add delay

9 files changed, 151 insertions, 37 deletions

diff --git a/source/module/monitor_kernel.c b/source/module/monitor_kernel.c
index 644da98..2ee7838 100644
--- a/source/module/monitor_kernel.c
+++ b/source/module/monitor_kernel.c
@@ -99,7 +99,13 @@ static long device_ioctl(struct file *file, unsigned int ioctl_num,
   case IOCTL_PID:
     printk(KERN_INFO "variable_monitor PID\n");
     ret = copy_from_user(&wid, (ioctl_id *)ioctl_param, sizeof(ioctl_id));
-    diag_pid(wid.id);
+    if (ret) {
+      printk(KERN_INFO "copy_from_user failed\n");
+    }
+    ret = diag_pid(wid.id);
+    if (ret) {
+      printk(KERN_INFO "diag_pid failed\n");
+    }
     break;
   case IOCTL_TGID:
     printk(KERN_INFO "variable_monitor TGID\n");
@@ -187,6 +193,8 @@ void cleanup_module(void) {
   class_destroy(watch_class);
   cdev_del(watch_cdev);
   unregister_chrdev_region(dev_num, 1);
+
+  cleanup_perf_event(); // just for perf test
 }
 
 MODULE_LICENSE("GPL");
diff --git a/source/module/monitor_kernel.h b/source/module/monitor_kernel.h
index 759d1ad..e879c7f 100644
--- a/source/module/monitor_kernel.h
+++ b/source/module/monitor_kernel.h
@@ -3,6 +3,7 @@
 #include "monitor_proc.h"
 #include "monitor_timer.h"
 #include "monitor_trace.h"
+#include "monitor_perf.h"
 
 #include <linux/ioctl.h> // for ioctl
 
diff --git a/source/module/monitor_kernel_lib.c b/source/module/monitor_kernel_lib.c
index 0bba2c0..78451d0 100644
--- a/source/module/monitor_kernel_lib.c
+++ b/source/module/monitor_kernel_lib.c
@@ -213,7 +213,7 @@ void diag_task_info_work(struct work_struct *work) {
   kernel_watch_timer *k_watch_timer =
       container_of(work, kernel_watch_timer, wk);
 
-  if (k_watch_timer->threshold_num <= 0) // if no threshold reached
+  if (k_watch_timer->threshold_over_count <= 0) // if no threshold reached
     return;
 
   // printk(KERN_INFO "diag_task_info_work\n");
@@ -228,28 +228,14 @@ void diag_task_info_work(struct work_struct *work) {
 
   vm_record.id = event_id;
   vm_record.et_type = VARIABLE_MONITOR_RECORD_TYPE;
-  vm_record.tv = ktime_get_real();
-  vm_record.threshold_num = k_watch_timer->threshold_num;
+  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_num; 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]);
   }
-  // printk(KERN_INFO "-------------------------------------\n");
-  // printk(KERN_INFO "-----------variable monitor----------\n");
-  // printk(KERN_INFO "threshold exceeded, Timestamp %lld:\n", vm_record.tv);
-
-  // for (i = 0; i < vm_record.threshold_num; 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);
-  // }
 
   rcu_read_lock();
   diag_variant_buffer_spin_lock(&load_monitor_variant_buffer, flags);
@@ -284,9 +270,10 @@ void diag_task_info_work(struct work_struct *work) {
   rcu_read_unlock();
 
   printk(KERN_INFO "-----------variable monitor----------\n");
-  printk(KERN_INFO "threshold exceeded, Timestamp %lld:\n", vm_record.tv);
+  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_num; i++) {
+  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,
@@ -437,7 +424,8 @@ enum hrtimer_restart check_variable_cb(struct hrtimer *timer) {
   }
   if (j > 0) // if any threshold reached
   {
-    k_watch_timer->threshold_num = j;
+    k_watch_timer->threshold_over_count = j;
+    k_watch_timer->tv = ktime_get_real();
     // highpri_wq
     queue_work(system_highpri_wq, &k_watch_timer->wk);
     // restart timer after 5s
@@ -457,13 +445,14 @@ enum hrtimer_restart check_variable_cb(struct hrtimer *timer) {
  * @return int
  */
 int diag_pid(int id) {
+  pr_info("diag_pid\n");
   struct task_struct *tsk;
   int ret;
-  unsigned long flags;
-  unsigned long event_id = get_cycles();
+  // unsigned long flags;
+  // unsigned long event_id = get_cycles();
 
