load-monitor: 新增输出各CPU上正在运行进程的内核态和用户态堆栈功能

4 files changed, 80 insertions, 1 deletions

diff --git a/SOURCE/diagnose-tools/load.cc b/SOURCE/diagnose-tools/load.cc
index 0def903..5a34790 100644
--- a/SOURCE/diagnose-tools/load.cc
+++ b/SOURCE/diagnose-tools/load.cc
@@ -74,6 +74,7 @@ static void do_activate(const char *arg)
 	settings.verbose = parse.int_value("verbose");
 	settings.style = parse.int_value("style");
 	settings.mass = parse.int_value("mass");
+	settings.cpu_run = parse.int_value("cpu.run");
 
 	if (run_in_host) {
 		ret = diag_call_ioctl(DIAG_IOCTL_LOAD_MONITOR_SET, (long)&settings);
@@ -180,6 +181,7 @@ static int load_monitor_extract(void *buf, unsigned int len, void *)
 	int *et_type;
 	struct load_monitor_detail *detail;
 	struct load_monitor_task *tsk_info;
+	struct load_monitor_cpu_run *cpu_run;
 	static int seq = 0;
 
 	if (len == 0)
@@ -229,6 +231,30 @@ static int load_monitor_extract(void *buf, unsigned int len, void *)
 		tsk_info++;
 
 		break;
+	case et_load_monitor_cpu_run:
+		if (len < sizeof(struct load_monitor_cpu_run))
+			break;
+		cpu_run = (struct load_monitor_cpu_run *)buf;
+		printf("输出各CPU当前运行任务: CPU:%lu task:%s(%u)\n",
+			cpu_run->cpu, cpu_run->task.comm, cpu_run->task.pid);
+
+		printf("##CGROUP:[%s]  %u      [%03d]  采样命中[%s]\n",
+				cpu_run->task.cgroup_buf,
+				cpu_run->task.pid,
+				seq,
+				cpu_run->task.state == 0 ? "R" : "D");
+
+		diag_printf_kern_stack(&cpu_run->kern_stack);
+		diag_printf_user_stack(cpu_run->task.tgid,
+				       cpu_run->task.container_tgid,
+				       cpu_run->task.comm,
+				       &cpu_run->user_stack);
+		printf("#*        0xffffffffffffff %s (UNKNOWN)\n",
+			cpu_run->task.comm);
+		printf("#*        0xffffffffffffff cpu:%lu (UNKNOWN)\n", cpu_run->cpu);
+		printf("##\n");
+
+		break;
 	default:
 		break;
 	}
@@ -436,6 +462,9 @@ static void do_dump(const char *arg)
 	out_json = parse.int_value("json", 0);
 	out_flame = parse.int_value("flame", 1);
 
+	g_symbol_parser.java_only = parse.int_value("java-only", 0);
+	g_symbol_parser.user_symbol = parse.int_value("user-symbol", 1);
+
 	if (run_in_host) {
 		ret = diag_call_ioctl(DIAG_IOCTL_LOAD_MONITOR_DUMP, (long)&dump_param);
 	} else {
diff --git a/SOURCE/module/kernel/load.c b/SOURCE/module/kernel/load.c
index 9b4af2f..09f4f42 100755
--- a/SOURCE/module/kernel/load.c
+++ b/SOURCE/module/kernel/load.c
@@ -40,7 +40,7 @@
 
 static atomic64_t diag_nr_running = ATOMIC64_INIT(0);
 
-struct diag_load_monitor_settings load_monitor_settings;
+static struct diag_load_monitor_settings load_monitor_settings;
 
 static unsigned int load_monitor_alloced;
 
@@ -49,6 +49,7 @@ static struct mm_tree mm_tree;
 unsigned long *orig_avenrun_r;
 unsigned long *orig_avenrun;
 
