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 static void smp_ext_bitcall(int, ec_bit_sig);
63 static void smp_ext_bitcall_others(ec_bit_sig);
66 * Structure and data for smp_call_function(). This is designed to minimise
67 * static memory requirements. It also looks cleaner.
69 static DEFINE_SPINLOCK(call_lock);
71 struct call_data_struct {
72 void (*func) (void *info);
79 static struct call_data_struct * call_data;
82 * 'Call function' interrupt callback
84 static void do_call_function(void)
86 void (*func) (void *info) = call_data->func;
87 void *info = call_data->info;
88 int wait = call_data->wait;
90 atomic_inc(&call_data->started);
93 atomic_inc(&call_data->finished);
97 * this function sends a 'generic call function' IPI to all other CPUs
101 int smp_call_function (void (*func) (void *info), void *info, int nonatomic,
104 * [SUMMARY] Run a function on all other CPUs.
105 * <func> The function to run. This must be fast and non-blocking.
106 * <info> An arbitrary pointer to pass to the function.
107 * <nonatomic> currently unused.
108 * <wait> If true, wait (atomically) until function has completed on other CPUs.
109 * [RETURNS] 0 on success, else a negative status code. Does not return until
110 * remote CPUs are nearly ready to execute <<func>> or are or have executed.
112 * You must not call this function with disabled interrupts or from a
113 * hardware interrupt handler or from a bottom half handler.
116 struct call_data_struct data;
117 int cpus = num_online_cpus()-1;
122 /* Can deadlock when called with interrupts disabled */
123 WARN_ON(irqs_disabled());
127 atomic_set(&data.started, 0);
130 atomic_set(&data.finished, 0);
132 spin_lock(&call_lock);
134 /* Send a message to all other CPUs and wait for them to respond */
135 smp_ext_bitcall_others(ec_call_function);
137 /* Wait for response */
138 while (atomic_read(&data.started) != cpus)
142 while (atomic_read(&data.finished) != cpus)
144 spin_unlock(&call_lock);
150 * Call a function on one CPU
151 * cpu : the CPU the function should be executed on
153 * You must not call this function with disabled interrupts or from a
154 * hardware interrupt handler. You may call it from a bottom half.
156 * It is guaranteed that the called function runs on the specified CPU,
157 * preemption is disabled.
159 int smp_call_function_on(void (*func) (void *info), void *info,
160 int nonatomic, int wait, int cpu)
162 struct call_data_struct data;
165 if (!cpu_online(cpu))
168 /* disable preemption for local function call */
169 curr_cpu = get_cpu();
171 if (curr_cpu == cpu) {
172 /* direct call to function */
180 atomic_set(&data.started, 0);
183 atomic_set(&data.finished, 0);
185 spin_lock_bh(&call_lock);
187 smp_ext_bitcall(cpu, ec_call_function);
189 /* Wait for response */
190 while (atomic_read(&data.started) != 1)
194 while (atomic_read(&data.finished) != 1)
197 spin_unlock_bh(&call_lock);
201 EXPORT_SYMBOL(smp_call_function_on);
203 static inline void do_send_stop(void)
207 /* stop all processors */
208 for_each_online_cpu(cpu) {
209 if (cpu == smp_processor_id())
212 rc = signal_processor(cpu, sigp_stop);
213 } while (rc == sigp_busy);
217 static inline void do_store_status(void)
221 /* store status of all processors in their lowcores (real 0) */
222 for_each_online_cpu(cpu) {
223 if (cpu == smp_processor_id())
226 rc = signal_processor_p(
227 (__u32)(unsigned long) lowcore_ptr[cpu], cpu,
228 sigp_store_status_at_address);
229 } while(rc == sigp_busy);
234 * this function sends a 'stop' sigp to all other CPUs in the system.
235 * it goes straight through.
237 void smp_send_stop(void)
239 /* write magic number to zero page (absolute 0) */
240 lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
242 /* stop other processors. */
245 /* store status of other processors. */
250 * Reboot, halt and power_off routines for SMP.
