git.ipfire.org Git - thirdparty/openembedded/openembedded-core-contrib.git/blob

   1 From 8e52fd71e693619f7a58de2692e59f0c826e9988 Mon Sep 17 00:00:00 2001
   2 From: Michael Jeanson <mjeanson@efficios.com>
   3 Date: Mon, 4 Apr 2022 13:52:57 -0400
   4 Subject: [PATCH 03/10] fix: sched/tracing: Don't re-read p->state when
   5  emitting sched_switch event (v5.18)
   6
   7 See upstream commit :
   8
   9   commit fa2c3254d7cfff5f7a916ab928a562d1165f17bb
  10   Author: Valentin Schneider <valentin.schneider@arm.com>
  11   Date:   Thu Jan 20 16:25:19 2022 +0000
  12
  13     sched/tracing: Don't re-read p->state when emitting sched_switch event
  14
  15     As of commit
  16
  17       c6e7bd7afaeb ("sched/core: Optimize ttwu() spinning on p->on_cpu")
  18
  19     the following sequence becomes possible:
  20
  21                           p->__state = TASK_INTERRUPTIBLE;
  22                           __schedule()
  23                             deactivate_task(p);
  24       ttwu()
  25         READ !p->on_rq
  26         p->__state=TASK_WAKING
  27                             trace_sched_switch()
  28                               __trace_sched_switch_state()
  29                                 task_state_index()
  30                                   return 0;
  31
  32     TASK_WAKING isn't in TASK_REPORT, so the task appears as TASK_RUNNING in
  33     the trace event.
  34
  35     Prevent this by pushing the value read from __schedule() down the trace
  36     event.
  37
  38 Upstream-Status: Backport
  39
  40 Change-Id: I46743cd006be4b4d573cae2d77df7d6d16744d04
  41 Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
  42 Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  43 ---
  44  include/instrumentation/events/sched.h | 88 +++++++++++++++++++++++---
  45  1 file changed, 78 insertions(+), 10 deletions(-)
  46
  47 diff --git a/include/instrumentation/events/sched.h b/include/instrumentation/events/sched.h
  48 index 91953a6f..339bec94 100644
  49 --- a/include/instrumentation/events/sched.h
  50 +++ b/include/instrumentation/events/sched.h
  51 @@ -20,7 +20,37 @@
  52  #ifndef _TRACE_SCHED_DEF_
  53  #define _TRACE_SCHED_DEF_
  54
  55 -#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,15,0))
  56 +#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,18,0))
  57 +
  58 +static inline long __trace_sched_switch_state(bool preempt,
  59 +               unsigned int prev_state,
  60 +               struct task_struct *p)
  61 +{
  62 +        unsigned int state;
  63 +
  64 +#ifdef CONFIG_SCHED_DEBUG
  65 +        BUG_ON(p != current);
  66 +#endif /* CONFIG_SCHED_DEBUG */
  67 +
  68 +        /*
  69 +         * Preemption ignores task state, therefore preempted tasks are always
  70 +         * RUNNING (we will not have dequeued if state != RUNNING).
  71 +         */
  72 +        if (preempt)
  73 +                return TASK_REPORT_MAX;
  74 +
  75 +        /*
  76 +         * task_state_index() uses fls() and returns a value from 0-8 range.
  77 +         * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
  78 +         * it for left shift operation to get the correct task->state
  79 +         * mapping.
