Fix: update sched prev_state instrumentation for upstream kernel
[lttng-modules.git] / instrumentation / events / lttng-module / sched.h
index fd78f24a8eedf6b4512d8ac2922467dd17792f86..5b4313a98983bc71cbc649c3e84812aca9921acd 100644 (file)
+/* SPDX-License-Identifier: GPL-2.0 */
 #undef TRACE_SYSTEM
 #define TRACE_SYSTEM sched
 
-#if !defined(_TRACE_SCHED_H) || defined(TRACE_HEADER_MULTI_READ)
-#define _TRACE_SCHED_H
+#if !defined(LTTNG_TRACE_SCHED_H) || defined(TRACE_HEADER_MULTI_READ)
+#define LTTNG_TRACE_SCHED_H
 
+#include <probes/lttng-tracepoint-event.h>
 #include <linux/sched.h>
-#include <linux/tracepoint.h>
+#include <linux/pid_namespace.h>
+#include <linux/binfmts.h>
+#include <linux/version.h>
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0))
+#include <linux/sched/rt.h>
+#endif
+
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,19,0))
+#define lttng_proc_inum ns.inum
+#else
+#define lttng_proc_inum proc_inum
+#endif
+
+#define LTTNG_MAX_PID_NS_LEVEL 32
 
 #ifndef _TRACE_SCHED_DEF_
 #define _TRACE_SCHED_DEF_
 
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,15,0))
+
+static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
+{
+        unsigned int state;
+
+#ifdef CONFIG_SCHED_DEBUG
+        BUG_ON(p != current);
+#endif /* CONFIG_SCHED_DEBUG */
+
+        /*
+         * Preemption ignores task state, therefore preempted tasks are always
+         * RUNNING (we will not have dequeued if state != RUNNING).
+         */
+        if (preempt)
+                return TASK_REPORT_MAX;
+
+        /*
+         * task_state_index() uses fls() and returns a value from 0-8 range.
+         * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
+         * it for left shift operation to get the correct task->state
+         * mapping.
+         */
+        state = task_state_index(p);
+
+        return state ? (1 << (state - 1)) : state;
+}
+
+#elif (LINUX_VERSION_CODE >= KERNEL_VERSION(4,14,0))
+
+static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
+{
+        unsigned int state;
+
+#ifdef CONFIG_SCHED_DEBUG
+        BUG_ON(p != current);
+#endif /* CONFIG_SCHED_DEBUG */
+
+        /*
+         * Preemption ignores task state, therefore preempted tasks are always
+         * RUNNING (we will not have dequeued if state != RUNNING).
+         */
+        if (preempt)
+                return TASK_REPORT_MAX;
+
+        /*
+         * __get_task_state() uses fls() and returns a value from 0-8 range.
+         * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
+         * it for left shift operation to get the correct task->state
+         * mapping.
+         */
+        state = __get_task_state(p);
+
+        return state ? (1 << (state - 1)) : state;
+}
+
+#elif (LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0))
+
+static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
+{
+#ifdef CONFIG_SCHED_DEBUG
+       BUG_ON(p != current);
+#endif /* CONFIG_SCHED_DEBUG */
+       /*
+        * Preemption ignores task state, therefore preempted tasks are always RUNNING
+        * (we will not have dequeued if state != RUNNING).
+        */
+       return preempt ? TASK_RUNNING | TASK_STATE_MAX : p->state;
+}
+
+#elif (LINUX_VERSION_CODE >= KERNEL_VERSION(3,19,0))
+
+static inline long __trace_sched_switch_state(struct task_struct *p)
+{
+       long state = p->state;
+
+#ifdef CONFIG_PREEMPT
+#ifdef CONFIG_SCHED_DEBUG
+       BUG_ON(p != current);
+#endif /* CONFIG_SCHED_DEBUG */
+       /*
+        * For all intents and purposes a preempted task is a running task.
+        */
+       if (preempt_count() & PREEMPT_ACTIVE)
+               state = TASK_RUNNING | TASK_STATE_MAX;
+#endif /* CONFIG_PREEMPT */
+
+       return state;
+}
+
+#elif (LINUX_VERSION_CODE >= KERNEL_VERSION(3,13,0))
+
+static inline long __trace_sched_switch_state(struct task_struct *p)
+{
+       long state = p->state;
+
+#ifdef CONFIG_PREEMPT
+       /*
+        * For all intents and purposes a preempted task is a running task.
+        */
+       if (task_preempt_count(p) & PREEMPT_ACTIVE)
+               state = TASK_RUNNING | TASK_STATE_MAX;
+#endif
+
+       return state;
+}
+
+#elif (LINUX_VERSION_CODE >= KERNEL_VERSION(3,2,0))
+
+static inline long __trace_sched_switch_state(struct task_struct *p)
+{
+       long state = p->state;
+
+#ifdef CONFIG_PREEMPT
+       /*
+        * For all intents and purposes a preempted task is a running task.
+        */
+       if (task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)
+               state = TASK_RUNNING | TASK_STATE_MAX;
+#endif
+
+       return state;
+}
+
+#else
+
 static inline long __trace_sched_switch_state(struct task_struct *p)
 {
        long state = p->state;
@@ -25,286 +166,341 @@ static inline long __trace_sched_switch_state(struct task_struct *p)
        return state;
 }
 
