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/err.h>
27 #include <linux/spinlock.h>
28 #include <linux/kernel_stat.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(&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
))
134 while (!cpus_equal(map
, data
.finished
))
136 spin_unlock(&call_lock
);
147 * @func: the function to run; this must be fast and non-blocking
148 * @info: an arbitrary pointer to pass to the function
150 * @wait: if true, wait (atomically) until function has completed on other CPUs
152 * Run a function on all other CPUs.
154 * You must not call this function with disabled interrupts, from a
155 * hardware interrupt handler or from a bottom half.
157 int smp_call_function(void (*func
) (void *info
), void *info
, int nonatomic
,
163 map
= cpu_online_map
;
164 cpu_clear(smp_processor_id(), map
);
165 __smp_call_function_map(func
, info
, nonatomic
, wait
, map
);
169 EXPORT_SYMBOL(smp_call_function
);
172 * smp_call_function_single:
173 * @cpu: the CPU where func should run
174 * @func: the function to run; this must be fast and non-blocking
175 * @info: an arbitrary pointer to pass to the function
177 * @wait: if true, wait (atomically) until function has completed on other CPUs
179 * Run a function on one processor.
181 * You must not call this function with disabled interrupts, from a
182 * hardware interrupt handler or from a bottom half.
184 int smp_call_function_single(int cpu
, void (*func
) (void *info
), void *info
,
185 int nonatomic
, int wait
)
188 __smp_call_function_map(func
, info
, nonatomic
, wait
,
189 cpumask_of_cpu(cpu
));
193 EXPORT_SYMBOL(smp_call_function_single
);
195 static void do_send_stop(void)
199 /* stop all processors */
200 for_each_online_cpu(cpu
) {
201 if (cpu
== smp_processor_id())
204 rc
= signal_processor(cpu
, sigp_stop
);
205 } while (rc
== sigp_busy
);
209 static void do_store_status(void)
213 /* store status of all processors in their lowcores (real 0) */
214 for_each_online_cpu(cpu
) {
215 if (cpu
== smp_processor_id())
218 rc
= signal_processor_p(
219 (__u32
)(unsigned long) lowcore_ptr
[cpu
], cpu
,
220 sigp_store_status_at_address
);
221 } while (rc
== sigp_busy
);
225 static void do_wait_for_stop(void)
229 /* Wait for all other cpus to enter stopped state */
230 for_each_online_cpu(cpu
) {
231 if (cpu
== smp_processor_id())
233 while (!smp_cpu_not_running(cpu
))
239 * this function sends a 'stop' sigp to all other CPUs in the system.
240 * it goes straight through.
242 void smp_send_stop(void)
244 /* Disable all interrupts/machine checks */
245 __load_psw_mask(psw_kernel_bits
& ~PSW_MASK_MCHECK
);
247 /* write magic number to zero page (absolute 0) */
248 lowcore_ptr
[smp_processor_id()]->panic_magic
= __PANIC_MAGIC
;
250 /* stop other processors. */
253 /* wait until other processors are stopped */
256 /* store status of other processors. */
261 * Reboot, halt and power_off routines for SMP.
263 void machine_restart_smp(char *__unused
)
269 void machine_halt_smp(void)
272 if (MACHINE_IS_VM
&& strlen(vmhalt_cmd
) > 0)
273 __cpcmd(vmhalt_cmd
, NULL
, 0, NULL
);
274 signal_processor(smp_processor_id(), sigp_stop_and_store_status
);
278 void machine_power_off_smp(void)
281 if (MACHINE_IS_VM
&& strlen(vmpoff_cmd
) > 0)
282 __cpcmd(vmpoff_cmd
, NULL
, 0, NULL
);
283 signal_processor(smp_processor_id(), sigp_stop_and_store_status
);
288 * This is the main routine where commands issued by other
292 static void do_ext_call_interrupt(__u16 code
)
297 * handle bit signal external calls
299 * For the ec_schedule signal we have to do nothing. All the work
300 * is done automatically when we return from the interrupt.
