15 files changed, 383 insertions, 164 deletions

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index e903c71f7e6..0952ecd60ec 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -26,6 +26,7 @@ config X86
 	select HAVE_AOUT if X86_32
 	select HAVE_UNSTABLE_SCHED_CLOCK
 	select ARCH_SUPPORTS_NUMA_BALANCING
+	select ARCH_SUPPORTS_INT128 if X86_64
 	select ARCH_WANTS_PROT_NUMA_PROT_NONE
 	select HAVE_IDE
 	select HAVE_OPROFILE
diff --git a/arch/x86/include/asm/pgtable.h b/arch/x86/include/asm/pgtable.h
index 3d199945870..bbc8b12fa44 100644
--- a/arch/x86/include/asm/pgtable.h
+++ b/arch/x86/include/asm/pgtable.h
@@ -452,9 +452,16 @@ static inline int pte_present(pte_t a)
 }
 
 #define pte_accessible pte_accessible
-static inline int pte_accessible(pte_t a)
+static inline bool pte_accessible(struct mm_struct *mm, pte_t a)
 {
-	return pte_flags(a) & _PAGE_PRESENT;
+	if (pte_flags(a) & _PAGE_PRESENT)
+		return true;
+
+	if ((pte_flags(a) & (_PAGE_PROTNONE | _PAGE_NUMA)) &&
+			mm_tlb_flush_pending(mm))
+		return true;
+
+	return false;
 }
 
 static inline int pte_hidden(pte_t pte)
diff --git a/arch/x86/include/asm/preempt.h b/arch/x86/include/asm/preempt.h
index 8729723636f..c8b051933b1 100644
--- a/arch/x86/include/asm/preempt.h
+++ b/arch/x86/include/asm/preempt.h
@@ -8,6 +8,12 @@
 DECLARE_PER_CPU(int, __preempt_count);
 
 /*
+ * We use the PREEMPT_NEED_RESCHED bit as an inverted NEED_RESCHED such
+ * that a decrement hitting 0 means we can and should reschedule.
+ */
+#define PREEMPT_ENABLED	(0 + PREEMPT_NEED_RESCHED)
+
+/*
  * We mask the PREEMPT_NEED_RESCHED bit so as not to confuse all current users
  * that think a non-zero value indicates we cannot preempt.
  */
@@ -74,6 +80,11 @@ static __always_inline void __preempt_count_sub(int val)
 	__this_cpu_add_4(__preempt_count, -val);
 }
 
+/*
+ * Because we keep PREEMPT_NEED_RESCHED set when we do _not_ need to reschedule
+ * a decrement which hits zero means we have no preempt_count and should
+ * reschedule.
+ */
 static __always_inline bool __preempt_count_dec_and_test(void)
 {
 	GEN_UNARY_RMWcc("decl", __preempt_count, __percpu_arg(0), "e");
diff --git a/arch/x86/include/asm/timer.h b/arch/x86/include/asm/timer.h
index 34baa0eb5d0..3de54ef0aea 100644
--- a/arch/x86/include/asm/timer.h
+++ b/arch/x86/include/asm/timer.h
@@ -4,6 +4,7 @@
 #include <linux/pm.h>
 #include <linux/percpu.h>
 #include <linux/interrupt.h>
+#include <linux/math64.h>
 
 #define TICK_SIZE (tick_nsec / 1000)
 
@@ -12,68 +13,26 @@ extern int recalibrate_cpu_khz(void);
 
 extern int no_timer_check;
 
