smp: Avoid using two cache lines for struct call_single_data

struct call_single_data is used in IPIs to transfer information between
CPUs.  Its size is bigger than sizeof(unsigned long) and less than
cache line size.  Currently it is not allocated with any explicit alignment
requirements.  This makes it possible for allocated call_single_data to
cross two cache lines, which results in double the number of the cache lines
that need to be transferred among CPUs.

This can be fixed by requiring call_single_data to be aligned with the
size of call_single_data. Currently the size of call_single_data is the
power of 2.  If we add new fields to call_single_data, we may need to
add padding to make sure the size of new definition is the power of 2
as well.

Fortunately, this is enforced by GCC, which will report bad sizes.

To set alignment requirements of call_single_data to the size of
call_single_data, a struct definition and a typedef is used.

To test the effect of the patch, I used the vm-scalability multiple
thread swap test case (swap-w-seq-mt).  The test will create multiple
threads and each thread will eat memory until all RAM and part of swap
is used, so that huge number of IPIs are triggered when unmapping
memory.  In the test, the throughput of memory writing improves ~5%
compared with misaligned call_single_data, because of faster IPIs.

Suggested-by: Peter Zijlstra <[email protected]>
Signed-off-by: Huang, Ying <[email protected]>
[ Add call_single_data_t and align with size of call_single_data. ]
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Cc: Aaron Lu <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: Eric Dumazet <[email protected]>
Cc: Juergen Gross <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>

Author: yhuang-intel

arch/mips/kernel/smp.c

@@ -648,12 +648,12 @@ EXPORT_SYMBOL(flush_tlb_one);
 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 
 static DEFINE_PER_CPU(atomic_t, tick_broadcast_count);
-static DEFINE_PER_CPU(struct call_single_data, tick_broadcast_csd);
+static DEFINE_PER_CPU(call_single_data_t, tick_broadcast_csd);
 
 void tick_broadcast(const struct cpumask *mask)
 {
 	atomic_t *count;
-	struct call_single_data *csd;
+	call_single_data_t *csd;
 	int cpu;
 
 	for_each_cpu(cpu, mask) {
@@ -674,7 +674,7 @@ static void tick_broadcast_callee(void *info)
 
 static int __init tick_broadcast_init(void)
 {
-	struct call_single_data *csd;
+	call_single_data_t *csd;
 	int cpu;
 
 	for (cpu = 0; cpu < NR_CPUS; cpu++) {

block/blk-softirq.c

@@ -60,7 +60,7 @@ static void trigger_softirq(void *data)
 static int raise_blk_irq(int cpu, struct request *rq)
 {
 	if (cpu_online(cpu)) {
-		struct call_single_data *data = &rq->csd;
+		call_single_data_t *data = &rq->csd;
 
 		data->func = trigger_softirq;
 		data->info = rq;

drivers/block/null_blk.c

@@ -13,7 +13,7 @@
 struct nullb_cmd {
 	struct list_head list;
 	struct llist_node ll_list;
-	struct call_single_data csd;
+	call_single_data_t csd;
 	struct request *rq;
 	struct bio *bio;
 	unsigned int tag;

drivers/cpuidle/coupled.c

@@ -119,13 +119,13 @@ struct cpuidle_coupled {
 
 #define CPUIDLE_COUPLED_NOT_IDLE	(-1)
 
-static DEFINE_PER_CPU(struct call_single_data, cpuidle_coupled_poke_cb);
+static DEFINE_PER_CPU(call_single_data_t, cpuidle_coupled_poke_cb);
 
 /*
  * The cpuidle_coupled_poke_pending mask is used to avoid calling
- * __smp_call_function_single with the per cpu call_single_data struct already
+ * __smp_call_function_single with the per cpu call_single_data_t struct already
  * in use.  This prevents a deadlock where two cpus are waiting for each others
- * call_single_data struct to be available
+ * call_single_data_t struct to be available
  */
 static cpumask_t cpuidle_coupled_poke_pending;
 