302 bits
= xchg(&S390_lowcore
.ext_call_fast
, 0);
304 if (test_bit(ec_call_function
, &bits
))
309 * Send an external call sigp to another cpu and return without waiting
310 * for its completion.
312 static void smp_ext_bitcall(int cpu
, ec_bit_sig sig
)
315 * Set signaling bit in lowcore of target cpu and kick it
317 set_bit(sig
, (unsigned long *) &lowcore_ptr
[cpu
]->ext_call_fast
);
318 while (signal_processor(cpu
, sigp_emergency_signal
) == sigp_busy
)
324 * this function sends a 'purge tlb' signal to another CPU.
326 void smp_ptlb_callback(void *info
)
331 void smp_ptlb_all(void)
333 on_each_cpu(smp_ptlb_callback
, NULL
, 0, 1);
335 EXPORT_SYMBOL(smp_ptlb_all
);
336 #endif /* ! CONFIG_64BIT */
339 * this function sends a 'reschedule' IPI to another CPU.
340 * it goes straight through and wastes no time serializing
341 * anything. Worst case is that we lose a reschedule ...
343 void smp_send_reschedule(int cpu
)
345 smp_ext_bitcall(cpu
, ec_schedule
);
349 * parameter area for the set/clear control bit callbacks
351 struct ec_creg_mask_parms
{
352 unsigned long orvals
[16];
353 unsigned long andvals
[16];
357 * callback for setting/clearing control bits
359 static void smp_ctl_bit_callback(void *info
)
361 struct ec_creg_mask_parms
*pp
= info
;
362 unsigned long cregs
[16];
365 __ctl_store(cregs
, 0, 15);
366 for (i
= 0; i
<= 15; i
++)
367 cregs
[i
] = (cregs
[i
] & pp
->andvals
[i
]) | pp
->orvals
[i
];
368 __ctl_load(cregs
, 0, 15);
372 * Set a bit in a control register of all cpus
374 void smp_ctl_set_bit(int cr
, int bit
)
376 struct ec_creg_mask_parms parms
;
378 memset(&parms
.orvals
, 0, sizeof(parms
.orvals
));
379 memset(&parms
.andvals
, 0xff, sizeof(parms
.andvals
));
380 parms
.orvals
[cr
] = 1 << bit
;
381 on_each_cpu(smp_ctl_bit_callback
, &parms
, 0, 1);
383 EXPORT_SYMBOL(smp_ctl_set_bit
);
386 * Clear a bit in a control register of all cpus
388 void smp_ctl_clear_bit(int cr
, int bit
)
390 struct ec_creg_mask_parms parms
;
392 memset(&parms
.orvals
, 0, sizeof(parms
.orvals
));
393 memset(&parms
.andvals
, 0xff, sizeof(parms
.andvals
));
394 parms
.andvals
[cr
] = ~(1L << bit
);
395 on_each_cpu(smp_ctl_bit_callback
, &parms
, 0, 1);
397 EXPORT_SYMBOL(smp_ctl_clear_bit
);
399 #if defined(CONFIG_ZFCPDUMP) || defined(CONFIG_ZFCPDUMP_MODULE)
402 * zfcpdump_prefix_array holds prefix registers for the following scenario:
403 * 64 bit zfcpdump kernel and 31 bit kernel which is to be dumped. We have to
404 * save its prefix registers, since they get lost, when switching from 31 bit
407 unsigned int zfcpdump_prefix_array
[NR_CPUS
+ 1] \
408 __attribute__((__section__(".data")));
410 static void __init
smp_get_save_area(unsigned int cpu
, unsigned int phy_cpu
)
412 if (ipl_info
.type
!= IPL_TYPE_FCP_DUMP
)
414 if (cpu
>= NR_CPUS
) {
415 printk(KERN_WARNING
"Registers for cpu %i not saved since dump "
416 "kernel was compiled with NR_CPUS=%i\n", cpu
, NR_CPUS
);
419 zfcpdump_save_areas
[cpu
] = alloc_bootmem(sizeof(union save_area
));
420 __cpu_logical_map
[1] = (__u16
) phy_cpu
;
421 while (signal_processor(1, sigp_stop_and_store_status
) == sigp_busy
)
423 memcpy(zfcpdump_save_areas
[cpu
],
424 (void *)(unsigned long) store_prefix() + SAVE_AREA_BASE
,
427 /* copy original prefix register */
428 zfcpdump_save_areas
[cpu
]->s390x
.pref_reg
= zfcpdump_prefix_array
[cpu
];
432 union save_area
*zfcpdump_save_areas
[NR_CPUS
+ 1];
433 EXPORT_SYMBOL_GPL(zfcpdump_save_areas
);
437 static inline void smp_get_save_area(unsigned int cpu
, unsigned int phy_cpu
) { }
439 #endif /* CONFIG_ZFCPDUMP || CONFIG_ZFCPDUMP_MODULE */
442 * Lets check how many CPUs we have.
