Merge branches 'release', 'asus', 'sony-laptop' and 'thinkpad' into release
[pandora-kernel.git] / arch / s390 / kernel / smp.c
index 264ea90..8506065 100644 (file)
@@ -42,6 +42,7 @@
 #include <asm/tlbflush.h>
 #include <asm/timer.h>
 #include <asm/lowcore.h>
+#include <asm/sclp.h>
 #include <asm/cpu.h>
 
 /*
@@ -53,11 +54,27 @@ EXPORT_SYMBOL(lowcore_ptr);
 cpumask_t cpu_online_map = CPU_MASK_NONE;
 EXPORT_SYMBOL(cpu_online_map);
 
-cpumask_t cpu_possible_map = CPU_MASK_NONE;
+cpumask_t cpu_possible_map = CPU_MASK_ALL;
 EXPORT_SYMBOL(cpu_possible_map);
 
 static struct task_struct *current_set[NR_CPUS];
 
+static u8 smp_cpu_type;
+static int smp_use_sigp_detection;
+
+enum s390_cpu_state {
+       CPU_STATE_STANDBY,
+       CPU_STATE_CONFIGURED,
+};
+
+#ifdef CONFIG_HOTPLUG_CPU
+static DEFINE_MUTEX(smp_cpu_state_mutex);
+#endif
+static int smp_cpu_state[NR_CPUS];
+
+static DEFINE_PER_CPU(struct cpu, cpu_devices);
+DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
+
 static void smp_ext_bitcall(int, ec_bit_sig);
 
 /*
@@ -193,6 +210,32 @@ int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
 }
 EXPORT_SYMBOL(smp_call_function_single);
 
+/**
+ * smp_call_function_mask(): Run a function on a set of other CPUs.
+ * @mask: The set of cpus to run on.  Must not include the current cpu.
+ * @func: The function to run. This must be fast and non-blocking.
+ * @info: An arbitrary pointer to pass to the function.
+ * @wait: If true, wait (atomically) until function has completed on other CPUs.
+ *
+ * Returns 0 on success, else a negative status code.
+ *
+ * If @wait is true, then returns once @func has returned; otherwise
+ * it returns just before the target cpu calls @func.
+ *
+ * You must not call this function with disabled interrupts or from a
+ * hardware interrupt handler or from a bottom half handler.
+ */
+int smp_call_function_mask(cpumask_t mask, void (*func)(void *), void *info,
+                          int wait)
+{
+       preempt_disable();
+       cpu_clear(smp_processor_id(), mask);
+       __smp_call_function_map(func, info, 0, wait, mask);
+       preempt_enable();
+       return 0;
+}
+EXPORT_SYMBOL(smp_call_function_mask);
+
 void smp_send_stop(void)
 {
        int cpu, rc;
@@ -216,33 +259,6 @@ void smp_send_stop(void)
        }
 }
 
-/*
- * Reboot, halt and power_off routines for SMP.
- */
-void machine_restart_smp(char *__unused)
-{
-       smp_send_stop();
-       do_reipl();
-}
-
-void machine_halt_smp(void)
-{
-       smp_send_stop();
-       if (MACHINE_IS_VM && strlen(vmhalt_cmd) > 0)
-               __cpcmd(vmhalt_cmd, NULL, 0, NULL);
-       signal_processor(smp_processor_id(), sigp_stop_and_store_status);
-       for (;;);
-}
-
-void machine_power_off_smp(void)
-{
-       smp_send_stop();
-       if (MACHINE_IS_VM && strlen(vmpoff_cmd) > 0)
-               __cpcmd(vmpoff_cmd, NULL, 0, NULL);
-       signal_processor(smp_processor_id(), sigp_stop_and_store_status);
-       for (;;);
-}
-
 /*
  * This is the main routine where commands issued by other
  * cpus are handled.
@@ -355,6 +371,13 @@ void smp_ctl_clear_bit(int cr, int bit)
 }
 EXPORT_SYMBOL(smp_ctl_clear_bit);
 
+/*
+ * In early ipl state a temp. logically cpu number is needed, so the sigp
+ * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
+ * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
+ */
+#define CPU_INIT_NO    1
+
 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
 
