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];
53 cycles_t cacheflush_time=0;
54 int smp_threads_ready=0; /* Set when the idlers are all forked. */
56 cpumask_t cpu_online_map;
57 cpumask_t cpu_possible_map;
58 unsigned long cache_decay_ticks = 0;
60 static struct task_struct *current_set[NR_CPUS];
62 EXPORT_SYMBOL(cpu_online_map);
65 * Reboot, halt and power_off routines for SMP.
67 extern char vmhalt_cmd[];
68 extern char vmpoff_cmd[];
70 extern void reipl(unsigned long devno);
72 static void smp_ext_bitcall(int, ec_bit_sig);
73 static void smp_ext_bitcall_others(ec_bit_sig);
76 * Structure and data for smp_call_function(). This is designed to minimise
77 * static memory requirements. It also looks cleaner.
79 static DEFINE_SPINLOCK(call_lock);
81 struct call_data_struct {
82 void (*func) (void *info);
89 static struct call_data_struct * call_data;
92 * 'Call function' interrupt callback
94 static void do_call_function(void)
96 void (*func) (void *info) = call_data->func;
97 void *info = call_data->info;
98 int wait = call_data->wait;
100 atomic_inc(&call_data->started);
103 atomic_inc(&call_data->finished);
107 * this function sends a 'generic call function' IPI to all other CPUs
111 int smp_call_function (void (*func) (void *info), void *info, int nonatomic,
114 * [SUMMARY] Run a function on all other CPUs.
115 * <func> The function to run. This must be fast and non-blocking.
116 * <info> An arbitrary pointer to pass to the function.
117 * <nonatomic> currently unused.
118 * <wait> If true, wait (atomically) until function has completed on other CPUs.
119 * [RETURNS] 0 on success, else a negative status code. Does not return until
120 * remote CPUs are nearly ready to execute <<func>> or are or have executed.
122 * You must not call this function with disabled interrupts or from a
123 * hardware interrupt handler or from a bottom half handler.
126 struct call_data_struct data;
127 int cpus = num_online_cpus()-1;
132 /* Can deadlock when called with interrupts disabled */
133 WARN_ON(irqs_disabled());
137 atomic_set(&data.started, 0);
140 atomic_set(&data.finished, 0);
142 spin_lock(&call_lock);
144 /* Send a message to all other CPUs and wait for them to respond */
145 smp_ext_bitcall_others(ec_call_function);
147 /* Wait for response */
148 while (atomic_read(&data.started) != cpus)
152 while (atomic_read(&data.finished) != cpus)
154 spin_unlock(&call_lock);
160 * Call a function on one CPU
161 * cpu : the CPU the function should be executed on
163 * You must not call this function with disabled interrupts or from a
164 * hardware interrupt handler. You may call it from a bottom half.
166 * It is guaranteed that the called function runs on the specified CPU,
167 * preemption is disabled.
169 int smp_call_function_on(void (*func) (void *info), void *info,
170 int nonatomic, int wait, int cpu)
172 struct call_data_struct data;
175 if (!cpu_online(cpu))
178 /* disable preemption for local function call */
179 curr_cpu = get_cpu();
181 if (curr_cpu == cpu) {
182 /* direct call to function */
190 atomic_set(&data.started, 0);
193 atomic_set(&data.finished, 0);
195 spin_lock_bh(&call_lock);
197 smp_ext_bitcall(cpu, ec_call_function);
199 /* Wait for response */
200 while (atomic_read(&data.started) != 1)
204 while (atomic_read(&data.finished) != 1)
207 spin_unlock_bh(&call_lock);
211 EXPORT_SYMBOL(smp_call_function_on);
213 static inline void do_send_stop(void)
217 /* stop all processors */
218 for (i = 0; i < NR_CPUS; i++) {
219 if (!cpu_online(i) || smp_processor_id() == i)
222 rc = signal_processor(i, sigp_stop);
223 } while (rc == sigp_busy);
227 static inline void do_store_status(void)
231 /* store status of all processors in their lowcores (real 0) */
232 for (i = 0; i < NR_CPUS; i++) {
233 if (!cpu_online(i) || smp_processor_id() == i)
236 rc = signal_processor_p(
237 (__u32)(unsigned long) lowcore_ptr[i], i,
238 sigp_store_status_at_address);
239 } while(rc == sigp_busy);
244 * this function sends a 'stop' sigp to all other CPUs in the system.
