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>
27 #include <linux/spinlock.h>
28 #include <linux/kernel_stat.h>
29 #include <linux/smp_lock.h>
31 #include <linux/delay.h>
32 #include <linux/cache.h>
33 #include <linux/interrupt.h>
34 #include <linux/cpu.h>
37 #include <asm/pgalloc.h>
39 #include <asm/s390_ext.h>
40 #include <asm/cpcmd.h>
41 #include <asm/tlbflush.h>
43 extern volatile int __cpu_logical_map[];
46 * An array with a pointer the lowcore of every CPU.
49 struct _lowcore *lowcore_ptr[NR_CPUS];
51 cpumask_t cpu_online_map = CPU_MASK_NONE;
52 cpumask_t cpu_possible_map = CPU_MASK_NONE;
54 static struct task_struct *current_set[NR_CPUS];
57 * Reboot, halt and power_off routines for SMP.
59 extern char vmhalt_cmd[];
60 extern char vmpoff_cmd[];
62 extern void reipl(unsigned long devno);
63 extern void reipl_diag(void);
65 static void smp_ext_bitcall(int, ec_bit_sig);
66 static void smp_ext_bitcall_others(ec_bit_sig);
69 * Structure and data for smp_call_function(). This is designed to minimise
70 * static memory requirements. It also looks cleaner.
72 static DEFINE_SPINLOCK(call_lock);
74 struct call_data_struct {
75 void (*func) (void *info);
82 static struct call_data_struct * call_data;
85 * 'Call function' interrupt callback
87 static void do_call_function(void)
89 void (*func) (void *info) = call_data->func;
90 void *info = call_data->info;
91 int wait = call_data->wait;
93 atomic_inc(&call_data->started);
96 atomic_inc(&call_data->finished);
100 * this function sends a 'generic call function' IPI to all other CPUs
104 int smp_call_function (void (*func) (void *info), void *info, int nonatomic,
107 * [SUMMARY] Run a function on all other CPUs.
108 * <func> The function to run. This must be fast and non-blocking.
109 * <info> An arbitrary pointer to pass to the function.
110 * <nonatomic> currently unused.
111 * <wait> If true, wait (atomically) until function has completed on other CPUs.
112 * [RETURNS] 0 on success, else a negative status code. Does not return until
113 * remote CPUs are nearly ready to execute <<func>> or are or have executed.
115 * You must not call this function with disabled interrupts or from a
116 * hardware interrupt handler or from a bottom half handler.
119 struct call_data_struct data;
120 int cpus = num_online_cpus()-1;
125 /* Can deadlock when called with interrupts disabled */
126 WARN_ON(irqs_disabled());
130 atomic_set(&data.started, 0);
133 atomic_set(&data.finished, 0);
135 spin_lock(&call_lock);
137 /* Send a message to all other CPUs and wait for them to respond */
138 smp_ext_bitcall_others(ec_call_function);
140 /* Wait for response */
141 while (atomic_read(&data.started) != cpus)
145 while (atomic_read(&data.finished) != cpus)
147 spin_unlock(&call_lock);
153 * Call a function on one CPU
154 * cpu : the CPU the function should be executed on
156 * You must not call this function with disabled interrupts or from a
157 * hardware interrupt handler. You may call it from a bottom half.
159 * It is guaranteed that the called function runs on the specified CPU,
160 * preemption is disabled.
162 int smp_call_function_on(void (*func) (void *info), void *info,
163 int nonatomic, int wait, int cpu)
165 struct call_data_struct data;
168 if (!cpu_online(cpu))
171 /* disable preemption for local function call */
172 curr_cpu = get_cpu();
174 if (curr_cpu == cpu) {
175 /* direct call to function */
183 atomic_set(&data.started, 0);
186 atomic_set(&data.finished, 0);
188 spin_lock_bh(&call_lock);
190 smp_ext_bitcall(cpu, ec_call_function);
192 /* Wait for response */
193 while (atomic_read(&data.started) != 1)
197 while (atomic_read(&data.finished) != 1)
200 spin_unlock_bh(&call_lock);
204 EXPORT_SYMBOL(smp_call_function_on);
206 static inline void do_send_stop(void)
210 /* stop all processors */
211 for_each_online_cpu(cpu) {
212 if (cpu == smp_processor_id())
215 rc = signal_processor(cpu, sigp_stop);
216 } while (rc == sigp_busy);
220 static inline void do_store_status(void)
224 /* store status of all processors in their lowcores (real 0) */
225 for_each_online_cpu(cpu) {
226 if (cpu == smp_processor_id())
229 rc = signal_processor_p(
230 (__u32)(unsigned long) lowcore_ptr[cpu], cpu,
231 sigp_store_status_at_address);
232 } while(rc == sigp_busy);
237 * this function sends a 'stop' sigp to all other CPUs in the system.