 /*
@@ -375,9 +398,10 @@ static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
                       "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
                return;
        }
-       zfcpdump_save_areas[cpu] = alloc_bootmem(sizeof(union save_area));
-       __cpu_logical_map[1] = (__u16) phy_cpu;
-       while (signal_processor(1, sigp_stop_and_store_status) == sigp_busy)
+       zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
+       __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
+       while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
+              sigp_busy)
                cpu_relax();
        memcpy(zfcpdump_save_areas[cpu],
               (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
@@ -397,32 +421,155 @@ static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
 
 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
 
-/*
- * Lets check how many CPUs we have.
- */
-static unsigned int __init smp_count_cpus(void)
+static int cpu_stopped(int cpu)
 {
-       unsigned int cpu, num_cpus;
-       __u16 boot_cpu_addr;
+       __u32 status;
 
-       /*
-        * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
-        */
+       /* Check for stopped state */
+       if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
+           sigp_status_stored) {
+               if (status & 0x40)
+                       return 1;
+       }
+       return 0;
+}
+
+static int cpu_known(int cpu_id)
+{
+       int cpu;
+
+       for_each_present_cpu(cpu) {
+               if (__cpu_logical_map[cpu] == cpu_id)
+                       return 1;
+       }
+       return 0;
+}
+
+static int smp_rescan_cpus_sigp(cpumask_t avail)
+{
+       int cpu_id, logical_cpu;
+
+       logical_cpu = first_cpu(avail);
+       if (logical_cpu == NR_CPUS)
+               return 0;
+       for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
+               if (cpu_known(cpu_id))
+                       continue;
+               __cpu_logical_map[logical_cpu] = cpu_id;
+               if (!cpu_stopped(logical_cpu))
+                       continue;
+               cpu_set(logical_cpu, cpu_present_map);
+               smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
+               logical_cpu = next_cpu(logical_cpu, avail);
+               if (logical_cpu == NR_CPUS)
+                       break;
+       }
+       return 0;
+}
+
+static int smp_rescan_cpus_sclp(cpumask_t avail)
+{
+       struct sclp_cpu_info *info;
+       int cpu_id, logical_cpu, cpu;
+       int rc;
+
+       logical_cpu = first_cpu(avail);
+       if (logical_cpu == NR_CPUS)
+               return 0;
+       info = kmalloc(sizeof(*info), GFP_KERNEL);
+       if (!info)
+               return -ENOMEM;
+       rc = sclp_get_cpu_info(info);
+       if (rc)
+               goto out;
+       for (cpu = 0; cpu < info->combined; cpu++) {
+               if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
+                       continue;
+               cpu_id = info->cpu[cpu].address;
+               if (cpu_known(cpu_id))
+                       continue;
+               __cpu_logical_map[logical_cpu] = cpu_id;
+               cpu_set(logical_cpu, cpu_present_map);
+               if (cpu >= info->configured)
+                       smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
+               else
+                       smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
+               logical_cpu = next_cpu(logical_cpu, avail);
+               if (logical_cpu == NR_CPUS)
+                       break;
+       }
+out:
+       kfree(info);
+       return rc;
+}
+
+static int smp_rescan_cpus(void)
+{
+       cpumask_t avail;
+
+       cpus_xor(avail, cpu_possible_map, cpu_present_map);
+       if (smp_use_sigp_detection)
+               return smp_rescan_cpus_sigp(avail);
+       else
+               return smp_rescan_cpus_sclp(avail);
+}
+
+static void __init smp_detect_cpus(void)
+{
+       unsigned int cpu, c_cpus, s_cpus;
+       struct sclp_cpu_info *info;
+       u16 boot_cpu_addr, cpu_addr;
+
+       c_cpus = 1;
+       s_cpus = 0;
        boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
-       current_thread_info()->cpu = 0;
-       num_cpus = 1;
-       for (cpu = 0; cpu <= 65535; cpu++) {
-               if ((__u16) cpu == boot_cpu_addr)
+       info = kmalloc(sizeof(*info), GFP_KERNEL);
+       if (!info)
+               panic("smp_detect_cpus failed to allocate memory\n");
+       /* Use sigp detection algorithm if sclp doesn't work. */
+       if (sclp_get_cpu_info(info)) {
+               smp_use_sigp_detection = 1;
+               for (cpu = 0; cpu <= 65535; cpu++) {
+                       if (cpu == boot_cpu_addr)
+                               continue;
+                       __cpu_logical_map[CPU_INIT_NO] = cpu;
+                       if (!cpu_stopped(CPU_INIT_NO))
+                               continue;
+                       smp_get_save_area(c_cpus, cpu);
+                       c_cpus++;
+               }
+               goto out;
+       }
+
+       if (info->has_cpu_type) {
+               for (cpu = 0; cpu < info->combined; cpu++) {
+                       if (info->cpu[cpu].address == boot_cpu_addr) {
+                               smp_cpu_type = info->cpu[cpu].type;
+                               break;
+                       }
+               }
+       }
+
+       for (cpu = 0; cpu < info->combined; cpu++) {
+               if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
+                       continue;
+               cpu_addr = info->cpu[cpu].address;
+               if (cpu_addr == boot_cpu_addr)
                        continue;
-               __cpu_logical_map[1] = (__u16) cpu;
-               if (signal_processor(1, sigp_sense) == sigp_not_operational)
+               __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
+               if (!cpu_stopped(CPU_INIT_NO)) {
+                       s_cpus++;
                        continue;
-               smp_get_save_area(num_cpus, cpu);
-               num_cpus++;
+               }
+               smp_get_save_area(c_cpus, cpu_addr);
+               c_cpus++;
        }
-       printk("Detected %d CPU's\n", (int) num_cpus);
-       printk("Boot cpu address %2X\n", boot_cpu_addr);
-       return num_cpus;
+out:
+       kfree(info);
+       printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
+       get_online_cpus();
+       smp_rescan_cpus();
+       put_online_cpus();
 }
 
