1 files changed, 159 insertions, 21 deletions

diff --git a/kernel/smp.c b/kernel/smp.c
index d0ae8eb6bf8b..aeb0adfa0606 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -14,17 +14,21 @@
 #include <linux/export.h>
 #include <linux/percpu.h>
 #include <linux/init.h>
+#include <linux/interrupt.h>
 #include <linux/gfp.h>
 #include <linux/smp.h>
 #include <linux/cpu.h>
 #include <linux/sched.h>
 #include <linux/sched/idle.h>
 #include <linux/hypervisor.h>
+#include <linux/sched/clock.h>
+#include <linux/nmi.h>
+#include <linux/sched/debug.h>
 
 #include "smpboot.h"
 #include "sched/smp.h"
 
-#define CSD_TYPE(_csd)	((_csd)->flags & CSD_FLAG_TYPE_MASK)
+#define CSD_TYPE(_csd)	((_csd)->node.u_flags & CSD_FLAG_TYPE_MASK)
 
 struct call_function_data {
 	call_single_data_t	__percpu *csd;
@@ -96,6 +100,103 @@ void __init call_function_init(void)
 	smpcfd_prepare_cpu(smp_processor_id());
 }
 
+#ifdef CONFIG_CSD_LOCK_WAIT_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);
+
+#define CSD_LOCK_TIMEOUT (5ULL * NSEC_PER_SEC)
+static atomic_t csd_bug_count = ATOMIC_INIT(0);
+
+/* Record current CSD work for current CPU, NULL to erase. */
+static void csd_lock_record(call_single_data_t *csd)
+{
+	if (!csd) {
+		smp_mb(); /* NULL cur_csd after unlock. */
+		__this_cpu_write(cur_csd, NULL);
+		return;
+	}
+	__this_cpu_write(cur_csd_func, csd->func);
+	__this_cpu_write(cur_csd_info, csd->info);
+	smp_wmb(); /* func and info before csd. */
+	__this_cpu_write(cur_csd, csd);
+	smp_mb(); /* Update cur_csd before function call. */
+		  /* Or before unlock, as the case may be. */
+}
+
+static __always_inline int csd_lock_wait_getcpu(call_single_data_t *csd)
+{
+	unsigned int csd_type;
+
+	csd_type = CSD_TYPE(csd);
+	if (csd_type == CSD_TYPE_ASYNC || csd_type == CSD_TYPE_SYNC)
+		return csd->node.dst; /* Other CSD_TYPE_ values might not have ->dst. */
+	return -1;
+}
+
+/*
+ * 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)
+{
+	int cpu = -1;
+	int cpux;
+	bool firsttime;
+	u64 ts2, ts_delta;
+	call_single_data_t *cpu_cur_csd;
+	unsigned int flags = READ_ONCE(csd->node.u_flags);
+
+	if (!(flags & CSD_FLAG_LOCK)) {
+		if (!unlikely(*bug_id))
+			return true;
+		cpu = csd_lock_wait_getcpu(csd);
+		pr_alert("csd: CSD lock (#%d) got unstuck on CPU#%02d, CPU#%02d released the lock.\n",
+			 *bug_id, raw_smp_processor_id(), cpu);
+		return true;
+	}
+
+	ts2 = sched_clock();
+	ts_delta = ts2 - *ts1;
+	if (likely(ts_delta <= CSD_LOCK_TIMEOUT))
+		return false;
+
+	firsttime = !*bug_id;
+	if (firsttime)
+		*bug_id = atomic_inc_return(&csd_bug_count);
+	cpu = csd_lock_wait_getcpu(csd);
+	if (WARN_ONCE(cpu < 0 || cpu >= nr_cpu_ids, "%s: cpu = %d\n", __func__, cpu))
+		cpux = 0;
+	else
+		cpux = cpu;
+	cpu_cur_csd = smp_load_acquire(&per_cpu(cur_csd, cpux)); /* Before func and info. */
+	pr_alert("csd: %s non-responsive CSD lock (#%d) on CPU#%d, waiting %llu ns for CPU#%02d %pS(%ps).\n",
+		 firsttime ? "Detected" : "Continued", *bug_id, raw_smp_processor_id(), ts2 - ts0,
+		 cpu, csd->func, csd->info);
+	if (cpu_cur_csd && csd != cpu_cur_csd) {
+		pr_alert("\tcsd: CSD lock (#%d) handling prior %pS(%ps) request.\n",
+			 *bug_id, READ_ONCE(per_cpu(cur_csd_func, cpux)),
+			 READ_ONCE(per_cpu(cur_csd_info, cpux)));
+	} else {
+		pr_alert("\tcsd: CSD lock (#%d) %s.\n",
+			 *bug_id, !cpu_cur_csd ? "unresponsive" : "handling this request");
+	}
+	if (cpu >= 0) {
+		if (!trigger_single_cpu_backtrace(cpu))
+			dump_cpu_task(cpu);
+		if (!cpu_cur_csd) {
+			pr_alert("csd: Re-sending CSD lock (#%d) IPI from CPU#%02d to CPU#%02d\n", *bug_id, raw_smp_processor_id(), cpu);
+			arch_send_call_function_single_ipi(cpu);
+		}
+	}
+	dump_stack();
+	*ts1 = ts2;
+
+	return false;
+}
+
 /*
  * csd_lock/csd_unlock used to serialize access to per-cpu csd resources
  *
@@ -105,13 +206,33 @@ void __init call_function_init(void)
  */
 static __always_inline void csd_lock_wait(call_single_data_t *csd)
 {
-	smp_cond_load_acquire(&csd->flags, !(VAL & CSD_FLAG_LOCK));
+	int bug_id = 0;
+	u64 ts0, ts1;
+
+	ts1 = ts0 = sched_clock();
+	for (;;) {
+		if (csd_lock_wait_toolong(csd, ts0, &ts1, &bug_id))
+			break;
+		cpu_relax();
+	}
+	smp_acquire__after_ctrl_dep();
+}
+
+#else
+static void csd_lock_record(call_single_data_t *csd)
+{
+}
+
+static __always_inline void csd_lock_wait(call_single_data_t *csd)
+{
+	smp_cond_load_acquire(&csd->node.u_flags, !(VAL & CSD_FLAG_LOCK));
 }
+#endif
 
