2 * arch/s390/kernel/smp.c
4 * Copyright IBM Corp. 1999,2007
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 #include <linux/module.h>
24 #include <linux/init.h>
26 #include <linux/err.h>
27 #include <linux/spinlock.h>
28 #include <linux/kernel_stat.h>
29 #include <linux/delay.h>
30 #include <linux/cache.h>
31 #include <linux/interrupt.h>
32 #include <linux/cpu.h>
33 #include <linux/timex.h>
34 #include <linux/bootmem.h>
36 #include <asm/setup.h>
38 #include <asm/pgalloc.h>
40 #include <asm/s390_ext.h>
41 #include <asm/cpcmd.h>
42 #include <asm/tlbflush.h>
43 #include <asm/timer.h>
44 #include <asm/lowcore.h>
50 * An array with a pointer the lowcore of every CPU.
52 struct _lowcore *lowcore_ptr[NR_CPUS];
53 EXPORT_SYMBOL(lowcore_ptr);
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);
75 * Structure and data for __smp_call_function_map(). This is designed to
76 * minimise static memory requirements. It also looks cleaner.
78 static DEFINE_SPINLOCK(call_lock);
80 struct call_data_struct {
81 void (*func) (void *info);
88 static struct call_data_struct *call_data;
91 * 'Call function' interrupt callback
93 static void do_call_function(void)
95 void (*func) (void *info) = call_data->func;
96 void *info = call_data->info;
97 int wait = call_data->wait;
99 cpu_set(smp_processor_id(), call_data->started);
102 cpu_set(smp_processor_id(), call_data->finished);;
105 static void __smp_call_function_map(void (*func) (void *info), void *info,
106 int wait, cpumask_t map)
108 struct call_data_struct data;
112 * Can deadlock when interrupts are disabled or if in wrong context.
114 WARN_ON(irqs_disabled() || in_irq());
117 * Check for local function call. We have to have the same call order
118 * as in on_each_cpu() because of machine_restart_smp().
120 if (cpu_isset(smp_processor_id(), map)) {
122 cpu_clear(smp_processor_id(), map);
125 cpus_and(map, map, cpu_online_map);
131 data.started = CPU_MASK_NONE;
134 data.finished = CPU_MASK_NONE;
138 for_each_cpu_mask(cpu, map)
139 smp_ext_bitcall(cpu, ec_call_function);
141 /* Wait for response */
142 while (!cpus_equal(map, data.started))
145 while (!cpus_equal(map, data.finished))
157 * @func: the function to run; this must be fast and non-blocking
158 * @info: an arbitrary pointer to pass to the function
159 * @wait: if true, wait (atomically) until function has completed on other CPUs
161 * Run a function on all other CPUs.
163 * You must not call this function with disabled interrupts, from a
164 * hardware interrupt handler or from a bottom half.
166 int smp_call_function(void (*func) (void *info), void *info, int wait)
170 spin_lock(&call_lock);
171 map = cpu_online_map;
172 cpu_clear(smp_processor_id(), map);
173 __smp_call_function_map(func, info, wait, map);
174 spin_unlock(&call_lock);
177 EXPORT_SYMBOL(smp_call_function);
180 * smp_call_function_single:
181 * @cpu: the CPU where func should run
182 * @func: the function to run; this must be fast and non-blocking
183 * @info: an arbitrary pointer to pass to the function
184 * @wait: if true, wait (atomically) until function has completed on other CPUs
186 * Run a function on one processor.
188 * You must not call this function with disabled interrupts, from a
189 * hardware interrupt handler or from a bottom half.
191 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
194 spin_lock(&call_lock);
195 __smp_call_function_map(func, info, wait, cpumask_of_cpu(cpu));
196 spin_unlock(&call_lock);
199 EXPORT_SYMBOL(smp_call_function_single);
202 * smp_call_function_mask(): Run a function on a set of other CPUs.
203 * @mask: The set of cpus to run on. Must not include the current cpu.
204 * @func: The function to run. This must be fast and non-blocking.
205 * @info: An arbitrary pointer to pass to the function.
206 * @wait: If true, wait (atomically) until function has completed on other CPUs.
208 * Returns 0 on success, else a negative status code.
210 * If @wait is true, then returns once @func has returned; otherwise
211 * it returns just before the target cpu calls @func.
