From 6e2c81d4bb75860be1094608a2519f4ea5f0f631 Mon Sep 17 00:00:00 2001
From: Armin Luntzer <armin.luntzer@univie.ac.at>
Date: Tue, 22 Oct 2019 14:40:51 +0200
Subject: [PATCH] looks like EDF admission is somewhat working now
---
include/kernel/kthread.h | 2 -
init/main.c | 43 ++++++----
kernel/sched/edf.c | 178 ++++++++++++++++++++++++---------------
3 files changed, 137 insertions(+), 86 deletions(-)
diff --git a/include/kernel/kthread.h b/include/kernel/kthread.h
index a72c686..ab0c7a8 100644
--- a/include/kernel/kthread.h
+++ b/include/kernel/kthread.h
@@ -84,8 +84,6 @@ struct task_struct {
ktime total;
unsigned long slices;
- ktime first_wake;
- ktime first_dead;
/* Tasks may have a parent and any number of siblings or children.
* If the parent is killed or terminated, so are all siblings and
diff --git a/init/main.c b/init/main.c
index c78a8d8..0607103 100644
--- a/init/main.c
+++ b/init/main.c
@@ -113,7 +113,6 @@ int task0(void *data)
- printk("HELLO! %d\n", leon3_cpuid());
bzero(buf1, 64);
sys_irq = sysset_find_obj(sys_set, "/sys/irl/primary");
@@ -130,7 +129,15 @@ int task0(void *data)
c = xc;
sched_time = sched_last_time;
curr = atoi(buf1)/2;
- printk("%d %d %d %llu irq: %s %d per sec; sched %llu us %llu per call, calls %d cpu %d\n", a, b, c, ktime_get(), buf1, (curr -last), ktime_to_us(sched_last_time), sched_last_time /sched_ev, sched_ev - last_call, leon3_cpuid());
+ printk("%d %d %d %llu ", a, b, c, ktime_get());
+ printk("irq: %s %d per sec; ", buf1, (curr -last));
+// printk("sched %llu us ", ktime_to_us(sched_last_time));
+ printk("%llu per call ", sched_last_time /sched_ev);
+// printk("calls %d ", sched_ev - last_call);
+ printk("cpu %d", leon3_cpuid());
+
+ printk("\n");
+
last = curr;
last_call = sched_ev;
sched_yield();
@@ -285,12 +292,12 @@ int kernel_main(void)
#if 1
- t = kthread_create(task0, NULL, 0, "res");
+ t = kthread_create(task0, NULL, KTHREAD_CPU_AFFINITY_NONE, "res");
sched_get_attr(t, &attr);
attr.policy = SCHED_EDF;
- attr.period = ms_to_ktime(1000);
- attr.deadline_rel = ms_to_ktime(900);
- attr.wcet = ms_to_ktime(200);
+ attr.period = ms_to_ktime(100);
+ attr.deadline_rel = ms_to_ktime(90);
+ attr.wcet = ms_to_ktime(30);
sched_set_attr(t, &attr);
kthread_wake_up(t);
#endif
@@ -298,34 +305,34 @@ int kernel_main(void)
#if 1
- t = kthread_create(task1, NULL, 1, "task1");
+ t = kthread_create(task1, NULL, KTHREAD_CPU_AFFINITY_NONE, "task1");
sched_get_attr(t, &attr);
attr.policy = SCHED_EDF;
- attr.period = ms_to_ktime(500);
- attr.deadline_rel = ms_to_ktime(300);
- attr.wcet = ms_to_ktime(200);
+ attr.period = ms_to_ktime(50);
+ attr.deadline_rel = ms_to_ktime(40);
+ attr.wcet = ms_to_ktime(31);
sched_set_attr(t, &attr);
kthread_wake_up(t);
#endif
#if 1
- t = kthread_create(task2, NULL, 1, "task2");
+ t = kthread_create(task2, NULL, KTHREAD_CPU_AFFINITY_NONE, "task2");
sched_get_attr(t, &attr);
attr.policy = SCHED_EDF;
- attr.period = ms_to_ktime(400);
- attr.deadline_rel = ms_to_ktime(200);
- attr.wcet = ms_to_ktime(100);
+ attr.period = ms_to_ktime(40);
+ attr.deadline_rel = ms_to_ktime(25);
+ attr.wcet = ms_to_ktime(20);
sched_set_attr(t, &attr);
kthread_wake_up(t);
#endif
#if 1
- t = kthread_create(task3, NULL, 0, "task3");
+ t = kthread_create(task3, NULL, KTHREAD_CPU_AFFINITY_NONE, "task3");
sched_get_attr(t, &attr);
attr.policy = SCHED_EDF;
- attr.period = ms_to_ktime(800);
- attr.deadline_rel = ms_to_ktime(700);
