sched: cleanup: refactor normalize_rt_tasks
[linux-2.6/kvm.git] / arch / powerpc / kernel / smp.c
blobd30f08fa029772dff196d73b6177924e27aeebc4
1 /*
2 * SMP support for ppc.
4 * Written by Cort Dougan (cort@cs.nmt.edu) borrowing a great
5 * deal of code from the sparc and intel versions.
7 * Copyright (C) 1999 Cort Dougan <cort@cs.nmt.edu>
9 * PowerPC-64 Support added by Dave Engebretsen, Peter Bergner, and
10 * Mike Corrigan {engebret|bergner|mikec}@us.ibm.com
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
18 #undef DEBUG
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/sched.h>
23 #include <linux/smp.h>
24 #include <linux/interrupt.h>
25 #include <linux/delay.h>
26 #include <linux/init.h>
27 #include <linux/spinlock.h>
28 #include <linux/cache.h>
29 #include <linux/err.h>
30 #include <linux/sysdev.h>
31 #include <linux/cpu.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
35 #include <asm/ptrace.h>
36 #include <asm/atomic.h>
37 #include <asm/irq.h>
38 #include <asm/page.h>
39 #include <asm/pgtable.h>
40 #include <asm/prom.h>
41 #include <asm/smp.h>
42 #include <asm/time.h>
43 #include <asm/machdep.h>
44 #include <asm/cputable.h>
45 #include <asm/system.h>
46 #include <asm/mpic.h>
47 #include <asm/vdso_datapage.h>
48 #ifdef CONFIG_PPC64
49 #include <asm/paca.h>
50 #endif
52 #ifdef DEBUG
53 #include <asm/udbg.h>
54 #define DBG(fmt...) udbg_printf(fmt)
55 #else
56 #define DBG(fmt...)
57 #endif
59 int smp_hw_index[NR_CPUS];
60 struct thread_info *secondary_ti;
62 cpumask_t cpu_possible_map = CPU_MASK_NONE;
63 cpumask_t cpu_online_map = CPU_MASK_NONE;
64 cpumask_t cpu_sibling_map[NR_CPUS] = { [0 ... NR_CPUS-1] = CPU_MASK_NONE };
66 EXPORT_SYMBOL(cpu_online_map);
67 EXPORT_SYMBOL(cpu_possible_map);
68 EXPORT_SYMBOL(cpu_sibling_map);
70 /* SMP operations for this machine */
71 struct smp_ops_t *smp_ops;
73 static volatile unsigned int cpu_callin_map[NR_CPUS];
75 void smp_call_function_interrupt(void);
77 int smt_enabled_at_boot = 1;
79 static void (*crash_ipi_function_ptr)(struct pt_regs *) = NULL;
81 #ifdef CONFIG_PPC64
82 void __devinit smp_generic_kick_cpu(int nr)
84 BUG_ON(nr < 0 || nr >= NR_CPUS);
87 * The processor is currently spinning, waiting for the
88 * cpu_start field to become non-zero After we set cpu_start,
89 * the processor will continue on to secondary_start
91 paca[nr].cpu_start = 1;
92 smp_mb();
94 #endif
96 void smp_message_recv(int msg)
98 switch(msg) {
99 case PPC_MSG_CALL_FUNCTION:
100 smp_call_function_interrupt();
101 break;
102 case PPC_MSG_RESCHEDULE:
103 /* XXX Do we have to do this? */
104 set_need_resched();
105 break;
106 case PPC_MSG_DEBUGGER_BREAK:
107 if (crash_ipi_function_ptr) {
108 crash_ipi_function_ptr(get_irq_regs());
109 break;
111 #ifdef CONFIG_DEBUGGER
112 debugger_ipi(get_irq_regs());
113 break;
114 #endif /* CONFIG_DEBUGGER */
115 /* FALLTHROUGH */
116 default:
117 printk("SMP %d: smp_message_recv(): unknown msg %d\n",
118 smp_processor_id(), msg);
119 break;
123 void smp_send_reschedule(int cpu)
125 if (likely(smp_ops))
126 smp_ops->message_pass(cpu, PPC_MSG_RESCHEDULE);
129 #ifdef CONFIG_DEBUGGER
130 void smp_send_debugger_break(int cpu)
132 if (likely(smp_ops))
133 smp_ops->message_pass(cpu, PPC_MSG_DEBUGGER_BREAK);
135 #endif
137 #ifdef CONFIG_KEXEC
138 void crash_send_ipi(void (*crash_ipi_callback)(struct pt_regs *))
140 crash_ipi_function_ptr = crash_ipi_callback;
141 if (crash_ipi_callback && smp_ops) {
142 mb();
143 smp_ops->message_pass(MSG_ALL_BUT_SELF, PPC_MSG_DEBUGGER_BREAK);
146 #endif
148 static void stop_this_cpu(void *dummy)
150 local_irq_disable();
151 while (1)
156 * Structure and data for smp_call_function(). This is designed to minimise
157 * static memory requirements. It also looks cleaner.
