include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit...
[safe/jmp/linux-2.6] / arch / s390 / kernel / smp.c
index c0cd255..e4d98de 100644 (file)
@@ -1,12 +1,12 @@
 /*
  *  arch/s390/kernel/smp.c
  *
- *    Copyright (C) IBM Corp. 1999,2006
+ *    Copyright IBM Corp. 1999, 2009
  *    Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
- *               Martin Schwidefsky (schwidefsky@de.ibm.com)
- *               Heiko Carstens (heiko.carstens@de.ibm.com)
+ *              Martin Schwidefsky (schwidefsky@de.ibm.com)
+ *              Heiko Carstens (heiko.carstens@de.ibm.com)
  *
- *  based on other smp stuff by 
+ *  based on other smp stuff by
  *    (c) 1995 Alan Cox, CymruNET Ltd  <alan@cymru.net>
  *    (c) 1998 Ingo Molnar
  *
  * cpu_number_map in other architectures.
  */
 
+#define KMSG_COMPONENT "cpu"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
 #include <linux/module.h>
 #include <linux/init.h>
-
 #include <linux/mm.h>
+#include <linux/err.h>
 #include <linux/spinlock.h>
 #include <linux/kernel_stat.h>
-#include <linux/smp_lock.h>
-
 #include <linux/delay.h>
 #include <linux/cache.h>
 #include <linux/interrupt.h>
+#include <linux/irqflags.h>
 #include <linux/cpu.h>
-
+#include <linux/timex.h>
+#include <linux/bootmem.h>
+#include <linux/slab.h>
+#include <asm/asm-offsets.h>
+#include <asm/ipl.h>
+#include <asm/setup.h>
 #include <asm/sigp.h>
 #include <asm/pgalloc.h>
 #include <asm/irq.h>
 #include <asm/s390_ext.h>
 #include <asm/cpcmd.h>
 #include <asm/tlbflush.h>
+#include <asm/timer.h>
+#include <asm/lowcore.h>
+#include <asm/sclp.h>
+#include <asm/cputime.h>
+#include <asm/vdso.h>
+#include <asm/cpu.h>
+#include "entry.h"
 
-extern volatile int __cpu_logical_map[];
-
-/*
- * An array with a pointer the lowcore of every CPU.
- */
-
-struct _lowcore *lowcore_ptr[NR_CPUS];
-
-cpumask_t cpu_online_map = CPU_MASK_NONE;
-cpumask_t cpu_possible_map = CPU_MASK_NONE;
+/* logical cpu to cpu address */
+unsigned short __cpu_logical_map[NR_CPUS];
 
 static struct task_struct *current_set[NR_CPUS];
 
-/*
- * Reboot, halt and power_off routines for SMP.
- */
-extern char vmhalt_cmd[];
-extern char vmpoff_cmd[];
+static u8 smp_cpu_type;
+static int smp_use_sigp_detection;
 
-static void smp_ext_bitcall(int, ec_bit_sig);
-static void smp_ext_bitcall_others(ec_bit_sig);
-
-/*
-5B * Structure and data for smp_call_function(). This is designed to minimise
- * static memory requirements. It also looks cleaner.
- */
-static DEFINE_SPINLOCK(call_lock);
-
-struct call_data_struct {
-       void (*func) (void *info);
-       void *info;
-       atomic_t started;
-       atomic_t finished;
-       int wait;
+enum s390_cpu_state {
+       CPU_STATE_STANDBY,
+       CPU_STATE_CONFIGURED,
 };
 
-static struct call_data_struct * call_data;
-
-/*
- * 'Call function' interrupt callback
- */
-static void do_call_function(void)
-{
-       void (*func) (void *info) = call_data->func;
-       void *info = call_data->info;
-       int wait = call_data->wait;
+DEFINE_MUTEX(smp_cpu_state_mutex);
+int smp_cpu_polarization[NR_CPUS];
+static int smp_cpu_state[NR_CPUS];
+static int cpu_management;
 
-       atomic_inc(&call_data->started);
-       (*func)(info);
-       if (wait)
-               atomic_inc(&call_data->finished);
-}
+static DEFINE_PER_CPU(struct cpu, cpu_devices);
 
-/*
- * this function sends a 'generic call function' IPI to all other CPUs
- * in the system.
- */
+static void smp_ext_bitcall(int, int);
 
-int smp_call_function (void (*func) (void *info), void *info, int nonatomic,
-                       int wait)
-/*
- * [SUMMARY] Run a function on all other CPUs.
- * <func> The function to run. This must be fast and non-blocking.
- * <info> An arbitrary pointer to pass to the function.
- * <nonatomic> currently unused.
- * <wait> If true, wait (atomically) until function has completed on other CPUs.
- * [RETURNS] 0 on success, else a negative status code. Does not return until
- * remote CPUs are nearly ready to execute <<func>> or are or have executed.
- *
- * You must not call this function with disabled interrupts or from a
- * hardware interrupt handler or from a bottom half handler.
- */
+static int raw_cpu_stopped(int cpu)
 {
-       struct call_data_struct data;
-       int cpus = num_online_cpus()-1;
-
-       if (cpus <= 0)
-               return 0;
-
-       /* Can deadlock when called with interrupts disabled */
-       WARN_ON(irqs_disabled());
-
-       data.func = func;
-       data.info = info;
-       atomic_set(&data.started, 0);
-       data.wait = wait;
-       if (wait)
-               atomic_set(&data.finished, 0);
-
-       spin_lock(&call_lock);
-       call_data = &data;
-       /* Send a message to all other CPUs and wait for them to respond */
-        smp_ext_bitcall_others(ec_call_function);
-
-       /* Wait for response */
-       while (atomic_read(&data.started) != cpus)
-               cpu_relax();
-
-       if (wait)
-               while (atomic_read(&data.finished) != cpus)
-                       cpu_relax();
-       spin_unlock(&call_lock);
+       u32 status;
 
