer.c b/xen/common/ac_timer.c
index f4752ca3fb..e456e83624 100644
--- a/xen/common/ac_timer.c
+++ b/xen/common/ac_timer.c
@@ -241,88 +241,35 @@ void mod_ac_timer(struct ac_timer *timer, s_time_t new_time)
 }
 
 
-void do_ac_timer(void)
+static void ac_timer_softirq_action(struct softirq_action *a)
 {
     int              cpu = smp_processor_id();
-    unsigned long    flags;
     struct ac_timer *t, **heap;
-    s_time_t         diff, now = NOW();
-    long             max;
-
-    spin_lock_irqsave(&ac_timers[cpu].lock, flags);
-
- do_timer_again:
-    TRC(printk("ACT  [%02d] do(): now=%lld\n", cpu, NOW()));
-
-    heap = ac_timers[cpu].heap;
-
-    while ( (GET_HEAP_SIZE(heap) != 0) &&
-            ((t = heap[1])->expires < (NOW() + TIMER_SLOP)) )
-    {
-        remove_entry(heap, t);
-
-        /* Do some stats collection. */
-        diff = (now - t->expires);
-        if ( diff > 0x7fffffff ) 
-            diff =  0x7fffffff; /* THIS IS BAD! */
-        max = perfc_valuea(ac_timer_max, cpu);
-        if ( diff > max ) 
-            perfc_seta(ac_timer_max, cpu, diff);
-
-        spin_unlock_irqrestore(&ac_timers[cpu].lock, flags);
-        if ( t->function != NULL ) 
-            t->function(t->data);
-        spin_lock_irqsave(&ac_timers[cpu].lock, flags);
-
-        /* Heap may have grown while the lock was released. */
-        heap = ac_timers[cpu].heap;
-    }
-
-    if ( GET_HEAP_SIZE(heap) != 0 )
-    {
-        if ( !reprogram_ac_timer(heap[1]->expires) )
-            goto do_timer_again;
-    } 
-    else 
-    {
-        reprogram_ac_timer(0);
-    }
-
-    spin_unlock_irqrestore(&ac_timers[cpu].lock, flags);
-    TRC(printk("ACT  [%02d] do(): end\n", cpu));
-}
-
-
-static void ac_timer_softirq_action(struct softirq_action *a)
-{
-    int               cpu = smp_processor_id();
-    unsigned long     flags;
-    struct ac_timer  *t;
-    int               process_timer_list = 0;
+    s_time_t         now;
 
-    spin_lock_irqsave(&ac_timers[cpu].lock, flags);
+    spin_lock_irq(&ac_timers[cpu].lock);
     
-    if ( GET_HEAP_SIZE(ac_timers[cpu].heap) != 0 )
-    {
-        /*
-         * Reprogram timer with earliest deadline. If that has already passed
-         * then we will process the timer list as soon as we release the lock.
-         */
-        t = ac_timers[cpu].heap[1];
-        if ( (t->expires < (NOW() + TIMER_SLOP)) ||
-             !reprogram_ac_timer(t->expires) )
-            process_timer_list = 1;
-    }
-    else
-    {
-        /* No deadline to program the timer with.*/
-        reprogram_ac_timer((s_time_t)0);
+    do {
+        heap = ac_timers[cpu].heap;
+        now  = NOW();
+        
+        while ( (GET_HEAP_SIZE(heap) != 0) &&
+                ((t = heap[1])->expires < (now + TIMER_SLOP)) )
+        {
+            remove_entry(heap, t);
+                        
+            spin_unlock_irq(&ac_timers[cpu].lock);
+            if ( t->function != NULL ) 
+                t->function(t->data);
+            spin_lock_irq(&ac_timers[cpu].lock);
+            
+            /* Heap may have grown while the lock was released. */
+            heap = ac_timers[cpu].heap;
+        }
     }
+    while ( !reprogram_ac_timer(GET_HEAP_SIZE(heap) ? heap[1]->expires : 0) );
 