245 * it goes straight through.
247 void smp_send_stop(void)
249 /* write magic number to zero page (absolute 0) */
250 lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
252 /* stop other processors. */
255 /* store status of other processors. */
260 * Reboot, halt and power_off routines for SMP.
262 static cpumask_t cpu_restart_map;
264 static void do_machine_restart(void * __unused)
266 static atomic_t cpuid = ATOMIC_INIT(-1);
268 cpu_clear(smp_processor_id(), cpu_restart_map);
269 if (atomic_compare_and_swap(-1, smp_processor_id(), &cpuid) == 0) {
270 /* Wait for all other cpus to enter do_machine_restart. */
271 while (!cpus_empty(cpu_restart_map))
273 /* Store status of other cpus. */
276 * Finally call reipl. Because we waited for all other
277 * cpus to enter this function we know that they do
278 * not hold any s390irq-locks (the cpus have been
279 * interrupted by an external interrupt and s390irq
280 * locks are always held disabled).
283 cpcmd ("IPL", NULL, 0);
285 reipl (0x10000 | S390_lowcore.ipl_device);
287 signal_processor(smp_processor_id(), sigp_stop);
290 void machine_restart_smp(char * __unused)
292 cpu_restart_map = cpu_online_map;
293 on_each_cpu(do_machine_restart, NULL, 0, 0);
296 static void do_wait_for_stop(void)
298 unsigned long cr[16];
300 __ctl_store(cr, 0, 15);
303 __ctl_load(cr, 0, 15);
308 static void do_machine_halt(void * __unused)
310 static atomic_t cpuid = ATOMIC_INIT(-1);
312 if (atomic_compare_and_swap(-1, smp_processor_id(), &cpuid) == 0) {
314 if (MACHINE_IS_VM && strlen(vmhalt_cmd) > 0)
315 cpcmd(vmhalt_cmd, NULL, 0);
316 signal_processor(smp_processor_id(),
317 sigp_stop_and_store_status);
322 void machine_halt_smp(void)
324 on_each_cpu(do_machine_halt, NULL, 0, 0);
327 static void do_machine_power_off(void * __unused)
329 static atomic_t cpuid = ATOMIC_INIT(-1);
331 if (atomic_compare_and_swap(-1, smp_processor_id(), &cpuid) == 0) {
333 if (MACHINE_IS_VM && strlen(vmpoff_cmd) > 0)
334 cpcmd(vmpoff_cmd, NULL, 0);
335 signal_processor(smp_processor_id(),
336 sigp_stop_and_store_status);
341 void machine_power_off_smp(void)
343 on_each_cpu(do_machine_power_off, NULL, 0, 0);
347 * This is the main routine where commands issued by other
351 void do_ext_call_interrupt(struct pt_regs *regs, __u16 code)
356 * handle bit signal external calls
358 * For the ec_schedule signal we have to do nothing. All the work
359 * is done automatically when we return from the interrupt.
361 bits = xchg(&S390_lowcore.ext_call_fast, 0);
363 if (test_bit(ec_call_function, &bits))
368 * Send an external call sigp to another cpu and return without waiting
369 * for its completion.
371 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
374 * Set signaling bit in lowcore of target cpu and kick it
376 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
377 while(signal_processor(cpu, sigp_external_call) == sigp_busy)
382 * Send an external call sigp to every other cpu in the system and
383 * return without waiting for its completion.
