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>
48 * An array with a pointer the lowcore of every CPU.
50 struct _lowcore *lowcore_ptr[NR_CPUS];
51 EXPORT_SYMBOL(lowcore_ptr);
53 cpumask_t cpu_online_map = CPU_MASK_NONE;
54 EXPORT_SYMBOL(cpu_online_map);
56 cpumask_t cpu_possible_map = CPU_MASK_NONE;
57 EXPORT_SYMBOL(cpu_possible_map);
59 static struct task_struct *current_set[NR_CPUS];
61 static void smp_ext_bitcall(int, ec_bit_sig);
64 * Structure and data for __smp_call_function_map(). This is designed to
65 * minimise static memory requirements. It also looks cleaner.
67 static DEFINE_SPINLOCK(call_lock);
69 struct call_data_struct {
70 void (*func) (void *info);
77 static struct call_data_struct *call_data;
80 * 'Call function' interrupt callback
82 static void do_call_function(void)
84 void (*func) (void *info) = call_data->func;
85 void *info = call_data->info;
86 int wait = call_data->wait;
88 cpu_set(smp_processor_id(), call_data->started);
91 cpu_set(smp_processor_id(), call_data->finished);;
94 static void __smp_call_function_map(void (*func) (void *info), void *info,
95 int nonatomic, int wait, cpumask_t map)
97 struct call_data_struct data;
101 * Can deadlock when interrupts are disabled or if in wrong context.
103 WARN_ON(irqs_disabled() || in_irq());
106 * Check for local function call. We have to have the same call order
107 * as in on_each_cpu() because of machine_restart_smp().
109 if (cpu_isset(smp_processor_id(), map)) {
111 cpu_clear(smp_processor_id(), map);
114 cpus_and(map, map, cpu_online_map);
120 data.started = CPU_MASK_NONE;
123 data.finished = CPU_MASK_NONE;
125 spin_lock(&call_lock);
128 for_each_cpu_mask(cpu, map)
129 smp_ext_bitcall(cpu, ec_call_function);
131 /* Wait for response */
132 while (!cpus_equal(map, data.started))
135 while (!cpus_equal(map, data.finished))
137 spin_unlock(&call_lock);
148 * @func: the function to run; this must be fast and non-blocking
149 * @info: an arbitrary pointer to pass to the function
151 * @wait: if true, wait (atomically) until function has completed on other CPUs
153 * Run a function on all other CPUs.
155 * You must not call this function with disabled interrupts, from a
156 * hardware interrupt handler or from a bottom half.
158 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
164 map = cpu_online_map;
165 cpu_clear(smp_processor_id(), map);
166 __smp_call_function_map(func, info, nonatomic, wait, map);
170 EXPORT_SYMBOL(smp_call_function);
173 * smp_call_function_single:
174 * @cpu: the CPU where func should run
175 * @func: the function to run; this must be fast and non-blocking
176 * @info: an arbitrary pointer to pass to the function
178 * @wait: if true, wait (atomically) until function has completed on other CPUs
180 * Run a function on one processor.
182 * You must not call this function with disabled interrupts, from a
183 * hardware interrupt handler or from a bottom half.
185 int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
186 int nonatomic, int wait)
189 __smp_call_function_map(func, info, nonatomic, wait,
190 cpumask_of_cpu(cpu));
194 EXPORT_SYMBOL(smp_call_function_single);
196 static void do_send_stop(void)
200 /* stop all processors */
201 for_each_online_cpu(cpu) {
202 if (cpu == smp_processor_id())
205 rc = signal_processor(cpu, sigp_stop);
206 } while (rc == sigp_busy);
210 static void do_store_status(void)
214 /* store status of all processors in their lowcores (real 0) */
215 for_each_online_cpu(cpu) {
216 if (cpu == smp_processor_id())
219 rc = signal_processor_p(
220 (__u32)(unsigned long) lowcore_ptr[cpu], cpu,
221 sigp_store_status_at_address);
222 } while (rc == sigp_busy);
226 static void do_wait_for_stop(void)
230 /* Wait for all other cpus to enter stopped state */
231 for_each_online_cpu(cpu) {
232 if (cpu == smp_processor_id())
234 while (!smp_cpu_not_running(cpu))
240 * this function sends a 'stop' sigp to all other CPUs in the system.
