2 * arch/s390/kernel/smp.c
4 * Copyright IBM Corp. 1999,2007
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>
34 #include <linux/bootmem.h>
36 #include <asm/setup.h>
38 #include <asm/pgalloc.h>
40 #include <asm/s390_ext.h>
41 #include <asm/cpcmd.h>
42 #include <asm/tlbflush.h>
43 #include <asm/timer.h>
44 #include <asm/lowcore.h>
47 * An array with a pointer the lowcore of every CPU.
49 struct _lowcore *lowcore_ptr[NR_CPUS];
50 EXPORT_SYMBOL(lowcore_ptr);
52 cpumask_t cpu_online_map = CPU_MASK_NONE;
53 EXPORT_SYMBOL(cpu_online_map);
55 cpumask_t cpu_possible_map = CPU_MASK_NONE;
56 EXPORT_SYMBOL(cpu_possible_map);
58 static struct task_struct *current_set[NR_CPUS];
60 static void smp_ext_bitcall(int, ec_bit_sig);
63 * Structure and data for __smp_call_function_map(). This is designed to
64 * minimise static memory requirements. It also looks cleaner.
66 static DEFINE_SPINLOCK(call_lock);
68 struct call_data_struct {
69 void (*func) (void *info);
76 static struct call_data_struct *call_data;
79 * 'Call function' interrupt callback
81 static void do_call_function(void)
83 void (*func) (void *info) = call_data->func;
84 void *info = call_data->info;
85 int wait = call_data->wait;
87 cpu_set(smp_processor_id(), call_data->started);
90 cpu_set(smp_processor_id(), call_data->finished);;
93 static void __smp_call_function_map(void (*func) (void *info), void *info,
94 int nonatomic, int wait, cpumask_t map)
96 struct call_data_struct data;
100 * Can deadlock when interrupts are disabled or if in wrong context.
102 WARN_ON(irqs_disabled() || in_irq());
105 * Check for local function call. We have to have the same call order
106 * as in on_each_cpu() because of machine_restart_smp().
108 if (cpu_isset(smp_processor_id(), map)) {
110 cpu_clear(smp_processor_id(), map);
113 cpus_and(map, map, cpu_online_map);
119 data.started = CPU_MASK_NONE;
122 data.finished = CPU_MASK_NONE;
124 spin_lock_bh(&call_lock);
127 for_each_cpu_mask(cpu, map)
128 smp_ext_bitcall(cpu, ec_call_function);
130 /* Wait for response */
131 while (!cpus_equal(map, data.started))
135 while (!cpus_equal(map, data.finished))
138 spin_unlock_bh(&call_lock);
149 * @func: the function to run; this must be fast and non-blocking
150 * @info: an arbitrary pointer to pass to the function
152 * @wait: if true, wait (atomically) until function has completed on other CPUs
154 * Run a function on all other CPUs.
156 * You must not call this function with disabled interrupts, from a
157 * hardware interrupt handler or from a bottom half.
159 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
165 map = cpu_online_map;
166 cpu_clear(smp_processor_id(), map);
167 __smp_call_function_map(func, info, nonatomic, wait, map);
171 EXPORT_SYMBOL(smp_call_function);
174 * smp_call_function_on:
175 * @func: the function to run; this must be fast and non-blocking
176 * @info: an arbitrary pointer to pass to the function
178 * @wait: if true, wait (atomically) until function has completed on other CPUs
179 * @cpu: the CPU where func should run
181 * Run a function on one processor.
183 * You must not call this function with disabled interrupts, from a
184 * hardware interrupt handler or from a bottom half.
