LTTV_STATE_EXIT,
LTTV_STATE_ZOMBIE,
LTTV_STATE_WAIT,
- LTTV_STATE_RUN;
+ LTTV_STATE_RUN,
+ LTTV_STATE_DEAD;
static GQuark
LTTV_STATE_TRACEFILES,
if(unlikely(process->state->s == LTTV_STATE_EXIT)) {
process->state->s = LTTV_STATE_ZOMBIE;
+ process->state->change = s->parent.timestamp;
} else {
if(unlikely(state_out == 0)) process->state->s = LTTV_STATE_WAIT_CPU;
else process->state->s = LTTV_STATE_WAIT;
- } /* FIXME : we do not remove process here, because the kernel
- * still has them : they may be zombies. We need to know
- * exactly when release_task is executed on the PID to
- * know when the zombie is destroyed.
- */
- //else
- // exit_process(s, process);
-
- process->state->change = s->parent.timestamp;
+ process->state->change = s->parent.timestamp;
+ }
+
+ if(state_out == 32)
+ exit_process(s, process); /* EXIT_DEAD */
+ /* see sched.h for states */
}
process = ts->running_process[cpu] =
lttv_state_find_process_or_create(
/* Mathieu : it seems like the process might have been scheduled in before the
* fork, and, in a rare case, might be the current process. This might happen
- * in a SMP case where we don't have enough precision on the clocks */
-#if 0
+ * in a SMP case where we don't have enough precision on the clocks.
+ *
+ * Test reenabled after precision fixes on time. (Mathieu) */
+#if 0
zombie_process = lttv_state_find_process(ts, ANY_CPU, child_pid);
if(unlikely(zombie_process != NULL)) {
guint num_cpus = ltt_trace_get_num_cpu(ts->parent.t);
guint i;
for(i=0; i< num_cpus; i++) {
- g_assert(process != ts->running_process[i]);
+ g_assert(zombie_process != ts->running_process[i]);
}
exit_process(s, zombie_process);
child_pid, &s->parent.timestamp);
} else {
/* The process has already been created : due to time imprecision between
- * multiple CPUs : it has been scheduled in before creation.
+ * multiple CPUs : it has been scheduled in before creation. Note that we
+ * shouldn't have this kind of imprecision.
*
* Simply put a correct parent.
*/
+ g_assert(0); /* This is a problematic case : the process has been created
+ before the fork event */
child_process->ppid = process->pid;
}
if(likely(process != NULL)) {
/* release_task is happening at kernel level : we can now safely release
* the data structure of the process */
+ //This test is fun, though, as it may happen that
+ //at time t : CPU 0 : process_free
+ //at time t+150ns : CPU 1 : schedule out
+ //Clearly due to time imprecision, we disable it. (Mathieu)
+ //If this weird case happen, we have no choice but to put the
+ //Currently running process on the cpu to 0.
+ //I re-enable it following time precision fixes. (Mathieu)
+ //Well, in the case where an process is freed by a process on another CPU
+ //and still scheduled, it happens that this is the schedchange that will
+ //drop the last reference count. Do not free it here!
guint num_cpus = ltt_trace_get_num_cpu(ts->parent.t);
guint i;
for(i=0; i< num_cpus; i++) {
- g_assert(process != ts->running_process[i]);
+ //g_assert(process != ts->running_process[i]);
+ if(process == ts->running_process[i]) {
+ //ts->running_process[i] = lttv_state_find_process(ts, i, 0);
+ break;
+ }
}
- exit_process(s, process);
+ if(i == num_cpus) /* process is not scheduled */
+ exit_process(s, process);
}
return FALSE;
LTTV_STATE_ZOMBIE = g_quark_from_string("zombie");
LTTV_STATE_WAIT = g_quark_from_string("wait for I/O");
LTTV_STATE_RUN = g_quark_from_string("running");
+ LTTV_STATE_DEAD = g_quark_from_string("dead");
LTTV_STATE_TRACEFILES = g_quark_from_string("tracefiles");
LTTV_STATE_PROCESSES = g_quark_from_string("processes");
LTTV_STATE_PROCESS = g_quark_from_string("process");