-/* Accelerators for sched_clock()
- * convert from cycles(64bits) => nanoseconds (64bits)
- *  basic equation:
- *		ns = cycles / (freq / ns_per_sec)
- *		ns = cycles * (ns_per_sec / freq)
- *		ns = cycles * (10^9 / (cpu_khz * 10^3))
- *		ns = cycles * (10^6 / cpu_khz)
+/*
+ * We use the full linear equation: f(x) = a + b*x, in order to allow
+ * a continuous function in the face of dynamic freq changes.
  *
- *	Then we use scaling math (suggested by george@mvista.com) to get:
- *		ns = cycles * (10^6 * SC / cpu_khz) / SC
- *		ns = cycles * cyc2ns_scale / SC
+ * Continuity means that when our frequency changes our slope (b); we want to
+ * ensure that: f(t) == f'(t), which gives: a + b*t == a' + b'*t.
  *
- *	And since SC is a constant power of two, we can convert the div
- *  into a shift.
+ * Without an offset (a) the above would not be possible.
  *
- *  We can use khz divisor instead of mhz to keep a better precision, since
- *  cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
- *  (mathieu.desnoyers@polymtl.ca)
- *
- *			-johnstul@us.ibm.com "math is hard, lets go shopping!"
- *
- * In:
- *
- * ns = cycles * cyc2ns_scale / SC
- *
- * Although we may still have enough bits to store the value of ns,
- * in some cases, we may not have enough bits to store cycles * cyc2ns_scale,
- * leading to an incorrect result.
- *
- * To avoid this, we can decompose 'cycles' into quotient and remainder
- * of division by SC.  Then,
- *
- * ns = (quot * SC + rem) * cyc2ns_scale / SC
- *    = quot * cyc2ns_scale + (rem * cyc2ns_scale) / SC
- *
- *			- sqazi@google.com
+ * See the comment near cycles_2_ns() for details on how we compute (b).
  */
-
-DECLARE_PER_CPU(unsigned long, cyc2ns);
-DECLARE_PER_CPU(unsigned long long, cyc2ns_offset);
-
-#define CYC2NS_SCALE_FACTOR 10 /* 2^10, carefully chosen */
-
-static inline unsigned long long __cycles_2_ns(unsigned long long cyc)
-{
-	int cpu = smp_processor_id();
-	unsigned long long ns = per_cpu(cyc2ns_offset, cpu);
-	ns += mult_frac(cyc, per_cpu(cyc2ns, cpu),
-			(1UL << CYC2NS_SCALE_FACTOR));
-	return ns;
-}
-
-static inline unsigned long long cycles_2_ns(unsigned long long cyc)
-{
-	unsigned long long ns;
-	unsigned long flags;
-
-	local_irq_save(flags);
-	ns = __cycles_2_ns(cyc);
-	local_irq_restore(flags);
-
-	return ns;
-}
+struct cyc2ns_data {
+	u32 cyc2ns_mul;
+	u32 cyc2ns_shift;
+	u64 cyc2ns_offset;
+	u32 __count;
+	/* u32 hole */
+}; /* 24 bytes -- do not grow */
+
+extern struct cyc2ns_data *cyc2ns_read_begin(void);
+extern void cyc2ns_read_end(struct cyc2ns_data *);
 
 #endif /* _ASM_X86_TIMER_H */
diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
index bca023bdd6b..8bc79cddd9a 100644
--- a/arch/x86/kernel/cpu/amd.c
+++ b/arch/x86/kernel/cpu/amd.c
@@ -487,7 +487,7 @@ static void early_init_amd(struct cpuinfo_x86 *c)
 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
 		if (!check_tsc_unstable())
-			sched_clock_stable = 1;
+			set_sched_clock_stable();
 	}
 
 #ifdef CONFIG_X86_64
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index ea04b342c02..1a439c047ff 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -93,7 +93,7 @@ static void early_init_intel(struct cpuinfo_x86 *c)
 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
 		if (!check_tsc_unstable())
-			sched_clock_stable = 1;
+			set_sched_clock_stable();
 	}
 
 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
index 8e132931614..b88645191fe 100644
--- a/arch/x86/kernel/cpu/perf_event.c
+++ b/arch/x86/kernel/cpu/perf_event.c
@@ -1883,21 +1883,27 @@ static struct pmu pmu = {
 
 void arch_perf_update_userpage(struct perf_event_mmap_page *userpg, u64 now)
 {
+	struct cyc2ns_data *data;
+
 	userpg->cap_user_time = 0;
 	userpg->cap_user_time_zero = 0;
 	userpg->cap_user_rdpmc = x86_pmu.attr_rdpmc;
 	userpg->pmc_width = x86_pmu.cntval_bits;
 