213 * You must not call this function with disabled interrupts or from a
214 * hardware interrupt handler or from a bottom half handler.
216 int smp_call_function_mask(cpumask_t mask, void (*func)(void *), void *info,
219 spin_lock(&call_lock);
220 cpu_clear(smp_processor_id(), mask);
221 __smp_call_function_map(func, info, wait, mask);
222 spin_unlock(&call_lock);
225 EXPORT_SYMBOL(smp_call_function_mask);
227 void smp_send_stop(void)
231 /* Disable all interrupts/machine checks */
232 __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
234 /* write magic number to zero page (absolute 0) */
235 lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
237 /* stop all processors */
238 for_each_online_cpu(cpu) {
239 if (cpu == smp_processor_id())
242 rc = signal_processor(cpu, sigp_stop);
243 } while (rc == sigp_busy);
245 while (!smp_cpu_not_running(cpu))
251 * This is the main routine where commands issued by other
255 static void do_ext_call_interrupt(__u16 code)
260 * handle bit signal external calls
262 * For the ec_schedule signal we have to do nothing. All the work
263 * is done automatically when we return from the interrupt.
265 bits = xchg(&S390_lowcore.ext_call_fast, 0);
267 if (test_bit(ec_call_function, &bits))
272 * Send an external call sigp to another cpu and return without waiting
273 * for its completion.
275 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
278 * Set signaling bit in lowcore of target cpu and kick it
280 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
281 while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
287 * this function sends a 'purge tlb' signal to another CPU.
289 static void smp_ptlb_callback(void *info)
294 void smp_ptlb_all(void)
296 on_each_cpu(smp_ptlb_callback, NULL, 1);
298 EXPORT_SYMBOL(smp_ptlb_all);
299 #endif /* ! CONFIG_64BIT */
302 * this function sends a 'reschedule' IPI to another CPU.
303 * it goes straight through and wastes no time serializing
304 * anything. Worst case is that we lose a reschedule ...
306 void smp_send_reschedule(int cpu)
308 smp_ext_bitcall(cpu, ec_schedule);
312 * parameter area for the set/clear control bit callbacks
314 struct ec_creg_mask_parms {
315 unsigned long orvals[16];
316 unsigned long andvals[16];
320 * callback for setting/clearing control bits
322 static void smp_ctl_bit_callback(void *info)
324 struct ec_creg_mask_parms *pp = info;
325 unsigned long cregs[16];
328 __ctl_store(cregs, 0, 15);
329 for (i = 0; i <= 15; i++)
330 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
331 __ctl_load(cregs, 0, 15);
335 * Set a bit in a control register of all cpus
337 void smp_ctl_set_bit(int cr, int bit)
339 struct ec_creg_mask_parms parms;
341 memset(&parms.orvals, 0, sizeof(parms.orvals));
342 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
343 parms.orvals[cr] = 1 << bit;
344 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
346 EXPORT_SYMBOL(smp_ctl_set_bit);
349 * Clear a bit in a control register of all cpus
351 void smp_ctl_clear_bit(int cr, int bit)
353 struct ec_creg_mask_parms parms;
355 memset(&parms.orvals, 0, sizeof(parms.orvals));
356 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
357 parms.andvals[cr] = ~(1L << bit);
358 on_each_cpu(smp_ctl_bit_callback, &parms, 1);
360 EXPORT_SYMBOL(smp_ctl_clear_bit);
363 * In early ipl state a temp. logically cpu number is needed, so the sigp
364 * functions can be used to sense other cpus. Since NR_CPUS is >= 2 on
365 * CONFIG_SMP and the ipl cpu is logical cpu 0, it must be 1.
