linux: Support new 'fast suspend' mode which does not require us to

hotplug all auxiliary CPUs.
Signed-off-by: Keir Fraser <keir@xensource.com>

4 files changed, 128 insertions, 74 deletions

diff --git a/linux-2.6-xen-sparse/drivers/xen/core/evtchn.c b/linux-2.6-xen-sparse/drivers/xen/core/evtchn.c
index e0657a6c87..fda12a3f32 100644
--- a/linux-2.6-xen-sparse/drivers/xen/core/evtchn.c
+++ b/linux-2.6-xen-sparse/drivers/xen/core/evtchn.c
@@ -888,48 +888,20 @@ void unmask_evtchn(int port)
 }
 EXPORT_SYMBOL_GPL(unmask_evtchn);
 
-void irq_resume(void)
+static void restore_cpu_virqs(int cpu)
 {
 	struct evtchn_bind_virq bind_virq;
-	struct evtchn_bind_ipi  bind_ipi;
-	int cpu, pirq, virq, ipi, irq, evtchn;
-
-	init_evtchn_cpu_bindings();
-
-	/* New event-channel space is not 'live' yet. */
-	for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
-		mask_evtchn(evtchn);
+	int virq, irq, evtchn;
 
-	/* Check that no PIRQs are still bound. */
-	for (pirq = 0; pirq < NR_PIRQS; pirq++)
-		BUG_ON(irq_info[pirq_to_irq(pirq)] != IRQ_UNBOUND);
-
-	/* Secondary CPUs must have no VIRQ or IPI bindings. */
-	for_each_possible_cpu(cpu) {
-		if (cpu == 0)
-			continue;
-		for (virq = 0; virq < NR_VIRQS; virq++)
-			BUG_ON(per_cpu(virq_to_irq, cpu)[virq] != -1);
-		for (ipi = 0; ipi < NR_IPIS; ipi++)
-			BUG_ON(per_cpu(ipi_to_irq, cpu)[ipi] != -1);
-	}
-
-	/* No IRQ <-> event-channel mappings. */
-	for (irq = 0; irq < NR_IRQS; irq++)
-		irq_info[irq] &= ~0xFFFF; /* zap event-channel binding */
-	for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
-		evtchn_to_irq[evtchn] = -1;
-
-	/* Primary CPU: rebind VIRQs automatically. */
 	for (virq = 0; virq < NR_VIRQS; virq++) {
-		if ((irq = per_cpu(virq_to_irq, 0)[virq]) == -1)
+		if ((irq = per_cpu(virq_to_irq, cpu)[virq]) == -1)
 			continue;
 
 		BUG_ON(irq_info[irq] != mk_irq_info(IRQT_VIRQ, virq, 0));
 
 		/* Get a new binding from Xen. */
 		bind_virq.virq = virq;
-		bind_virq.vcpu = 0;
+		bind_virq.vcpu = cpu;
 		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_virq,
 						&bind_virq) != 0)
 			BUG();
@@ -938,20 +910,26 @@ void irq_resume(void)
 		/* Record the new mapping. */
 		evtchn_to_irq[evtchn] = irq;
 		irq_info[irq] = mk_irq_info(IRQT_VIRQ, virq, evtchn);
+		bind_evtchn_to_cpu(evtchn, cpu);
 
 		/* Ready for use. */
 		unmask_evtchn(evtchn);
 	}
+}
+
+static void restore_cpu_ipis(int cpu)
+{
+	struct evtchn_bind_ipi bind_ipi;
+	int ipi, irq, evtchn;
 
-	/* Primary CPU: rebind IPIs automatically. */
 	for (ipi = 0; ipi < NR_IPIS; ipi++) {
-		if ((irq = per_cpu(ipi_to_irq, 0)[ipi]) == -1)
+		if ((irq = per_cpu(ipi_to_irq, cpu)[ipi]) == -1)
 			continue;
 
 		BUG_ON(irq_info[irq] != mk_irq_info(IRQT_IPI, ipi, 0));
 
 		/* Get a new binding from Xen. */
-		bind_ipi.vcpu = 0;
+		bind_ipi.vcpu = cpu;
 		if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_ipi,
 						&bind_ipi) != 0)
 			BUG();
@@ -960,12 +938,41 @@ void irq_resume(void)
 		/* Record the new mapping. */
 		evtchn_to_irq[evtchn] = irq;
 		irq_info[irq] = mk_irq_info(IRQT_IPI, ipi, evtchn);
+		bind_evtchn_to_cpu(evtchn, cpu);
 