241 * it goes straight through.
243 void smp_send_stop(void)
245 /* Disable all interrupts/machine checks */
246 __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
248 /* write magic number to zero page (absolute 0) */
249 lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
251 /* stop other processors. */
254 /* wait until other processors are stopped */
257 /* store status of other processors. */
262 * Reboot, halt and power_off routines for SMP.
264 void machine_restart_smp(char *__unused)
270 void machine_halt_smp(void)
273 if (MACHINE_IS_VM && strlen(vmhalt_cmd) > 0)
274 __cpcmd(vmhalt_cmd, NULL, 0, NULL);
275 signal_processor(smp_processor_id(), sigp_stop_and_store_status);
279 void machine_power_off_smp(void)
282 if (MACHINE_IS_VM && strlen(vmpoff_cmd) > 0)
283 __cpcmd(vmpoff_cmd, NULL, 0, NULL);
284 signal_processor(smp_processor_id(), sigp_stop_and_store_status);
289 * This is the main routine where commands issued by other
293 static void do_ext_call_interrupt(__u16 code)
298 * handle bit signal external calls
300 * For the ec_schedule signal we have to do nothing. All the work
301 * is done automatically when we return from the interrupt.
303 bits = xchg(&S390_lowcore.ext_call_fast, 0);
305 if (test_bit(ec_call_function, &bits))
310 * Send an external call sigp to another cpu and return without waiting
311 * for its completion.
313 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
316 * Set signaling bit in lowcore of target cpu and kick it
318 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
319 while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
325 * this function sends a 'purge tlb' signal to another CPU.
327 void smp_ptlb_callback(void *info)
332 void smp_ptlb_all(void)
334 on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
336 EXPORT_SYMBOL(smp_ptlb_all);
337 #endif /* ! CONFIG_64BIT */
340 * this function sends a 'reschedule' IPI to another CPU.
341 * it goes straight through and wastes no time serializing
342 * anything. Worst case is that we lose a reschedule ...
344 void smp_send_reschedule(int cpu)
346 smp_ext_bitcall(cpu, ec_schedule);
350 * parameter area for the set/clear control bit callbacks
352 struct ec_creg_mask_parms {
353 unsigned long orvals[16];
354 unsigned long andvals[16];
358 * callback for setting/clearing control bits
360 static void smp_ctl_bit_callback(void *info)
362 struct ec_creg_mask_parms *pp = info;
363 unsigned long cregs[16];
366 __ctl_store(cregs, 0, 15);
367 for (i = 0; i <= 15; i++)
368 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
369 __ctl_load(cregs, 0, 15);
373 * Set a bit in a control register of all cpus
375 void smp_ctl_set_bit(int cr, int bit)
377 struct ec_creg_mask_parms parms;
379 memset(&parms.orvals, 0, sizeof(parms.orvals));
380 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
381 parms.orvals[cr] = 1 << bit;
382 on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
384 EXPORT_SYMBOL(smp_ctl_set_bit);
387 * Clear a bit in a control register of all cpus
389 void smp_ctl_clear_bit(int cr, int bit)
391 struct ec_creg_mask_parms parms;
393 memset(&parms.orvals, 0, sizeof(parms.orvals));
394 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
395 parms.andvals[cr] = ~(1L << bit);
396 on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
398 EXPORT_SYMBOL(smp_ctl_clear_bit);
400 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
403 * zfcpdump_prefix_array holds prefix registers for the following scenario:
404 * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
405 * save its prefix registers, since they get lost, when switching from 31 bit
408 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
409 __attribute__((__section__(".data")));
411 static void __init smp_get_save_area(unsigned int cpu, unsigned int phy_cpu)
413 if (ipl_info.type != IPL_TYPE_FCP_DUMP)
415 if (cpu >= NR_CPUS) {
416 printk(KERN_WARNING "Registers for cpu %i not saved since dump "
417 "kernel was compiled with NR_CPUS=%i\n", cpu, NR_CPUS);
420 zfcpdump_save_areas[cpu] = alloc_bootmem(sizeof(union save_area));
421 __cpu_logical_map[1] = (__u16) phy_cpu;
422 while (signal_processor(1, sigp_stop_and_store_status) == sigp_busy)
424 memcpy(zfcpdump_save_areas[cpu],
425 (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE,
428 /* copy original prefix register */
429 zfcpdump_save_areas[cpu]->s390x.pref_reg = zfcpdump_prefix_array[cpu];
433 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
434 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
438 static inline void smp_get_save_area(unsigned int cpu, unsigned int phy_cpu) { }
440 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
443 * Lets check how many CPUs we have.