186 int smp_call_function_on(void (*func) (void *info), void *info, int nonatomic,
189 cpumask_t map = CPU_MASK_NONE;
193 __smp_call_function_map(func, info, nonatomic, wait, map);
197 EXPORT_SYMBOL(smp_call_function_on);
199 static void do_send_stop(void)
203 /* stop all processors */
204 for_each_online_cpu(cpu) {
205 if (cpu == smp_processor_id())
208 rc = signal_processor(cpu, sigp_stop);
209 } while (rc == sigp_busy);
213 static void do_store_status(void)
217 /* store status of all processors in their lowcores (real 0) */
218 for_each_online_cpu(cpu) {
219 if (cpu == smp_processor_id())
222 rc = signal_processor_p(
223 (__u32)(unsigned long) lowcore_ptr[cpu], cpu,
224 sigp_store_status_at_address);
225 } while (rc == sigp_busy);
229 static void do_wait_for_stop(void)
233 /* Wait for all other cpus to enter stopped state */
234 for_each_online_cpu(cpu) {
235 if (cpu == smp_processor_id())
237 while (!smp_cpu_not_running(cpu))
243 * this function sends a 'stop' sigp to all other CPUs in the system.
244 * it goes straight through.
246 void smp_send_stop(void)
248 /* Disable all interrupts/machine checks */
249 __load_psw_mask(psw_kernel_bits & ~PSW_MASK_MCHECK);
251 /* write magic number to zero page (absolute 0) */
252 lowcore_ptr[smp_processor_id()]->panic_magic = __PANIC_MAGIC;
254 /* stop other processors. */
257 /* wait until other processors are stopped */
260 /* store status of other processors. */
265 * Reboot, halt and power_off routines for SMP.
267 void machine_restart_smp(char *__unused)
273 void machine_halt_smp(void)
276 if (MACHINE_IS_VM && strlen(vmhalt_cmd) > 0)
277 __cpcmd(vmhalt_cmd, NULL, 0, NULL);
278 signal_processor(smp_processor_id(), sigp_stop_and_store_status);
282 void machine_power_off_smp(void)
285 if (MACHINE_IS_VM && strlen(vmpoff_cmd) > 0)
286 __cpcmd(vmpoff_cmd, NULL, 0, NULL);
287 signal_processor(smp_processor_id(), sigp_stop_and_store_status);
292 * This is the main routine where commands issued by other
296 static void do_ext_call_interrupt(__u16 code)
301 * handle bit signal external calls
303 * For the ec_schedule signal we have to do nothing. All the work
304 * is done automatically when we return from the interrupt.
306 bits = xchg(&S390_lowcore.ext_call_fast, 0);
308 if (test_bit(ec_call_function, &bits))
313 * Send an external call sigp to another cpu and return without waiting
314 * for its completion.
316 static void smp_ext_bitcall(int cpu, ec_bit_sig sig)
319 * Set signaling bit in lowcore of target cpu and kick it
321 set_bit(sig, (unsigned long *) &lowcore_ptr[cpu]->ext_call_fast);
322 while (signal_processor(cpu, sigp_emergency_signal) == sigp_busy)
328 * this function sends a 'purge tlb' signal to another CPU.
330 void smp_ptlb_callback(void *info)
335 void smp_ptlb_all(void)
337 on_each_cpu(smp_ptlb_callback, NULL, 0, 1);
339 EXPORT_SYMBOL(smp_ptlb_all);
340 #endif /* ! CONFIG_64BIT */
343 * this function sends a 'reschedule' IPI to another CPU.
344 * it goes straight through and wastes no time serializing
345 * anything. Worst case is that we lose a reschedule ...