-    spin_unlock_irqrestore(&ac_timers[cpu].lock, flags);
-
-    if ( process_timer_list )
-        do_ac_timer();
+    spin_unlock_irq(&ac_timers[cpu].lock);
 }
 
 
diff --git a/xen/common/softirq.c b/xen/common/softirq.c
index 951fb91395..f209c6668d 100644
--- a/xen/common/softirq.c
+++ b/xen/common/softirq.c
@@ -12,9 +12,7 @@
 #include <linux/config.h>
 #include <linux/mm.h>
 #include <linux/sched.h>
-//#include <linux/kernel_stat.h>
 #include <linux/interrupt.h>
-//#include <linux/smp_lock.h>
 #include <linux/init.h>
 #include <linux/tqueue.h>
 
@@ -51,48 +49,30 @@ asmlinkage void do_softirq()
     int cpu = smp_processor_id();
     struct softirq_action *h;
     __u32 pending;
-    long flags;
 
-    if (in_interrupt())
-        return;
-
-    local_irq_save(flags);
-
-    pending = xchg(&softirq_pending(cpu), 0);
-    if ( !pending ) goto out;
+    if ( in_interrupt() )
+        BUG();
 
     local_bh_disable();
 
-    do {
-        local_irq_enable();
-        
+    while ( (pending = xchg(&softirq_pending(cpu), 0)) != 0 )
+    {
         h = softirq_vec;
-        
-        do {
+        while ( pending )
+        {
             if (pending & 1)
                 h->action(h);
             h++;
             pending >>= 1;
-        } while (pending);
-        
-        local_irq_disable();
-        
-        pending = xchg(&softirq_pending(cpu), 0);
-    } while ( pending );
+        }
+    }
 
     __local_bh_enable();
-
-out:
-    local_irq_restore(flags);
 }
 
-/*
- * This function must run with irq disabled!
- */
 inline void cpu_raise_softirq(unsigned int cpu, unsigned int nr)
 {
     __cpu_raise_softirq(cpu, nr);
-
 #ifdef CONFIG_SMP
     if ( cpu != smp_processor_id() )
         smp_send_event_check_cpu(cpu);
@@ -101,11 +81,7 @@ inline void cpu_raise_softirq(unsigned int cpu, unsigned int nr)
 
 void raise_softirq(unsigned int nr)
 {
-    long flags;
-
-    local_irq_save(flags);
     cpu_raise_softirq(smp_processor_id(), nr);
-    local_irq_restore(flags);
 }
 
 void open_softirq(int nr, void (*action)(struct softirq_action*), void *data)
@@ -224,7 +200,7 @@ void tasklet_init(struct tasklet_struct *t,
 void tasklet_kill(struct tasklet_struct *t)
 {
     if (in_interrupt())
-        printk("Attempt to kill tasklet from interrupt\n");
+        BUG();
     while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state))
         while (test_bit(TASKLET_STATE_SCHED, &t->state))
             do_softirq();
diff --git a/xen/drivers/scsi/aacraid/aacraid.h b/xen/drivers/scsi/aacraid/aacraid.h
index eb38cd1361..ecf56413cd 100644
--- a/xen/drivers/scsi/aacraid/aacraid.h
+++ b/xen/drivers/scsi/aacraid/aacraid.h
@@ -3,11 +3,8 @@
 # define dprintk(x)
 #endif
 
-/* Start of Xen additions XXX */
 #include <asm/byteorder.h>
 #include <xeno/interrupt.h>
-#define TRY_TASKLET
-/* End of Xen additions XXX */
 
 /*------------------------------------------------------------------------------
  *              D E F I N E S
@@ -1416,12 +1413,7 @@ int aac_rx_init(struct aac_dev *dev, unsigned long devNumber);
 int aac_sa_init(struct aac_dev *dev, unsigned long devNumber);
 unsigned int aac_response_normal(struct aac_queue * q);
 unsigned int aac_command_normal(struct aac_queue * q);
-#ifdef TRY_TASKLET
-extern struct tasklet_struct aac_command_tasklet;
-void aac_command_thread(unsigned long data);
-#else
-int aac_command_thread(struct aac_dev * dev);
-#endif
+void aac_command_thread(struct aac_dev * dev);
 int aac_close_fib_context(struct aac_dev * dev, struct aac_fib_context *fibctx);
 int fib_adapter_complete(struct fib * fibptr, unsigned short size);
 struct aac_driver_ident* aac_get_driver_ident(int devtype);
diff --git a/xen/drivers/scsi/aacraid/commsup.c b/xen/drivers/scsi/aacraid/commsup.c
index 84c77464f5..4450dcbe8b 100644
--- a/xen/drivers/scsi/aacraid/commsup.c
+++ b/xen/drivers/scsi/aacraid/commsup.c
@@ -523,24 +523,8 @@ int fib_send(u16 command, struct fib * fibptr, unsigned long size,  int priority
     