445 static unsigned int __init smp_count_cpus(void)
447 unsigned int cpu, num_cpus;
451 * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
453 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
454 current_thread_info()->cpu = 0;
456 for (cpu = 0; cpu <= 65535; cpu++) {
457 if ((__u16) cpu == boot_cpu_addr)
459 __cpu_logical_map[1] = (__u16) cpu;
460 if (signal_processor(1, sigp_sense) == sigp_not_operational)
462 smp_get_save_area(num_cpus, cpu);
465 printk("Detected %d CPU's\n", (int) num_cpus);
466 printk("Boot cpu address %2X\n", boot_cpu_addr);
471 * Activate a secondary processor.
473 int __cpuinit start_secondary(void *cpuvoid)
478 /* Enable TOD clock interrupts on the secondary cpu. */
480 #ifdef CONFIG_VIRT_TIMER
481 /* Enable cpu timer interrupts on the secondary cpu. */
484 /* Enable pfault pseudo page faults on this cpu. */
487 /* Mark this cpu as online */
488 cpu_set(smp_processor_id(), cpu_online_map);
489 /* Switch on interrupts */
491 /* Print info about this processor */
492 print_cpu_info(&S390_lowcore.cpu_data);
493 /* cpu_idle will call schedule for us */
498 DEFINE_PER_CPU(struct s390_idle_data, s390_idle);
500 static void __init smp_create_idle(unsigned int cpu)
502 struct task_struct *p;
505 * don't care about the psw and regs settings since we'll never
506 * reschedule the forked task.
510 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
511 current_set[cpu] = p;
512 spin_lock_init(&(&per_cpu(s390_idle, cpu))->lock);
515 static int cpu_stopped(int cpu)
519 /* Check for stopped state */
520 if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
521 sigp_status_stored) {
528 /* Upping and downing of CPUs */
530 int __cpu_up(unsigned int cpu)
532 struct task_struct *idle;
533 struct _lowcore *cpu_lowcore;
534 struct stack_frame *sf;
538 for (curr_cpu = 0; curr_cpu <= 65535; curr_cpu++) {
539 __cpu_logical_map[cpu] = (__u16) curr_cpu;
540 if (cpu_stopped(cpu))
544 if (!cpu_stopped(cpu))
547 ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
548 cpu, sigp_set_prefix);
550 printk("sigp_set_prefix failed for cpu %d "
551 "with condition code %d\n",
552 (int) cpu, (int) ccode);
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 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
561 - sizeof(struct pt_regs)
562 - sizeof(struct stack_frame));
563 memset(sf, 0, sizeof(struct stack_frame));
564 sf->gprs[9] = (unsigned long) sf;
565 cpu_lowcore->save_area[15] = (unsigned long) sf;
566 __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
569 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
570 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
571 cpu_lowcore->current_task = (unsigned long) idle;
572 cpu_lowcore->cpu_data.cpu_nr = cpu;
575 while (signal_processor(cpu, sigp_restart) == sigp_busy)
578 while (!cpu_online(cpu))
583 static unsigned int __initdata additional_cpus;
584 static unsigned int __initdata possible_cpus;
586 void __init smp_setup_cpu_possible_map(void)
588 unsigned int phy_cpus, pos_cpus, cpu;
590 phy_cpus = smp_count_cpus();
591 pos_cpus = min(phy_cpus + additional_cpus, (unsigned int) NR_CPUS);
594 pos_cpus = min(possible_cpus, (unsigned int) NR_CPUS);
596 for (cpu = 0; cpu < pos_cpus; cpu++)
597 cpu_set(cpu, cpu_possible_map);
599 phy_cpus = min(phy_cpus, pos_cpus);
601 for (cpu = 0; cpu < phy_cpus; cpu++)
602 cpu_set(cpu, cpu_present_map);