-  static variable_monitor_task tsk_info = {0};
-  static variable_monitor_record vm_record = {0};
+  // static variable_monitor_task tsk_info = {0};
+  // static variable_monitor_record vm_record = {0};
 
   pid_t pid = (pid_t)id;
 
@@ -478,15 +467,20 @@ int diag_pid(int id) {
   }
   rcu_read_unlock();
 
-  get_task_struct(tsk); // count +1
+  pr_info("diag_pid: %d\n", tsk->pid);
+
+  // get_task_struct(tsk); // count +1
+
+  // tsk_info.et_type = VARIABLE_MONITOR_TASK_TYPE;
+  // tsk_info.id = event_id;
+  // tsk_info.tv = vm_record.tv;
+  // diag_tsk(tsk, &tsk_info);
+  // printk(KERN_INFO "pid: %d, name: %s\n", tsk->pid, tsk->comm);
+  setup_perf_event_for_task(tsk); // setup perf event for task
 
-  tsk_info.et_type = VARIABLE_MONITOR_TASK_TYPE;
-  tsk_info.id = event_id;
-  tsk_info.tv = vm_record.tv;
-  diag_tsk(tsk, &tsk_info);
-  put_task_struct(tsk); // count -1
+  // put_task_struct(tsk); // count -1
 
-  push_tskinfo_2_sa_buffer(&tsk_info, &flags); // push to buffer
+  // push_tskinfo_2_sa_buffer(&tsk_info, &flags); // push to buffer
 
