[PATCH][2/3] Sched batch for staircase

From: Con Kolivas
Date: Tue Aug 03 2004 - 19:51:57 EST

Next message: Andi Kleen: "Re: XFS: how to NOT null files on fsck?"
Previous message: Con Kolivas: "[PATCH][1/3] Schedrange"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

This implements idle scheduling for the staircase scheduler.

SCHED_BATCH tasks are scheduled only when idle time is available, or transiently elevated to normal scheduling while in kernel mode.
Batch tasks have 10 times larger timeslices than normal policy tasks, so in compute mode (which already enlarges them 10 times) a nice 0 batch task has a 20 second long timeslice.
Nice is supported between batch tasks.
Batch tasks are hyperthread aware and will not schedule if an SMT sibling is not idle or not running a batch tasks.

Schedtools and the heirloom toolchest currently have userspace support for the setting of sched_batch policy.

Signed-off-by: Con Kolivas <kernel@xxxxxxxxxxx>

Index: linux-2.6.8-rc2-mm2/include/linux/init_task.h
===================================================================
--- linux-2.6.8-rc2-mm2.orig/include/linux/init_task.h 2004-08-03 01:52:34.388711380 +1000
+++ linux-2.6.8-rc2-mm2/include/linux/init_task.h 2004-08-03 01:54:34.136953777 +1000
@@ -72,8 +72,8 @@
.usage = ATOMIC_INIT(2), \
.flags = 0, \
.lock_depth = -1, \
- .prio = MAX_PRIO-20, \
- .static_prio = MAX_PRIO-20, \
+ .prio = MAX_PRIO-21, \
+ .static_prio = MAX_PRIO-21, \
.policy = SCHED_NORMAL, \
.cpus_allowed = CPU_MASK_ALL, \
.mm = NULL, \
Index: linux-2.6.8-rc2-mm2/include/linux/sched.h
===================================================================
--- linux-2.6.8-rc2-mm2.orig/include/linux/sched.h 2004-08-03 01:54:32.489156849 +1000
+++ linux-2.6.8-rc2-mm2/include/linux/sched.h 2004-08-03 01:54:34.156951313 +1000
@@ -126,9 +126,10 @@
#define SCHED_NORMAL 0
#define SCHED_FIFO 1
#define SCHED_RR 2
+#define SCHED_BATCH 3

#define SCHED_MIN 0
-#define SCHED_MAX 2
+#define SCHED_MAX 3

#define SCHED_RANGE(policy) ((policy) >= SCHED_MIN && \
(policy) <= SCHED_MAX)
@@ -327,9 +328,10 @@
#define MAX_USER_RT_PRIO 100
#define MAX_RT_PRIO MAX_USER_RT_PRIO

-#define MAX_PRIO (MAX_RT_PRIO + 40)
+#define MAX_PRIO (MAX_RT_PRIO + 41)

-#define rt_task(p) (unlikely((p)->prio < MAX_RT_PRIO))
+#define rt_task(p) ((p)->prio < MAX_RT_PRIO)
+#define batch_task(p) ((p)->policy == SCHED_BATCH)

/*
* Some day this will be a full-fledged user tracking system..
Index: linux-2.6.8-rc2-mm2/init/main.c
===================================================================
--- linux-2.6.8-rc2-mm2.orig/init/main.c 2004-08-03 01:29:29.000000000 +1000
+++ linux-2.6.8-rc2-mm2/init/main.c 2004-08-03 01:54:50.729908885 +1000
@@ -685,7 +685,7 @@
static int init(void * unused)
{
lock_kernel();
- current->prio = MAX_PRIO - 1;
+ current->prio = MAX_PRIO - 2;
/*
* Tell the world that we're going to be the grim
* reaper of innocent orphaned children.
Index: linux-2.6.8-rc2-mm2/kernel/sched.c
===================================================================
--- linux-2.6.8-rc2-mm2.orig/kernel/sched.c 2004-08-03 01:54:32.490156726 +1000
+++ linux-2.6.8-rc2-mm2/kernel/sched.c 2004-08-03 01:54:34.158951066 +1000
@@ -49,7 +49,7 @@

/*
* Convert user-nice values [ -20 ... 0 ... 19 ]
- * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ],
+ * to static priority [ MAX_RT_PRIO..MAX_PRIO-2 ],
* and back.
*/
#define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20)
@@ -59,7 +59,7 @@
/*
* 'User priority' is the nice value converted to something we
* can work with better when scaling various scheduler parameters,
- * it's a [ 0 ... 39 ] range.
+ * it's a [ 0 ... 40 ] range.
*/
#define USER_PRIO(p) ((p)-MAX_RT_PRIO)
#define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio)
@@ -274,8 +274,10 @@
static unsigned int slice(task_t *p)
{
unsigned int slice = RR_INTERVAL();
- if (likely(!rt_task(p)))
+ if (likely(!rt_task(p) && !batch_task(p)))
slice += burst(p) * RR_INTERVAL();
+ else if (batch_task(p))
+ slice *= 40 - TASK_USER_PRIO(p);
return slice;
}

