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 static struct task_struct *current_set[NR_CPUS];
57 static u8 smp_cpu_type;
58 static int smp_use_sigp_detection;
65 DEFINE_MUTEX(smp_cpu_state_mutex);
66 int smp_cpu_polarization[NR_CPUS];
67 static int smp_cpu_state[NR_CPUS];
68 static int cpu_management;
70 static DEFINE_PER_CPU(struct cpu, cpu_devices);
72 static void smp_ext_bitcall(int, ec_bit_sig);
74 void smp_send_stop(void)
78 /* Disable all interrupts/machine checks */
79 __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
82 /* stop all processors */
83 for_each_online_cpu(cpu) {
84 if (cpu == smp_processor_id())
87 rc = signal_processor(cpu, sigp_stop);
88 } while (rc == sigp_busy);
90 while (!smp_cpu_not_running(cpu))
96 * This is the main routine where commands issued by other
100 static void do_ext_call_interrupt(__u16 code)
105 * handle bit signal external calls
107 * For the ec_schedule signal we have to do nothing. All the work
108 * is done automatically when we return from the interrupt.
110 bits = xchg(&S390_lowcore.ext_call_fast, 0);
112 if (test_bit(ec_call_function, &bits))
113 generic_smp_call_function_interrupt();
115 if (test_bit(ec_call_function_single, &bits))
116 generic_smp_call_function_single_interrupt();
120 * Send an external call sigp to another cpu and return without waiting
121 * for its completion.
123 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
126 * Set signaling bit in lowcore of target cpu and kick it
128 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
129 while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
133 void arch_send_call_function_ipi(cpumask_t mask)
137 for_each_cpu_mask(cpu, mask)
138 smp_ext_bitcall(cpu, ec_call_function);
141 void arch_send_call_function_single_ipi(int cpu)
143 smp_ext_bitcall(cpu, ec_call_function_single);
148 * this function sends a 'purge tlb' signal to another CPU.
150 static void smp_ptlb_callback(void *info)
155 void smp_ptlb_all(void)
157 on_each_cpu(smp_ptlb_callback, NULL, 1);
159 EXPORT_SYMBOL(smp_ptlb_all);
160 #endif /* ! CONFIG_64BIT */
163 * this function sends a 'reschedule' IPI to another CPU.
164 * it goes straight through and wastes no time serializing
165 * anything. Worst case is that we lose a reschedule ...
167 void smp_send_reschedule(int cpu)
169 smp_ext_bitcall(cpu, ec_schedule);
173 * parameter area for the set/clear control bit callbacks
175 struct ec_creg_mask_parms {
176 unsigned long orvals[16];
177 unsigned long andvals[16];
181 * callback for setting/clearing control bits
183 static void smp_ctl_bit_callback(void *info)
185 struct ec_creg_mask_parms *pp = info;
186 unsigned long cregs[16];
189 __ctl_store(cregs, 0, 15);
190 for (i = 0; i <= 15; i++)
191 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
192 __ctl_load(cregs, 0, 15);
196 * Set a bit in a control register of all cpus
198 void smp_ctl_set_bit(int cr, int bit)
200 struct ec_creg_mask_parms parms;
202 memset(&parms.orvals, 0, sizeof(parms.orvals));
203 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
204 parms.orvals[cr] = 1 << bit;
205 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
207 EXPORT_SYMBOL(smp_ctl_set_bit);
210 * Clear a bit in a control register of all cpus
212 void smp_ctl_clear_bit(int cr, int bit)
214 struct ec_creg_mask_parms parms;
216 memset(&parms.orvals, 0, sizeof(parms.orvals));
217 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
218 parms.andvals[cr] = ~(1L << bit);
219 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
221 EXPORT_SYMBOL(smp_ctl_clear_bit);
224 * In early ipl state a temp. logically cpu number is needed, so the sigp
225 * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
226 * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
228 #define CPU_INIT_NO 1
230 #ifdef CONFIG_ZFCPDUMP
233 * zfcpdump_prefix_array holds prefix registers for the following scenario:
234 * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
235 * save its prefix registers, since they get lost, when switching from 31 bit
