[gcv v4 01/38] x86: Use this_cpu_inc/dec for debug registers

[gcv v4 01/38] x86: Use this_cpu_inc/dec for debug registers

[Christoph Lameter]

I think this patch already exists in one form or another in some x86 tree.

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/arch/x86/include/asm/debugreg.h
===================================================================
--- linux.orig/arch/x86/include/asm/debugreg.h	2013-08-06 13:52:23.740092873 -0500
+++ linux/arch/x86/include/asm/debugreg.h	2013-08-06 13:52:23.740092873 -0500
@@ -97,11 +97,11 @@ extern void hw_breakpoint_restore(void);
 DECLARE_PER_CPU(int, debug_stack_usage);
 static inline void debug_stack_usage_inc(void)
 {
-	__get_cpu_var(debug_stack_usage)++;
+	__this_cpu_inc(debug_stack_usage);
 }
 static inline void debug_stack_usage_dec(void)
 {
-	__get_cpu_var(debug_stack_usage)--;
+	__this_cpu_dec(debug_stack_usage);
 }
 int is_debug_stack(unsigned long addr);
 void debug_stack_set_zero(void);
Index: linux/arch/x86/kernel/cpu/common.c
===================================================================
--- linux.orig/arch/x86/kernel/cpu/common.c	2013-07-19 09:06:36.850047837 -0500
+++ linux/arch/x86/kernel/cpu/common.c	2013-08-06 13:54:36.426384889 -0500
@@ -1144,9 +1144,9 @@ DEFINE_PER_CPU(int, debug_stack_usage);
 
 int is_debug_stack(unsigned long addr)
 {
-	return __get_cpu_var(debug_stack_usage) ||
-		(addr <= __get_cpu_var(debug_stack_addr) &&
-		 addr > (__get_cpu_var(debug_stack_addr) - DEBUG_STKSZ));
+	return __this_cpu_read(debug_stack_usage) ||
+		(addr <= __this_cpu_read(debug_stack_addr) &&
+		 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
 }
 
 DEFINE_PER_CPU(u32, debug_idt_ctr);

[gcv v4 02/38] percpu: Make __verify_pcu_ptr handle per cpu pointers to arrays

[Christoph Lameter]

__verify_pcpu_ptr() will cause a compilation failure if the type of the
pointer is a pointer to a fixed array of objects. Adding zero to the
pointer converts the type of pointer to that pointing to a single
object of the array.

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/include/linux/percpu-defs.h
===================================================================
--- linux.orig/include/linux/percpu-defs.h	2013-08-22 13:39:04.000000000 -0500
+++ linux/include/linux/percpu-defs.h	2013-08-22 13:41:15.333140537 -0500
@@ -22,9 +22,12 @@
  * Macro which verifies @ptr is a percpu pointer without evaluating
  * @ptr.  This is to be used in percpu accessors to verify that the
  * input parameter is a percpu pointer.
+ *
+ * + 0 is required in order to convert the pointer type from a
+ * potential array type to a pointer to a single item of the array.
  */
 #define __verify_pcpu_ptr(ptr)	do {					\
-	const void __percpu *__vpp_verify = (typeof(ptr))NULL;		\
+	const void __percpu *__vpp_verify = (typeof((ptr) + 0))NULL;	\
 	(void)__vpp_verify;						\
 } while (0)
 

[gcv v4 04/38] Use raw_cpu ops so that my kvm session runs fine

[Christoph Lameter]

These location triggered during testing with KVM. The list here
is by no means extensive.

These are fetches without preemption off where we judged that
to be more performance efficient or where other means of
providing synchronization (BH handling) are avalt ilable.

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/include/linux/topology.h
===================================================================
--- linux.orig/include/linux/topology.h	2013-09-03 13:41:00.701563523 -0500
+++ linux/include/linux/topology.h	2013-09-03 13:41:00.693563604 -0500
@@ -182,7 +182,7 @@ DECLARE_PER_CPU(int, numa_node);
 /* Returns the number of the current Node. */
 static inline int numa_node_id(void)
 {
-	return __this_cpu_read(numa_node);
+	return raw_cpu_read(numa_node);
 }
 #endif
 
Index: linux/include/net/snmp.h
===================================================================
--- linux.orig/include/net/snmp.h	2013-09-03 13:41:00.701563523 -0500
+++ linux/include/net/snmp.h	2013-09-03 13:49:05.840592016 -0500
@@ -126,7 +126,7 @@ struct linux_xfrm_mib {
 	extern __typeof__(type) __percpu *name[SNMP_ARRAY_SZ]
 
 #define SNMP_INC_STATS_BH(mib, field)	\
-			__this_cpu_inc(mib[0]->mibs[field])
+			raw_cpu_inc(mib[0]->mibs[field])
 
 #define SNMP_INC_STATS_USER(mib, field)	\
 			this_cpu_inc(mib[0]->mibs[field])
@@ -141,7 +141,7 @@ struct linux_xfrm_mib {
 			this_cpu_dec(mib[0]->mibs[field])
 
 #define SNMP_ADD_STATS_BH(mib, field, addend)	\
-			__this_cpu_add(mib[0]->mibs[field], addend)
+			raw_cpu_add(mib[0]->mibs[field], addend)
 
 #define SNMP_ADD_STATS_USER(mib, field, addend)	\
 			this_cpu_add(mib[0]->mibs[field], addend)
Index: linux/kernel/hrtimer.c
===================================================================
--- linux.orig/kernel/hrtimer.c	2013-09-03 13:41:00.701563523 -0500
+++ linux/kernel/hrtimer.c	2013-09-03 13:41:00.693563604 -0500
@@ -538,7 +538,7 @@ static inline int hrtimer_is_hres_enable
  */
 static inline int hrtimer_hres_active(void)
 {
-	return __this_cpu_read(hrtimer_bases.hres_active);
+	return raw_cpu_read(hrtimer_bases.hres_active);
 }
 
 /*

[gcv v4 07/38] net: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/net/core/dev.c
===================================================================
--- linux.orig/net/core/dev.c	2013-08-26 14:16:59.000000000 -0500
+++ linux/net/core/dev.c	2013-08-26 14:18:37.214005168 -0500
@@ -2129,7 +2129,7 @@ static inline void __netif_reschedule(st
 	unsigned long flags;
 
 	local_irq_save(flags);
-	sd = &__get_cpu_var(softnet_data);
+	sd = this_cpu_ptr(&softnet_data);
 	q->next_sched = NULL;
 	*sd->output_queue_tailp = q;
 	sd->output_queue_tailp = &q->next_sched;
@@ -2151,7 +2151,7 @@ void dev_kfree_skb_irq(struct sk_buff *s
 		unsigned long flags;
 
 		local_irq_save(flags);
-		sd = &__get_cpu_var(softnet_data);
+		sd = this_cpu_ptr(&softnet_data);
 		skb->next = sd->completion_queue;
 		sd->completion_queue = skb;
 		raise_softirq_irqoff(NET_TX_SOFTIRQ);
@@ -3111,7 +3111,7 @@ static void rps_trigger_softirq(void *da
 static int rps_ipi_queued(struct softnet_data *sd)
 {
 #ifdef CONFIG_RPS
-	struct softnet_data *mysd = &__get_cpu_var(softnet_data);
+	struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
 
 	if (sd != mysd) {
 		sd->rps_ipi_next = mysd->rps_ipi_list;
@@ -3138,7 +3138,7 @@ static bool skb_flow_limit(struct sk_buf
 	if (qlen < (netdev_max_backlog >> 1))
 		return false;
 
-	sd = &__get_cpu_var(softnet_data);
+	sd = this_cpu_ptr(&softnet_data);
 
 	rcu_read_lock();
 	fl = rcu_dereference(sd->flow_limit);
@@ -3280,7 +3280,7 @@ EXPORT_SYMBOL(netif_rx_ni);
 
 static void net_tx_action(struct softirq_action *h)
 {
-	struct softnet_data *sd = &__get_cpu_var(softnet_data);
+	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
 
 	if (sd->completion_queue) {
 		struct sk_buff *clist;
@@ -3700,7 +3700,7 @@ EXPORT_SYMBOL(netif_receive_skb);
 static void flush_backlog(void *arg)
 {
 	struct net_device *dev = arg;
-	struct softnet_data *sd = &__get_cpu_var(softnet_data);
+	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
 	struct sk_buff *skb, *tmp;
 
 	rps_lock(sd);
@@ -4146,7 +4146,7 @@ void __napi_schedule(struct napi_struct
 	unsigned long flags;
 
 	local_irq_save(flags);
-	____napi_schedule(&__get_cpu_var(softnet_data), n);
+	____napi_schedule(this_cpu_ptr(&softnet_data), n);
 	local_irq_restore(flags);
 }
 EXPORT_SYMBOL(__napi_schedule);
@@ -4274,7 +4274,7 @@ EXPORT_SYMBOL(netif_napi_del);
 
 static void net_rx_action(struct softirq_action *h)
 {
-	struct softnet_data *sd = &__get_cpu_var(softnet_data);
+	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
 	unsigned long time_limit = jiffies + 2;
 	int budget = netdev_budget;
 	void *have;
Index: linux/net/core/drop_monitor.c
===================================================================
--- linux.orig/net/core/drop_monitor.c	2013-08-26 14:16:59.000000000 -0500
+++ linux/net/core/drop_monitor.c	2013-08-26 14:18:37.218005126 -0500
@@ -142,7 +142,7 @@ static void trace_drop_common(struct sk_
 	unsigned long flags;
 
 	local_irq_save(flags);
-	data = &__get_cpu_var(dm_cpu_data);
+	data = this_cpu_ptr(&dm_cpu_data);
 	spin_lock(&data->lock);
 	dskb = data->skb;
 