+       switch (raw_sigp_ps(&status, 0, cpu, sigp_sense)) {
+       case sigp_status_stored:
+               /* Check for stopped and check stop state */
+               if (status & 0x50)
+                       return 1;
+               break;
+       default:
+               break;
+       }
        return 0;
 }
 
-/*
- * Call a function on one CPU
- * cpu : the CPU the function should be executed on
- *
- * You must not call this function with disabled interrupts or from a
- * hardware interrupt handler. You may call it from a bottom half.
- *
- * It is guaranteed that the called function runs on the specified CPU,
- * preemption is disabled.
- */
-int smp_call_function_on(void (*func) (void *info), void *info,
-                        int nonatomic, int wait, int cpu)
+static inline int cpu_stopped(int cpu)
 {
-       struct call_data_struct data;
-       int curr_cpu;
-
-       if (!cpu_online(cpu))
-               return -EINVAL;
-
-       /* disable preemption for local function call */
-       curr_cpu = get_cpu();
-
-       if (curr_cpu == cpu) {
-               /* direct call to function */
-               func(info);
-               put_cpu();
-               return 0;
-       }
-
-       data.func = func;
-       data.info = info;
-       atomic_set(&data.started, 0);
-       data.wait = wait;
-       if (wait)
-               atomic_set(&data.finished, 0);
-
-       spin_lock_bh(&call_lock);
-       call_data = &data;
-       smp_ext_bitcall(cpu, ec_call_function);
-
-       /* Wait for response */
-       while (atomic_read(&data.started) != 1)
-               cpu_relax();
-
-       if (wait)
-               while (atomic_read(&data.finished) != 1)
-                       cpu_relax();
-
-       spin_unlock_bh(&call_lock);
-       put_cpu();
-       return 0;
+       return raw_cpu_stopped(cpu_logical_map(cpu));
 }
-EXPORT_SYMBOL(smp_call_function_on);
 
-static inline void do_send_stop(void)
+void smp_switch_to_ipl_cpu(void (*func)(void *), void *data)
 {
-        int cpu, rc;
-
-        /* stop all processors */
-       for_each_online_cpu(cpu) {
-               if (cpu == smp_processor_id())
-                       continue;
-               do {
-                       rc = signal_processor(cpu, sigp_stop);
-               } while (rc == sigp_busy);
-       }
+       struct _lowcore *lc, *current_lc;
+       struct stack_frame *sf;
+       struct pt_regs *regs;
+       unsigned long sp;
+
+       if (smp_processor_id() == 0)
+               func(data);
+       __load_psw_mask(PSW_BASE_BITS | PSW_DEFAULT_KEY);
+       /* Disable lowcore protection */
+       __ctl_clear_bit(0, 28);
+       current_lc = lowcore_ptr[smp_processor_id()];
+       lc = lowcore_ptr[0];
+       if (!lc)
+               lc = current_lc;
+       lc->restart_psw.mask = PSW_BASE_BITS | PSW_DEFAULT_KEY;
+       lc->restart_psw.addr = PSW_ADDR_AMODE | (unsigned long) smp_restart_cpu;
+       if (!cpu_online(0))
+               smp_switch_to_cpu(func, data, 0, stap(), __cpu_logical_map[0]);
+       while (sigp(0, sigp_stop_and_store_status) == sigp_busy)
+               cpu_relax();
+       sp = lc->panic_stack;
+       sp -= sizeof(struct pt_regs);
+       regs = (struct pt_regs *) sp;
+       memcpy(&regs->gprs, &current_lc->gpregs_save_area, sizeof(regs->gprs));
+       regs->psw = lc->psw_save_area;
+       sp -= STACK_FRAME_OVERHEAD;
+       sf = (struct stack_frame *) sp;
+       sf->back_chain = regs->gprs[15];
+       smp_switch_to_cpu(func, data, sp, stap(), __cpu_logical_map[0]);
 }
 
-static inline void do_store_status(void)
+void smp_send_stop(void)
 {
-        int cpu, rc;
+       int cpu, rc;
 