253 static void do_machine_restart(void * __unused)
256 static atomic_t cpuid = ATOMIC_INIT(-1);
258 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) != -1)
259 signal_processor(smp_processor_id(), sigp_stop);
261 /* Wait for all other cpus to enter stopped state */
262 for_each_online_cpu(cpu) {
263 if (cpu == smp_processor_id())
265 while(!smp_cpu_not_running(cpu))
269 /* Store status of other cpus. */
273 * Finally call reipl. Because we waited for all other
274 * cpus to enter this function we know that they do
275 * not hold any s390irq-locks (the cpus have been
276 * interrupted by an external interrupt and s390irq
277 * locks are always held disabled).
282 void machine_restart_smp(char * __unused)
284 on_each_cpu(do_machine_restart, NULL, 0, 0);
287 static void do_wait_for_stop(void)
289 unsigned long cr[16];
291 __ctl_store(cr, 0, 15);
294 __ctl_load(cr, 0, 15);
299 static void do_machine_halt(void * __unused)
301 static atomic_t cpuid = ATOMIC_INIT(-1);
303 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) == -1) {
305 if (MACHINE_IS_VM && strlen(vmhalt_cmd) > 0)
306 cpcmd(vmhalt_cmd, NULL, 0, NULL);
307 signal_processor(smp_processor_id(),
308 sigp_stop_and_store_status);
313 void machine_halt_smp(void)
315 on_each_cpu(do_machine_halt, NULL, 0, 0);
318 static void do_machine_power_off(void * __unused)
320 static atomic_t cpuid = ATOMIC_INIT(-1);
322 if (atomic_cmpxchg(&cpuid, -1, smp_processor_id()) == -1) {
324 if (MACHINE_IS_VM && strlen(vmpoff_cmd) > 0)
325 cpcmd(vmpoff_cmd, NULL, 0, NULL);
326 signal_processor(smp_processor_id(),
327 sigp_stop_and_store_status);
332 void machine_power_off_smp(void)
334 on_each_cpu(do_machine_power_off, NULL, 0, 0);
338 * This is the main routine where commands issued by other
342 void do_ext_call_interrupt(struct pt_regs *regs, __u16 code)
347 * handle bit signal external calls
349 * For the ec_schedule signal we have to do nothing. All the work
350 * is done automatically when we return from the interrupt.
352 bits = xchg(&S390_lowcore.ext_call_fast, 0);
354 if (test_bit(ec_call_function, &bits))
359 * Send an external call sigp to another cpu and return without waiting
360 * for its completion.
362 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
365 * Set signaling bit in lowcore of target cpu and kick it
367 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
368 while(signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
373 * Send an external call sigp to every other cpu in the system and
374 * return without waiting for its completion.
376 static void smp_ext_bitcall_others(ec_bit_sig sig)
380 for_each_online_cpu(cpu) {
381 if (cpu == smp_processor_id())
384 * Set signaling bit in lowcore of target cpu and kick it
386 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
387 while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
394 * this function sends a 'purge tlb' signal to another CPU.
396 void smp_ptlb_callback(void *info)
401 void smp_ptlb_all(void)
403 on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
405 EXPORT_SYMBOL(smp_ptlb_all);
406 #endif /* ! CONFIG_64BIT */
409 * this function sends a 'reschedule' IPI to another CPU.
410 * it goes straight through and wastes no time serializing
411 * anything. Worst case is that we lose a reschedule ...
