I'm using do_gettimeofday in the scheduler, once per schedule.
Ocassioanlly, i'll get what seems to be a random garbage number.
For example:
tv_sec tv_usec
1032378775 627501
1032378775 627521
3433001412 1
1032378775 627573
1032378775 627617
1032378775 627638
or
1032379233 253008
1032379233 253028
3387638236 1
1032379233 253068
1032379233 253125
The garbage number tv_sec always seems to begin with a 3 and is followed by 9
digits. The tv_usec garbage number is always 1.
Most of the code I changed in sched.c follows. It attempts to allow a normal
process to run in place of a SCHED_RR process with a certain limited
frequency and within a certain window of time. A starvation threshold is
added which allows a normal process to run disregarding the requested window.
The patch is very constrained in it's application. It's for a program in
which a fifo thread runs once every millisecond, and round robin threads take
up the remainder of the time.
=====================================================
#define NORM_THRESH 1000
#define NORM_STARVE 20000
#define NORM_WINDOW 800
#ifdef CONFIG_SCHED_RR_NONSTARVE
int inline rr_should_yield(runqueue_t *rq, struct timeval now)
{
unsigned long normdiff, secdiff, fifodiff;
secdiff = now.tv_sec - rq->last_normal.tv_sec;
normdiff = now.tv_usec - ((secdiff * 1000000) + rq->last_normal.tv_usec);
if( normdiff > NORM_STARVE)
{
return 1;
}
secdiff = now.tv_sec - rq->last_fifo.tv_sec;
fifodiff = now.tv_usec - ((secdiff * 1000000) + rq->last_fifo.tv_usec);
if( (normdiff > NORM_THRESH) && (fifodiff > NORM_WINDOW) )
{
return 1;
}
return 0;
}
#endif /*CONFIG_RT_NONSTARVE*/
/*
* 'schedule()' is the main scheduler function.
*/
asmlinkage void schedule(void)
{
task_t *prev, *next;
runqueue_t *rq;
prio_array_t *array;
list_t *queue;
int idx;
#ifdef CONFIG_SCHED_RR_NONSTARVE
struct timeval now; /* schedule timestamp */
int idx2;
#endif
#ifdef CONFIG_RT_SWITCH_PRINT
int policy1;
int policy2;
struct timeval tv;
unsigned long flags;
int allowed = 0;
#endif
if (unlikely(in_interrupt()))
BUG();
need_resched:
preempt_disable();
prev = current;
rq = this_rq();
release_kernel_lock(prev, smp_processor_id());
prev->sleep_timestamp = jiffies;
spin_lock_irq(&rq->lock);
/*
* if entering from preempt_schedule, off a kernel preemption,
* go straight to picking the next task.
*/
if (unlikely(preempt_get_count() & PREEMPT_ACTIVE))
goto pick_next_task;
#if CONFIG_RR_RESCHED_SWITCH || CONFIG_SCHED_RR_NONSTARVE
/*
* Move a SCHED_RR task to end of RQ each time schedule is called
*/
do_gettimeofday(&now);
if (prev->policy == SCHED_RR){
#ifdef CONFIG_SCHED_RR_NONSTARVE
if(unlikely(rq->non_rt_starving)){
/* if all normal processes are expired
* we dequeue the round robin processes
* instead of putting them at the end
* of their queue in the active array
* This will trigger a active/expired switch
* eventually
*/
printk("moving RR to expired\n");
/* dequeue_task(prev, rq->active);
prev->first_time_slice = 0;
enqueue_task(prev, rq->expired);*/
//rq->expired_timestamp = jiffies;
goto pick_next_task;
}
#endif /*CONFIG_SCHED_RR_NONSTARVE*/
#ifdef CONFIG_RR_RESCHED_SWITCH
dequeue_task(prev, rq->active);
enqueue_task(prev, rq->active);
#endif /*CONFIG_RR_RESCHED_SWITCH*/
out_rr_move:
}
#endif /*CONFIG_RR_RESCHED_SWITCH || CONFIG_SCHED_RR_NONSTARVE*/
switch (prev->state) {
case TASK_INTERRUPTIBLE:
if (unlikely(signal_pending(prev))) {
prev->state = TASK_RUNNING;
break;
}
default:
deactivate_task(prev, rq);
case TASK_RUNNING:
;
}
pick_next_task:
if (unlikely(!rq->nr_running)) {
#if CONFIG_SMP
load_balance(rq, 1);
if (rq->nr_running)
goto pick_next_task;
#endif
/*
* Pick a task from the batch queue if available.
