2 * arch/s390/kernel/smp.c
4 * Copyright (C) IBM Corp. 1999,2006
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/spinlock.h>
27 #include <linux/kernel_stat.h>
28 #include <linux/smp_lock.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>
35 #include <asm/setup.h>
37 #include <asm/pgalloc.h>
39 #include <asm/s390_ext.h>
40 #include <asm/cpcmd.h>
41 #include <asm/tlbflush.h>
42 #include <asm/timer.h>
44 extern volatile int __cpu_logical_map[];
47 * An array with a pointer the lowcore of every CPU.
50 struct _lowcore *lowcore_ptr[NR_CPUS];
52 cpumask_t cpu_online_map = CPU_MASK_NONE;
53 cpumask_t cpu_possible_map = CPU_MASK_NONE;
55 static struct task_struct *current_set[NR_CPUS];
57 static void smp_ext_bitcall(int, ec_bit_sig);
60 * Structure and data for __smp_call_function_map(). This is designed to
61 * minimise static memory requirements. It also looks cleaner.
63 static DEFINE_SPINLOCK(call_lock);
65 struct call_data_struct {
66 void (*func) (void *info);
73 static struct call_data_struct * call_data;
76 * 'Call function' interrupt callback
78 static void do_call_function(void)
80 void (*func) (void *info) = call_data->func;
81 void *info = call_data->info;
82 int wait = call_data->wait;
84 cpu_set(smp_processor_id(), call_data->started);
87 cpu_set(smp_processor_id(), call_data->finished);;
90 static void __smp_call_function_map(void (*func) (void *info), void *info,
91 int nonatomic, int wait, cpumask_t map)
93 struct call_data_struct data;
97 * Can deadlock when interrupts are disabled or if in wrong context,
98 * caller must disable preemption
100 WARN_ON(irqs_disabled() || in_irq() || preemptible());
103 * Check for local function call. We have to have the same call order
104 * as in on_each_cpu() because of machine_restart_smp().
106 if (cpu_isset(smp_processor_id(), map)) {
108 cpu_clear(smp_processor_id(), map);
111 cpus_and(map, map, cpu_online_map);
117 data.started = CPU_MASK_NONE;
120 data.finished = CPU_MASK_NONE;
122 spin_lock_bh(&call_lock);
125 for_each_cpu_mask(cpu, map)
126 smp_ext_bitcall(cpu, ec_call_function);
128 /* Wait for response */
129 while (!cpus_equal(map, data.started))
133 while (!cpus_equal(map, data.finished))
136 spin_unlock_bh(&call_lock);
147 * @func: the function to run; this must be fast and non-blocking
148 * @info: an arbitrary pointer to pass to the function
150 * @wait: if true, wait (atomically) until function has completed on other CPUs
152 * Run a function on all other CPUs.
154 * You must not call this function with disabled interrupts or from a
155 * hardware interrupt handler. Must be called with preemption disabled.
156 * You may call it from a bottom half.
158 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
163 map = cpu_online_map;
164 cpu_clear(smp_processor_id(), map);
165 __smp_call_function_map(func, info, nonatomic, wait, map);
168 EXPORT_SYMBOL(smp_call_function);
171 * smp_call_function_on:
172 * @func: the function to run; this must be fast and non-blocking
173 * @info: an arbitrary pointer to pass to the function
175 * @wait: if true, wait (atomically) until function has completed on other CPUs
176 * @cpu: the CPU where func should run
178 * Run a function on one processor.
180 * You must not call this function with disabled interrupts or from a
181 * hardware interrupt handler. Must be called with preemption disabled.
182 * You may call it from a bottom half.