444 static unsigned int __init
smp_count_cpus(void)
446 unsigned int cpu
, num_cpus
;
450 * cpu 0 is the boot cpu. See smp_prepare_boot_cpu.
452 boot_cpu_addr
= S390_lowcore
.cpu_data
.cpu_addr
;
453 current_thread_info()->cpu
= 0;
455 for (cpu
= 0; cpu
<= 65535; cpu
++) {
456 if ((__u16
) cpu
== boot_cpu_addr
)
458 __cpu_logical_map
[1] = (__u16
) cpu
;
459 if (signal_processor(1, sigp_sense
) == sigp_not_operational
)
461 smp_get_save_area(num_cpus
, cpu
);
464 printk("Detected %d CPU's\n", (int) num_cpus
);
465 printk("Boot cpu address %2X\n", boot_cpu_addr
);
470 * Activate a secondary processor.
472 int __cpuinit
start_secondary(void *cpuvoid
)
477 /* Enable TOD clock interrupts on the secondary cpu. */
479 #ifdef CONFIG_VIRT_TIMER
480 /* Enable cpu timer interrupts on the secondary cpu. */
483 /* Enable pfault pseudo page faults on this cpu. */
486 /* Mark this cpu as online */
487 cpu_set(smp_processor_id(), cpu_online_map
);
488 /* Switch on interrupts */
490 /* Print info about this processor */
491 print_cpu_info(&S390_lowcore
.cpu_data
);
492 /* cpu_idle will call schedule for us */
497 static void __init
smp_create_idle(unsigned int cpu
)
499 struct task_struct
*p
;
502 * don't care about the psw and regs settings since we'll never
503 * reschedule the forked task.
507 panic("failed fork for CPU %u: %li", cpu
, PTR_ERR(p
));
508 current_set
[cpu
] = p
;
511 static int cpu_stopped(int cpu
)
515 /* Check for stopped state */
516 if (signal_processor_ps(&status
, 0, cpu
, sigp_sense
) ==
517 sigp_status_stored
) {
524 /* Upping and downing of CPUs */
526 int __cpu_up(unsigned int cpu
)
528 struct task_struct
*idle
;
529 struct _lowcore
*cpu_lowcore
;
530 struct stack_frame
*sf
;
534 for (curr_cpu
= 0; curr_cpu
<= 65535; curr_cpu
++) {
535 __cpu_logical_map
[cpu
] = (__u16
) curr_cpu
;
536 if (cpu_stopped(cpu
))
540 if (!cpu_stopped(cpu
))
543 ccode
= signal_processor_p((__u32
)(unsigned long)(lowcore_ptr
[cpu
]),
544 cpu
, sigp_set_prefix
);
546 printk("sigp_set_prefix failed for cpu %d "
547 "with condition code %d\n",
548 (int) cpu
, (int) ccode
);
552 idle
= current_set
[cpu
];
553 cpu_lowcore
= lowcore_ptr
[cpu
];
554 cpu_lowcore
->kernel_stack
= (unsigned long)
555 task_stack_page(idle
) + THREAD_SIZE
;
556 sf
= (struct stack_frame
*) (cpu_lowcore
->kernel_stack
557 - sizeof(struct pt_regs
)
558 - sizeof(struct stack_frame
));
559 memset(sf
, 0, sizeof(struct stack_frame
));
560 sf
->gprs
[9] = (unsigned long) sf
;
561 cpu_lowcore
->save_area