385 static void smp_ext_bitcall_others(ec_bit_sig sig)
389 for (i = 0; i < NR_CPUS; i++) {
390 if (!cpu_online(i) || smp_processor_id() == i)
393 * Set signaling bit in lowcore of target cpu and kick it
395 set_bit(sig, (unsigned long *) &lowcore_ptr[i]->ext_call_fast);
396 while (signal_processor(i, sigp_external_call) == sigp_busy)
401 #ifndef CONFIG_ARCH_S390X
403 * this function sends a 'purge tlb' signal to another CPU.
405 void smp_ptlb_callback(void *info)
410 void smp_ptlb_all(void)
412 on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
414 EXPORT_SYMBOL(smp_ptlb_all);
415 #endif /* ! CONFIG_ARCH_S390X */
418 * this function sends a 'reschedule' IPI to another CPU.
419 * it goes straight through and wastes no time serializing
420 * anything. Worst case is that we lose a reschedule ...
422 void smp_send_reschedule(int cpu)
424 smp_ext_bitcall(cpu, ec_schedule);
428 * parameter area for the set/clear control bit callbacks
434 unsigned long orvals[16];
435 unsigned long andvals[16];
436 } ec_creg_mask_parms;
439 * callback for setting/clearing control bits
441 void smp_ctl_bit_callback(void *info) {
442 ec_creg_mask_parms *pp;
443 unsigned long cregs[16];
446 pp = (ec_creg_mask_parms *) info;
447 __ctl_store(cregs[pp->start_ctl], pp->start_ctl, pp->end_ctl);
448 for (i = pp->start_ctl; i <= pp->end_ctl; i++)
449 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
450 __ctl_load(cregs[pp->start_ctl], pp->start_ctl, pp->end_ctl);
454 * Set a bit in a control register of all cpus
456 void smp_ctl_set_bit(int cr, int bit) {
457 ec_creg_mask_parms parms;
459 parms.start_ctl = cr;
461 parms.orvals[cr] = 1 << bit;
462 parms.andvals[cr] = -1L;
464 smp_call_function(smp_ctl_bit_callback, &parms, 0, 1);
465 __ctl_set_bit(cr, bit);
470 * Clear a bit in a control register of all cpus
472 void smp_ctl_clear_bit(int cr, int bit) {
473 ec_creg_mask_parms parms;
475 parms.start_ctl = cr;
477 parms.orvals[cr] = 0;
478 parms.andvals[cr] = ~(1L << bit);
480 smp_call_function(smp_ctl_bit_callback, &parms, 0, 1);
481 __ctl_clear_bit(cr, bit);
486 * Lets check how many CPUs we have.
490 __init smp_check_cpus(unsigned int max_cpus)
496 * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
499 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
500 __cpu_logical_map[0] = boot_cpu_addr;
501 current_thread_info()->cpu = 0;
503 for (cpu = 0; cpu <= 65535 && num_cpus < max_cpus; cpu++) {
504 if ((__u16) cpu == boot_cpu_addr)
506 __cpu_logical_map[num_cpus] = (__u16) cpu;
507 if (signal_processor(num_cpus, sigp_sense) ==
508 sigp_not_operational)
510 cpu_set(num_cpus, cpu_present_map);
514 for (cpu = 1; cpu < max_cpus; cpu++)
515 cpu_set(cpu, cpu_possible_map);
517 printk("Detected %d CPU's\n",(int) num_cpus);
518 printk("Boot cpu address %2X\n", boot_cpu_addr);
522 * Activate a secondary processor.
524 extern void init_cpu_timer(void);
525 extern void init_cpu_vtimer(void);
526 extern int pfault_init(void);
527 extern void pfault_fini(void);
529 int __devinit start_secondary(void *cpuvoid)
533 /* init per CPU timer */
535 #ifdef CONFIG_VIRT_TIMER
539 /* Enable pfault pseudo page faults on this cpu. */
542 /* Mark this cpu as online */
543 cpu_set(smp_processor_id(), cpu_online_map);
544 /* Switch on interrupts */
546 /* Print info about this processor */
547 print_cpu_info(&S390_lowcore.cpu_data);
548 /* cpu_idle will call schedule for us */
553 static void __init smp_create_idle(unsigned int cpu)
555 struct task_struct *p;
558 * don't care about the psw and regs settings since we'll never
559 * reschedule the forked task.