184 int smp_call_function_on(void (*func) (void *info), void *info, int nonatomic,
187 cpumask_t map = CPU_MASK_NONE;
190 __smp_call_function_map(func, info, nonatomic, wait, map);
193 EXPORT_SYMBOL(smp_call_function_on);
195 static void do_send_stop(void)
199 /* stop all processors */
200 for_each_online_cpu(cpu) {
201 if (cpu == smp_processor_id())
204 rc = signal_processor(cpu, sigp_stop);
205 } while (rc == sigp_busy);
209 static void do_store_status(void)
213 /* store status of all processors in their lowcores (real 0) */
214 for_each_online_cpu(cpu) {
215 if (cpu == smp_processor_id())
218 rc = signal_processor_p(
219 (__u32)(unsigned long) lowcore_ptr[cpu], cpu,
220 sigp_store_status_at_address);
221 } while(rc == sigp_busy);
225 static void do_wait_for_stop(void)
229 /* Wait for all other cpus to enter stopped state */
230 for_each_online_cpu(cpu) {
231 if (cpu == smp_processor_id())
233 while(!smp_cpu_not_running(cpu))
239 * this function sends a 'stop' sigp to all other CPUs in the system.
240 * it goes straight through.
242 void smp_send_stop(void)
244 /* Disable all interrupts/machine checks */
245 __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
247 /* write magic number to zero page (absolute 0) */
248 lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
250 /* stop other processors. */
253 /* wait until other processors are stopped */
256 /* store status of other processors. */
261 * 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) {
361 struct ec_creg_mask_parms *pp = info;
362 unsigned long cregs[16];
365 __ctl_store(cregs, 0, 15);
366 for (i = 0; i <= 15; i++)
367 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
368 __ctl_load(cregs, 0, 15);
372 * Set a bit in a control register of all cpus
374 void smp_ctl_set_bit(int cr, int bit)
376 struct ec_creg_mask_parms parms;
378 memset(&parms.orvals, 0, sizeof(parms.orvals));
379 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
380 parms.orvals[cr] = 1 << bit;
381 on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
385 * Clear a bit in a control register of all cpus
387 void smp_ctl_clear_bit(int cr, int bit)
389 struct ec_creg_mask_parms parms;
391 memset(&parms.orvals, 0, sizeof(parms.orvals));
392 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
393 parms.andvals[cr] = ~(1L << bit);
394 on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
398 * Lets check how many CPUs we have.
402 __init smp_count_cpus(void)
404 unsigned int cpu, num_cpus;
408 * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
411 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
412 current_thread_info()->cpu = 0;
414 for (cpu = 0; cpu <= 65535; cpu++) {
415 if ((__u16) cpu == boot_cpu_addr)
417 __cpu_logical_map[1] = (__u16) cpu;
418 if (signal_processor(1, sigp_sense) ==
419 sigp_not_operational)
424 printk("Detected %d CPU's\n",(int) num_cpus);
425 printk("Boot cpu address %2X\n", boot_cpu_addr);
431 * Activate a secondary processor.
433 int __devinit start_secondary(void *cpuvoid)
438 /* Enable TOD clock interrupts on the secondary cpu. */
440 #ifdef CONFIG_VIRT_TIMER
441 /* Enable cpu timer interrupts on the secondary cpu. */
444 /* Enable pfault pseudo page faults on this cpu. */
447 /* Mark this cpu as online */
448 cpu_set(smp_processor_id(), cpu_online_map);
449 /* Switch on interrupts */
451 /* Print info about this processor */
452 print_cpu_info(&S390_lowcore.cpu_data);
453 /* cpu_idle will call schedule for us */
458 static void __init smp_create_idle(unsigned int cpu)
460 struct task_struct *p;
463 * don't care about the psw and regs settings since we'll never
464 * reschedule the forked task.
