Hotplug CPU toy for i386

diff -urpN --exclude TAGS -X /home/rusty/devel/kernel/kernel-patches/current-dontdiff --minimal .24305-2.6.3-mm1-atomic-cpudown-i386.pre/arch/i386/kernel/apic.c .24305-2.6.3-mm1-atomic-cpudown-i386/arch/i386/kernel/apic.c
@@ -26,6 +26,7 @@
 #include <linux/mc146818rtc.h>
 #include <linux/kernel_stat.h>
 #include <linux/sysdev.h>
+#include <linux/cpu.h>
 
 #include <asm/atomic.h>
 #include <asm/smp.h>

@@ -1035,6 +1036,10 @@ inline void smp_local_timer_interrupt(st
 {
 	int cpu = smp_processor_id();
 
+	/* FIXME: Actually remove timer interrupt in __cpu_disable() --RR */
+	if (cpu_is_offline(cpu))
+		return;
+
 	x86_do_profile(regs);
 
 	if (--per_cpu(prof_counter, cpu) <= 0) {

diff -urpN --exclude TAGS -X /home/rusty/devel/kernel/kernel-patches/current-dontdiff --minimal .24305-2.6.3-mm1-atomic-cpudown-i386.pre/arch/i386/kernel/io_apic.c .24305-2.6.3-mm1-atomic-cpudown-i386/arch/i386/kernel/io_apic.c
@@ -614,7 +614,10 @@ repeat:
 	if (time_after(jiffies, prev_balance_time+balanced_irq_interval)) {
 		Dprintk("balanced_irq: calling do_irq_balance() %lu\n",
 					jiffies);
+		/* Disable preempt to stop cpus going down. */
+		preempt_disable();
 		do_irq_balance();
+		preempt_enable();
 		prev_balance_time = jiffies;
 		time_remaining = balanced_irq_interval;
 	}

117 lines not on this page · raw message

diff -urpN --exclude TAGS -X /home/rusty/devel/kernel/kernel-patches/current-dontdiff --minimal .24305-2.6.3-mm1-atomic-cpudown-i386.pre/arch/i386/kernel/process.c .24305-2.6.3-mm1-atomic-cpudown-i386/arch/i386/kernel/process.c
@@ -14,6 +14,7 @@
 #define __KERNEL_SYSCALLS__
 #include <stdarg.h>
 
+#include <linux/cpu.h>
 #include <linux/errno.h>
 #include <linux/sched.h>
 #include <linux/fs.h>

@@ -55,6 +56,9 @@
 #include <linux/irq.h>
 #include <linux/err.h>
 
+#include <asm/tlbflush.h>
+#include <asm/cpu.h>
+
 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
 
 int hlt_counter;

@@ -133,6 +137,35 @@ static void poll_idle (void)
 	}
 }
 
+#ifdef CONFIG_HOTPLUG_CPU
+/* We don't actually take CPU down, just spin without interrupts. */
+static inline void play_dead(void)
+{
+	/* Ack it */
+	__get_cpu_var(cpu_state) = CPU_DEAD;
+
+	disable_timer_nmi_watchdog();
+
+	/* We shouldn't have to disable interrupts while dead, but
+	 * some interrupts just don't seem to go away, and this makes
+	 * it "work" for testing purposes. */
+	/* Death loop */
+	while (__get_cpu_var(cpu_state) != CPU_UP_PREPARE)
+		cpu_relax();
+
+	local_irq_disable();
+	__flush_tlb_all();
+	cpu_set(smp_processor_id(), cpu_online_map);
+	local_irq_enable();
+	enable_timer_nmi_watchdog();
+}
+#else
+static inline void play_dead(void)
+{
+	BUG();
+}
+#endif /* CONFIG_HOTPLUG_CPU */
+
 /*
  * The idle thread. There's no useful work to be
  * done, so just try to conserve power and have a

@@ -149,6 +182,8 @@ void cpu_idle (void)
 			if (!idle)
 				idle = default_idle;
 
+			if (cpu_is_offline(smp_processor_id()))
+				play_dead();
 			irq_stat[smp_processor_id()].idle_timestamp = jiffies;
 			idle();
 		}

diff -urpN --exclude TAGS -X /home/rusty/devel/kernel/kernel-patches/current-dontdiff --minimal .24305-2.6.3-mm1-atomic-cpudown-i386.pre/arch/i386/kernel/smp.c .24305-2.6.3-mm1-atomic-cpudown-i386/arch/i386/kernel/smp.c
@@ -19,6 +19,7 @@
 #include <linux/mc146818rtc.h>
 #include <linux/cache.h>
 #include <linux/interrupt.h>
+#include <linux/cpu.h>
 