158 * Stolen from the i386 version.
160 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(call_lock);
162 static struct call_data_struct {
163 void (*func) (void *info);
164 void *info;
165 atomic_t started;
166 atomic_t finished;
167 int wait;
168 } *call_data;
170 /* delay of at least 8 seconds */
171 #define SMP_CALL_TIMEOUT 8
174 * These functions send a 'generic call function' IPI to other online
175 * CPUS in the system.
177 * [SUMMARY] Run a function on other CPUs.
178 * <func> The function to run. This must be fast and non-blocking.
179 * <info> An arbitrary pointer to pass to the function.
180 * <nonatomic> currently unused.
181 * <wait> If true, wait (atomically) until function has completed on other CPUs.
182 * [RETURNS] 0 on success, else a negative status code. Does not return until
183 * remote CPUs are nearly ready to execute <<func>> or are or have executed.
185 * You must not call this function with disabled interrupts or from a
186 * hardware interrupt handler or from a bottom half handler.
188 int smp_call_function_map(void (*func) (void *info), void *info, int nonatomic,
189 int wait, cpumask_t map)
191 struct call_data_struct data;
192 int ret = -1, num_cpus;
193 int cpu;
194 u64 timeout;
196 if (unlikely(smp_ops == NULL))
197 return ret;
199 data.func = func;
200 data.info = info;
201 atomic_set(&data.started, 0);
202 data.wait = wait;
203 if (wait)
204 atomic_set(&data.finished, 0);
206 spin_lock(&call_lock);
208 /* remove 'self' from the map */
209 if (cpu_isset(smp_processor_id(), map))
210 cpu_clear(smp_processor_id(), map);
212 /* sanity check the map, remove any non-online processors. */
213 cpus_and(map, map, cpu_online_map);
215 num_cpus = cpus_weight(map);
216 if (!num_cpus)
217 goto done;
219 call_data = &data;
220 smp_wmb();
221 /* Send a message to all CPUs in the map */
222 for_each_cpu_mask(cpu, map)
223 smp_ops->message_pass(cpu, PPC_MSG_CALL_FUNCTION);
225 timeout = get_tb() + (u64) SMP_CALL_TIMEOUT * tb_ticks_per_sec;
227 /* Wait for indication that they have received the message */
228 while (atomic_read(&data.started) != num_cpus) {
229 HMT_low();
230 if (get_tb() >= timeout) {
231 printk("smp_call_function on cpu %d: other cpus not "
232 "responding (%d)\n", smp_processor_id(),
233 atomic_read(&data.started));
234 debugger(NULL);
235 goto out;
239 /* optionally wait for the CPUs to complete */
240 if (wait) {
241 while (atomic_read(&data.finished) != num_cpus) {
242 HMT_low();
243 if (get_tb() >= timeout) {
244 printk("smp_call_function on cpu %d: other "
245 "cpus not finishing (%d/%d)\n",
246 smp_processor_id(),
247 atomic_read(&data.finished),
248 atomic_read(&data.started));
249 debugger(NULL);
250 goto out;
255 done:
256 ret = 0;
258 out:
259 call_data = NULL;
260 HMT_medium();
261 spin_unlock(&call_lock);
262 return ret;
265 static int __smp_call_function(void (*func)(void *info), void *info,
266 int nonatomic, int wait)
268 return smp_call_function_map(func,info,nonatomic,wait,cpu_online_map);
271 int smp_call_function(void (*func) (void *info), void *info, int nonatomic,
272 int wait)
274 /* Can deadlock when called with interrupts disabled */
275 WARN_ON(irqs_disabled());
277 return __smp_call_function(func, info, nonatomic, wait);
279 EXPORT_SYMBOL(smp_call_function);
281 int smp_call_function_single(int cpu, void (*func) (void *info), void *info, int nonatomic,
282 int wait)
284 cpumask_t map = CPU_MASK_NONE;
285 int ret = 0;
287 /* Can deadlock when called with interrupts disabled */
288 WARN_ON(irqs_disabled());
290 if (!cpu_online(cpu))
291 return -EINVAL;
293 cpu_set(cpu, map);
294 if (cpu != get_cpu())
295 ret = smp_call_function_map(func,info,nonatomic,wait,map);
296 else {
297 local_irq_disable();
298 func(info);
299 local_irq_enable();
301 put_cpu();
302 return ret;
304 EXPORT_SYMBOL(smp_call_function_single);
306 void smp_send_stop(void)
308 __smp_call_function(stop_this_cpu, NULL, 1, 0);
311 void smp_call_function_interrupt(void)
313 void (*func) (void *info);
314 void *info;
315 int wait;
317 /* call_data will be NULL if the sender timed out while
318 * waiting on us to receive the call.