468 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
469 current_set[cpu] = p;
472 /* Reserving and releasing of CPUs */
474 static DEFINE_SPINLOCK(smp_reserve_lock);
475 static int smp_cpu_reserved[NR_CPUS];
478 smp_get_cpu(cpumask_t cpu_mask)
483 spin_lock_irqsave(&smp_reserve_lock, flags);
484 /* Try to find an already reserved cpu. */
485 for_each_cpu_mask(cpu, cpu_mask) {
486 if (smp_cpu_reserved[cpu] != 0) {
487 smp_cpu_reserved[cpu]++;
492 /* Reserve a new cpu from cpu_mask. */
493 for_each_cpu_mask(cpu, cpu_mask) {
494 if (cpu_online(cpu)) {
495 smp_cpu_reserved[cpu]++;
501 spin_unlock_irqrestore(&smp_reserve_lock, flags);
510 spin_lock_irqsave(&smp_reserve_lock, flags);
511 smp_cpu_reserved[cpu]--;
512 spin_unlock_irqrestore(&smp_reserve_lock, flags);
520 /* Check for stopped state */
521 if (signal_processor_ps(&status, 0, cpu, sigp_sense) == sigp_status_stored) {
528 /* Upping and downing of CPUs */
531 __cpu_up(unsigned int cpu)
533 struct task_struct *idle;
534 struct _lowcore *cpu_lowcore;
535 struct stack_frame *sf;
539 for (curr_cpu = 0; curr_cpu <= 65535; curr_cpu++) {
540 __cpu_logical_map[cpu] = (__u16) curr_cpu;
541 if (cpu_stopped(cpu))
545 if (!cpu_stopped(cpu))
548 ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
549 cpu, sigp_set_prefix);
551 printk("sigp_set_prefix failed for cpu %d "
552 "with condition code %d\n",
553 (int) cpu, (int) ccode);
557 idle = current_set[cpu];
558 cpu_lowcore = lowcore_ptr[cpu];
559 cpu_lowcore->kernel_stack = (unsigned long)
560 task_stack_page(idle) + (THREAD_SIZE);
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], 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_data.cpu_nr = cpu;
576 while (signal_processor(cpu,sigp_restart) == sigp_busy)
579 while (!cpu_online(cpu))
584 static unsigned int __initdata additional_cpus;
585 static unsigned int __initdata possible_cpus;
587 void __init smp_setup_cpu_possible_map(void)
589 unsigned int phy_cpus, pos_cpus, cpu;
591 phy_cpus = smp_count_cpus();
592 pos_cpus = min(phy_cpus + additional_cpus, (unsigned int) NR_CPUS);
595 pos_cpus = min(possible_cpus, (unsigned int) NR_CPUS);
597 for (cpu = 0; cpu < pos_cpus; cpu++)
598 cpu_set(cpu, cpu_possible_map);
600 phy_cpus = min(phy_cpus, pos_cpus);
602 for (cpu = 0; cpu < phy_cpus; cpu++)
603 cpu_set(cpu, cpu_present_map);
606 #ifdef CONFIG_HOTPLUG_CPU
608 static int __init setup_additional_cpus(char *s)
610 additional_cpus = simple_strtoul(s, NULL, 0);
613 early_param("additional_cpus", setup_additional_cpus);
615 static int __init setup_possible_cpus(char *s)
617 possible_cpus = simple_strtoul(s, NULL, 0);
620 early_param("possible_cpus", setup_possible_cpus);
626 struct ec_creg_mask_parms cr_parms;
627 int cpu = smp_processor_id();
629 spin_lock_irqsave(&smp_reserve_lock, flags);
630 if (smp_cpu_reserved[cpu] != 0) {
631 spin_unlock_irqrestore(&smp_reserve_lock, flags);
634 cpu_clear(cpu, cpu_online_map);
636 /* Disable pfault pseudo page faults on this cpu. */
639 memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
640 memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
642 /* disable all external interrupts */
643 cr_parms.orvals[0] = 0;
644 cr_parms.andvals[0] = ~(1<<15 | 1<<14 | 1<<13 | 1<<12 |
645 1<<11 | 1<<10 | 1<< 6 | 1<< 4);
646 /* disable all I/O interrupts */
647 cr_parms.orvals[6] = 0;
648 cr_parms.andvals[6] = ~(1<<31 | 1<<30 | 1<<29 | 1<<28 |
649 1<<27 | 1<<26 | 1<<25 | 1<<24);
650 /* disable most machine checks */
651 cr_parms.orvals[14] = 0;
652 cr_parms.andvals[14] = ~(1<<28 | 1<<27 | 1<<26 | 1<<25 | 1<<24);
654 smp_ctl_bit_callback(&cr_parms);
656 spin_unlock_irqrestore(&smp_reserve_lock, flags);
661 __cpu_die(unsigned int cpu)
663 /* Wait until target cpu is down */
664 while (!smp_cpu_not_running(cpu))
666 printk("Processor %d spun down\n", cpu);
673 signal_processor(smp_processor_id(), sigp_stop);
678 #endif /* CONFIG_HOTPLUG_CPU */
681 * Cycle through the processors and setup structures.