-        /* store status of all processors in their lowcores (real 0) */
+       /* Disable all interrupts/machine checks */
+       __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
+       trace_hardirqs_off();
+
+       /* stop all processors */
        for_each_online_cpu(cpu) {
                if (cpu == smp_processor_id())
                        continue;
                do {
-                       rc = signal_processor_p(
-                               (__u32)(unsigned long) lowcore_ptr[cpu], cpu,
-                               sigp_store_status_at_address);
-               } while(rc == sigp_busy);
-        }
-}
-
-static inline void do_wait_for_stop(void)
-{
-       int cpu;
+                       rc = sigp(cpu, sigp_stop);
+               } while (rc == sigp_busy);
 
-       /* Wait for all other cpus to enter stopped state */
-       for_each_online_cpu(cpu) {
-               if (cpu == smp_processor_id())
-                       continue;
-               while(!smp_cpu_not_running(cpu))
+               while (!cpu_stopped(cpu))
                        cpu_relax();
        }
 }
 
 /*
- * this function sends a 'stop' sigp to all other CPUs in the system.
- * it goes straight through.
- */
-void smp_send_stop(void)
-{
-       /* Disable all interrupts/machine checks */
-       __load_psw_mask(PSW_KERNEL_BITS & ~PSW_MASK_MCHECK);
-
-        /* write magic number to zero page (absolute 0) */
-       lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
-
-       /* stop other processors. */
-       do_send_stop();
-
-       /* wait until other processors are stopped */
-       do_wait_for_stop();
-
-       /* store status of other processors. */
-       do_store_status();
-}
-
-/*
- * 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.
  */
 
-void do_ext_call_interrupt(__u16 code)
+static void do_ext_call_interrupt(__u16 code)
 {
-        unsigned long bits;
+       unsigned long bits;
 
-        /*
-         * handle bit signal external calls
-         *
-         * For the ec_schedule signal we have to do nothing. All the work
-         * is done automatically when we return from the interrupt.
-         */
+       /*
+        * handle bit signal external calls
+        *
+        * For the ec_schedule signal we have to do nothing. All the work
+        * is done automatically when we return from the interrupt.
+        */
        bits = xchg(&S390_lowcore.ext_call_fast, 0);
 
-       if (test_bit(ec_call_function, &bits)) 
-               do_call_function();
+       if (test_bit(ec_call_function, &bits))
+               generic_smp_call_function_interrupt();
+
+       if (test_bit(ec_call_function_single, &bits))
+               generic_smp_call_function_single_interrupt();
 }
 
 /*
  * Send an external call sigp to another cpu and return without waiting
  * for its completion.
  */
-static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
+static void smp_ext_bitcall(int cpu, int sig)
 {
-        /*
-         * Set signaling bit in lowcore of target cpu and kick it
-         */
+       /*
+        * Set signaling bit in lowcore of target cpu and kick it
+        */
        set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
-       while(signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
+       while (sigp(cpu, sigp_emergency_signal) == sigp_busy)
                udelay(10);
 }
 
-/*
- * Send an external call sigp to every other cpu in the system and
- * return without waiting for its completion.
- */
-static void smp_ext_bitcall_others(ec_bit_sig sig)
+void arch_send_call_function_ipi_mask(const struct cpumask *mask)
 {
-        int cpu;
+       int cpu;
 
-       for_each_online_cpu(cpu) {
-               if (cpu == smp_processor_id())
-                        continue;
-                /*
-                 * Set signaling bit in lowcore of target cpu and kick it
-                 */
-               set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
-               while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
-                       udelay(10);
-        }
+       for_each_cpu(cpu, mask)
+               smp_ext_bitcall(cpu, ec_call_function);
+}
+
+void arch_send_call_function_single_ipi(int cpu)
+{
+       smp_ext_bitcall(cpu, ec_call_function_single);
 }
 
 #ifndef CONFIG_64BIT
 /*
  * this function sends a 'purge tlb' signal to another CPU.
  */
-void smp_ptlb_callback(void *info)
+static void smp_ptlb_callback(void *info)
 {
-       local_flush_tlb();
+       __tlb_flush_local();
 }
 
 void smp_ptlb_all(void)
 {
-        on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
+       on_each_cpu(smp_ptlb_callback, NULL, 1);
 }
 EXPORT_SYMBOL(smp_ptlb_all);
 #endif /* ! CONFIG_64BIT */
@@ -371,7 +225,7 @@ EXPORT_SYMBOL(smp_ptlb_all);
  */
 void smp_send_reschedule(int cpu)
 {
-        smp_ext_bitcall(cpu, ec_schedule);
+       smp_ext_bitcall(cpu, ec_schedule);
 }
 
 /*
@@ -385,11 +239,12 @@ struct ec_creg_mask_parms {
 /*
  * callback for setting/clearing control bits
  */
-void smp_ctl_bit_callback(void *info) {
+static void smp_ctl_bit_callback(void *info)
+{
        struct ec_creg_mask_parms *pp = info;
        unsigned long cregs[16];
        int i;
-       
+
        __ctl_store(cregs, 0, 15);
        for (i = 0; i <= 15; i++)
                cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
@@ -406,8 +261,9 @@ void smp_ctl_set_bit(int cr, int bit)
        memset(&parms.orvals, 0, sizeof(parms.orvals));
        memset(&parms.andvals, 0xff, sizeof(parms.andvals));
        parms.orvals[cr] = 1 << bit;
-       on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
+       on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 }
+EXPORT_SYMBOL(smp_ctl_set_bit);
 