 		/* Ready for use. */
 		unmask_evtchn(evtchn);
+
 	}
 }
 
+void irq_resume(void)
+{
+	int cpu, pirq, irq, evtchn;
+
+	init_evtchn_cpu_bindings();
+
+	/* New event-channel space is not 'live' yet. */
+	for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
+		mask_evtchn(evtchn);
+
+	/* Check that no PIRQs are still bound. */
+	for (pirq = 0; pirq < NR_PIRQS; pirq++)
+		BUG_ON(irq_info[pirq_to_irq(pirq)] != IRQ_UNBOUND);
+
+	/* No IRQ <-> event-channel mappings. */
+	for (irq = 0; irq < NR_IRQS; irq++)
+		irq_info[irq] &= ~0xFFFF; /* zap event-channel binding */
+	for (evtchn = 0; evtchn < NR_EVENT_CHANNELS; evtchn++)
+		evtchn_to_irq[evtchn] = -1;
+
+	for_each_possible_cpu(cpu) {
+		restore_cpu_virqs(cpu);
+		restore_cpu_ipis(cpu);
+	}
+
+}
+
 void __init xen_init_IRQ(void)
 {
 	int i;
diff --git a/linux-2.6-xen-sparse/drivers/xen/core/machine_reboot.c b/linux-2.6-xen-sparse/drivers/xen/core/machine_reboot.c
index 95337b8077..d8667b336a 100644
--- a/linux-2.6-xen-sparse/drivers/xen/core/machine_reboot.c
+++ b/linux-2.6-xen-sparse/drivers/xen/core/machine_reboot.c
@@ -1,4 +1,3 @@
-#define __KERNEL_SYSCALLS__
 #include <linux/version.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
@@ -7,6 +6,7 @@
 #include <linux/reboot.h>
 #include <linux/sysrq.h>
 #include <linux/stringify.h>
+#include <linux/stop_machine.h>
 #include <asm/irq.h>
 #include <asm/mmu_context.h>
 #include <xen/evtchn.h>
@@ -18,6 +18,7 @@
 #include <xen/gnttab.h>
 #include <xen/xencons.h>
 #include <xen/cpu_hotplug.h>
+#include <xen/interface/vcpu.h>
 
 #if defined(__i386__) || defined(__x86_64__)
 
@@ -98,7 +99,6 @@ static void post_suspend(int suspend_cancelled)
 		xen_start_info->console.domU.mfn =
 			pfn_to_mfn(xen_start_info->console.domU.mfn);
 	} else {
-		extern cpumask_t cpu_initialized_map;
 		cpu_initialized_map = cpumask_of_cpu(0);
 	}
 	
@@ -133,44 +133,35 @@ static void post_suspend(int suspend_cancelled)
 
 #endif
 
-int __xen_suspend(void)
+static int take_machine_down(void *p_fast_suspend)
 {
-	int err, suspend_cancelled;
-
+	int fast_suspend = *(int *)p_fast_suspend;
+	int suspend_cancelled, err, cpu;
 	extern void time_resume(void);
 
-	BUG_ON(smp_processor_id() != 0);
-	BUG_ON(in_interrupt());
-
-#if defined(__i386__) || defined(__x86_64__)
-	if (xen_feature(XENFEAT_auto_translated_physmap)) {
-		printk(KERN_WARNING "Cannot suspend in "
-		       "auto_translated_physmap mode.\n");
-		return -EOPNOTSUPP;
-	}
-#endif
-
-	for (;;) {
-		err = smp_suspend();
-		if (err)
-			return err;
-
-		xenbus_suspend();
+	if (fast_suspend) {
 		preempt_disable();
+	} else {
+		for (;;) {
+			err = smp_suspend();
+			if (err)
+				return err;
 
-		if (num_online_cpus() == 1)
-			break;
+			xenbus_suspend();
+			preempt_disable();
 
-		preempt_enable();
-		xenbus_suspend_cancel();
+			if (num_online_cpus() == 1)
+				break;
+
+			preempt_enable();
+			xenbus_suspend_cancel();
+		}
 	}
 
 	mm_pin_all();
 	local_irq_disable();
 	preempt_enable();
-
 	gnttab_suspend();
-
 	pre_suspend();
 
 	/*
@@ -180,18 +171,56 @@ int __xen_suspend(void)
 	suspend_cancelled = HYPERVISOR_suspend(virt_to_mfn(xen_start_info));
 
 	post_suspend(suspend_cancelled);
-
 	gnttab_resume();
-
 	if (!suspend_cancelled)
 		irq_resume();
-
 	time_resume();
-
 	switch_idle_mm();
-
 	local_irq_enable();
 