 /*
@@ -453,8 +600,6 @@ int __cpuinit start_secondary(void *cpuvoid)
        return 0;
 }
 
-DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
-
 static void __init smp_create_idle(unsigned int cpu)
 {
        struct task_struct *p;
@@ -470,37 +615,82 @@ static void __init smp_create_idle(unsigned int cpu)
        spin_lock_init(&(&per_cpu(s390_idle, cpu))->lock);
 }
 
-static int cpu_stopped(int cpu)
+static int __cpuinit smp_alloc_lowcore(int cpu)
 {
-       __u32 status;
+       unsigned long async_stack, panic_stack;
+       struct _lowcore *lowcore;
+       int lc_order;
+
+       lc_order = sizeof(long) == 8 ? 1 : 0;
+       lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
+       if (!lowcore)
+               return -ENOMEM;
+       async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
+       if (!async_stack)
+               goto out_async_stack;
+       panic_stack = __get_free_page(GFP_KERNEL);
+       if (!panic_stack)
+               goto out_panic_stack;
+
+       *lowcore = S390_lowcore;
+       lowcore->async_stack = async_stack + ASYNC_SIZE;
+       lowcore->panic_stack = panic_stack + PAGE_SIZE;
 
-       /* Check for stopped state */
-       if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
-           sigp_status_stored) {
-               if (status & 0x40)
-                       return 1;
+#ifndef CONFIG_64BIT
+       if (MACHINE_HAS_IEEE) {
+               unsigned long save_area;
+
+               save_area = get_zeroed_page(GFP_KERNEL);
+               if (!save_area)
+                       goto out_save_area;
+               lowcore->extended_save_area_addr = (u32) save_area;
        }
+#endif
+       lowcore_ptr[cpu] = lowcore;
        return 0;
+
+#ifndef CONFIG_64BIT
+out_save_area:
+       free_page(panic_stack);
+#endif
+out_panic_stack:
+       free_pages(async_stack, ASYNC_ORDER);
+out_async_stack:
+       free_pages((unsigned long) lowcore, lc_order);
+       return -ENOMEM;
 }
 
-/* Upping and downing of CPUs */
+#ifdef CONFIG_HOTPLUG_CPU
+static void smp_free_lowcore(int cpu)
+{
+       struct _lowcore *lowcore;
+       int lc_order;
+
+       lc_order = sizeof(long) == 8 ? 1 : 0;
+       lowcore = lowcore_ptr[cpu];
+#ifndef CONFIG_64BIT
+       if (MACHINE_HAS_IEEE)
+               free_page((unsigned long) lowcore->extended_save_area_addr);
+#endif
+       free_page(lowcore->panic_stack - PAGE_SIZE);
+       free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
+       free_pages((unsigned long) lowcore, lc_order);
+       lowcore_ptr[cpu] = NULL;
+}
+#endif /* CONFIG_HOTPLUG_CPU */
 
