2 * linux/arch/arm/kernel/smp.c
4 * Copyright (C) 2002 ARM Limited, All Rights Reserved.
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.
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>
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>
28 #include <asm/atomic.h>
29 #include <asm/cacheflush.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>
41 * as from 2.5, kernels no longer have an init_tasks structure
42 * so we need some other way of telling a new secondary core
43 * where to place its SVC stack
45 struct secondary_data secondary_data;
48 * structures for inter-processor calls
49 * - A collection of single bit ipi messages.
53 unsigned long ipi_count;
57 static DEFINE_PER_CPU(struct ipi_data, ipi_data) = {
58 .lock = SPIN_LOCK_UNLOCKED,
69 int __cpuinit __cpu_up(unsigned int cpu)
71 struct cpuinfo_arm *ci = &per_cpu(cpu_data, cpu);
72 struct task_struct *idle = ci->idle;
78 * Spawn a new process manually, if not already done.
79 * Grab a pointer to its task struct so we can mess with it
82 idle = fork_idle(cpu);
84 printk(KERN_ERR "CPU%u: fork() failed\n", cpu);
91 * Allocate initial page tables to allow the new CPU to
92 * enable the MMU safely. This essentially means a set
93 * of our "standard" page tables, with the addition of
94 * a 1:1 mapping for the physical address of the kernel.
96 pgd = pgd_alloc(&init_mm);
97 pmd = pmd_offset(pgd + pgd_index(PHYS_OFFSET), PHYS_OFFSET);
98 *pmd = __pmd((PHYS_OFFSET & PGDIR_MASK) |
99 PMD_TYPE_SECT | PMD_SECT_AP_WRITE);
100 flush_pmd_entry(pmd);
103 * We need to tell the secondary core where to find
104 * its stack and the page tables.
106 secondary_data.stack = task_stack_page(idle) + THREAD_START_SP;
107 secondary_data.pgdir = virt_to_phys(pgd);
111 * Now bring the CPU into our world.
113 ret = boot_secondary(cpu, idle);
115 unsigned long timeout;
118 * CPU was successfully started, wait for it
119 * to come online or time out.
121 timeout = jiffies + HZ;
122 while (time_before(jiffies, timeout)) {
130 if (!cpu_online(cpu))
134 secondary_data.stack = NULL;
135 secondary_data.pgdir = 0;
138 clean_pmd_entry(pmd);
139 pgd_free(&init_mm, pgd);
142 printk(KERN_CRIT "CPU%u: processor failed to boot\n", cpu);
145 * FIXME: We need to clean up the new idle thread. --rmk
152 #ifdef CONFIG_HOTPLUG_CPU
154 * __cpu_disable runs on the processor to be shutdown.
156 int __cpuexit __cpu_disable(void)
158 unsigned int cpu = smp_processor_id();
159 struct task_struct *p;
162 ret = mach_cpu_disable(cpu);
167 * Take this CPU offline. Once we clear this, we can't return,
168 * and we must not schedule until we're ready to give up the cpu.
170 set_cpu_online(cpu, false);
173 * OK - migrate IRQs away from this CPU
178 * Stop the local timer for this CPU.
183 * Flush user cache and TLB mappings, and then remove this CPU
184 * from the vm mask set of all processes.
187 local_flush_tlb_all();
189 read_lock(&tasklist_lock);
190 for_each_process(p) {
192 cpumask_clear_cpu(cpu, mm_cpumask(p->mm));
194 read_unlock(&tasklist_lock);
200 * called on the thread which is asking for a CPU to be shutdown -
201 * waits until shutdown has completed, or it is timed out.
203 void __cpuexit __cpu_die(unsigned int cpu)
205 if (!platform_cpu_kill(cpu))
206 printk("CPU%u: unable to kill\n", cpu);
210 * Called from the idle thread for the CPU which has been shutdown.
212 * Note that we disable IRQs here, but do not re-enable them
213 * before returning to the caller. This is also the behaviour
214 * of the other hotplug-cpu capable cores, so presumably coming
215 * out of idle fixes this.
217 void __cpuexit cpu_die(void)
219 unsigned int cpu = smp_processor_id();
225 * actual CPU shutdown procedure is at least platform (if not
228 platform_cpu_die(cpu);
231 * Do not return to the idle loop - jump back to the secondary
232 * cpu initialisation. There's some initialisation which needs
233 * to be repeated to undo the effects of taking the CPU offline.