605 #ifdef CONFIG_HOTPLUG_CPU
607 static int __init setup_additional_cpus(char *s)
609 additional_cpus = simple_strtoul(s, NULL, 0);
612 early_param("additional_cpus", setup_additional_cpus);
614 static int __init setup_possible_cpus(char *s)
616 possible_cpus = simple_strtoul(s, NULL, 0);
619 early_param("possible_cpus", setup_possible_cpus);
621 int __cpu_disable(void)
623 struct ec_creg_mask_parms cr_parms;
624 int cpu = smp_processor_id();
626 cpu_clear(cpu, cpu_online_map);
628 /* Disable pfault pseudo page faults on this cpu. */
631 memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
632 memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
634 /* disable all external interrupts */
635 cr_parms.orvals[0] = 0;
636 cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
637 1 << 11 | 1 << 10 | 1 << 6 | 1 << 4);
638 /* disable all I/O interrupts */
639 cr_parms.orvals[6] = 0;
640 cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
641 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
642 /* disable most machine checks */
643 cr_parms.orvals[14] = 0;
644 cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
647 smp_ctl_bit_callback(&cr_parms);
652 void __cpu_die(unsigned int cpu)
654 /* Wait until target cpu is down */
655 while (!smp_cpu_not_running(cpu))
657 printk("Processor %d spun down\n", cpu);
663 signal_processor(smp_processor_id(), sigp_stop);
668 #endif /* CONFIG_HOTPLUG_CPU */
671 * Cycle through the processors and setup structures.
674 void __init smp_prepare_cpus(unsigned int max_cpus)
680 /* request the 0x1201 emergency signal external interrupt */
681 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
682 panic("Couldn't request external interrupt 0x1201");
683 memset(lowcore_ptr, 0, sizeof(lowcore_ptr));
685 * Initialize prefix pages and stacks for all possible cpus
687 print_cpu_info(&S390_lowcore.cpu_data);
689 for_each_possible_cpu(i) {
690 lowcore_ptr[i] = (struct _lowcore *)
691 __get_free_pages(GFP_KERNEL | GFP_DMA,
692 sizeof(void*) == 8 ? 1 : 0);
693 stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
694 if (!lowcore_ptr[i] || !stack)
695 panic("smp_boot_cpus failed to allocate memory\n");
697 *(lowcore_ptr[i]) = S390_lowcore;
698 lowcore_ptr[i]->async_stack = stack + ASYNC_SIZE;
699 stack = __get_free_pages(GFP_KERNEL, 0);
701 panic("smp_boot_cpus failed to allocate memory\n");
702 lowcore_ptr[i]->panic_stack = stack + PAGE_SIZE;
704 if (MACHINE_HAS_IEEE) {
705 lowcore_ptr[i]->extended_save_area_addr =
706 (__u32) __get_free_pages(GFP_KERNEL, 0);
707 if (!lowcore_ptr[i]->extended_save_area_addr)
708 panic("smp_boot_cpus failed to "
709 "allocate memory\n");
714 if (MACHINE_HAS_IEEE)
715 ctl_set_bit(14, 29); /* enable extended save area */
717 set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
719 for_each_possible_cpu(cpu)
720 if (cpu != smp_processor_id())
721 smp_create_idle(cpu);
724 void __init smp_prepare_boot_cpu(void)
726 BUG_ON(smp_processor_id() != 0);
728 cpu_set(0, cpu_online_map);
729 S390_lowcore.percpu_offset = __per_cpu_offset[0];
730 current_set[0] = current;
731 spin_lock_init(&(&__get_cpu_var(s390_idle))->lock);
734 void __init smp_cpus_done(unsigned int max_cpus)
736 cpu_present_map = cpu_possible_map;
740 * the frequency of the profiling timer can be changed
741 * by writing a multiplier value into /proc/profile.