238 * it goes straight through.
240 void smp_send_stop(void)
242 /* write magic number to zero page (absolute 0) */
243 lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
245 /* stop other processors. */
248 /* store status of other processors. */
253 * Reboot, halt and power_off routines for SMP.
256 static void do_machine_restart(void * __unused)
259 static atomic_t cpuid = ATOMIC_INIT(-1);
261 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) != -1)
262 signal_processor(smp_processor_id(), sigp_stop);
264 /* Wait for all other cpus to enter stopped state */
265 for_each_online_cpu(cpu) {
266 if (cpu == smp_processor_id())
268 while(!smp_cpu_not_running(cpu))
272 /* 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).
285 cpcmd ("IPL", NULL, 0, NULL);
287 reipl (0x10000 | S390_lowcore.ipl_device);
290 void machine_restart_smp(char * __unused)
292 on_each_cpu(do_machine_restart, NULL, 0, 0);
295 static void do_wait_for_stop(void)
297 unsigned long cr[16];
299 __ctl_store(cr, 0, 15);
302 __ctl_load(cr, 0, 15);
307 static void do_machine_halt(void * __unused)
309 static atomic_t cpuid = ATOMIC_INIT(-1);
311 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) == -1) {
313 if (MACHINE_IS_VM && strlen(vmhalt_cmd) > 0)
314 cpcmd(vmhalt_cmd, NULL, 0, NULL);
315 signal_processor(smp_processor_id(),
316 sigp_stop_and_store_status);
321 void machine_halt_smp(void)
323 on_each_cpu(do_machine_halt, NULL, 0, 0);
326 static void do_machine_power_off(void * __unused)
328 static atomic_t cpuid = ATOMIC_INIT(-1);
330 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) == -1) {
332 if (MACHINE_IS_VM && strlen(vmpoff_cmd) > 0)
333 cpcmd(vmpoff_cmd, NULL, 0, NULL);
334 signal_processor(smp_processor_id(),
335 sigp_stop_and_store_status);
340 void machine_power_off_smp(void)
342 on_each_cpu(do_machine_power_off, NULL, 0, 0);
346 * This is the main routine where commands issued by other
350 void do_ext_call_interrupt(struct pt_regs *regs, __u16 code)
355 * handle bit signal external calls
357 * For the ec_schedule signal we have to do nothing. All the work
358 * is done automatically when we return from the interrupt.
360 bits = xchg(&S390_lowcore.ext_call_fast, 0);
362 if (test_bit(ec_call_function, &bits))
367 * Send an external call sigp to another cpu and return without waiting
368 * for its completion.
370 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
373 * Set signaling bit in lowcore of target cpu and kick it
375 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
376 while(signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
381 * Send an external call sigp to every other cpu in the system and
382 * return without waiting for its completion.
384 static void smp_ext_bitcall_others(ec_bit_sig sig)
388 for_each_online_cpu(cpu) {
389 if (cpu == smp_processor_id())
392 * Set signaling bit in lowcore of target cpu and kick it
394 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
395 while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
402 * this function sends a 'purge tlb' signal to another CPU.
404 void smp_ptlb_callback(void *info)
409 void smp_ptlb_all(void)
411 on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
413 EXPORT_SYMBOL(smp_ptlb_all);
414 #endif /* ! CONFIG_64BIT */
417 * this function sends a 'reschedule' IPI to another CPU.
418 * it goes straight through and wastes no time serializing
419 * anything. Worst case is that we lose a reschedule ...