+#endif
+
 #endif /* _TRACE_SCHED_DEF_ */
 
 /*
  * Tracepoint for calling kthread_stop, performed to end a kthread:
  */
-TRACE_EVENT(sched_kthread_stop,
+LTTNG_TRACEPOINT_EVENT(sched_kthread_stop,
 
        TP_PROTO(struct task_struct *t),
 
        TP_ARGS(t),
 
-       TP_STRUCT__entry(
-               __array_text(   char,   comm,   TASK_COMM_LEN   )
-               __field(        pid_t,  pid                     )
-       ),
-
-       TP_fast_assign(
-               tp_memcpy(comm, t->comm, TASK_COMM_LEN)
-               tp_assign(pid, t->pid)
-       ),
-
-       TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
+       TP_FIELDS(
+               ctf_array_text(char, comm, t->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, t->pid)
+       )
 )
 
 /*
  * Tracepoint for the return value of the kthread stopping:
  */
-TRACE_EVENT(sched_kthread_stop_ret,
+LTTNG_TRACEPOINT_EVENT(sched_kthread_stop_ret,
 
        TP_PROTO(int ret),
 
        TP_ARGS(ret),
 
-       TP_STRUCT__entry(
-               __field(        int,    ret     )
-       ),
-
-       TP_fast_assign(
-               tp_assign(ret, ret)
-       ),
-
-       TP_printk("ret=%d", __entry->ret)
+       TP_FIELDS(
+               ctf_integer(int, ret, ret)
+       )
 )
 
 /*
  * Tracepoint for waking up a task:
  */
-DECLARE_EVENT_CLASS(sched_wakeup_template,
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0) || \
+       LTTNG_RT_KERNEL_RANGE(4,1,10,11, 4,2,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,18,27,26, 3,19,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,14,61,63, 3,15,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,12,54,73, 3,13,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,10,97,106, 3,11,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,4,110,139, 3,5,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,2,77,111, 3,3,0,0))
+LTTNG_TRACEPOINT_EVENT_CLASS(sched_wakeup_template,
+
+       TP_PROTO(struct task_struct *p),
+
+       TP_ARGS(p),
+
+       TP_FIELDS(
+               ctf_array_text(char, comm, p->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, p->pid)
+               ctf_integer(int, prio, p->prio - MAX_RT_PRIO)
+               ctf_integer(int, target_cpu, task_cpu(p))
+       )
+)
+#else /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0)) */
+LTTNG_TRACEPOINT_EVENT_CLASS(sched_wakeup_template,
 