238 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
239 __attribute__((__section__(".data")));
241 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
243 if (ipl_info.type != IPL_TYPE_FCP_DUMP)
245 if (cpu >= NR_CPUS) {
246 pr_warning("CPU %i exceeds the maximum %i and is excluded from "
247 "the dump\n", cpu, NR_CPUS - 1);
250 zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
251 __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
252 while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
255 memcpy(zfcpdump_save_areas[cpu],
256 (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
259 /* copy original prefix register */
260 zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
264 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
265 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
269 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
271 #endif /* CONFIG_ZFCPDUMP */
273 static int cpu_stopped(int cpu)
277 /* Check for stopped state */
278 if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
279 sigp_status_stored) {
286 static int cpu_known(int cpu_id)
290 for_each_present_cpu(cpu) {
291 if (__cpu_logical_map[cpu] == cpu_id)
297 static int smp_rescan_cpus_sigp(cpumask_t avail)
299 int cpu_id, logical_cpu;
301 logical_cpu = cpumask_first(&avail);
302 if (logical_cpu >= nr_cpu_ids)
304 for (cpu_id = 0; cpu_id <= MAX_CPU_ADDRESS; cpu_id++) {
305 if (cpu_known(cpu_id))
307 __cpu_logical_map[logical_cpu] = cpu_id;
308 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
309 if (!cpu_stopped(logical_cpu))
311 cpu_set(logical_cpu, cpu_present_map);
312 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
313 logical_cpu = cpumask_next(logical_cpu, &avail);
314 if (logical_cpu >= nr_cpu_ids)
320 static int smp_rescan_cpus_sclp(cpumask_t avail)
322 struct sclp_cpu_info *info;
323 int cpu_id, logical_cpu, cpu;
326 logical_cpu = cpumask_first(&avail);
327 if (logical_cpu >= nr_cpu_ids)
329 info = kmalloc(sizeof(*info), GFP_KERNEL);
332 rc = sclp_get_cpu_info(info);
335 for (cpu = 0; cpu < info->combined; cpu++) {
336 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
338 cpu_id = info->cpu[cpu].address;
339 if (cpu_known(cpu_id))
341 __cpu_logical_map[logical_cpu] = cpu_id;
342 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
343 cpu_set(logical_cpu, cpu_present_map);
344 if (cpu >= info->configured)
345 smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
347 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
348 logical_cpu = cpumask_next(logical_cpu, &avail);
349 if (logical_cpu >= nr_cpu_ids)
357 static int __smp_rescan_cpus(void)
361 cpus_xor(avail, cpu_possible_map, cpu_present_map);
362 if (smp_use_sigp_detection)
363 return smp_rescan_cpus_sigp(avail);
365 return smp_rescan_cpus_sclp(avail);
368 static void __init smp_detect_cpus(void)
370 unsigned int cpu, c_cpus, s_cpus;
371 struct sclp_cpu_info *info;
372 u16 boot_cpu_addr, cpu_addr;
376 boot_cpu_addr = __cpu_logical_map[0];
377 info = kmalloc(sizeof(*info), GFP_KERNEL);
379 panic("smp_detect_cpus failed to allocate memory\n");
380 /* Use sigp detection algorithm if sclp doesn't work. */
381 if (sclp_get_cpu_info(info)) {
382 smp_use_sigp_detection = 1;
383 for (cpu = 0; cpu <= MAX_CPU_ADDRESS; cpu++) {
384 if (cpu == boot_cpu_addr)
386 __cpu_logical_map[CPU_INIT_NO] = cpu;
387 if (!cpu_stopped(CPU_INIT_NO))
389 smp_get_save_area(c_cpus, cpu);
395 if (info->has_cpu_type) {
396 for (cpu = 0; cpu < info->combined; cpu++) {
397 if (info->cpu[cpu].address == boot_cpu_addr) {
398 smp_cpu_type = info->cpu[cpu].type;
404 for (cpu = 0; cpu < info->combined; cpu++) {
405 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
407 cpu_addr = info->cpu[cpu].address;
408 if (cpu_addr == boot_cpu_addr)
410 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
411 if (!cpu_stopped(CPU_INIT_NO)) {
415 smp_get_save_area(c_cpus, cpu_addr);
420 pr_info("%d configured CPUs, %d standby CPUs\n", c_cpus, s_cpus);
427 * Activate a secondary processor.