-int __cpu_up(unsigned int cpu)
+/* Upping and downing of CPUs */
+int __cpuinit __cpu_up(unsigned int cpu)
 {
        struct task_struct *idle;
        struct _lowcore *cpu_lowcore;
        struct stack_frame *sf;
        sigp_ccode ccode;
-       int curr_cpu;
 
-       for (curr_cpu = 0; curr_cpu <= 65535; curr_cpu++) {
-               __cpu_logical_map[cpu] = (__u16) curr_cpu;
-               if (cpu_stopped(cpu))
-                       break;
-       }
-
-       if (!cpu_stopped(cpu))
-               return -ENODEV;
+       if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
+               return -EIO;
+       if (smp_alloc_lowcore(cpu))
+               return -ENOMEM;
 
        ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
                                   cpu, sigp_set_prefix);
@@ -515,6 +705,7 @@ int __cpu_up(unsigned int cpu)
        cpu_lowcore = lowcore_ptr[cpu];
        cpu_lowcore->kernel_stack = (unsigned long)
                task_stack_page(idle) + THREAD_SIZE;
+       cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
        sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
                                     - sizeof(struct pt_regs)
                                     - sizeof(struct stack_frame));
@@ -528,6 +719,8 @@ int __cpu_up(unsigned int cpu)
        cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
        cpu_lowcore->current_task = (unsigned long) idle;
        cpu_lowcore->cpu_data.cpu_nr = cpu;
+       cpu_lowcore->softirq_pending = 0;
+       cpu_lowcore->ext_call_fast = 0;
        eieio();
 
        while (signal_processor(cpu, sigp_restart) == sigp_busy)
@@ -538,44 +731,20 @@ int __cpu_up(unsigned int cpu)
        return 0;
 }
 
-static unsigned int __initdata additional_cpus;
-static unsigned int __initdata possible_cpus;
-
-void __init smp_setup_cpu_possible_map(void)
+static int __init setup_possible_cpus(char *s)
 {
-       unsigned int phy_cpus, pos_cpus, cpu;
-
-       phy_cpus = smp_count_cpus();
-       pos_cpus = min(phy_cpus + additional_cpus, (unsigned int) NR_CPUS);
-
-       if (possible_cpus)
-               pos_cpus = min(possible_cpus, (unsigned int) NR_CPUS);
+       int pcpus, cpu;
 
-       for (cpu = 0; cpu < pos_cpus; cpu++)
+       pcpus = simple_strtoul(s, NULL, 0);
+       cpu_possible_map = cpumask_of_cpu(0);
+       for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
                cpu_set(cpu, cpu_possible_map);
-
-       phy_cpus = min(phy_cpus, pos_cpus);
-
-       for (cpu = 0; cpu < phy_cpus; cpu++)
-               cpu_set(cpu, cpu_present_map);
-}
-
-#ifdef CONFIG_HOTPLUG_CPU
-
-static int __init setup_additional_cpus(char *s)
-{
-       additional_cpus = simple_strtoul(s, NULL, 0);
-       return 0;
-}
-early_param("additional_cpus", setup_additional_cpus);
-
-static int __init setup_possible_cpus(char *s)
-{
-       possible_cpus = simple_strtoul(s, NULL, 0);
        return 0;
 }
 early_param("possible_cpus", setup_possible_cpus);
 
+#ifdef CONFIG_HOTPLUG_CPU
+
 int __cpu_disable(void)
 {
        struct ec_creg_mask_parms cr_parms;
@@ -612,7 +781,8 @@ void __cpu_die(unsigned int cpu)
        /* Wait until target cpu is down */
        while (!smp_cpu_not_running(cpu))
                cpu_relax();
-       printk("Processor %d spun down\n", cpu);
+       smp_free_lowcore(cpu);
+       printk(KERN_INFO "Processor %d spun down\n", cpu);
 }
 
 void cpu_die(void)
@@ -625,49 +795,19 @@ void cpu_die(void)
 
 #endif /* CONFIG_HOTPLUG_CPU */
 
-/*
- *     Cycle through the processors and setup structures.
- */
-
 void __init smp_prepare_cpus(unsigned int max_cpus)
 {
-       unsigned long stack;
        unsigned int cpu;
-       int i;
+
+       smp_detect_cpus();
 