*/
if (rq->nr_batch) {
list_t *tmp = rq->batch_queue.next;
next = list_entry(tmp, task_t, run_list);
activate_batch_task(next, rq);
} else
next = rq->idle;
rq->expired_timestamp = 0;
goto switch_tasks;
}
array = rq->active;
if (unlikely(!array->nr_active)) {
/*
* Switch the active and expired arrays.
*/
rq->active = rq->expired;
rq->expired = array;
array = rq->active;
rq->expired_timestamp = 0;
#ifdef CONFIG_SCHED_RR_NONSTARVE
if(rq->non_rt_starving){
rq->non_rt_starving = 0;
printk("exchange!\n");
}
#endif /* CONFIG_SCHED_RR_NONSTARVE*/
}
idx = sched_find_first_bit(array->bitmap);
queue = array->queue + idx;
next = list_entry(queue->next, task_t, run_list);
#ifdef CONFIG_SCHED_RR_NONSTARVE
/*
* if SCHED_RR task should yeild and next task is SCHED_RR
* let a SCHED_NORMAL process take the CPU
*/
switch (next->policy){
case SCHED_FIFO:
rq->last_fifo = now;
break;
case SCHED_NORMAL:
rq->last_normal = now;
break;
case SCHED_RR:
if( rr_should_yield(rq, now) ){
/*find non RT process*/
idx2 = find_next_bit(array->bitmap, MAX_PRIO,MAX_RT_PRIO);
/*if nothing found, they must be expired*/
if(idx2 == MAX_PRIO && EXPIRED_STARVING(rq)){
rq->non_rt_starving = 1;
goto switch_tasks;
}
/*use process list at idx2*/
if(idx2 < MAX_PRIO){
queue = array->queue + idx2;
next = list_entry(queue->next, task_t, run_list);
rq->last_normal = now;
printk("allowing %d\n", next->pid);
allowed = 1;
}
}
}
#endif /*CONFIG_SCHED_RR_NONSTARVE*/
switch_tasks:
#ifdef CONFIG_RT_SWITCH_PRINT
policy1 = prev->policy;
policy2 = next->policy;
if(allowed || (policy1 != SCHED_NORMAL || policy2 != SCHED_NORMAL)){
switch (policy1){
case SCHED_FIFO:
printk("[{%d} ",prev->pid);
break;
case SCHED_RR:
printk("[(%d) ",prev->pid);
break;
default:
printk("[%d ",prev->pid);
}
switch (policy2){
case SCHED_FIFO:
printk("{%d} - %lu %lu]\n",next->pid, now.tv_sec,
now.tv_usec);
break;
case SCHED_RR:
printk("(%d) - %lu %lu]",next->pid, now.tv_sec, now.tv_usec);
break;
default:
printk("%d - %lu %lu]",next->pid, now.tv_sec, now.tv_usec);
}
#ifdef CONFIG_RT_PRINT_DEFAULT
printk("\n");
#endif /* CONFIG_RT_PRINT_DEFAULT*/
}
#endif /*CONFIG_RT_SWITCH_PRINT */
prefetch(next);
clear_tsk_need_resched(prev);
if (likely(prev != next)) {
rq->nr_switches++;
rq->curr = next;
prepare_arch_switch(rq, next);
prev = context_switch(prev, next);
barrier();
rq = this_rq();
finish_arch_switch(rq, prev);
} else
spin_unlock_irq(&rq->lock);
reacquire_kernel_lock(current);
preempt_enable_no_resched();
if (need_resched())
goto need_resched;
}
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Sep 23 2002 - 22:00:24 EST