429 int __cpuinit start_secondary(void *cpuvoid)
434 /* Enable TOD clock interrupts on the secondary cpu. */
436 /* Enable cpu timer interrupts on the secondary cpu. */
438 /* Enable pfault pseudo page faults on this cpu. */
441 /* call cpu notifiers */
442 notify_cpu_starting(smp_processor_id());
443 /* Mark this cpu as online */
445 cpu_set(smp_processor_id(), cpu_online_map);
447 /* Switch on interrupts */
449 /* Print info about this processor */
451 /* cpu_idle will call schedule for us */
456 static void __init smp_create_idle(unsigned int cpu)
458 struct task_struct *p;
461 * don't care about the psw and regs settings since we'll never
462 * reschedule the forked task.
466 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
467 current_set[cpu] = p;
470 static int __cpuinit smp_alloc_lowcore(int cpu)
472 unsigned long async_stack, panic_stack;
473 struct _lowcore *lowcore;
476 lc_order = sizeof(long) == 8 ? 1 : 0;
477 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
480 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
481 panic_stack = __get_free_page(GFP_KERNEL);
482 if (!panic_stack || !async_stack)
484 memcpy(lowcore, &S390_lowcore, 512);
485 memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
486 lowcore->async_stack = async_stack + ASYNC_SIZE;
487 lowcore->panic_stack = panic_stack + PAGE_SIZE;
490 if (MACHINE_HAS_IEEE) {
491 unsigned long save_area;
493 save_area = get_zeroed_page(GFP_KERNEL);
496 lowcore->extended_save_area_addr = (u32) save_area;
499 if (vdso_alloc_per_cpu(cpu, lowcore))
502 lowcore_ptr[cpu] = lowcore;
506 free_page(panic_stack);
507 free_pages(async_stack, ASYNC_ORDER);
508 free_pages((unsigned long) lowcore, lc_order);
512 static void smp_free_lowcore(int cpu)
514 struct _lowcore *lowcore;
517 lc_order = sizeof(long) == 8 ? 1 : 0;
518 lowcore = lowcore_ptr[cpu];
520 if (MACHINE_HAS_IEEE)
521 free_page((unsigned long) lowcore->extended_save_area_addr);
523 vdso_free_per_cpu(cpu, lowcore);
525 free_page(lowcore->panic_stack - PAGE_SIZE);
526 free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
527 free_pages((unsigned long) lowcore, lc_order);
528 lowcore_ptr[cpu] = NULL;
531 /* Upping and downing of CPUs */
532 int __cpuinit __cpu_up(unsigned int cpu)
534 struct task_struct *idle;
535 struct _lowcore *cpu_lowcore;
536 struct stack_frame *sf;
540 if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
542 if (smp_alloc_lowcore(cpu))
545 ccode = signal_processor(cpu, sigp_initial_cpu_reset);
546 if (ccode == sigp_busy)
548 if (ccode == sigp_not_operational)
550 } while (ccode == sigp_busy);
552 lowcore = (u32)(unsigned long)lowcore_ptr[cpu];
553 while (signal_processor_p(lowcore, cpu, sigp_set_prefix) == sigp_busy)
556 idle = current_set[cpu];
557 cpu_lowcore = lowcore_ptr[cpu];
558 cpu_lowcore->kernel_stack = (unsigned long)
559 task_stack_page(idle) + THREAD_SIZE;
560 cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
561 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
562 - sizeof(struct pt_regs)
563 - sizeof(struct stack_frame));
564 memset(sf, 0, sizeof(struct stack_frame));
565 sf->gprs[9] = (unsigned long) sf;
566 cpu_lowcore->save_area[15] = (unsigned long) sf;
567 __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
570 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
571 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
572 cpu_lowcore->current_task = (unsigned long) idle;
573 cpu_lowcore->cpu_nr = cpu;
