4 files changed, 75 insertions, 36 deletions
diff --git a/arch/x86/lib/copy_user_64.S b/arch/x86/lib/copy_user_64.S
index 70bebd31040..ee1c3f63515 100644
--- a/arch/x86/lib/copy_user_64.S
+++ b/arch/x86/lib/copy_user_64.S
@@ -217,19 +217,19 @@ ENTRY(copy_user_generic_unrolled)
 	/* table sorted by exception address */
 	.section __ex_table,"a"
 	.align 8
-	.quad .Ls1,.Ls1e
-	.quad .Ls2,.Ls2e
-	.quad .Ls3,.Ls3e
-	.quad .Ls4,.Ls4e
-	.quad .Ld1,.Ls1e
+	.quad .Ls1,.Ls1e	/* Ls1-Ls4 have copied zero bytes */
+	.quad .Ls2,.Ls1e
+	.quad .Ls3,.Ls1e
+	.quad .Ls4,.Ls1e
+	.quad .Ld1,.Ls1e	/* Ld1-Ld4 have copied 0-24 bytes */
 	.quad .Ld2,.Ls2e
 	.quad .Ld3,.Ls3e
 	.quad .Ld4,.Ls4e
-	.quad .Ls5,.Ls5e
-	.quad .Ls6,.Ls6e
-	.quad .Ls7,.Ls7e
-	.quad .Ls8,.Ls8e
-	.quad .Ld5,.Ls5e
+	.quad .Ls5,.Ls5e	/* Ls5-Ls8 have copied 32 bytes */
+	.quad .Ls6,.Ls5e
+	.quad .Ls7,.Ls5e
+	.quad .Ls8,.Ls5e
+	.quad .Ld5,.Ls5e	/* Ld5-Ld8 have copied 32-56 bytes */
 	.quad .Ld6,.Ls6e
 	.quad .Ld7,.Ls7e
 	.quad .Ld8,.Ls8e
@@ -244,11 +244,8 @@ ENTRY(copy_user_generic_unrolled)
 	.quad .Le5,.Le_zero
 	.previous
 
-	/* compute 64-offset for main loop. 8 bytes accuracy with error on the
-	   pessimistic side. this is gross. it would be better to fix the
-	   interface. */
 	/* eax: zero, ebx: 64 */
-.Ls1e: 	addl $8,%eax
+.Ls1e: 	addl $8,%eax		/* eax is bytes left uncopied within the loop (Ls1e: 64 .. Ls8e: 8) */
 .Ls2e: 	addl $8,%eax
 .Ls3e: 	addl $8,%eax
 .Ls4e: 	addl $8,%eax
diff --git a/arch/x86/lib/copy_user_nocache_64.S b/arch/x86/lib/copy_user_nocache_64.S
index 5196762b3b0..9d3d1ab8376 100644
--- a/arch/x86/lib/copy_user_nocache_64.S
+++ b/arch/x86/lib/copy_user_nocache_64.S
@@ -145,19 +145,19 @@ ENTRY(__copy_user_nocache)
 	/* table sorted by exception address */
 	.section __ex_table,"a"
 	.align 8
-	.quad .Ls1,.Ls1e
-	.quad .Ls2,.Ls2e
-	.quad .Ls3,.Ls3e
-	.quad .Ls4,.Ls4e
-	.quad .Ld1,.Ls1e
+	.quad .Ls1,.Ls1e	/* .Ls[1-4] - 0 bytes copied */
+	.quad .Ls2,.Ls1e
+	.quad .Ls3,.Ls1e
+	.quad .Ls4,.Ls1e
+	.quad .Ld1,.Ls1e	/* .Ld[1-4] - 0..24 bytes coped */
 	.quad .Ld2,.Ls2e
 	.quad .Ld3,.Ls3e
 	.quad .Ld4,.Ls4e
-	.quad .Ls5,.Ls5e
-	.quad .Ls6,.Ls6e
-	.quad .Ls7,.Ls7e
-	.quad .Ls8,.Ls8e
-	.quad .Ld5,.Ls5e
+	.quad .Ls5,.Ls5e	/* .Ls[5-8] - 32 bytes copied */
+	.quad .Ls6,.Ls5e
+	.quad .Ls7,.Ls5e
+	.quad .Ls8,.Ls5e
+	.quad .Ld5,.Ls5e	/* .Ld[5-8] - 32..56 bytes copied */
 	.quad .Ld6,.Ls6e
 	.quad .Ld7,.Ls7e
 	.quad .Ld8,.Ls8e
@@ -172,11 +172,8 @@ ENTRY(__copy_user_nocache)
 	.quad .Le5,.Le_zero
 	.previous
 