 /*
  * Clear a bit in a control register of all cpus
@@ -419,70 +275,204 @@ void smp_ctl_clear_bit(int cr, int bit)
        memset(&parms.orvals, 0, sizeof(parms.orvals));
        memset(&parms.andvals, 0xff, sizeof(parms.andvals));
        parms.andvals[cr] = ~(1L << bit);
-       on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
+       on_each_cpu(smp_ctl_bit_callback, &parms, 1);
 }
+EXPORT_SYMBOL(smp_ctl_clear_bit);
 
-/*
- * Lets check how many CPUs we have.
- */
+#ifdef CONFIG_ZFCPDUMP
 
-static unsigned int
-__init smp_count_cpus(void)
+static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
 {
-       unsigned int cpu, num_cpus;
-       __u16 boot_cpu_addr;
+       if (ipl_info.type != IPL_TYPE_FCP_DUMP)
+               return;
+       if (cpu >= NR_CPUS) {
+               pr_warning("CPU %i exceeds the maximum %i and is excluded from "
+                          "the dump\n", cpu, NR_CPUS - 1);
+               return;
+       }
+       zfcpdump_save_areas[cpu] = kmalloc(sizeof(struct save_area), GFP_KERNEL);
+       while (raw_sigp(phy_cpu, sigp_stop_and_store_status) == sigp_busy)
+               cpu_relax();
+       memcpy_real(zfcpdump_save_areas[cpu],
+                   (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
+                   sizeof(struct save_area));
+}
 
-       /*
-        * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
-        */
+struct save_area *zfcpdump_save_areas[NR_CPUS + 1];
+EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
 
-       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)
+#else
+
+static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
+
+#endif /* CONFIG_ZFCPDUMP */
+
+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 = cpumask_first(&avail);
+       if (logical_cpu >= nr_cpu_ids)
+               return 0;
+       for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
+               if (cpu_known(cpu_id))
                        continue;
-               __cpu_logical_map[1] = (__u16) cpu;
-               if (signal_processor(1, sigp_sense) ==
-                   sigp_not_operational)
+               __cpu_logical_map[logical_cpu] = cpu_id;
+               smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
+               if (!cpu_stopped(logical_cpu))
                        continue;
-               num_cpus++;
+               cpu_set(logical_cpu, cpu_present_map);
+               smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
+               logical_cpu = cpumask_next(logical_cpu, &avail);
+               if (logical_cpu >= nr_cpu_ids)
+                       break;
        }
+       return 0;
+}
 
-       printk("Detected %d CPU's\n",(int) num_cpus);
-       printk("Boot cpu address %2X\n", boot_cpu_addr);
+static int smp_rescan_cpus_sclp(cpumask_t avail)
+{
+       struct sclp_cpu_info *info;
+       int cpu_id, logical_cpu, cpu;
+       int rc;
 
-       return num_cpus;
+       logical_cpu = cpumask_first(&avail);
+       if (logical_cpu >= nr_cpu_ids)
+               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;
+               smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
+               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 = cpumask_next(logical_cpu, &avail);
+               if (logical_cpu >= nr_cpu_ids)
+                       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 = __cpu_logical_map[0];
+       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 <= MAX_CPU_ADDRESS; cpu++) {
+                       if (cpu == boot_cpu_addr)
+                               continue;
+                       if (!raw_cpu_stopped(cpu))
+                               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;
+               if (!raw_cpu_stopped(cpu_addr)) {
+                       s_cpus++;
+                       continue;
+               }
+               smp_get_save_area(c_cpus, cpu_addr);
+               c_cpus++;
+       }
+out:
+       kfree(info);
+       pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
+       get_online_cpus();
+       __smp_rescan_cpus();
+       put_online_cpus();
 }
 
 /*
- *      Activate a secondary processor.
+ *     Activate a secondary processor.
  */
-extern void init_cpu_timer(void);
-extern void init_cpu_vtimer(void);
-
-int __devinit start_secondary(void *cpuvoid)
+int __cpuinit start_secondary(void *cpuvoid)
 {
-        /* Setup the cpu */
-        cpu_init();
+       /* Setup the cpu */
+       cpu_init();
        preempt_disable();
-        /* init per CPU timer */
-        init_cpu_timer();
-#ifdef CONFIG_VIRT_TIMER
-        init_cpu_vtimer();
-#endif
+       /* Enable TOD clock interrupts on the secondary cpu. */
+       init_cpu_timer();
+       /* Enable cpu timer interrupts on the secondary cpu. */
+       init_cpu_vtimer();
        /* Enable pfault pseudo page faults on this cpu. */
        pfault_init();
 
+       /* call cpu notifiers */
+       notify_cpu_starting(smp_processor_id());
        /* Mark this cpu as online */
+       ipi_call_lock();
        cpu_set(smp_processor_id(), cpu_online_map);
+       ipi_call_unlock();
        /* Switch on interrupts */
        local_irq_enable();
-        /* Print info about this processor */
-        print_cpu_info(&S390_lowcore.cpu_data);
-        /* cpu_idle will call schedule for us */
-        cpu_idle();
-        return 0;
+       /* Print info about this processor */
+       print_cpu_info();
+       /* cpu_idle will call schedule for us */
+       cpu_idle();
+       return 0;
 }
 
 static void __init smp_create_idle(unsigned int cpu)
@@ -499,168 +489,142 @@ static void __init smp_create_idle(unsigned int cpu)
        current_set[cpu] = p;
 }
 