421 void smp_send_reschedule(int cpu)
423 smp_ext_bitcall(cpu, ec_schedule);
427 * parameter area for the set/clear control bit callbacks
433 unsigned long orvals[16];
434 unsigned long andvals[16];
435 } ec_creg_mask_parms;
438 * callback for setting/clearing control bits
440 void smp_ctl_bit_callback(void *info) {
441 ec_creg_mask_parms *pp;
442 unsigned long cregs[16];
445 pp = (ec_creg_mask_parms *) info;
446 __ctl_store(cregs[pp->start_ctl], pp->start_ctl, pp->end_ctl);
447 for (i = pp->start_ctl; i <= pp->end_ctl; i++)
448 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
449 __ctl_load(cregs[pp->start_ctl], pp->start_ctl, pp->end_ctl);
453 * Set a bit in a control register of all cpus
455 void smp_ctl_set_bit(int cr, int bit) {
456 ec_creg_mask_parms parms;
458 parms.start_ctl = cr;
460 parms.orvals[cr] = 1 << bit;
461 parms.andvals[cr] = -1L;
463 smp_call_function(smp_ctl_bit_callback, &parms, 0, 1);
464 __ctl_set_bit(cr, bit);
469 * Clear a bit in a control register of all cpus
471 void smp_ctl_clear_bit(int cr, int bit) {
472 ec_creg_mask_parms parms;
474 parms.start_ctl = cr;
476 parms.orvals[cr] = 0;
477 parms.andvals[cr] = ~(1L << bit);
479 smp_call_function(smp_ctl_bit_callback, &parms, 0, 1);
480 __ctl_clear_bit(cr, bit);
485 * Lets check how many CPUs we have.
489 __init smp_count_cpus(void)
491 unsigned int cpu, num_cpus;
495 * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
498 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
499 current_thread_info()->cpu = 0;
501 for (cpu = 0; cpu <= 65535; cpu++) {
502 if ((__u16) cpu == boot_cpu_addr)
504 __cpu_logical_map[1] = (__u16) cpu;
505 if (signal_processor(1, sigp_sense) ==
506 sigp_not_operational)
511 printk("Detected %d CPU's\n",(int) num_cpus);
512 printk("Boot cpu address %2X\n", boot_cpu_addr);
518 * Activate a secondary processor.
520 extern void init_cpu_timer(void);
521 extern void init_cpu_vtimer(void);
522 extern int pfault_init(void);
523 extern void pfault_fini(void);
525 int __devinit start_secondary(void *cpuvoid)
530 /* init per CPU timer */
532 #ifdef CONFIG_VIRT_TIMER
536 /* Enable pfault pseudo page faults on this cpu. */
540 /* Mark this cpu as online */
541 cpu_set(smp_processor_id(), cpu_online_map);
542 /* Switch on interrupts */
544 /* Print info about this processor */
545 print_cpu_info(&S390_lowcore.cpu_data);
546 /* cpu_idle will call schedule for us */
551 static void __init smp_create_idle(unsigned int cpu)
553 struct task_struct *p;
556 * don't care about the psw and regs settings since we'll never
557 * reschedule the forked task.
