ucli perf software irq

12 files changed, 207 insertions, 84 deletions

diff --git a/source/Makefile b/source/Makefile
index c0fb1d6..5f0f0c4 100644
--- a/source/Makefile
+++ b/source/Makefile
@@ -7,6 +7,7 @@ variable_monitor-objs := module/monitor_kernel.o \
                          module/monitor_timer.o \
                          module/monitor_trace.o \
                          module/monitor_proc.o \
+                         module/monitor_perf.o \
                          buffer/trace_buffer.o \
                          buffer/variant_buffer.o \
 
diff --git a/source/module/monitor_kallsyms.c b/source/module/monitor_kallsyms.c
index 182412e..7111967 100644
--- a/source/module/monitor_kallsyms.c
+++ b/source/module/monitor_kallsyms.c
@@ -19,6 +19,9 @@ int (*orig_access_remote_vm)(struct mm_struct *mm, unsigned long addr,
                              void *buf, int len, unsigned int gup_flags);
 struct task_struct *(*orig_find_task_by_vpid)(pid_t nr);
 
+void (*orig_open_softirq)(int nr, void (*action)(struct softirq_action *));
+void (*orig_raise_softirq)(unsigned int nr);
+
 /**
  * @brief diag_kallsyms_lookup_name init
  * 
@@ -59,5 +62,7 @@ int init_orig_fun(void) {
       get_task_type); // get_task_type | this fun is not available on 5.17.15
   LOOKUP_SYMS_NORET(kernfs_name); // kernfs_name
   LOOKUP_SYMS_NORET(find_task_by_vpid);
+  LOOKUP_SYMS_NORET(open_softirq);
+  LOOKUP_SYMS_NORET(raise_softirq);
   return 0;
 }
diff --git a/source/module/monitor_kallsyms.h b/source/module/monitor_kallsyms.h
index c77e7bd..cab6893 100644
--- a/source/module/monitor_kallsyms.h
+++ b/source/module/monitor_kallsyms.h
@@ -43,4 +43,7 @@ extern int (*orig_access_remote_vm)(
     struct mm_struct *mm, unsigned long addr, void *buf, int len,
     unsigned int gup_flags); // read remote memory
 extern struct task_struct *(*orig_find_task_by_vpid)(
-    pid_t nr); // find task by pid
\ No newline at end of file
+    pid_t nr); // find task by pid
+
+extern void (*orig_open_softirq)(int nr, void (*action)(struct softirq_action *));
+extern void (*orig_raise_softirq)(unsigned int nr);
\ No newline at end of file
diff --git a/source/module/monitor_kernel.c b/source/module/monitor_kernel.c
index 2ee7838..2695266 100644
--- a/source/module/monitor_kernel.c
+++ b/source/module/monitor_kernel.c
@@ -10,6 +10,19 @@
 
 #define DEVICE_NAME "variable_monitor"
 
+#include <linux/interrupt.h>
+
+enum {
+    MY_SOFTIRQ = 11, // 这个值是示例，确保它没有被其他软中断使用
+};
+
+static void my_softirq_handler(struct softirq_action *action)
+{
+    // 这是软中断处理函数，它不能睡眠，必须快速执行
+    // 在这里调用 get_raw_stack
+    diag_task_all();
+}
+
 // for character device
 static dev_t dev_num;
 static struct cdev *watch_cdev;
@@ -177,6 +190,8 @@ int init_module(void) {
   // orig_X | buffer
   monitor_init();
 
+  orig_open_softirq(MY_SOFTIRQ, my_softirq_handler);
+
   return 0;
 }
 
diff --git a/source/module/monitor_kernel.h b/source/module/monitor_kernel.h
index e879c7f..ae927df 100644
--- a/source/module/monitor_kernel.h
+++ b/source/module/monitor_kernel.h
@@ -34,5 +34,9 @@ void clear_watch(pid_t pid);              // for release
 enum hrtimer_restart
 check_variable_cb(struct hrtimer *timer); // hrtimer callback
 