684 void __init smp_prepare_cpus(unsigned int max_cpus)
690 /* request the 0x1201 emergency signal external interrupt */
691 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
692 panic("Couldn't request external interrupt 0x1201");
693 memset(lowcore_ptr,0,sizeof(lowcore_ptr));
695 * Initialize prefix pages and stacks for all possible cpus
697 print_cpu_info(&S390_lowcore.cpu_data);
699 for_each_possible_cpu(i) {
700 lowcore_ptr[i] = (struct _lowcore *)
701 __get_free_pages(GFP_KERNEL|GFP_DMA,
702 sizeof(void*) == 8 ? 1 : 0);
703 stack = __get_free_pages(GFP_KERNEL,ASYNC_ORDER);
704 if (lowcore_ptr[i] == NULL || stack == 0ULL)
705 panic("smp_boot_cpus failed to allocate memory\n");
707 *(lowcore_ptr[i]) = S390_lowcore;
708 lowcore_ptr[i]->async_stack = stack + (ASYNC_SIZE);
709 stack = __get_free_pages(GFP_KERNEL,0);
711 panic("smp_boot_cpus failed to allocate memory\n");
712 lowcore_ptr[i]->panic_stack = stack + (PAGE_SIZE);
714 if (MACHINE_HAS_IEEE) {
715 lowcore_ptr[i]->extended_save_area_addr =
716 (__u32) __get_free_pages(GFP_KERNEL,0);
717 if (lowcore_ptr[i]->extended_save_area_addr == 0)
718 panic("smp_boot_cpus failed to "
719 "allocate memory\n");
724 if (MACHINE_HAS_IEEE)
725 ctl_set_bit(14, 29); /* enable extended save area */
727 set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
729 for_each_possible_cpu(cpu)
730 if (cpu != smp_processor_id())
731 smp_create_idle(cpu);
734 void __devinit smp_prepare_boot_cpu(void)
736 BUG_ON(smp_processor_id() != 0);
738 cpu_set(0, cpu_online_map);
739 S390_lowcore.percpu_offset = __per_cpu_offset[0];
740 current_set[0] = current;
743 void smp_cpus_done(unsigned int max_cpus)
745 cpu_present_map = cpu_possible_map;
749 * the frequency of the profiling timer can be changed
750 * by writing a multiplier value into /proc/profile.
752 * usually you want to run this on all CPUs ;)
754 int setup_profiling_timer(unsigned int multiplier)
759 static DEFINE_PER_CPU(struct cpu, cpu_devices);
761 static int __init topology_init(void)
766 for_each_possible_cpu(cpu) {
767 struct cpu *c = &per_cpu(cpu_devices, cpu);
770 ret = register_cpu(c, cpu);
772 printk(KERN_WARNING "topology_init: register_cpu %d "
773 "failed (%d)\n", cpu, ret);
778 subsys_initcall(topology_init);
780 EXPORT_SYMBOL(cpu_online_map);
781 EXPORT_SYMBOL(cpu_possible_map);
782 EXPORT_SYMBOL(lowcore_ptr);
783 EXPORT_SYMBOL(smp_ctl_set_bit);
784 EXPORT_SYMBOL(smp_ctl_clear_bit);
785 EXPORT_SYMBOL(smp_get_cpu);
786 EXPORT_SYMBOL(smp_put_cpu);