        TP_PROTO(struct task_struct *p, int success),
 
        TP_ARGS(p, success),
 
-       TP_STRUCT__entry(
-               __array_text(   char,   comm,   TASK_COMM_LEN   )
-               __field(        pid_t,  pid                     )
-               __field(        int,    prio                    )
-               __field(        int,    success                 )
-               __field(        int,    target_cpu              )
-       ),
+       TP_FIELDS(
+               ctf_array_text(char, comm, p->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, p->pid)
+               ctf_integer(int, prio, p->prio - MAX_RT_PRIO)
+               ctf_integer(int, success, success)
+               ctf_integer(int, target_cpu, task_cpu(p))
+       )
+)
+#endif /* #else #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0)) */
 
-       TP_fast_assign(
-               tp_memcpy(comm, p->comm, TASK_COMM_LEN)
-               tp_assign(pid, p->pid)
-               tp_assign(prio, p->prio)
-               tp_assign(success, success)
-               tp_assign(target_cpu, task_cpu(p))
-       ),
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0) || \
+       LTTNG_RT_KERNEL_RANGE(4,1,10,11, 4,2,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,18,27,26, 3,19,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,14,61,63, 3,15,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,12,54,73, 3,13,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,10,97,106, 3,11,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,4,110,139, 3,5,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(3,2,77,111, 3,3,0,0))
 
-       TP_printk("comm=%s pid=%d prio=%d success=%d target_cpu=%03d",
-                 __entry->comm, __entry->pid, __entry->prio,
-                 __entry->success, __entry->target_cpu)
-)
+/*
+ * Tracepoint called when waking a task; this tracepoint is guaranteed to be
+ * called from the waking context.
+ */
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_waking,
+            TP_PROTO(struct task_struct *p),
+            TP_ARGS(p))
 
-DEFINE_EVENT(sched_wakeup_template, sched_wakeup,
+/*
+ * Tracepoint called when the task is actually woken; p->state == TASK_RUNNNG.
+ * It it not always called from the waking context.
+ */
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup,
+            TP_PROTO(struct task_struct *p),
+            TP_ARGS(p))
+
+/*
+ * Tracepoint for waking up a new task:
+ */
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup_new,
+            TP_PROTO(struct task_struct *p),
+            TP_ARGS(p))
+
+#else
+
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup,
             TP_PROTO(struct task_struct *p, int success),
             TP_ARGS(p, success))
 
 /*
  * Tracepoint for waking up a new task:
  */
-DEFINE_EVENT(sched_wakeup_template, sched_wakeup_new,
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup_new,
             TP_PROTO(struct task_struct *p, int success),
             TP_ARGS(p, success))
 
+#endif /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,3,0)) */
+
 /*
  * Tracepoint for task switches, performed by the scheduler:
  */
-TRACE_EVENT(sched_switch,
+LTTNG_TRACEPOINT_EVENT(sched_switch,
+
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0))
+       TP_PROTO(bool preempt,
+                struct task_struct *prev,
+                struct task_struct *next),
 
+       TP_ARGS(preempt, prev, next),
+#else
        TP_PROTO(struct task_struct *prev,
                 struct task_struct *next),
 
        TP_ARGS(prev, next),
-
-       TP_STRUCT__entry(
-               __array_text(   char,   prev_comm,      TASK_COMM_LEN   )
-               __field(        pid_t,  prev_pid                        )
-               __field(        int,    prev_prio                       )
-               __field(        long,   prev_state                      )
-               __array_text(   char,   next_comm,      TASK_COMM_LEN   )
-               __field(        pid_t,  next_pid                        )
-               __field(        int,    next_prio                       )
-       ),
-
-       TP_fast_assign(
-               tp_memcpy(next_comm, next->comm, TASK_COMM_LEN)
-               tp_assign(prev_pid, prev->pid)
-               tp_assign(prev_prio, prev->prio)
-               tp_assign(prev_state, __trace_sched_switch_state(prev))
-               tp_memcpy(prev_comm, prev->comm, TASK_COMM_LEN)
-               tp_assign(next_pid, next->pid)
-               tp_assign(next_prio, next->prio)
-       ),
-
-       TP_printk("prev_comm=%s prev_pid=%d prev_prio=%d prev_state=%s ==> next_comm=%s next_pid=%d next_prio=%d",
-               __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
-               __entry->prev_state ?
-                 __print_flags(__entry->prev_state, "|",
-                               { 1, "S"} , { 2, "D" }, { 4, "T" }, { 8, "t" },
-                               { 16, "Z" }, { 32, "X" }, { 64, "x" },
-                               { 128, "W" }) : "R",
-               __entry->next_comm, __entry->next_pid, __entry->next_prio)
+#endif /* #if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0)) */
+
+       TP_FIELDS(
+               ctf_array_text(char, prev_comm, prev->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, prev_tid, prev->pid)
+               ctf_integer(int, prev_prio, prev->prio - MAX_RT_PRIO)
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0))
+               ctf_integer(long, prev_state, __trace_sched_switch_state(preempt, prev))
+#else
+               ctf_integer(long, prev_state, __trace_sched_switch_state(prev))
+#endif
+               ctf_array_text(char, next_comm, next->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, next_tid, next->pid)
+               ctf_integer(int, next_prio, next->prio - MAX_RT_PRIO)
+       )
 )
 