+// for test
 int diag_pid(int id);  // for test
-int diag_tgid(int id); // for test
\ No newline at end of file
+int diag_tgid(int id); // for test
+void diag_task_info(kernel_watch_timer *k_watch_timer);
+void diag_task_all(void);
+void diag_task_by_tgid(pid_t tgid);
\ No newline at end of file
diff --git a/source/module/monitor_kernel_lib.c b/source/module/monitor_kernel_lib.c
index 508e35b..5285807 100644
--- a/source/module/monitor_kernel_lib.c
+++ b/source/module/monitor_kernel_lib.c
@@ -2,6 +2,7 @@
 
 #include <linux/sched.h>
 #include <linux/stacktrace.h>
+#include <linux/interrupt.h>
 // #include <linux/sched/task.h>
 // #include <linux/sched/mm.h>
 
@@ -119,11 +120,14 @@ static void diag_tsk(struct task_struct *p, variable_monitor_task *tsk_info) {
   unsigned int nr_bt;
   // printk(KERN_INFO "diag_tsk\n");
   diag_task_brief(p, &tsk_info->task); // task brief
-  // diag_task_user_stack(p, &tsk_info->user_stack);         // user stack
-  nr_bt = diag_task_kern_stack(p, &tsk_info->kern_stack); // kernel stack
-  dump_proc_chains_argv(1, p, &mm_tree_struct,
-                        &tsk_info->proc_chains); // proc chains
-  diag_task_raw_stack(p, &tsk_info->raw_stack);  // raw stack
+  if (tsk_info->task.sys_task == 1) { // system task
+    nr_bt = diag_task_kern_stack(p, &tsk_info->kern_stack); // kernel stack
+  } else { // other task
+    nr_bt = diag_task_kern_stack(p, &tsk_info->kern_stack); // kernel stack
+    dump_proc_chains_argv(1, p, &mm_tree_struct,
+                          &tsk_info->proc_chains); // proc chains
+    diag_task_raw_stack(p, &tsk_info->raw_stack);  // raw stack
+  }
 }
 
 static void push_tskinfo_2_buffer_orig(variable_monitor_task *tsk_info,
@@ -203,31 +207,25 @@ static void clear_all_watch(void) {
   memset(kernel_wtimer_list, 0, sizeof(kernel_wtimer_list));
 }
 
-void diag_task_info(kernel_watch_timer *k_watch_timer) {
-  if (k_watch_timer->threshold_over_count <= 0) // if no threshold reached
-    return;
-
-  // printk(KERN_INFO "diag_task_info_work\n");
+static void diag_vm_record(kernel_watch_timer *k_watch_timer,
+                           unsigned char is_print) {
+  static variable_monitor_record vm_record;
 
-  struct task_struct *g, *p; // g: task group; p: task
+  kernel_watch_arg *kwarg;
+  int i;
   unsigned long flags;
-  unsigned long event_id = get_cycles();
 
-  static variable_monitor_task tsk_info = {0};
-  static variable_monitor_record vm_record = {0};
-  kernel_watch_arg *kwarg;
+  unsigned long event_id = get_cycles();
 
   vm_record.id = event_id;
   vm_record.et_type = VARIABLE_MONITOR_RECORD_TYPE;
   vm_record.tv = k_watch_timer->tv;
   vm_record.threshold_over_count = k_watch_timer->threshold_over_count;
 
-  int i;
   for (i = 0; i < vm_record.threshold_over_count; i++) {
     kwarg = &k_watch_timer->k_watch_args[k_watch_timer->threshold_buffer[i]];
     k_w_arg2threshold(kwarg, &vm_record.threshold_record[i]);
   }
-
   rcu_read_lock();
   diag_variant_buffer_spin_lock(&load_monitor_variant_buffer, flags);
 
@@ -236,21 +234,92 @@ void diag_task_info(kernel_watch_timer *k_watch_timer) {
   diag_variant_buffer_write_nolock(&load_monitor_variant_buffer, &vm_record,
                                    sizeof(variable_monitor_record));
   diag_variant_buffer_seal(&load_monitor_variant_buffer);
-  // diag_variant_buffer_spin_unlock(&load_monitor_variant_buffer, flags);
 