Index: linux/net/core/skbuff.c
===================================================================
--- linux.orig/net/core/skbuff.c	2013-08-26 14:16:59.000000000 -0500
+++ linux/net/core/skbuff.c	2013-08-26 14:18:37.218005126 -0500
@@ -371,7 +371,7 @@ static void *__netdev_alloc_frag(unsigne
 	unsigned long flags;
 
 	local_irq_save(flags);
-	nc = &__get_cpu_var(netdev_alloc_cache);
+	nc = this_cpu_ptr(&netdev_alloc_cache);
 	if (unlikely(!nc->frag.page)) {
 refill:
 		for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
Index: linux/net/ipv4/syncookies.c
===================================================================
--- linux.orig/net/ipv4/syncookies.c	2013-08-26 14:16:59.000000000 -0500
+++ linux/net/ipv4/syncookies.c	2013-08-26 14:18:37.210005210 -0500
@@ -44,7 +44,7 @@ static DEFINE_PER_CPU(__u32 [16 + 5 + SH
 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
 		       u32 count, int c)
 {
-	__u32 *tmp = __get_cpu_var(ipv4_cookie_scratch);
+	__u32 *tmp = this_cpu_ptr(ipv4_cookie_scratch);
 
 	memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
 	tmp[0] = (__force u32)saddr;
Index: linux/net/ipv4/tcp_output.c
===================================================================
--- linux.orig/net/ipv4/tcp_output.c	2013-08-26 14:16:59.000000000 -0500
+++ linux/net/ipv4/tcp_output.c	2013-08-26 14:18:37.218005126 -0500
@@ -810,7 +810,7 @@ void tcp_wfree(struct sk_buff *skb)
 
 		/* queue this socket to tasklet queue */
 		local_irq_save(flags);
-		tsq = &__get_cpu_var(tsq_tasklet);
+		tsq = this_cpu_ptr(&tsq_tasklet);
 		list_add(&tp->tsq_node, &tsq->head);
 		tasklet_schedule(&tsq->tasklet);
 		local_irq_restore(flags);
Index: linux/net/ipv6/syncookies.c
===================================================================
--- linux.orig/net/ipv6/syncookies.c	2013-08-26 14:16:59.000000000 -0500
+++ linux/net/ipv6/syncookies.c	2013-08-26 14:18:37.214005168 -0500
@@ -66,7 +66,7 @@ static DEFINE_PER_CPU(__u32 [16 + 5 + SH
 static u32 cookie_hash(const struct in6_addr *saddr, const struct in6_addr *daddr,
 		       __be16 sport, __be16 dport, u32 count, int c)
 {
-	__u32 *tmp = __get_cpu_var(ipv6_cookie_scratch);
+	__u32 *tmp = this_cpu_ptr(ipv6_cookie_scratch);
 
 	/*
 	 * we have 320 bits of information to hash, copy in the remaining
Index: linux/net/rds/ib_rdma.c
===================================================================
--- linux.orig/net/rds/ib_rdma.c	2013-08-26 14:16:59.000000000 -0500
+++ linux/net/rds/ib_rdma.c	2013-08-26 14:18:37.218005126 -0500
@@ -267,7 +267,7 @@ static inline struct rds_ib_mr *rds_ib_r
 	unsigned long *flag;
 
 	preempt_disable();
-	flag = &__get_cpu_var(clean_list_grace);
+	flag = this_cpu_ptr(&clean_list_grace);
 	set_bit(CLEAN_LIST_BUSY_BIT, flag);
 	ret = llist_del_first(&pool->clean_list);
 	if (ret)
Index: linux/include/net/netfilter/nf_conntrack.h
===================================================================
--- linux.orig/include/net/netfilter/nf_conntrack.h	2013-08-26 14:23:46.000000000 -0500
+++ linux/include/net/netfilter/nf_conntrack.h	2013-08-26 14:24:17.190395755 -0500
@@ -243,7 +243,7 @@ extern s16 (*nf_ct_nat_offset)(const str
 DECLARE_PER_CPU(struct nf_conn, nf_conntrack_untracked);
 static inline struct nf_conn *nf_ct_untracked_get(void)
 {
-	return &__raw_get_cpu_var(nf_conntrack_untracked);
+	return __this_cpu_ptr(&nf_conntrack_untracked);
 }
 extern void nf_ct_untracked_status_or(unsigned long bits);
 

[gcv v4 15/38] watchdog: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/kernel/watchdog.c
===================================================================
--- linux.orig/kernel/watchdog.c	2013-08-26 14:28:39.000000000 -0500
+++ linux/kernel/watchdog.c	2013-08-26 14:29:34.471034451 -0500
@@ -174,8 +174,8 @@ EXPORT_SYMBOL(touch_nmi_watchdog);
 
 void touch_softlockup_watchdog_sync(void)
 {
-	__raw_get_cpu_var(softlockup_touch_sync) = true;
-	__raw_get_cpu_var(watchdog_touch_ts) = 0;
+	__this_cpu_write(softlockup_touch_sync, 1);
+	__this_cpu_write(watchdog_touch_ts, 0);
 }
 
 #ifdef CONFIG_HARDLOCKUP_DETECTOR
@@ -341,7 +341,7 @@ static void watchdog_set_prio(unsigned i
 
 static void watchdog_enable(unsigned int cpu)
 {
-	struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
+	struct hrtimer *hrtimer = __this_cpu_ptr(&watchdog_hrtimer);
 
 	/* kick off the timer for the hardlockup detector */
 	hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
@@ -361,7 +361,7 @@ static void watchdog_enable(unsigned int
 
 static void watchdog_disable(unsigned int cpu)
 {
-	struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
+	struct hrtimer *hrtimer = __this_cpu_ptr(&watchdog_hrtimer);
 
 	watchdog_set_prio(SCHED_NORMAL, 0);
 	hrtimer_cancel(hrtimer);

[gcv v4 09/38] scheduler: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)


Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/include/linux/kernel_stat.h
===================================================================
--- linux.orig/include/linux/kernel_stat.h	2013-08-26 14:21:39.000000000 -0500
+++ linux/include/linux/kernel_stat.h	2013-08-26 14:23:08.883120293 -0500
@@ -47,8 +47,8 @@ DECLARE_PER_CPU(struct kernel_stat, ksta
 DECLARE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
 
 /* Must have preemption disabled for this to be meaningful. */
-#define kstat_this_cpu (&__get_cpu_var(kstat))
-#define kcpustat_this_cpu (&__get_cpu_var(kernel_cpustat))
+#define kstat_this_cpu this_cpu_ptr(&kstat)
+#define kcpustat_this_cpu this_cpu_ptr(&kernel_cpustat)
 #define kstat_cpu(cpu) per_cpu(kstat, cpu)
 #define kcpustat_cpu(cpu) per_cpu(kernel_cpustat, cpu)
 
Index: linux/kernel/events/callchain.c
===================================================================
--- linux.orig/kernel/events/callchain.c	2013-08-26 14:21:39.000000000 -0500
+++ linux/kernel/events/callchain.c	2013-08-26 14:23:08.875120379 -0500
@@ -134,7 +134,7 @@ static struct perf_callchain_entry *get_
 	int cpu;
 	struct callchain_cpus_entries *entries;
 
-	*rctx = get_recursion_context(__get_cpu_var(callchain_recursion));
+	*rctx = get_recursion_context(this_cpu_ptr(callchain_recursion));
 	if (*rctx == -1)
 		return NULL;
 
@@ -150,7 +150,7 @@ static struct perf_callchain_entry *get_
 static void
 put_callchain_entry(int rctx)
 {
-	put_recursion_context(__get_cpu_var(callchain_recursion), rctx);
+	put_recursion_context(this_cpu_ptr(callchain_recursion), rctx);
 }
 
 struct perf_callchain_entry *
Index: linux/kernel/events/core.c
===================================================================
--- linux.orig/kernel/events/core.c	2013-08-26 14:21:39.000000000 -0500
+++ linux/kernel/events/core.c	2013-08-26 14:23:08.875120379 -0500
@@ -238,10 +238,10 @@ void perf_sample_event_took(u64 sample_l
 		return;
 
 	/* decay the counter by 1 average sample */
-	local_samples_len = __get_cpu_var(running_sample_length);
+	local_samples_len = __this_cpu_read(running_sample_length);
 	local_samples_len -= local_samples_len/NR_ACCUMULATED_SAMPLES;
 	local_samples_len += sample_len_ns;
-	__get_cpu_var(running_sample_length) = local_samples_len;
+	__this_cpu_write(running_sample_length, local_samples_len);
 
 	/*
 	 * note: this will be biased artifically low until we have
@@ -865,7 +865,7 @@ static DEFINE_PER_CPU(struct list_head,
 static void perf_pmu_rotate_start(struct pmu *pmu)
 {
 	struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
-	struct list_head *head = &__get_cpu_var(rotation_list);
+	struct list_head *head = this_cpu_ptr(&rotation_list);
 
 	WARN_ON(!irqs_disabled());
 
@@ -2321,7 +2321,7 @@ void __perf_event_task_sched_out(struct
 	 * to check if we have to switch out PMU state.
 	 * cgroup event are system-wide mode only
 	 */
-	if (atomic_read(&__get_cpu_var(perf_cgroup_events)))
+	if (atomic_read(this_cpu_ptr(&perf_cgroup_events)))
 		perf_cgroup_sched_out(task, next);
 }
 
@@ -2566,11 +2566,11 @@ void __perf_event_task_sched_in(struct t
 	 * to check if we have to switch in PMU state.
 	 * cgroup event are system-wide mode only
 	 */
-	if (atomic_read(&__get_cpu_var(perf_cgroup_events)))
+	if (atomic_read(this_cpu_ptr(&perf_cgroup_events)))
 		perf_cgroup_sched_in(prev, task);
 
 	/* check for system-wide branch_stack events */
-	if (atomic_read(&__get_cpu_var(perf_branch_stack_events)))
+	if (atomic_read(this_cpu_ptr(&perf_branch_stack_events)))
 		perf_branch_stack_sched_in(prev, task);
 }
 
@@ -2811,7 +2811,7 @@ done:
 #ifdef CONFIG_NO_HZ_FULL
 bool perf_event_can_stop_tick(void)
 {
-	if (list_empty(&__get_cpu_var(rotation_list)))
+	if (list_empty(this_cpu_ptr(&rotation_list)))
 		return true;
 	else
 		return false;
@@ -2820,7 +2820,7 @@ bool perf_event_can_stop_tick(void)
 
 void perf_event_task_tick(void)
 {
-	struct list_head *head = &__get_cpu_var(rotation_list);
+	struct list_head *head = this_cpu_ptr(&rotation_list);
 	struct perf_cpu_context *cpuctx, *tmp;
 	struct perf_event_context *ctx;
 	int throttled;
@@ -5414,7 +5414,7 @@ static void do_perf_sw_event(enum perf_t
 				    struct perf_sample_data *data,
 				    struct pt_regs *regs)
 {
-	struct swevent_htable *swhash = &__get_cpu_var(swevent_htable);
+	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
 	struct perf_event *event;
 	struct hlist_head *head;
 
