Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
[pandora-kernel.git] / arch / arm / kernel / smp.c
1 /*
2  *  linux/arch/arm/kernel/smp.c
3  *
4  *  Copyright (C) 2002 ARM Limited, All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 #include <linux/module.h>
11 #include <linux/delay.h>
12 #include <linux/init.h>
13 #include <linux/spinlock.h>
14 #include <linux/sched.h>
15 #include <linux/interrupt.h>
16 #include <linux/cache.h>
17 #include <linux/profile.h>
18 #include <linux/errno.h>
19 #include <linux/mm.h>
20 #include <linux/err.h>
21 #include <linux/cpu.h>
22 #include <linux/smp.h>
23 #include <linux/seq_file.h>
24 #include <linux/irq.h>
25 #include <linux/percpu.h>
26 #include <linux/clockchips.h>
27
28 #include <asm/atomic.h>
29 #include <asm/cacheflush.h>
30 #include <asm/cpu.h>
31 #include <asm/cputype.h>
32 #include <asm/mmu_context.h>
33 #include <asm/pgtable.h>
34 #include <asm/pgalloc.h>
35 #include <asm/processor.h>
36 #include <asm/tlbflush.h>
37 #include <asm/ptrace.h>
38 #include <asm/localtimer.h>
39 #include <asm/smp_plat.h>
40
41 /*
42  * as from 2.5, kernels no longer have an init_tasks structure
43  * so we need some other way of telling a new secondary core
44  * where to place its SVC stack
45  */
46 struct secondary_data secondary_data;
47
48 /*
49  * structures for inter-processor calls
50  * - A collection of single bit ipi messages.
51  */
52 struct ipi_data {
53         spinlock_t lock;
54         unsigned long ipi_count;
55         unsigned long bits;
56 };
57
58 static DEFINE_PER_CPU(struct ipi_data, ipi_data) = {
59         .lock   = SPIN_LOCK_UNLOCKED,
60 };
61
62 enum ipi_msg_type {
63         IPI_TIMER,
64         IPI_RESCHEDULE,
65         IPI_CALL_FUNC,
66         IPI_CALL_FUNC_SINGLE,
67         IPI_CPU_STOP,
68 };
69
70 int __cpuinit __cpu_up(unsigned int cpu)
71 {
72         struct cpuinfo_arm *ci = &per_cpu(cpu_data, cpu);
73         struct task_struct *idle = ci->idle;
74         pgd_t *pgd;
75         pmd_t *pmd;
76         int ret;
77
78         /*
79          * Spawn a new process manually, if not already done.
80          * Grab a pointer to its task struct so we can mess with it
81          */
82         if (!idle) {
83                 idle = fork_idle(cpu);
84                 if (IS_ERR(idle)) {
85                         printk(KERN_ERR "CPU%u: fork() failed\n", cpu);
86                         return PTR_ERR(idle);
87                 }
88                 ci->idle = idle;
89         }
90
91         /*
92          * Allocate initial page tables to allow the new CPU to
93          * enable the MMU safely.  This essentially means a set
94          * of our "standard" page tables, with the addition of
95          * a 1:1 mapping for the physical address of the kernel.
96          */
97         pgd = pgd_alloc(&init_mm);
98         pmd = pmd_offset(pgd + pgd_index(PHYS_OFFSET), PHYS_OFFSET);
99         *pmd = __pmd((PHYS_OFFSET & PGDIR_MASK) |
100                      PMD_TYPE_SECT | PMD_SECT_AP_WRITE);
101         flush_pmd_entry(pmd);
102         outer_clean_range(__pa(pmd), __pa(pmd + 1));
103
104         /*
105          * We need to tell the secondary core where to find
106          * its stack and the page tables.
107          */
108         secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
109         secondary_data.pgdir = virt_to_phys(pgd);
110         __cpuc_flush_dcache_area(&secondary_data, sizeof(secondary_data));
111         outer_clean_range(__pa(&secondary_data), __pa(&secondary_data + 1));
112
113         /*
114          * Now bring the CPU into our world.
115          */
116         ret = boot_secondary(cpu, idle);
117         if (ret == 0) {
118                 unsigned long timeout;
119
120                 /*
121                  * CPU was successfully started, wait for it
122                  * to come online or time out.