320 if (!call_data)
321 return;
323 func = call_data->func;
324 info = call_data->info;
325 wait = call_data->wait;
327 if (!wait)
328 smp_mb__before_atomic_inc();
331 * Notify initiating CPU that I've grabbed the data and am
332 * about to execute the function
334 atomic_inc(&call_data->started);
336 * At this point the info structure may be out of scope unless wait==1
338 (*func)(info);
339 if (wait) {
340 smp_mb__before_atomic_inc();
341 atomic_inc(&call_data->finished);
345 extern struct gettimeofday_struct do_gtod;
347 struct thread_info *current_set[NR_CPUS];
349 DECLARE_PER_CPU(unsigned int, pvr);
351 static void __devinit smp_store_cpu_info(int id)
353 per_cpu(pvr, id) = mfspr(SPRN_PVR);
356 static void __init smp_create_idle(unsigned int cpu)
358 struct task_struct *p;
360 /* create a process for the processor */
361 p = fork_idle(cpu);
362 if (IS_ERR(p))
363 panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
364 #ifdef CONFIG_PPC64
365 paca[cpu].__current = p;
366 #endif
367 current_set[cpu] = task_thread_info(p);
368 task_thread_info(p)->cpu = cpu;
371 void __init smp_prepare_cpus(unsigned int max_cpus)
373 unsigned int cpu;
375 DBG("smp_prepare_cpus\n");
378 * setup_cpu may need to be called on the boot cpu. We havent
379 * spun any cpus up but lets be paranoid.
381 BUG_ON(boot_cpuid != smp_processor_id());
383 /* Fixup boot cpu */
384 smp_store_cpu_info(boot_cpuid);
385 cpu_callin_map[boot_cpuid] = 1;
387 if (smp_ops)
388 max_cpus = smp_ops->probe();
389 else
390 max_cpus = 1;
392 smp_space_timers(max_cpus);
394 for_each_possible_cpu(cpu)
395 if (cpu != boot_cpuid)
396 smp_create_idle(cpu);
399 void __devinit smp_prepare_boot_cpu(void)
401 BUG_ON(smp_processor_id() != boot_cpuid);
403 cpu_set(boot_cpuid, cpu_online_map);
404 #ifdef CONFIG_PPC64
405 paca[boot_cpuid].__current = current;
406 #endif
407 current_set[boot_cpuid] = task_thread_info(current);
410 #ifdef CONFIG_HOTPLUG_CPU
411 /* State of each CPU during hotplug phases */
412 DEFINE_PER_CPU(int, cpu_state) = { 0 };
414 int generic_cpu_disable(void)
416 unsigned int cpu = smp_processor_id();
418 if (cpu == boot_cpuid)
419 return -EBUSY;
421 cpu_clear(cpu, cpu_online_map);
422 #ifdef CONFIG_PPC64
423 vdso_data->processorCount--;
424 fixup_irqs(cpu_online_map);
425 #endif
426 return 0;
429 int generic_cpu_enable(unsigned int cpu)
431 /* Do the normal bootup if we haven't
432 * already bootstrapped. */
433 if (system_state != SYSTEM_RUNNING)
434 return -ENOSYS;
436 /* get the target out of it's holding state */
437 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
438 smp_wmb();
440 while (!cpu_online(cpu))
441 cpu_relax();
443 #ifdef CONFIG_PPC64
444 fixup_irqs(cpu_online_map);
445 /* counter the irq disable in fixup_irqs */
446 local_irq_enable();
447 #endif
448 return 0;
451 void generic_cpu_die(unsigned int cpu)
453 int i;
455 for (i = 0; i < 100; i++) {
456 smp_rmb();
457 if (per_cpu(cpu_state, cpu) == CPU_DEAD)
458 return;
459 msleep(100);
461 printk(KERN_ERR "CPU%d didn't die...\n", cpu);
464 void generic_mach_cpu_die(void)
466 unsigned int cpu;
468 local_irq_disable();
469 cpu = smp_processor_id();
470 printk(KERN_DEBUG "CPU%d offline\n", cpu);
