Hi, all.
I want to dynamically use irqbalance on X86 processor. My design is
like the following:
1) if we boot kernel with "noirqbalance", then irqbalance is always
disabled.
2) if we boot kernel without "noirqbalance", we can enable/disable
irqbalance in runtime.
I create a proc_fs entry /proc/sys/kernel/irqbalance
This symbol I adding is in file arch/i386/kernel/io_apic.c
/proc/sys/kernel/irqbalance = 0 /* disable irqbalance in runtime */
/proc/sys/kernel/irqbalance = 1 /* enable irqbalance in runtime */
The core function is like the following:
irqbalance_irq_flag = -1; /* this is the last time irqbalance used */
irqbalance_enable = 1; /* set it to enable by default */
static int balanced_irq(void *unused)
{
int i;
unsigned long prev_balance_time = jiffies;
long time_remaining = balanced_irq_interval;
daemonize("kirqd");
/* push everything to CPU 0 to give us a starting point. */
for (i = 0 ; i < NR_IRQS ; i++) {
irq_desc[i].pending_mask = cpumask_of_cpu(0);
set_pending_irq(i, cpumask_of_cpu(0));
}
for ( ; ; ) {
time_remaining = schedule_timeout_interruptible(time_remaining);
try_to_freeze();
if(irqbalance_enable) {
if (time_after(jiffies,
prev_balance_time+balanced_irq_interval)) {
preempt_disable();
do_irq_balance();
prev_balance_time = jiffies;
time_remaining = balanced_irq_interval;
preempt_enable();
irqbalance_irq_flag = 1;
}
} else if (irqbalance_irq_flag != 0) {
/* Is it SAFE to do so? */
for (i = 0 ; i < NR_IRQS ; i++)
set_pending_irq(i, cpumask_of_cpu(0));
irqbalance_irq_flag = 0;
}
}
return 0;
}
When we change /proc/sys/kernel/irqbalance from 1 to 0, I move all the
irqs to CPU#0, Is it safe to do so? or will lead to some very dangerous
thing?
Please help me to review my design, thanks in advance.
BTW: I can't join the linux-kernel maillist, for our mail server now has
some problems. If anyone replies my mail. Pls CC to me.
-