563 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
564 current_set[cpu] = p;
567 /* Reserving and releasing of CPUs */
569 static DEFINE_SPINLOCK(smp_reserve_lock);
570 static int smp_cpu_reserved[NR_CPUS];
573 smp_get_cpu(cpumask_t cpu_mask)
578 spin_lock_irqsave(&smp_reserve_lock, flags);
579 /* Try to find an already reserved cpu. */
580 for_each_cpu_mask(cpu, cpu_mask) {
581 if (smp_cpu_reserved[cpu] != 0) {
582 smp_cpu_reserved[cpu]++;
587 /* Reserve a new cpu from cpu_mask. */
588 for_each_cpu_mask(cpu, cpu_mask) {
589 if (cpu_online(cpu)) {
590 smp_cpu_reserved[cpu]++;
596 spin_unlock_irqrestore(&smp_reserve_lock, flags);
605 spin_lock_irqsave(&smp_reserve_lock, flags);
606 smp_cpu_reserved[cpu]--;
607 spin_unlock_irqrestore(&smp_reserve_lock, flags);
615 /* Check for stopped state */
616 if (signal_processor_ps(&status, 0, cpu, sigp_sense) == sigp_status_stored) {
623 /* Upping and downing of CPUs */
626 __cpu_up(unsigned int cpu)
628 struct task_struct *idle;
629 struct _lowcore *cpu_lowcore;
630 struct stack_frame *sf;
634 for (curr_cpu = 0; curr_cpu <= 65535; curr_cpu++) {
635 __cpu_logical_map[cpu] = (__u16) curr_cpu;
636 if (cpu_stopped(cpu))
640 if (!cpu_stopped(cpu))
643 ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
644 cpu, sigp_set_prefix);
646 printk("sigp_set_prefix failed for cpu %d "
647 "with condition code %d\n",
648 (int) cpu, (int) ccode);
652 idle = current_set[cpu];
653 cpu_lowcore = lowcore_ptr[cpu];
654 cpu_lowcore->kernel_stack = (unsigned long)
655 idle->thread_info + (THREAD_SIZE);
656 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
657 - sizeof(struct pt_regs)
658 - sizeof(struct stack_frame));
659 memset(sf, 0, sizeof(struct stack_frame));
660 sf->gprs[9] = (unsigned long) sf;
661 cpu_lowcore->save_area[15] = (unsigned long) sf;
662 __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
663 __asm__ __volatile__("stam 0,15,0(%0)"
664 : : "a" (&cpu_lowcore->access_regs_save_area)
666 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
667 cpu_lowcore->current_task = (unsigned long) idle;
668 cpu_lowcore->cpu_data.cpu_nr = cpu;
670 signal_processor(cpu,sigp_restart);
672 while (!cpu_online(cpu))
681 ec_creg_mask_parms cr_parms;
683 spin_lock_irqsave(&smp_reserve_lock, flags);
684 if (smp_cpu_reserved[smp_processor_id()] != 0) {
685 spin_unlock_irqrestore(&smp_reserve_lock, flags);
690 /* Disable pfault pseudo page faults on this cpu. */
694 /* disable all external interrupts */
696 cr_parms.start_ctl = 0;
697 cr_parms.end_ctl = 0;
698 cr_parms.orvals[0] = 0;
699 cr_parms.andvals[0] = ~(1<<15 | 1<<14 | 1<<13 | 1<<12 |
700 1<<11 | 1<<10 | 1<< 6 | 1<< 4);
701 smp_ctl_bit_callback(&cr_parms);
703 /* disable all I/O interrupts */
705 cr_parms.start_ctl = 6;
706 cr_parms.end_ctl = 6;
707 cr_parms.orvals[6] = 0;
708 cr_parms.andvals[6] = ~(1<<31 | 1<<30 | 1<<29 | 1<<28 |
709 1<<27 | 1<<26 | 1<<25 | 1<<24);
710 smp_ctl_bit_callback(&cr_parms);
712 /* disable most machine checks */
714 cr_parms.start_ctl = 14;
715 cr_parms.end_ctl = 14;
716 cr_parms.orvals[14] = 0;
717 cr_parms.andvals[14] = ~(1<<28 | 1<<27 | 1<<26 | 1<<25 | 1<<24);
718 smp_ctl_bit_callback(&cr_parms);
720 spin_unlock_irqrestore(&smp_reserve_lock, flags);
725 __cpu_die(unsigned int cpu)
727 /* Wait until target cpu is down */
728 while (!cpu_stopped(cpu))
730 printk("Processor %d spun down\n", cpu);
737 signal_processor(smp_processor_id(), sigp_stop);
743 * Cycle through the processors and setup structures.
