diff -urN 2.3.17/arch/i386/kernel/irq.c shm/arch/i386/kernel/irq.c
--- 2.3.17/arch/i386/kernel/irq.c	Wed Sep  8 00:25:40 1999
+++ shm/arch/i386/kernel/irq.c	Thu Sep  9 20:31:43 1999
@@ -151,10 +151,7 @@
 static inline void check_smp_invalidate(int cpu)
 {
 	if (test_bit(cpu, &smp_invalidate_needed)) {
-		struct mm_struct *mm = current->mm;
 		clear_bit(cpu, &smp_invalidate_needed);
-		if (mm)
-			atomic_set_mask(1 << cpu, &mm->cpu_vm_mask);
 		local_flush_tlb();
 	}
 }
diff -urN 2.3.17/arch/i386/kernel/smp.c shm/arch/i386/kernel/smp.c
--- 2.3.17/arch/i386/kernel/smp.c	Wed Sep  8 18:18:45 1999
+++ shm/arch/i386/kernel/smp.c	Fri Sep 10 02:19:33 1999
@@ -1635,10 +1635,7 @@
 			 * Take care of "crossing" invalidates
 			 */
 			if (test_bit(cpu, &smp_invalidate_needed)) {
-				struct mm_struct *mm = current->mm;
 				clear_bit(cpu, &smp_invalidate_needed);
-				if (mm)
-					atomic_set_mask(1 << cpu, &mm->cpu_vm_mask);
 				local_flush_tlb();
 			}
 			--stuck;
@@ -1660,39 +1657,33 @@
  */	
 void flush_tlb_current_task(void)
 {
-	unsigned long vm_mask = 1 << current->processor;
-	struct mm_struct *mm = current->mm;
-	unsigned long cpu_mask = mm->cpu_vm_mask & ~vm_mask;
+	unsigned long cpu_mask;
 
-	mm->cpu_vm_mask = vm_mask;
+	rmb();
+	cpu_mask = current->mm->cpu_vm_mask & ~(1 << smp_processor_id());
 	flush_tlb_others(cpu_mask);
 	local_flush_tlb();
 }
 
 void flush_tlb_mm(struct mm_struct * mm)
 {
-	unsigned long vm_mask = 1 << current->processor;
-	unsigned long cpu_mask = mm->cpu_vm_mask & ~vm_mask;
+	unsigned long cpu_mask;
 
-	mm->cpu_vm_mask = 0;
-	if (current->active_mm == mm) {
-		mm->cpu_vm_mask = vm_mask;
+	rmb();
+	cpu_mask = mm->cpu_vm_mask & ~(1 << smp_processor_id());
+	if (current->active_mm == mm)
 		local_flush_tlb();
-	}
 	flush_tlb_others(cpu_mask);
 }
 
 void flush_tlb_page(struct vm_area_struct * vma, unsigned long va)
 {
-	unsigned long vm_mask = 1 << current->processor;
-	struct mm_struct *mm = vma->vm_mm;
-	unsigned long cpu_mask = mm->cpu_vm_mask & ~vm_mask;
+	unsigned long cpu_mask;
 
-	mm->cpu_vm_mask = 0;
-	if (current->active_mm == mm) {
+	rmb();
+	cpu_mask = vma->vm_mm->cpu_vm_mask & ~(1 << smp_processor_id());
+	if (current->active_mm == vma->vm_mm)
 		__flush_tlb_one(va);
-		mm->cpu_vm_mask = vm_mask;
-	}
 	flush_tlb_others(cpu_mask);
 }
 
@@ -1934,12 +1925,8 @@
 	struct task_struct *tsk = current;
 	unsigned int cpu = tsk->processor;
 
-	if (test_and_clear_bit(cpu, &smp_invalidate_needed)) {
-		struct mm_struct *mm = tsk->mm;
-		if (mm)
-			atomic_set_mask(1 << cpu, &mm->cpu_vm_mask);
+	if (test_and_clear_bit(cpu, &smp_invalidate_needed))
 		local_flush_tlb();
-	}
 	ack_APIC_irq();
 
 }
diff -urN 2.3.17/include/asm-i386/mmu_context.h shm/include/asm-i386/mmu_context.h
--- 2.3.17/include/asm-i386/mmu_context.h	Thu Sep  9 20:45:45 1999
+++ shm/include/asm-i386/mmu_context.h	Fri Sep 10 02:03:19 1999
@@ -12,7 +12,6 @@
 
 static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next, struct task_struct *tsk, unsigned cpu)
 {
-
 	if (prev != next) {
 		/*
 		 * Re-load LDT if necessary
@@ -20,11 +19,13 @@
 		if (prev->segments != next->segments)
 			load_LDT(next);
 
+		clear_bit(cpu, &prev->cpu_vm_mask);
+		set_bit(cpu, &next->cpu_vm_mask);
+		wmb();
+
 		/* Re-load page tables */
 		asm volatile("movl %0,%%cr3": :"r" (__pa(next->pgd)));
-		clear_bit(cpu, &prev->cpu_vm_mask);
 	}
-	set_bit(cpu, &next->cpu_vm_mask);
 }
 
 #define activate_mm(prev, next) \