 	if (wait) {
 		spin_unlock_irqrestore(&fibptr->event_lock, flags);
-#if 0
-		down(&fibptr->event_wait);
-		if(fibptr->done == 0)
-			BUG();
-#endif
-#ifdef TRY_TASKLET
-	while (!fibptr->done) { 
-		tasklet_schedule(&aac_command_tasklet);
-		do_softirq(); /* force execution */
-	}
-#else 
-	while (!fibptr->done) { 
-		mdelay(100); 
-		aac_command_thread(dev); 
-	}
-#endif
-
-			
+		while (!fibptr->done)
+			aac_command_thread(dev); 
 		if((fibptr->flags & FIB_CONTEXT_FLAG_TIMED_OUT))
 			return -ETIMEDOUT;
 		else
@@ -924,62 +908,27 @@ static void aac_handle_aif(struct aac_dev * dev, struct fib * fibptr)
  *	until the queue is empty. When the queue is empty it will wait for
  *	more FIBs.
  */
-#ifndef TRY_TASKLET
-int aac_command_thread(struct aac_dev * dev)
-{
-#else
-DECLARE_TASKLET_DISABLED(aac_command_tasklet, aac_command_thread, 0);
-void aac_command_thread(unsigned long data)
-#define return(_x) return 
+void aac_command_thread(struct aac_dev * dev)
 {
-	struct aac_dev *dev = (struct aac_dev *)data; 
-#endif
 	struct hw_fib *hw_fib, *hw_newfib;
 	struct fib *fib, *newfib;
 	struct aac_queue_block *queues = dev->queues;
 	struct aac_fib_context *fibctx;
 	unsigned long flags;
-#if 0
-	DECLARE_WAITQUEUE(wait, current);
-#endif
+	static spinlock_t lock = SPIN_LOCK_UNLOCKED;
+
+	spin_lock_irqsave(&lock, flags);
 
-	/*
-	 *	We can only have one thread per adapter for AIF's.
-	 */
-	if (dev->aif_thread)
-		return(-EINVAL);
-#if 0
-	/*
-	 *	Set up the name that will appear in 'ps'
-	 *	stored in  task_struct.comm[16].
-	 */
-	sprintf(current->comm, "aacraid");
-	daemonize();
-#endif
-	/*
-	 *	Let the DPC know it has a place to send the AIF's to.
-	 */
-	dev->aif_thread = 1;
-#if 0
-	add_wait_queue(&queues->queue[HostNormCmdQueue].cmdready, &wait);
-	set_current_state(TASK_INTERRUPTIBLE);
-	dprintk ((KERN_INFO "aac_command_thread start\n"));
-	while(1) 
-#endif
 	{
-		spin_lock_irqsave(queues->queue[HostNormCmdQueue].lock, flags);
+		spin_lock(queues->queue[HostNormCmdQueue].lock);
 		while(!list_empty(&(queues->queue[HostNormCmdQueue].cmdq))) {
 			struct list_head *entry;
 			struct aac_aifcmd * aifcmd;
 
-#if 0
-			set_current_state(TASK_RUNNING);
-#endif
-
 			entry = queues->queue[HostNormCmdQueue].cmdq.next;
 			list_del(entry);
 	