367 #define CPU_INIT_NO 1
369 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
372 * zfcpdump_prefix_array holds prefix registers for the following scenario:
373 * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
374 * save its prefix registers, since they get lost, when switching from 31 bit
377 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
378 __attribute__((__section__(".data")));
380 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
382 if (ipl_info.type != IPL_TYPE_FCP_DUMP)
384 if (cpu >= NR_CPUS) {
385 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
386 "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
389 zfcpdump_save_areas[cpu] = kmalloc(sizeof(union save_area), GFP_KERNEL);
390 __cpu_logical_map[CPU_INIT_NO] = (__u16) phy_cpu;
391 while (signal_processor(CPU_INIT_NO, sigp_stop_and_store_status) ==
394 memcpy(zfcpdump_save_areas[cpu],
395 (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
398 /* copy original prefix register */
399 zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
403 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
404 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
408 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
410 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
412 static int cpu_stopped(int cpu)
416 /* Check for stopped state */
417 if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
418 sigp_status_stored) {
425 static int cpu_known(int cpu_id)
429 for_each_present_cpu(cpu) {
430 if (__cpu_logical_map[cpu] == cpu_id)
436 static int smp_rescan_cpus_sigp(cpumask_t avail)
438 int cpu_id, logical_cpu;
440 logical_cpu = first_cpu(avail);
441 if (logical_cpu == NR_CPUS)
443 for (cpu_id = 0; cpu_id <= 65535; cpu_id++) {
444 if (cpu_known(cpu_id))
446 __cpu_logical_map[logical_cpu] = cpu_id;
447 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
448 if (!cpu_stopped(logical_cpu))
450 cpu_set(logical_cpu, cpu_present_map);
451 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
452 logical_cpu = next_cpu(logical_cpu, avail);
453 if (logical_cpu == NR_CPUS)
459 static int smp_rescan_cpus_sclp(cpumask_t avail)
461 struct sclp_cpu_info *info;
462 int cpu_id, logical_cpu, cpu;
465 logical_cpu = first_cpu(avail);
466 if (logical_cpu == NR_CPUS)
468 info = kmalloc(sizeof(*info), GFP_KERNEL);
471 rc = sclp_get_cpu_info(info);
474 for (cpu = 0; cpu < info->combined; cpu++) {
475 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
477 cpu_id = info->cpu[cpu].address;
478 if (cpu_known(cpu_id))
480 __cpu_logical_map[logical_cpu] = cpu_id;
481 smp_cpu_polarization[logical_cpu] = POLARIZATION_UNKNWN;
482 cpu_set(logical_cpu, cpu_present_map);
483 if (cpu >= info->configured)
484 smp_cpu_state[logical_cpu] = CPU_STATE_STANDBY;
486 smp_cpu_state[logical_cpu] = CPU_STATE_CONFIGURED;
487 logical_cpu = next_cpu(logical_cpu, avail);
488 if (logical_cpu == NR_CPUS)
496 static int __smp_rescan_cpus(void)
500 cpus_xor(avail, cpu_possible_map, cpu_present_map);
501 if (smp_use_sigp_detection)
502 return smp_rescan_cpus_sigp(avail);
504 return smp_rescan_cpus_sclp(avail);
507 static void __init smp_detect_cpus(void)
509 unsigned int cpu, c_cpus, s_cpus;
510 struct sclp_cpu_info *info;
511 u16 boot_cpu_addr, cpu_addr;
515 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
516 info = kmalloc(sizeof(*info), GFP_KERNEL);
518 panic("smp_detect_cpus failed to allocate memory\n");
519 /* Use sigp detection algorithm if sclp doesn't work. */
520 if (sclp_get_cpu_info(info)) {
521 smp_use_sigp_detection = 1;
522 for (cpu = 0; cpu <= 65535; cpu++) {
523 if (cpu == boot_cpu_addr)
525 __cpu_logical_map[CPU_INIT_NO] = cpu;
526 if (!cpu_stopped(CPU_INIT_NO))
528 smp_get_save_area(c_cpus, cpu);
534 if (info->has_cpu_type) {
535 for (cpu = 0; cpu < info->combined; cpu++) {
536 if (info->cpu[cpu].address == boot_cpu_addr) {
537 smp_cpu_type = info->cpu[cpu].type;
543 for (cpu = 0; cpu < info->combined; cpu++) {
544 if (info->has_cpu_type && info->cpu[cpu].type != smp_cpu_type)
546 cpu_addr = info->cpu[cpu].address;
547 if (cpu_addr == boot_cpu_addr)
549 __cpu_logical_map[CPU_INIT_NO] = cpu_addr;
550 if (!cpu_stopped(CPU_INIT_NO)) {
554 smp_get_save_area(c_cpus, cpu_addr);
559 printk(KERN_INFO "CPUs: %d configured, %d standby\n", c_cpus, s_cpus);
566 * Activate a secondary processor.