 /*
  * Tracepoint for a task being migrated:
  */
-TRACE_EVENT(sched_migrate_task,
+LTTNG_TRACEPOINT_EVENT(sched_migrate_task,
 
        TP_PROTO(struct task_struct *p, int dest_cpu),
 
        TP_ARGS(p, dest_cpu),
 
-       TP_STRUCT__entry(
-               __array_text(   char,   comm,   TASK_COMM_LEN   )
-               __field(        pid_t,  pid                     )
-               __field(        int,    prio                    )
-               __field(        int,    orig_cpu                )
-               __field(        int,    dest_cpu                )
-       ),
-
-       TP_fast_assign(
-               tp_memcpy(comm, p->comm, TASK_COMM_LEN)
-               tp_assign(pid, p->pid)
-               tp_assign(prio, p->prio)
-               tp_assign(orig_cpu, task_cpu(p))
-               tp_assign(dest_cpu, dest_cpu)
-       ),
-
-       TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d",
-                 __entry->comm, __entry->pid, __entry->prio,
-                 __entry->orig_cpu, __entry->dest_cpu)
+       TP_FIELDS(
+               ctf_array_text(char, comm, p->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, p->pid)
+               ctf_integer(int, prio, p->prio - MAX_RT_PRIO)
+               ctf_integer(int, orig_cpu, task_cpu(p))
+               ctf_integer(int, dest_cpu, dest_cpu)
+       )
 )
 
-DECLARE_EVENT_CLASS(sched_process_template,
+LTTNG_TRACEPOINT_EVENT_CLASS(sched_process_template,
 
        TP_PROTO(struct task_struct *p),
 
        TP_ARGS(p),
 
-       TP_STRUCT__entry(
-               __array_text(   char,   comm,   TASK_COMM_LEN   )
-               __field(        pid_t,  pid                     )
-               __field(        int,    prio                    )
-       ),
-
-       TP_fast_assign(
-               tp_memcpy(comm, p->comm, TASK_COMM_LEN)
-               tp_assign(pid, p->pid)
-               tp_assign(prio, p->prio)
-       ),
-
-       TP_printk("comm=%s pid=%d prio=%d",
-                 __entry->comm, __entry->pid, __entry->prio)
+       TP_FIELDS(
+               ctf_array_text(char, comm, p->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, p->pid)
+               ctf_integer(int, prio, p->prio - MAX_RT_PRIO)
+       )
 )
 
 /*
  * Tracepoint for freeing a task:
  */
-DEFINE_EVENT(sched_process_template, sched_process_free,
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_process_template, sched_process_free,
             TP_PROTO(struct task_struct *p),
             TP_ARGS(p))
-            
+
 
 /*
  * Tracepoint for a task exiting:
  */
-DEFINE_EVENT(sched_process_template, sched_process_exit,
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_process_template, sched_process_exit,
             TP_PROTO(struct task_struct *p),
             TP_ARGS(p))
 