- attr.wcet = ms_to_ktime(100);
+ attr.period = ms_to_ktime(80);
+ attr.deadline_rel = ms_to_ktime(70);
+ attr.wcet = ms_to_ktime(10);
sched_set_attr(t, &attr);
kthread_wake_up(t);
#endif
diff --git a/kernel/sched/edf.c b/kernel/sched/edf.c
index bc70a09..5617fb3 100644
--- a/kernel/sched/edf.c
+++ b/kernel/sched/edf.c
@@ -27,7 +27,7 @@ void sched_print_edf_list_internal(struct task_queue *tq, int cpu, ktime now)
ktime prev = 0;
ktime prevd = 0;
- printk("\nktime: %lld\n", ktime_to_us(now));
+ printk("\nktime: %lld\n", ktime_to_ms(now));
printk("S\tDeadline\tWakeup\t\tdelta W\tdelta P\tt_rem\ttotal\tslices\tName\twcet\tavg\n");
printk("---------------------------------------------------------------------------------------------------------------------------------\n");
list_for_each_entry_safe(tsk, tmp, &tq->run, node) {
@@ -51,11 +51,11 @@ void sched_print_edf_list_internal(struct task_queue *tq, int cpu, ktime now)
tsk->slices = 1;
printk("%c\t%lld\t\t%lld\t\t%lld\t%lld\t%lld\t%lld\t%d\t%s\t|\t%lld\t%lld\n",
- state, ktime_to_us(tsk->deadline), ktime_to_us(tsk->wakeup),
- ktime_to_us(rel_wait), ktime_to_us(rel_deadline), ktime_to_us(tsk->runtime), ktime_to_us(tsk->total),
+ state, ktime_to_ms(tsk->deadline), ktime_to_ms(tsk->wakeup),
+ ktime_to_ms(rel_wait), ktime_to_ms(rel_deadline), ktime_to_ms(tsk->runtime), ktime_to_ms(tsk->total),
tsk->slices, tsk->name,
- ktime_to_us(tsk->attr.wcet),
- ktime_to_us(tsk->total/tsk->slices));
+ ktime_to_ms(tsk->attr.wcet),
+ ktime_to_ms(tsk->total/tsk->slices));
prev = tsk->wakeup;
prevd = tsk->deadline;
@@ -89,12 +89,9 @@ void sched_print_edf_list_internal(struct task_queue *tq, int cpu, ktime now)
/**
- * @brief check if an EDF task can still execute given its deadline
+ * @brief check if an EDF task can still execute given its deadline and
+ * remaining runtime
*
- * @note effectively checks
- * wcet remaining runtime in slot
- * ------ < --------------------------
- * period remaining time to deadline
*
* @returns true if can still execute before deadline
*/
@@ -106,33 +103,17 @@ static inline bool schedule_edf_can_execute(struct task_struct *tsk, int cpu, kt
- if (tsk->runtime <= 0)
+ /* should to consider twice the min tick period for overhead */
+ if (tsk->runtime <= (tick_get_period_min_ns() << 1))
return false;
- if (ktime_before(tsk->deadline, now)) {
-#if 1
- sched_print_edf_list_internal(&tsk->sched->tq[cpu], cpu, now);
- printk("%s violated, %lld %lld, dead %lld wake %lld now %lld start %lld\n", tsk->name,
- tsk->runtime, ktime_us_delta(tsk->deadline, now),
- tsk->deadline, tsk->wakeup, now, tsk->exec_start);
- // sched_print_edf_list_internal(now);
- BUG();
-#endif
- return false;
- }
-
to_deadline = ktime_delta(tsk->deadline, now);
- if (to_deadline <= 0)
+ /* should to consider twice the min tick period for overhead */
+ if (to_deadline <= (tick_get_period_min_ns() << 1))
return false;
-
-
- if (tsk->attr.wcet * to_deadline < tsk->attr.period * tsk->runtime)
- return true;
-
-
- return false;
+ return true;
}
@@ -142,17 +123,17 @@ static inline void schedule_edf_reinit_task(struct task_struct *tsk, ktime now)
tsk->wakeup = ktime_add(tsk->wakeup, tsk->attr.period);
#if 1
- if (ktime_after(now, tsk->wakeup)) {
+ /* need FDIR procedure for this situation */
+ if (ktime_after(now, tsk->wakeup)){ /* deadline missed earlier? */
printk("%s delta %lld\n",tsk->name, ktime_us_delta(tsk->wakeup, now));