123                  */
124                 timeout = jiffies + HZ;
125                 while (time_before(jiffies, timeout)) {
126                         if (cpu_online(cpu))
127                                 break;
128
129                         udelay(10);
130                         barrier();
131                 }
132
133                 if (!cpu_online(cpu))
134                         ret = -EIO;
135         }
136
137         secondary_data.stack = NULL;
138         secondary_data.pgdir = 0;
139
140         *pmd = __pmd(0);
141         clean_pmd_entry(pmd);
142         pgd_free(&init_mm, pgd);
143
144         if (ret) {
145                 printk(KERN_CRIT "CPU%u: processor failed to boot\n", cpu);
146
147                 /*
148                  * FIXME: We need to clean up the new idle thread. --rmk
149                  */
150         }
151
152         return ret;
153 }
154
155 #ifdef CONFIG_HOTPLUG_CPU
156 /*
157  * __cpu_disable runs on the processor to be shutdown.
158  */
159 int __cpu_disable(void)
160 {
161         unsigned int cpu = smp_processor_id();
162         struct task_struct *p;
163         int ret;
164
165         ret = mach_cpu_disable(cpu);
166         if (ret)
167                 return ret;
168
169         /*
170          * Take this CPU offline.  Once we clear this, we can't return,
171          * and we must not schedule until we're ready to give up the cpu.
172          */
173         set_cpu_online(cpu, false);
174
175         /*
176          * OK - migrate IRQs away from this CPU
177          */
178         migrate_irqs();
179
180         /*
181          * Stop the local timer for this CPU.
182          */
183         local_timer_stop();
184
185         /*
186          * Flush user cache and TLB mappings, and then remove this CPU
187          * from the vm mask set of all processes.
188          */
189         flush_cache_all();
190         local_flush_tlb_all();
191
192         read_lock(&tasklist_lock);
193         for_each_process(p) {
194                 if (p->mm)
195                         cpumask_clear_cpu(cpu, mm_cpumask(p->mm));
196         }
197         read_unlock(&tasklist_lock);
198
199         return 0;
200 }
201
202 /*
203  * called on the thread which is asking for a CPU to be shutdown -
204  * waits until shutdown has completed, or it is timed out.
205  */
206 void __cpu_die(unsigned int cpu)
207 {
208         if (!platform_cpu_kill(cpu))
209                 printk("CPU%u: unable to kill\n", cpu);
210 }
211
212 /*
213  * Called from the idle thread for the CPU which has been shutdown.
214  *
215  * Note that we disable IRQs here, but do not re-enable them
216  * before returning to the caller. This is also the behaviour
217  * of the other hotplug-cpu capable cores, so presumably coming
218  * out of idle fixes this.
219  */
220 void __ref cpu_die(void)
221 {
222         unsigned int cpu = smp_processor_id();
223
224         local_irq_disable();
225         idle_task_exit();
226
227         /*
228          * actual CPU shutdown procedure is at least platform (if not
229          * CPU) specific
230          */
231         platform_cpu_die(cpu);
232
233         /*
234          * Do not return to the idle loop - jump back to the secondary
235          * cpu initialisation.  There's some initialisation which needs
236          * to be repeated to undo the effects of taking the CPU offline.
237          */
238         __asm__("mov    sp, %0\n"
239         "       b       secondary_start_kernel"
240                 :
241                 : "r" (task_stack_page(current) + THREAD_SIZE - 8));
242 }
243 #endif /* CONFIG_HOTPLUG_CPU */
244
245 /*
246  * This is the secondary CPU boot entry.  We're using this CPUs
247  * idle thread stack, but a set of temporary page tables.
248  */
249 asmlinkage void __cpuinit secondary_start_kernel(void)
250 {
251         struct mm_struct *mm = &init_mm;
252         unsigned int cpu = smp_processor_id();
253
254         printk("CPU%u: Booted secondary processor\n", cpu);
255
256         /*
257          * All kernel threads share the same mm context; grab a
258          * reference and switch to it.
259          */
260         atomic_inc(&mm->mm_users);
261         atomic_inc(&mm->mm_count);
262         current->active_mm = mm;
263         cpumask_set_cpu(cpu, mm_cpumask(mm));
264         cpu_switch_mm(mm->pgd, mm);
265         enter_lazy_tlb(mm, current);
266         local_flush_tlb_all();
267
268         cpu_init();
269         preempt_disable();
270
271         /*
272          * Give the platform a chance to do its own initialisation.
