2 * arch/s390/kernel/smp.c
5 * Copyright (C) 1999,2000 IBM Deutschland Entwicklung GmbH, IBM Corporation
6 * Author(s): Denis Joseph Barrow (djbarrow@de.ibm.com,barrow_dj@yahoo.com),
7 * Martin Schwidefsky (schwidefsky@de.ibm.com)
8 * Heiko Carstens (heiko.carstens@de.ibm.com)
10 * based on other smp stuff by
11 * (c) 1995 Alan Cox, CymruNET Ltd <alan@cymru.net>
12 * (c) 1998 Ingo Molnar
14 * We work with logical cpu numbering everywhere we can. The only
15 * functions using the real cpu address (got from STAP) are the sigp
16 * functions. For all other functions we use the identity mapping.
17 * That means that cpu_number_map[i] == i for every cpu. cpu_number_map is
18 * used e.g. to find the idle task belonging to a logical cpu. Every array
19 * in the kernel is sorted by the logical cpu number and not by the physical
20 * one which is causing all the confusion with __cpu_logical_map and
21 * cpu_number_map in other architectures.
24 #include <linux/module.h>
25 #include <linux/init.h>
28 #include <linux/spinlock.h>
29 #include <linux/kernel_stat.h>
30 #include <linux/smp_lock.h>
32 #include <linux/delay.h>
33 #include <linux/cache.h>
34 #include <linux/interrupt.h>
35 #include <linux/cpu.h>
38 #include <asm/pgalloc.h>
40 #include <asm/s390_ext.h>
41 #include <asm/cpcmd.h>
42 #include <asm/tlbflush.h>
46 extern volatile int __cpu_logical_map[];
49 * An array with a pointer the lowcore of every CPU.
52 struct _lowcore *lowcore_ptr[NR_CPUS];
54 cpumask_t cpu_online_map;
55 cpumask_t cpu_possible_map;
57 static struct task_struct *current_set[NR_CPUS];
59 EXPORT_SYMBOL(cpu_online_map);
62 * Reboot, halt and power_off routines for SMP.
64 extern char vmhalt_cmd[];
65 extern char vmpoff_cmd[];
67 extern void reipl(unsigned long devno);
68 extern void reipl_diag(void);
70 static void smp_ext_bitcall(int, ec_bit_sig);
71 static void smp_ext_bitcall_others(ec_bit_sig);
74 * Structure and data for smp_call_function(). This is designed to minimise
75 * static memory requirements. It also looks cleaner.
77 static DEFINE_SPINLOCK(call_lock);
79 struct call_data_struct {
80 void (*func) (void *info);
87 static struct call_data_struct * call_data;
90 * 'Call function' interrupt callback
92 static void do_call_function(void)
94 void (*func) (void *info) = call_data->func;
95 void *info = call_data->info;
96 int wait = call_data->wait;
98 atomic_inc(&call_data->started);
101 atomic_inc(&call_data->finished);
105 * this function sends a 'generic call function' IPI to all other CPUs
109 int smp_call_function (void (*func) (void *info), void *info, int nonatomic,
112 * [SUMMARY] Run a function on all other CPUs.
113 * <func> The function to run. This must be fast and non-blocking.
114 * <info> An arbitrary pointer to pass to the function.
115 * <nonatomic> currently unused.
116 * <wait> If true, wait (atomically) until function has completed on other CPUs.
117 * [RETURNS] 0 on success, else a negative status code. Does not return until
118 * remote CPUs are nearly ready to execute <<func>> or are or have executed.
120 * You must not call this function with disabled interrupts or from a
121 * hardware interrupt handler or from a bottom half handler.
124 struct call_data_struct data;
125 int cpus = num_online_cpus()-1;
130 /* Can deadlock when called with interrupts disabled */
131 WARN_ON(irqs_disabled());
135 atomic_set(&data.started, 0);
138 atomic_set(&data.finished, 0);
140 spin_lock(&call_lock);
142 /* Send a message to all other CPUs and wait for them to respond */
143 smp_ext_bitcall_others(ec_call_function);
145 /* Wait for response */
146 while (atomic_read(&data.started) != cpus)
150 while (atomic_read(&data.finished) != cpus)
152 spin_unlock(&call_lock);
158 * Call a function on one CPU
159 * cpu : the CPU the function should be executed on
161 * You must not call this function with disabled interrupts or from a
162 * hardware interrupt handler. You may call it from a bottom half.