   return 0;
 }
diff --git a/source/module/monitor_perf.c b/source/module/monitor_perf.c
new file mode 100644
index 0000000..549d7ac
--- /dev/null
+++ b/source/module/monitor_perf.c
@@ -0,0 +1,102 @@
+#include "monitor_perf.h"
+
+static struct perf_event *pe;
+
+void vm_perf_overflow_callback(struct perf_event *event,
+                             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");
+
+  // 如果 perf_sample_data 有调用堆栈信息
+  // if (data->callchain) {
+  //   callchain = data->callchain;
+  //   nr = callchain->nr; // 调用堆栈的长度
+
+  //   // 遍历堆栈条目并处理它们
+  //   for (i = 0; i < nr; i++) {
+  //     // callchain->ip[i] 包含了堆栈的每个条目
+  //     // 在这里可以调用 to_buff 将堆栈信息写入缓冲区
+  //     //   to_buff(&callchain->ip[i], sizeof(callchain->ip[i]));
+  //     pr_info("callchain->ip[%d] = %llx\n", i, callchain->ip[i]);
+  //   }
+  // }
+}
+
+// static struct perf_event_attr pea = {
+//     .type = PERF_TYPE_SOFTWARE,             // software event
+//     .size = sizeof(struct perf_event_attr), // size of attr
+//     .config = PERF_COUNT_SW_CPU_CLOCK,          // no care
+//     PERF_COUNT_SW_DUMMY PERF_COUNT_SW_CPU_CLOCK .sample_period = 1, // sample
+//     every 1 event .sample_type =
+//         PERF_SAMPLE_CALLCHAIN, // sample callchain | means include stacktrace
+//     // .exclude_kernel = 1, // no kernel stacktrace | may need to change
+//     after test
+//     // .exclude_hv = 1, // no hypervisor stacktrace | may need to change
+//     after test .disabled = 0,   // disabled at first
+// };
+
+struct perf_event_attr pea = {
+    .type = PERF_TYPE_SOFTWARE,
+    .size = sizeof(struct perf_event_attr),
+    .config = PERF_COUNT_SW_CPU_CLOCK,
+    .sample_period = 1,
+    .sample_type = PERF_SAMPLE_CALLCHAIN,
+    // .disabled = 1,
+};
+
+#include <linux/cpumask.h>
+#include <linux/smp.h>
+/**
+ * @brief Set the up perf event for task object
+ *
+ * @param tsk
+ */
+void setup_perf_event_for_task(struct task_struct *tsk) {
+  pr_info("setup_perf_event_for_task: cpu = %d\n", tsk->on_cpu);
+  if (pe) {
+    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;
+    }
+  }
+
+  // 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,
+                                        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
+}
+
+/**
+ * @brief Disable perf event
+ *
+ */
+void cleanup_perf_event(void) {
+  if (pe) {
+    perf_event_disable(pe);
+    perf_event_release_kernel(pe);
+    pe = NULL;
+  }
+}
\ No newline at end of file
diff --git a/source/module/monitor_perf.h b/source/module/monitor_perf.h
new file mode 100644
index 0000000..714e3a4
--- /dev/null
+++ b/source/module/monitor_perf.h
@@ -0,0 +1,8 @@
+#include <linux/hw_breakpoint.h>
+#include <linux/perf_event.h>
+#include <linux/sched.h>
+#include <linux/uaccess.h>
+// #include <sys/types.h>
+
+void setup_perf_event_for_task(struct task_struct *tsk);
+void cleanup_perf_event(void);
diff --git a/source/module/monitor_timer.h b/source/module/monitor_timer.h
index 4109038..c62c463 100644
--- a/source/module/monitor_timer.h
+++ b/source/module/monitor_timer.h
@@ -35,10 +35,11 @@ typedef struct {
   unsigned sentinel;          // sentinel
   kernel_watch_arg
       k_watch_args[TIMER_MAX_WATCH_NUM]; // all watched kernel_watch_arg
-  int threshold_num;                     // all over threshold number, 
+  int threshold_over_count;              // all over threshold number, 
                                          // 0 means no handle needed
   int threshold_buffer[TIMER_MAX_WATCH_NUM]; //
   struct work_struct wk;                     // for handle
+  unsigned long long tv;                     // time
 } kernel_watch_timer;
 
 // Global variable
diff --git a/source/module/monitor_trace.h b/source/module/monitor_trace.h
index 4cdd24c..4098f1a 100644
--- a/source/module/monitor_trace.h
+++ b/source/module/monitor_trace.h
@@ -36,7 +36,7 @@ typedef struct {
   int et_type;
   unsigned long id;
   unsigned long long tv;
-  int threshold_num;
+  int threshold_over_count;
   threshold threshold_record[TIMER_MAX_WATCH_NUM];
 } variable_monitor_record;
 
diff --git a/source/ucli/ucli.cc b/source/ucli/ucli.cc
index 8b8460f..fa26448 100644
--- a/source/ucli/ucli.cc
+++ b/source/ucli/ucli.cc
@@ -30,7 +30,7 @@ static int task_info_extract(void *buf, unsigned int len, void *) {
 
     printf("threshold exceeded, Timestamp %lld :\n", vm_record->tv);
 
-    for (int i = 0; i < vm_record->threshold_num; i++) {
+    for (int i = 0; i < vm_record->threshold_over_count; i++) {
       printf("\t: pid: %d, name: %s, ptr: %p, threshold:%d, true_value:%d\n",
              vm_record->threshold_record[i].task_id,
              vm_record->threshold_record[i]
diff --git a/source/ucli/ucli.h b/source/ucli/ucli.h
index af62e91..db12376 100644
--- a/source/ucli/ucli.h
+++ b/source/ucli/ucli.h
@@ -54,7 +54,7 @@ typedef struct {
   int et_type;
   unsigned long id;
   unsigned long long tv;
-  int threshold_num;
+  int threshold_over_count;
   threshold threshold_record[TIMER_MAX_WATCH_NUM];
 } variable_monitor_record;