347 void smp_send_reschedule(int cpu)
349 smp_ext_bitcall(cpu, ec_schedule);
353 * parameter area for the set/clear control bit callbacks
355 struct ec_creg_mask_parms {
356 unsigned long orvals[16];
357 unsigned long andvals[16];
361 * callback for setting/clearing control bits
363 static void smp_ctl_bit_callback(void *info)
365 struct ec_creg_mask_parms *pp = info;
366 unsigned long cregs[16];
369 __ctl_store(cregs, 0, 15);
370 for (i = 0; i <= 15; i++)
371 cregs[i] = (cregs[i] & pp->andvals[i]) | pp->orvals[i];
372 __ctl_load(cregs, 0, 15);
376 * Set a bit in a control register of all cpus
378 void smp_ctl_set_bit(int cr, int bit)
380 struct ec_creg_mask_parms parms;
382 memset(&parms.orvals, 0, sizeof(parms.orvals));
383 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
384 parms.orvals[cr] = 1 << bit;
385 on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
387 EXPORT_SYMBOL(smp_ctl_set_bit);
390 * Clear a bit in a control register of all cpus
392 void smp_ctl_clear_bit(int cr, int bit)
394 struct ec_creg_mask_parms parms;
396 memset(&parms.orvals, 0, sizeof(parms.orvals));
397 memset(&parms.andvals, 0xff, sizeof(parms.andvals));
398 parms.andvals[cr] = ~(1L << bit);
399 on_each_cpu(smp_ctl_bit_callback, &parms, 0, 1);
401 EXPORT_SYMBOL(smp_ctl_clear_bit);
403 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
406 * zfcpdump_prefix_array holds prefix registers for the following scenario:
407 * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
408 * save its prefix registers, since they get lost, when switching from 31 bit
411 unsigned int zfcpdump_prefix_array[NR_CPUS + 1] \
412 __attribute__((__section__(".data")));
414 static void __init smp_get_save_areas(void)
416 unsigned int cpu, cpu_num, rc;
419 if (ipl_info.type != IPL_TYPE_FCP_DUMP)
421 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
423 for (cpu = 0; cpu <= 65535; cpu++) {
424 if ((u16) cpu == boot_cpu_addr)
426 __cpu_logical_map[1] = (__u16) cpu;
427 if (signal_processor(1, sigp_sense) == sigp_not_operational)
429 if (cpu_num >= NR_CPUS) {
430 printk("WARNING: Registers for cpu %i are not "
431 "saved, since dump kernel was compiled with"
432 "NR_CPUS=%i!\n", cpu_num, NR_CPUS);
435 zfcpdump_save_areas[cpu_num] =
436 alloc_bootmem(sizeof(union save_area));
438 rc = signal_processor(1, sigp_stop_and_store_status);
443 memcpy(zfcpdump_save_areas[cpu_num],
444 (void *)(unsigned long) store_prefix() +
445 SAVE_AREA_BASE, SAVE_AREA_SIZE);
447 /* copy original prefix register */
448 zfcpdump_save_areas[cpu_num]->s390x.pref_reg =
449 zfcpdump_prefix_array[cpu_num];
455 union save_area *zfcpdump_save_areas[NR_CPUS + 1];
456 EXPORT_SYMBOL_GPL(zfcpdump_save_areas);
459 #define smp_get_save_areas() do { } while (0)
463 * Lets check how many CPUs we have.
466 static unsigned int __init smp_count_cpus(void)
468 unsigned int cpu, num_cpus;
472 * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
475 boot_cpu_addr = S390_lowcore.cpu_data.cpu_addr;
476 current_thread_info()->cpu = 0;
478 for (cpu = 0; cpu <= 65535; cpu++) {
479 if ((__u16) cpu == boot_cpu_addr)
481 __cpu_logical_map[1] = (__u16) cpu;
482 if (signal_processor(1, sigp_sense) == sigp_not_operational)
487 printk("Detected %d CPU's\n", (int) num_cpus);
488 printk("Boot cpu address %2X\n", boot_cpu_addr);
494 * Activate a secondary processor.
496 int __devinit start_secondary(void *cpuvoid)
501 /* Enable TOD clock interrupts on the secondary cpu. */
503 #ifdef CONFIG_VIRT_TIMER
504 /* Enable cpu timer interrupts on the secondary cpu. */
507 /* Enable pfault pseudo page faults on this cpu. */
510 /* Mark this cpu as online */
511 cpu_set(smp_processor_id(), cpu_online_map);
512 /* Switch on interrupts */
514 /* Print info about this processor */
515 print_cpu_info(&S390_lowcore.cpu_data);
516 /* cpu_idle will call schedule for us */
521 static void __init smp_create_idle(unsigned int cpu)
523 struct task_struct *p;
526 * don't care about the psw and regs settings since we'll never
527 * reschedule the forked task.
