2 * arch/s390/kernel/smp.c
4 * Copyright IBM Corp. 1999, 2009
5 * Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
6 * Martin Schwidefsky (schwidefsky@de.ibm.com)
7 * Heiko Carstens (heiko.carstens@de.ibm.com)
9 * based on other smp stuff by
10 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
11 * (c) 1998 Ingo Molnar
13 * We work with logical cpu numbering everywhere we can. The only
14 * functions using the real cpu address (got from STAP) are the sigp
15 * functions. For all other functions we use the identity mapping.
16 * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
17 * used e.g. to find the idle task belonging to a logical cpu. Every array
18 * in the kernel is sorted by the logical cpu number and not by the physical
19 * one which is causing all the confusion with __cpu_logical_map and
20 * cpu_number_map in other architectures.
23 #define KMSG_COMPONENT "cpu"
24 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
26 #include <linux/module.h>
27 #include <linux/init.h>
29 #include <linux/err.h>
30 #include <linux/spinlock.h>
31 #include <linux/kernel_stat.h>
32 #include <linux/delay.h>
33 #include <linux/cache.h>
34 #include <linux/interrupt.h>
35 #include <linux/irqflags.h>
36 #include <linux/cpu.h>
37 #include <linux/timex.h>
38 #include <linux/bootmem.h>
40 #include <asm/setup.h>
42 #include <asm/pgalloc.h>
44 #include <asm/s390_ext.h>
45 #include <asm/cpcmd.h>
46 #include <asm/tlbflush.h>
47 #include <asm/timer.h>
48 #include <asm/lowcore.h>
50 #include <asm/cputime.h>
55 /* logical cpu to cpu address */
56 int __cpu_logical_map[NR_CPUS];
58 static struct task_struct *current_set[NR_CPUS];
60 static u8 smp_cpu_type;
61 static int smp_use_sigp_detection;
68 DEFINE_MUTEX(smp_cpu_state_mutex);
69 int smp_cpu_polarization[NR_CPUS];
70 static int smp_cpu_state[NR_CPUS];
71 static int cpu_management;
73 static DEFINE_PER_CPU(struct cpu, cpu_devices);
75 static void smp_ext_bitcall(int, ec_bit_sig);
77 static int cpu_stopped(int cpu)
81 switch (signal_processor_ps(&status, 0, cpu, sigp_sense)) {
82 case sigp_status_stored:
83 /* Check for stopped and check stop state */
93 void smp_send_stop(void)
97 /* Disable all interrupts/machine checks */
98 __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
101 /* stop all processors */
102 for_each_online_cpu(cpu) {
103 if (cpu == smp_processor_id())
106 rc = signal_processor(cpu, sigp_stop);
107 } while (rc == sigp_busy);
109 while (!cpu_stopped(cpu))
115 * This is the main routine where commands issued by other
119 static void do_ext_call_interrupt(__u16 code)
124 * handle bit signal external calls
126 * For the ec_schedule signal we have to do nothing. All the work
127 * is done automatically when we return from the interrupt.
129 bits = xchg(&S390_lowcore.ext_call_fast, 0);
131 if (test_bit(ec_call_function, &bits))
132 generic_smp_call_function_interrupt();
134 if (test_bit(ec_call_function_single, &bits))
135 generic_smp_call_function_single_interrupt();
139 * Send an external call sigp to another cpu and return without waiting
140 * for its completion.
142 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
145 * Set signaling bit in lowcore of target cpu and kick it
147 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
148 while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
152 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
156 for_each_cpu(cpu, mask)
157 smp_ext_bitcall(cpu, ec_call_function);
160 void arch_send_call_function_single_ipi(int cpu)
162 smp_ext_bitcall(cpu, ec_call_function_single);
167 * this function sends a 'purge tlb' signal to another CPU.
169 static void smp_ptlb_callback(void *info)
174 void smp_ptlb_all(void)
176 on_each_cpu(smp_ptlb_callback, NULL, 1);
178 EXPORT_SYMBOL(smp_ptlb_all);
179 #endif /* ! CONFIG_64BIT */
182 * this function sends a 'reschedule' IPI to another CPU.