-  // for task info
+  diag_variant_buffer_spin_unlock(&load_monitor_variant_buffer, flags);
+  rcu_read_unlock();
+
+  if (is_print) {
+    printk(KERN_INFO "-----------variable monitor----------\n");
+    printk(KERN_INFO "threshold exceeded, Timestamp %lld, Delay %lld:\n",
+           vm_record.tv, ktime_get_real() - vm_record.tv);
+
+    for (i = 0; i < vm_record.threshold_over_count; i++) {
+      printk(
+          KERN_INFO
+          "\t: pid: %d, name: %s, ptr: %p, threshold:%lld, true_value:%lld\n",
+          vm_record.threshold_record[i].task_id,
+          vm_record.threshold_record[i]
+              .name, // Assuming name is a null-terminated string
+          vm_record.threshold_record[i].ptr,
+          vm_record.threshold_record[i].threshold,
+          vm_record.threshold_record[i].true_value);
+    }
+
+    printk(KERN_INFO "-------------------------------------\n");
+  }
+}
+
+void diag_task_by_tgid(pid_t tgid){
+
+  struct task_struct *tsk;
+  int ret;
+  unsigned long flags;
+
+  static variable_monitor_task tsk_info;
+
+  tsk = NULL;
+  rcu_read_lock();
+  tsk = NULL;
+  if (orig_find_task_by_vpid)
+    tsk = orig_find_task_by_vpid(tgid);
+  if (!tsk) {
+    ret = -EINVAL;
+    rcu_read_unlock();
+    return ;
+  }
+
+  diag_variant_buffer_spin_lock(&load_monitor_variant_buffer, flags);
+
+  struct task_struct *thread = tsk;
+  unsigned long event_id = get_cycles();
+
+  while_each_thread(tsk, thread) {
+
+    tsk_info.et_type = VARIABLE_MONITOR_TASK_TYPE;
+    tsk_info.id = event_id;
+    tsk_info.tv = ktime_get_real();
+    diag_tsk(tsk, &tsk_info);
+
+    push_tskinfo_22_buffer(&tsk_info, &flags); // push to buffer
+  }
+  diag_variant_buffer_spin_unlock(&load_monitor_variant_buffer, flags);
+  rcu_read_unlock();
+}
+
+void diag_task_all(void){
+  pr_info("diag_task_all, tv %lld\n", ktime_get_real());
+
+  static variable_monitor_task tsk_info;
+
+  unsigned long event_id = get_cycles();
+  struct task_struct *g, *p; // g: task group; p: task
+  unsigned long flags;
+
+  rcu_read_lock();
+  diag_variant_buffer_spin_lock(&load_monitor_variant_buffer, flags);
+
   do_each_thread(g, p) {
     if (p->__state == TASK_RUNNING || __task_contributes_to_load(p) ||
         ((READ_ONCE(p->__state) & TASK_IDLE) != 0)) {
 
-      get_task_struct(p); // count +1
+      // get_task_struct(p); // count +1
 
       tsk_info.et_type = VARIABLE_MONITOR_TASK_TYPE;
       tsk_info.id = event_id;
-      tsk_info.tv = vm_record.tv;
+      tsk_info.tv = ktime_get_real();
       diag_tsk(p, &tsk_info);
 
-      put_task_struct(p); // count -1
+      // put_task_struct(p); // count -1
 
       push_tskinfo_22_buffer(&tsk_info, &flags); // push to buffer
     }
@@ -259,24 +328,14 @@ void diag_task_info(kernel_watch_timer *k_watch_timer) {
 
   diag_variant_buffer_spin_unlock(&load_monitor_variant_buffer, flags);
   rcu_read_unlock();
+  pr_info("diag_task_all, finish tv %lld\n", ktime_get_real());
+}
 
