Merge branch 'develop' of github.com:liuslient/lmp into develop

.github/workflows/ebpf_cpu_watcher.yml

@@ -31,3 +31,10 @@ 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

.github/workflows/ebpf_net_watcher.yml

@@ -46,4 +46,6 @@ jobs:
           sudo timeout -s SIGINT 5 ./netwatcher -k -T || if [[ $? != 124 && $? != 0 ]];then exit $?;fi
           sudo timeout -s SIGINT 5 ./netwatcher -I || if [[ $? != 124 && $? != 0 ]];then exit $?;fi
           sudo timeout -s SIGINT 5 ./netwatcher -S || if [[ $? != 124 && $? != 0 ]];then exit $?;fi
+          sudo timeout -s SIGINT 5 ./netwatcher -D || if [[ $? != 124 && $? != 0 ]];then exit $?;fi
+          sudo timeout -s SIGINT 5 ./netwatcher -M || if [[ $? != 124 && $? != 0 ]];then exit $?;fi
         timeout-minutes: 5

eBPF_Supermarket/CPU_Subsystem/cpu_watcher/Makefile

@@ -44,6 +44,7 @@ ALL_LDFLAGS := $(LDFLAGS) $(EXTRA_LDFLAGS)
 
 APPS =cs_delay sar sc_delay preempt schedule_delay mq_delay
 TARGETS=cpu_watcher
+CONTROLLER := controller
 
 SRC_DIR = ./include
 
@@ -81,12 +82,12 @@ $(call allow-override,CC,$(CROSS_COMPILE)cc)
 $(call allow-override,LD,$(CROSS_COMPILE)ld)
 
 .PHONY: all
-all: $(TARGETS)
+all: $(CONTROLLER) $(TARGETS)
 
 .PHONY: clean
 clean:
 	$(call msg,CLEAN)
-	$(Q)rm -rf $(OUTPUT) $(TARGETS)
+	$(Q)rm -rf $(OUTPUT) $(TARGETS) $(CONTROLLER)
 
 $(OUTPUT) $(OUTPUT)/libbpf $(BPFTOOL_OUTPUT):
 	$(call msg,MKDIR,$@)
@@ -132,11 +133,19 @@ $(OUTPUT)/%.o: $(SRC_DIR)/%.c | $(OUTPUT)
 	$(call msg,CC,$@)
 	$(Q)$(CC) $(CFLAGS) $(INCLUDES) -c $< -o $@
 
+$(OUTPUT)/%.o: $(CONTROLLER).c | $(OUTPUT)
+	$(call msg,CC,$@)
+	$(Q)$(CC) $(CFLAGS) $(INCLUDES) -c $< -o $@
+
 $(OUTPUT)/$(TARGETS).o: $(TARGETS).c $(APPS) | $(OUTPUT)
 	$(call msg,CC,$@)
 	$(Q)$(CC) $(CFLAGS) $(INCLUDES) -c $(filter %.c,$^) -o $@
 
 # Build application binary
+$(CONTROLLER): %: $(OUTPUT)/%.o $(COMMON_OBJ) $(LIBBPF_OBJ) | $(OUTPUT)
+	$(call msg,BINARY,$@)
+	$(Q)$(CC) $^ $(ALL_LDFLAGS) -lstdc++ -lelf -lz -o $@
+
 $(TARGETS): %: $(OUTPUT)/%.o $(COMMON_OBJ) $(LIBBPF_OBJ) | $(OUTPUT)
 	$(call msg,BINARY,$@)
 	$(Q)$(CC) $(CFLAGS) $^ $(ALL_LDFLAGS) -lstdc++ -lelf -lz -o $@

eBPF_Supermarket/CPU_Subsystem/cpu_watcher/bpf/cs_delay.bpf.c

@@ -21,9 +21,10 @@
 #include "cpu_watcher.h"
 
 char LICENSE[] SEC("license") = "Dual BSD/GPL";
-
+const int ctrl_key = 0;
 //记录时间戳；
 BPF_ARRAY(start,int,u64,1);
+BPF_ARRAY(cs_ctrl_map,int,struct cs_ctrl,1);
 struct {
 	__uint(type, BPF_MAP_TYPE_RINGBUF);
 	__uint(max_entries, 256 * 1024);