561 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
562 current_set[cpu] = p;
565 /* Reserving and releasing of CPUs */
567 static DEFINE_SPINLOCK(smp_reserve_lock);
568 static int smp_cpu_reserved[NR_CPUS];
571 smp_get_cpu(cpumask_t cpu_mask)
576 spin_lock_irqsave(&smp_reserve_lock, flags);
577 /* Try to find an already reserved cpu. */
578 for_each_cpu_mask(cpu, cpu_mask) {
579 if (smp_cpu_reserved[cpu] != 0) {
580 smp_cpu_reserved[cpu]++;
585 /* Reserve a new cpu from cpu_mask. */
586 for_each_cpu_mask(cpu, cpu_mask) {
587 if (cpu_online(cpu)) {
588 smp_cpu_reserved[cpu]++;
594 spin_unlock_irqrestore(&smp_reserve_lock, flags);
603 spin_lock_irqsave(&smp_reserve_lock, flags);
604 smp_cpu_reserved[cpu]--;
605 spin_unlock_irqrestore(&smp_reserve_lock, flags);
613 /* Check for stopped state */
614 if (signal_processor_ps(&status, 0, cpu, sigp_sense) == sigp_status_stored) {
621 /* Upping and downing of CPUs */
624 __cpu_up(unsigned int cpu)
626 struct task_struct *idle;
627 struct _lowcore *cpu_lowcore;
628 struct stack_frame *sf;
632 for (curr_cpu = 0; curr_cpu <= 65535; curr_cpu++) {
633 __cpu_logical_map[cpu] = (__u16) curr_cpu;
634 if (cpu_stopped(cpu))
638 if (!cpu_stopped(cpu))
641 ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
642 cpu, sigp_set_prefix);
644 printk("sigp_set_prefix failed for cpu %d "
645 "with condition code %d\n",
646 (int) cpu, (int) ccode);
650 idle = current_set[cpu];
651 cpu_lowcore = lowcore_ptr[cpu];
652 cpu_lowcore->kernel_stack = (unsigned long)
653 task_stack_page(idle) + (THREAD_SIZE);
654 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
655 - sizeof(struct pt_regs)
656 - sizeof(struct stack_frame));
657 memset(sf, 0, sizeof(struct stack_frame));
658 sf->gprs[9] = (unsigned long) sf;
659 cpu_lowcore->save_area[15] = (unsigned long) sf;
660 __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
661 __asm__ __volatile__("stam 0,15,0(%0)"
662 : : "a" (&cpu_lowcore->access_regs_save_area)
664 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
665 cpu_lowcore->current_task = (unsigned long) idle;
666 cpu_lowcore->cpu_data.cpu_nr = cpu;
669 while (signal_processor(cpu,sigp_restart) == sigp_busy)
672 while (!cpu_online(cpu))
677 static unsigned int __initdata additional_cpus;
678 static unsigned int __initdata possible_cpus;
680 void __init smp_setup_cpu_possible_map(void)
682 unsigned int phy_cpus, pos_cpus, cpu;
684 phy_cpus = smp_count_cpus();
685 pos_cpus = min(phy_cpus + additional_cpus, (unsigned int) NR_CPUS);
688 pos_cpus = min(possible_cpus, (unsigned int) NR_CPUS);
690 for (cpu = 0; cpu < pos_cpus; cpu++)
691 cpu_set(cpu, cpu_possible_map);
693 phy_cpus = min(phy_cpus, pos_cpus);
695 for (cpu = 0; cpu < phy_cpus; cpu++)
696 cpu_set(cpu, cpu_present_map);
699 #ifdef CONFIG_HOTPLUG_CPU
701 static int __init setup_additional_cpus(char *s)
703 additional_cpus = simple_strtoul(s, NULL, 0);
706 early_param("additional_cpus", setup_additional_cpus);
708 static int __init setup_possible_cpus(char *s)
710 possible_cpus = simple_strtoul(s, NULL, 0);
713 early_param("possible_cpus", setup_possible_cpus);
719 ec_creg_mask_parms cr_parms;
720 int cpu = smp_processor_id();
722 spin_lock_irqsave(&smp_reserve_lock, flags);
723 if (smp_cpu_reserved[cpu] != 0) {
724 spin_unlock_irqrestore(&smp_reserve_lock, flags);
727 cpu_clear(cpu, cpu_online_map);
730 /* Disable pfault pseudo page faults on this cpu. */
735 /* disable all external interrupts */
737 cr_parms.start_ctl = 0;
738 cr_parms.end_ctl = 0;
739 cr_parms.orvals[0] = 0;
740 cr_parms.andvals[0] = ~(1<<15 | 1<<14 | 1<<13 | 1<<12 |
741 1<<11 | 1<<10 | 1<< 6 | 1<< 4);
742 smp_ctl_bit_callback(&cr_parms);
744 /* disable all I/O interrupts */
746 cr_parms.start_ctl = 6;
747 cr_parms.end_ctl = 6;
748 cr_parms.orvals[6] = 0;
749 cr_parms.andvals[6] = ~(1<<31 | 1<<30 | 1<<29 | 1<<28 |
750 1<<27 | 1<<26 | 1<<25 | 1<<24);
751 smp_ctl_bit_callback(&cr_parms);
753 /* disable most machine checks */
755 cr_parms.start_ctl = 14;
756 cr_parms.end_ctl = 14;
757 cr_parms.orvals[14] = 0;
758 cr_parms.andvals[14] = ~(1<<28 | 1<<27 | 1<<26 | 1<<25 | 1<<24);
759 smp_ctl_bit_callback(&cr_parms);
761 spin_unlock_irqrestore(&smp_reserve_lock, flags);
766 __cpu_die(unsigned int cpu)
768 /* Wait until target cpu is down */
769 while (!smp_cpu_not_running(cpu))
771 printk("Processor %d spun down\n", cpu);
778 signal_processor(smp_processor_id(), sigp_stop);
783 #endif /* CONFIG_HOTPLUG_CPU */
786 * Cycle through the processors and setup structures.