-  printk(KERN_INFO "-----------variable monitor----------\n");
-  printk(KERN_INFO "threshold exceeded, Timestamp %lld, Delay %lld:\n",
-         vm_record.tv, ktime_get_real() - vm_record.tv);
-
-  for (i = 0; i < vm_record.threshold_over_count; i++) {
-    printk(KERN_INFO
-           "\t: pid: %d, name: %s, ptr: %p, threshold:%lld, true_value:%lld\n",
-           vm_record.threshold_record[i].task_id,
-           vm_record.threshold_record[i]
-               .name, // Assuming name is a null-terminated string
-           vm_record.threshold_record[i].ptr,
-           vm_record.threshold_record[i].threshold,
-           vm_record.threshold_record[i].true_value);
-  }
-
-  printk(KERN_INFO "-------------------------------------\n");
-  return;
+void diag_task_info(kernel_watch_timer *k_watch_timer) {
+  if (k_watch_timer->threshold_over_count <= 0) // if no threshold reached
+    return;
+  diag_task_all();
+  diag_vm_record(k_watch_timer, 1);
 }
 
 /**
@@ -400,6 +459,10 @@ void clear_watch(pid_t pid) {
   start_all_hrTimer(); // restart timer
 }
 
+// enum {
+//     MY_SOFTIRQ = 11, // 这个值是示例，确保它没有被其他软中断使用
+// };
+
 /**
  * @brief main callback function
  *
@@ -429,9 +492,13 @@ enum hrtimer_restart check_variable_cb(struct hrtimer *timer) {
   {
     k_watch_timer->threshold_over_count = j;
     k_watch_timer->tv = ktime_get_real();
+    pr_info("threshold reached, tv %lld\n", k_watch_timer->tv);
     // highpri_wq
-    queue_work(system_highpri_wq, &k_watch_timer->wk);
+    // queue_work(system_highpri_wq, &k_watch_timer->wk);
     diag_task_info(k_watch_timer);
+    
+    // orig_raise_softirq(MY_SOFTIRQ); // for test
+
     // restart timer after dump_reset_sec sec
     hrtimer_forward(timer, timer->base->get_time(),
                     ktime_set(dump_reset_sec, 0));
diff --git a/source/module/monitor_perf.c b/source/module/monitor_perf.c
index 549d7ac..76df6f3 100644
--- a/source/module/monitor_perf.c
+++ b/source/module/monitor_perf.c
@@ -3,13 +3,15 @@
 static struct perf_event *pe;
 
 void vm_perf_overflow_callback(struct perf_event *event,
-                             struct perf_sample_data *data,
-                             struct pt_regs *regs) {
+                               struct perf_sample_data *data,
+                               struct pt_regs *regs) {
   // handle perf event data
   // struct perf_callchain_entry *callchain;
   // int nr, i;
 
   pr_info("perf event callback\n");
+  
+  // perf_event_disable(event);
 
   // 如果 perf_sample_data 有调用堆栈信息
   // if (data->callchain) {
@@ -42,10 +44,10 @@ void vm_perf_overflow_callback(struct perf_event *event,
 struct perf_event_attr pea = {
     .type = PERF_TYPE_SOFTWARE,
     .size = sizeof(struct perf_event_attr),
-    .config = PERF_COUNT_SW_CPU_CLOCK,
+    .config = PERF_COUNT_SW_DUMMY,
     .sample_period = 1,
     .sample_type = PERF_SAMPLE_CALLCHAIN,
-    // .disabled = 1,
+    .disabled = 1,
 };
 