[15] = (unsigned long) sf
;
562 __ctl_store(cpu_lowcore
->cregs_save_area
[0], 0, 15);
565 : : "a" (&cpu_lowcore
->access_regs_save_area
) : "memory");
566 cpu_lowcore
->percpu_offset
= __per_cpu_offset
[cpu
];
567 cpu_lowcore
->current_task
= (unsigned long) idle
;
568 cpu_lowcore
->cpu_data
.cpu_nr
= cpu
;
571 while (signal_processor(cpu
, sigp_restart
) == sigp_busy
)
574 while (!cpu_online(cpu
))
579 static unsigned int __initdata additional_cpus
;
580 static unsigned int __initdata possible_cpus
;
582 void __init
smp_setup_cpu_possible_map(void)
584 unsigned int phy_cpus
, pos_cpus
, cpu
;
586 phy_cpus
= smp_count_cpus();
587 pos_cpus
= min(phy_cpus
+ additional_cpus
, (unsigned int) NR_CPUS
);
590 pos_cpus
= min(possible_cpus
, (unsigned int) NR_CPUS
);
592 for (cpu
= 0; cpu
< pos_cpus
; cpu
++)
593 cpu_set(cpu
, cpu_possible_map
);
595 phy_cpus
= min(phy_cpus
, pos_cpus
);
597 for (cpu
= 0; cpu
< phy_cpus
; cpu
++)
598 cpu_set(cpu
, cpu_present_map
);
601 #ifdef CONFIG_HOTPLUG_CPU
603 static int __init
setup_additional_cpus(char *s
)
605 additional_cpus
= simple_strtoul(s
, NULL
, 0);
608 early_param("additional_cpus", setup_additional_cpus
);
610 static int __init
setup_possible_cpus(char *s
)
612 possible_cpus
= simple_strtoul(s
, NULL
, 0);
615 early_param("possible_cpus", setup_possible_cpus
);
617 int __cpu_disable(void)
619 struct ec_creg_mask_parms cr_parms
;
620 int cpu
= smp_processor_id();
622 cpu_clear(cpu
, cpu_online_map
);
624 /* Disable pfault pseudo page faults on this cpu. */
627 memset(&cr_parms
.orvals
, 0, sizeof(cr_parms
.orvals
));
628 memset(&cr_parms
.andvals
, 0xff, sizeof(cr_parms
.andvals
));
630 /* disable all external interrupts */
631 cr_parms
.orvals
[0] = 0;
632 cr_parms
.andvals
[0] = ~(1 << 15 | 1 << 14 | 1 << 13 | 1 << 12 |
633 1 << 11 | 1 << 10 | 1 << 6 | 1 << 4);
634 /* disable all I/O interrupts */
635 cr_parms
.orvals
[6] = 0;
636 cr_parms
.andvals
[6] = ~(1 << 31 | 1 << 30 | 1 << 29 | 1 << 28 |
637 1 << 27 | 1 << 26 | 1 << 25 | 1 << 24);
638 /* disable most machine checks */
639 cr_parms
.orvals
[14] = 0;
640 cr_parms
.andvals
[14] = ~(1 << 28 | 1 << 27 | 1 << 26 |
643 smp_ctl_bit_callback(&cr_parms
);
648 void __cpu_die(unsigned int cpu
)
650 /* Wait until target cpu is down */
651 while (!smp_cpu_not_running(cpu
))
653 printk("Processor %d spun down\n", cpu
);
659 signal_processor(smp_processor_id(), sigp_stop
);
664 #endif /* CONFIG_HOTPLUG_CPU */
667 * Cycle through the processors and setup structures.