568 int __cpuinit start_secondary(void *cpuvoid)
573 /* Enable TOD clock interrupts on the secondary cpu. */
575 #ifdef CONFIG_VIRT_TIMER
576 /* Enable cpu timer interrupts on the secondary cpu. */
579 /* Enable pfault pseudo page faults on this cpu. */
582 /* call cpu notifiers */
583 notify_cpu_starting(smp_processor_id());
584 /* Mark this cpu as online */
585 spin_lock(&call_lock);
586 cpu_set(smp_processor_id(), cpu_online_map);
587 spin_unlock(&call_lock);
588 /* Switch on interrupts */
590 /* Print info about this processor */
591 print_cpu_info(&S390_lowcore.cpu_data);
592 /* cpu_idle will call schedule for us */
597 static void __init smp_create_idle(unsigned int cpu)
599 struct task_struct *p;
602 * don't care about the psw and regs settings since we'll never
603 * reschedule the forked task.
607 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
608 current_set[cpu] = p;
611 static int __cpuinit smp_alloc_lowcore(int cpu)
613 unsigned long async_stack, panic_stack;
614 struct _lowcore *lowcore;
617 lc_order = sizeof(long) == 8 ? 1 : 0;
618 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
621 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
622 panic_stack = __get_free_page(GFP_KERNEL);
623 if (!panic_stack || !async_stack)
625 memcpy(lowcore, &S390_lowcore, 512);
626 memset((char *)lowcore + 512, 0, sizeof(*lowcore) - 512);
627 lowcore->async_stack = async_stack + ASYNC_SIZE;
628 lowcore->panic_stack = panic_stack + PAGE_SIZE;
631 if (MACHINE_HAS_IEEE) {
632 unsigned long save_area;
634 save_area = get_zeroed_page(GFP_KERNEL);
637 lowcore->extended_save_area_addr = (u32) save_area;
640 lowcore_ptr[cpu] = lowcore;
645 free_page(panic_stack);
648 free_pages(async_stack, ASYNC_ORDER);
649 free_pages((unsigned long) lowcore, lc_order);
653 #ifdef CONFIG_HOTPLUG_CPU
654 static void smp_free_lowcore(int cpu)
656 struct _lowcore *lowcore;
659 lc_order = sizeof(long) == 8 ? 1 : 0;
660 lowcore = lowcore_ptr[cpu];
662 if (MACHINE_HAS_IEEE)
663 free_page((unsigned long) lowcore->extended_save_area_addr);
665 free_page(lowcore->panic_stack - PAGE_SIZE);
666 free_pages(lowcore->async_stack - ASYNC_SIZE, ASYNC_ORDER);
667 free_pages((unsigned long) lowcore, lc_order);
668 lowcore_ptr[cpu] = NULL;
670 #endif /* CONFIG_HOTPLUG_CPU */
672 /* Upping and downing of CPUs */
673 int __cpuinit __cpu_up(unsigned int cpu)
675 struct task_struct *idle;
676 struct _lowcore *cpu_lowcore;
677 struct stack_frame *sf;
680 if (smp_cpu_state[cpu] != CPU_STATE_CONFIGURED)
682 if (smp_alloc_lowcore(cpu))
685 ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
686 cpu, sigp_set_prefix);
688 printk("sigp_set_prefix failed for cpu %d "
689 "with condition code %d\n",
690 (int) cpu, (int) ccode);
694 idle = current_set[cpu];
695 cpu_lowcore = lowcore_ptr[cpu];
696 cpu_lowcore->kernel_stack = (unsigned long)
697 task_stack_page(idle) + THREAD_SIZE;
698 cpu_lowcore->thread_info = (unsigned long) task_thread_info(idle);
699 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
700 - sizeof(struct pt_regs)
701 - sizeof(struct stack_frame));
702 memset(sf, 0, sizeof(struct stack_frame));
703 sf->gprs[9] = (unsigned long) sf;
704 cpu_lowcore->save_area[15] = (unsigned long) sf;
705 __ctl_store(cpu_lowcore->cregs_save_area, 0, 15);
708 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
709 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