 #include <linux/cpumask.h>
@@ -61,32 +63,38 @@ void setup_perf_event_for_task(struct task_struct *tsk) {
     pr_info("Perf event already created\n");
     return;
   }
-  int cpu;
-  struct perf_event **events;
-  for_each_possible_cpu(cpu) {
-    struct perf_event **event = per_cpu_ptr(events, cpu);
-    if (cpu_is_offline(cpu)) {
-      *event = NULL;
-      continue;
-    }
-    *event = perf_event_create_kernel_counter(&pea, cpu, tsk,
-                                              vm_perf_overflow_callback, NULL);
-    if (IS_ERR(*event)) {
-      printk(KERN_INFO "create perf event failure\n");
-      // return -1;
-    }
-  }
+  // int cpu;
+  // struct perf_event **events;
+  // for_each_possible_cpu(cpu) {
+  //   struct perf_event **event = per_cpu_ptr(events, cpu);
+  //   if (cpu_is_offline(cpu)) {
+  //     pr_info("cpu %d is offline\n", cpu);
+  //     *event = NULL;
+  //     continue;
+  //   }
+  //   *event = perf_event_create_kernel_counter(&pea, cpu, tsk,
+  //                                             vm_perf_overflow_callback, NULL);
+
+  //   // perf_event_create_kernel_counter(&pea, cpu, tsk,
+  //   // vm_perf_overflow_callback,
+  //   //                                  NULL);
+  //   if (IS_ERR(*event)) {
+  //     printk(KERN_INFO "create perf event failure\n");
+  //     // return -1;
+  //   }
+  //   // perf_event_enable(*event);
+  // }
 
   // pe = perf_event_create_kernel_counter(&pea, tsk->on_cpu, tsk,
   //                                         vm_perf_callback, NULL);
-  pe = perf_event_create_kernel_counter(&pea, tsk->on_cpu, tsk,
+  pe = perf_event_create_kernel_counter(&pea, -1, tsk,
                                         vm_perf_overflow_callback, NULL);
   if (IS_ERR(pe)) {
     pr_info("Error in perf_event_create_kernel_counter\n");
     return;
   }
 
-  // perf_event_enable(pe); // enable perf event
+  perf_event_enable(pe); // enable perf event
 }
 
 /**
diff --git a/source/module/monitor_proc.c b/source/module/monitor_proc.c
index 9603d5b..f7960f8 100644
--- a/source/module/monitor_proc.c
+++ b/source/module/monitor_proc.c
@@ -8,6 +8,7 @@
 const char* proc_dir = "variable_monitor";
 int def_interval_ns = DEFAULT_INTERVAL_NS;
 int dump_reset_sec = DEFAULT_DUMP_RESET_SEC;
+int sample_all = DEFAULT_SAMPLE_ALL;
 
 static ssize_t read_proc(struct file *file, char __user *buf, size_t count,
                          loff_t *offset, int *var) {
@@ -54,6 +55,16 @@ static ssize_t write_proc_dump_reset_sec(struct file *file,
   return write_proc(file, buf, count, offset, &dump_reset_sec);
 }
 
+static ssize_t read_proc_sample_all(struct file *file, char __user *buf,
+                                    size_t count, loff_t *offset) {
+  return read_proc(file, buf, count, offset, &sample_all);
+}
+static ssize_t write_proc_sample_all(struct file *file,
+                                     const char __user *buf, size_t count,
+                                     loff_t *offset) {
+  return write_proc(file, buf, count, offset, &sample_all);
+}
+
 static const struct proc_ops proc_def_interval_ns_ops = {
     .proc_read = read_proc_def_interval_ns,
     .proc_write = write_proc_def_interval_ns,
@@ -64,6 +75,11 @@ static const struct proc_ops proc_dump_reset_sec_ops = {
     .proc_write = write_proc_dump_reset_sec,
 };
 
+static const struct proc_ops proc_sample_all_ops = {
+    .proc_read = read_proc_sample_all,
+    .proc_write = write_proc_sample_all,
+};
+
 int monitor_proc_init(void) {
   struct proc_dir_entry *dir;
 
@@ -75,6 +91,7 @@ int monitor_proc_init(void) {
 
   proc_create("def_interval_ns", 0666, dir, &proc_def_interval_ns_ops);
   proc_create("dump_reset_sec", 0666, dir, &proc_dump_reset_sec_ops);
+  proc_create("sample_all", 0666, dir, &proc_sample_all_ops);
 
   return 0;
 }
diff --git a/source/module/monitor_proc.h b/source/module/monitor_proc.h
index 33a3ead..8ed4efc 100644
--- a/source/module/monitor_proc.h
+++ b/source/module/monitor_proc.h
@@ -2,10 +2,12 @@
 #define MODULE_MONITOR_PROC_H
 