183 * it goes straight through and wastes no time serializing
184 * anything. Worst case is that we lose a reschedule ...
186 void smp_send_reschedule(int cpu)
188 smp_ext_bitcall(cpu, ec_schedule);
192 * parameter area for the set/clear control bit callbacks
194 struct ec_creg_mask_parms {
195 unsigned long orvals[16];
196 unsigned long andvals[16];
200 * callback for setting/clearing control bits
202 static void smp_ctl_bit_callback(void *info)
204 struct ec_creg_mask_parms *pp = info;
205 unsigned long cregs[16];
208 __ctl_store(cregs, 0, 15);
209 for (i = 0; i <= 15; i++)
210 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
211 __ctl_load(cregs, 0, 15);
215 * Set a bit in a control register of all cpus
217 void smp_ctl_set_bit(int cr, int bit)
219 struct ec_creg_mask_parms parms;
221 memset(&parms.orvals, 0, sizeof(parms.orvals));
222 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
223 parms.orvals[cr] = 1 << bit;
224 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
226 EXPORT_SYMBOL(smp_ctl_set_bit);
229 * Clear a bit in a control register of all cpus
231 void smp_ctl_clear_bit(int cr, int bit)
233 struct ec_creg_mask_parms parms;
235 memset(&parms.orvals, 0, sizeof(parms.orvals));
236 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
237 parms.andvals[cr] = ~(1L << bit);
238 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
240 EXPORT_SYMBOL(smp_ctl_clear_bit);
243 * In early ipl state a temp. logically cpu number is needed, so the sigp
244 * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
245 * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
247 #define CPU_INIT_NO 1
249 #ifdef CONFIG_ZFCPDUMP
252 * zfcpdump_prefix_array holds prefix registers for the following scenario:
253 * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
254 * save its prefix registers, since they get lost, when switching from 31 bit
257 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
258 __attribute__((__section__(".data")));
260 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
262 if (ipl_info.type != IPL_TYPE_FCP_DUMP)
264 if (cpu >= NR_CPUS) {
265 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
266 "the dump\n", cpu, NR_CPUS - 1);
269 zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
270 __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
271 while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
274 memcpy(zfcpdump_save_areas[cpu],
275 (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
278 /* copy original prefix register */
279 zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
283 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
284 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
288 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
290 #endif /* CONFIG_ZFCPDUMP */
292 static int cpu_known(int cpu_id)
296 for_each_present_cpu(cpu) {
297 if (__cpu_logical_map[cpu] == cpu_id)
303 static int smp_rescan_cpus_sigp(cpumask_t avail)
305 int cpu_id, logical_cpu;
307 logical_cpu = cpumask_first(&avail);
308 if (logical_cpu >= nr_cpu_ids)
310 for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
311 if (cpu_known(cpu_id))
313 __cpu_logical_map[logical_cpu] = cpu_id;
314 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
315 if (!cpu_stopped(logical_cpu))
317 cpu_set(logical_cpu, cpu_present_map);
318 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
319 logical_cpu = cpumask_next(logical_cpu, &avail);
320 if (logical_cpu >= nr_cpu_ids)
326 static int smp_rescan_cpus_sclp(cpumask_t avail)
328 struct sclp_cpu_info *info;
329 int cpu_id, logical_cpu, cpu;
332 logical_cpu = cpumask_first(&avail);
333 if (logical_cpu >= nr_cpu_ids)
335 info = kmalloc(sizeof(*info), GFP_KERNEL);
338 rc = sclp_get_cpu_info(info);
341 for (cpu = 0; cpu < info->combined; cpu++) {
342 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