-	if (!sched_clock_stable)
+	if (!sched_clock_stable())
 		return;
 
+	data = cyc2ns_read_begin();
+
 	userpg->cap_user_time = 1;
-	userpg->time_mult = this_cpu_read(cyc2ns);
-	userpg->time_shift = CYC2NS_SCALE_FACTOR;
-	userpg->time_offset = this_cpu_read(cyc2ns_offset) - now;
+	userpg->time_mult = data->cyc2ns_mul;
+	userpg->time_shift = data->cyc2ns_shift;
+	userpg->time_offset = data->cyc2ns_offset - now;
 
 	userpg->cap_user_time_zero = 1;
-	userpg->time_zero = this_cpu_read(cyc2ns_offset);
+	userpg->time_zero = data->cyc2ns_offset;
+
+	cyc2ns_read_end(data);
 }
 
 /*
diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h
index fd00bb29425..c1a861829d8 100644
--- a/arch/x86/kernel/cpu/perf_event.h
+++ b/arch/x86/kernel/cpu/perf_event.h
@@ -262,11 +262,20 @@ struct cpu_hw_events {
 	__EVENT_CONSTRAINT(c, n, INTEL_ARCH_EVENT_MASK, \
 			  HWEIGHT(n), 0, PERF_X86_EVENT_PEBS_ST_HSW)
 
-#define EVENT_CONSTRAINT_END		\
-	EVENT_CONSTRAINT(0, 0, 0)
+/*
+ * We define the end marker as having a weight of -1
+ * to enable blacklisting of events using a counter bitmask
+ * of zero and thus a weight of zero.
+ * The end marker has a weight that cannot possibly be
+ * obtained from counting the bits in the bitmask.
+ */
+#define EVENT_CONSTRAINT_END { .weight = -1 }
 
+/*
+ * Check for end marker with weight == -1
+ */
 #define for_each_event_constraint(e, c)	\
-	for ((e) = (c); (e)->weight; (e)++)
+	for ((e) = (c); (e)->weight != -1; (e)++)
 
 /*
  * Extra registers for specific events.
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index 930e5d48f56..6377fb28b95 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -11,6 +11,7 @@
 #include <linux/clocksource.h>
 #include <linux/percpu.h>
 #include <linux/timex.h>
+#include <linux/static_key.h>
 
 #include <asm/hpet.h>
 #include <asm/timer.h>
@@ -37,13 +38,244 @@ static int __read_mostly tsc_unstable;
    erroneous rdtsc usage on !cpu_has_tsc processors */
 static int __read_mostly tsc_disabled = -1;
 