413 void smp_send_reschedule(int cpu)
415 smp_ext_bitcall(cpu, ec_schedule);
419 * parameter area for the set/clear control bit callbacks
425 unsigned long orvals[16];
426 unsigned long andvals[16];
427 } ec_creg_mask_parms;
430 * callback for setting/clearing control bits
432 void smp_ctl_bit_callback(void *info) {
433 ec_creg_mask_parms *pp;
434 unsigned long cregs[16];
437 pp = (ec_creg_mask_parms *) info;
438 __ctl_store(cregs[pp->start_ctl], pp->start_ctl, pp->end_ctl);
439 for (i = pp->start_ctl; i <= pp->end_ctl; i++)
440 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
441 __ctl_load(cregs[pp->start_ctl], pp->start_ctl, pp->end_ctl);
445 * Set a bit in a control register of all cpus
447 void smp_ctl_set_bit(int cr, int bit) {
448 ec_creg_mask_parms parms;
450 parms.start_ctl = cr;
452 parms.orvals[cr] = 1 << bit;
453 parms.andvals[cr] = -1L;
455 smp_call_function(smp_ctl_bit_callback, &parms, 0, 1);
456 __ctl_set_bit(cr, bit);
461 * Clear a bit in a control register of all cpus
463 void smp_ctl_clear_bit(int cr, int bit) {
464 ec_creg_mask_parms parms;
466 parms.start_ctl = cr;
468 parms.orvals[cr] = 0;
469 parms.andvals[cr] = ~(1L << bit);
471 smp_call_function(smp_ctl_bit_callback, &parms, 0, 1);
472 __ctl_clear_bit(cr, bit);
477 * Lets check how many CPUs we have.
481 __init smp_count_cpus(void)
483 unsigned int cpu, num_cpus;
487 * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
490 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
491 current_thread_info()->cpu = 0;
493 for (cpu = 0; cpu <= 65535; cpu++) {
494 if ((__u16) cpu == boot_cpu_addr)
496 __cpu_logical_map[1] = (__u16) cpu;
497 if (signal_processor(1, sigp_sense) ==
498 sigp_not_operational)
503 printk("Detected %d CPU's\n",(int) num_cpus);
504 printk("Boot cpu address %2X\n", boot_cpu_addr);
510 * Activate a secondary processor.
512 extern void init_cpu_timer(void);
513 extern void init_cpu_vtimer(void);
514 extern int pfault_init(void);
515 extern void pfault_fini(void);
517 int __devinit start_secondary(void *cpuvoid)
522 /* init per CPU timer */
524 #ifdef CONFIG_VIRT_TIMER
528 /* Enable pfault pseudo page faults on this cpu. */
532 /* Mark this cpu as online */
533 cpu_set(smp_processor_id(), cpu_online_map);
534 /* Switch on interrupts */
536 /* Print info about this processor */
537 print_cpu_info(&S390_lowcore.cpu_data);
538 /* cpu_idle will call schedule for us */
543 static void __init smp_create_idle(unsigned int cpu)
545 struct task_struct *p;
548 * don't care about the psw and regs settings since we'll never
549 * reschedule the forked task.
553 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
554 current_set[cpu] = p;
557 /* Reserving and releasing of CPUs */
559 static DEFINE_SPINLOCK(smp_reserve_lock);
560 static int smp_cpu_reserved[NR_CPUS];
563 smp_get_cpu(cpumask_t cpu_mask)
568 spin_lock_irqsave(&smp_reserve_lock, flags);
569 /* Try to find an already reserved cpu. */
570 for_each_cpu_mask(cpu, cpu_mask) {
571 if (smp_cpu_reserved[cpu] != 0) {
572 smp_cpu_reserved[cpu]++;
577 /* Reserve a new cpu from cpu_mask. */
578 for_each_cpu_mask(cpu, cpu_mask) {
579 if (cpu_online(cpu)) {
580 smp_cpu_reserved[cpu]++;
586 spin_unlock_irqrestore(&smp_reserve_lock, flags);
595 spin_lock_irqsave(&smp_reserve_lock, flags);
596 smp_cpu_reserved[cpu]--;
597 spin_unlock_irqrestore(&smp_reserve_lock, flags);
605 /* Check for stopped state */
606 if (signal_processor_ps(&status, 0, cpu, sigp_sense) == sigp_status_stored) {
613 /* Upping and downing of CPUs */
616 __cpu_up(unsigned int cpu)