-/* Reserving and releasing of CPUs */
-
-static DEFINE_SPINLOCK(smp_reserve_lock);
-static int smp_cpu_reserved[NR_CPUS];
-
-int
-smp_get_cpu(cpumask_t cpu_mask)
+static int __cpuinit smp_alloc_lowcore(int cpu)
 {
-       unsigned long flags;
-       int cpu;
+       unsigned long async_stack, panic_stack;
+       struct _lowcore *lowcore;
+
+       lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
+       if (!lowcore)
+               return -ENOMEM;
+       async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
+       panic_stack = __get_free_page(GFP_KERNEL);
+       if (!panic_stack || !async_stack)
+               goto out;
+       memcpy(lowcore, &S390_lowcore, 512);
+       memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
+       lowcore->async_stack = async_stack + ASYNC_SIZE;
+       lowcore->panic_stack = panic_stack + PAGE_SIZE;
 
-       spin_lock_irqsave(&smp_reserve_lock, flags);
-       /* Try to find an already reserved cpu. */
-       for_each_cpu_mask(cpu, cpu_mask) {
-               if (smp_cpu_reserved[cpu] != 0) {
-                       smp_cpu_reserved[cpu]++;
-                       /* Found one. */
-                       goto out;
-               }
-       }
-       /* Reserve a new cpu from cpu_mask. */
-       for_each_cpu_mask(cpu, cpu_mask) {
-               if (cpu_online(cpu)) {
-                       smp_cpu_reserved[cpu]++;
+#ifndef CONFIG_64BIT
+       if (MACHINE_HAS_IEEE) {
+               unsigned long save_area;
+
+               save_area = get_zeroed_page(GFP_KERNEL);
+               if (!save_area)
                        goto out;
-               }
+               lowcore->extended_save_area_addr = (u32) save_area;
        }
-       cpu = -ENODEV;
-out:
-       spin_unlock_irqrestore(&smp_reserve_lock, flags);
-       return cpu;
-}
-
-void
-smp_put_cpu(int cpu)
-{
-       unsigned long flags;
+#else
+       if (vdso_alloc_per_cpu(cpu, lowcore))
+               goto out;
+#endif
+       lowcore_ptr[cpu] = lowcore;
+       return 0;
 
-       spin_lock_irqsave(&smp_reserve_lock, flags);
-       smp_cpu_reserved[cpu]--;
-       spin_unlock_irqrestore(&smp_reserve_lock, flags);
+out:
+       free_page(panic_stack);
+       free_pages(async_stack, ASYNC_ORDER);
+       free_pages((unsigned long) lowcore, LC_ORDER);
+       return -ENOMEM;
 }
 
-static inline int
-cpu_stopped(int cpu)
+static void smp_free_lowcore(int cpu)
 {
-       __u32 status;
+       struct _lowcore *lowcore;
 
-       /* Check for stopped state */
-       if (signal_processor_ps(&status, 0, cpu, sigp_sense) == sigp_status_stored) {
-               if (status & 0x40)
-                       return 1;
-       }
-       return 0;
+       lowcore = lowcore_ptr[cpu];
+#ifndef CONFIG_64BIT
+       if (MACHINE_HAS_IEEE)
+               free_page((unsigned long) lowcore->extended_save_area_addr);
+#else
+       vdso_free_per_cpu(cpu, lowcore);
+#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;
 }
 
 /* Upping and downing of CPUs */
-
-int
-__cpu_up(unsigned int cpu)
+int __cpuinit __cpu_up(unsigned int cpu)
 {
+       struct _lowcore *cpu_lowcore;
        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;
-       }
+       u32 lowcore;
+       int ccode;
 
-       if (!cpu_stopped(cpu))
-               return -ENODEV;
-
-       ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
-                                  cpu, sigp_set_prefix);
-       if (ccode){
-               printk("sigp_set_prefix failed for cpu %d "
-                      "with condition code %d\n",
-                      (int) cpu, (int) ccode);
+       if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
                return -EIO;
-       }
+       if (smp_alloc_lowcore(cpu))
+               return -ENOMEM;
+       do {
+               ccode = sigp(cpu, sigp_initial_cpu_reset);
+               if (ccode == sigp_busy)
+                       udelay(10);
+               if (ccode == sigp_not_operational)
+                       goto err_out;
+       } while (ccode == sigp_busy);
+
+       lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
+       while (sigp_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
+               udelay(10);
 
        idle = current_set[cpu];
-        cpu_lowcore = lowcore_ptr[cpu];
+       cpu_lowcore = lowcore_ptr[cpu];
        cpu_lowcore->kernel_stack = (unsigned long)
-               task_stack_page(idle) + (THREAD_SIZE);
+               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));
        memset(sf, 0, sizeof(struct stack_frame));
        sf->gprs[9] = (unsigned long) sf;
        cpu_lowcore->save_area[15] = (unsigned long) sf;
-       __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
+       __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
        asm volatile(
                "       stam    0,15,0(%0)"
                : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
        cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
-        cpu_lowcore->current_task = (unsigned long) idle;
-        cpu_lowcore->cpu_data.cpu_nr = cpu;
+       cpu_lowcore->current_task = (unsigned long) idle;
+       cpu_lowcore->cpu_nr = cpu;
+       cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
+       cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
+       cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
        eieio();
 