+static struct static_key __use_tsc = STATIC_KEY_INIT;
+
 int tsc_clocksource_reliable;
+
+/*
+ * Use a ring-buffer like data structure, where a writer advances the head by
+ * writing a new data entry and a reader advances the tail when it observes a
+ * new entry.
+ *
+ * Writers are made to wait on readers until there's space to write a new
+ * entry.
+ *
+ * This means that we can always use an {offset, mul} pair to compute a ns
+ * value that is 'roughly' in the right direction, even if we're writing a new
+ * {offset, mul} pair during the clock read.
+ *
+ * The down-side is that we can no longer guarantee strict monotonicity anymore
+ * (assuming the TSC was that to begin with), because while we compute the
+ * intersection point of the two clock slopes and make sure the time is
+ * continuous at the point of switching; we can no longer guarantee a reader is
+ * strictly before or after the switch point.
+ *
+ * It does mean a reader no longer needs to disable IRQs in order to avoid
+ * CPU-Freq updates messing with his times, and similarly an NMI reader will
+ * no longer run the risk of hitting half-written state.
+ */
+
+struct cyc2ns {
+	struct cyc2ns_data data[2];	/*  0 + 2*24 = 48 */
+	struct cyc2ns_data *head;	/* 48 + 8    = 56 */
+	struct cyc2ns_data *tail;	/* 56 + 8    = 64 */
+}; /* exactly fits one cacheline */
+
+static DEFINE_PER_CPU_ALIGNED(struct cyc2ns, cyc2ns);
+
+struct cyc2ns_data *cyc2ns_read_begin(void)
+{
+	struct cyc2ns_data *head;
+
+	preempt_disable();
+
+	head = this_cpu_read(cyc2ns.head);
+	/*
+	 * Ensure we observe the entry when we observe the pointer to it.
+	 * matches the wmb from cyc2ns_write_end().
+	 */
+	smp_read_barrier_depends();
+	head->__count++;
+	barrier();
+
+	return head;
+}
+
+void cyc2ns_read_end(struct cyc2ns_data *head)
+{
+	barrier();
+	/*
+	 * If we're the outer most nested read; update the tail pointer
+	 * when we're done. This notifies possible pending writers
+	 * that we've observed the head pointer and that the other
+	 * entry is now free.
+	 */
+	if (!--head->__count) {
+		/*
+		 * x86-TSO does not reorder writes with older reads;
+		 * therefore once this write becomes visible to another
+		 * cpu, we must be finished reading the cyc2ns_data.
+		 *
+		 * matches with cyc2ns_write_begin().
+		 */
+		this_cpu_write(cyc2ns.tail, head);
+	}
+	preempt_enable();
+}
+
+/*
+ * Begin writing a new @data entry for @cpu.
+ *
+ * Assumes some sort of write side lock; currently 'provided' by the assumption
+ * that cpufreq will call its notifiers sequentially.
+ */
+static struct cyc2ns_data *cyc2ns_write_begin(int cpu)
+{
+	struct cyc2ns *c2n = &per_cpu(cyc2ns, cpu);
+	struct cyc2ns_data *data = c2n->data;
+
+	if (data == c2n->head)
+		data++;
+
+	/* XXX send an IPI to @cpu in order to guarantee a read? */
+
+	/*
+	 * When we observe the tail write from cyc2ns_read_end(),
+	 * the cpu must be done with that entry and its safe
+	 * to start writing to it.
+	 */
+	while (c2n->tail == data)
+		cpu_relax();
+
+	return data;
+}
+
+static void cyc2ns_write_end(int cpu, struct cyc2ns_data *data)
+{
+	struct cyc2ns *c2n = &per_cpu(cyc2ns, cpu);
+
+	/*
+	 * Ensure the @data writes are visible before we publish the
+	 * entry. Matches the data-depencency in cyc2ns_read_begin().
+	 */
+	smp_wmb();
+
+	ACCESS_ONCE(c2n->head) = data;
