From: Andrew Morton <akpm@osdl.org>

Arjan said cpu_relax() always implies a barrier.

Cc: Andreas Mohr <andi@lisas.de>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
---

 arch/i386/kernel/smpboot.c |   24 ++++++------------------
 1 file changed, 6 insertions(+), 18 deletions(-)

diff -puN arch/i386/kernel/smpboot.c~cpu_relax-smpbootc-fix arch/i386/kernel/smpboot.c
--- a/arch/i386/kernel/smpboot.c~cpu_relax-smpbootc-fix
+++ a/arch/i386/kernel/smpboot.c
@@ -251,10 +251,8 @@ static void __init synchronize_tsc_bp (v
 		/*
 		 * all APs synchronize but they loop on '== num_cpus'
 		 */
-		while (atomic_read(&tsc_count_start) != num_booting_cpus()-1) {
+		while (atomic_read(&tsc_count_start) != num_booting_cpus()-1)
 			cpu_relax();
-			mb();
-		}
 		atomic_set(&tsc_count_stop, 0);
 		wmb();
 		/*
@@ -272,10 +270,8 @@ static void __init synchronize_tsc_bp (v
 		/*
 		 * Wait for all APs to leave the synchronization point:
 		 */
-		while (atomic_read(&tsc_count_stop) != num_booting_cpus()-1) {
+		while (atomic_read(&tsc_count_stop) != num_booting_cpus()-1)
 			cpu_relax();
-			mb();
-		}
 		atomic_set(&tsc_count_start, 0);
 		wmb();
 		atomic_inc(&tsc_count_stop);
@@ -332,27 +328,21 @@ static void __init synchronize_tsc_ap (v
 	 * this gets called, so we first wait for the BP to
 	 * finish SMP initialization:
 	 */
-	while (!atomic_read(&tsc_start_flag)) {
+	while (!atomic_read(&tsc_start_flag))
 		cpu_relax();
-		mb();
-	}
 
 	for (i = 0; i < NR_LOOPS; i++) {
 		atomic_inc(&tsc_count_start);
-		while (atomic_read(&tsc_count_start) != num_booting_cpus()) {
+		while (atomic_read(&tsc_count_start) != num_booting_cpus())
 			cpu_relax();
-			mb();
-		}
 
 		rdtscll(tsc_values[smp_processor_id()]);
 		if (i == NR_LOOPS-1)
 			write_tsc(0, 0);
 
 		atomic_inc(&tsc_count_stop);
-		while (atomic_read(&tsc_count_stop) != num_booting_cpus()) {
+		while (atomic_read(&tsc_count_stop) != num_booting_cpus())
 			cpu_relax();
-			mb();
-		}
 	}
 }
 #undef NR_LOOPS
@@ -1453,10 +1443,8 @@ int __devinit __cpu_up(unsigned int cpu)
 	per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
 	/* Unleash the CPU! */
 	cpu_set(cpu, smp_commenced_mask);
-	while (!cpu_isset(cpu, cpu_online_map)) {
+	while (!cpu_isset(cpu, cpu_online_map))
 		cpu_relax();
-		mb();
-	}
 	return 0;
 }
 
_