Merge branch 'locking/core' into x86/mm, to resolve conflict

There's a non-trivial conflict between the parallel TLB flush
framework and the IPI flush debugging code - merge them
manually.

Conflicts:
	kernel/smp.c

Signed-off-by: Ingo Molnar <mingo@kernel.org>
diff --git a/kernel/smp.c b/kernel/smp.c
index af0d51da..e210749 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -24,14 +24,70 @@
 #include <linux/sched/clock.h>
 #include <linux/nmi.h>
 #include <linux/sched/debug.h>
+#include <linux/jump_label.h>
 
 #include "smpboot.h"
 #include "sched/smp.h"
 
 #define CSD_TYPE(_csd)	((_csd)->node.u_flags & CSD_FLAG_TYPE_MASK)
 
+#ifdef CONFIG_CSD_LOCK_WAIT_DEBUG
+union cfd_seq_cnt {
+	u64		val;
+	struct {
+		u64	src:16;
+		u64	dst:16;
+#define CFD_SEQ_NOCPU	0xffff
+		u64	type:4;
+#define CFD_SEQ_QUEUE	0
+#define CFD_SEQ_IPI	1
+#define CFD_SEQ_NOIPI	2
+#define CFD_SEQ_PING	3
+#define CFD_SEQ_PINGED	4
+#define CFD_SEQ_HANDLE	5
+#define CFD_SEQ_DEQUEUE	6
+#define CFD_SEQ_IDLE	7
+#define CFD_SEQ_GOTIPI	8
+#define CFD_SEQ_HDLEND	9
+		u64	cnt:28;
+	}		u;
+};
+
+static char *seq_type[] = {
+	[CFD_SEQ_QUEUE]		= "queue",
+	[CFD_SEQ_IPI]		= "ipi",
+	[CFD_SEQ_NOIPI]		= "noipi",
+	[CFD_SEQ_PING]		= "ping",
+	[CFD_SEQ_PINGED]	= "pinged",
+	[CFD_SEQ_HANDLE]	= "handle",
+	[CFD_SEQ_DEQUEUE]	= "dequeue (src CPU 0 == empty)",
+	[CFD_SEQ_IDLE]		= "idle",
+	[CFD_SEQ_GOTIPI]	= "gotipi",
+	[CFD_SEQ_HDLEND]	= "hdlend (src CPU 0 == early)",
+};
+
+struct cfd_seq_local {
+	u64	ping;
+	u64	pinged;
+	u64	handle;
+	u64	dequeue;
+	u64	idle;
+	u64	gotipi;
+	u64	hdlend;
+};
+#endif
+
+struct cfd_percpu {
+	call_single_data_t	csd;
+#ifdef CONFIG_CSD_LOCK_WAIT_DEBUG
+	u64	seq_queue;
+	u64	seq_ipi;
+	u64	seq_noipi;
+#endif
+};
+
 struct call_function_data {
-	call_single_data_t	__percpu *csd;
+	struct cfd_percpu	__percpu *pcpu;
 	cpumask_var_t		cpumask;
 	cpumask_var_t		cpumask_ipi;
 };
@@ -54,8 +110,8 @@ int smpcfd_prepare_cpu(unsigned int cpu)
 		free_cpumask_var(cfd->cpumask);
 		return -ENOMEM;
 	}
-	cfd->csd = alloc_percpu(call_single_data_t);
-	if (!cfd->csd) {
+	cfd->pcpu = alloc_percpu(struct cfd_percpu);
+	if (!cfd->pcpu) {
 		free_cpumask_var(cfd->cpumask);
 		free_cpumask_var(cfd->cpumask_ipi);
 		return -ENOMEM;
@@ -70,7 +126,7 @@ int smpcfd_dead_cpu(unsigned int cpu)
 
 	free_cpumask_var(cfd->cpumask);
 	free_cpumask_var(cfd->cpumask_ipi);
-	free_percpu(cfd->csd);
+	free_percpu(cfd->pcpu);
 	return 0;
 }
 
@@ -102,15 +158,60 @@ void __init call_function_init(void)
 