  80 +         */
  81 +       state = __task_state_index(prev_state, p->exit_state);
  82 +
  83 +        return state ? (1 << (state - 1)) : state;
  84 +}
  85 +
  86 +#elif (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,15,0))
  87
  88  static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
  89  {
  90 @@ -321,43 +351,81 @@ LTTNG_TRACEPOINT_EVENT_INSTANCE(sched_wakeup_template, sched_wakeup_new,
  91  /*
  92   * Tracepoint for task switches, performed by the scheduler:
  93   */
  94 +
  95 +#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(5,18,0))
  96  LTTNG_TRACEPOINT_EVENT(sched_switch,
  97
  98 -#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,4,0))
  99         TP_PROTO(bool preempt,
 100 -                struct task_struct *prev,
 101 -                struct task_struct *next),
 102 +               unsigned int prev_state,
 103 +               struct task_struct *prev,
 104 +               struct task_struct *next),
 105
 106 -       TP_ARGS(preempt, prev, next),
 107 +       TP_ARGS(preempt, prev_state, prev, next),
 108 +
 109 +       TP_FIELDS(
 110 +               ctf_array_text(char, prev_comm, prev->comm, TASK_COMM_LEN)
 111 +               ctf_integer(pid_t, prev_tid, prev->pid)
 112 +               ctf_integer(int, prev_prio, prev->prio - MAX_RT_PRIO)
 113 +#ifdef CONFIG_LTTNG_EXPERIMENTAL_BITWISE_ENUM
 114 +               ctf_enum(task_state, long, prev_state, __trace_sched_switch_state(preempt, prev_state, prev))
 115  #else
 116 -       TP_PROTO(struct task_struct *prev,
 117 +               ctf_integer(long, prev_state, __trace_sched_switch_state(preempt, prev_state, prev))
 118 +#endif
 119 +               ctf_array_text(char, next_comm, next->comm, TASK_COMM_LEN)
 120 +               ctf_integer(pid_t, next_tid, next->pid)
 121 +               ctf_integer(int, next_prio, next->prio - MAX_RT_PRIO)
 122 +       )
 123 +)
 124 +
 125 +#elif (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,4,0))
 126 +
 127 +LTTNG_TRACEPOINT_EVENT(sched_switch,
 128 +
 129 +       TP_PROTO(bool preempt,
 130 +                struct task_struct *prev,
 131                  struct task_struct *next),
 132
 133 -       TP_ARGS(prev, next),
 134 -#endif /* #if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,4,0)) */
 135 +       TP_ARGS(preempt, prev, next),
 136
 137         TP_FIELDS(
 138                 ctf_array_text(char, prev_comm, prev->comm, TASK_COMM_LEN)
 139                 ctf_integer(pid_t, prev_tid, prev->pid)
 140                 ctf_integer(int, prev_prio, prev->prio - MAX_RT_PRIO)
 141 -#if (LTTNG_LINUX_VERSION_CODE >= LTTNG_KERNEL_VERSION(4,4,0))
 142  #ifdef CONFIG_LTTNG_EXPERIMENTAL_BITWISE_ENUM
 143                 ctf_enum(task_state, long, prev_state, __trace_sched_switch_state(preempt, prev))
 144  #else
 145                 ctf_integer(long, prev_state, __trace_sched_switch_state(preempt, prev))
 146  #endif
 147 +               ctf_array_text(char, next_comm, next->comm, TASK_COMM_LEN)
 148 +               ctf_integer(pid_t, next_tid, next->pid)
 149 +               ctf_integer(int, next_prio, next->prio - MAX_RT_PRIO)
 150 +       )
 151 +)
 152 +
 153  #else
 154 +
 155 +LTTNG_TRACEPOINT_EVENT(sched_switch,
 156 +
 157 +       TP_PROTO(struct task_struct *prev,
 158 +                struct task_struct *next),
 159 +
 160 +       TP_ARGS(prev, next),
 161 +
 162 +       TP_FIELDS(
 163 +               ctf_array_text(char, prev_comm, prev->comm, TASK_COMM_LEN)
 164 +               ctf_integer(pid_t, prev_tid, prev->pid)
 165 +               ctf_integer(int, prev_prio, prev->prio - MAX_RT_PRIO)
 166  #ifdef CONFIG_LTTNG_EXPERIMENTAL_BITWISE_ENUM
 167                 ctf_enum(task_state, long, prev_state, __trace_sched_switch_state(prev))
 168  #else
 169                 ctf_integer(long, prev_state, __trace_sched_switch_state(prev))
 170 -#endif
 171  #endif
 172                 ctf_array_text(char, next_comm, next->comm, TASK_COMM_LEN)
 173                 ctf_integer(pid_t, next_tid, next->pid)
 174                 ctf_integer(int, next_prio, next->prio - MAX_RT_PRIO)
 175         )
 176  )
 177 +#endif
 178
 179  /*
 180   * Tracepoint for a task being migrated:
 181 --
 182 2.19.1
 183