164 * It is guaranteed that the called function runs on the specified CPU,
165 * preemption is disabled.
167 int smp_call_function_on(void (*func) (void *info), void *info,
168 int nonatomic, int wait, int cpu)
170 struct call_data_struct data;
173 if (!cpu_online(cpu))
176 /* disable preemption for local function call */
177 curr_cpu = get_cpu();
179 if (curr_cpu == cpu) {
180 /* direct call to function */
188 atomic_set(&data.started, 0);
191 atomic_set(&data.finished, 0);
193 spin_lock_bh(&call_lock);
195 smp_ext_bitcall(cpu, ec_call_function);
197 /* Wait for response */
198 while (atomic_read(&data.started) != 1)
202 while (atomic_read(&data.finished) != 1)
205 spin_unlock_bh(&call_lock);
209 EXPORT_SYMBOL(smp_call_function_on);
211 static inline void do_send_stop(void)
215 /* stop all processors */
216 for_each_online_cpu(cpu) {
217 if (cpu == smp_processor_id())
220 rc = signal_processor(cpu, sigp_stop);
221 } while (rc == sigp_busy);
225 static inline void do_store_status(void)
229 /* store status of all processors in their lowcores (real 0) */
230 for_each_online_cpu(cpu) {
231 if (cpu == smp_processor_id())
234 rc = signal_processor_p(
235 (__u32)(unsigned long) lowcore_ptr[cpu], cpu,
236 sigp_store_status_at_address);
237 } while(rc == sigp_busy);
242 * this function sends a 'stop' sigp to all other CPUs in the system.
243 * it goes straight through.
245 void smp_send_stop(void)
247 /* write magic number to zero page (absolute 0) */
248 lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
250 /* stop other processors. */
253 /* store status of other processors. */
258 * Reboot, halt and power_off routines for SMP.
261 static void do_machine_restart(void * __unused)
264 static atomic_t cpuid = ATOMIC_INIT(-1);
266 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) != -1)
267 signal_processor(smp_processor_id(), sigp_stop);
269 /* Wait for all other cpus to enter stopped state */
270 for_each_online_cpu(cpu) {
271 if (cpu == smp_processor_id())
273 while(!smp_cpu_not_running(cpu))
277 /* Store status of other cpus. */
281 * Finally call reipl. Because we waited for all other
282 * cpus to enter this function we know that they do
283 * not hold any s390irq-locks (the cpus have been
284 * interrupted by an external interrupt and s390irq
285 * locks are always held disabled).
290 cpcmd ("IPL", NULL, 0, NULL);
292 reipl (0x10000 | S390_lowcore.ipl_device);
295 void machine_restart_smp(char * __unused)
297 on_each_cpu(do_machine_restart, NULL, 0, 0);
300 static void do_wait_for_stop(void)
302 unsigned long cr[16];
304 __ctl_store(cr, 0, 15);
307 __ctl_load(cr, 0, 15);
312 static void do_machine_halt(void * __unused)
314 static atomic_t cpuid = ATOMIC_INIT(-1);
316 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) == -1) {
318 if (MACHINE_IS_VM && strlen(vmhalt_cmd) > 0)
319 cpcmd(vmhalt_cmd, NULL, 0, NULL);
320 signal_processor(smp_processor_id(),
321 sigp_stop_and_store_status);
326 void machine_halt_smp(void)
328 on_each_cpu(do_machine_halt, NULL, 0, 0);
331 static void do_machine_power_off(void * __unused)
333 static atomic_t cpuid = ATOMIC_INIT(-1);
335 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) == -1) {
337 if (MACHINE_IS_VM && strlen(vmpoff_cmd) > 0)
338 cpcmd(vmpoff_cmd, NULL, 0, NULL);
339 signal_processor(smp_processor_id(),
340 sigp_stop_and_store_status);
345 void machine_power_off_smp(void)
347 on_each_cpu(do_machine_power_off, NULL, 0, 0);
351 * This is the main routine where commands issued by other
355 void do_ext_call_interrupt(struct pt_regs *regs, __u16 code)
360 * handle bit signal external calls
362 * For the ec_schedule signal we have to do nothing. All the work
363 * is done automatically when we return from the interrupt.