-       while (signal_processor(cpu,sigp_restart) == sigp_busy)
+       while (sigp(cpu, sigp_restart) == sigp_busy)
                udelay(10);
 
        while (!cpu_online(cpu))
                cpu_relax();
        return 0;
-}
-
-static unsigned int __initdata additional_cpus;
-static unsigned int __initdata possible_cpus;
-
-void __init smp_setup_cpu_possible_map(void)
-{
-       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);
-
-       for (cpu = 0; cpu < pos_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);
+err_out:
+       smp_free_lowcore(cpu);
+       return -EIO;
 }
 
-#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);
+       int pcpus, cpu;
+
+       pcpus = simple_strtoul(s, NULL, 0);
+       init_cpu_possible(cpumask_of(0));
+       for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
+               set_cpu_possible(cpu, true);
        return 0;
 }
 early_param("possible_cpus", setup_possible_cpus);
 
-int
-__cpu_disable(void)
+#ifdef CONFIG_HOTPLUG_CPU
+
+int __cpu_disable(void)
 {
-       unsigned long flags;
        struct ec_creg_mask_parms cr_parms;
        int cpu = smp_processor_id();
 
-       spin_lock_irqsave(&smp_reserve_lock, flags);
-       if (smp_cpu_reserved[cpu] != 0) {
-               spin_unlock_irqrestore(&smp_reserve_lock, flags);
-               return -EBUSY;
-       }
        cpu_clear(cpu, cpu_online_map);
 
        /* Disable pfault pseudo page faults on this cpu. */
@@ -671,108 +635,111 @@ __cpu_disable(void)
 
        /* disable all external interrupts */
        cr_parms.orvals[0] = 0;
-       cr_parms.andvals[0] = ~(1<<15 | 1<<14 | 1<<13 | 1<<12 |
-                               1<<11 | 1<<10 | 1<< 6 | 1<< 4);
+       cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
+                               1 << 11 | 1 << 10 | 1 <<  6 | 1 <<  4);
        /* disable all I/O interrupts */
        cr_parms.orvals[6] = 0;
-       cr_parms.andvals[6] = ~(1<<31 | 1<<30 | 1<<29 | 1<<28 |
-                               1<<27 | 1<<26 | 1<<25 | 1<<24);
+       cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
+                               1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
        /* disable most machine checks */
        cr_parms.orvals[14] = 0;
-       cr_parms.andvals[14] = ~(1<<28 | 1<<27 | 1<<26 | 1<<25 | 1<<24);
+       cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
+                                1 << 25 | 1 << 24);
 
        smp_ctl_bit_callback(&cr_parms);
 
-       spin_unlock_irqrestore(&smp_reserve_lock, flags);
        return 0;
 }
 
-void
-__cpu_die(unsigned int cpu)
+void __cpu_die(unsigned int cpu)
 {
        /* Wait until target cpu is down */
-       while (!smp_cpu_not_running(cpu))
+       while (!cpu_stopped(cpu))
                cpu_relax();
-       printk("Processor %d spun down\n", cpu);
+       while (sigp_p(0, cpu, sigp_set_prefix) == sigp_busy)
+               udelay(10);
+       smp_free_lowcore(cpu);
+       pr_info("Processor %d stopped\n", cpu);
 }
 
-void
-cpu_die(void)
+void cpu_die(void)
 {
        idle_task_exit();
-       signal_processor(smp_processor_id(), sigp_stop);
-       BUG();
-       for(;;);
+       while (sigp(smp_processor_id(), sigp_stop) == sigp_busy)
+               cpu_relax();
+       for (;;);
 }
 