574 cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
575 cpu_lowcore->machine_flags = S390_lowcore.machine_flags;
576 cpu_lowcore->ftrace_func = S390_lowcore.ftrace_func;
579 while (signal_processor(cpu, sigp_restart) == sigp_busy)
582 while (!cpu_online(cpu))
587 smp_free_lowcore(cpu);
591 static int __init setup_possible_cpus(char *s)
595 pcpus = simple_strtoul(s, NULL, 0);
596 init_cpu_possible(cpumask_of(0));
597 for (cpu = 1; cpu < pcpus && cpu < nr_cpu_ids; cpu++)
598 set_cpu_possible(cpu, true);
601 early_param("possible_cpus", setup_possible_cpus);
603 #ifdef CONFIG_HOTPLUG_CPU
605 int __cpu_disable(void)
607 struct ec_creg_mask_parms cr_parms;
608 int cpu = smp_processor_id();
610 cpu_clear(cpu, cpu_online_map);
612 /* Disable pfault pseudo page faults on this cpu. */
615 memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
616 memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
618 /* disable all external interrupts */
619 cr_parms.orvals[0] = 0;
620 cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
621 1 << 11 | 1 << 10 | 1 << 6 | 1 << 4);
622 /* disable all I/O interrupts */
623 cr_parms.orvals[6] = 0;
624 cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
625 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
626 /* disable most machine checks */
627 cr_parms.orvals[14] = 0;
628 cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
631 smp_ctl_bit_callback(&cr_parms);
636 void __cpu_die(unsigned int cpu)
638 /* Wait until target cpu is down */
639 while (!smp_cpu_not_running(cpu))
641 smp_free_lowcore(cpu);
642 pr_info("Processor %d stopped\n", cpu);
648 signal_processor(smp_processor_id(), sigp_stop);
653 #endif /* CONFIG_HOTPLUG_CPU */
655 void __init smp_prepare_cpus(unsigned int max_cpus)
658 unsigned long save_area = 0;
660 unsigned long async_stack, panic_stack;
661 struct _lowcore *lowcore;
667 /* request the 0x1201 emergency signal external interrupt */
668 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
669 panic("Couldn't request external interrupt 0x1201");
672 /* Reallocate current lowcore, but keep its contents. */
673 lc_order = sizeof(long) == 8 ? 1 : 0;
674 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
675 panic_stack = __get_free_page(GFP_KERNEL);
676 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
677 BUG_ON(!lowcore || !panic_stack || !async_stack);
679 if (MACHINE_HAS_IEEE)
680 save_area = get_zeroed_page(GFP_KERNEL);
683 local_mcck_disable();
684 lowcore_ptr[smp_processor_id()] = lowcore;
685 *lowcore = S390_lowcore;
686 lowcore->panic_stack = panic_stack + PAGE_SIZE;
687 lowcore->async_stack = async_stack + ASYNC_SIZE;
689 if (MACHINE_HAS_IEEE)
690 lowcore->extended_save_area_addr = (u32) save_area;
692 set_prefix((u32)(unsigned long) lowcore);
696 if (vdso_alloc_per_cpu(smp_processor_id(), &S390_lowcore))
699 for_each_possible_cpu(cpu)
700 if (cpu != smp_processor_id())
701 smp_create_idle(cpu);
704 void __init smp_prepare_boot_cpu(void)
706 BUG_ON(smp_processor_id() != 0);
708 current_thread_info()->cpu = 0;
709 cpu_set(0, cpu_present_map);
710 cpu_set(0, cpu_online_map);
711 S390_lowcore.percpu_offset = __per_cpu_offset[0];
712 current_set[0] = current;
713 smp_cpu_state[0] = CPU_STATE_CONFIGURED;
714 smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
717 void __init smp_cpus_done(unsigned int max_cpus)
722 * the frequency of the profiling timer can be changed
723 * by writing a multiplier value into /proc/profile.