 #ifdef CONFIG_CSD_LOCK_WAIT_DEBUG
 
+static DEFINE_STATIC_KEY_FALSE(csdlock_debug_enabled);
+static DEFINE_STATIC_KEY_FALSE(csdlock_debug_extended);
+
+static int __init csdlock_debug(char *str)
+{
+	unsigned int val = 0;
+
+	if (str && !strcmp(str, "ext")) {
+		val = 1;
+		static_branch_enable(&csdlock_debug_extended);
+	} else
+		get_option(&str, &val);
+
+	if (val)
+		static_branch_enable(&csdlock_debug_enabled);
+
+	return 0;
+}
+early_param("csdlock_debug", csdlock_debug);
+
 static DEFINE_PER_CPU(call_single_data_t *, cur_csd);
 static DEFINE_PER_CPU(smp_call_func_t, cur_csd_func);
 static DEFINE_PER_CPU(void *, cur_csd_info);
+static DEFINE_PER_CPU(struct cfd_seq_local, cfd_seq_local);
 
 #define CSD_LOCK_TIMEOUT (5ULL * NSEC_PER_SEC)
 static atomic_t csd_bug_count = ATOMIC_INIT(0);
+static u64 cfd_seq;
+
+#define CFD_SEQ(s, d, t, c)	\
+	(union cfd_seq_cnt){ .u.src = s, .u.dst = d, .u.type = t, .u.cnt = c }
+
+static u64 cfd_seq_inc(unsigned int src, unsigned int dst, unsigned int type)
+{
+	union cfd_seq_cnt new, old;
+
+	new = CFD_SEQ(src, dst, type, 0);
+
+	do {
+		old.val = READ_ONCE(cfd_seq);
+		new.u.cnt = old.u.cnt + 1;
+	} while (cmpxchg(&cfd_seq, old.val, new.val) != old.val);
+
+	return old.val;
+}
+
+#define cfd_seq_store(var, src, dst, type)				\
+	do {								\
+		if (static_branch_unlikely(&csdlock_debug_extended))	\
+			var = cfd_seq_inc(src, dst, type);		\
+	} while (0)
 
 /* Record current CSD work for current CPU, NULL to erase. */
-static void csd_lock_record(call_single_data_t *csd)
+static void __csd_lock_record(call_single_data_t *csd)
 {
 	if (!csd) {
 		smp_mb(); /* NULL cur_csd after unlock. */
@@ -125,7 +226,13 @@ static void csd_lock_record(call_single_data_t *csd)
 		  /* Or before unlock, as the case may be. */
 }
 
-static __always_inline int csd_lock_wait_getcpu(call_single_data_t *csd)
+static __always_inline void csd_lock_record(call_single_data_t *csd)
+{
+	if (static_branch_unlikely(&csdlock_debug_enabled))
+		__csd_lock_record(csd);
+}
+
+static int csd_lock_wait_getcpu(call_single_data_t *csd)
 {
 	unsigned int csd_type;
 
@@ -135,12 +242,86 @@ static __always_inline int csd_lock_wait_getcpu(call_single_data_t *csd)
 	return -1;
 }
 