        /* request the 0x1201 emergency signal external interrupt */
        if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
                panic("Couldn't request external interrupt 0x1201");
        memset(lowcore_ptr, 0, sizeof(lowcore_ptr));
-       /*
-        *  Initialize prefix pages and stacks for all possible cpus
-        */
        print_cpu_info(&S390_lowcore.cpu_data);
+       smp_alloc_lowcore(smp_processor_id());
 
-       for_each_possible_cpu(i) {
-               lowcore_ptr[i] = (struct _lowcore *)
-                       __get_free_pages(GFP_KERNEL | GFP_DMA,
-                                        sizeof(void*) == 8 ? 1 : 0);
-               stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
-               if (!lowcore_ptr[i] || !stack)
-                       panic("smp_boot_cpus failed to allocate memory\n");
-
-               *(lowcore_ptr[i]) = S390_lowcore;
-               lowcore_ptr[i]->async_stack = stack + ASYNC_SIZE;
-               stack = __get_free_pages(GFP_KERNEL, 0);
-               if (!stack)
-                       panic("smp_boot_cpus failed to allocate memory\n");
-               lowcore_ptr[i]->panic_stack = stack + PAGE_SIZE;
-#ifndef CONFIG_64BIT
-               if (MACHINE_HAS_IEEE) {
-                       lowcore_ptr[i]->extended_save_area_addr =
-                               (__u32) __get_free_pages(GFP_KERNEL, 0);
-                       if (!lowcore_ptr[i]->extended_save_area_addr)
-                               panic("smp_boot_cpus failed to "
-                                     "allocate memory\n");
-               }
-#endif
-       }
 #ifndef CONFIG_64BIT
        if (MACHINE_HAS_IEEE)
                ctl_set_bit(14, 29); /* enable extended save area */
@@ -683,15 +823,17 @@ void __init smp_prepare_boot_cpu(void)
 {
        BUG_ON(smp_processor_id() != 0);
 
+       current_thread_info()->cpu = 0;
+       cpu_set(0, cpu_present_map);
        cpu_set(0, cpu_online_map);
        S390_lowcore.percpu_offset = __per_cpu_offset[0];
        current_set[0] = current;
+       smp_cpu_state[0] = CPU_STATE_CONFIGURED;
        spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
 }
 
 void __init smp_cpus_done(unsigned int max_cpus)
 {
-       cpu_present_map = cpu_possible_map;
 }
 
 /*
@@ -705,7 +847,79 @@ int setup_profiling_timer(unsigned int multiplier)
        return 0;
 }
 
-static DEFINE_PER_CPU(struct cpu, cpu_devices);
+#ifdef CONFIG_HOTPLUG_CPU
+static ssize_t cpu_configure_show(struct sys_device *dev, char *buf)
+{
+       ssize_t count;
+
+       mutex_lock(&smp_cpu_state_mutex);
+       count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
+       mutex_unlock(&smp_cpu_state_mutex);
+       return count;
+}
+
+static ssize_t cpu_configure_store(struct sys_device *dev, const char *buf,
+                                  size_t count)
+{
+       int cpu = dev->id;
+       int val, rc;
+       char delim;
+
+       if (sscanf(buf, "%d %c", &val, &delim) != 1)
+               return -EINVAL;
+       if (val != 0 && val != 1)
+               return -EINVAL;
+
+       mutex_lock(&smp_cpu_state_mutex);
+       get_online_cpus();
+       rc = -EBUSY;
+       if (cpu_online(cpu))
+               goto out;
+       rc = 0;
+       switch (val) {
+       case 0:
+               if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
+                       rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
+                       if (!rc)
+                               smp_cpu_state[cpu] = CPU_STATE_STANDBY;
+               }
+               break;
+       case 1:
+               if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
+                       rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
+                       if (!rc)
+                               smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
+               }
+               break;
+       default:
+               break;
+       }
+out:
+       put_online_cpus();
+       mutex_unlock(&smp_cpu_state_mutex);
+       return rc ? rc : count;
+}
+static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
+#endif /* CONFIG_HOTPLUG_CPU */
+
+static ssize_t show_cpu_address(struct sys_device *dev, char *buf)
+{
+       return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
+}
+static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
+
+
+static struct attribute *cpu_common_attrs[] = {
+#ifdef CONFIG_HOTPLUG_CPU
+       &attr_configure.attr,
+#endif
+       &attr_address.attr,
+       NULL,
+};
+
+static struct attribute_group cpu_common_attr_group = {
+       .attrs = cpu_common_attrs,
+};
 
 static ssize_t show_capability(struct sys_device *dev, char *buf)
 {
@@ -750,15 +964,15 @@ static ssize_t show_idle_time(struct sys_device *dev, char *buf)
 }
 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
 