@@ -5433,7 +5433,7 @@ end:
 
 int perf_swevent_get_recursion_context(void)
 {
-	struct swevent_htable *swhash = &__get_cpu_var(swevent_htable);
+	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
 
 	return get_recursion_context(swhash->recursion);
 }
@@ -5441,7 +5441,7 @@ EXPORT_SYMBOL_GPL(perf_swevent_get_recur
 
 inline void perf_swevent_put_recursion_context(int rctx)
 {
-	struct swevent_htable *swhash = &__get_cpu_var(swevent_htable);
+	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
 
 	put_recursion_context(swhash->recursion, rctx);
 }
@@ -5470,7 +5470,7 @@ static void perf_swevent_read(struct per
 
 static int perf_swevent_add(struct perf_event *event, int flags)
 {
-	struct swevent_htable *swhash = &__get_cpu_var(swevent_htable);
+	struct swevent_htable *swhash = this_cpu_ptr(&swevent_htable);
 	struct hw_perf_event *hwc = &event->hw;
 	struct hlist_head *head;
 
Index: linux/kernel/sched/cputime.c
===================================================================
--- linux.orig/kernel/sched/cputime.c	2013-08-26 14:21:39.000000000 -0500
+++ linux/kernel/sched/cputime.c	2013-08-26 14:23:08.879120335 -0500
@@ -121,7 +121,7 @@ static inline void task_group_account_fi
 	 * is the only cgroup, then nothing else should be necessary.
 	 *
 	 */
-	__get_cpu_var(kernel_cpustat).cpustat[index] += tmp;
+	__this_cpu_add(kernel_cpustat.cpustat[index], tmp);
 
 	cpuacct_account_field(p, index, tmp);
 }
Index: linux/kernel/sched/fair.c
===================================================================
--- linux.orig/kernel/sched/fair.c	2013-08-26 14:21:39.000000000 -0500
+++ linux/kernel/sched/fair.c	2013-08-26 14:23:08.879120335 -0500
@@ -5057,7 +5057,7 @@ static int load_balance(int this_cpu, st
 	struct sched_group *group;
 	struct rq *busiest;
 	unsigned long flags;
-	struct cpumask *cpus = __get_cpu_var(load_balance_mask);
+	struct cpumask *cpus = this_cpu_ptr(load_balance_mask);
 
 	struct lb_env env = {
 		.sd		= sd,
Index: linux/kernel/sched/rt.c
===================================================================
--- linux.orig/kernel/sched/rt.c	2013-08-26 14:21:39.000000000 -0500
+++ linux/kernel/sched/rt.c	2013-08-26 14:23:08.883120293 -0500
@@ -1389,7 +1389,7 @@ static DEFINE_PER_CPU(cpumask_var_t, loc
 static int find_lowest_rq(struct task_struct *task)
 {
 	struct sched_domain *sd;
-	struct cpumask *lowest_mask = __get_cpu_var(local_cpu_mask);
+	struct cpumask *lowest_mask = this_cpu_ptr(local_cpu_mask);
 	int this_cpu = smp_processor_id();
 	int cpu      = task_cpu(task);
 
Index: linux/kernel/sched/sched.h
===================================================================
--- linux.orig/kernel/sched/sched.h	2013-08-26 14:21:39.000000000 -0500
+++ linux/kernel/sched/sched.h	2013-08-26 14:23:08.883120293 -0500
@@ -538,10 +538,10 @@ static inline int cpu_of(struct rq *rq)
 DECLARE_PER_CPU(struct rq, runqueues);
 
 #define cpu_rq(cpu)		(&per_cpu(runqueues, (cpu)))
-#define this_rq()		(&__get_cpu_var(runqueues))
+#define this_rq()		this_cpu_ptr(&runqueues)
 #define task_rq(p)		cpu_rq(task_cpu(p))
 #define cpu_curr(cpu)		(cpu_rq(cpu)->curr)
-#define raw_rq()		(&__raw_get_cpu_var(runqueues))
+#define raw_rq()		__this_cpu_ptr(&runqueues)
 
 static inline u64 rq_clock(struct rq *rq)
 {
Index: linux/kernel/user-return-notifier.c
===================================================================
--- linux.orig/kernel/user-return-notifier.c	2013-08-26 14:21:39.000000000 -0500
+++ linux/kernel/user-return-notifier.c	2013-08-26 14:23:08.883120293 -0500
@@ -14,7 +14,7 @@ static DEFINE_PER_CPU(struct hlist_head,
 void user_return_notifier_register(struct user_return_notifier *urn)
 {
 	set_tsk_thread_flag(current, TIF_USER_RETURN_NOTIFY);
-	hlist_add_head(&urn->link, &__get_cpu_var(return_notifier_list));
+	hlist_add_head(&urn->link, this_cpu_ptr(&return_notifier_list));
 }
 EXPORT_SYMBOL_GPL(user_return_notifier_register);
 
@@ -25,7 +25,7 @@ EXPORT_SYMBOL_GPL(user_return_notifier_r
 void user_return_notifier_unregister(struct user_return_notifier *urn)
 {
 	hlist_del(&urn->link);
-	if (hlist_empty(&__get_cpu_var(return_notifier_list)))
+	if (hlist_empty(this_cpu_ptr(&return_notifier_list)))
 		clear_tsk_thread_flag(current, TIF_USER_RETURN_NOTIFY);
 }
 EXPORT_SYMBOL_GPL(user_return_notifier_unregister);

[gcv v4 12/38] block: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/block/blk-iopoll.c
===================================================================
--- linux.orig/block/blk-iopoll.c	2013-08-26 14:36:07.858875667 -0500
+++ linux/block/blk-iopoll.c	2013-08-26 14:36:07.850875751 -0500
@@ -35,7 +35,7 @@ void blk_iopoll_sched(struct blk_iopoll
 	unsigned long flags;
 
 	local_irq_save(flags);
-	list_add_tail(&iop->list, &__get_cpu_var(blk_cpu_iopoll));
+	list_add_tail(&iop->list, this_cpu_ptr(&blk_cpu_iopoll));
 	__raise_softirq_irqoff(BLOCK_IOPOLL_SOFTIRQ);
 	local_irq_restore(flags);
 }
@@ -79,7 +79,7 @@ EXPORT_SYMBOL(blk_iopoll_complete);
 
 static void blk_iopoll_softirq(struct softirq_action *h)
 {
-	struct list_head *list = &__get_cpu_var(blk_cpu_iopoll);
+	struct list_head *list = this_cpu_ptr(&blk_cpu_iopoll);
 	int rearm = 0, budget = blk_iopoll_budget;
 	unsigned long start_time = jiffies;
 
@@ -201,7 +201,7 @@ static int blk_iopoll_cpu_notify(struct
 
 		local_irq_disable();
 		list_splice_init(&per_cpu(blk_cpu_iopoll, cpu),
-				 &__get_cpu_var(blk_cpu_iopoll));
+				 this_cpu_ptr(&blk_cpu_iopoll));
 		__raise_softirq_irqoff(BLOCK_IOPOLL_SOFTIRQ);
 		local_irq_enable();
 	}
Index: linux/block/blk-softirq.c
===================================================================
--- linux.orig/block/blk-softirq.c	2013-08-26 14:36:07.858875667 -0500
+++ linux/block/blk-softirq.c	2013-08-26 14:36:07.854875709 -0500
@@ -23,7 +23,7 @@ static void blk_done_softirq(struct soft
 	struct list_head *cpu_list, local_list;
 
 	local_irq_disable();
-	cpu_list = &__get_cpu_var(blk_cpu_done);
+	cpu_list = this_cpu_ptr(&blk_cpu_done);
 	list_replace_init(cpu_list, &local_list);
 	local_irq_enable();
 
@@ -44,7 +44,7 @@ static void trigger_softirq(void *data)
 	struct list_head *list;
 
 	local_irq_save(flags);
-	list = &__get_cpu_var(blk_cpu_done);
+	list = this_cpu_ptr(&blk_cpu_done);
 	list_add_tail(&rq->csd.list, list);
 
 	if (list->next == &rq->csd.list)
@@ -90,7 +90,7 @@ static int blk_cpu_notify(struct notifie
 
 		local_irq_disable();
 		list_splice_init(&per_cpu(blk_cpu_done, cpu),
-				 &__get_cpu_var(blk_cpu_done));
+				 this_cpu_ptr(&blk_cpu_done));
 		raise_softirq_irqoff(BLOCK_SOFTIRQ);
 		local_irq_enable();
 	}
@@ -135,7 +135,7 @@ void __blk_complete_request(struct reque
 	if (ccpu == cpu || shared) {
 		struct list_head *list;
 do_local:
-		list = &__get_cpu_var(blk_cpu_done);
+		list = this_cpu_ptr(&blk_cpu_done);
 		list_add_tail(&req->csd.list, list);
 
 		/*
Index: linux/fs/fscache/object.c
===================================================================
--- linux.orig/fs/fscache/object.c	2013-08-26 14:36:07.858875667 -0500
+++ linux/fs/fscache/object.c	2013-08-26 14:36:07.854875709 -0500
@@ -796,7 +796,7 @@ void fscache_enqueue_object(struct fscac
  */
 bool fscache_object_sleep_till_congested(signed long *timeoutp)
 {
-	wait_queue_head_t *cong_wq = &__get_cpu_var(fscache_object_cong_wait);
+	wait_queue_head_t *cong_wq = this_cpu_ptr(&fscache_object_cong_wait);
 	DEFINE_WAIT(wait);
 
 	if (fscache_object_congested())

[gcv v4 03/38] Subject; percpu: Add raw_cpu_ops

[Christoph Lameter]

The following patches add preemption checks to __this_cpu ops so we need
to have an alternative way to use this cpu operations without all checks.

raw_cpu ops rely on the __this_cpu_xxx_1/2/4/8 operations having no
preemption checks. We therefore do not have to duplicate these functions
but can affort to only duplicate the higher level macros.