+static void cfd_seq_data_add(u64 val, unsigned int src, unsigned int dst,
+			     unsigned int type, union cfd_seq_cnt *data,
+			     unsigned int *n_data, unsigned int now)
+{
+	union cfd_seq_cnt new[2];
+	unsigned int i, j, k;
+
+	new[0].val = val;
+	new[1] = CFD_SEQ(src, dst, type, new[0].u.cnt + 1);
+
+	for (i = 0; i < 2; i++) {
+		if (new[i].u.cnt <= now)
+			new[i].u.cnt |= 0x80000000U;
+		for (j = 0; j < *n_data; j++) {
+			if (new[i].u.cnt == data[j].u.cnt) {
+				/* Direct read value trumps generated one. */
+				if (i == 0)
+					data[j].val = new[i].val;
+				break;
+			}
+			if (new[i].u.cnt < data[j].u.cnt) {
+				for (k = *n_data; k > j; k--)
+					data[k].val = data[k - 1].val;
+				data[j].val = new[i].val;
+				(*n_data)++;
+				break;
+			}
+		}
+		if (j == *n_data) {
+			data[j].val = new[i].val;
+			(*n_data)++;
+		}
+	}
+}
+
+static const char *csd_lock_get_type(unsigned int type)
+{
+	return (type >= ARRAY_SIZE(seq_type)) ? "?" : seq_type[type];
+}
+
+static void csd_lock_print_extended(call_single_data_t *csd, int cpu)
+{
+	struct cfd_seq_local *seq = &per_cpu(cfd_seq_local, cpu);
+	unsigned int srccpu = csd->node.src;
+	struct call_function_data *cfd = per_cpu_ptr(&cfd_data, srccpu);
+	struct cfd_percpu *pcpu = per_cpu_ptr(cfd->pcpu, cpu);
+	unsigned int now;
+	union cfd_seq_cnt data[2 * ARRAY_SIZE(seq_type)];
+	unsigned int n_data = 0, i;
+
+	data[0].val = READ_ONCE(cfd_seq);
+	now = data[0].u.cnt;
+
+	cfd_seq_data_add(pcpu->seq_queue,			srccpu, cpu,	       CFD_SEQ_QUEUE,  data, &n_data, now);
+	cfd_seq_data_add(pcpu->seq_ipi,				srccpu, cpu,	       CFD_SEQ_IPI,    data, &n_data, now);
+	cfd_seq_data_add(pcpu->seq_noipi,			srccpu, cpu,	       CFD_SEQ_NOIPI,  data, &n_data, now);
+
+	cfd_seq_data_add(per_cpu(cfd_seq_local.ping, srccpu),	srccpu, CFD_SEQ_NOCPU, CFD_SEQ_PING,   data, &n_data, now);
+	cfd_seq_data_add(per_cpu(cfd_seq_local.pinged, srccpu), srccpu, CFD_SEQ_NOCPU, CFD_SEQ_PINGED, data, &n_data, now);
+
+	cfd_seq_data_add(seq->idle,    CFD_SEQ_NOCPU, cpu, CFD_SEQ_IDLE,    data, &n_data, now);
+	cfd_seq_data_add(seq->gotipi,  CFD_SEQ_NOCPU, cpu, CFD_SEQ_GOTIPI,  data, &n_data, now);
+	cfd_seq_data_add(seq->handle,  CFD_SEQ_NOCPU, cpu, CFD_SEQ_HANDLE,  data, &n_data, now);
+	cfd_seq_data_add(seq->dequeue, CFD_SEQ_NOCPU, cpu, CFD_SEQ_DEQUEUE, data, &n_data, now);
+	cfd_seq_data_add(seq->hdlend,  CFD_SEQ_NOCPU, cpu, CFD_SEQ_HDLEND,  data, &n_data, now);
+
+	for (i = 0; i < n_data; i++) {
+		pr_alert("\tcsd: cnt(%07x): %04x->%04x %s\n",
+			 data[i].u.cnt & ~0x80000000U, data[i].u.src,
+			 data[i].u.dst, csd_lock_get_type(data[i].u.type));
+	}
+	pr_alert("\tcsd: cnt now: %07x\n", now);
+}
+
 /*
  * Complain if too much time spent waiting.  Note that only
  * the CSD_TYPE_SYNC/ASYNC types provide the destination CPU,
  * so waiting on other types gets much less information.
  */
-static __always_inline bool csd_lock_wait_toolong(call_single_data_t *csd, u64 ts0, u64 *ts1, int *bug_id)
+static bool csd_lock_wait_toolong(call_single_data_t *csd, u64 ts0, u64 *ts1, int *bug_id)
 {
 	int cpu = -1;
 	int cpux;
@@ -184,6 +365,8 @@ static __always_inline bool csd_lock_wait_toolong(call_single_data_t *csd, u64 t
 			 *bug_id, !cpu_cur_csd ? "unresponsive" : "handling this request");
 	}
 	if (cpu >= 0) {
+		if (static_branch_unlikely(&csdlock_debug_extended))
+			csd_lock_print_extended(csd, cpu);
 		if (!trigger_single_cpu_backtrace(cpu))
 			dump_cpu_task(cpu);
 		if (!cpu_cur_csd) {
@@ -204,7 +387,7 @@ static __always_inline bool csd_lock_wait_toolong(call_single_data_t *csd, u64 t
  * previous function call. For multi-cpu calls its even more interesting
  * as we'll have to ensure no other cpu is observing our csd.
  */
-static __always_inline void csd_lock_wait(call_single_data_t *csd)
+static void __csd_lock_wait(call_single_data_t *csd)
 {
 	int bug_id = 0;
 	u64 ts0, ts1;
@@ -218,7 +401,36 @@ static __always_inline void csd_lock_wait(call_single_data_t *csd)
 	smp_acquire__after_ctrl_dep();
 }
 