725 * usually you want to run this on all CPUs ;)
727 int setup_profiling_timer(unsigned int multiplier)
732 #ifdef CONFIG_HOTPLUG_CPU
733 static ssize_t cpu_configure_show(struct sys_device *dev,
734 struct sysdev_attribute *attr, char *buf)
738 mutex_lock(&smp_cpu_state_mutex);
739 count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
740 mutex_unlock(&smp_cpu_state_mutex);
744 static ssize_t cpu_configure_store(struct sys_device *dev,
745 struct sysdev_attribute *attr,
746 const char *buf, size_t count)
752 if (sscanf(buf, "%d %c", &val, &delim) != 1)
754 if (val != 0 && val != 1)
758 mutex_lock(&smp_cpu_state_mutex);
765 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
766 rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
768 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
769 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
774 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
775 rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
777 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
778 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
786 mutex_unlock(&smp_cpu_state_mutex);
788 return rc ? rc : count;
790 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
791 #endif /* CONFIG_HOTPLUG_CPU */
793 static ssize_t cpu_polarization_show(struct sys_device *dev,
794 struct sysdev_attribute *attr, char *buf)
799 mutex_lock(&smp_cpu_state_mutex);
800 switch (smp_cpu_polarization[cpu]) {
801 case POLARIZATION_HRZ:
802 count = sprintf(buf, "horizontal\n");
804 case POLARIZATION_VL:
805 count = sprintf(buf, "vertical:low\n");
807 case POLARIZATION_VM:
808 count = sprintf(buf, "vertical:medium\n");
810 case POLARIZATION_VH:
811 count = sprintf(buf, "vertical:high\n");
814 count = sprintf(buf, "unknown\n");
817 mutex_unlock(&smp_cpu_state_mutex);
820 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
822 static ssize_t show_cpu_address(struct sys_device *dev,
823 struct sysdev_attribute *attr, char *buf)
825 return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
827 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
830 static struct attribute *cpu_common_attrs[] = {
831 #ifdef CONFIG_HOTPLUG_CPU
832 &attr_configure.attr,
835 &attr_polarization.attr,
839 static struct attribute_group cpu_common_attr_group = {
840 .attrs = cpu_common_attrs,
843 static ssize_t show_capability(struct sys_device *dev,
844 struct sysdev_attribute *attr, char *buf)
846 unsigned int capability;
849 rc = get_cpu_capability(&capability);
852 return sprintf(buf, "%u\n", capability);
854 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
856 static ssize_t show_idle_count(struct sys_device *dev,
857 struct sysdev_attribute *attr, char *buf)
859 struct s390_idle_data *idle;
860 unsigned long long idle_count;
861 unsigned int sequence;
863 idle = &per_cpu(s390_idle, dev->id);
865 sequence = idle->sequence;
869 idle_count = idle->idle_count;
870 if (idle->idle_enter)
873 if (idle->sequence != sequence)
875 return sprintf(buf, "%llu\n", idle_count);
877 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
879 static ssize_t show_idle_time(struct sys_device *dev,
880 struct sysdev_attribute *attr, char *buf)
882 struct s390_idle_data *idle;
883 unsigned long long now, idle_time, idle_enter;
884 unsigned int sequence;
886 idle = &per_cpu(s390_idle, dev->id);
889 sequence = idle->sequence;
893 idle_time = idle->idle_time;
894 idle_enter = idle->idle_enter;
895 if (idle_enter != 0ULL && idle_enter < now)
896 idle_time += now - idle_enter;
898 if (idle->sequence != sequence)
900 return sprintf(buf, "%llu\n", idle_time >> 12);
902 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
904 static struct attribute *cpu_online_attrs[] = {
905 &attr_capability.attr,
906 &attr_idle_count.attr,