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/include/linux/percpu.h
===================================================================
--- linux.orig/include/linux/percpu.h	2013-09-03 13:31:40.519232161 -0500
+++ linux/include/linux/percpu.h	2013-09-03 13:37:30.511691516 -0500
@@ -520,17 +520,7 @@ do {									\
 
 /*
  * Generic percpu operations for context that are safe from preemption/interrupts.
- * Either we do not care about races or the caller has the
- * responsibility of handling preemption/interrupt issues. Arch code can still
- * override these instructions since the arch per cpu code may be more
- * efficient and may actually get race freeness for free (that is the
- * case for x86 for example).
- *
- * If there is no other protection through preempt disable and/or
- * disabling interupts then one of these RMW operations can show unexpected
- * behavior because the execution thread was rescheduled on another processor
- * or an interrupt occurred and the same percpu variable was modified from
- * the interrupt context.
+ * These functions verify that preemption has been disabled.
  */
 #ifndef __this_cpu_read
 # ifndef __this_cpu_read_1
@@ -755,4 +745,74 @@ do {									\
 	__pcpu_double_call_return_bool(__this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
 #endif
 
+/*
+ * Generic percpu operations for context where we do not want to do
+ * any checks for preemptiosn.
+ *
+ * If there is no other protection through preempt disable and/or
+ * disabling interupts then one of these RMW operations can show unexpected
+ * behavior because the execution thread was rescheduled on another processor
+ * or an interrupt occurred and the same percpu variable was modified from
+ * the interrupt context.
+ */
+#ifndef raw_cpu_read
+# define raw_cpu_read(pcp)	__pcpu_size_call_return(__this_cpu_read_, (pcp))
+#endif
+
+#ifndef raw_cpu_write
+# define raw_cpu_write(pcp, val)	__pcpu_size_call(__this_cpu_write_, (pcp), (val))
+#endif
+
+#ifndef raw_cpu_add
+# define raw_cpu_add(pcp, val)	__pcpu_size_call(__this_cpu_add_, (pcp), (val))
+#endif
+
+#ifndef raw_cpu_sub
+# define raw_cpu_sub(pcp, val)	raw_cpu_add((pcp), -(val))
+#endif
+
+#ifndef raw_cpu_inc
+# define raw_cpu_inc(pcp)		raw_cpu_add((pcp), 1)
+#endif
+
+#ifndef raw_cpu_dec
+# define raw_cpu_dec(pcp)		raw_cpu_sub((pcp), 1)
+#endif
+
+#ifndef raw_cpu_and
+# define raw_cpu_and(pcp, val)	__pcpu_size_call(__this_cpu_and_, (pcp), (val))
+#endif
+
+#ifndef raw_cpu_or
+# define raw_cpu_or(pcp, val)	__pcpu_size_call(__this_cpu_or_, (pcp), (val))
+#endif
+
+#ifndef raw_cpu_xor
+# define raw_cpu_xor(pcp, val)	__pcpu_size_call(__this_cpu_xor_, (pcp), (val))
+#endif
+
+#ifndef raw_cpu_add_return
+# define raw_cpu_add_return(pcp, val)	\
+	__pcpu_size_call_return2(__this_cpu_add_return_, pcp, val)
+#endif
+
+#define raw_cpu_sub_return(pcp, val)	raw_cpu_add_return(pcp, -(val))
+#define raw_cpu_inc_return(pcp)	raw_cpu_add_return(pcp, 1)
+#define raw_cpu_dec_return(pcp)	raw_cpu_add_return(pcp, -1)
+
+#ifndef raw_cpu_xchg
+# define raw_cpu_xchg(pcp, nval)	\
+	__pcpu_size_call_return2(__this_cpu_xchg_, (pcp), nval)
+#endif
+
+#ifndef raw_cpu_cmpxchg
+# define raw_cpu_cmpxchg(pcp, oval, nval)	\
+	__pcpu_size_call_return2(__this_cpu_cmpxchg_, pcp, oval, nval)
+#endif
+
+#ifndef raw_cpu_cmpxchg_double
+# define raw_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
+	__pcpu_double_call_return_bool(__this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
+#endif
+
 #endif /* __LINUX_PERCPU_H */

[gcv v4 06/38] Coccinelle script for __get_cpu_var conversion

[Christoph Lameter]

Much of the work was done by this coccinelle script.

Wont catch everything (since coccinelle aborts when it no longer can make
sense out of macros, sigh) but it covers a lot of ground.

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/scripts/coccinelle/convert/__get_cpu_var.cocci
===================================================================
--- /dev/null	1970-01-01 00:00:00.000000000 +0000
+++ linux/scripts/coccinelle/convert/__get_cpu_var.cocci	2013-08-22 14:41:13.976518846 -0500
@@ -0,0 +1,70 @@
+// Convert all &__get_cpu_var to this_cpu_ptr
+//
+// _get_cpu_var is defined as:
+//
+// #define __get_cpu_var(var) (*this_cpu_ptr(&(var)))
+//
+//&__get_cpu_var() is therefore
+//
+//	this_cpu_ptr(&var);
+//
+// The use of this_cpu_ptr() instead of __get_cpu_var clarifies
+// that we are relocating a per cpu offset and that there are
+// no get/put semantics involved.
+//
+//
+// Convert all &__get_cpu_vars (and also related functions)
+// to this_cpu_ptr(). Multiple transformations are provided
+// to be able to beautify the source a bit.
+//
+// Christoph Lameter <cl@linux.com> August 20, 2013
+//
+
+@A@ expression E; @@
+
+...
+-(&__get_cpu_var(E))
++this_cpu_ptr(&E)
+...
+
+@B@ expression E; @@
+
+...
+-(&__raw_get_cpu_var(E))
++__this_cpu_ptr(&E)
+...
+
+@C@ expression E; @@
+
+...
+-&__get_cpu_var(E)
++this_cpu_ptr(&E)
+...
+
+@D@ expression E; @@
+
+...
+-&__raw_get_cpu_var(E)
++__this_cpu_ptr(&E)
+...
+
+@A2@ expression E; @@
+
+-(&__get_cpu_var(E))
++this_cpu_ptr(&E)
+
+@B2@ expression E; @@
+
+-(&__raw_get_cpu_var(E))
++__this_cpu_ptr(&E)
+
+@C2@ expression E; @@
+
+-&__get_cpu_var(E)
++this_cpu_ptr(&E)
+
+@D2@ expression E; @@
+
+-&__raw_get_cpu_var(E)
++__this_cpu_ptr(&E)
+

[gcv v4 13/38] rcu: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/kernel/rcutree.c
===================================================================
--- linux.orig/kernel/rcutree.c	2013-08-26 14:27:35.000000000 -0500
+++ linux/kernel/rcutree.c	2013-08-26 14:28:13.959886770 -0500
@@ -383,7 +383,7 @@ static void rcu_eqs_enter(bool user)
 	long long oldval;
 	struct rcu_dynticks *rdtp;
 
-	rdtp = &__get_cpu_var(rcu_dynticks);
+	rdtp = this_cpu_ptr(&rcu_dynticks);
 	oldval = rdtp->dynticks_nesting;
 	WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0);
 	if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE)
@@ -443,7 +443,7 @@ void rcu_user_enter_after_irq(void)
 	struct rcu_dynticks *rdtp;
 
 	local_irq_save(flags);
-	rdtp = &__get_cpu_var(rcu_dynticks);
+	rdtp = this_cpu_ptr(&rcu_dynticks);
 	/* Ensure this irq is interrupting a non-idle RCU state.  */
 	WARN_ON_ONCE(!(rdtp->dynticks_nesting & DYNTICK_TASK_MASK));
 	rdtp->dynticks_nesting = 1;
@@ -474,7 +474,7 @@ void rcu_irq_exit(void)
 	struct rcu_dynticks *rdtp;
 
 	local_irq_save(flags);
-	rdtp = &__get_cpu_var(rcu_dynticks);
+	rdtp = this_cpu_ptr(&rcu_dynticks);
 	oldval = rdtp->dynticks_nesting;
 	rdtp->dynticks_nesting--;
 	WARN_ON_ONCE(rdtp->dynticks_nesting < 0);
@@ -523,7 +523,7 @@ static void rcu_eqs_exit(bool user)
 	struct rcu_dynticks *rdtp;
 	long long oldval;
 
-	rdtp = &__get_cpu_var(rcu_dynticks);
+	rdtp = this_cpu_ptr(&rcu_dynticks);
 	oldval = rdtp->dynticks_nesting;
 	WARN_ON_ONCE(oldval < 0);
 	if (oldval & DYNTICK_TASK_NEST_MASK)
@@ -581,7 +581,7 @@ void rcu_user_exit_after_irq(void)
 	struct rcu_dynticks *rdtp;
 
 	local_irq_save(flags);
-	rdtp = &__get_cpu_var(rcu_dynticks);
+	rdtp = this_cpu_ptr(&rcu_dynticks);
 	/* Ensure we are interrupting an RCU idle mode. */
 	WARN_ON_ONCE(rdtp->dynticks_nesting & DYNTICK_TASK_NEST_MASK);
 	rdtp->dynticks_nesting += DYNTICK_TASK_EXIT_IDLE;
@@ -615,7 +615,7 @@ void rcu_irq_enter(void)
 	long long oldval;
 
 	local_irq_save(flags);
-	rdtp = &__get_cpu_var(rcu_dynticks);
+	rdtp = this_cpu_ptr(&rcu_dynticks);
 	oldval = rdtp->dynticks_nesting;
 	rdtp->dynticks_nesting++;
 	WARN_ON_ONCE(rdtp->dynticks_nesting == 0);
@@ -635,7 +635,7 @@ void rcu_irq_enter(void)
  */
 void rcu_nmi_enter(void)
 {
-	struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
+	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
 
 	if (rdtp->dynticks_nmi_nesting == 0 &&
 	    (atomic_read(&rdtp->dynticks) & 0x1))
@@ -657,7 +657,7 @@ void rcu_nmi_enter(void)
  */
 void rcu_nmi_exit(void)
 {
-	struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
+	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
 
 	if (rdtp->dynticks_nmi_nesting == 0 ||
 	    --rdtp->dynticks_nmi_nesting != 0)
@@ -680,7 +680,7 @@ int rcu_is_cpu_idle(void)
 	int ret;
 
 	preempt_disable();
-	ret = (atomic_read(&__get_cpu_var(rcu_dynticks).dynticks) & 0x1) == 0;
+	ret = (atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1) == 0;
 	preempt_enable();
 	return ret;
 }
@@ -718,7 +718,7 @@ bool rcu_lockdep_current_cpu_online(void
 	if (in_nmi())
 		return 1;
 	preempt_disable();
-	rdp = &__get_cpu_var(rcu_sched_data);
+	rdp = this_cpu_ptr(&rcu_sched_data);
 	rnp = rdp->mynode;
 	ret = (rdp->grpmask & rnp->qsmaskinit) ||
 	      !rcu_scheduler_fully_active;
@@ -738,7 +738,7 @@ EXPORT_SYMBOL_GPL(rcu_lockdep_current_cp
  */
 static int rcu_is_cpu_rrupt_from_idle(void)
 {
-	return __get_cpu_var(rcu_dynticks).dynticks_nesting <= 1;
+	return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1;
 }
 