printk("%s violated, %lld %lld, dead %lld wake %lld now %lld start %lld\n", tsk->name,
tsk->runtime, ktime_us_delta(tsk->deadline, now),
tsk->deadline, tsk->wakeup, now, tsk->exec_start);
+ sched_print_edf_list_internal(&tsk->sched->tq[tsk->on_cpu], tsk->on_cpu, now);
+ BUG();
}
-
-
-
- BUG_ON(ktime_after(now, tsk->wakeup)); /* deadline missed earlier? */
#endif
+
tsk->deadline = ktime_add(tsk->wakeup, tsk->attr.deadline_rel);
tsk->runtime = tsk->attr.wcet;
@@ -166,7 +147,6 @@ static inline void schedule_edf_reinit_task(struct task_struct *tsk, ktime now)
#define MSG "SCHED_EDF: "
-
static ktime edf_hyperperiod(struct task_queue tq[], int cpu, const struct task_struct *task)
{
ktime lcm = 1;
@@ -227,16 +207,6 @@ static ktime edf_hyperperiod(struct task_queue tq[], int cpu, const struct task_
-
-
-
-
-
-
-
-
-
-
/**
* @brief EDF schedulability test
*
@@ -282,13 +252,13 @@ static ktime edf_hyperperiod(struct task_queue tq[], int cpu, const struct task_
*
* XXX function needs adaptation
*/
-
+#define UTIL_MAX 0.8
static int edf_schedulable(struct task_queue tq[], const struct task_struct *task)
{
struct task_struct *tsk = NULL;
struct task_struct *tmp;
- int cpu;
+ int cpu = -EINVAL;
double u = 0.0; /* utilisation */
@@ -323,7 +293,8 @@ static int edf_schedulable(struct task_queue tq[], const struct task_struct *tas
}
- if (util > 0.9)
+ printk("UTIL %g\n", util);
+ if (util > UTIL_MAX)
continue;
if (util > util_max) {
@@ -331,16 +302,24 @@ static int edf_schedulable(struct task_queue tq[], const struct task_struct *tas
cpu = i;
}
+
+ }
+ if (cpu == -EINVAL) {
+ printk("---- WILL NOT FIT ----\n");
+ return -EINVAL;
}
+
printk("best fit is %d\n", cpu);
} else {
cpu = task->on_cpu;
}
+retry:
/******/
-if (0)
+if (1)
{
+ int nogo = 0;
printk("\n\n\n");
ktime p;
ktime h;
@@ -373,8 +352,8 @@ if (0)
}
/* add tasks queued in run */
- if (!list_empty(&tq[cpu].wake)) {
- list_for_each_entry_safe(tsk, tmp, &tq[cpu].wake, node) {
+ if (!list_empty(&tq[cpu].run)) {
+ list_for_each_entry_safe(tsk, tmp, &tq[cpu].run, node) {
if (tsk->attr.period > max) {
t0 = tsk;
max = tsk->attr.period;
@@ -382,6 +361,7 @@ if (0)
}
}
+ printk("t0: %s (cpu %d)\n", t0->name, cpu);
h = p / t0->attr.period;
printk("Period factor %lld, duration %lld actual period: %lld\n", h, ktime_to_ms(p), ktime_to_ms(t0->attr.period));
@@ -394,7 +374,8 @@ if (0)
printk("max UH: %lld, UT: %lld\n", ktime_to_ms(uh), ktime_to_ms(ut));
-
+#if 0
+ /* XXX already implicitly subtracted! */
/* subtract longest period thread from head, its slices must always
* be used before the deadline
*/
@@ -406,7 +387,7 @@ if (0)
uh = uh - sh;
printk("%s UH: %lld, UT: %lld\n", t0->name, ktime_to_ms(uh), ktime_to_ms(ut));
printk("%s SH: %lld, ST: %lld\n", t0->name, ktime_to_ms(sh), ktime_to_ms(st));
-
+#endif
/* tasks queued in wakeup */
@@ -416,6 +397,7 @@ if (0)
if (tsk == t0)
continue;
+ printk("tsk wake: %s\n", tsk->name);
if (tsk->attr.deadline_rel <= t0->attr.deadline_rel) {
/* slots before deadline of T0 */
@@ -424,6 +406,7 @@ if (0)
shmin = sh;
if (sh > uh) {
printk("WARN: NOT SCHEDULABLE in head: %s\n", tsk->name);
+ nogo = 1;
}
uh = uh - sh;
}
@@ -435,6 +418,7 @@ if (0)
if (st > ut) {
printk("WARN: NOT SCHEDULABLE in tail: %s\n", tsk->name);
+ nogo = 1;
}
ut = ut - st;
@@ -454,6 +438,7 @@ if (0)
if (tsk == t0)
continue;
+ printk("tsk run: %s\n", tsk->name);
if (tsk->attr.deadline_rel <= t0->attr.deadline_rel) {