 #define DEFAULT_INTERVAL_NS 10000 // 10us
-#define DEFAULT_DUMP_RESET_SEC 60  // 60s
+#define DEFAULT_DUMP_RESET_SEC 10  // 60s
+#define DEFAULT_SAMPLE_ALL 0
 
 extern int def_interval_ns;
 extern int dump_reset_sec;
+extern int sample_all;
 
 int monitor_proc_init(void);
 int monitor_proc_exit(void);
diff --git a/source/module/monitor_trace.c b/source/module/monitor_trace.c
index 8a0d59a..d9344ed 100644
--- a/source/module/monitor_trace.c
+++ b/source/module/monitor_trace.c
@@ -135,17 +135,17 @@ static inline void save_stack_trace_user_remote(struct task_struct *tsk,
   const void __user *fp = (const void __user *)regs->bp;
   int count = 0;
 
-  if (in_atomic()) {
-    printk(KERN_INFO "save_stack_trace_user_remote %d: task in_atomic\n",
-           tsk->pid);
-    return;
-  }
-
-  if (irqs_disabled()) {
-    printk(KERN_INFO "save_stack_trace_user_remote %d: task in irqs_disabled\n",
-           tsk->pid);
-    return;
-  }
+  // if (in_atomic()) {
+  //   printk(KERN_INFO "save_stack_trace_user_remote %d: task in_atomic\n",
+  //          tsk->pid);
+  //   return;
+  // }
+
+  // if (irqs_disabled()) {
+  //   printk(KERN_INFO "save_stack_trace_user_remote %d: task in irqs_disabled\n",
+  //          tsk->pid);
+  //   return;
+  // }
 
   if (trace->nr_entries < trace->max_entries)
     trace->entries[trace->nr_entries++] = regs->ip;
@@ -288,15 +288,15 @@ static int diagnose_task_raw_stack_remote(struct task_struct *tsk, void *to,
   int ret;
   struct mm_struct *mm;
 
-  if (in_atomic()) {
-    // printk(KERN_INFO "task_raw_stack_remote %d in_atomic\n", tsk->pid);
-    return 0;
-  }
+  // if (in_atomic()) {
+  //   printk(KERN_INFO "task_raw_stack_remote %d in_atomic\n", tsk->pid);
+  //   return 0;
+  // }
 
-  if (irqs_disabled()) {
-    // printk(KERN_INFO "task_raw_stack_remote %d irqs_disabled\n", tsk->pid);
-    return 0;
-  }
+  // if (irqs_disabled()) {
+  //   printk(KERN_INFO "task_raw_stack_remote %d irqs_disabled\n", tsk->pid);
+  //   return 0;
+  // }
 
   if (in_atomic() || irqs_disabled()) {
     return 0;
diff --git a/source/module/monitor_trace.h b/source/module/monitor_trace.h
index 4098f1a..a0bc1c5 100644
--- a/source/module/monitor_trace.h
+++ b/source/module/monitor_trace.h
@@ -59,6 +59,7 @@ typedef struct {
   unsigned long sys_task;
   /**
    * 1->user mode 0->sys mode -1->unknown
+   * may not be accurate
    */
   unsigned long user_mode;
   char comm[TASK_COMM_LEN];
diff --git a/source/ucli/Makefile b/source/ucli/Makefile
index dd0bf10..ae7625d 100644
--- a/source/ucli/Makefile
+++ b/source/ucli/Makefile
@@ -1,6 +1,6 @@
 OUTPUT_DIR ?= .
 TARGET_EXE=ucli
-SOURCES=ucli.cc ucli-lib.cc unwind.cc symbol.cc accessors.cc elf.cc
+SOURCES=ucli.cc ucli-lib.cc unwind.cc symbol.cc accessors.cc elf.cc helpfun.cc
 OBJECTS=$(SOURCES:.cc=.o) 
 
 CFLAGS=-g -O0