344 cpu_id = info->cpu[cpu].address;
345 if (cpu_known(cpu_id))
347 __cpu_logical_map[logical_cpu] = cpu_id;
348 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
349 cpu_set(logical_cpu, cpu_present_map);
350 if (cpu >= info->configured)
351 smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
353 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
354 logical_cpu = cpumask_next(logical_cpu, &avail);
355 if (logical_cpu >= nr_cpu_ids)
363 static int __smp_rescan_cpus(void)
367 cpus_xor(avail, cpu_possible_map, cpu_present_map);
368 if (smp_use_sigp_detection)
369 return smp_rescan_cpus_sigp(avail);
371 return smp_rescan_cpus_sclp(avail);
374 static void __init smp_detect_cpus(void)
376 unsigned int cpu, c_cpus, s_cpus;
377 struct sclp_cpu_info *info;
378 u16 boot_cpu_addr, cpu_addr;
382 boot_cpu_addr = __cpu_logical_map[0];
383 info = kmalloc(sizeof(*info), GFP_KERNEL);
385 panic("smp_detect_cpus failed to allocate memory\n");
386 /* Use sigp detection algorithm if sclp doesn't work. */
387 if (sclp_get_cpu_info(info)) {
388 smp_use_sigp_detection = 1;
389 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
390 if (cpu == boot_cpu_addr)
392 __cpu_logical_map[CPU_INIT_NO] = cpu;
393 if (!cpu_stopped(CPU_INIT_NO))
395 smp_get_save_area(c_cpus, cpu);
401 if (info->has_cpu_type) {
402 for (cpu = 0; cpu < info->combined; cpu++) {
403 if (info->cpu[cpu].address == boot_cpu_addr) {
404 smp_cpu_type = info->cpu[cpu].type;
410 for (cpu = 0; cpu < info->combined; cpu++) {
411 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
413 cpu_addr = info->cpu[cpu].address;
414 if (cpu_addr == boot_cpu_addr)
416 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
417 if (!cpu_stopped(CPU_INIT_NO)) {
421 smp_get_save_area(c_cpus, cpu_addr);
426 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
433 * Activate a secondary processor.
435 int __cpuinit start_secondary(void *cpuvoid)
440 /* Enable TOD clock interrupts on the secondary cpu. */
442 /* Enable cpu timer interrupts on the secondary cpu. */
444 /* Enable pfault pseudo page faults on this cpu. */
447 /* call cpu notifiers */
448 notify_cpu_starting(smp_processor_id());
449 /* Mark this cpu as online */
451 cpu_set(smp_processor_id(), cpu_online_map);
453 /* Switch on interrupts */
455 /* Print info about this processor */
457 /* cpu_idle will call schedule for us */
462 static void __init smp_create_idle(unsigned int cpu)
464 struct task_struct *p;
467 * don't care about the psw and regs settings since we'll never
468 * reschedule the forked task.
472 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
473 current_set[cpu] = p;
476 static int __cpuinit smp_alloc_lowcore(int cpu)
478 unsigned long async_stack, panic_stack;
479 struct _lowcore *lowcore;
481 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
484 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
485 panic_stack = __get_free_page(GFP_KERNEL);
486 if (!panic_stack || !async_stack)
488 memcpy(lowcore, &S390_lowcore, 512);
489 memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
490 lowcore->async_stack = async_stack + ASYNC_SIZE;
491 lowcore->panic_stack = panic_stack + PAGE_SIZE;
494 if (MACHINE_HAS_IEEE) {
495 unsigned long save_area;
497 save_area = get_zeroed_page(GFP_KERNEL);
500 lowcore->extended_save_area_addr = (u32) save_area;
503 if (vdso_alloc_per_cpu(cpu, lowcore))
506 lowcore_ptr[cpu] = lowcore;
510 free_page(panic_stack);
511 free_pages(async_stack, ASYNC_ORDER);
512 free_pages((unsigned long) lowcore, LC_ORDER);
516 static void smp_free_lowcore(int cpu)
518 struct _lowcore *lowcore;
520 lowcore = lowcore_ptr[cpu];
522 if (MACHINE_HAS_IEEE)
523 free_page((unsigned long) lowcore->extended_save_area_addr);
525 vdso_free_per_cpu(cpu, lowcore);
527 free_page(lowcore->panic_stack - PAGE_SIZE);
528 free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