365 bits = xchg(&S390_lowcore.ext_call_fast, 0);
367 if (test_bit(ec_call_function, &bits))
372 * Send an external call sigp to another cpu and return without waiting
373 * for its completion.
375 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
378 * Set signaling bit in lowcore of target cpu and kick it
380 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
381 while(signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
386 * Send an external call sigp to every other cpu in the system and
387 * return without waiting for its completion.
389 static void smp_ext_bitcall_others(ec_bit_sig sig)
393 for_each_online_cpu(cpu) {
394 if (cpu == smp_processor_id())
397 * Set signaling bit in lowcore of target cpu and kick it
399 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
400 while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
407 * this function sends a 'purge tlb' signal to another CPU.
409 void smp_ptlb_callback(void *info)
414 void smp_ptlb_all(void)
416 on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
418 EXPORT_SYMBOL(smp_ptlb_all);
419 #endif /* ! CONFIG_64BIT */
422 * this function sends a 'reschedule' IPI to another CPU.
423 * it goes straight through and wastes no time serializing
424 * anything. Worst case is that we lose a reschedule ...
426 void smp_send_reschedule(int cpu)
428 smp_ext_bitcall(cpu, ec_schedule);
432 * parameter area for the set/clear control bit callbacks
438 unsigned long orvals[16];
439 unsigned long andvals[16];
440 } ec_creg_mask_parms;
443 * callback for setting/clearing control bits
445 void smp_ctl_bit_callback(void *info) {
446 ec_creg_mask_parms *pp;
447 unsigned long cregs[16];
450 pp = (ec_creg_mask_parms *) info;
451 __ctl_store(cregs[pp->start_ctl], pp->start_ctl, pp->end_ctl);
452 for (i = pp->start_ctl; i <= pp->end_ctl; i++)
453 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
454 __ctl_load(cregs[pp->start_ctl], pp->start_ctl, pp->end_ctl);
458 * Set a bit in a control register of all cpus
460 void smp_ctl_set_bit(int cr, int bit) {
461 ec_creg_mask_parms parms;
463 parms.start_ctl = cr;
465 parms.orvals[cr] = 1 << bit;
466 parms.andvals[cr] = -1L;
468 smp_call_function(smp_ctl_bit_callback, &parms, 0, 1);
469 __ctl_set_bit(cr, bit);
474 * Clear a bit in a control register of all cpus
476 void smp_ctl_clear_bit(int cr, int bit) {
477 ec_creg_mask_parms parms;
479 parms.start_ctl = cr;
481 parms.orvals[cr] = 0;
482 parms.andvals[cr] = ~(1L << bit);
484 smp_call_function(smp_ctl_bit_callback, &parms, 0, 1);
485 __ctl_clear_bit(cr, bit);
490 * Lets check how many CPUs we have.
494 __init smp_check_cpus(unsigned int max_cpus)
500 * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
503 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
504 current_thread_info()->cpu = 0;
506 for (cpu = 0; cpu <= 65535 && num_cpus < max_cpus; cpu++) {
507 if ((__u16) cpu == boot_cpu_addr)
509 __cpu_logical_map[num_cpus] = (__u16) cpu;
510 if (signal_processor(num_cpus, sigp_sense) ==
511 sigp_not_operational)
513 cpu_set(num_cpus, cpu_present_map);
517 for (cpu = 1; cpu < max_cpus; cpu++)
518 cpu_set(cpu, cpu_possible_map);
520 printk("Detected %d CPU's\n",(int) num_cpus);
521 printk("Boot cpu address %2X\n", boot_cpu_addr);
525 * Activate a secondary processor.
527 extern void init_cpu_timer(void);
528 extern void init_cpu_vtimer(void);
529 extern int pfault_init(void);
530 extern void pfault_fini(void);
532 int __devinit start_secondary(void *cpuvoid)
537 /* init per CPU timer */
539 #ifdef CONFIG_VIRT_TIMER
543 /* Enable pfault pseudo page faults on this cpu. */
547 /* Mark this cpu as online */
548 cpu_set(smp_processor_id(), cpu_online_map);
549 /* Switch on interrupts */
551 /* Print info about this processor */
552 print_cpu_info(&S390_lowcore.cpu_data);
553 /* cpu_idle will call schedule for us */
558 static void __init smp_create_idle(unsigned int cpu)
560 struct task_struct *p;
563 * don't care about the psw and regs settings since we'll never
564 * reschedule the forked task.