 #include <asm/mtrr.h>
 #include <asm/pgalloc.h>

@@ -164,6 +165,8 @@ inline void send_IPI_mask_bitmask(cpumas
 	unsigned long cfg;
 	unsigned long flags;
 
+	WARN_ON(mask & ~cpus_coerce(cpu_online_map));
+
 	local_irq_save(flags);
 		
 	/*

@@ -347,21 +350,21 @@ out:
 static void flush_tlb_others(cpumask_t cpumask, struct mm_struct *mm,
 						unsigned long va)
 {
-	cpumask_t tmp;
 	/*
 	 * A couple of (to be removed) sanity checks:
 	 *
-	 * - we do not send IPIs to not-yet booted CPUs.
 	 * - current CPU must not be in mask
 	 * - mask must exist :)
 	 */
 	BUG_ON(cpus_empty(cpumask));
-
-	cpus_and(tmp, cpumask, cpu_online_map);
-	BUG_ON(!cpus_equal(cpumask, tmp));
 	BUG_ON(cpu_isset(smp_processor_id(), cpumask));
 	BUG_ON(!mm);
 
+	/* If a CPU which we ran on has gone down, OK. */
+	cpus_and(cpumask, cpumask, cpu_online_map);
+	if (cpus_empty(cpumask))
+		return;
+
 	/*
 	 * i'm not happy about this global shared spinlock in the
 	 * MM hot path, but we'll see how contended it is.

@@ -474,6 +477,7 @@ void smp_send_nmi_allbutself(void)
  */
 void smp_send_reschedule(int cpu)
 {
+	WARN_ON(cpu_is_offline(cpu));
 	send_IPI_mask(cpumask_of_cpu(cpu), RESCHEDULE_VECTOR);
 }
 

@@ -514,10 +518,16 @@ int smp_call_function (void (*func) (voi
  */
 {
 	struct call_data_struct data;
-	int cpus = num_online_cpus()-1;
+	int cpus;
 
-	if (!cpus)
+	/* Holding any lock stops cpus from going down. */
+	spin_lock(&call_lock);
+	cpus = num_online_cpus()-1;
+
+	if (!cpus) {
+		spin_unlock(&call_lock);
 		return 0;
+	}
 
 	data.func = func;
 	data.info = info;

@@ -526,7 +536,6 @@ int smp_call_function (void (*func) (voi
 	if (wait)
 		atomic_set(&data.finished, 0);
 
-	spin_lock(&call_lock);
 	call_data = &data;
 	mb();
 	

118 lines not on this page · raw message

diff -urpN --exclude TAGS -X /home/rusty/devel/kernel/kernel-patches/current-dontdiff --minimal .24305-2.6.3-mm1-atomic-cpudown-i386.pre/include/asm-i386/cpu.h .24305-2.6.3-mm1-atomic-cpudown-i386/include/asm-i386/cpu.h
@@ -4,6 +4,7 @@
 #include <linux/device.h>
 #include <linux/cpu.h>
 #include <linux/topology.h>
+#include <linux/percpu.h>
 
 #include <asm/node.h>
 

@@ -23,4 +24,5 @@ static inline int arch_register_cpu(int 
 	return register_cpu(&cpu_devices[num].cpu, num, parent);
 }
 
+DECLARE_PER_CPU(int, cpu_state);
 #endif /* _ASM_I386_CPU_H_ */

diff -urpN --exclude TAGS -X /home/rusty/devel/kernel/kernel-patches/current-dontdiff --minimal .24305-2.6.3-mm1-atomic-cpudown-i386.pre/include/asm-i386/smp.h .24305-2.6.3-mm1-atomic-cpudown-i386/include/asm-i386/smp.h
@@ -84,6 +84,9 @@ static __inline int logical_smp_processo
 }
 
 #endif
+
+extern int __cpu_disable(void);
+extern void __cpu_die(unsigned int cpu);
 #endif /* !__ASSEMBLY__ */
 
 #define NO_PROC_ID		0xFF		/* No processor magic marker */