 /*
Index: linux/kernel/rcutree_plugin.h
===================================================================
--- linux.orig/kernel/rcutree_plugin.h	2013-08-26 14:27:35.000000000 -0500
+++ linux/kernel/rcutree_plugin.h	2013-08-26 14:28:13.963886728 -0500
@@ -662,7 +662,7 @@ static void rcu_preempt_check_callbacks(
 
 static void rcu_preempt_do_callbacks(void)
 {
-	rcu_do_batch(&rcu_preempt_state, &__get_cpu_var(rcu_preempt_data));
+	rcu_do_batch(&rcu_preempt_state, this_cpu_ptr(&rcu_preempt_data));
 }
 
 #endif /* #ifdef CONFIG_RCU_BOOST */
@@ -1334,7 +1334,7 @@ static void invoke_rcu_callbacks_kthread
  */
 static bool rcu_is_callbacks_kthread(void)
 {
-	return __get_cpu_var(rcu_cpu_kthread_task) == current;
+	return __this_cpu_read(rcu_cpu_kthread_task) == current;
 }
 
 #define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
@@ -1384,8 +1384,8 @@ static int rcu_spawn_one_boost_kthread(s
 
 static void rcu_kthread_do_work(void)
 {
-	rcu_do_batch(&rcu_sched_state, &__get_cpu_var(rcu_sched_data));
-	rcu_do_batch(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
+	rcu_do_batch(&rcu_sched_state, this_cpu_ptr(&rcu_sched_data));
+	rcu_do_batch(&rcu_bh_state, this_cpu_ptr(&rcu_bh_data));
 	rcu_preempt_do_callbacks();
 }
 
@@ -1404,7 +1404,7 @@ static void rcu_cpu_kthread_park(unsigne
 
 static int rcu_cpu_kthread_should_run(unsigned int cpu)
 {
-	return __get_cpu_var(rcu_cpu_has_work);
+	return __this_cpu_read(rcu_cpu_has_work);
 }
 
 /*
@@ -1414,8 +1414,8 @@ static int rcu_cpu_kthread_should_run(un
  */
 static void rcu_cpu_kthread(unsigned int cpu)
 {
-	unsigned int *statusp = &__get_cpu_var(rcu_cpu_kthread_status);
-	char work, *workp = &__get_cpu_var(rcu_cpu_has_work);
+	unsigned int *statusp = this_cpu_ptr(&rcu_cpu_kthread_status);
+	char work, *workp = this_cpu_ptr(&rcu_cpu_has_work);
 	int spincnt;
 
 	for (spincnt = 0; spincnt < 10; spincnt++) {

[gcv v4 17/38] drivers/char: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)



Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/drivers/char/random.c
===================================================================
--- linux.orig/drivers/char/random.c	2013-08-27 14:31:00.000000000 -0500
+++ linux/drivers/char/random.c	2013-08-27 14:31:34.275916299 -0500
@@ -744,7 +744,7 @@ static DEFINE_PER_CPU(struct fast_pool,
 void add_interrupt_randomness(int irq, int irq_flags)
 {
 	struct entropy_store	*r;
-	struct fast_pool	*fast_pool = &__get_cpu_var(irq_randomness);
+	struct fast_pool	*fast_pool = this_cpu_ptr(&irq_randomness);
 	struct pt_regs		*regs = get_irq_regs();
 	unsigned long		now = jiffies;
 	__u32			input[4], cycles = get_cycles();

[gcv v4 16/38] kernel misc: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)


Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/kernel/printk/printk.c
===================================================================
--- linux.orig/kernel/printk/printk.c	2013-08-26 14:30:25.846490787 -0500
+++ linux/kernel/printk/printk.c	2013-08-26 14:30:25.846490787 -0500
@@ -2441,7 +2441,7 @@ static void wake_up_klogd_work_func(stru
 	int pending = __this_cpu_xchg(printk_pending, 0);
 
 	if (pending & PRINTK_PENDING_SCHED) {
-		char *buf = __get_cpu_var(printk_sched_buf);
+		char *buf = this_cpu_ptr(printk_sched_buf);
 		printk(KERN_WARNING "[sched_delayed] %s", buf);
 	}
 
@@ -2459,7 +2459,7 @@ void wake_up_klogd(void)
 	preempt_disable();
 	if (waitqueue_active(&log_wait)) {
 		this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
-		irq_work_queue(&__get_cpu_var(wake_up_klogd_work));
+		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
 	}
 	preempt_enable();
 }
@@ -2472,14 +2472,14 @@ int printk_sched(const char *fmt, ...)
 	int r;
 
 	local_irq_save(flags);
-	buf = __get_cpu_var(printk_sched_buf);
+	buf = this_cpu_ptr(printk_sched_buf);
 
 	va_start(args, fmt);
 	r = vsnprintf(buf, PRINTK_BUF_SIZE, fmt, args);
 	va_end(args);
 
 	__this_cpu_or(printk_pending, PRINTK_PENDING_SCHED);
-	irq_work_queue(&__get_cpu_var(wake_up_klogd_work));
+	irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
 	local_irq_restore(flags);
 
 	return r;
Index: linux/kernel/smp.c
===================================================================
--- linux.orig/kernel/smp.c	2013-08-26 14:30:25.846490787 -0500
+++ linux/kernel/smp.c	2013-08-26 14:30:25.846490787 -0500
@@ -172,7 +172,7 @@ void generic_exec_single(int cpu, struct
  */
 void generic_smp_call_function_single_interrupt(void)
 {
-	struct call_single_queue *q = &__get_cpu_var(call_single_queue);
+	struct call_single_queue *q = this_cpu_ptr(&call_single_queue);
 	LIST_HEAD(list);
 
 	/*
@@ -252,7 +252,7 @@ int smp_call_function_single(int cpu, sm
 			struct call_single_data *csd = &d;
 
 			if (!wait)
-				csd = &__get_cpu_var(csd_data);
+				csd = this_cpu_ptr(&csd_data);
 
 			csd_lock(csd);
 
@@ -401,7 +401,7 @@ void smp_call_function_many(const struct
 		return;
 	}
 
-	cfd = &__get_cpu_var(cfd_data);
+	cfd = this_cpu_ptr(&cfd_data);
 
 	cpumask_and(cfd->cpumask, mask, cpu_online_mask);
 	cpumask_clear_cpu(this_cpu, cfd->cpumask);

[gcv v4 10/38] mm: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/lib/radix-tree.c
===================================================================
--- linux.orig/lib/radix-tree.c	2013-08-26 14:24:48.000000000 -0500
+++ linux/lib/radix-tree.c	2013-08-26 14:25:30.709616290 -0500
@@ -215,7 +215,7 @@ radix_tree_node_alloc(struct radix_tree_
 		 * succeed in getting a node here (and never reach
 		 * kmem_cache_alloc)
 		 */
-		rtp = &__get_cpu_var(radix_tree_preloads);
+		rtp = this_cpu_ptr(&radix_tree_preloads);
 		if (rtp->nr) {
 			ret = rtp->nodes[rtp->nr - 1];
 			rtp->nodes[rtp->nr - 1] = NULL;
@@ -271,14 +271,14 @@ int radix_tree_preload(gfp_t gfp_mask)
 	int ret = -ENOMEM;
 
 	preempt_disable();
-	rtp = &__get_cpu_var(radix_tree_preloads);
+	rtp = this_cpu_ptr(&radix_tree_preloads);
 	while (rtp->nr < ARRAY_SIZE(rtp->nodes)) {
 		preempt_enable();
 		node = kmem_cache_alloc(radix_tree_node_cachep, gfp_mask);
 		if (node == NULL)
 			goto out;
 		preempt_disable();
-		rtp = &__get_cpu_var(radix_tree_preloads);
+		rtp = this_cpu_ptr(&radix_tree_preloads);
 		if (rtp->nr < ARRAY_SIZE(rtp->nodes))
 			rtp->nodes[rtp->nr++] = node;
 		else
Index: linux/mm/memcontrol.c
===================================================================
--- linux.orig/mm/memcontrol.c	2013-08-26 14:24:48.000000000 -0500
+++ linux/mm/memcontrol.c	2013-08-26 14:25:30.713616248 -0500
@@ -2398,7 +2398,7 @@ static void drain_stock(struct memcg_sto
  */
 static void drain_local_stock(struct work_struct *dummy)
 {
-	struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock);
+	struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock);
 	drain_stock(stock);
 	clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags);
 }
Index: linux/mm/memory-failure.c
===================================================================
--- linux.orig/mm/memory-failure.c	2013-08-26 14:24:48.000000000 -0500
+++ linux/mm/memory-failure.c	2013-08-26 14:25:30.713616248 -0500
@@ -1279,7 +1279,7 @@ static void memory_failure_work_func(str
 	unsigned long proc_flags;
 	int gotten;
 