710 cpu_lowcore->current_task = (unsigned long) idle;
711 cpu_lowcore->cpu_data.cpu_nr = cpu;
712 cpu_lowcore->kernel_asce = S390_lowcore.kernel_asce;
713 cpu_lowcore->ipl_device = S390_lowcore.ipl_device;
716 while (signal_processor(cpu, sigp_restart) == sigp_busy)
719 while (!cpu_online(cpu))
724 static int __init setup_possible_cpus(char *s)
728 pcpus = simple_strtoul(s, NULL, 0);
729 cpu_possible_map = cpumask_of_cpu(0);
730 for (cpu = 1; cpu < pcpus && cpu < NR_CPUS; cpu++)
731 cpu_set(cpu, cpu_possible_map);
734 early_param("possible_cpus", setup_possible_cpus);
736 #ifdef CONFIG_HOTPLUG_CPU
738 int __cpu_disable(void)
740 struct ec_creg_mask_parms cr_parms;
741 int cpu = smp_processor_id();
743 cpu_clear(cpu, cpu_online_map);
745 /* Disable pfault pseudo page faults on this cpu. */
748 memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
749 memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
751 /* disable all external interrupts */
752 cr_parms.orvals[0] = 0;
753 cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
754 1 << 11 | 1 << 10 | 1 << 6 | 1 << 4);
755 /* disable all I/O interrupts */
756 cr_parms.orvals[6] = 0;
757 cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
758 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
759 /* disable most machine checks */
760 cr_parms.orvals[14] = 0;
761 cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
764 smp_ctl_bit_callback(&cr_parms);
769 void __cpu_die(unsigned int cpu)
771 /* Wait until target cpu is down */
772 while (!smp_cpu_not_running(cpu))
774 smp_free_lowcore(cpu);
775 printk(KERN_INFO "Processor %d spun down\n", cpu);
781 signal_processor(smp_processor_id(), sigp_stop);
786 #endif /* CONFIG_HOTPLUG_CPU */
788 void __init smp_prepare_cpus(unsigned int max_cpus)
791 unsigned long save_area = 0;
793 unsigned long async_stack, panic_stack;
794 struct _lowcore *lowcore;
800 /* request the 0x1201 emergency signal external interrupt */
801 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
802 panic("Couldn't request external interrupt 0x1201");
803 print_cpu_info(&S390_lowcore.cpu_data);
805 /* Reallocate current lowcore, but keep its contents. */
806 lc_order = sizeof(long) == 8 ? 1 : 0;
807 lowcore = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, lc_order);
808 panic_stack = __get_free_page(GFP_KERNEL);
809 async_stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
811 if (MACHINE_HAS_IEEE)
812 save_area = get_zeroed_page(GFP_KERNEL);
815 local_mcck_disable();
816 lowcore_ptr[smp_processor_id()] = lowcore;
817 *lowcore = S390_lowcore;
818 lowcore->panic_stack = panic_stack + PAGE_SIZE;
819 lowcore->async_stack = async_stack + ASYNC_SIZE;
821 if (MACHINE_HAS_IEEE)
822 lowcore->extended_save_area_addr = (u32) save_area;
824 set_prefix((u32)(unsigned long) lowcore);
827 for_each_possible_cpu(cpu)
828 if (cpu != smp_processor_id())
829 smp_create_idle(cpu);
832 void __init smp_prepare_boot_cpu(void)
834 BUG_ON(smp_processor_id() != 0);
836 current_thread_info()->cpu = 0;
837 cpu_set(0, cpu_present_map);
838 cpu_set(0, cpu_online_map);
839 S390_lowcore.percpu_offset = __per_cpu_offset[0];
840 current_set[0] = current;
841 smp_cpu_state[0] = CPU_STATE_CONFIGURED;
842 smp_cpu_polarization[0] = POLARIZATION_UNKNWN;
845 void __init smp_cpus_done(unsigned int max_cpus)
850 * the frequency of the profiling timer can be changed
851 * by writing a multiplier value into /proc/profile.