 static __always_inline void csd_lock(call_single_data_t *csd)
 {
 	csd_lock_wait(csd);
-	csd->flags |= CSD_FLAG_LOCK;
+	csd->node.u_flags |= CSD_FLAG_LOCK;
 
 	/*
 	 * prevent CPU from reordering the above assignment
@@ -123,12 +244,12 @@ static __always_inline void csd_lock(call_single_data_t *csd)
 
 static __always_inline void csd_unlock(call_single_data_t *csd)
 {
-	WARN_ON(!(csd->flags & CSD_FLAG_LOCK));
+	WARN_ON(!(csd->node.u_flags & CSD_FLAG_LOCK));
 
 	/*
 	 * ensure we're all done before releasing data:
 	 */
-	smp_store_release(&csd->flags, 0);
+	smp_store_release(&csd->node.u_flags, 0);
 }
 
 static DEFINE_PER_CPU_SHARED_ALIGNED(call_single_data_t, csd_data);
@@ -166,9 +287,11 @@ static int generic_exec_single(int cpu, call_single_data_t *csd)
 		 * We can unlock early even for the synchronous on-stack case,
 		 * since we're doing this from the same CPU..
 		 */
+		csd_lock_record(csd);
 		csd_unlock(csd);
 		local_irq_save(flags);
 		func(info);
+		csd_lock_record(NULL);
 		local_irq_restore(flags);
 		return 0;
 	}
@@ -178,7 +301,7 @@ static int generic_exec_single(int cpu, call_single_data_t *csd)
 		return -ENXIO;
 	}
 
-	__smp_call_single_queue(cpu, &csd->llist);
+	__smp_call_single_queue(cpu, &csd->node.llist);
 
 	return 0;
 }
@@ -231,7 +354,7 @@ static void flush_smp_call_function_queue(bool warn_cpu_offline)
 		 * We don't have to use the _safe() variant here
 		 * because we are not invoking the IPI handlers yet.
 		 */
-		llist_for_each_entry(csd, entry, llist) {
+		llist_for_each_entry(csd, entry, node.llist) {
 			switch (CSD_TYPE(csd)) {
 			case CSD_TYPE_ASYNC:
 			case CSD_TYPE_SYNC:
@@ -256,22 +379,24 @@ static void flush_smp_call_function_queue(bool warn_cpu_offline)
 	 * First; run all SYNC callbacks, people are waiting for us.
 	 */
 	prev = NULL;
-	llist_for_each_entry_safe(csd, csd_next, entry, llist) {
+	llist_for_each_entry_safe(csd, csd_next, entry, node.llist) {
 		/* Do we wait until *after* callback? */
 		if (CSD_TYPE(csd) == CSD_TYPE_SYNC) {
 			smp_call_func_t func = csd->func;
 			void *info = csd->info;
 
 			if (prev) {
-				prev->next = &csd_next->llist;
+				prev->next = &csd_next->node.llist;
 			} else {
-				entry = &csd_next->llist;
+				entry = &csd_next->node.llist;
 			}
 