529 free_pages((unsigned long) lowcore, LC_ORDER);
530 lowcore_ptr[cpu] = NULL;
533 /* Upping and downing of CPUs */
534 int __cpuinit __cpu_up(unsigned int cpu)
536 struct task_struct *idle;
537 struct _lowcore *cpu_lowcore;
538 struct stack_frame *sf;
542 if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
544 if (smp_alloc_lowcore(cpu))
547 ccode = signal_processor(cpu, sigp_initial_cpu_reset);
548 if (ccode == sigp_busy)
550 if (ccode == sigp_not_operational)
552 } while (ccode == sigp_busy);
554 lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
555 while (signal_processor_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
558 idle = current_set[cpu];
559 cpu_lowcore = lowcore_ptr[cpu];
560 cpu_lowcore->kernel_stack = (unsigned long)
561 task_stack_page(idle) + THREAD_SIZE;
562 cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
563 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
564 - sizeof(struct pt_regs)
565 - sizeof(struct stack_frame));
566 memset(sf, 0, sizeof(struct stack_frame));
567 sf->gprs[9] = (unsigned long) sf;
568 cpu_lowcore->save_area[15] = (unsigned long) sf;
569 __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
572 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
573 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
574 cpu_lowcore->current_task = (unsigned long) idle;
575 cpu_lowcore->cpu_nr = cpu;
576 cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
577 cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
578 cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
581 while (signal_processor(cpu, sigp_restart) == sigp_busy)
584 while (!cpu_online(cpu))
589 smp_free_lowcore(cpu);
593 static int __init setup_possible_cpus(char *s)
597 pcpus = simple_strtoul(s, NULL, 0);
598 init_cpu_possible(cpumask_of(0));
599 for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
600 set_cpu_possible(cpu, true);
603 early_param("possible_cpus", setup_possible_cpus);
605 #ifdef CONFIG_HOTPLUG_CPU
607 int __cpu_disable(void)
609 struct ec_creg_mask_parms cr_parms;
610 int cpu = smp_processor_id();
612 cpu_clear(cpu, cpu_online_map);
614 /* Disable pfault pseudo page faults on this cpu. */
617 memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
618 memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
620 /* disable all external interrupts */
621 cr_parms.orvals[0] = 0;
622 cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
623 1 << 11 | 1 << 10 | 1 << 6 | 1 << 4);
624 /* disable all I/O interrupts */
625 cr_parms.orvals[6] = 0;
626 cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
627 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
628 /* disable most machine checks */
629 cr_parms.orvals[14] = 0;
630 cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
633 smp_ctl_bit_callback(&cr_parms);
638 void __cpu_die(unsigned int cpu)
640 /* Wait until target cpu is down */
641 while (!cpu_stopped(cpu))
643 while (signal_processor_p(0, cpu, sigp_set_prefix) == sigp_busy)
645 smp_free_lowcore(cpu);
646 pr_info("Processor %d stopped\n", cpu);
652 while (signal_processor(smp_processor_id(), sigp_stop) == sigp_busy)
657 #endif /* CONFIG_HOTPLUG_CPU */
659 void __init smp_prepare_cpus(unsigned int max_cpus)
662 unsigned long save_area = 0;
664 unsigned long async_stack, panic_stack;
665 struct _lowcore *lowcore;
670 /* request the 0x1201 emergency signal external interrupt */
671 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
672 panic("Couldn't request external interrupt 0x1201");
675 /* Reallocate current lowcore, but keep its contents. */
676 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, LC_ORDER);
677 panic_stack = __get_free_page(GFP_KERNEL);
678 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
679 BUG_ON(!lowcore || !panic_stack || !async_stack);
681 if (MACHINE_HAS_IEEE)