907 &attr_idle_time_us.attr,
911 static struct attribute_group cpu_online_attr_group = {
912 .attrs = cpu_online_attrs,
915 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
916 unsigned long action, void *hcpu)
918 unsigned int cpu = (unsigned int)(long)hcpu;
919 struct cpu *c = &per_cpu(cpu_devices, cpu);
920 struct sys_device *s = &c->sysdev;
921 struct s390_idle_data *idle;
925 case CPU_ONLINE_FROZEN:
926 idle = &per_cpu(s390_idle, cpu);
927 memset(idle, 0, sizeof(struct s390_idle_data));
928 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
932 case CPU_DEAD_FROZEN:
933 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
939 static struct notifier_block __cpuinitdata smp_cpu_nb = {
940 .notifier_call = smp_cpu_notify,
943 static int __devinit smp_add_present_cpu(int cpu)
945 struct cpu *c = &per_cpu(cpu_devices, cpu);
946 struct sys_device *s = &c->sysdev;
950 rc = register_cpu(c, cpu);
953 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
956 if (!cpu_online(cpu))
958 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
961 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
963 #ifdef CONFIG_HOTPLUG_CPU
970 #ifdef CONFIG_HOTPLUG_CPU
972 int __ref smp_rescan_cpus(void)
979 mutex_lock(&smp_cpu_state_mutex);
980 newcpus = cpu_present_map;
981 rc = __smp_rescan_cpus();
984 cpus_andnot(newcpus, cpu_present_map, newcpus);
985 for_each_cpu_mask(cpu, newcpus) {
986 rc = smp_add_present_cpu(cpu);
988 cpu_clear(cpu, cpu_present_map);
992 mutex_unlock(&smp_cpu_state_mutex);
994 if (!cpus_empty(newcpus))
995 topology_schedule_update();
999 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1004 rc = smp_rescan_cpus();
1005 return rc ? rc : count;
1007 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1008 #endif /* CONFIG_HOTPLUG_CPU */
1010 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1014 mutex_lock(&smp_cpu_state_mutex);
1015 count = sprintf(buf, "%d\n", cpu_management);
1016 mutex_unlock(&smp_cpu_state_mutex);
1020 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1026 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1028 if (val != 0 && val != 1)
1032 mutex_lock(&smp_cpu_state_mutex);
1033 if (cpu_management == val)
1035 rc = topology_set_cpu_management(val);
1037 cpu_management = val;
1039 mutex_unlock(&smp_cpu_state_mutex);
1041 return rc ? rc : count;
1043 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1047 * If the resume kernel runs on another cpu than the suspended kernel,
1048 * we have to switch the cpu IDs in the logical map.
1050 void smp_switch_boot_cpu_in_resume(u32 resume_phys_cpu_id,
1051 struct _lowcore *suspend_lowcore)
1053 int cpu, suspend_cpu_id, resume_cpu_id;
1054 u32 suspend_phys_cpu_id;
1056 suspend_phys_cpu_id = __cpu_logical_map[suspend_lowcore->cpu_nr];
1057 suspend_cpu_id = suspend_lowcore->cpu_nr;
1059 for_each_present_cpu(cpu) {
1060 if (__cpu_logical_map[cpu] == resume_phys_cpu_id) {
1061 resume_cpu_id = cpu;
1065 panic("Could not find resume cpu in logical map.\n");
1068 printk("Resume cpu ID: %i/%i\n", resume_phys_cpu_id, resume_cpu_id);
1069 printk("Suspend cpu ID: %i/%i\n", suspend_phys_cpu_id, suspend_cpu_id);
1071 __cpu_logical_map[resume_cpu_id] = suspend_phys_cpu_id;
1072 __cpu_logical_map[suspend_cpu_id] = resume_phys_cpu_id;
1074 lowcore_ptr[suspend_cpu_id]->cpu_addr = resume_phys_cpu_id;
1077 u32 smp_get_phys_cpu_id(void)
1079 return __cpu_logical_map[smp_processor_id()];
1082 static int __init topology_init(void)
1087 register_cpu_notifier(&smp_cpu_nb);
1089 #ifdef CONFIG_HOTPLUG_CPU
1090 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1094 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1097 for_each_present_cpu(cpu) {
1098 rc = smp_add_present_cpu(cpu);
1104 subsys_initcall(topology_init);