568 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
569 current_set[cpu] = p;
572 /* Reserving and releasing of CPUs */
574 static DEFINE_SPINLOCK(smp_reserve_lock);
575 static int smp_cpu_reserved[NR_CPUS];
578 smp_get_cpu(cpumask_t cpu_mask)
583 spin_lock_irqsave(&smp_reserve_lock, flags);
584 /* Try to find an already reserved cpu. */
585 for_each_cpu_mask(cpu, cpu_mask) {
586 if (smp_cpu_reserved[cpu] != 0) {
587 smp_cpu_reserved[cpu]++;
592 /* Reserve a new cpu from cpu_mask. */
593 for_each_cpu_mask(cpu, cpu_mask) {
594 if (cpu_online(cpu)) {
595 smp_cpu_reserved[cpu]++;
601 spin_unlock_irqrestore(&smp_reserve_lock, flags);
610 spin_lock_irqsave(&smp_reserve_lock, flags);
611 smp_cpu_reserved[cpu]--;
612 spin_unlock_irqrestore(&smp_reserve_lock, flags);
620 /* Check for stopped state */
621 if (signal_processor_ps(&status, 0, cpu, sigp_sense) == sigp_status_stored) {
628 /* Upping and downing of CPUs */
631 __cpu_up(unsigned int cpu)
633 struct task_struct *idle;
634 struct _lowcore *cpu_lowcore;
635 struct stack_frame *sf;
639 for (curr_cpu = 0; curr_cpu <= 65535; curr_cpu++) {
640 __cpu_logical_map[cpu] = (__u16) curr_cpu;
641 if (cpu_stopped(cpu))
645 if (!cpu_stopped(cpu))
648 ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
649 cpu, sigp_set_prefix);
651 printk("sigp_set_prefix failed for cpu %d "
652 "with condition code %d\n",
653 (int) cpu, (int) ccode);
657 idle = current_set[cpu];
658 cpu_lowcore = lowcore_ptr[cpu];
659 cpu_lowcore->kernel_stack = (unsigned long)
660 task_stack_page(idle) + (THREAD_SIZE);
661 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
662 - sizeof(struct pt_regs)
663 - sizeof(struct stack_frame));
664 memset(sf, 0, sizeof(struct stack_frame));
665 sf->gprs[9] = (unsigned long) sf;
666 cpu_lowcore->save_area[15] = (unsigned long) sf;
667 __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
668 __asm__ __volatile__("stam 0,15,0(%0)"
669 : : "a" (&cpu_lowcore->access_regs_save_area)
671 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
672 cpu_lowcore->current_task = (unsigned long) idle;
673 cpu_lowcore->cpu_data.cpu_nr = cpu;
675 signal_processor(cpu,sigp_restart);
677 while (!cpu_online(cpu))
686 ec_creg_mask_parms cr_parms;
687 int cpu = smp_processor_id();
689 spin_lock_irqsave(&smp_reserve_lock, flags);
690 if (smp_cpu_reserved[cpu] != 0) {
691 spin_unlock_irqrestore(&smp_reserve_lock, flags);
694 cpu_clear(cpu, cpu_online_map);
697 /* Disable pfault pseudo page faults on this cpu. */
702 /* disable all external interrupts */
704 cr_parms.start_ctl = 0;
705 cr_parms.end_ctl = 0;
706 cr_parms.orvals[0] = 0;
707 cr_parms.andvals[0] = ~(1<<15 | 1<<14 | 1<<13 | 1<<12 |
708 1<<11 | 1<<10 | 1<< 6 | 1<< 4);
709 smp_ctl_bit_callback(&cr_parms);
711 /* disable all I/O interrupts */
713 cr_parms.start_ctl = 6;
714 cr_parms.end_ctl = 6;
715 cr_parms.orvals[6] = 0;
716 cr_parms.andvals[6] = ~(1<<31 | 1<<30 | 1<<29 | 1<<28 |
717 1<<27 | 1<<26 | 1<<25 | 1<<24);
718 smp_ctl_bit_callback(&cr_parms);
720 /* disable most machine checks */
722 cr_parms.start_ctl = 14;
723 cr_parms.end_ctl = 14;
724 cr_parms.orvals[14] = 0;
725 cr_parms.andvals[14] = ~(1<<28 | 1<<27 | 1<<26 | 1<<25 | 1<<24);
726 smp_ctl_bit_callback(&cr_parms);
728 spin_unlock_irqrestore(&smp_reserve_lock, flags);
733 __cpu_die(unsigned int cpu)
735 /* Wait until target cpu is down */
736 while (!smp_cpu_not_running(cpu))
738 printk("Processor %d spun down\n", cpu);
745 signal_processor(smp_processor_id(), sigp_stop);
751 * Cycle through the processors and setup structures.