@@ -297,6 +299,17 @@
unsigned int best_burst;
if (rt_task(p))
return p->prio;
+ if (batch_task(p)) {
+ if (unlikely(p->flags & PF_UISLEEP)) {
+ /*
+ * If batch is waking up from uninterruptible sleep
+ * reschedule at a normal priority to begin with.
+ */
+ p->flags |= PF_YIELDED;
+ return MAX_PRIO - 2;
+ }
+ return MAX_PRIO - 1;
+ }

best_burst = burst(p);
full_slice = slice(p);
@@ -306,13 +319,13 @@
first_slice = RR_INTERVAL();
if (sched_interactive && !sched_compute)
first_slice *= (p->burst + 1);
- prio = MAX_PRIO - 1 - best_burst;
+ prio = MAX_PRIO - 2 - best_burst;

if (used_slice < first_slice)
return prio;
prio += 1 + (used_slice - first_slice) / RR_INTERVAL();
- if (prio > MAX_PRIO - 1)
- prio = MAX_PRIO - 1;
+ if (prio > MAX_PRIO - 2)
+ prio = MAX_PRIO - 2;
return prio;
}

@@ -361,9 +374,11 @@
#endif
p->slice = slice(p);
recalc_task_prio(p, now);
- p->flags &= ~PF_UISLEEP;
p->prio = effective_prio(p);
+ p->flags &= ~PF_UISLEEP;
p->time_slice = RR_INTERVAL();
+ if (batch_task(p))
+ p->time_slice = p->slice;
p->timestamp = now;
__activate_task(p, rq);
}
@@ -1717,7 +1732,7 @@
rebalance_tick(cpu, rq, IDLE);
return;
}
- if (TASK_NICE(p) > 0)
+ if (TASK_NICE(p) > 0 || batch_task(p))
cpustat->nice += user_ticks;
else
cpustat->user += user_ticks;
@@ -1820,8 +1835,9 @@
* physical cpu's resources. -ck
*/
if (((smt_curr->slice * (100 - sd->per_cpu_gain) / 100) >
- slice(p) || rt_task(smt_curr)) &&
- p->mm && smt_curr->mm && !rt_task(p))
+ slice(p) || rt_task(smt_curr) || batch_task(p)) &&
+ p->mm && smt_curr->mm && !rt_task(p) &&
+ !batch_task(smt_curr))
ret = 1;

/*
@@ -1830,8 +1846,9 @@
* reasons.
*/
if ((((p->slice * (100 - sd->per_cpu_gain) / 100) >
- slice(smt_curr) || rt_task(p)) &&
- smt_curr->mm && p->mm && !rt_task(smt_curr)) ||
+ slice(smt_curr) || rt_task(p) || batch_task(smt_curr)) &&
+ smt_curr->mm && p->mm && !rt_task(smt_curr) &&
+ !batch_task(p)) ||
(smt_curr == smt_rq->idle && smt_rq->nr_running))
resched_task(smt_curr);
}
@@ -2229,8 +2246,9 @@
* If the task increased its priority or is running and
* lowered its priority, then reschedule its CPU:
*/
- if (delta < 0 || (delta > 0 && task_running(rq, p)))
- resched_task(rq->curr);
+ if (delta < 0 || ((delta > 0 || batch_task(p)) &&
+ task_running(rq, p)))
+ resched_task(rq->curr);
}
out_unlock:
task_rq_unlock(rq, &flags);
@@ -2406,6 +2424,12 @@
!capable(CAP_SYS_NICE))
goto out_unlock;

+ if (!(p->mm) && policy == SCHED_BATCH)
+ /*
+ * Don't allow kernel threads to be SCHED_BATCH.
+ */
+ goto out_unlock;
+
retval = security_task_setscheduler(p, policy, &lp);
if (retval)
goto out_unlock;
@@ -2640,9 +2664,9 @@
dequeue_task(current, rq);
current->slice = slice(current);
current->time_slice = RR_INTERVAL();
- if (likely(!rt_task(current))) {
+ if (likely(!rt_task(current) && !batch_task(current))) {
current->flags |= PF_YIELDED;
- current->prio = MAX_PRIO - 1;
+ current->prio = MAX_PRIO - 2;
}
current->burst = 0;
enqueue_task(current, rq);
@@ -2727,6 +2751,7 @@
ret = MAX_USER_RT_PRIO-1;
break;
case SCHED_NORMAL:
+ case SCHED_BATCH:
ret = 0;
break;
}
@@ -2750,6 +2775,7 @@
ret = 1;
break;
case SCHED_NORMAL:
+ case SCHED_BATCH:
ret = 0;
}
return ret;

Next message: Andi Kleen: "Re: XFS: how to NOT null files on fsck?"
Previous message: Con Kolivas: "[PATCH][1/3] Schedrange"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]