853 * usually you want to run this on all CPUs ;)
855 int setup_profiling_timer(unsigned int multiplier)
860 #ifdef CONFIG_HOTPLUG_CPU
861 static ssize_t cpu_configure_show(struct sys_device *dev,
862 struct sysdev_attribute *attr, char *buf)
866 mutex_lock(&smp_cpu_state_mutex);
867 count = sprintf(buf, "%d\n", smp_cpu_state[dev->id]);
868 mutex_unlock(&smp_cpu_state_mutex);
872 static ssize_t cpu_configure_store(struct sys_device *dev,
873 struct sysdev_attribute *attr,
874 const char *buf, size_t count)
880 if (sscanf(buf, "%d %c", &val, &delim) != 1)
882 if (val != 0 && val != 1)
886 mutex_lock(&smp_cpu_state_mutex);
893 if (smp_cpu_state[cpu] == CPU_STATE_CONFIGURED) {
894 rc = sclp_cpu_deconfigure(__cpu_logical_map[cpu]);
896 smp_cpu_state[cpu] = CPU_STATE_STANDBY;
897 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
902 if (smp_cpu_state[cpu] == CPU_STATE_STANDBY) {
903 rc = sclp_cpu_configure(__cpu_logical_map[cpu]);
905 smp_cpu_state[cpu] = CPU_STATE_CONFIGURED;
906 smp_cpu_polarization[cpu] = POLARIZATION_UNKNWN;
914 mutex_unlock(&smp_cpu_state_mutex);
916 return rc ? rc : count;
918 static SYSDEV_ATTR(configure, 0644, cpu_configure_show, cpu_configure_store);
919 #endif /* CONFIG_HOTPLUG_CPU */
921 static ssize_t cpu_polarization_show(struct sys_device *dev,
922 struct sysdev_attribute *attr, char *buf)
927 mutex_lock(&smp_cpu_state_mutex);
928 switch (smp_cpu_polarization[cpu]) {
929 case POLARIZATION_HRZ:
930 count = sprintf(buf, "horizontal\n");
932 case POLARIZATION_VL:
933 count = sprintf(buf, "vertical:low\n");
935 case POLARIZATION_VM:
936 count = sprintf(buf, "vertical:medium\n");
938 case POLARIZATION_VH:
939 count = sprintf(buf, "vertical:high\n");
942 count = sprintf(buf, "unknown\n");
945 mutex_unlock(&smp_cpu_state_mutex);
948 static SYSDEV_ATTR(polarization, 0444, cpu_polarization_show, NULL);
950 static ssize_t show_cpu_address(struct sys_device *dev,
951 struct sysdev_attribute *attr, char *buf)
953 return sprintf(buf, "%d\n", __cpu_logical_map[dev->id]);
955 static SYSDEV_ATTR(address, 0444, show_cpu_address, NULL);
958 static struct attribute *cpu_common_attrs[] = {
959 #ifdef CONFIG_HOTPLUG_CPU
960 &attr_configure.attr,
963 &attr_polarization.attr,
967 static struct attribute_group cpu_common_attr_group = {
968 .attrs = cpu_common_attrs,
971 static ssize_t show_capability(struct sys_device *dev,
972 struct sysdev_attribute *attr, char *buf)
974 unsigned int capability;
977 rc = get_cpu_capability(&capability);
980 return sprintf(buf, "%u\n", capability);
982 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
984 static ssize_t show_idle_count(struct sys_device *dev,
985 struct sysdev_attribute *attr, char *buf)
987 struct s390_idle_data *idle;
988 unsigned long long idle_count;
990 idle = &per_cpu(s390_idle, dev->id);
991 spin_lock_irq(&idle->lock);
992 idle_count = idle->idle_count;
993 spin_unlock_irq(&idle->lock);
994 return sprintf(buf, "%llu\n", idle_count);
996 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
998 static ssize_t show_idle_time(struct sys_device *dev,
999 struct sysdev_attribute *attr, char *buf)
1001 struct s390_idle_data *idle;
1002 unsigned long long new_time;
1004 idle = &per_cpu(s390_idle, dev->id);
1005 spin_lock_irq(&idle->lock);
1006 if (idle->in_idle) {
1007 new_time = get_clock();
1008 idle->idle_time += new_time - idle->idle_enter;