-			spin_unlock_irqrestore(queues->queue[HostNormCmdQueue].lock, flags);
+			spin_unlock(queues->queue[HostNormCmdQueue].lock);
 			fib = list_entry(entry, struct fib, fiblink);
 			/*
 			 *	We will process the FIB here or pass it to a 
@@ -1074,13 +1023,6 @@ void aac_command_thread(unsigned long data)
 						 */
 						list_add_tail(&newfib->fiblink, &fibctx->fib_list);
 						fibctx->count++;
-#if 0
-						/* 
-						 * Set the event to wake up the
-						 * thread that will waiting.
-						 */
-						up(&fibctx->wait_sem);
-#endif
 					} else {
 						printk(KERN_WARNING "aifd: didn't allocate NewFib.\n");
 						if(newfib)
@@ -1097,27 +1039,14 @@ void aac_command_thread(unsigned long data)
 				fib_adapter_complete(fib, sizeof(u32));
 				spin_unlock_irqrestore(&dev->fib_lock, flagv);
 			}
-			spin_lock_irqsave(queues->queue[HostNormCmdQueue].lock, flags);
+			spin_lock(queues->queue[HostNormCmdQueue].lock);
 			kfree(fib);
 		}
 		/*
 		 *	There are no more AIF's
 		 */
-		spin_unlock_irqrestore(queues->queue[HostNormCmdQueue].lock, flags);
-#if 0
-		schedule();
-
-		if(signal_pending(current))
-			break;
-		set_current_state(TASK_INTERRUPTIBLE);
-#endif
+		spin_unlock(queues->queue[HostNormCmdQueue].lock);
 	}
-#if 0
-	remove_wait_queue(&queues->queue[HostNormCmdQueue].cmdready, &wait);
-	dev->aif_thread = 0;
-	complete_and_exit(&dev->aif_completion, 0);
-#else
-	mdelay(50); 
-	dev->aif_thread = 0;
-#endif
+
+	spin_unlock_irqrestore(&lock, flags);
 }
diff --git a/xen/drivers/scsi/aacraid/rx.c b/xen/drivers/scsi/aacraid/rx.c
index ebb797041e..0cb17691d1 100644
--- a/xen/drivers/scsi/aacraid/rx.c
+++ b/xen/drivers/scsi/aacraid/rx.c
@@ -406,20 +406,6 @@ int aac_rx_init(struct aac_dev *dev, unsigned long num)
 
 	if (aac_init_adapter(dev) == NULL)
 		return -1;
-#ifdef TRY_TASKLET
-	aac_command_tasklet.data = (unsigned long)dev;
-	tasklet_enable(&aac_command_tasklet);
-#else
-	/*
-	 *	Start any kernel threads needed
-	 */
-	dev->thread_pid = kernel_thread((int (*)(void *))aac_command_thread, dev, 0);
-	if(dev->thread_pid < 0)
-	{
-		printk(KERN_ERR "aacraid: Unable to create rx thread.\n");
-		return -1;
-	}	
-#endif
 	/*
 	 *	Tell the adapter that all is configured, and it can start
 	 *	accepting requests
diff --git a/xen/drivers/scsi/aic7xxx/aic7xxx_osm.c b/xen/drivers/scsi/aic7xxx/aic7xxx_osm.c
index e868f32bbf..4f4ce01c4a 100644
--- a/xen/drivers/scsi/aic7xxx/aic7xxx_osm.c
+++ b/xen/drivers/scsi/aic7xxx/aic7xxx_osm.c
@@ -1903,7 +1903,6 @@ ahc_linux_isr(int irq, void *dev_id, struct pt_regs * regs)
 	if (next_dev) {
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
 		tasklet_schedule(&ahc->platform_data->runq_tasklet);
-		do_softirq();
 #else
 		ahc_runq_tasklet((unsigned long)ahc);
 #endif
@@ -2535,7 +2534,6 @@ ahc_linux_release_sim_queue(u_long arg)
 		scsi_unblock_requests(ahc->platform_data->host);
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
 		tasklet_schedule(&ahc->platform_data->runq_tasklet);
-		do_softirq();
 #else
 		ahc_runq_tasklet((unsigned long)ahc);
 #endif
diff --git a/xen/drivers/scsi/scsi.c b/xen/drivers/scsi/scsi.c
index 3826d4ed1e..ed869dd6aa 100644
--- a/xen/drivers/scsi/scsi.c
+++ b/xen/drivers/scsi/scsi.c
@@ -840,6 +840,7 @@ void scsi_wait_req (Scsi_Request * SRpnt, const void *cmnd ,
        if it hasn't been done already. This is not the correct behaviour 
        in xen ... hmm .. how to fix? */
     while(wait) { 
+        do_softirq(); /* XXX KAF: this is safe, and necessary!! */
         udelay(500); 
         usecs += 500; 
         if(usecs > 1000000) {
diff --git a/xen/include/asm-i386/softirq.h b/xen/include/asm-i386/softirq.h
index 254224411b..4f37cfdb7d 100644
--- a/xen/include/asm-i386/softirq.h
+++ b/xen/include/asm-i386/softirq.h
@@ -9,40 +9,10 @@
 #define cpu_bh_disable(cpu) \
 		do { local_bh_count(cpu)++; barrier(); } while (0)
 