@@ -339,7 +339,7 @@ static void cpuidle_coupled_handle_poke(void *info)
  */
 static void cpuidle_coupled_poke(int cpu)
 {
-	struct call_single_data *csd = &per_cpu(cpuidle_coupled_poke_cb, cpu);
+	call_single_data_t *csd = &per_cpu(cpuidle_coupled_poke_cb, cpu);
 
 	if (!cpumask_test_and_set_cpu(cpu, &cpuidle_coupled_poke_pending))
 		smp_call_function_single_async(cpu, csd);
@@ -651,7 +651,7 @@ int cpuidle_coupled_register_device(struct cpuidle_device *dev)
 {
 	int cpu;
 	struct cpuidle_device *other_dev;
-	struct call_single_data *csd;
+	call_single_data_t *csd;
 	struct cpuidle_coupled *coupled;
 
 	if (cpumask_empty(&dev->coupled_cpus))

drivers/net/ethernet/cavium/liquidio/lio_main.c

@@ -2468,7 +2468,7 @@ static void liquidio_napi_drv_callback(void *arg)
 	if (OCTEON_CN23XX_PF(oct) || droq->cpu_id == this_cpu) {
 		napi_schedule_irqoff(&droq->napi);
 	} else {
-		struct call_single_data *csd = &droq->csd;
+		call_single_data_t *csd = &droq->csd;
 
 		csd->func = napi_schedule_wrapper;
 		csd->info = &droq->napi;

drivers/net/ethernet/cavium/liquidio/octeon_droq.h

@@ -328,7 +328,7 @@ struct octeon_droq {
 
 	u32 cpu_id;
 
-	struct call_single_data csd;
+	call_single_data_t csd;
 };
 
 #define OCT_DROQ_SIZE   (sizeof(struct octeon_droq))

include/linux/blkdev.h

@@ -134,7 +134,7 @@ typedef __u32 __bitwise req_flags_t;
 struct request {
 	struct list_head queuelist;
 	union {
-		struct call_single_data csd;
+		call_single_data_t csd;
 		u64 fifo_time;
 	};
 

include/linux/netdevice.h

@@ -2774,7 +2774,7 @@ struct softnet_data {
 	unsigned int		input_queue_head ____cacheline_aligned_in_smp;
 
 	/* Elements below can be accessed between CPUs for RPS/RFS */
-	struct call_single_data	csd ____cacheline_aligned_in_smp;
+	call_single_data_t	csd ____cacheline_aligned_in_smp;
 	struct softnet_data	*rps_ipi_next;
 	unsigned int		cpu;
 	unsigned int		input_queue_tail;

include/linux/smp.h

@@ -14,13 +14,17 @@
 #include <linux/llist.h>
 
 typedef void (*smp_call_func_t)(void *info);
-struct call_single_data {
+struct __call_single_data {
 	struct llist_node llist;
 	smp_call_func_t func;
 	void *info;
 	unsigned int flags;
 };
 
+/* Use __aligned() to avoid to use 2 cache lines for 1 csd */
+typedef struct __call_single_data call_single_data_t
+	__aligned(sizeof(struct __call_single_data));
+
 /* total number of cpus in this system (may exceed NR_CPUS) */
 extern unsigned int total_cpus;
 
@@ -48,7 +52,7 @@ void on_each_cpu_cond(bool (*cond_func)(int cpu, void *info),
 		smp_call_func_t func, void *info, bool wait,
 		gfp_t gfp_flags);
 
-int smp_call_function_single_async(int cpu, struct call_single_data *csd);
+int smp_call_function_single_async(int cpu, call_single_data_t *csd);
 
 #ifdef CONFIG_SMP
 

kernel/sched/sched.h

@@ -769,7 +769,7 @@ struct rq {
 #ifdef CONFIG_SCHED_HRTICK
 #ifdef CONFIG_SMP
 	int hrtick_csd_pending;
-	struct call_single_data hrtick_csd;
+	call_single_data_t hrtick_csd;
 #endif
 	struct hrtimer hrtick_timer;
 #endif