618 struct task_struct *idle;
619 struct _lowcore *cpu_lowcore;
620 struct stack_frame *sf;
624 for (curr_cpu = 0; curr_cpu <= 65535; curr_cpu++) {
625 __cpu_logical_map[cpu] = (__u16) curr_cpu;
626 if (cpu_stopped(cpu))
630 if (!cpu_stopped(cpu))
633 ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
634 cpu, sigp_set_prefix);
636 printk("sigp_set_prefix failed for cpu %d "
637 "with condition code %d\n",
638 (int) cpu, (int) ccode);
642 idle = current_set[cpu];
643 cpu_lowcore = lowcore_ptr[cpu];
644 cpu_lowcore->kernel_stack = (unsigned long)
645 task_stack_page(idle) + (THREAD_SIZE);
646 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
647 - sizeof(struct pt_regs)
648 - sizeof(struct stack_frame));
649 memset(sf, 0, sizeof(struct stack_frame));
650 sf->gprs[9] = (unsigned long) sf;
651 cpu_lowcore->save_area[15] = (unsigned long) sf;
652 __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
653 __asm__ __volatile__("stam 0,15,0(%0)"
654 : : "a" (&cpu_lowcore->access_regs_save_area)
656 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
657 cpu_lowcore->current_task = (unsigned long) idle;
658 cpu_lowcore->cpu_data.cpu_nr = cpu;
661 while (signal_processor(cpu,sigp_restart) == sigp_busy)
664 while (!cpu_online(cpu))
669 static unsigned int __initdata additional_cpus;
670 static unsigned int __initdata possible_cpus;
672 void __init smp_setup_cpu_possible_map(void)
674 unsigned int phy_cpus, pos_cpus, cpu;
676 phy_cpus = smp_count_cpus();
677 pos_cpus = min(phy_cpus + additional_cpus, (unsigned int) NR_CPUS);
680 pos_cpus = min(possible_cpus, (unsigned int) NR_CPUS);
682 for (cpu = 0; cpu < pos_cpus; cpu++)
683 cpu_set(cpu, cpu_possible_map);
685 phy_cpus = min(phy_cpus, pos_cpus);
687 for (cpu = 0; cpu < phy_cpus; cpu++)
688 cpu_set(cpu, cpu_present_map);
691 #ifdef CONFIG_HOTPLUG_CPU
693 static int __init setup_additional_cpus(char *s)
695 additional_cpus = simple_strtoul(s, NULL, 0);
698 early_param("additional_cpus", setup_additional_cpus);
700 static int __init setup_possible_cpus(char *s)
702 possible_cpus = simple_strtoul(s, NULL, 0);
705 early_param("possible_cpus", setup_possible_cpus);
711 ec_creg_mask_parms cr_parms;
712 int cpu = smp_processor_id();
714 spin_lock_irqsave(&smp_reserve_lock, flags);
715 if (smp_cpu_reserved[cpu] != 0) {
716 spin_unlock_irqrestore(&smp_reserve_lock, flags);
719 cpu_clear(cpu, cpu_online_map);
722 /* Disable pfault pseudo page faults on this cpu. */
727 /* disable all external interrupts */
729 cr_parms.start_ctl = 0;
730 cr_parms.end_ctl = 0;
731 cr_parms.orvals[0] = 0;
732 cr_parms.andvals[0] = ~(1<<15 | 1<<14 | 1<<13 | 1<<12 |
733 1<<11 | 1<<10 | 1<< 6 | 1<< 4);
734 smp_ctl_bit_callback(&cr_parms);
736 /* disable all I/O interrupts */
738 cr_parms.start_ctl = 6;
739 cr_parms.end_ctl = 6;
740 cr_parms.orvals[6] = 0;
741 cr_parms.andvals[6] = ~(1<<31 | 1<<30 | 1<<29 | 1<<28 |
742 1<<27 | 1<<26 | 1<<25 | 1<<24);
743 smp_ctl_bit_callback(&cr_parms);
745 /* disable most machine checks */
747 cr_parms.start_ctl = 14;
748 cr_parms.end_ctl = 14;
749 cr_parms.orvals[14] = 0;
750 cr_parms.andvals[14] = ~(1<<28 | 1<<27 | 1<<26 | 1<<25 | 1<<24);
751 smp_ctl_bit_callback(&cr_parms);
753 spin_unlock_irqrestore(&smp_reserve_lock, flags);
758 __cpu_die(unsigned int cpu)
760 /* Wait until target cpu is down */
761 while (!smp_cpu_not_running(cpu))
763 printk("Processor %d spun down\n", cpu);
770 signal_processor(smp_processor_id(), sigp_stop);
775 #endif /* CONFIG_HOTPLUG_CPU */
778 * Cycle through the processors and setup structures.