789 void __init smp_prepare_cpus(unsigned int max_cpus)
795 /* request the 0x1201 emergency signal external interrupt */
796 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
797 panic("Couldn't request external interrupt 0x1201");
798 memset(lowcore_ptr,0,sizeof(lowcore_ptr));
800 * Initialize prefix pages and stacks for all possible cpus
802 print_cpu_info(&S390_lowcore.cpu_data);
805 lowcore_ptr[i] = (struct _lowcore *)
806 __get_free_pages(GFP_KERNEL|GFP_DMA,
807 sizeof(void*) == 8 ? 1 : 0);
808 stack = __get_free_pages(GFP_KERNEL,ASYNC_ORDER);
809 if (lowcore_ptr[i] == NULL || stack == 0ULL)
810 panic("smp_boot_cpus failed to allocate memory\n");
812 *(lowcore_ptr[i]) = S390_lowcore;
813 lowcore_ptr[i]->async_stack = stack + (ASYNC_SIZE);
814 stack = __get_free_pages(GFP_KERNEL,0);
816 panic("smp_boot_cpus failed to allocate memory\n");
817 lowcore_ptr[i]->panic_stack = stack + (PAGE_SIZE);
819 if (MACHINE_HAS_IEEE) {
820 lowcore_ptr[i]->extended_save_area_addr =
821 (__u32) __get_free_pages(GFP_KERNEL,0);
822 if (lowcore_ptr[i]->extended_save_area_addr == 0)
823 panic("smp_boot_cpus failed to "
824 "allocate memory\n");
829 if (MACHINE_HAS_IEEE)
830 ctl_set_bit(14, 29); /* enable extended save area */
832 set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
835 if (cpu != smp_processor_id())
836 smp_create_idle(cpu);
839 void __devinit smp_prepare_boot_cpu(void)
841 BUG_ON(smp_processor_id() != 0);
843 cpu_set(0, cpu_online_map);
844 S390_lowcore.percpu_offset = __per_cpu_offset[0];
845 current_set[0] = current;
848 void smp_cpus_done(unsigned int max_cpus)
850 cpu_present_map = cpu_possible_map;
854 * the frequency of the profiling timer can be changed
855 * by writing a multiplier value into /proc/profile.
857 * usually you want to run this on all CPUs ;)
859 int setup_profiling_timer(unsigned int multiplier)
864 static DEFINE_PER_CPU(struct cpu, cpu_devices);
866 static int __init topology_init(void)
872 ret = register_cpu(&per_cpu(cpu_devices, cpu), cpu, NULL);
874 printk(KERN_WARNING "topology_init: register_cpu %d "
875 "failed (%d)\n", cpu, ret);
880 subsys_initcall(topology_init);
882 EXPORT_SYMBOL(cpu_online_map);
883 EXPORT_SYMBOL(cpu_possible_map);
884 EXPORT_SYMBOL(lowcore_ptr);
885 EXPORT_SYMBOL(smp_ctl_set_bit);
886 EXPORT_SYMBOL(smp_ctl_clear_bit);
887 EXPORT_SYMBOL(smp_call_function);
888 EXPORT_SYMBOL(smp_get_cpu);
889 EXPORT_SYMBOL(smp_put_cpu);