682 save_area = get_zeroed_page(GFP_KERNEL);
685 local_mcck_disable();
686 lowcore_ptr[smp_processor_id()] = lowcore;
687 *lowcore = S390_lowcore;
688 lowcore->panic_stack = panic_stack + PAGE_SIZE;
689 lowcore->async_stack = async_stack + ASYNC_SIZE;
691 if (MACHINE_HAS_IEEE)
692 lowcore->extended_save_area_addr = (u32) save_area;
694 set_prefix((u32)(unsigned long) lowcore);
698 if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
701 for_each_possible_cpu(cpu)
702 if (cpu != smp_processor_id())
703 smp_create_idle(cpu);
706 void __init smp_prepare_boot_cpu(void)
708 BUG_ON(smp_processor_id() != 0);
710 current_thread_info()->cpu = 0;
711 cpu_set(0, cpu_present_map);
712 cpu_set(0, cpu_online_map);
713 S390_lowcore.percpu_offset = __per_cpu_offset[0];
714 current_set[0] = current;
715 smp_cpu_state[0] = CPU_STATE_CONFIGURED;
716 smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
719 void __init smp_cpus_done(unsigned int max_cpus)
723 void __init smp_setup_processor_id(void)
725 S390_lowcore.cpu_nr = 0;
726 __cpu_logical_map[0] = stap();
730 * the frequency of the profiling timer can be changed
731 * by writing a multiplier value into /proc/profile.
733 * usually you want to run this on all CPUs ;)
735 int setup_profiling_timer(unsigned int multiplier)
740 #ifdef CONFIG_HOTPLUG_CPU
741 static ssize_t cpu_configure_show(struct sys_device *dev,
742 struct sysdev_attribute *attr, char *buf)
746 mutex_lock(&smp_cpu_state_mutex);
747 count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
748 mutex_unlock(&smp_cpu_state_mutex);
752 static ssize_t cpu_configure_store(struct sys_device *dev,
753 struct sysdev_attribute *attr,
754 const char *buf, size_t count)
760 if (sscanf(buf, "%d %c", &val, &delim) != 1)
762 if (val != 0 && val != 1)
766 mutex_lock(&smp_cpu_state_mutex);
773 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
774 rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
776 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
777 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
782 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
783 rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
785 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
786 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
794 mutex_unlock(&smp_cpu_state_mutex);
796 return rc ? rc : count;
798 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
799 #endif /* CONFIG_HOTPLUG_CPU */
801 static ssize_t cpu_polarization_show(struct sys_device *dev,
802 struct sysdev_attribute *attr, char *buf)
807 mutex_lock(&smp_cpu_state_mutex);
808 switch (smp_cpu_polarization[cpu]) {
809 case POLARIZATION_HRZ:
810 count = sprintf(buf, "horizontal\n");
812 case POLARIZATION_VL:
813 count = sprintf(buf, "vertical:low\n");
815 case POLARIZATION_VM:
816 count = sprintf(buf, "vertical:medium\n");
818 case POLARIZATION_VH:
819 count = sprintf(buf, "vertical:high\n");
822 count = sprintf(buf, "unknown\n");
825 mutex_unlock(&smp_cpu_state_mutex);
828 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
830 static ssize_t show_cpu_address(struct sys_device *dev,
831 struct sysdev_attribute *attr, char *buf)
833 return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
835 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
838 static struct attribute *cpu_common_attrs[] = {
839 #ifdef CONFIG_HOTPLUG_CPU
840 &attr_configure.attr,
843 &attr_polarization.attr,
847 static struct attribute_group cpu_common_attr_group = {
848 .attrs = cpu_common_attrs,
851 static ssize_t show_capability(struct sys_device *dev,
852 struct sysdev_attribute *attr, char *buf)
854 unsigned int capability;
857 rc = get_cpu_capability(&capability);
860 return sprintf(buf, "%u\n", capability);
862 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
864 static ssize_t show_idle_count(struct sys_device *dev,