-	mf_cpu = &__get_cpu_var(memory_failure_cpu);
+	mf_cpu = this_cpu_ptr(&memory_failure_cpu);
 	for (;;) {
 		spin_lock_irqsave(&mf_cpu->lock, proc_flags);
 		gotten = kfifo_get(&mf_cpu->fifo, &entry);
Index: linux/mm/page-writeback.c
===================================================================
--- linux.orig/mm/page-writeback.c	2013-08-26 14:24:48.000000000 -0500
+++ linux/mm/page-writeback.c	2013-08-26 14:25:30.713616248 -0500
@@ -1487,7 +1487,7 @@ void balance_dirty_pages_ratelimited(str
 	 * 1000+ tasks, all of them start dirtying pages at exactly the same
 	 * time, hence all honoured too large initial task->nr_dirtied_pause.
 	 */
-	p =  &__get_cpu_var(bdp_ratelimits);
+	p =  this_cpu_ptr(&bdp_ratelimits);
 	if (unlikely(current->nr_dirtied >= ratelimit))
 		*p = 0;
 	else if (unlikely(*p >= ratelimit_pages)) {
@@ -1499,7 +1499,7 @@ void balance_dirty_pages_ratelimited(str
 	 * short-lived tasks (eg. gcc invocations in a kernel build) escaping
 	 * the dirty throttling and livelock other long-run dirtiers.
 	 */
-	p = &__get_cpu_var(dirty_throttle_leaks);
+	p = this_cpu_ptr(&dirty_throttle_leaks);
 	if (*p > 0 && current->nr_dirtied < ratelimit) {
 		unsigned long nr_pages_dirtied;
 		nr_pages_dirtied = min(*p, ratelimit - current->nr_dirtied);
Index: linux/mm/swap.c
===================================================================
--- linux.orig/mm/swap.c	2013-08-26 14:24:48.000000000 -0500
+++ linux/mm/swap.c	2013-08-26 14:25:30.717616206 -0500
@@ -359,7 +359,7 @@ void rotate_reclaimable_page(struct page
 
 		page_cache_get(page);
 		local_irq_save(flags);
-		pvec = &__get_cpu_var(lru_rotate_pvecs);
+		pvec = this_cpu_ptr(&lru_rotate_pvecs);
 		if (!pagevec_add(pvec, page))
 			pagevec_move_tail(pvec);
 		local_irq_restore(flags);
Index: linux/mm/vmalloc.c
===================================================================
--- linux.orig/mm/vmalloc.c	2013-08-26 14:24:48.000000000 -0500
+++ linux/mm/vmalloc.c	2013-08-26 14:25:30.717616206 -0500
@@ -1487,7 +1487,7 @@ void vfree(const void *addr)
 	if (!addr)
 		return;
 	if (unlikely(in_interrupt())) {
-		struct vfree_deferred *p = &__get_cpu_var(vfree_deferred);
+		struct vfree_deferred *p = this_cpu_ptr(&vfree_deferred);
 		if (llist_add((struct llist_node *)addr, &p->list))
 			schedule_work(&p->wq);
 	} else
Index: linux/mm/vmstat.c
===================================================================
--- linux.orig/mm/vmstat.c	2013-08-26 14:24:48.000000000 -0500
+++ linux/mm/vmstat.c	2013-08-26 14:25:30.717616206 -0500
@@ -1178,7 +1178,7 @@ int sysctl_stat_interval __read_mostly =
 static void vmstat_update(struct work_struct *w)
 {
 	refresh_cpu_vm_stats(smp_processor_id());
-	schedule_delayed_work(&__get_cpu_var(vmstat_work),
+	schedule_delayed_work(this_cpu_ptr(&vmstat_work),
 		round_jiffies_relative(sysctl_stat_interval));
 }
 

[gcv v4 08/38] time: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/kernel/hrtimer.c
===================================================================
--- linux.orig/kernel/hrtimer.c	2013-08-26 14:19:14.000000000 -0500
+++ linux/kernel/hrtimer.c	2013-08-26 14:21:24.540227691 -0500
@@ -597,7 +597,7 @@ hrtimer_force_reprogram(struct hrtimer_c
 static int hrtimer_reprogram(struct hrtimer *timer,
 			     struct hrtimer_clock_base *base)
 {
-	struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
+	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
 	ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
 	int res;
 
@@ -680,7 +680,7 @@ static inline ktime_t hrtimer_update_bas
  */
 static void retrigger_next_event(void *arg)
 {
-	struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
+	struct hrtimer_cpu_base *base = this_cpu_ptr(&hrtimer_bases);
 
 	if (!hrtimer_hres_active())
 		return;
@@ -954,7 +954,7 @@ remove_hrtimer(struct hrtimer *timer, st
 		 */
 		debug_deactivate(timer);
 		timer_stats_hrtimer_clear_start_info(timer);
-		reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
+		reprogram = base->cpu_base == this_cpu_ptr(&hrtimer_bases);
 		/*
 		 * We must preserve the CALLBACK state flag here,
 		 * otherwise we could move the timer base in
@@ -1009,7 +1009,7 @@ int __hrtimer_start_range_ns(struct hrti
 	 *
 	 * XXX send_remote_softirq() ?
 	 */
-	if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases)
+	if (leftmost && new_base->cpu_base == this_cpu_ptr(&hrtimer_bases)
 		&& hrtimer_enqueue_reprogram(timer, new_base)) {
 		if (wakeup) {
 			/*
@@ -1142,7 +1142,7 @@ EXPORT_SYMBOL_GPL(hrtimer_get_remaining)
  */
 ktime_t hrtimer_get_next_event(void)
 {
-	struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
+	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
 	struct hrtimer_clock_base *base = cpu_base->clock_base;
 	ktime_t delta, mindelta = { .tv64 = KTIME_MAX };
 	unsigned long flags;
@@ -1183,7 +1183,7 @@ static void __hrtimer_init(struct hrtime
 
 	memset(timer, 0, sizeof(struct hrtimer));
 
-	cpu_base = &__raw_get_cpu_var(hrtimer_bases);
+	cpu_base = __this_cpu_ptr(&hrtimer_bases);
 
 	if (clock_id == CLOCK_REALTIME && mode != HRTIMER_MODE_ABS)
 		clock_id = CLOCK_MONOTONIC;
@@ -1226,7 +1226,7 @@ int hrtimer_get_res(const clockid_t whic
 	struct hrtimer_cpu_base *cpu_base;
 	int base = hrtimer_clockid_to_base(which_clock);
 
-	cpu_base = &__raw_get_cpu_var(hrtimer_bases);
+	cpu_base = __this_cpu_ptr(&hrtimer_bases);
 	*tp = ktime_to_timespec(cpu_base->clock_base[base].resolution);
 
 	return 0;
@@ -1281,7 +1281,7 @@ static void __run_hrtimer(struct hrtimer
  */
 void hrtimer_interrupt(struct clock_event_device *dev)
 {
-	struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
+	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
 	ktime_t expires_next, now, entry_time, delta;
 	int i, retries = 0;
 
@@ -1415,7 +1415,7 @@ static void __hrtimer_peek_ahead_timers(
 	if (!hrtimer_hres_active())
 		return;
 
-	td = &__get_cpu_var(tick_cpu_device);
+	td = this_cpu_ptr(&tick_cpu_device);
 	if (td && td->evtdev)
 		hrtimer_interrupt(td->evtdev);
 }
@@ -1479,7 +1479,7 @@ void hrtimer_run_pending(void)
 void hrtimer_run_queues(void)
 {
 	struct timerqueue_node *node;
-	struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
+	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
 	struct hrtimer_clock_base *base;
 	int index, gettime = 1;
 
@@ -1717,7 +1717,7 @@ static void migrate_hrtimers(int scpu)
 
 	local_irq_disable();
 	old_base = &per_cpu(hrtimer_bases, scpu);
-	new_base = &__get_cpu_var(hrtimer_bases);
+	new_base = this_cpu_ptr(&hrtimer_bases);
 	/*
 	 * The caller is globally serialized and nobody else
 	 * takes two locks at once, deadlock is not possible.
Index: linux/kernel/irq_work.c
===================================================================
--- linux.orig/kernel/irq_work.c	2013-08-26 14:19:14.000000000 -0500
+++ linux/kernel/irq_work.c	2013-08-26 14:21:24.540227691 -0500
@@ -70,7 +70,7 @@ void irq_work_queue(struct irq_work *wor
 	/* Queue the entry and raise the IPI if needed. */
 	preempt_disable();
 
-	llist_add(&work->llnode, &__get_cpu_var(irq_work_list));
+	llist_add(&work->llnode, this_cpu_ptr(&irq_work_list));
 
 	/*
 	 * If the work is not "lazy" or the tick is stopped, raise the irq
@@ -90,7 +90,7 @@ bool irq_work_needs_cpu(void)
 {
 	struct llist_head *this_list;
 
-	this_list = &__get_cpu_var(irq_work_list);
+	this_list = this_cpu_ptr(&irq_work_list);
 	if (llist_empty(this_list))
 		return false;
 
@@ -115,7 +115,7 @@ static void __irq_work_run(void)
 	__this_cpu_write(irq_work_raised, 0);
 	barrier();
 
-	this_list = &__get_cpu_var(irq_work_list);
+	this_list = this_cpu_ptr(&irq_work_list);
 	if (llist_empty(this_list))
 		return;
 
Index: linux/kernel/sched/clock.c
===================================================================
--- linux.orig/kernel/sched/clock.c	2013-08-26 14:19:14.000000000 -0500
+++ linux/kernel/sched/clock.c	2013-08-26 14:21:24.540227691 -0500
@@ -94,7 +94,7 @@ static DEFINE_PER_CPU_SHARED_ALIGNED(str
 
 static inline struct sched_clock_data *this_scd(void)
 {
-	return &__get_cpu_var(sched_clock_data);
+	return this_cpu_ptr(&sched_clock_data);
 }
 
 static inline struct sched_clock_data *cpu_sdc(int cpu)
Index: linux/kernel/softirq.c
===================================================================
--- linux.orig/kernel/softirq.c	2013-08-26 14:19:14.000000000 -0500
+++ linux/kernel/softirq.c	2013-08-26 14:21:24.540227691 -0500
@@ -466,7 +466,7 @@ static void tasklet_action(struct softir
 	local_irq_disable();
 	list = __this_cpu_read(tasklet_vec.head);
 	__this_cpu_write(tasklet_vec.head, NULL);
-	__this_cpu_write(tasklet_vec.tail, &__get_cpu_var(tasklet_vec).head);
+	__this_cpu_write(tasklet_vec.tail, this_cpu_ptr(&tasklet_vec.head));
 	local_irq_enable();
 
 	while (list) {
@@ -501,7 +501,7 @@ static void tasklet_hi_action(struct sof
 	local_irq_disable();
 	list = __this_cpu_read(tasklet_hi_vec.head);
 	__this_cpu_write(tasklet_hi_vec.head, NULL);
-	__this_cpu_write(tasklet_hi_vec.tail, &__get_cpu_var(tasklet_hi_vec).head);
+	__this_cpu_write(tasklet_hi_vec.tail, this_cpu_ptr(&tasklet_hi_vec.head));
 	local_irq_enable();
 
 	while (list) {
@@ -618,7 +618,7 @@ EXPORT_PER_CPU_SYMBOL(softirq_work_list)
 
 static void __local_trigger(struct call_single_data *cp, int softirq)
 {
-	struct list_head *head = &__get_cpu_var(softirq_work_list[softirq]);
+	struct list_head *head = this_cpu_ptr(&softirq_work_list[softirq]);
 
 	list_add_tail(&cp->list, head);
 