+}
+
+/*
+ * Accelerators for sched_clock()
+ * convert from cycles(64bits) => nanoseconds (64bits)
+ *  basic equation:
+ *              ns = cycles / (freq / ns_per_sec)
+ *              ns = cycles * (ns_per_sec / freq)
+ *              ns = cycles * (10^9 / (cpu_khz * 10^3))
+ *              ns = cycles * (10^6 / cpu_khz)
+ *
+ *      Then we use scaling math (suggested by george@mvista.com) to get:
+ *              ns = cycles * (10^6 * SC / cpu_khz) / SC
+ *              ns = cycles * cyc2ns_scale / SC
+ *
+ *      And since SC is a constant power of two, we can convert the div
+ *  into a shift.
+ *
+ *  We can use khz divisor instead of mhz to keep a better precision, since
+ *  cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
+ *  (mathieu.desnoyers@polymtl.ca)
+ *
+ *                      -johnstul@us.ibm.com "math is hard, lets go shopping!"
+ */
+
+#define CYC2NS_SCALE_FACTOR 10 /* 2^10, carefully chosen */
+
+static void cyc2ns_data_init(struct cyc2ns_data *data)
+{
+	data->cyc2ns_mul = 1U << CYC2NS_SCALE_FACTOR;
+	data->cyc2ns_shift = CYC2NS_SCALE_FACTOR;
+	data->cyc2ns_offset = 0;
+	data->__count = 0;
+}
+
+static void cyc2ns_init(int cpu)
+{
+	struct cyc2ns *c2n = &per_cpu(cyc2ns, cpu);
+
+	cyc2ns_data_init(&c2n->data[0]);
+	cyc2ns_data_init(&c2n->data[1]);
+
+	c2n->head = c2n->data;
+	c2n->tail = c2n->data;
+}
+
+static inline unsigned long long cycles_2_ns(unsigned long long cyc)
+{
+	struct cyc2ns_data *data, *tail;
+	unsigned long long ns;
+
+	/*
+	 * See cyc2ns_read_*() for details; replicated in order to avoid
+	 * an extra few instructions that came with the abstraction.
+	 * Notable, it allows us to only do the __count and tail update
+	 * dance when its actually needed.
+	 */
+
+	preempt_disable();
+	data = this_cpu_read(cyc2ns.head);
+	tail = this_cpu_read(cyc2ns.tail);
+
+	if (likely(data == tail)) {
+		ns = data->cyc2ns_offset;
+		ns += mul_u64_u32_shr(cyc, data->cyc2ns_mul, CYC2NS_SCALE_FACTOR);
+	} else {
+		data->__count++;
+
+		barrier();
+
+		ns = data->cyc2ns_offset;
+		ns += mul_u64_u32_shr(cyc, data->cyc2ns_mul, CYC2NS_SCALE_FACTOR);
+
+		barrier();
+
+		if (!--data->__count)
+			this_cpu_write(cyc2ns.tail, data);
+	}
+	preempt_enable();
+
+	return ns;
+}
+
+/* XXX surely we already have this someplace in the kernel?! */
+#define DIV_ROUND(n, d) (((n) + ((d) / 2)) / (d))
+
+static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu)
+{
+	unsigned long long tsc_now, ns_now;
+	struct cyc2ns_data *data;
+	unsigned long flags;
+
+	local_irq_save(flags);
+	sched_clock_idle_sleep_event();
+
+	if (!cpu_khz)
+		goto done;
+
+	data = cyc2ns_write_begin(cpu);
+
+	rdtscll(tsc_now);
+	ns_now = cycles_2_ns(tsc_now);
+
+	/*
+	 * Compute a new multiplier as per the above comment and ensure our
+	 * time function is continuous; see the comment near struct
+	 * cyc2ns_data.
+	 */
+	data->cyc2ns_mul = DIV_ROUND(NSEC_PER_MSEC << CYC2NS_SCALE_FACTOR, cpu_khz);
+	data->cyc2ns_shift = CYC2NS_SCALE_FACTOR;
+	data->cyc2ns_offset = ns_now -
+		mul_u64_u32_shr(tsc_now, data->cyc2ns_mul, CYC2NS_SCALE_FACTOR);
+
+	cyc2ns_write_end(cpu, data);
+
+done:
+	sched_clock_idle_wakeup_event(0);
+	local_irq_restore(flags);
+}
 /*
  * Scheduler clock - returns current time in nanosec units.
  */
 u64 native_sched_clock(void)
 {
-	u64 this_offset;
+	u64 tsc_now;
 