273          */
274         platform_secondary_init(cpu);
275
276         /*
277          * Enable local interrupts.
278          */
279         notify_cpu_starting(cpu);
280         local_irq_enable();
281         local_fiq_enable();
282
283         /*
284          * Setup the percpu timer for this CPU.
285          */
286         percpu_timer_setup();
287
288         calibrate_delay();
289
290         smp_store_cpu_info(cpu);
291
292         /*
293          * OK, now it's safe to let the boot CPU continue
294          */
295         set_cpu_online(cpu, true);
296
297         /*
298          * OK, it's off to the idle thread for us
299          */
300         cpu_idle();
301 }
302
303 /*
304  * Called by both boot and secondaries to move global data into
305  * per-processor storage.
306  */
307 void __cpuinit smp_store_cpu_info(unsigned int cpuid)
308 {
309         struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
310
311         cpu_info->loops_per_jiffy = loops_per_jiffy;
312 }
313
314 void __init smp_cpus_done(unsigned int max_cpus)
315 {
316         int cpu;
317         unsigned long bogosum = 0;
318
319         for_each_online_cpu(cpu)
320                 bogosum += per_cpu(cpu_data, cpu).loops_per_jiffy;
321
322         printk(KERN_INFO "SMP: Total of %d processors activated "
323                "(%lu.%02lu BogoMIPS).\n",
324                num_online_cpus(),
325                bogosum / (500000/HZ),
326                (bogosum / (5000/HZ)) % 100);
327 }
328
329 void __init smp_prepare_boot_cpu(void)
330 {
331         unsigned int cpu = smp_processor_id();
332
333         per_cpu(cpu_data, cpu).idle = current;
334 }
335
336 static void send_ipi_message(const struct cpumask *mask, enum ipi_msg_type msg)
337 {
338         unsigned long flags;
339         unsigned int cpu;
340
341         local_irq_save(flags);
342
343         for_each_cpu(cpu, mask) {
344                 struct ipi_data *ipi = &per_cpu(ipi_data, cpu);
345
346                 spin_lock(&ipi->lock);
347                 ipi->bits |= 1 << msg;
348                 spin_unlock(&ipi->lock);
349         }
350
351         /*
352          * Call the platform specific cross-CPU call function.
353          */
354         smp_cross_call(mask);
355
356         local_irq_restore(flags);
357 }
358
359 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
360 {
361         send_ipi_message(mask, IPI_CALL_FUNC);
362 }
363
364 void arch_send_call_function_single_ipi(int cpu)
365 {
366         send_ipi_message(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
367 }
368
369 void show_ipi_list(struct seq_file *p)
370 {
371         unsigned int cpu;
372
373         seq_puts(p, "IPI:");
374
375         for_each_present_cpu(cpu)
376                 seq_printf(p, " %10lu", per_cpu(ipi_data, cpu).ipi_count);
377
378         seq_putc(p, '\n');
379 }
380
381 void show_local_irqs(struct seq_file *p)
382 {
383         unsigned int cpu;
384
385         seq_printf(p, "LOC: ");
386
387         for_each_present_cpu(cpu)
388                 seq_printf(p, "%10u ", irq_stat[cpu].local_timer_irqs);
389
390         seq_putc(p, '\n');
391 }
392
393 /*
394  * Timer (local or broadcast) support
395  */
396 static DEFINE_PER_CPU(struct clock_event_device, percpu_clockevent);
397
398 static void ipi_timer(void)
399 {
400         struct clock_event_device *evt = &__get_cpu_var(percpu_clockevent);
401         irq_enter();
402         evt->event_handler(evt);
403         irq_exit();
404 }
405
406 #ifdef CONFIG_LOCAL_TIMERS
407 asmlinkage void __exception do_local_timer(struct pt_regs *regs)
408 {
409         struct pt_regs *old_regs = set_irq_regs(regs);
410         int cpu = smp_processor_id();
411
412         if (local_timer_ack()) {
413                 irq_stat[cpu].local_timer_irqs++;
414                 ipi_timer();
415         }
416
417         set_irq_regs(old_regs);
418 }
419 #endif
420
421 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
422 static void smp_timer_broadcast(const struct cpumask *mask)
423 {
424         send_ipi_message(mask, IPI_TIMER);