670 void __init
smp_prepare_cpus(unsigned int max_cpus
)
676 /* request the 0x1201 emergency signal external interrupt */
677 if (register_external_interrupt(0x1201, do_ext_call_interrupt
) != 0)
678 panic("Couldn't request external interrupt 0x1201");
679 memset(lowcore_ptr
, 0, sizeof(lowcore_ptr
));
681 * Initialize prefix pages and stacks for all possible cpus
683 print_cpu_info(&S390_lowcore
.cpu_data
);
685 for_each_possible_cpu(i
) {
686 lowcore_ptr
[i
] = (struct _lowcore
*)
687 __get_free_pages(GFP_KERNEL
| GFP_DMA
,
688 sizeof(void*) == 8 ? 1 : 0);
689 stack
= __get_free_pages(GFP_KERNEL
, ASYNC_ORDER
);
690 if (!lowcore_ptr
[i
] || !stack
)
691 panic("smp_boot_cpus failed to allocate memory\n");
693 *(lowcore_ptr
[i
]) = S390_lowcore
;
694 lowcore_ptr
[i
]->async_stack
= stack
+ ASYNC_SIZE
;
695 stack
= __get_free_pages(GFP_KERNEL
, 0);
697 panic("smp_boot_cpus failed to allocate memory\n");
698 lowcore_ptr
[i
]->panic_stack
= stack
+ PAGE_SIZE
;
700 if (MACHINE_HAS_IEEE
) {
701 lowcore_ptr
[i
]->extended_save_area_addr
=
702 (__u32
) __get_free_pages(GFP_KERNEL
, 0);
703 if (!lowcore_ptr
[i
]->extended_save_area_addr
)
704 panic("smp_boot_cpus failed to "
705 "allocate memory\n");
710 if (MACHINE_HAS_IEEE
)
711 ctl_set_bit(14, 29); /* enable extended save area */
713 set_prefix((u32
)(unsigned long) lowcore_ptr
[smp_processor_id()]);
715 for_each_possible_cpu(cpu
)
716 if (cpu
!= smp_processor_id())
717 smp_create_idle(cpu
);
720 void __init
smp_prepare_boot_cpu(void)
722 BUG_ON(smp_processor_id() != 0);
724 cpu_set(0, cpu_online_map
);
725 S390_lowcore
.percpu_offset
= __per_cpu_offset
[0];
726 current_set
[0] = current
;
729 void __init
smp_cpus_done(unsigned int max_cpus
)
731 cpu_present_map
= cpu_possible_map
;
735 * the frequency of the profiling timer can be changed
736 * by writing a multiplier value into /proc/profile.
738 * usually you want to run this on all CPUs ;)
740 int setup_profiling_timer(unsigned int multiplier
)
745 static DEFINE_PER_CPU(struct cpu
, cpu_devices
);
747 static ssize_t
show_capability(struct sys_device
*dev
, char *buf
)
749 unsigned int capability
;
752 rc
= get_cpu_capability(&capability
);
755 return sprintf(buf
, "%u\n", capability
);
757 static SYSDEV_ATTR(capability
, 0444, show_capability
, NULL
);
759 static int __cpuinit
smp_cpu_notify(struct notifier_block
*self
,
760 unsigned long action
, void *hcpu
)
762 unsigned int cpu
= (unsigned int)(long)hcpu
;
763 struct cpu
*c
= &per_cpu(cpu_devices
, cpu
);
764 struct sys_device
*s
= &c
->sysdev
;
768 case CPU_ONLINE_FROZEN
:
769 if (sysdev_create_file(s
, &attr_capability
))
773 case CPU_DEAD_FROZEN
:
774 sysdev_remove_file(s
, &attr_capability
);
780 static struct notifier_block __cpuinitdata smp_cpu_nb
= {
781 .notifier_call
= smp_cpu_notify
,
784 static int __init
topology_init(void)
788 register_cpu_notifier(&smp_cpu_nb
);
790 for_each_possible_cpu(cpu
) {
791 struct cpu
*c
= &per_cpu(cpu_devices
, cpu
);
792 struct sys_device
*s
= &c
->sysdev
;
795 register_cpu(c
, cpu
);
796 if (!cpu_online(cpu
))
799 sysdev_create_file(s
, &attr_capability
);
803 subsys_initcall(topology_init
);