 	/*
 	 * Fall back to jiffies if there's no TSC available:
@@ -53,16 +285,16 @@ u64 native_sched_clock(void)
 	 *   very important for it to be as fast as the platform
 	 *   can achieve it. )
 	 */
-	if (unlikely(tsc_disabled)) {
+	if (!static_key_false(&__use_tsc)) {
 		/* No locking but a rare wrong value is not a big deal: */
 		return (jiffies_64 - INITIAL_JIFFIES) * (1000000000 / HZ);
 	}
 
 	/* read the Time Stamp Counter: */
-	rdtscll(this_offset);
+	rdtscll(tsc_now);
 
 	/* return the value in ns */
-	return __cycles_2_ns(this_offset);
+	return cycles_2_ns(tsc_now);
 }
 
 /* We need to define a real function for sched_clock, to override the
@@ -589,61 +821,11 @@ int recalibrate_cpu_khz(void)
 EXPORT_SYMBOL(recalibrate_cpu_khz);
 
 
-/* Accelerators for sched_clock()
- * convert from cycles(64bits) => nanoseconds (64bits)
- *  basic equation:
- *              ns = cycles / (freq / ns_per_sec)
- *              ns = cycles * (ns_per_sec / freq)
- *              ns = cycles * (10^9 / (cpu_khz * 10^3))
- *              ns = cycles * (10^6 / cpu_khz)
- *
- *      Then we use scaling math (suggested by george@mvista.com) to get:
- *              ns = cycles * (10^6 * SC / cpu_khz) / SC
- *              ns = cycles * cyc2ns_scale / SC
- *
- *      And since SC is a constant power of two, we can convert the div
- *  into a shift.
- *
- *  We can use khz divisor instead of mhz to keep a better precision, since
- *  cyc2ns_scale is limited to 10^6 * 2^10, which fits in 32 bits.
- *  (mathieu.desnoyers@polymtl.ca)
- *
- *                      -johnstul@us.ibm.com "math is hard, lets go shopping!"
- */
-
-DEFINE_PER_CPU(unsigned long, cyc2ns);
-DEFINE_PER_CPU(unsigned long long, cyc2ns_offset);
-
-static void set_cyc2ns_scale(unsigned long cpu_khz, int cpu)
-{
-	unsigned long long tsc_now, ns_now, *offset;
-	unsigned long flags, *scale;
-
-	local_irq_save(flags);
-	sched_clock_idle_sleep_event();
-
-	scale = &per_cpu(cyc2ns, cpu);
-	offset = &per_cpu(cyc2ns_offset, cpu);
-
-	rdtscll(tsc_now);
-	ns_now = __cycles_2_ns(tsc_now);
-
-	if (cpu_khz) {
-		*scale = ((NSEC_PER_MSEC << CYC2NS_SCALE_FACTOR) +
-				cpu_khz / 2) / cpu_khz;
-		*offset = ns_now - mult_frac(tsc_now, *scale,
-					     (1UL << CYC2NS_SCALE_FACTOR));
-	}
-
-	sched_clock_idle_wakeup_event(0);
-	local_irq_restore(flags);
-}
-
 static unsigned long long cyc2ns_suspend;
 
 void tsc_save_sched_clock_state(void)
 {
-	if (!sched_clock_stable)
+	if (!sched_clock_stable())
 		return;
 
 	cyc2ns_suspend = sched_clock();
@@ -663,16 +845,26 @@ void tsc_restore_sched_clock_state(void)
 	unsigned long flags;
 	int cpu;
 
-	if (!sched_clock_stable)
+	if (!sched_clock_stable())
 		return;
 
 	local_irq_save(flags);
 
-	__this_cpu_write(cyc2ns_offset, 0);
+	/*
+	 * We're comming out of suspend, there's no concurrency yet; don't
+	 * bother being nice about the RCU stuff, just write to both
+	 * data fields.
+	 */
+
+	this_cpu_write(cyc2ns.data[0].cyc2ns_offset, 0);
+	this_cpu_write(cyc2ns.data[1].cyc2ns_offset, 0);
+
 	offset = cyc2ns_suspend - sched_clock();
 
-	for_each_possible_cpu(cpu)
-		per_cpu(cyc2ns_offset, cpu) = offset;
+	for_each_possible_cpu(cpu) {
+		per_cpu(cyc2ns.data[0].cyc2ns_offset, cpu) = offset;
+		per_cpu(cyc2ns.data[1].cyc2ns_offset, cpu) = offset;
+	}
 
 	local_irq_restore(flags);
 }
@@ -795,7 +987,7 @@ void mark_tsc_unstable(char *reason)
 {
 	if (!tsc_unstable) {
 		tsc_unstable = 1;
-		sched_clock_stable = 0;
+		clear_sched_clock_stable();
 		disable_sched_clock_irqtime();
 		pr_info("Marking TSC unstable due to %s\n", reason);
 		/* Change only the rating, when not registered */
@@ -995,14 +1187,18 @@ void __init tsc_init(void)
 	 * speed as the bootup CPU. (cpufreq notifiers will fix this
 	 * up if their speed diverges)
 	 */
-	for_each_possible_cpu(cpu)
+	for_each_possible_cpu(cpu) {
+		cyc2ns_init(cpu);
 		set_cyc2ns_scale(cpu_khz, cpu);
+	}
 
 	if (tsc_disabled > 0)
 		return;
 