/* slots before deadline of T0 */
@@ -462,6 +447,7 @@ if (0)
shmin = sh;
if (sh > uh) {
printk("WARN: NOT SCHEDULABLE in head: %s\n", tsk->name);
+ nogo = 1;
}
uh = uh - sh;
}
@@ -473,6 +459,7 @@ if (0)
if (st > ut) {
printk("WARN: NOT SCHEDULABLE in tail: %s\n", tsk->name);
+ nogo = 1;
}
ut = ut - st;
@@ -488,6 +475,51 @@ if (0)
+ if (task != t0) {
+
+ if (task->attr.deadline_rel <= t0->attr.deadline_rel) {
+ printk("task: %s\n", task->name);
+
+ /* slots before deadline of T0 */
+ sh = h * task->attr.wcet * t0->attr.deadline_rel / task->attr.period;
+ if (sh < shmin)
+ shmin = sh;
+ if (sh > uh) {
+ printk("xWARN: NOT SCHEDULABLE in head: %s\n", task->name);
+ nogo = 1;
+ }
+ uh = uh - sh;
+ }
+
+ /* slots after deadline of T0 */
+ st = h * task->attr.wcet * f1 / task->attr.period;
+ if (st < stmin)
+ stmin = st;
+
+ if (st > ut) {
+ printk("xWARN: NOT SCHEDULABLE in tail: %s\n", task->name);
+ nogo = 1;
+ }
+
+ ut = ut - st;
+
+ printk("x %s UH: %lld, UT: %lld\n", task->name, ktime_to_ms(uh), ktime_to_ms(ut));
+
+ printk("x %s SH: %lld, ST: %lld\n", task->name, ktime_to_ms(sh), ktime_to_ms(st));
+
+ }
+
+ if (nogo == 1) {
+ if (cpu == 0) {
+ cpu = 1;
+ printk("RETRY\n");
+ goto retry;
+ } else {
+ printk("RETURN: I am NOT schedul-ableh: %f ", u);
+ return -EINVAL;
+ }
+ }
+
printk("\n\n\n");
}
@@ -515,14 +547,14 @@ if (0)
u += (double) (int32_t) tsk->attr.wcet / (double) (int32_t) tsk->attr.period;
}
- if (u > 0.9) {
+ if (u > UTIL_MAX) {
printk("I am NOT schedul-ableh: %f ", u);
BUG();
return -EINVAL;
printk("changed task mode to RR\n", u);
}
- printk("Utilisation: %g\n", u);
+ printk("Utilisation: %g\n---\n", u);
/* TODO check against projected interrupt rate, we really need a limit
@@ -659,6 +691,12 @@ static void edf_wake_next(struct task_queue *tq, int cpu, ktime now)
struct task_struct *t;
struct task_struct *prev = NULL;
+
+ ktime max = 0;
+
+
+ struct task_struct *after = NULL;
+
ktime wake;
@@ -668,12 +706,22 @@ static void edf_wake_next(struct task_queue *tq, int cpu, ktime now)
kthread_lock();
last = now;
-#if 0 /* OK */
+
+
list_for_each_entry_safe(task, tmp, &tq->run, node) {
- if (last < task->deadline)
- last = task->deadline;
+
+ if (max > task->attr.period)
+ continue;
+
+ max = task->attr.period;
+ after = task;
}
-#endif
+
+
+ if (after) {
+ last = ktime_add(after->wakeup, after->attr.period);
+ }
+
#if 1 /* better */
task = list_first_entry(&tq[cpu].wake, struct task_struct, node);
@@ -696,7 +744,7 @@ static void edf_wake_next(struct task_queue *tq, int cpu, ktime now)
list_for_each_entry_safe(t, tmp, &tq[cpu].run, node) {
- if (t->state == TASK_BUSY)
+ if (t->state != TASK_IDLE)
continue;
if (ktime_before (t->wakeup, now))
@@ -726,14 +774,12 @@ static void edf_wake_next(struct task_queue *tq, int cpu, ktime now)
task->state = TASK_IDLE;
/* initially furthest deadline as wakeup */
+ last = ktime_add(last, 5000000000ULL);
task->wakeup = ktime_add(last, task->attr.period);
task->deadline = ktime_add(task->wakeup, task->attr.deadline_rel);
- /* XXX unneeded, remove */
- task->first_wake = task->wakeup;
- task->first_dead = task->deadline;
- printk("%s now %lld, last %lld, wake at %lld dead at %lld\n", task->name, now, last, task->wakeup, task->deadline);
+// printk("%s now %lld, last %lld, wake at %lld dead at %lld\n", task->name, now, last, task->wakeup, task->deadline);
list_move_tail(&task->node, &tq[cpu].run);
kthread_unlock();
--
GitLab