746 void __init smp_prepare_cpus(unsigned int max_cpus)
752 /* request the 0x1202 external interrupt */
753 if (register_external_interrupt(0x1202, do_ext_call_interrupt) != 0)
754 panic("Couldn't request external interrupt 0x1202");
755 smp_check_cpus(max_cpus);
756 memset(lowcore_ptr,0,sizeof(lowcore_ptr));
758 * Initialize prefix pages and stacks for all possible cpus
760 print_cpu_info(&S390_lowcore.cpu_data);
762 for(i = 0; i < NR_CPUS; i++) {
763 if (!cpu_possible(i))
765 lowcore_ptr[i] = (struct _lowcore *)
766 __get_free_pages(GFP_KERNEL|GFP_DMA,
767 sizeof(void*) == 8 ? 1 : 0);
768 stack = __get_free_pages(GFP_KERNEL,ASYNC_ORDER);
769 if (lowcore_ptr[i] == NULL || stack == 0ULL)
770 panic("smp_boot_cpus failed to allocate memory\n");
772 *(lowcore_ptr[i]) = S390_lowcore;
773 lowcore_ptr[i]->async_stack = stack + (ASYNC_SIZE);
774 #ifdef CONFIG_CHECK_STACK
775 stack = __get_free_pages(GFP_KERNEL,0);
777 panic("smp_boot_cpus failed to allocate memory\n");
778 lowcore_ptr[i]->panic_stack = stack + (PAGE_SIZE);
781 set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
784 if (cpu != smp_processor_id())
785 smp_create_idle(cpu);
788 void __devinit smp_prepare_boot_cpu(void)
790 BUG_ON(smp_processor_id() != 0);
792 cpu_set(0, cpu_online_map);
793 cpu_set(0, cpu_possible_map);
794 S390_lowcore.percpu_offset = __per_cpu_offset[0];
795 current_set[0] = current;
798 void smp_cpus_done(unsigned int max_cpus)
800 cpu_present_map = cpu_possible_map;
804 * the frequency of the profiling timer can be changed
805 * by writing a multiplier value into /proc/profile.
807 * usually you want to run this on all CPUs ;)
809 int setup_profiling_timer(unsigned int multiplier)
814 static DEFINE_PER_CPU(struct cpu, cpu_devices);
816 static int __init topology_init(void)
822 ret = register_cpu(&per_cpu(cpu_devices, cpu), cpu, NULL);
824 printk(KERN_WARNING "topology_init: register_cpu %d "
825 "failed (%d)\n", cpu, ret);
830 subsys_initcall(topology_init);
832 EXPORT_SYMBOL(cpu_possible_map);
833 EXPORT_SYMBOL(lowcore_ptr);
834 EXPORT_SYMBOL(smp_ctl_set_bit);
835 EXPORT_SYMBOL(smp_ctl_clear_bit);
836 EXPORT_SYMBOL(smp_call_function);
837 EXPORT_SYMBOL(smp_get_cpu);
838 EXPORT_SYMBOL(smp_put_cpu);