-#define local_bh_disable()	cpu_bh_disable(smp_processor_id())
-#define __local_bh_enable()	__cpu_bh_enable(smp_processor_id())
+#define local_bh_disable()  cpu_bh_disable(smp_processor_id())
+#define __local_bh_enable() __cpu_bh_enable(smp_processor_id())
+#define local_bh_enable()   __local_bh_enable()
 
 #define in_softirq() (local_bh_count(smp_processor_id()) != 0)
 
-/*
- * NOTE: this assembly code assumes:
- *
- *    (char *)&local_bh_count - 8 == (char *)&softirq_pending
- *
- * If you change the offsets in irq_stat then you have to
- * update this code as well.
- */
-#define local_bh_enable()						\
-do {									\
-	unsigned int *ptr = &local_bh_count(smp_processor_id());	\
-									\
-	barrier();							\
-	if (!--*ptr)							\
-		__asm__ __volatile__ (					\
-			"cmpl $0, -8(%0);"				\
-			"jnz 2f;"					\
-			"1:;"						\
-									\
-			".section .text.lock,\"ax\";"			\
-			"2: pushl %%eax; pushl %%ecx; pushl %%edx;"	\
-			"call %c1;"					\
-			"popl %%edx; popl %%ecx; popl %%eax;"		\
-			"jmp 1b;"					\
-			".previous;"					\
-									\
-		: /* no output */					\
-		: "r" (ptr), "i" (do_softirq)				\
-		/* no registers clobbered */ );				\
-} while (0)
-
 #endif	/* __ASM_SOFTIRQ_H */
diff --git a/xen/include/xeno/ac_timer.h b/xen/include/xeno/ac_timer.h
index 78b380d76f..987fbeed33 100644
--- a/xen/include/xeno/ac_timer.h
+++ b/xen/include/xeno/ac_timer.h
@@ -67,6 +67,5 @@ static __inline__ int active_ac_timer(struct ac_timer *timer)
 
 /* interface used by programmable timer, implemented hardware dependent */
 extern int  reprogram_ac_timer(s_time_t timeout);
-extern void do_ac_timer(void);
 
 #endif /* _AC_TIMER_H_ */
diff --git a/xen/include/xeno/netdevice.h b/xen/include/xeno/netdevice.h
index e06faaaf64..de639a884d 100644
--- a/xen/include/xeno/netdevice.h
+++ b/xen/include/xeno/netdevice.h
@@ -466,19 +466,6 @@ extern int		dev_ioctl(unsigned int cmd, void *);
 extern int		dev_change_flags(struct net_device *, unsigned);
 extern void		dev_init(void);
 
-extern int		netdev_nit;
-
-/* Post buffer to the network code from _non interrupt_ context.
- * see net/core/dev.c for netif_rx description.
- */
-static inline int netif_rx_ni(struct sk_buff *skb)
-{
-       int err = netif_rx(skb);
-       if (softirq_pending(smp_processor_id()))
-               do_softirq();
-       return err;
-}
-
 extern int netdev_finish_unregister(struct net_device *dev);
 
 static inline void dev_put(struct net_device *dev)