 	/* now allow native_sched_clock() to use rdtsc */
+
 	tsc_disabled = 0;
+	static_key_slow_inc(&__use_tsc);
 
 	if (!no_sched_irq_time)
 		enable_sched_clock_irqtime();
diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index dec48bfaddb..1673940cf9c 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -1350,6 +1350,10 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
 		return;
 	}
 
+	if (!kvm_vcpu_is_bsp(apic->vcpu))
+		value &= ~MSR_IA32_APICBASE_BSP;
+	vcpu->arch.apic_base = value;
+
 	/* update jump label if enable bit changes */
 	if ((vcpu->arch.apic_base ^ value) & MSR_IA32_APICBASE_ENABLE) {
 		if (value & MSR_IA32_APICBASE_ENABLE)
@@ -1359,10 +1363,6 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
 		recalculate_apic_map(vcpu->kvm);
 	}
 
-	if (!kvm_vcpu_is_bsp(apic->vcpu))
-		value &= ~MSR_IA32_APICBASE_BSP;
-
-	vcpu->arch.apic_base = value;
 	if ((old_value ^ value) & X2APIC_ENABLE) {
 		if (value & X2APIC_ENABLE) {
 			u32 id = kvm_apic_id(apic);
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index b2fe1c252f3..da7837e1349 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -8283,8 +8283,7 @@ static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
 	vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
 	kvm_set_cr4(vcpu, vmcs12->host_cr4);
 
-	if (nested_cpu_has_ept(vmcs12))
-		nested_ept_uninit_mmu_context(vcpu);
+	nested_ept_uninit_mmu_context(vcpu);
 
 	kvm_set_cr3(vcpu, vmcs12->host_cr3);
 	kvm_mmu_reset_context(vcpu);
diff --git a/arch/x86/mm/gup.c b/arch/x86/mm/gup.c
index dd74e46828c..0596e8e0cc1 100644
--- a/arch/x86/mm/gup.c
+++ b/arch/x86/mm/gup.c
@@ -83,6 +83,12 @@ static noinline int gup_pte_range(pmd_t pmd, unsigned long addr,
 		pte_t pte = gup_get_pte(ptep);
 		struct page *page;
 
+		/* Similar to the PMD case, NUMA hinting must take slow path */
+		if (pte_numa(pte)) {
+			pte_unmap(ptep);
+			return 0;
+		}
+
 		if ((pte_flags(pte) & (mask | _PAGE_SPECIAL)) != mask) {
 			pte_unmap(ptep);
 			return 0;
@@ -167,6 +173,13 @@ static int gup_pmd_range(pud_t pud, unsigned long addr, unsigned long end,
 		if (pmd_none(pmd) || pmd_trans_splitting(pmd))
 			return 0;
 		if (unlikely(pmd_large(pmd))) {
+			/*
+			 * NUMA hinting faults need to be handled in the GUP
+			 * slowpath for accounting purposes and so that they
+			 * can be serialised against THP migration.
+			 */
+			if (pmd_numa(pmd))
+				return 0;
 			if (!gup_huge_pmd(pmd, addr, next, write, pages, nr))
 				return 0;
 		} else {
diff --git a/arch/x86/platform/uv/tlb_uv.c b/arch/x86/platform/uv/tlb_uv.c
index efe4d722039..dfe605ac1bc 100644
--- a/arch/x86/platform/uv/tlb_uv.c
+++ b/arch/x86/platform/uv/tlb_uv.c
@@ -433,15 +433,49 @@ static void reset_with_ipi(struct pnmask *distribution, struct bau_control *bcp)
 	return;
 }
 
-static inline unsigned long cycles_2_us(unsigned long long cyc)
+/*
+ * Not to be confused with cycles_2_ns() from tsc.c; this gives a relative
+ * number, not an absolute. It converts a duration in cycles to a duration in
+ * ns.
+ */
+static inline unsigned long long cycles_2_ns(unsigned long long cyc)
 {
+	struct cyc2ns_data *data = cyc2ns_read_begin();
 	unsigned long long ns;
-	unsigned long us;
-	int cpu = smp_processor_id();
 