@@ -718,7 +718,7 @@ static int remote_softirq_cpu_notify(str
 			if (list_empty(head))
 				continue;
 
-			local_head = &__get_cpu_var(softirq_work_list[i]);
+			local_head = this_cpu_ptr(&softirq_work_list[i]);
 			list_splice_init(head, local_head);
 			raise_softirq_irqoff(i);
 		}
Index: linux/kernel/time/tick-common.c
===================================================================
--- linux.orig/kernel/time/tick-common.c	2013-08-26 14:19:14.000000000 -0500
+++ linux/kernel/time/tick-common.c	2013-08-26 14:21:24.544227649 -0500
@@ -208,7 +208,7 @@ static void tick_setup_device(struct tic
 
 void tick_install_replacement(struct clock_event_device *newdev)
 {
-	struct tick_device *td = &__get_cpu_var(tick_cpu_device);
+	struct tick_device *td = this_cpu_ptr(&tick_cpu_device);
 	int cpu = smp_processor_id();
 
 	clockevents_exchange_device(td->evtdev, newdev);
@@ -358,14 +358,14 @@ void tick_shutdown(unsigned int *cpup)
 
 void tick_suspend(void)
 {
-	struct tick_device *td = &__get_cpu_var(tick_cpu_device);
+	struct tick_device *td = this_cpu_ptr(&tick_cpu_device);
 
 	clockevents_shutdown(td->evtdev);
 }
 
 void tick_resume(void)
 {
-	struct tick_device *td = &__get_cpu_var(tick_cpu_device);
+	struct tick_device *td = this_cpu_ptr(&tick_cpu_device);
 	int broadcast = tick_resume_broadcast();
 
 	clockevents_set_mode(td->evtdev, CLOCK_EVT_MODE_RESUME);
Index: linux/kernel/time/tick-oneshot.c
===================================================================
--- linux.orig/kernel/time/tick-oneshot.c	2013-08-26 14:19:14.000000000 -0500
+++ linux/kernel/time/tick-oneshot.c	2013-08-26 14:21:24.544227649 -0500
@@ -59,7 +59,7 @@ void tick_setup_oneshot(struct clock_eve
  */
 int tick_switch_to_oneshot(void (*handler)(struct clock_event_device *))
 {
-	struct tick_device *td = &__get_cpu_var(tick_cpu_device);
+	struct tick_device *td = this_cpu_ptr(&tick_cpu_device);
 	struct clock_event_device *dev = td->evtdev;
 
 	if (!dev || !(dev->features & CLOCK_EVT_FEAT_ONESHOT) ||
Index: linux/kernel/time/tick-sched.c
===================================================================
--- linux.orig/kernel/time/tick-sched.c	2013-08-26 14:19:14.000000000 -0500
+++ linux/kernel/time/tick-sched.c	2013-08-26 14:21:24.540227691 -0500
@@ -199,7 +199,7 @@ static void tick_nohz_restart_sched_tick
  */
 void tick_nohz_full_check(void)
 {
-	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
+	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
 	if (tick_nohz_full_cpu(smp_processor_id())) {
 		if (ts->tick_stopped && !is_idle_task(current)) {
@@ -225,7 +225,7 @@ static DEFINE_PER_CPU(struct irq_work, n
 void tick_nohz_full_kick(void)
 {
 	if (tick_nohz_full_cpu(smp_processor_id()))
-		irq_work_queue(&__get_cpu_var(nohz_full_kick_work));
+		irq_work_queue(this_cpu_ptr(&nohz_full_kick_work));
 }
 
 static void nohz_full_kick_ipi(void *info)
@@ -536,7 +536,7 @@ static ktime_t tick_nohz_stop_sched_tick
 	unsigned long seq, last_jiffies, next_jiffies, delta_jiffies;
 	ktime_t last_update, expires, ret = { .tv64 = 0 };
 	unsigned long rcu_delta_jiffies;
-	struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev;
+	struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev);
 	u64 time_delta;
 
 	/* Read jiffies and the time when jiffies were updated last */
@@ -801,7 +801,7 @@ void tick_nohz_idle_enter(void)
 
 	local_irq_disable();
 
-	ts = &__get_cpu_var(tick_cpu_sched);
+	ts = this_cpu_ptr(&tick_cpu_sched);
 	/*
 	 * set ts->inidle unconditionally. even if the system did not
 	 * switch to nohz mode the cpu frequency governers rely on the
@@ -824,7 +824,7 @@ EXPORT_SYMBOL_GPL(tick_nohz_idle_enter);
  */
 void tick_nohz_irq_exit(void)
 {
-	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
+	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
 	if (ts->inidle)
 		__tick_nohz_idle_enter(ts);
@@ -839,7 +839,7 @@ void tick_nohz_irq_exit(void)
  */
 ktime_t tick_nohz_get_sleep_length(void)
 {
-	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
+	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
 	return ts->sleep_length;
 }
@@ -953,7 +953,7 @@ static int tick_nohz_reprogram(struct ti
  */
 static void tick_nohz_handler(struct clock_event_device *dev)
 {
-	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
+	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 	struct pt_regs *regs = get_irq_regs();
 	ktime_t now = ktime_get();
 
@@ -973,7 +973,7 @@ static void tick_nohz_handler(struct clo
  */
 static void tick_nohz_switch_to_nohz(void)
 {
-	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
+	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 	ktime_t next;
 
 	if (!tick_nohz_enabled)
@@ -1111,7 +1111,7 @@ early_param("skew_tick", skew_tick);
  */
 void tick_setup_sched_timer(void)
 {
-	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
+	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 	ktime_t now = ktime_get();
 
 	/*
@@ -1178,7 +1178,7 @@ void tick_clock_notify(void)
  */
 void tick_oneshot_notify(void)
 {
-	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
+	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
 	set_bit(0, &ts->check_clocks);
 }
@@ -1193,7 +1193,7 @@ void tick_oneshot_notify(void)
  */
 int tick_check_oneshot_change(int allow_nohz)
 {
-	struct tick_sched *ts = &__get_cpu_var(tick_cpu_sched);
+	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
 	if (!test_and_clear_bit(0, &ts->check_clocks))
 		return 0;
Index: linux/kernel/timer.c
===================================================================
--- linux.orig/kernel/timer.c	2013-08-26 14:19:14.000000000 -0500
+++ linux/kernel/timer.c	2013-08-26 14:21:24.544227649 -0500
@@ -621,7 +621,7 @@ static inline void debug_assert_init(str
 static void do_init_timer(struct timer_list *timer, unsigned int flags,
 			  const char *name, struct lock_class_key *key)
 {
-	struct tvec_base *base = __raw_get_cpu_var(tvec_bases);
+	struct tvec_base *base = __this_cpu_read(tvec_bases);
 
 	timer->entry.next = NULL;
 	timer->base = (void *)((unsigned long)base | flags);

[gcv v4 14/38] percpu: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/include/linux/percpu.h
===================================================================
--- linux.orig/include/linux/percpu.h	2013-08-26 14:29:49.000000000 -0500
+++ linux/include/linux/percpu.h	2013-08-26 14:30:18.302570608 -0500
@@ -28,7 +28,7 @@
  */
 #define get_cpu_var(var) (*({				\
 	preempt_disable();				\
-	&__get_cpu_var(var); }))
+	this_cpu_ptr(&var); }))
 
 /*
  * The weird & is necessary because sparse considers (void)(var) to be

[gcv v4 11/38] tracing: Replace __get_cpu_var uses

[Christoph Lameter]

__get_cpu_var() is used for multiple purposes in the kernel source. One of them is
address calculation via the form &__get_cpu_var(x). This calculates the address for
the instance of the percpu variable of the current processor based on an offset.

Other use cases are for storing and retrieving data from the current processors percpu area.
__get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment.

__get_cpu_var() is defined as :


#define __get_cpu_var(var) (*this_cpu_ptr(&(var)))



__get_cpu_var() always only does an address determination. However, store and retrieve operations
could use a segment prefix (or global register on other platforms) to avoid the address calculation.

this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use
optimized assembly code to read and write per cpu variables.


This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr()
or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided
and less registers are used when code is generated.

At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too.

The patch set includes passes over all arches as well. Once these operations are used throughout then
specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by
f.e. using a global register that may be set to the per cpu base.




Transformations done to __get_cpu_var()


1. Determine the address of the percpu instance of the current processor.

	DEFINE_PER_CPU(int, y);
	int *x = &__get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(&y);


2. Same as #1 but this time an array structure is involved.

	DEFINE_PER_CPU(int, y[20]);
	int *x = __get_cpu_var(y);

    Converts to

	int *x = this_cpu_ptr(y);


3. Retrieve the content of the current processors instance of a per cpu variable.

	DEFINE_PER_CPU(int, y);
	int x = __get_cpu_var(y)

   Converts to

	int x = __this_cpu_read(y);


4. Retrieve the content of a percpu struct

	DEFINE_PER_CPU(struct mystruct, y);
	struct mystruct x = __get_cpu_var(y);

   Converts to

	memcpy(&x, this_cpu_ptr(&y), sizeof(x));


5. Assignment to a per cpu variable

	DEFINE_PER_CPU(int, y)
	__get_cpu_var(y) = x;

   Converts to

	this_cpu_write(y, x);


6. Increment/Decrement etc of a per cpu variable

	DEFINE_PER_CPU(int, y);
	__get_cpu_var(y)++

   Converts to

	this_cpu_inc(y)


Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/include/linux/kprobes.h
===================================================================
--- linux.orig/include/linux/kprobes.h	2013-08-26 14:25:53.000000000 -0500
+++ linux/include/linux/kprobes.h	2013-08-26 14:26:29.460993659 -0500
@@ -329,7 +329,7 @@ static inline void reset_current_kprobe(
 
 static inline struct kprobe_ctlblk *get_kprobe_ctlblk(void)
 {
-	return (&__get_cpu_var(kprobe_ctlblk));
+	return this_cpu_ptr(&kprobe_ctlblk);
 }
 
 int register_kprobe(struct kprobe *p);
Index: linux/kernel/trace/ftrace.c
===================================================================
--- linux.orig/kernel/trace/ftrace.c	2013-08-26 14:25:53.000000000 -0500
+++ linux/kernel/trace/ftrace.c	2013-08-26 14:26:29.460993659 -0500
@@ -870,7 +870,7 @@ function_profile_call(unsigned long ip,
 
 	local_irq_save(flags);
 