+static struct load_monitor_cpu_run ld_mon_cpu_run[NR_CPUS];
 static struct diag_variant_buffer load_monitor_variant_buffer;
 
 static ktime_t last_dump;
@@ -77,6 +78,38 @@ void diag_load_timer(struct diag_percpu_context *context)
 	return;
 }
 #else
+
+static void load_monitor_ipi(void *ignore)
+{
+	struct load_monitor_cpu_run *cpu_run;
+	struct task_struct *tsk;
+	unsigned long flags;
+	int cpu;
+
+	tsk = current;
+	cpu = smp_processor_id();
+	if (cpu >= NR_CPUS)
+		return;
+
+	cpu_run = &ld_mon_cpu_run[cpu];
+
+	cpu_run->id = get_cycles();
+	cpu_run->et_type = et_load_monitor_cpu_run;
+	cpu_run->cpu = cpu;
+	do_gettimeofday(&cpu_run->tv);
+
+	diag_task_brief(tsk, &cpu_run->task);
+	diag_task_kern_stack(tsk, &cpu_run->kern_stack);
+	diag_task_user_stack(tsk, &cpu_run->user_stack);
+
+	diag_variant_buffer_spin_lock(&load_monitor_variant_buffer, flags);
+	diag_variant_buffer_reserve(&load_monitor_variant_buffer, sizeof(struct load_monitor_cpu_run));
+	diag_variant_buffer_write_nolock(&load_monitor_variant_buffer,
+					 cpu_run, sizeof(struct load_monitor_cpu_run));
+	diag_variant_buffer_seal(&load_monitor_variant_buffer);
+	diag_variant_buffer_spin_unlock(&load_monitor_variant_buffer, flags);
+}
+
 void diag_load_timer(struct diag_percpu_context *context)
 {
 	u64 ms;
@@ -190,6 +223,11 @@ void diag_load_timer(struct diag_percpu_context *context)
 			}
 		} while_each_thread(g, p);
 		rcu_read_unlock();
+
+		if (!load_monitor_settings.mass && load_monitor_settings.cpu_run) {
+			smp_call_function(load_monitor_ipi, NULL, 1);
+			load_monitor_ipi(NULL);
+		}
 	}
 }
 #endif
diff --git a/SOURCE/uapi/ali_diagnose.h b/SOURCE/uapi/ali_diagnose.h
index 9d79bff..1ba7763 100644
--- a/SOURCE/uapi/ali_diagnose.h
+++ b/SOURCE/uapi/ali_diagnose.h
@@ -395,6 +395,7 @@ enum diag_record_id {
 	et_load_monitor_base = et_kprobe_base + DIAG_EVENT_TYPE_INTERVAL,
 	et_load_monitor_detail,
 	et_load_monitor_task,
+	et_load_monitor_cpu_run,
 
 	et_mm_leak_base = et_load_monitor_base + DIAG_EVENT_TYPE_INTERVAL,
 	et_mm_leak_detail,
diff --git a/SOURCE/uapi/load_monitor.h b/SOURCE/uapi/load_monitor.h
index bff499a..f4f91e9 100644
--- a/SOURCE/uapi/load_monitor.h
+++ b/SOURCE/uapi/load_monitor.h
@@ -27,6 +27,7 @@ struct diag_load_monitor_settings {
 	unsigned int verbose;
 	unsigned int style;
 	unsigned int mass;
+	unsigned int cpu_run;
 	unsigned int threshold_load;
 	unsigned int threshold_load_r;
 	unsigned int threshold_load_d;
@@ -42,6 +43,16 @@ struct load_monitor_task {
 	struct diag_proc_chains_detail proc_chains;
 };
 
+struct load_monitor_cpu_run {
+	int et_type;
+	unsigned long id;
+	unsigned long cpu;
+	struct timeval tv;
+	struct diag_task_detail task;
+	struct diag_kern_stack_detail kern_stack;
+	struct diag_user_stack_detail user_stack;
+};
+
 struct load_monitor_detail {
 	int et_type;
 	unsigned long id;