-	ns =  (cyc * per_cpu(cyc2ns, cpu)) >> CYC2NS_SCALE_FACTOR;
-	us = ns / 1000;
-	return us;
+	ns = mul_u64_u32_shr(cyc, data->cyc2ns_mul, data->cyc2ns_shift);
+
+	cyc2ns_read_end(data);
+	return ns;
+}
+
+/*
+ * The reverse of the above; converts a duration in ns to a duration in cycles.
+ */ 
+static inline unsigned long long ns_2_cycles(unsigned long long ns)
+{
+	struct cyc2ns_data *data = cyc2ns_read_begin();
+	unsigned long long cyc;
+
+	cyc = (ns << data->cyc2ns_shift) / data->cyc2ns_mul;
+
+	cyc2ns_read_end(data);
+	return cyc;
+}
+
+static inline unsigned long cycles_2_us(unsigned long long cyc)
+{
+	return cycles_2_ns(cyc) / NSEC_PER_USEC;
+}
+
+static inline cycles_t sec_2_cycles(unsigned long sec)
+{
+	return ns_2_cycles(sec * NSEC_PER_SEC);
+}
+
+static inline unsigned long long usec_2_cycles(unsigned long usec)
+{
+	return ns_2_cycles(usec * NSEC_PER_USEC);
 }
 
 /*
@@ -668,16 +702,6 @@ static int wait_completion(struct bau_desc *bau_desc,
 								bcp, try);
 }
 
-static inline cycles_t sec_2_cycles(unsigned long sec)
-{
-	unsigned long ns;
-	cycles_t cyc;
-
-	ns = sec * 1000000000;
-	cyc = (ns << CYC2NS_SCALE_FACTOR)/(per_cpu(cyc2ns, smp_processor_id()));
-	return cyc;
-}
-
 /*
  * Our retries are blocked by all destination sw ack resources being
  * in use, and a timeout is pending. In that case hardware immediately
@@ -1327,16 +1351,6 @@ static void ptc_seq_stop(struct seq_file *file, void *data)
 {
 }
 
-static inline unsigned long long usec_2_cycles(unsigned long microsec)
-{
-	unsigned long ns;
-	unsigned long long cyc;
-
-	ns = microsec * 1000;
-	cyc = (ns << CYC2NS_SCALE_FACTOR)/(per_cpu(cyc2ns, smp_processor_id()));
-	return cyc;
-}
-
 /*
  * Display the statistics thru /proc/sgi_uv/ptc_statistics
  * 'data' points to the cpu number
diff --git a/arch/x86/syscalls/syscall_32.tbl b/arch/x86/syscalls/syscall_32.tbl
index aabfb8380a1..96bc506ac6d 100644
--- a/arch/x86/syscalls/syscall_32.tbl
+++ b/arch/x86/syscalls/syscall_32.tbl
@@ -357,3 +357,5 @@
 348	i386	process_vm_writev	sys_process_vm_writev		compat_sys_process_vm_writev
 349	i386	kcmp			sys_kcmp
 350	i386	finit_module		sys_finit_module
+351	i386	sched_setattr		sys_sched_setattr
+352	i386	sched_getattr		sys_sched_getattr
diff --git a/arch/x86/syscalls/syscall_64.tbl b/arch/x86/syscalls/syscall_64.tbl
index 38ae65dfd14..a12bddc7cce 100644
--- a/arch/x86/syscalls/syscall_64.tbl
+++ b/arch/x86/syscalls/syscall_64.tbl
@@ -320,6 +320,8 @@
 311	64	process_vm_writev	sys_process_vm_writev
 312	common	kcmp			sys_kcmp
 313	common	finit_module		sys_finit_module
+314	common	sched_setattr		sys_sched_setattr
+315	common	sched_getattr		sys_sched_getattr
 
 #
 # x32-specific system call numbers start at 512 to avoid cache impact