531 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
532 current_set[cpu] = p;
535 static int cpu_stopped(int cpu)
539 /* Check for stopped state */
540 if (signal_processor_ps(&status, 0, cpu, sigp_sense) ==
541 sigp_status_stored) {
548 /* Upping and downing of CPUs */
550 int __cpu_up(unsigned int cpu)
552 struct task_struct *idle;
553 struct _lowcore *cpu_lowcore;
554 struct stack_frame *sf;
558 for (curr_cpu = 0; curr_cpu <= 65535; curr_cpu++) {
559 __cpu_logical_map[cpu] = (__u16) curr_cpu;
560 if (cpu_stopped(cpu))
564 if (!cpu_stopped(cpu))
567 ccode = signal_processor_p((__u32)(unsigned long)(lowcore_ptr[cpu]),
568 cpu, sigp_set_prefix);
570 printk("sigp_set_prefix failed for cpu %d "
571 "with condition code %d\n",
572 (int) cpu, (int) ccode);
576 idle = current_set[cpu];
577 cpu_lowcore = lowcore_ptr[cpu];
578 cpu_lowcore->kernel_stack = (unsigned long)
579 task_stack_page(idle) + THREAD_SIZE;
580 sf = (struct stack_frame *) (cpu_lowcore->kernel_stack
581 - sizeof(struct pt_regs)
582 - sizeof(struct stack_frame));
583 memset(sf, 0, sizeof(struct stack_frame));
584 sf->gprs[9] = (unsigned long) sf;
585 cpu_lowcore->save_area[15] = (unsigned long) sf;
586 __ctl_store(cpu_lowcore->cregs_save_area[0], 0, 15);
589 : : "a" (&cpu_lowcore->access_regs_save_area) : "memory");
590 cpu_lowcore->percpu_offset = __per_cpu_offset[cpu];
591 cpu_lowcore->current_task = (unsigned long) idle;
592 cpu_lowcore->cpu_data.cpu_nr = cpu;
595 while (signal_processor(cpu, sigp_restart) == sigp_busy)
598 while (!cpu_online(cpu))
603 static unsigned int __initdata additional_cpus;
604 static unsigned int __initdata possible_cpus;
606 void __init smp_setup_cpu_possible_map(void)
608 unsigned int phy_cpus, pos_cpus, cpu;
610 smp_get_save_areas();
611 phy_cpus = smp_count_cpus();
612 pos_cpus = min(phy_cpus + additional_cpus, (unsigned int) NR_CPUS);
615 pos_cpus = min(possible_cpus, (unsigned int) NR_CPUS);
617 for (cpu = 0; cpu < pos_cpus; cpu++)
618 cpu_set(cpu, cpu_possible_map);
620 phy_cpus = min(phy_cpus, pos_cpus);
622 for (cpu = 0; cpu < phy_cpus; cpu++)
623 cpu_set(cpu, cpu_present_map);
626 #ifdef CONFIG_HOTPLUG_CPU
628 static int __init setup_additional_cpus(char *s)
630 additional_cpus = simple_strtoul(s, NULL, 0);
633 early_param("additional_cpus", setup_additional_cpus);
635 static int __init setup_possible_cpus(char *s)
637 possible_cpus = simple_strtoul(s, NULL, 0);
640 early_param("possible_cpus", setup_possible_cpus);
642 int __cpu_disable(void)
644 struct ec_creg_mask_parms cr_parms;
645 int cpu = smp_processor_id();
647 cpu_clear(cpu, cpu_online_map);
649 /* Disable pfault pseudo page faults on this cpu. */
652 memset(&cr_parms.orvals, 0, sizeof(cr_parms.orvals));
653 memset(&cr_parms.andvals, 0xff, sizeof(cr_parms.andvals));
655 /* disable all external interrupts */
656 cr_parms.orvals[0] = 0;
657 cr_parms.andvals[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
658 1 << 11 | 1 << 10 | 1 << 6 | 1 << 4);
659 /* disable all I/O interrupts */
660 cr_parms.orvals[6] = 0;
661 cr_parms.andvals[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
662 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
663 /* disable most machine checks */
664 cr_parms.orvals[14] = 0;
665 cr_parms.andvals[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
668 smp_ctl_bit_callback(&cr_parms);
673 void __cpu_die(unsigned int cpu)
675 /* Wait until target cpu is down */
676 while (!smp_cpu_not_running(cpu))
678 printk("Processor %d spun down\n", cpu);
684 signal_processor(smp_processor_id(), sigp_stop);
689 #endif /* CONFIG_HOTPLUG_CPU */
692 * Cycle through the processors and setup structures.
