diff -urN 2.4.5pre2/include/asm-alpha/hardirq.h ksoftirqd/include/asm-alpha/hardirq.h
--- 2.4.5pre2/include/asm-alpha/hardirq.h	Sun Apr  1 20:11:14 2001
+++ ksoftirqd/include/asm-alpha/hardirq.h	Tue May 15 22:29:17 2001
@@ -11,6 +11,7 @@
 	unsigned int __local_irq_count;
 	unsigned int __local_bh_count;
 	unsigned int __syscall_count;
+	struct task_struct * __ksoftirqd_task;
 } ____cacheline_aligned irq_cpustat_t;
 
 #include <linux/irq_cpustat.h>	/* Standard mappings for irq_cpustat_t above */
diff -urN 2.4.5pre2/include/asm-i386/hardirq.h ksoftirqd/include/asm-i386/hardirq.h
--- 2.4.5pre2/include/asm-i386/hardirq.h	Fri May 11 04:56:08 2001
+++ ksoftirqd/include/asm-i386/hardirq.h	Tue May 15 22:29:17 2001
@@ -12,6 +12,7 @@
 	unsigned int __local_irq_count;
 	unsigned int __local_bh_count;
 	unsigned int __syscall_count;
+	struct task_struct * __ksoftirqd_task; /* waitqueue is too large */
 	unsigned int __nmi_count;	/* arch dependent */
 } ____cacheline_aligned irq_cpustat_t;
 
diff -urN 2.4.5pre2/include/asm-sparc/hardirq.h ksoftirqd/include/asm-sparc/hardirq.h
--- 2.4.5pre2/include/asm-sparc/hardirq.h	Tue Aug 29 06:20:03 2000
+++ ksoftirqd/include/asm-sparc/hardirq.h	Tue May 15 22:29:29 2001
@@ -23,6 +23,7 @@
 #endif
 	unsigned int __local_bh_count;
 	unsigned int __syscall_count;
+        struct task_struct * __ksoftirqd_task;
 } ____cacheline_aligned irq_cpustat_t;
 
 #include <linux/irq_cpustat.h>	/* Standard mappings for irq_cpustat_t above */
diff -urN 2.4.5pre2/include/asm-sparc64/hardirq.h ksoftirqd/include/asm-sparc64/hardirq.h
--- 2.4.5pre2/include/asm-sparc64/hardirq.h	Wed Aug 23 18:30:13 2000
+++ ksoftirqd/include/asm-sparc64/hardirq.h	Tue May 15 22:29:29 2001
@@ -22,6 +22,7 @@
 #endif
 	unsigned int __local_bh_count;
 	unsigned int __syscall_count;
+        struct task_struct * __ksoftirqd_task;
 } ____cacheline_aligned irq_cpustat_t;
 
 #include <linux/irq_cpustat.h>	/* Standard mappings for irq_cpustat_t above */
diff -urN 2.4.5pre2/include/linux/irq_cpustat.h ksoftirqd/include/linux/irq_cpustat.h
--- 2.4.5pre2/include/linux/irq_cpustat.h	Fri May 11 04:56:08 2001
+++ ksoftirqd/include/linux/irq_cpustat.h	Tue May 15 22:29:17 2001
@@ -31,6 +31,7 @@
 #define local_irq_count(cpu)	__IRQ_STAT((cpu), __local_irq_count)
 #define local_bh_count(cpu)	__IRQ_STAT((cpu), __local_bh_count)
 #define syscall_count(cpu)	__IRQ_STAT((cpu), __syscall_count)
+#define ksoftirqd_task(cpu)	__IRQ_STAT((cpu), __ksoftirqd_task)
   /* arch dependent irq_stat fields */
 #define nmi_count(cpu)		__IRQ_STAT((cpu), __nmi_count)		/* i386, ia64 */
 
diff -urN 2.4.5pre2/kernel/softirq.c ksoftirqd/kernel/softirq.c
--- 2.4.5pre2/kernel/softirq.c	Tue Jan  2 17:41:22 2001
+++ ksoftirqd/kernel/softirq.c	Tue May 15 22:29:17 2001
@@ -86,6 +86,17 @@
 		if ((active &= mask) != 0)
 			goto retry;
 	}
+	if (softirq_active(cpu) & softirq_mask(cpu)) {
+		/*
+		 * we cannot loop indefinitely here to avoid userspace starvation,
+		 * but we also don't want to introduce a worst case 1/HZ latency
+		 * to the pending events, so lets the scheduler to balance
+		 * the irq load for us.
+		 */
+		struct task_struct * tsk = ksoftirqd_task(cpu);
+		if (tsk && tsk->state != TASK_RUNNING)
+			wake_up_process(tsk);
+	}
 
 	local_bh_enable();
 
@@ -315,3 +326,61 @@
 			f(data);
 	}
 }
+
+static int ksoftirqd(void * __bind_cpu)
+{
+	int bind_cpu = *(int *) __bind_cpu;
+	int cpu = cpu_logical_map(bind_cpu);
+
+	daemonize();
+	current->nice = 19;
+	sigfillset(&current->blocked);
+
+	/* Migrate to the right CPU */
+	current->cpus_allowed = 1UL << cpu;
+	while (smp_processor_id() != cpu)
+		schedule();
+
+	sprintf(current->comm, "ksoftirqd_CPU%d", bind_cpu);
+
+	__set_current_state(TASK_INTERRUPTIBLE);
+	mb();
+
+	ksoftirqd_task(cpu) = current;
+
+	for (;;) {
+		if (!(softirq_active(cpu) & softirq_mask(cpu)))
+			schedule();
+
+		__set_current_state(TASK_RUNNING);
+
+		while (softirq_active(cpu) & softirq_mask(cpu)) {
+			do_softirq();
+			if (current->need_resched)
+				schedule();
+		}
+
+		__set_current_state(TASK_INTERRUPTIBLE);
+	}
+}
+
+static __init int spawn_ksoftirqd(void)
+{
+	int cpu;
+
+	for (cpu = 0; cpu < smp_num_cpus; cpu++) {
+		if (kernel_thread(ksoftirqd, (void *) &cpu,
+				  CLONE_FS | CLONE_FILES | CLONE_SIGNAL) < 0)
+			printk("spawn_ksoftirqd() failed for cpu %d\n", cpu);
+		else {
+			while (!ksoftirqd_task(cpu_logical_map(cpu))) {
+				current->policy |= SCHED_YIELD;
+				schedule();
+			}
+		}
+	}
+
+	return 0;
+}
+
+__initcall(spawn_ksoftirqd);