471 __get_cpu_var(cpu_state) = CPU_DEAD;
472 smp_wmb();
473 while (__get_cpu_var(cpu_state) != CPU_UP_PREPARE)
474 cpu_relax();
475 cpu_set(cpu, cpu_online_map);
476 local_irq_enable();
478 #endif
480 static int __devinit cpu_enable(unsigned int cpu)
482 if (smp_ops && smp_ops->cpu_enable)
483 return smp_ops->cpu_enable(cpu);
485 return -ENOSYS;
488 int __cpuinit __cpu_up(unsigned int cpu)
490 int c;
492 secondary_ti = current_set[cpu];
493 if (!cpu_enable(cpu))
494 return 0;
496 if (smp_ops == NULL ||
497 (smp_ops->cpu_bootable && !smp_ops->cpu_bootable(cpu)))
498 return -EINVAL;
500 /* Make sure callin-map entry is 0 (can be leftover a CPU
501 * hotplug
503 cpu_callin_map[cpu] = 0;
505 /* The information for processor bringup must
506 * be written out to main store before we release
507 * the processor.
509 smp_mb();
511 /* wake up cpus */
512 DBG("smp: kicking cpu %d\n", cpu);
513 smp_ops->kick_cpu(cpu);
516 * wait to see if the cpu made a callin (is actually up).
517 * use this value that I found through experimentation.
518 * -- Cort
520 if (system_state < SYSTEM_RUNNING)
521 for (c = 50000; c && !cpu_callin_map[cpu]; c--)
522 udelay(100);
523 #ifdef CONFIG_HOTPLUG_CPU
524 else
526 * CPUs can take much longer to come up in the
527 * hotplug case. Wait five seconds.
529 for (c = 25; c && !cpu_callin_map[cpu]; c--) {
530 msleep(200);
532 #endif
534 if (!cpu_callin_map[cpu]) {
535 printk("Processor %u is stuck.\n", cpu);
536 return -ENOENT;
539 printk("Processor %u found.\n", cpu);
541 if (smp_ops->give_timebase)
542 smp_ops->give_timebase();
544 /* Wait until cpu puts itself in the online map */
545 while (!cpu_online(cpu))
546 cpu_relax();
548 return 0;
552 /* Activate a secondary processor. */
553 int __devinit start_secondary(void *unused)
555 unsigned int cpu = smp_processor_id();
557 atomic_inc(&init_mm.mm_count);
558 current->active_mm = &init_mm;
560 smp_store_cpu_info(cpu);
561 set_dec(tb_ticks_per_jiffy);
562 preempt_disable();
563 cpu_callin_map[cpu] = 1;
565 smp_ops->setup_cpu(cpu);
566 if (smp_ops->take_timebase)
567 smp_ops->take_timebase();
569 if (system_state > SYSTEM_BOOTING)
570 snapshot_timebase();
572 secondary_cpu_time_init();
574 spin_lock(&call_lock);
575 cpu_set(cpu, cpu_online_map);
576 spin_unlock(&call_lock);
578 local_irq_enable();
580 cpu_idle();
581 return 0;
584 int setup_profiling_timer(unsigned int multiplier)
586 return 0;
589 void __init smp_cpus_done(unsigned int max_cpus)
591 cpumask_t old_mask;
593 /* We want the setup_cpu() here to be called from CPU 0, but our
594 * init thread may have been "borrowed" by another CPU in the meantime
595 * se we pin us down to CPU 0 for a short while
597 old_mask = current->cpus_allowed;
598 set_cpus_allowed(current, cpumask_of_cpu(boot_cpuid));
600 if (smp_ops)
601 smp_ops->setup_cpu(boot_cpuid);
603 set_cpus_allowed(current, old_mask);
605 snapshot_timebases();
607 dump_numa_cpu_topology();
610 #ifdef CONFIG_HOTPLUG_CPU
611 int __cpu_disable(void)
613 if (smp_ops->cpu_disable)
614 return smp_ops->cpu_disable();
616 return -ENOSYS;
619 void __cpu_die(unsigned int cpu)
621 if (smp_ops->cpu_die)
622 smp_ops->cpu_die(cpu);
624 #endif