695 void __init smp_prepare_cpus(unsigned int max_cpus)
701 /* request the 0x1201 emergency signal external interrupt */
702 if (register_external_interrupt(0x1201, do_ext_call_interrupt) != 0)
703 panic("Couldn't request external interrupt 0x1201");
704 memset(lowcore_ptr, 0, sizeof(lowcore_ptr));
706 * Initialize prefix pages and stacks for all possible cpus
708 print_cpu_info(&S390_lowcore.cpu_data);
710 for_each_possible_cpu(i) {
711 lowcore_ptr[i] = (struct _lowcore *)
712 __get_free_pages(GFP_KERNEL | GFP_DMA,
713 sizeof(void*) == 8 ? 1 : 0);
714 stack = __get_free_pages(GFP_KERNEL, ASYNC_ORDER);
715 if (!lowcore_ptr[i] || !stack)
716 panic("smp_boot_cpus failed to allocate memory\n");
718 *(lowcore_ptr[i]) = S390_lowcore;
719 lowcore_ptr[i]->async_stack = stack + ASYNC_SIZE;
720 stack = __get_free_pages(GFP_KERNEL, 0);
722 panic("smp_boot_cpus failed to allocate memory\n");
723 lowcore_ptr[i]->panic_stack = stack + PAGE_SIZE;
725 if (MACHINE_HAS_IEEE) {
726 lowcore_ptr[i]->extended_save_area_addr =
727 (__u32) __get_free_pages(GFP_KERNEL, 0);
728 if (!lowcore_ptr[i]->extended_save_area_addr)
729 panic("smp_boot_cpus failed to "
730 "allocate memory\n");
735 if (MACHINE_HAS_IEEE)
736 ctl_set_bit(14, 29); /* enable extended save area */
738 set_prefix((u32)(unsigned long) lowcore_ptr[smp_processor_id()]);
740 for_each_possible_cpu(cpu)
741 if (cpu != smp_processor_id())
742 smp_create_idle(cpu);
745 void __devinit smp_prepare_boot_cpu(void)
747 BUG_ON(smp_processor_id() != 0);
749 cpu_set(0, cpu_online_map);
750 S390_lowcore.percpu_offset = __per_cpu_offset[0];
751 current_set[0] = current;
754 void smp_cpus_done(unsigned int max_cpus)
756 cpu_present_map = cpu_possible_map;
760 * the frequency of the profiling timer can be changed
761 * by writing a multiplier value into /proc/profile.
763 * usually you want to run this on all CPUs ;)
765 int setup_profiling_timer(unsigned int multiplier)
770 static DEFINE_PER_CPU(struct cpu, cpu_devices);
772 static ssize_t show_capability(struct sys_device *dev, char *buf)
774 unsigned int capability;
777 rc = get_cpu_capability(&capability);
780 return sprintf(buf, "%u\n", capability);
782 static SYSDEV_ATTR(capability, 0444, show_capability, NULL);
784 static int __cpuinit smp_cpu_notify(struct notifier_block *self,
785 unsigned long action, void *hcpu)
787 unsigned int cpu = (unsigned int)(long)hcpu;
788 struct cpu *c = &per_cpu(cpu_devices, cpu);
789 struct sys_device *s = &c->sysdev;
793 if (sysdev_create_file(s, &attr_capability))
797 sysdev_remove_file(s, &attr_capability);
803 static struct notifier_block __cpuinitdata smp_cpu_nb = {
804 .notifier_call = smp_cpu_notify,
807 static int __init topology_init(void)
811 register_cpu_notifier(&smp_cpu_nb);
813 for_each_possible_cpu(cpu) {
814 struct cpu *c = &per_cpu(cpu_devices, cpu);
815 struct sys_device *s = &c->sysdev;
818 register_cpu(c, cpu);
819 if (!cpu_online(cpu))
822 sysdev_create_file(s, &attr_capability);
826 subsys_initcall(topology_init);