eBPF_Supermarket/CPU_Subsystem/cpu_watcher/bpf/mq_delay.bpf.c

@@ -23,10 +23,11 @@
 
 
 char LICENSE[] SEC("license") = "Dual BSD/GPL";
-
+const int ctrl_key = 0;
 BPF_HASH(send_msg1,pid_t,struct send_events,1024);//记录pid->u_msg_ptr的关系；do_mq_timedsend入参
 BPF_HASH(send_msg2,u64,struct send_events,1024);//记录msg->time的关系；
 BPF_HASH(rcv_msg1,pid_t,struct rcv_events,1024);//记录pid->u_msg_ptr的关系；do_mq_timedsend入参
+BPF_ARRAY(mq_ctrl_map,int,struct mq_ctrl,1);
 struct {
 	__uint(type, BPF_MAP_TYPE_RINGBUF);
 	__uint(max_entries, 256 * 1024);

eBPF_Supermarket/CPU_Subsystem/cpu_watcher/bpf/schedule_delay.bpf.c

@@ -23,14 +23,20 @@
 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);
-
+const int ctrl_key = 0;
+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
+BPF_ARRAY(schedule_ctrl_map,int,struct schedule_ctrl,1);
 
 SEC("tp_btf/sched_wakeup")
 int BPF_PROG(sched_wakeup, struct task_struct *p) {
-    pid_t pid = BPF_CORE_READ(p, pid);
+    struct schedule_ctrl *sched_ctrl;
+	sched_ctrl = bpf_map_lookup_elem(&schedule_ctrl_map,&ctrl_key);
+	if(!sched_ctrl || !sched_ctrl->schedule_func)
+		return 0;
+    pid_t pid = p->pid;
     int cpu = bpf_get_smp_processor_id();
     struct schedule_event *schedule_event;
     struct proc_id id= {};
@@ -56,7 +62,11 @@ 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);
+    struct schedule_ctrl *sched_ctrl;
+	sched_ctrl = bpf_map_lookup_elem(&schedule_ctrl_map,&ctrl_key);
+	if(!sched_ctrl || !sched_ctrl->schedule_func)
+		return 0;
+    pid_t pid = p->pid;
     int cpu = bpf_get_smp_processor_id();
     struct proc_id id= {};
     u64 current_time = bpf_ktime_get_ns();
@@ -76,6 +86,11 @@ int BPF_PROG(sched_wakeup_new, struct task_struct *p) {
 
 SEC("tp_btf/sched_switch")
 int BPF_PROG(sched_switch, bool preempt, struct task_struct *prev, struct task_struct *next) {
+    struct schedule_ctrl *sched_ctrl;
+	sched_ctrl = bpf_map_lookup_elem(&schedule_ctrl_map,&ctrl_key);
+	if(!sched_ctrl || !sched_ctrl->schedule_func)
+		return 0;
+
     u64 current_time = bpf_ktime_get_ns();
     pid_t prev_pid = prev->pid;
     unsigned int prev_state = prev->__state;
@@ -86,17 +101,17 @@ 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;
@@ -113,50 +128,60 @@ 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;
 }
 
 SEC("tracepoint/sched/sched_process_exit")
 int sched_process_exit(void *ctx) {
+    struct schedule_ctrl *sched_ctrl;
+	sched_ctrl = bpf_map_lookup_elem(&schedule_ctrl_map,&ctrl_key);
+	if(!sched_ctrl || !sched_ctrl->schedule_func)
+		return 0;
+
     struct task_struct *p = (struct task_struct *)bpf_get_current_task();
     pid_t pid = BPF_CORE_READ(p, pid);
     int cpu = bpf_get_smp_processor_id();
@@ -175,4 +200,4 @@ int sched_process_exit(void *ctx) {
         bpf_map_delete_elem(&has_scheduled, &id);
     }
     return 0;
-}
+}