-static struct attribute *cpu_attrs[] = {
+static struct attribute *cpu_online_attrs[] = {
        &attr_capability.attr,
        &attr_idle_count.attr,
        &attr_idle_time_us.attr,
        NULL,
 };
 
-static struct attribute_group cpu_attr_group = {
-       .attrs = cpu_attrs,
+static struct attribute_group cpu_online_attr_group = {
+       .attrs = cpu_online_attrs,
 };
 
 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
@@ -778,12 +992,12 @@ static int __cpuinit smp_cpu_notify(struct notifier_block *self,
                idle->idle_time = 0;
                idle->idle_count = 0;
                spin_unlock_irq(&idle->lock);
-               if (sysfs_create_group(&s->kobj, &cpu_attr_group))
+               if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
                        return NOTIFY_BAD;
                break;
        case CPU_DEAD:
        case CPU_DEAD_FROZEN:
-               sysfs_remove_group(&s->kobj, &cpu_attr_group);
+               sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
                break;
        }
        return NOTIFY_OK;
@@ -793,6 +1007,62 @@ static struct notifier_block __cpuinitdata smp_cpu_nb = {
        .notifier_call = smp_cpu_notify,
 };
 
+static int __devinit smp_add_present_cpu(int cpu)
+{
+       struct cpu *c = &per_cpu(cpu_devices, cpu);
+       struct sys_device *s = &c->sysdev;
+       int rc;
+
+       c->hotpluggable = 1;
+       rc = register_cpu(c, cpu);
+       if (rc)
+               goto out;
+       rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
+       if (rc)
+               goto out_cpu;
+       if (!cpu_online(cpu))
+               goto out;
+       rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
+       if (!rc)
+               return 0;
+       sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
+out_cpu:
+#ifdef CONFIG_HOTPLUG_CPU
+       unregister_cpu(c);
+#endif
+out:
+       return rc;
+}
+
+#ifdef CONFIG_HOTPLUG_CPU
+static ssize_t __ref rescan_store(struct sys_device *dev,
+                                 const char *buf, size_t count)
+{
+       cpumask_t newcpus;
+       int cpu;
+       int rc;
+
+       mutex_lock(&smp_cpu_state_mutex);
+       get_online_cpus();
+       newcpus = cpu_present_map;
+       rc = smp_rescan_cpus();
+       if (rc)
+               goto out;
+       cpus_andnot(newcpus, cpu_present_map, newcpus);
+       for_each_cpu_mask(cpu, newcpus) {
+               rc = smp_add_present_cpu(cpu);
+               if (rc)
+                       cpu_clear(cpu, cpu_present_map);
+       }
+       rc = 0;
+out:
+       put_online_cpus();
+       mutex_unlock(&smp_cpu_state_mutex);
+       return rc ? rc : count;
+}
+static SYSDEV_ATTR(rescan, 0200, NULL, rescan_store);
+#endif /* CONFIG_HOTPLUG_CPU */
+
 static int __init topology_init(void)
 {
        int cpu;
@@ -800,16 +1070,14 @@ static int __init topology_init(void)
 
        register_cpu_notifier(&smp_cpu_nb);
 
-       for_each_possible_cpu(cpu) {
-               struct cpu *c = &per_cpu(cpu_devices, cpu);
-               struct sys_device *s = &c->sysdev;
-
-               c->hotpluggable = 1;
-               register_cpu(c, cpu);
-               if (!cpu_online(cpu))
-                       continue;
-               s = &c->sysdev;
-               rc = sysfs_create_group(&s->kobj, &cpu_attr_group);
+#ifdef CONFIG_HOTPLUG_CPU
+       rc = sysfs_create_file(&cpu_sysdev_class.kset.kobj,
+                              &attr_rescan.attr);
+       if (rc)
+               return rc;
+#endif
+       for_each_present_cpu(cpu) {
+               rc = smp_add_present_cpu(cpu);
                if (rc)
                        return rc;
        }