+static __always_inline void csd_lock_wait(call_single_data_t *csd)
+{
+	if (static_branch_unlikely(&csdlock_debug_enabled)) {
+		__csd_lock_wait(csd);
+		return;
+	}
+
+	smp_cond_load_acquire(&csd->node.u_flags, !(VAL & CSD_FLAG_LOCK));
+}
+
+static void __smp_call_single_queue_debug(int cpu, struct llist_node *node)
+{
+	unsigned int this_cpu = smp_processor_id();
+	struct cfd_seq_local *seq = this_cpu_ptr(&cfd_seq_local);
+	struct call_function_data *cfd = this_cpu_ptr(&cfd_data);
+	struct cfd_percpu *pcpu = per_cpu_ptr(cfd->pcpu, cpu);
+
+	cfd_seq_store(pcpu->seq_queue, this_cpu, cpu, CFD_SEQ_QUEUE);
+	if (llist_add(node, &per_cpu(call_single_queue, cpu))) {
+		cfd_seq_store(pcpu->seq_ipi, this_cpu, cpu, CFD_SEQ_IPI);
+		cfd_seq_store(seq->ping, this_cpu, cpu, CFD_SEQ_PING);
+		send_call_function_single_ipi(cpu);
+		cfd_seq_store(seq->pinged, this_cpu, cpu, CFD_SEQ_PINGED);
+	} else {
+		cfd_seq_store(pcpu->seq_noipi, this_cpu, cpu, CFD_SEQ_NOIPI);
+	}
+}
 #else
+#define cfd_seq_store(var, src, dst, type)
+
 static void csd_lock_record(call_single_data_t *csd)
 {
 }
@@ -256,6 +468,19 @@ static DEFINE_PER_CPU_SHARED_ALIGNED(call_single_data_t, csd_data);
 
 void __smp_call_single_queue(int cpu, struct llist_node *node)
 {
+#ifdef CONFIG_CSD_LOCK_WAIT_DEBUG
+	if (static_branch_unlikely(&csdlock_debug_extended)) {
+		unsigned int type;
+
+		type = CSD_TYPE(container_of(node, call_single_data_t,
+					     node.llist));
+		if (type == CSD_TYPE_SYNC || type == CSD_TYPE_ASYNC) {
+			__smp_call_single_queue_debug(cpu, node);
+			return;
+		}
+	}
+#endif
+
 	/*
 	 * The list addition should be visible before sending the IPI
 	 * handler locks the list to pull the entry off it because of
@@ -314,6 +539,8 @@ static int generic_exec_single(int cpu, call_single_data_t *csd)
  */
 void generic_smp_call_function_single_interrupt(void)
 {
+	cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->gotipi, CFD_SEQ_NOCPU,
+		      smp_processor_id(), CFD_SEQ_GOTIPI);
 	flush_smp_call_function_queue(true);
 }
 