 /*
  * Tracepoint for waiting on task to unschedule:
  */
-DEFINE_EVENT(sched_process_template, sched_wait_task,
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_process_template, sched_wait_task,
        TP_PROTO(struct task_struct *p),
        TP_ARGS(p))
 
 /*
  * Tracepoint for a waiting task:
  */
-TRACE_EVENT(sched_process_wait,
+LTTNG_TRACEPOINT_EVENT(sched_process_wait,
 
        TP_PROTO(struct pid *pid),
 
        TP_ARGS(pid),
 
-       TP_STRUCT__entry(
-               __array_text(   char,   comm,   TASK_COMM_LEN   )
-               __field(        pid_t,  pid                     )
-               __field(        int,    prio                    )
-       ),
-
-       TP_fast_assign(
-               tp_memcpy(comm, current->comm, TASK_COMM_LEN)
-               tp_assign(pid, pid_nr(pid))
-               tp_assign(prio, current->prio)
-       ),
-
-       TP_printk("comm=%s pid=%d prio=%d",
-                 __entry->comm, __entry->pid, __entry->prio)
+       TP_FIELDS(
+               ctf_array_text(char, comm, current->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, pid_nr(pid))
+               ctf_integer(int, prio, current->prio - MAX_RT_PRIO)
+       )
 )
 
 /*
- * Tracepoint for do_fork:
+ * Tracepoint for do_fork.
+ * Saving both TID and PID information, especially for the child, allows
+ * trace analyzers to distinguish between creation of a new process and
+ * creation of a new thread. Newly created processes will have child_tid
+ * == child_pid, while creation of a thread yields to child_tid !=
+ * child_pid.
  */
-TRACE_EVENT(sched_process_fork,
+LTTNG_TRACEPOINT_EVENT_CODE(sched_process_fork,
 
        TP_PROTO(struct task_struct *parent, struct task_struct *child),
 
        TP_ARGS(parent, child),
 
-       TP_STRUCT__entry(
-               __array_text(   char,   parent_comm,    TASK_COMM_LEN   )
-               __field(        pid_t,  parent_pid                      )
-               __array_text(   char,   child_comm,     TASK_COMM_LEN   )
-               __field(        pid_t,  child_pid                       )
+       TP_locvar(
+               pid_t vtids[LTTNG_MAX_PID_NS_LEVEL];
+               unsigned int ns_level;
        ),
 
-       TP_fast_assign(
-               tp_memcpy(parent_comm, parent->comm, TASK_COMM_LEN)
-               tp_assign(parent_pid, parent->pid)
-               tp_memcpy(child_comm, child->comm, TASK_COMM_LEN)
-               tp_assign(child_pid, child->pid)
+       TP_code_pre(
+               if (child) {
+                       struct pid *child_pid;
+                       unsigned int i;
+
+                       child_pid = task_pid(child);
+                       tp_locvar->ns_level =
+                               min_t(unsigned int, child_pid->level + 1,
+                                       LTTNG_MAX_PID_NS_LEVEL);
+                       for (i = 0; i < tp_locvar->ns_level; i++)
+                               tp_locvar->vtids[i] = child_pid->numbers[i].nr;
+               }
        ),
 
-       TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
-               __entry->parent_comm, __entry->parent_pid,
-               __entry->child_comm, __entry->child_pid)
+       TP_FIELDS(
+               ctf_array_text(char, parent_comm, parent->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, parent_tid, parent->pid)
+               ctf_integer(pid_t, parent_pid, parent->tgid)
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,8,0))
+               ctf_integer(unsigned int, parent_ns_inum,
+                       ({
+                               unsigned int parent_ns_inum = 0;
+
+                               if (parent) {
+                                       struct pid_namespace *pid_ns;
+
+                                       pid_ns = task_active_pid_ns(parent);
+                                       if (pid_ns)
+                                               parent_ns_inum =
+                                                       pid_ns->lttng_proc_inum;
+                               }
+                               parent_ns_inum;
+                       }))
+#endif
+               ctf_array_text(char, child_comm, child->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, child_tid, child->pid)
+               ctf_sequence(pid_t, vtids, tp_locvar->vtids, u8, tp_locvar->ns_level)
+               ctf_integer(pid_t, child_pid, child->tgid)
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,8,0))
+               ctf_integer(unsigned int, child_ns_inum,
+                       ({
+                               unsigned int child_ns_inum = 0;
+
+                               if (child) {
+                                       struct pid_namespace *pid_ns;
+
+                                       pid_ns = task_active_pid_ns(child);
+                                       if (pid_ns)
+                                               child_ns_inum =
+                                                       pid_ns->lttng_proc_inum;
+                               }
+                               child_ns_inum;
+                       }))
+#endif
+       ),
+
+       TP_code_post()
 )
 