865 struct sysdev_attribute *attr, char *buf)
867 struct s390_idle_data *idle;
868 unsigned long long idle_count;
869 unsigned int sequence;
871 idle = &per_cpu(s390_idle, dev->id);
873 sequence = idle->sequence;
877 idle_count = idle->idle_count;
878 if (idle->idle_enter)
881 if (idle->sequence != sequence)
883 return sprintf(buf, "%llu\n", idle_count);
885 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
887 static ssize_t show_idle_time(struct sys_device *dev,
888 struct sysdev_attribute *attr, char *buf)
890 struct s390_idle_data *idle;
891 unsigned long long now, idle_time, idle_enter;
892 unsigned int sequence;
894 idle = &per_cpu(s390_idle, dev->id);
897 sequence = idle->sequence;
901 idle_time = idle->idle_time;
902 idle_enter = idle->idle_enter;
903 if (idle_enter != 0ULL && idle_enter < now)
904 idle_time += now - idle_enter;
906 if (idle->sequence != sequence)
908 return sprintf(buf, "%llu\n", idle_time >> 12);
910 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
912 static struct attribute *cpu_online_attrs[] = {
913 &attr_capability.attr,
914 &attr_idle_count.attr,
915 &attr_idle_time_us.attr,
919 static struct attribute_group cpu_online_attr_group = {
920 .attrs = cpu_online_attrs,
923 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
924 unsigned long action, void *hcpu)
926 unsigned int cpu = (unsigned int)(long)hcpu;
927 struct cpu *c = &per_cpu(cpu_devices, cpu);
928 struct sys_device *s = &c->sysdev;
929 struct s390_idle_data *idle;
933 case CPU_ONLINE_FROZEN:
934 idle = &per_cpu(s390_idle, cpu);
935 memset(idle, 0, sizeof(struct s390_idle_data));
936 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
940 case CPU_DEAD_FROZEN:
941 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
947 static struct notifier_block __cpuinitdata smp_cpu_nb = {
948 .notifier_call = smp_cpu_notify,
951 static int __devinit smp_add_present_cpu(int cpu)
953 struct cpu *c = &per_cpu(cpu_devices, cpu);
954 struct sys_device *s = &c->sysdev;
958 rc = register_cpu(c, cpu);
961 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
964 if (!cpu_online(cpu))
966 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
969 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
971 #ifdef CONFIG_HOTPLUG_CPU
978 #ifdef CONFIG_HOTPLUG_CPU
980 int __ref smp_rescan_cpus(void)
987 mutex_lock(&smp_cpu_state_mutex);
988 newcpus = cpu_present_map;
989 rc = __smp_rescan_cpus();
992 cpus_andnot(newcpus, cpu_present_map, newcpus);
993 for_each_cpu_mask(cpu, newcpus) {
994 rc = smp_add_present_cpu(cpu);
996 cpu_clear(cpu, cpu_present_map);
1000 mutex_unlock(&smp_cpu_state_mutex);
1002 if (!cpus_empty(newcpus))
1003 topology_schedule_update();
1007 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1012 rc = smp_rescan_cpus();
1013 return rc ? rc : count;
1015 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1016 #endif /* CONFIG_HOTPLUG_CPU */
1018 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1022 mutex_lock(&smp_cpu_state_mutex);
1023 count = sprintf(buf, "%d\n", cpu_management);
1024 mutex_unlock(&smp_cpu_state_mutex);
1028 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1034 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1036 if (val != 0 && val != 1)
1040 mutex_lock(&smp_cpu_state_mutex);
1041 if (cpu_management == val)
1043 rc = topology_set_cpu_management(val);
1045 cpu_management = val;
1047 mutex_unlock(&smp_cpu_state_mutex);
1049 return rc ? rc : count;
1051 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1054 static int __init topology_init(void)
1059 register_cpu_notifier(&smp_cpu_nb);
1061 #ifdef CONFIG_HOTPLUG_CPU
1062 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1066 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1069 for_each_present_cpu(cpu) {
1070 rc = smp_add_present_cpu(cpu);
1076 subsys_initcall(topology_init);