-	stat = &__get_cpu_var(ftrace_profile_stats);
+	stat = this_cpu_ptr(&ftrace_profile_stats);
 	if (!stat->hash || !ftrace_profile_enabled)
 		goto out;
 
@@ -901,7 +901,7 @@ static void profile_graph_return(struct
 	unsigned long flags;
 
 	local_irq_save(flags);
-	stat = &__get_cpu_var(ftrace_profile_stats);
+	stat = this_cpu_ptr(&ftrace_profile_stats);
 	if (!stat->hash || !ftrace_profile_enabled)
 		goto out;
 
Index: linux/kernel/trace/trace.c
===================================================================
--- linux.orig/kernel/trace/trace.c	2013-08-26 14:25:53.000000000 -0500
+++ linux/kernel/trace/trace.c	2013-08-26 14:26:29.464993617 -0500
@@ -1676,7 +1676,7 @@ static void __ftrace_trace_stack(struct
 	 */
 	barrier();
 	if (use_stack == 1) {
-		trace.entries		= &__get_cpu_var(ftrace_stack).calls[0];
+		trace.entries		= this_cpu_ptr(ftrace_stack.calls);
 		trace.max_entries	= FTRACE_STACK_MAX_ENTRIES;
 
 		if (regs)

[gcv v4 05/38] percpu: Add preemption checks to __this_cpu ops

[Christoph Lameter]

We define a check function in order to avoid trouble with the
include files. Then the higher level __this_cpu macros are
modified to involve the check before any operation.

Signed-off-by: Christoph Lameter <cl@linux.com>

Index: linux/include/linux/percpu.h
===================================================================
--- linux.orig/include/linux/percpu.h	2013-09-03 13:38:49.818888738 -0500
+++ linux/include/linux/percpu.h	2013-09-03 13:38:49.810888819 -0500
@@ -172,6 +172,12 @@ extern phys_addr_t per_cpu_ptr_to_phys(v
 
 extern void __bad_size_call_parameter(void);
 
+#ifdef CONFIG_PREEMPT
+extern void this_cpu_preempt_check(void);
+#else
+static inline void this_cpu_preempt_check(void) { }
+#endif
+
 #define __pcpu_size_call_return(stem, variable)				\
 ({	typeof(variable) pscr_ret__;					\
 	__verify_pcpu_ptr(&(variable));					\
@@ -556,7 +562,7 @@ do {									\
 # ifndef __this_cpu_write_8
 #  define __this_cpu_write_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), =)
 # endif
-# define __this_cpu_write(pcp, val)	__pcpu_size_call(__this_cpu_write_, (pcp), (val))
+# define __this_cpu_write(pcp, val)	do { this_cpu_preempt_check();__pcpu_size_call(__this_cpu_write_, (pcp), (val)); } while (0)
 #endif
 
 #ifndef __this_cpu_add
@@ -572,7 +578,7 @@ do {									\
 # ifndef __this_cpu_add_8
 #  define __this_cpu_add_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), +=)
 # endif
-# define __this_cpu_add(pcp, val)	__pcpu_size_call(__this_cpu_add_, (pcp), (val))
+# define __this_cpu_add(pcp, val)	do { this_cpu_preempt_check();__pcpu_size_call(__this_cpu_add_, (pcp), (val)); } while (0)
 #endif
 
 #ifndef __this_cpu_sub
@@ -600,7 +606,7 @@ do {									\
 # ifndef __this_cpu_and_8
 #  define __this_cpu_and_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), &=)
 # endif
-# define __this_cpu_and(pcp, val)	__pcpu_size_call(__this_cpu_and_, (pcp), (val))
+# define __this_cpu_and(pcp, val)	do { this_cpu_preempt_check();__pcpu_size_call(__this_cpu_and_, (pcp), (val)); } while (0)
 #endif
 
 #ifndef __this_cpu_or
@@ -616,7 +622,7 @@ do {									\
 # ifndef __this_cpu_or_8
 #  define __this_cpu_or_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), |=)
 # endif
-# define __this_cpu_or(pcp, val)	__pcpu_size_call(__this_cpu_or_, (pcp), (val))
+# define __this_cpu_or(pcp, val)	do { this_cpu_preempt_check();__pcpu_size_call(__this_cpu_or_, (pcp), (val)); } while (0)
 #endif
 
 #ifndef __this_cpu_xor
@@ -632,7 +638,7 @@ do {									\
 # ifndef __this_cpu_xor_8
 #  define __this_cpu_xor_8(pcp, val)	__this_cpu_generic_to_op((pcp), (val), ^=)
 # endif
-# define __this_cpu_xor(pcp, val)	__pcpu_size_call(__this_cpu_xor_, (pcp), (val))
+# define __this_cpu_xor(pcp, val)	do { this_cpu_preempt_check();__pcpu_size_call(__this_cpu_xor_, (pcp), (val)); } while (0)
 #endif
 
 #define __this_cpu_generic_add_return(pcp, val)				\
@@ -655,7 +661,7 @@ do {									\
 #  define __this_cpu_add_return_8(pcp, val)	__this_cpu_generic_add_return(pcp, val)
 # endif
 # define __this_cpu_add_return(pcp, val)	\
-	__pcpu_size_call_return2(__this_cpu_add_return_, pcp, val)
+	(this_cpu_preempt_check(),__pcpu_size_call_return2(__this_cpu_add_return_, pcp, val))
 #endif
 
 #define __this_cpu_sub_return(pcp, val)	__this_cpu_add_return(pcp, -(val))
@@ -683,7 +689,7 @@ do {									\
 #  define __this_cpu_xchg_8(pcp, nval)	__this_cpu_generic_xchg(pcp, nval)
 # endif
 # define __this_cpu_xchg(pcp, nval)	\
-	__pcpu_size_call_return2(__this_cpu_xchg_, (pcp), nval)
+	(this_cpu_preempt_check(),__pcpu_size_call_return2(__this_cpu_xchg_, (pcp), nval))
 #endif
 
 #define __this_cpu_generic_cmpxchg(pcp, oval, nval)			\
@@ -709,7 +715,7 @@ do {									\
 #  define __this_cpu_cmpxchg_8(pcp, oval, nval)	__this_cpu_generic_cmpxchg(pcp, oval, nval)
 # endif
 # define __this_cpu_cmpxchg(pcp, oval, nval)	\
-	__pcpu_size_call_return2(__this_cpu_cmpxchg_, pcp, oval, nval)
+	(this_cpu_preempt_check(),__pcpu_size_call_return2(__this_cpu_cmpxchg_, pcp, oval, nval))
 #endif
 
 #define __this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
@@ -742,7 +748,7 @@ do {									\
 	__this_cpu_generic_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)
 # endif
 # define __this_cpu_cmpxchg_double(pcp1, pcp2, oval1, oval2, nval1, nval2)	\
-	__pcpu_double_call_return_bool(__this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2))
+	(this_cpu_preempt_check(),__pcpu_double_call_return_bool(__this_cpu_cmpxchg_double_, (pcp1), (pcp2), (oval1), (oval2), (nval1), (nval2)))
 #endif
 
 /*
Index: linux/kernel/sched/core.c
===================================================================
--- linux.orig/kernel/sched/core.c	2013-09-03 13:38:49.818888738 -0500
+++ linux/kernel/sched/core.c	2013-09-03 13:38:49.814888779 -0500
@@ -2583,6 +2583,27 @@ asmlinkage void __sched preempt_schedule
 	exception_exit(prev_state);
 }
 
+/*
+ * This function is called if the kernel is compiled with preempt
+ * support for each __this_cpu operations. It verifies that
+ * preemption has been disabled.
+ *
+ * The function cannot be a macro due to the low level nature
+ * of the per cpu header files.
+ */
+void this_cpu_preempt_check(void)
+{
+	int p;
+
+	p = preemptible();
+	if (p) {
+		printk(KERN_ERR "__this_cpu but preempt is off."
+			" Preempt count=%d Interrupts=%d\n",
+			preempt_count(), irqs_disabled());
+		dump_stack();
+	}
+
+}
 #endif /* CONFIG_PREEMPT */
 
 int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags,

Re: [gcv v4 05/38] percpu: Add preemption checks to __this_cpu ops

[Peter Zijlstra]

On Tue, Sep 03, 2013 at 10:33:20PM +0000, Christoph Lameter wrote:
> We define a check function in order to avoid trouble with the
> include files. Then the higher level __this_cpu macros are
> modified to involve the check before any operation.
> 

So this_cpu_ptr() is the one with the check, and __this_cpu_ptr() is the
one without. But for the other this_cpu ops __this_cpu_$OP() is going to
be the one with a check and this_cpu_$OP() the one without?

Sounds like a bloody marvelous idea :/



> Index: linux/include/linux/percpu.h
> ===================================================================
> --- linux.orig/include/linux/percpu.h	2013-09-03 13:38:49.818888738 -0500
> +++ linux/include/linux/percpu.h	2013-09-03 13:38:49.810888819 -0500
> @@ -172,6 +172,12 @@ extern phys_addr_t per_cpu_ptr_to_phys(v
>  
>  extern void __bad_size_call_parameter(void);
>  
> +#ifdef CONFIG_PREEMPT
> +extern void this_cpu_preempt_check(void);
> +#else
> +static inline void this_cpu_preempt_check(void) { }
> +#endif

How about re-using debug_smp_processor_id() instead?

Re: [gcv v4 05/38] percpu: Add preemption checks to __this_cpu ops

[Christoph Lameter]

On Mon, 16 Sep 2013, Peter Zijlstra wrote:

> On Tue, Sep 03, 2013 at 10:33:20PM +0000, Christoph Lameter wrote:
> > We define a check function in order to avoid trouble with the
> > include files. Then the higher level __this_cpu macros are
> > modified to involve the check before any operation.
> >
>
> So this_cpu_ptr() is the one with the check, and __this_cpu_ptr() is the
> one without. But for the other this_cpu ops __this_cpu_$OP() is going to
> be the one with a check and this_cpu_$OP() the one without?
>
> Sounds like a bloody marvelous idea :/

Well it was the easiest way to get the preemption checks in given.
__this_cpu has no checks like __this_cpu_ptr before this patchset.

We could rename __this_cpu_ptr to raw_cpu_ptr to make it symmetric. A
simple alias would be good for starters.