Merge pull request #821 from albertxu216/develop

Cpu_watcher：增加ewma
linuxkerneltravel · Jun 7, 2024 · 42adaef · 42adaef
2 parents 9e6ed68 + b8e1249
commit 42adaef
Show file tree

Hide file tree

Showing 5 changed files with 278 additions and 141 deletions.
diff --git a/.github/workflows/ebpf_cpu_watcher.yml b/.github/workflows/ebpf_cpu_watcher.yml
@@ -31,3 +31,9 @@ jobs:
           cd eBPF_Supermarket/CPU_Subsystem/cpu_watcher/
           make 
           sudo ./cpu_watcher 
+          
+      - name: Run test_cpuwatcher
+        run: |
+          cd eBPF_Supermarket/CPU_Subsystem/cpu_watcher/test
+          make
+          ./test_cpuwatcher
diff --git a/eBPF_Supermarket/CPU_Subsystem/cpu_watcher/bpf/schedule_delay.bpf.c b/eBPF_Supermarket/CPU_Subsystem/cpu_watcher/bpf/schedule_delay.bpf.c
@@ -23,14 +23,14 @@
 char LICENSE[] SEC("license") = "Dual BSD/GPL";
 #define TASK_RUNNING			0x0000
 
-BPF_HASH(has_scheduled,struct proc_id, bool, 10240);
-BPF_HASH(enter_schedule,struct proc_id, struct schedule_event, 10240);
-BPF_ARRAY(sys_schedule,int,struct sum_schedule,1);
-
+BPF_HASH(has_scheduled,struct proc_id, bool, 10240);//记录该进程是否调度过
+BPF_HASH(enter_schedule,struct proc_id, struct schedule_event, 10240);//记录该进程上运行队列的时间
+BPF_ARRAY(sys_schedule,int,struct sum_schedule,1);//记录整个系统的调度延迟
+BPF_ARRAY(threshold_schedule,int,struct proc_schedule,10240);//记录每个进程的调度延迟
 
 SEC("tp_btf/sched_wakeup")
 int BPF_PROG(sched_wakeup, struct task_struct *p) {
-    pid_t pid = BPF_CORE_READ(p, pid);
+    pid_t pid = p->pid;
     int cpu = bpf_get_smp_processor_id();
     struct schedule_event *schedule_event;
     struct proc_id id= {};
@@ -56,7 +56,7 @@ int BPF_PROG(sched_wakeup, struct task_struct *p) {
 
 SEC("tp_btf/sched_wakeup_new")
 int BPF_PROG(sched_wakeup_new, struct task_struct *p) {
-    pid_t pid = BPF_CORE_READ(p, pid);
+    pid_t pid = p->pid;
     int cpu = bpf_get_smp_processor_id();
     struct proc_id id= {};
     u64 current_time = bpf_ktime_get_ns();
@@ -86,17 +86,17 @@ int BPF_PROG(sched_switch, bool preempt, struct task_struct *prev, struct task_s
     struct schedule_event *schedule_event;
     struct sum_schedule *sum_schedule;
     int key = 0;
-    struct proc_id next_id= {};
+    struct proc_id next_id = {};
     u64 delay;
     if (prev_state == TASK_RUNNING) {
-        struct proc_id prev_pd= {};
+        struct proc_id prev_pd = {};
         prev_pd.pid = prev_pid;
         if (prev_pid == 0) {
             prev_pd.cpu_id = prev_cpu;
-        }    
+        }
         schedule_event = bpf_map_lookup_elem(&enter_schedule, &prev_pd);
         if (!schedule_event) {
-            struct schedule_event schedule_event2 ;
+            struct schedule_event schedule_event2;
             bool issched = false;
             schedule_event2.pid = prev_pid;
             schedule_event2.count = 1;
@@ -113,44 +113,49 @@ int BPF_PROG(sched_switch, bool preempt, struct task_struct *prev, struct task_s
         next_id.cpu_id = next_cpu;
     }
     schedule_event = bpf_map_lookup_elem(&enter_schedule, &next_id);
-    if (!schedule_event)  return 0;
+    if (!schedule_event) return 0;
     issched = bpf_map_lookup_elem(&has_scheduled, &next_id);
-    if (!issched)  return 0;   
+    if (!issched) return 0;
     if (*issched) {
         schedule_event->count++;
     } else {
         *issched = true;
-    }   
+    }
     delay = current_time - schedule_event->enter_time;
+    struct proc_schedule proc_schedule;
+    proc_schedule.pid = next_pid;
+    proc_schedule.delay = delay;
+    bpf_probe_read_kernel_str(&proc_schedule.proc_name, sizeof(proc_schedule.proc_name), next->comm);
+    bpf_map_update_elem(&threshold_schedule, &key, &proc_schedule, BPF_ANY);
     sum_schedule = bpf_map_lookup_elem(&sys_schedule, &key);
     if (!sum_schedule) {
-        struct sum_schedule sum_schedule= {};
+        struct sum_schedule sum_schedule = {};
         sum_schedule.sum_count++;
         sum_schedule.sum_delay += delay;
-        if (delay > sum_schedule.max_delay){
+        if (delay > sum_schedule.max_delay) {
             sum_schedule.max_delay = delay;
-            if(next->pid!=0){
-                sum_schedule.pid_max = next->pid;
+            if (next->pid != 0) {
+                bpf_probe_read_kernel_str(&sum_schedule.proc_name_max, sizeof(sum_schedule.proc_name_max), next->comm);
             }
-        }else if (sum_schedule.min_delay == 0 || delay < sum_schedule.min_delay)
+        } else if (sum_schedule.min_delay == 0 || delay < sum_schedule.min_delay) {
             sum_schedule.min_delay = delay;
-            if(next->pid!=0){
-                sum_schedule.pid_min = next->pid;
+            if (next->pid != 0) {
+                bpf_probe_read_kernel_str(&sum_schedule.proc_name_min, sizeof(sum_schedule.proc_name_min), next->comm);
             }
+        }
         bpf_map_update_elem(&sys_schedule, &key, &sum_schedule, BPF_ANY);
     } else {
         sum_schedule->sum_count++;
         sum_schedule->sum_delay += delay;
-        if (delay > sum_schedule->max_delay){
+        if (delay > sum_schedule->max_delay) {
             sum_schedule->max_delay = delay;
-            if(next->pid!=0){
-                sum_schedule->pid_max = next->pid;
-            }
-        }else if (sum_schedule->min_delay == 0 || delay < sum_schedule->min_delay)
+            bpf_probe_read_kernel_str(&sum_schedule->proc_name_max, sizeof(sum_schedule->proc_name_max), next->comm);
+        } else if (sum_schedule->min_delay == 0 || delay < sum_schedule->min_delay) {
             sum_schedule->min_delay = delay;
-            if(next->pid!=0){
-                sum_schedule->pid_min = next->pid;
+            if (next->pid != 0) {
+                bpf_probe_read_kernel_str(&sum_schedule->proc_name_min, sizeof(sum_schedule->proc_name_min), next->comm);
             }
+        }
     }
     return 0;
 }
@@ -175,4 +180,4 @@ int sched_process_exit(void *ctx) {
         bpf_map_delete_elem(&has_scheduled, &id);
     }
     return 0;
-}
+}