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,4,0))
+/*
+ * Tracepoint for exec:
+ */
+LTTNG_TRACEPOINT_EVENT(sched_process_exec,
+
+       TP_PROTO(struct task_struct *p, pid_t old_pid,
+                struct linux_binprm *bprm),
+
+       TP_ARGS(p, old_pid, bprm),
+
+       TP_FIELDS(
+               ctf_string(filename, bprm->filename)
+               ctf_integer(pid_t, tid, p->pid)
+               ctf_integer(pid_t, old_tid, old_pid)
+       )
+)
+#endif
+
 /*
  * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
  *     adding sched_stat support to SCHED_FIFO/RR would be welcome.
  */
-DECLARE_EVENT_CLASS(sched_stat_template,
+LTTNG_TRACEPOINT_EVENT_CLASS(sched_stat_template,
 
        TP_PROTO(struct task_struct *tsk, u64 delay),
 
        TP_ARGS(tsk, delay),
 
-       TP_STRUCT__entry(
-               __array_text( char,     comm,   TASK_COMM_LEN   )
-               __field( pid_t, pid                     )
-               __field( u64,   delay                   )
-       ),
-
-       TP_fast_assign(
-               tp_memcpy(comm, tsk->comm, TASK_COMM_LEN)
-               tp_assign(pid,  tsk->pid)
-               tp_assign(delay, delay)
+       TP_FIELDS(
+               ctf_array_text(char, comm, tsk->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, tsk->pid)
+               ctf_integer(u64, delay, delay)
        )
-       TP_perf_assign(
-               __perf_count(delay)
-       ),
-
-       TP_printk("comm=%s pid=%d delay=%Lu [ns]",
-                       __entry->comm, __entry->pid,
-                       (unsigned long long)__entry->delay)
 )
 
 
@@ -312,7 +508,7 @@ DECLARE_EVENT_CLASS(sched_stat_template,
  * Tracepoint for accounting wait time (time the task is runnable
  * but not actually running due to scheduler contention).
  */
-DEFINE_EVENT(sched_stat_template, sched_stat_wait,
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_stat_template, sched_stat_wait,
             TP_PROTO(struct task_struct *tsk, u64 delay),
             TP_ARGS(tsk, delay))
 
@@ -320,7 +516,7 @@ DEFINE_EVENT(sched_stat_template, sched_stat_wait,
  * Tracepoint for accounting sleep time (time the task is not runnable,
  * including iowait, see below).
  */
-DEFINE_EVENT(sched_stat_template, sched_stat_sleep,
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_stat_template, sched_stat_sleep,
             TP_PROTO(struct task_struct *tsk, u64 delay),
             TP_ARGS(tsk, delay))
 
@@ -328,73 +524,78 @@ DEFINE_EVENT(sched_stat_template, sched_stat_sleep,
  * Tracepoint for accounting iowait time (time the task is not runnable
  * due to waiting on IO to complete).
  */
-DEFINE_EVENT(sched_stat_template, sched_stat_iowait,
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_stat_template, sched_stat_iowait,
+            TP_PROTO(struct task_struct *tsk, u64 delay),
+            TP_ARGS(tsk, delay))
+
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,3,0))
+/*
+ * Tracepoint for accounting blocked time (time the task is in uninterruptible).
+ */
+LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_stat_template, sched_stat_blocked,
             TP_PROTO(struct task_struct *tsk, u64 delay),
             TP_ARGS(tsk, delay))
+#endif
 