 #endif /* CONFIG_HOTPLUG_CPU */
 
-/*
- *     Cycle through the processors and setup structures.
- */
-
 void __init smp_prepare_cpus(unsigned int max_cpus)
 {
-       unsigned long stack;
+#ifndef CONFIG_64BIT
+       unsigned long save_area = 0;
+#endif
+       unsigned long async_stack, panic_stack;
+       struct _lowcore *lowcore;
        unsigned int cpu;
-        int i;
-
-        /* 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);
-
-        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] == NULL || stack == 0ULL)
-                       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 == 0ULL)
-                       panic("smp_boot_cpus failed to allocate memory\n");
-               lowcore_ptr[i]->panic_stack = stack + (PAGE_SIZE);
+
+       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");
+       print_cpu_info();
+
+       /* Reallocate current lowcore, but keep its contents. */
+       lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
+       panic_stack = __get_free_page(GFP_KERNEL);
+       async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
+       BUG_ON(!lowcore || !panic_stack || !async_stack);
 #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 == 0)
-                               panic("smp_boot_cpus failed to "
-                                     "allocate memory\n");
-               }
+       if (MACHINE_HAS_IEEE)
+               save_area = get_zeroed_page(GFP_KERNEL);
 #endif
-       }
+       local_irq_disable();
+       local_mcck_disable();
+       lowcore_ptr[smp_processor_id()] = lowcore;
+       *lowcore = S390_lowcore;
+       lowcore->panic_stack = panic_stack + PAGE_SIZE;
+       lowcore->async_stack = async_stack + ASYNC_SIZE;
 #ifndef CONFIG_64BIT
        if (MACHINE_HAS_IEEE)
-               ctl_set_bit(14, 29); /* enable extended save area */
+               lowcore->extended_save_area_addr = (u32) save_area;
+#endif
+       set_prefix((u32)(unsigned long) lowcore);
+       local_mcck_enable();
+       local_irq_enable();
+#ifdef CONFIG_64BIT
+       if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
+               BUG();
 #endif
-       set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
-
        for_each_possible_cpu(cpu)
                if (cpu != smp_processor_id())
                        smp_create_idle(cpu);
 }
 
-void __devinit smp_prepare_boot_cpu(void)
+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;
+       smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
 }
 
-void smp_cpus_done(unsigned int max_cpus)
+void __init smp_cpus_done(unsigned int max_cpus)
 {
-       cpu_present_map = cpu_possible_map;
+}
+
+void __init smp_setup_processor_id(void)
+{
+       S390_lowcore.cpu_nr = 0;
+       __cpu_logical_map[0] = stap();
 }
 
 /*
@@ -783,35 +750,350 @@ void smp_cpus_done(unsigned int max_cpus)
  */
 int setup_profiling_timer(unsigned int multiplier)
 {
-        return 0;
+       return 0;
 }
 