754 void __init smp_prepare_cpus(unsigned int max_cpus)
760 /* request the 0x1201 emergency signal external interrupt */
761 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
762 panic("Couldn't request external interrupt 0x1201");
763 smp_check_cpus(max_cpus);
764 memset(lowcore_ptr,0,sizeof(lowcore_ptr));
766 * Initialize prefix pages and stacks for all possible cpus
768 print_cpu_info(&S390_lowcore.cpu_data);
770 for(i = 0; i < NR_CPUS; i++) {
771 if (!cpu_possible(i))
773 lowcore_ptr[i] = (struct _lowcore *)
774 __get_free_pages(GFP_KERNEL|GFP_DMA,
775 sizeof(void*) == 8 ? 1 : 0);
776 stack = __get_free_pages(GFP_KERNEL,ASYNC_ORDER);
777 if (lowcore_ptr[i] == NULL || stack == 0ULL)
778 panic("smp_boot_cpus failed to allocate memory\n");
780 *(lowcore_ptr[i]) = S390_lowcore;
781 lowcore_ptr[i]->async_stack = stack + (ASYNC_SIZE);
782 stack = __get_free_pages(GFP_KERNEL,0);
784 panic("smp_boot_cpus failed to allocate memory\n");
785 lowcore_ptr[i]->panic_stack = stack + (PAGE_SIZE);
787 if (MACHINE_HAS_IEEE) {
788 lowcore_ptr[i]->extended_save_area_addr =
789 (__u32) __get_free_pages(GFP_KERNEL,0);
790 if (lowcore_ptr[i]->extended_save_area_addr == 0)
791 panic("smp_boot_cpus failed to "
792 "allocate memory\n");
797 if (MACHINE_HAS_IEEE)
798 ctl_set_bit(14, 29); /* enable extended save area */
800 set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
803 if (cpu != smp_processor_id())
804 smp_create_idle(cpu);
807 void __devinit smp_prepare_boot_cpu(void)
809 BUG_ON(smp_processor_id() != 0);
811 cpu_set(0, cpu_online_map);
812 cpu_set(0, cpu_present_map);
813 cpu_set(0, cpu_possible_map);
814 S390_lowcore.percpu_offset = __per_cpu_offset[0];
815 current_set[0] = current;
818 void smp_cpus_done(unsigned int max_cpus)
820 cpu_present_map = cpu_possible_map;
824 * the frequency of the profiling timer can be changed
825 * by writing a multiplier value into /proc/profile.
827 * usually you want to run this on all CPUs ;)
829 int setup_profiling_timer(unsigned int multiplier)
834 static DEFINE_PER_CPU(struct cpu, cpu_devices);
836 static int __init topology_init(void)
842 ret = register_cpu(&per_cpu(cpu_devices, cpu), cpu, NULL);
844 printk(KERN_WARNING "topology_init: register_cpu %d "
845 "failed (%d)\n", cpu, ret);
850 subsys_initcall(topology_init);
852 EXPORT_SYMBOL(cpu_possible_map);
853 EXPORT_SYMBOL(lowcore_ptr);
854 EXPORT_SYMBOL(smp_ctl_set_bit);
855 EXPORT_SYMBOL(smp_ctl_clear_bit);
856 EXPORT_SYMBOL(smp_call_function);
857 EXPORT_SYMBOL(smp_get_cpu);
858 EXPORT_SYMBOL(smp_put_cpu);