1009 idle->idle_enter = new_time;
1011 new_time = idle->idle_time;
1012 spin_unlock_irq(&idle->lock);
1013 return sprintf(buf, "%llu\n", new_time >> 12);
1015 static SYSDEV_ATTR(idle_time_us, 0444, show_idle_time, NULL);
1017 static struct attribute *cpu_online_attrs[] = {
1018 &attr_capability.attr,
1019 &attr_idle_count.attr,
1020 &attr_idle_time_us.attr,
1024 static struct attribute_group cpu_online_attr_group = {
1025 .attrs = cpu_online_attrs,
1028 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
1029 unsigned long action, void *hcpu)
1031 unsigned int cpu = (unsigned int)(long)hcpu;
1032 struct cpu *c = &per_cpu(cpu_devices, cpu);
1033 struct sys_device *s = &c->sysdev;
1034 struct s390_idle_data *idle;
1038 case CPU_ONLINE_FROZEN:
1039 idle = &per_cpu(s390_idle, cpu);
1040 spin_lock_irq(&idle->lock);
1041 idle->idle_enter = 0;
1042 idle->idle_time = 0;
1043 idle->idle_count = 0;
1044 spin_unlock_irq(&idle->lock);
1045 if (sysfs_create_group(&s->kobj, &cpu_online_attr_group))
1049 case CPU_DEAD_FROZEN:
1050 sysfs_remove_group(&s->kobj, &cpu_online_attr_group);
1056 static struct notifier_block __cpuinitdata smp_cpu_nb = {
1057 .notifier_call = smp_cpu_notify,
1060 static int __devinit smp_add_present_cpu(int cpu)
1062 struct cpu *c = &per_cpu(cpu_devices, cpu);
1063 struct sys_device *s = &c->sysdev;
1066 c->hotpluggable = 1;
1067 rc = register_cpu(c, cpu);
1070 rc = sysfs_create_group(&s->kobj, &cpu_common_attr_group);
1073 if (!cpu_online(cpu))
1075 rc = sysfs_create_group(&s->kobj, &cpu_online_attr_group);
1078 sysfs_remove_group(&s->kobj, &cpu_common_attr_group);
1080 #ifdef CONFIG_HOTPLUG_CPU
1087 #ifdef CONFIG_HOTPLUG_CPU
1089 int __ref smp_rescan_cpus(void)
1096 mutex_lock(&smp_cpu_state_mutex);
1097 newcpus = cpu_present_map;
1098 rc = __smp_rescan_cpus();
1101 cpus_andnot(newcpus, cpu_present_map, newcpus);
1102 for_each_cpu_mask(cpu, newcpus) {
1103 rc = smp_add_present_cpu(cpu);
1105 cpu_clear(cpu, cpu_present_map);
1109 mutex_unlock(&smp_cpu_state_mutex);
1111 if (!cpus_empty(newcpus))
1112 topology_schedule_update();
1116 static ssize_t __ref rescan_store(struct sysdev_class *class, const char *buf,
1121 rc = smp_rescan_cpus();
1122 return rc ? rc : count;
1124 static SYSDEV_CLASS_ATTR(rescan, 0200, NULL, rescan_store);
1125 #endif /* CONFIG_HOTPLUG_CPU */
1127 static ssize_t dispatching_show(struct sysdev_class *class, char *buf)
1131 mutex_lock(&smp_cpu_state_mutex);
1132 count = sprintf(buf, "%d\n", cpu_management);
1133 mutex_unlock(&smp_cpu_state_mutex);
1137 static ssize_t dispatching_store(struct sysdev_class *dev, const char *buf,
1143 if (sscanf(buf, "%d %c", &val, &delim) != 1)
1145 if (val != 0 && val != 1)
1149 mutex_lock(&smp_cpu_state_mutex);
1150 if (cpu_management == val)
1152 rc = topology_set_cpu_management(val);
1154 cpu_management = val;
1156 mutex_unlock(&smp_cpu_state_mutex);
1158 return rc ? rc : count;
1160 static SYSDEV_CLASS_ATTR(dispatching, 0644, dispatching_show,
1163 static int __init topology_init(void)
1168 register_cpu_notifier(&smp_cpu_nb);
1170 #ifdef CONFIG_HOTPLUG_CPU
1171 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_rescan);
1175 rc = sysdev_class_create_file(&cpu_sysdev_class, &attr_dispatching);
1178 for_each_present_cpu(cpu) {
1179 rc = smp_add_present_cpu(cpu);
1185 subsys_initcall(topology_init);