@@ -341,7 +568,13 @@ static void flush_smp_call_function_queue(bool warn_cpu_offline)
 	lockdep_assert_irqs_disabled();
 
 	head = this_cpu_ptr(&call_single_queue);
+	cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->handle, CFD_SEQ_NOCPU,
+		      smp_processor_id(), CFD_SEQ_HANDLE);
 	entry = llist_del_all(head);
+	cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->dequeue,
+		      /* Special meaning of source cpu: 0 == queue empty */
+		      entry ? CFD_SEQ_NOCPU : 0,
+		      smp_processor_id(), CFD_SEQ_DEQUEUE);
 	entry = llist_reverse_order(entry);
 
 	/* There shouldn't be any pending callbacks on an offline CPU. */
@@ -400,8 +633,12 @@ static void flush_smp_call_function_queue(bool warn_cpu_offline)
 		}
 	}
 
-	if (!entry)
+	if (!entry) {
+		cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->hdlend,
+			      0, smp_processor_id(),
+			      CFD_SEQ_HDLEND);
 		return;
+	}
 
 	/*
 	 * Second; run all !SYNC callbacks.
@@ -439,6 +676,9 @@ static void flush_smp_call_function_queue(bool warn_cpu_offline)
 	 */
 	if (entry)
 		sched_ttwu_pending(entry);
+
+	cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->hdlend, CFD_SEQ_NOCPU,
+		      smp_processor_id(), CFD_SEQ_HDLEND);
 }
 
 void flush_smp_call_function_from_idle(void)
@@ -448,6 +688,8 @@ void flush_smp_call_function_from_idle(void)
 	if (llist_empty(this_cpu_ptr(&call_single_queue)))
 		return;
 
+	cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->idle, CFD_SEQ_NOCPU,
+		      smp_processor_id(), CFD_SEQ_IDLE);
 	local_irq_save(flags);
 	flush_smp_call_function_queue(true);
 	if (local_softirq_pending())
@@ -667,7 +909,8 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 
 		cpumask_clear(cfd->cpumask_ipi);
 		for_each_cpu(cpu, cfd->cpumask) {
-			call_single_data_t *csd = per_cpu_ptr(cfd->csd, cpu);
+			struct cfd_percpu *pcpu = per_cpu_ptr(cfd->pcpu, cpu);
+			call_single_data_t *csd = &pcpu->csd;
 
 			if (cond_func && !cond_func(cpu, info))
 				continue;
@@ -681,13 +924,20 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 			csd->node.src = smp_processor_id();
 			csd->node.dst = cpu;
 #endif
+			cfd_seq_store(pcpu->seq_queue, this_cpu, cpu, CFD_SEQ_QUEUE);
 			if (llist_add(&csd->node.llist, &per_cpu(call_single_queue, cpu))) {
 				__cpumask_set_cpu(cpu, cfd->cpumask_ipi);
 				nr_cpus++;
 				last_cpu = cpu;
+
+				cfd_seq_store(pcpu->seq_ipi, this_cpu, cpu, CFD_SEQ_IPI);
+			} else {
+				cfd_seq_store(pcpu->seq_noipi, this_cpu, cpu, CFD_SEQ_NOIPI);
 			}
 		}
 
+		cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->ping, this_cpu, CFD_SEQ_NOCPU, CFD_SEQ_PING);
+
 		/*
 		 * Choose the most efficient way to send an IPI. Note that the
 		 * number of CPUs might be zero due to concurrent changes to the
@@ -697,6 +947,8 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 			send_call_function_single_ipi(last_cpu);
 		else if (likely(nr_cpus > 1))
 			arch_send_call_function_ipi_mask(cfd->cpumask_ipi);
+
+		cfd_seq_store(this_cpu_ptr(&cfd_seq_local)->pinged, this_cpu, CFD_SEQ_NOCPU, CFD_SEQ_PINGED);
 	}
 
 	if (run_local && (!cond_func || cond_func(this_cpu, info))) {
@@ -711,7 +963,7 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 		for_each_cpu(cpu, cfd->cpumask) {
 			call_single_data_t *csd;
 
-			csd = per_cpu_ptr(cfd->csd, cpu);
+			csd = &per_cpu_ptr(cfd->pcpu, cpu)->csd;
 			csd_lock_wait(csd);
 		}
 	}