425 }
426
427 static void broadcast_timer_set_mode(enum clock_event_mode mode,
428         struct clock_event_device *evt)
429 {
430 }
431
432 static void local_timer_setup(struct clock_event_device *evt)
433 {
434         evt->name       = "dummy_timer";
435         evt->features   = CLOCK_EVT_FEAT_ONESHOT |
436                           CLOCK_EVT_FEAT_PERIODIC |
437                           CLOCK_EVT_FEAT_DUMMY;
438         evt->rating     = 400;
439         evt->mult       = 1;
440         evt->set_mode   = broadcast_timer_set_mode;
441         evt->broadcast  = smp_timer_broadcast;
442
443         clockevents_register_device(evt);
444 }
445 #endif
446
447 void __cpuinit percpu_timer_setup(void)
448 {
449         unsigned int cpu = smp_processor_id();
450         struct clock_event_device *evt = &per_cpu(percpu_clockevent, cpu);
451
452         evt->cpumask = cpumask_of(cpu);
453
454         local_timer_setup(evt);
455 }
456
457 static DEFINE_SPINLOCK(stop_lock);
458
459 /*
460  * ipi_cpu_stop - handle IPI from smp_send_stop()
461  */
462 static void ipi_cpu_stop(unsigned int cpu)
463 {
464         spin_lock(&stop_lock);
465         printk(KERN_CRIT "CPU%u: stopping\n", cpu);
466         dump_stack();
467         spin_unlock(&stop_lock);
468
469         set_cpu_online(cpu, false);
470
471         local_fiq_disable();
472         local_irq_disable();
473
474         while (1)
475                 cpu_relax();
476 }
477
478 /*
479  * Main handler for inter-processor interrupts
480  *
481  * For ARM, the ipimask now only identifies a single
482  * category of IPI (Bit 1 IPIs have been replaced by a
483  * different mechanism):
484  *
485  *  Bit 0 - Inter-processor function call
486  */
487 asmlinkage void __exception do_IPI(struct pt_regs *regs)
488 {
489         unsigned int cpu = smp_processor_id();
490         struct ipi_data *ipi = &per_cpu(ipi_data, cpu);
491         struct pt_regs *old_regs = set_irq_regs(regs);
492
493         ipi->ipi_count++;
494
495         for (;;) {
496                 unsigned long msgs;
497
498                 spin_lock(&ipi->lock);
499                 msgs = ipi->bits;
500                 ipi->bits = 0;
501                 spin_unlock(&ipi->lock);
502
503                 if (!msgs)
504                         break;
505
506                 do {
507                         unsigned nextmsg;
508
509                         nextmsg = msgs & -msgs;
510                         msgs &= ~nextmsg;
511                         nextmsg = ffz(~nextmsg);
512
513                         switch (nextmsg) {
514                         case IPI_TIMER:
515                                 ipi_timer();
516                                 break;
517
518                         case IPI_RESCHEDULE:
519                                 /*
520                                  * nothing more to do - eveything is
521                                  * done on the interrupt return path
522                                  */
523                                 break;
524
525                         case IPI_CALL_FUNC:
526                                 generic_smp_call_function_interrupt();
527                                 break;
528
529                         case IPI_CALL_FUNC_SINGLE:
530                                 generic_smp_call_function_single_interrupt();
531                                 break;
532
533                         case IPI_CPU_STOP:
534                                 ipi_cpu_stop(cpu);
535                                 break;
536
537                         default:
538                                 printk(KERN_CRIT "CPU%u: Unknown IPI message 0x%x\n",
539                                        cpu, nextmsg);
540                                 break;
541                         }
542                 } while (msgs);
543         }
544
545         set_irq_regs(old_regs);
546 }
547
548 void smp_send_reschedule(int cpu)
549 {
550         send_ipi_message(cpumask_of(cpu), IPI_RESCHEDULE);
551 }
552
553 void smp_send_stop(void)
554 {