743 * usually you want to run this on all CPUs ;)
745 int setup_profiling_timer(unsigned int multiplier)
750 static DEFINE_PER_CPU(struct cpu, cpu_devices);
752 static ssize_t show_capability(struct sys_device *dev, char *buf)
754 unsigned int capability;
757 rc = get_cpu_capability(&capability);
760 return sprintf(buf, "%u\n", capability);
762 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
764 static ssize_t show_idle_count(struct sys_device *dev, char *buf)
766 struct s390_idle_data *idle;
767 unsigned long long idle_count;
769 idle = &per_cpu(s390_idle, dev->id);
770 spin_lock_irq(&idle->lock);
771 idle_count = idle->idle_count;
772 spin_unlock_irq(&idle->lock);
773 return sprintf(buf, "%llu\n", idle_count);
775 static SYSDEV_ATTR(idle_count, 0444, show_idle_count, NULL);
777 static ssize_t show_idle_time(struct sys_device *dev, char *buf)
779 struct s390_idle_data *idle;
780 unsigned long long new_time;
782 idle = &per_cpu(s390_idle, dev->id);
783 spin_lock_irq(&idle->lock);
785 new_time = get_clock();
786 idle->idle_time += new_time - idle->idle_enter;
787 idle->idle_enter = new_time;
789 new_time = idle->idle_time;
790 spin_unlock_irq(&idle->lock);
791 return sprintf(buf, "%llu us\n", new_time >> 12);
793 static SYSDEV_ATTR(idle_time, 0444, show_idle_time, NULL);
795 static struct attribute *cpu_attrs[] = {
796 &attr_capability.attr,
797 &attr_idle_count.attr,
798 &attr_idle_time.attr,
802 static struct attribute_group cpu_attr_group = {
806 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
807 unsigned long action, void *hcpu)
809 unsigned int cpu = (unsigned int)(long)hcpu;
810 struct cpu *c = &per_cpu(cpu_devices, cpu);
811 struct sys_device *s = &c->sysdev;
812 struct s390_idle_data *idle;
816 case CPU_ONLINE_FROZEN:
817 idle = &per_cpu(s390_idle, cpu);
818 spin_lock_irq(&idle->lock);
819 idle->idle_enter = 0;
821 idle->idle_count = 0;
822 spin_unlock_irq(&idle->lock);
823 if (sysfs_create_group(&s->kobj, &cpu_attr_group))
827 case CPU_DEAD_FROZEN:
828 sysfs_remove_group(&s->kobj, &cpu_attr_group);
834 static struct notifier_block __cpuinitdata smp_cpu_nb = {
835 .notifier_call = smp_cpu_notify,
838 static int __init topology_init(void)
843 register_cpu_notifier(&smp_cpu_nb);
845 for_each_possible_cpu(cpu) {
846 struct cpu *c = &per_cpu(cpu_devices, cpu);
847 struct sys_device *s = &c->sysdev;
850 register_cpu(c, cpu);
851 if (!cpu_online(cpu))
854 rc = sysfs_create_group(&s->kobj, &cpu_attr_group);
860 subsys_initcall(topology_init);