+			csd_lock_record(csd);
 			func(info);
 			csd_unlock(csd);
+			csd_lock_record(NULL);
 		} else {
-			prev = &csd->llist;
+			prev = &csd->node.llist;
 		}
 	}
 
@@ -282,28 +407,30 @@ static void flush_smp_call_function_queue(bool warn_cpu_offline)
 	 * Second; run all !SYNC callbacks.
 	 */
 	prev = NULL;
-	llist_for_each_entry_safe(csd, csd_next, entry, llist) {
+	llist_for_each_entry_safe(csd, csd_next, entry, node.llist) {
 		int type = CSD_TYPE(csd);
 
 		if (type != CSD_TYPE_TTWU) {
 			if (prev) {
-				prev->next = &csd_next->llist;
+				prev->next = &csd_next->node.llist;
 			} else {
-				entry = &csd_next->llist;
+				entry = &csd_next->node.llist;
 			}
 
 			if (type == CSD_TYPE_ASYNC) {
 				smp_call_func_t func = csd->func;
 				void *info = csd->info;
 
+				csd_lock_record(csd);
 				csd_unlock(csd);
 				func(info);
+				csd_lock_record(NULL);
 			} else if (type == CSD_TYPE_IRQ_WORK) {
 				irq_work_single(csd);
 			}
 
 		} else {
-			prev = &csd->llist;
+			prev = &csd->node.llist;
 		}
 	}
 
@@ -323,6 +450,9 @@ void flush_smp_call_function_from_idle(void)
 
 	local_irq_save(flags);
 	flush_smp_call_function_queue(true);
+	if (local_softirq_pending())
+		do_softirq();
+
 	local_irq_restore(flags);
 }
 
@@ -339,7 +469,7 @@ int smp_call_function_single(int cpu, smp_call_func_t func, void *info,
 {
 	call_single_data_t *csd;
 	call_single_data_t csd_stack = {
-		.flags = CSD_FLAG_LOCK | CSD_TYPE_SYNC,
+		.node = { .u_flags = CSD_FLAG_LOCK | CSD_TYPE_SYNC, },
 	};
 	int this_cpu;
 	int err;
@@ -375,6 +505,10 @@ int smp_call_function_single(int cpu, smp_call_func_t func, void *info,
 
 	csd->func = func;
 	csd->info = info;
+#ifdef CONFIG_CSD_LOCK_WAIT_DEBUG
+	csd->node.src = smp_processor_id();
+	csd->node.dst = cpu;
+#endif
 
 	err = generic_exec_single(cpu, csd);
 
@@ -414,12 +548,12 @@ int smp_call_function_single_async(int cpu, call_single_data_t *csd)
 
 	preempt_disable();
 
-	if (csd->flags & CSD_FLAG_LOCK) {
+	if (csd->node.u_flags & CSD_FLAG_LOCK) {
 		err = -EBUSY;
 		goto out;
 	}
 
-	csd->flags = CSD_FLAG_LOCK;
+	csd->node.u_flags = CSD_FLAG_LOCK;
 	smp_wmb();
 
 	err = generic_exec_single(cpu, csd);
@@ -537,10 +671,14 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
 
 		csd_lock(csd);
 		if (wait)
-			csd->flags |= CSD_TYPE_SYNC;
+			csd->node.u_flags |= CSD_TYPE_SYNC;
 		csd->func = func;
 		csd->info = info;
-		if (llist_add(&csd->llist, &per_cpu(call_single_queue, cpu)))
+#ifdef CONFIG_CSD_LOCK_WAIT_DEBUG
+		csd->node.src = smp_processor_id();
+		csd->node.dst = cpu;
+#endif
+		if (llist_add(&csd->node.llist, &per_cpu(call_single_queue, cpu)))
 			__cpumask_set_cpu(cpu, cfd->cpumask_ipi);
 	}
 
@@ -741,7 +879,7 @@ EXPORT_SYMBOL(on_each_cpu_mask);
  * for all the required CPUs to finish. This may include the local
  * processor.
  * @cond_func:	A callback function that is passed a cpu id and
- *		the the info parameter. The function is called
+ *		the info parameter. The function is called
  *		with preemption disabled. The function should
  *		return a blooean value indicating whether to IPI
  *		the specified CPU.