-	/* compute 64-offset for main loop. 8 bytes accuracy with error on the
-	   pessimistic side. this is gross. it would be better to fix the
-	   interface. */
 	/* eax: zero, ebx: 64 */
-.Ls1e: 	addl $8,%eax
+.Ls1e: 	addl $8,%eax	/* eax: bytes left uncopied: Ls1e: 64 .. Ls8e: 8 */
 .Ls2e: 	addl $8,%eax
 .Ls3e: 	addl $8,%eax
 .Ls4e: 	addl $8,%eax
diff --git a/arch/x86/lib/delay_32.c b/arch/x86/lib/delay_32.c
index 4535e6d147a..d710f2d167b 100644
--- a/arch/x86/lib/delay_32.c
+++ b/arch/x86/lib/delay_32.c
@@ -44,13 +44,36 @@ static void delay_loop(unsigned long loops)
 static void delay_tsc(unsigned long loops)
 {
 	unsigned long bclock, now;
+	int cpu;
 
-	preempt_disable();		/* TSC's are per-cpu */
+	preempt_disable();
+	cpu = smp_processor_id();
 	rdtscl(bclock);
-	do {
-		rep_nop();
+	for (;;) {
 		rdtscl(now);
-	} while ((now-bclock) < loops);
+		if ((now - bclock) >= loops)
+			break;
+
+		/* Allow RT tasks to run */
+		preempt_enable();
+		rep_nop();
+		preempt_disable();
+
+		/*
+		 * It is possible that we moved to another CPU, and
+		 * since TSC's are per-cpu we need to calculate
+		 * that. The delay must guarantee that we wait "at
+		 * least" the amount of time. Being moved to another
+		 * CPU could make the wait longer but we just need to
+		 * make sure we waited long enough. Rebalance the
+		 * counter for this CPU.
+		 */
+		if (unlikely(cpu != smp_processor_id())) {
+			loops -= (now - bclock);
+			cpu = smp_processor_id();
+			rdtscl(bclock);
+		}
+	}
 	preempt_enable();
 }
 
diff --git a/arch/x86/lib/delay_64.c b/arch/x86/lib/delay_64.c
index bbc61051851..4c441be9264 100644
--- a/arch/x86/lib/delay_64.c
+++ b/arch/x86/lib/delay_64.c
@@ -31,14 +31,36 @@ int __devinit read_current_timer(unsigned long *timer_value)
 void __delay(unsigned long loops)
 {
 	unsigned bclock, now;
+	int cpu;
 
-	preempt_disable();		/* TSC's are pre-cpu */
+	preempt_disable();
+	cpu = smp_processor_id();
 	rdtscl(bclock);
-	do {
-		rep_nop(); 
+	for (;;) {
 		rdtscl(now);
+		if ((now - bclock) >= loops)
+			break;
+
+		/* Allow RT tasks to run */
+		preempt_enable();
+		rep_nop();
+		preempt_disable();
+
+		/*
+		 * It is possible that we moved to another CPU, and
+		 * since TSC's are per-cpu we need to calculate
+		 * that. The delay must guarantee that we wait "at
+		 * least" the amount of time. Being moved to another
+		 * CPU could make the wait longer but we just need to
+		 * make sure we waited long enough. Rebalance the
+		 * counter for this CPU.
+		 */
+		if (unlikely(cpu != smp_processor_id())) {
+			loops -= (now - bclock);
+			cpu = smp_processor_id();
+			rdtscl(bclock);
+		}
 	}
-	while ((now-bclock) < loops);
 	preempt_enable();
 }
 EXPORT_SYMBOL(__delay);