+	if (fast_suspend && !suspend_cancelled) {
+		/*
+		 * In fast-suspend mode the APs may not be brought back online
+		 * when we resume. In that case we do it here.
+		 */
+		for_each_online_cpu(cpu) {
+			if (cpu == 0)
+				continue;
+			cpu_set_initialized(cpu);
+			err = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
+			BUG_ON(err);
+		}
+	}
+
+	return suspend_cancelled;
+}
+
+int __xen_suspend(int fast_suspend)
+{
+	int err, suspend_cancelled;
+
+	BUG_ON(smp_processor_id() != 0);
+	BUG_ON(in_interrupt());
+
+#if defined(__i386__) || defined(__x86_64__)
+	if (xen_feature(XENFEAT_auto_translated_physmap)) {
+		printk(KERN_WARNING "Cannot suspend in "
+		       "auto_translated_physmap mode.\n");
+		return -EOPNOTSUPP;
+	}
+#endif
+
+	if (fast_suspend) {
+		xenbus_suspend();
+		err = stop_machine_run(take_machine_down, &fast_suspend, 0);
+	} else {
+		err = take_machine_down(&fast_suspend);
+	}
+
+	if (err < 0)
+		return err;
+
+	suspend_cancelled = err;
 	if (!suspend_cancelled) {
 		xencons_resume();
 		xenbus_resume();
@@ -199,7 +228,8 @@ int __xen_suspend(void)
 		xenbus_suspend_cancel();
 	}
 
-	smp_resume();
+	if (!fast_suspend)
+		smp_resume();
 
-	return err;
+	return 0;
 }
diff --git a/linux-2.6-xen-sparse/drivers/xen/core/reboot.c b/linux-2.6-xen-sparse/drivers/xen/core/reboot.c
index aab244a738..0dcd2fa252 100644
--- a/linux-2.6-xen-sparse/drivers/xen/core/reboot.c
+++ b/linux-2.6-xen-sparse/drivers/xen/core/reboot.c
@@ -24,13 +24,16 @@ MODULE_LICENSE("Dual BSD/GPL");
 /* Ignore multiple shutdown requests. */
 static int shutting_down = SHUTDOWN_INVALID;
 
+/* Can we leave APs online when we suspend? */
+static int fast_suspend;
+
 static void __shutdown_handler(void *unused);
 static DECLARE_WORK(shutdown_work, __shutdown_handler, NULL);
 
 #ifdef CONFIG_XEN
-int __xen_suspend(void);
+int __xen_suspend(int fast_suspend);
 #else
-#define __xen_suspend() (void)0
+#define __xen_suspend(fast_suspend) 0
 #endif
 
 static int shutdown_process(void *__unused)
@@ -44,7 +47,8 @@ static int shutdown_process(void *__unused)
 
 	if ((shutting_down == SHUTDOWN_POWEROFF) ||
 	    (shutting_down == SHUTDOWN_HALT)) {
-		if (call_usermodehelper("/sbin/poweroff", poweroff_argv, envp, 0) < 0) {
+		if (call_usermodehelper("/sbin/poweroff", poweroff_argv,
+					envp, 0) < 0) {
 #ifdef CONFIG_XEN
 			sys_reboot(LINUX_REBOOT_MAGIC1,
 				   LINUX_REBOOT_MAGIC2,
@@ -61,7 +65,9 @@ static int shutdown_process(void *__unused)
 
 static int xen_suspend(void *__unused)
 {
-	__xen_suspend();
+	int err = __xen_suspend(fast_suspend);
+	if (err)
+		printk(KERN_ERR "Xen suspend failed (%d)\n", err);
 	shutting_down = SHUTDOWN_INVALID;
 	return 0;
 }
@@ -193,6 +199,10 @@ static int setup_shutdown_watcher(struct notifier_block *notifier,
 {
 	int err;
 
+	xenbus_scanf(XBT_NIL, "control",
+		     "platform-feature-multiprocessor-suspend",
+		     "%d", &fast_suspend);
+
 	err = register_xenbus_watch(&shutdown_watch);
 	if (err)
 		printk(KERN_ERR "Failed to set shutdown watcher\n");
diff --git a/linux-2.6-xen-sparse/include/xen/cpu_hotplug.h b/linux-2.6-xen-sparse/include/xen/cpu_hotplug.h
index 234503f1ea..8c9eeb2a98 100644
--- a/linux-2.6-xen-sparse/include/xen/cpu_hotplug.h
+++ b/linux-2.6-xen-sparse/include/xen/cpu_hotplug.h
@@ -4,6 +4,13 @@
 #include <linux/kernel.h>
 #include <linux/cpumask.h>
 
+#if defined(CONFIG_X86)
+extern cpumask_t cpu_initialized_map;
+#define cpu_set_initialized(cpu) cpu_set(cpu, cpu_initialized_map)
+#else
+#define cpu_set_initialized(cpu) ((void)0)
+#endif
+
 #if defined(CONFIG_HOTPLUG_CPU)
 
 int cpu_up_check(unsigned int cpu);