781 void __init smp_prepare_cpus(unsigned int max_cpus)
787 /* request the 0x1201 emergency signal external interrupt */
788 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
789 panic("Couldn't request external interrupt 0x1201");
790 memset(lowcore_ptr,0,sizeof(lowcore_ptr));
792 * Initialize prefix pages and stacks for all possible cpus
794 print_cpu_info(&S390_lowcore.cpu_data);
796 for_each_possible_cpu(i) {
797 lowcore_ptr[i] = (struct _lowcore *)
798 __get_free_pages(GFP_KERNEL|GFP_DMA,
799 sizeof(void*) == 8 ? 1 : 0);
800 stack = __get_free_pages(GFP_KERNEL,ASYNC_ORDER);
801 if (lowcore_ptr[i] == NULL || stack == 0ULL)
802 panic("smp_boot_cpus failed to allocate memory\n");
804 *(lowcore_ptr[i]) = S390_lowcore;
805 lowcore_ptr[i]->async_stack = stack + (ASYNC_SIZE);
806 stack = __get_free_pages(GFP_KERNEL,0);
808 panic("smp_boot_cpus failed to allocate memory\n");
809 lowcore_ptr[i]->panic_stack = stack + (PAGE_SIZE);
811 if (MACHINE_HAS_IEEE) {
812 lowcore_ptr[i]->extended_save_area_addr =
813 (__u32) __get_free_pages(GFP_KERNEL,0);
814 if (lowcore_ptr[i]->extended_save_area_addr == 0)
815 panic("smp_boot_cpus failed to "
816 "allocate memory\n");
821 if (MACHINE_HAS_IEEE)
822 ctl_set_bit(14, 29); /* enable extended save area */
824 set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
826 for_each_possible_cpu(cpu)
827 if (cpu != smp_processor_id())
828 smp_create_idle(cpu);
831 void __devinit smp_prepare_boot_cpu(void)
833 BUG_ON(smp_processor_id() != 0);
835 cpu_set(0, cpu_online_map);
836 S390_lowcore.percpu_offset = __per_cpu_offset[0];
837 current_set[0] = current;
840 void smp_cpus_done(unsigned int max_cpus)
842 cpu_present_map = cpu_possible_map;
846 * the frequency of the profiling timer can be changed
847 * by writing a multiplier value into /proc/profile.
849 * usually you want to run this on all CPUs ;)
851 int setup_profiling_timer(unsigned int multiplier)
856 static DEFINE_PER_CPU(struct cpu, cpu_devices);
858 static int __init topology_init(void)
863 for_each_possible_cpu(cpu) {
864 ret = register_cpu(&per_cpu(cpu_devices, cpu), cpu);
866 printk(KERN_WARNING "topology_init: register_cpu %d "
867 "failed (%d)\n", cpu, ret);
872 subsys_initcall(topology_init);
874 EXPORT_SYMBOL(cpu_online_map);
875 EXPORT_SYMBOL(cpu_possible_map);
876 EXPORT_SYMBOL(lowcore_ptr);
877 EXPORT_SYMBOL(smp_ctl_set_bit);
878 EXPORT_SYMBOL(smp_ctl_clear_bit);
879 EXPORT_SYMBOL(smp_call_function);
880 EXPORT_SYMBOL(smp_get_cpu);
881 EXPORT_SYMBOL(smp_put_cpu);