 /*
  * Tracepoint for accounting runtime (time the task is executing
  * on a CPU).
  */
-TRACE_EVENT(sched_stat_runtime,
+LTTNG_TRACEPOINT_EVENT(sched_stat_runtime,
 
        TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
 
        TP_ARGS(tsk, runtime, vruntime),
 
-       TP_STRUCT__entry(
-               __array_text( char,     comm,   TASK_COMM_LEN   )
-               __field( pid_t, pid                     )
-               __field( u64,   runtime                 )
-               __field( u64,   vruntime                        )
-       ),
-
-       TP_fast_assign(
-               tp_memcpy(comm, tsk->comm, TASK_COMM_LEN)
-               tp_assign(pid, tsk->pid)
-               tp_assign(runtime, runtime)
-               tp_assign(vruntime, vruntime)
+       TP_FIELDS(
+               ctf_array_text(char, comm, tsk->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, tsk->pid)
+               ctf_integer(u64, runtime, runtime)
+               ctf_integer(u64, vruntime, vruntime)
        )
-       TP_perf_assign(
-               __perf_count(runtime)
-       ),
-
-       TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]",
-                       __entry->comm, __entry->pid,
-                       (unsigned long long)__entry->runtime,
-                       (unsigned long long)__entry->vruntime)
 )
 
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(4,12,0) || \
+       LTTNG_RT_KERNEL_RANGE(4,9,27,18, 4,10,0,0) || \
+       LTTNG_RT_KERNEL_RANGE(4,11,5,1, 4,12,0,0))
 /*
  * Tracepoint for showing priority inheritance modifying a tasks
  * priority.
  */
-TRACE_EVENT(sched_pi_setprio,
+LTTNG_TRACEPOINT_EVENT(sched_pi_setprio,
 
-       TP_PROTO(struct task_struct *tsk, int newprio),
+       TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
 
-       TP_ARGS(tsk, newprio),
+       TP_ARGS(tsk, pi_task),
 
-       TP_STRUCT__entry(
-               __array_text( char,     comm,   TASK_COMM_LEN   )
-               __field( pid_t, pid                     )
-               __field( int,   oldprio                 )
-               __field( int,   newprio                 )
-       ),
+       TP_FIELDS(
+               ctf_array_text(char, comm, tsk->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, tsk->pid)
+               ctf_integer(int, oldprio, tsk->prio - MAX_RT_PRIO)
+               ctf_integer(int, newprio, pi_task ? pi_task->prio - MAX_RT_PRIO : tsk->prio - MAX_RT_PRIO)
+       )
+)
+#else
+/*
+ * Tracepoint for showing priority inheritance modifying a tasks
+ * priority.
+ */
+LTTNG_TRACEPOINT_EVENT(sched_pi_setprio,
 
-       TP_fast_assign(
-               tp_memcpy(comm, tsk->comm, TASK_COMM_LEN)
-               tp_assign(pid, tsk->pid)
-               tp_assign(oldprio, tsk->prio)
-               tp_assign(newprio, newprio)
-       ),
+       TP_PROTO(struct task_struct *tsk, int newprio),
+
+       TP_ARGS(tsk, newprio),
 
-       TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
-                       __entry->comm, __entry->pid,
-                       __entry->oldprio, __entry->newprio)
+       TP_FIELDS(
+               ctf_array_text(char, comm, tsk->comm, TASK_COMM_LEN)
+               ctf_integer(pid_t, tid, tsk->pid)
+               ctf_integer(int, oldprio, tsk->prio - MAX_RT_PRIO)
+               ctf_integer(int, newprio, newprio - MAX_RT_PRIO)
+       )
 )
+#endif
 
-#endif /* _TRACE_SCHED_H */
+#endif /* LTTNG_TRACE_SCHED_H */
 
 /* This part must be outside protection */
-#include "../../../probes/define_trace.h"
+#include <probes/define_trace.h>
This page took 0.031498 seconds and 4 git commands to generate.