555         cpumask_t mask = cpu_online_map;
556         cpu_clear(smp_processor_id(), mask);
557         send_ipi_message(&mask, IPI_CPU_STOP);
558 }
559
560 /*
561  * not supported here
562  */
563 int setup_profiling_timer(unsigned int multiplier)
564 {
565         return -EINVAL;
566 }
567
568 static void
569 on_each_cpu_mask(void (*func)(void *), void *info, int wait,
570                 const struct cpumask *mask)
571 {
572         preempt_disable();
573
574         smp_call_function_many(mask, func, info, wait);
575         if (cpumask_test_cpu(smp_processor_id(), mask))
576                 func(info);
577
578         preempt_enable();
579 }
580
581 /**********************************************************************/
582
583 /*
584  * TLB operations
585  */
586 struct tlb_args {
587         struct vm_area_struct *ta_vma;
588         unsigned long ta_start;
589         unsigned long ta_end;
590 };
591
592 static inline void ipi_flush_tlb_all(void *ignored)
593 {
594         local_flush_tlb_all();
595 }
596
597 static inline void ipi_flush_tlb_mm(void *arg)
598 {
599         struct mm_struct *mm = (struct mm_struct *)arg;
600
601         local_flush_tlb_mm(mm);
602 }
603
604 static inline void ipi_flush_tlb_page(void *arg)
605 {
606         struct tlb_args *ta = (struct tlb_args *)arg;
607
608         local_flush_tlb_page(ta->ta_vma, ta->ta_start);
609 }
610
611 static inline void ipi_flush_tlb_kernel_page(void *arg)
612 {
613         struct tlb_args *ta = (struct tlb_args *)arg;
614
615         local_flush_tlb_kernel_page(ta->ta_start);
616 }
617
618 static inline void ipi_flush_tlb_range(void *arg)
619 {
620         struct tlb_args *ta = (struct tlb_args *)arg;
621
622         local_flush_tlb_range(ta->ta_vma, ta->ta_start, ta->ta_end);
623 }
624
625 static inline void ipi_flush_tlb_kernel_range(void *arg)
626 {
627         struct tlb_args *ta = (struct tlb_args *)arg;
628
629         local_flush_tlb_kernel_range(ta->ta_start, ta->ta_end);
630 }
631
632 void flush_tlb_all(void)
633 {
634         if (tlb_ops_need_broadcast())
635                 on_each_cpu(ipi_flush_tlb_all, NULL, 1);
636         else
637                 local_flush_tlb_all();
638 }
639
640 void flush_tlb_mm(struct mm_struct *mm)
641 {
642         if (tlb_ops_need_broadcast())
643                 on_each_cpu_mask(ipi_flush_tlb_mm, mm, 1, mm_cpumask(mm));
644         else
645                 local_flush_tlb_mm(mm);
646 }
647
648 void flush_tlb_page(struct vm_area_struct *vma, unsigned long uaddr)
649 {
650         if (tlb_ops_need_broadcast()) {
651                 struct tlb_args ta;
652                 ta.ta_vma = vma;
653                 ta.ta_start = uaddr;
654                 on_each_cpu_mask(ipi_flush_tlb_page, &ta, 1, mm_cpumask(vma->vm_mm));
655         } else
656                 local_flush_tlb_page(vma, uaddr);
657 }
658
659 void flush_tlb_kernel_page(unsigned long kaddr)
660 {
661         if (tlb_ops_need_broadcast()) {
662                 struct tlb_args ta;
663                 ta.ta_start = kaddr;
664                 on_each_cpu(ipi_flush_tlb_kernel_page, &ta, 1);
665         } else
666                 local_flush_tlb_kernel_page(kaddr);
667 }
668
669 void flush_tlb_range(struct vm_area_struct *vma,
670                      unsigned long start, unsigned long end)
671 {
672         if (tlb_ops_need_broadcast()) {
673                 struct tlb_args ta;
674                 ta.ta_vma = vma;
675                 ta.ta_start = start;
676                 ta.ta_end = end;
677                 on_each_cpu_mask(ipi_flush_tlb_range, &ta, 1, mm_cpumask(vma->vm_mm));
678         } else
679                 local_flush_tlb_range(vma, start, end);
680 }
681
682 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
683 {
684         if (tlb_ops_need_broadcast()) {
685                 struct tlb_args ta;
686                 ta.ta_start = start;
687                 ta.ta_end = end;
688                 on_each_cpu(ipi_flush_tlb_kernel_range, &ta, 1);
689         } else
690                 local_flush_tlb_kernel_range(start, end);
691 }