-static DEFINE_PER_CPU(struct cpu, cpu_devices);
+#ifdef CONFIG_HOTPLUG_CPU
+static ssize_t cpu_configure_show(struct sys_device *dev,
+                               struct sysdev_attribute *attr, 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,
+                                 struct sysdev_attribute *attr,
+                                 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;
+
+       get_online_cpus();
+       mutex_lock(&smp_cpu_state_mutex);
+       rc = -EBUSY;
+       /* disallow configuration changes of online cpus and cpu 0 */
+       if (cpu_online(cpu) || cpu == 0)
+               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;
+                               smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
+                       }
+               }
+               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;
+                               smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
+                       }
+               }
+               break;
+       default:
+               break;
+       }
+out:
+       mutex_unlock(&smp_cpu_state_mutex);
+       put_online_cpus();
+       return rc ? rc : count;
+}
+static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
+#endif /* CONFIG_HOTPLUG_CPU */
+
+static ssize_t cpu_polarization_show(struct sys_device *dev,
+                                    struct sysdev_attribute *attr, char *buf)
+{
+       int cpu = dev->id;
+       ssize_t count;
+
+       mutex_lock(&smp_cpu_state_mutex);
+       switch (smp_cpu_polarization[cpu]) {
+       case POLARIZATION_HRZ:
+               count = sprintf(buf, "horizontal\n");
+               break;
+       case POLARIZATION_VL:
+               count = sprintf(buf, "vertical:low\n");
+               break;
+       case POLARIZATION_VM:
+               count = sprintf(buf, "vertical:medium\n");
+               break;
+       case POLARIZATION_VH:
+               count = sprintf(buf, "vertical:high\n");
+               break;
+       default:
+               count = sprintf(buf, "unknown\n");
+               break;
+       }
+       mutex_unlock(&smp_cpu_state_mutex);
+       return count;
+}
+static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
+
+static ssize_t show_cpu_address(struct sys_device *dev,
+                               struct sysdev_attribute *attr, 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,
+       &attr_polarization.attr,
+       NULL,
+};
+
+static struct attribute_group cpu_common_attr_group = {
+       .attrs = cpu_common_attrs,
+};
+
+static ssize_t show_capability(struct sys_device *dev,
+                               struct sysdev_attribute *attr, char *buf)
+{
+       unsigned int capability;
+       int rc;
+
+       rc = get_cpu_capability(&capability);
+       if (rc)
+               return rc;
+       return sprintf(buf, "%u\n", capability);
+}
+static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
+
+static ssize_t show_idle_count(struct sys_device *dev,
+                               struct sysdev_attribute *attr, char *buf)
+{
+       struct s390_idle_data *idle;
+       unsigned long long idle_count;
+       unsigned int sequence;
+
+       idle = &per_cpu(s390_idle, dev->id);
+repeat:
+       sequence = idle->sequence;
+       smp_rmb();
+       if (sequence & 1)
+               goto repeat;
+       idle_count = idle->idle_count;
+       if (idle->idle_enter)
+               idle_count++;
+       smp_rmb();
+       if (idle->sequence != sequence)
+               goto repeat;
+       return sprintf(buf, "%llu\n", idle_count);
+}
+static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
+
+static ssize_t show_idle_time(struct sys_device *dev,
+                               struct sysdev_attribute *attr, char *buf)
+{
+       struct s390_idle_data *idle;
+       unsigned long long now, idle_time, idle_enter;
+       unsigned int sequence;
+
+       idle = &per_cpu(s390_idle, dev->id);
+       now = get_clock();
+repeat:
+       sequence = idle->sequence;
+       smp_rmb();
+       if (sequence & 1)
+               goto repeat;
+       idle_time = idle->idle_time;
+       idle_enter = idle->idle_enter;
+       if (idle_enter != 0ULL && idle_enter < now)
+               idle_time += now - idle_enter;
+       smp_rmb();
+       if (idle->sequence != sequence)
+               goto repeat;
+       return sprintf(buf, "%llu\n", idle_time >> 12);
+}
+static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
+
+static struct attribute *cpu_online_attrs[] = {
+       &attr_capability.attr,
+       &attr_idle_count.attr,
+       &attr_idle_time_us.attr,
+       NULL,
+};
+
+static struct attribute_group cpu_online_attr_group = {
+       .attrs = cpu_online_attrs,
+};
+
+static int __cpuinit smp_cpu_notify(struct notifier_block *self,
+                                   unsigned long action, void *hcpu)
+{
+       unsigned int cpu = (unsigned int)(long)hcpu;
+       struct cpu *c = &per_cpu(cpu_devices, cpu);
+       struct sys_device *s = &c->sysdev;
+       struct s390_idle_data *idle;
+
+       switch (action) {
+       case CPU_ONLINE:
+       case CPU_ONLINE_FROZEN:
+               idle = &per_cpu(s390_idle, cpu);
+               memset(idle, 0, sizeof(struct s390_idle_data));
+               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_online_attr_group);
+               break;
+       }
+       return NOTIFY_OK;
+}
+
+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
+
+int __ref smp_rescan_cpus(void)
+{
+       cpumask_t newcpus;
+       int cpu;
+       int rc;
+
+       get_online_cpus();
+       mutex_lock(&smp_cpu_state_mutex);
+       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:
+       mutex_unlock(&smp_cpu_state_mutex);
+       put_online_cpus();
+       if (!cpus_empty(newcpus))
+               topology_schedule_update();
+       return rc;
+}
+
+static ssize_t __ref rescan_store(struct sysdev_class *class,
+                                 struct sysdev_class_attribute *attr,
+                                 const char *buf,
+                                 size_t count)
+{
+       int rc;
+
+       rc = smp_rescan_cpus();
+       return rc ? rc : count;
+}
+static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
+#endif /* CONFIG_HOTPLUG_CPU */
+
+static ssize_t dispatching_show(struct sysdev_class *class,
+                               struct sysdev_class_attribute *attr,
+                               char *buf)
+{
+       ssize_t count;
+
+       mutex_lock(&smp_cpu_state_mutex);
+       count = sprintf(buf, "%d\n", cpu_management);
+       mutex_unlock(&smp_cpu_state_mutex);
+       return count;
+}
+
+static ssize_t dispatching_store(struct sysdev_class *dev,
+                                struct sysdev_class_attribute *attr,
+                                const char *buf,
+                                size_t count)
+{
+       int val, rc;
+       char delim;
+
+       if (sscanf(buf, "%d %c", &val, &delim) != 1)
+               return -EINVAL;
+       if (val != 0 && val != 1)
+               return -EINVAL;
+       rc = 0;
+       get_online_cpus();
+       mutex_lock(&smp_cpu_state_mutex);
+       if (cpu_management == val)
+               goto out;
+       rc = topology_set_cpu_management(val);
+       if (!rc)
+               cpu_management = val;
+out:
+       mutex_unlock(&smp_cpu_state_mutex);
+       put_online_cpus();
+       return rc ? rc : count;
+}
+static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
+                        dispatching_store);
 
 static int __init topology_init(void)
 {
        int cpu;
-       int ret;
+       int rc;
 
-       for_each_possible_cpu(cpu) {
-               struct cpu *c = &per_cpu(cpu_devices, cpu);
+       register_cpu_notifier(&smp_cpu_nb);
 
-               c->hotpluggable = 1;
-               ret = register_cpu(c, cpu);
-               if (ret)
-                       printk(KERN_WARNING "topology_init: register_cpu %d "
-                              "failed (%d)\n", cpu, ret);
+#ifdef CONFIG_HOTPLUG_CPU
+       rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
+       if (rc)
+               return rc;
+#endif
+       rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
+       if (rc)
+               return rc;
+       for_each_present_cpu(cpu) {
+               rc = smp_add_present_cpu(cpu);
+               if (rc)
+                       return rc;
        }
        return 0;
 }
-
 subsys_initcall(topology_init);
-
-EXPORT_SYMBOL(cpu_online_map);
-EXPORT_SYMBOL(cpu_possible_map);
-EXPORT_SYMBOL(lowcore_ptr);
-EXPORT_SYMBOL(smp_ctl_set_bit);
-EXPORT_SYMBOL(smp_ctl_clear_bit);
-EXPORT_SYMBOL(smp_call_function);
-EXPORT_SYMBOL(smp_get_cpu);
-EXPORT_SYMBOL(smp_put_cpu);