235 __asm__("mov sp, %0\n"
236 " b secondary_start_kernel"
238 : "r" (task_stack_page(current) + THREAD_SIZE - 8));
240 #endif /* CONFIG_HOTPLUG_CPU */
243 * This is the secondary CPU boot entry. We're using this CPUs
244 * idle thread stack, but a set of temporary page tables.
246 asmlinkage void __cpuinit secondary_start_kernel(void)
248 struct mm_struct *mm = &init_mm;
249 unsigned int cpu = smp_processor_id();
251 printk("CPU%u: Booted secondary processor\n", cpu);
254 * All kernel threads share the same mm context; grab a
255 * reference and switch to it.
257 atomic_inc(&mm->mm_users);
258 atomic_inc(&mm->mm_count);
259 current->active_mm = mm;
260 cpumask_set_cpu(cpu, mm_cpumask(mm));
261 cpu_switch_mm(mm->pgd, mm);
262 enter_lazy_tlb(mm, current);
263 local_flush_tlb_all();
269 * Give the platform a chance to do its own initialisation.
271 platform_secondary_init(cpu);
274 * Enable local interrupts.
276 notify_cpu_starting(cpu);
281 * Setup the percpu timer for this CPU.
283 percpu_timer_setup();
287 smp_store_cpu_info(cpu);
290 * OK, now it's safe to let the boot CPU continue
292 set_cpu_online(cpu, true);
295 * OK, it's off to the idle thread for us
301 * Called by both boot and secondaries to move global data into
302 * per-processor storage.
304 void __cpuinit smp_store_cpu_info(unsigned int cpuid)
306 struct cpuinfo_arm *cpu_info = &per_cpu(cpu_data, cpuid);
308 cpu_info->loops_per_jiffy = loops_per_jiffy;
311 void __init smp_cpus_done(unsigned int max_cpus)
314 unsigned long bogosum = 0;
316 for_each_online_cpu(cpu)
317 bogosum += per_cpu(cpu_data, cpu).loops_per_jiffy;
319 printk(KERN_INFO "SMP: Total of %d processors activated "
320 "(%lu.%02lu BogoMIPS).\n",
322 bogosum / (500000/HZ),
323 (bogosum / (5000/HZ)) % 100);
326 void __init smp_prepare_boot_cpu(void)
328 unsigned int cpu = smp_processor_id();
330 per_cpu(cpu_data, cpu).idle = current;
333 static void send_ipi_message(const struct cpumask *mask, enum ipi_msg_type msg)
338 local_irq_save(flags);
340 for_each_cpu(cpu, mask) {
341 struct ipi_data *ipi = &per_cpu(ipi_data, cpu);
343 spin_lock(&ipi->lock);
344 ipi->bits |= 1 << msg;
345 spin_unlock(&ipi->lock);
349 * Call the platform specific cross-CPU call function.
351 smp_cross_call(mask);
353 local_irq_restore(flags);
356 void arch_send_call_function_ipi_mask(const struct cpumask *mask)
358 send_ipi_message(mask, IPI_CALL_FUNC);
361 void arch_send_call_function_single_ipi(int cpu)
363 send_ipi_message(cpumask_of(cpu), IPI_CALL_FUNC_SINGLE);
366 void show_ipi_list(struct seq_file *p)
372 for_each_present_cpu(cpu)
373 seq_printf(p, " %10lu", per_cpu(ipi_data, cpu).ipi_count);
378 void show_local_irqs(struct seq_file *p)
382 seq_printf(p, "LOC: ");
384 for_each_present_cpu(cpu)
385 seq_printf(p, "%10u ", irq_stat[cpu].local_timer_irqs);
391 * Timer (local or broadcast) support
393 static DEFINE_PER_CPU(struct clock_event_device, percpu_clockevent);
395 static void ipi_timer(void)
397 struct clock_event_device *evt = &__get_cpu_var(percpu_clockevent);
399 evt->event_handler(evt);
403 #ifdef CONFIG_LOCAL_TIMERS
404 asmlinkage void __exception do_local_timer(struct pt_regs *regs)
406 struct pt_regs *old_regs = set_irq_regs(regs);
407 int cpu = smp_processor_id();
409 if (local_timer_ack()) {
410 irq_stat[cpu].local_timer_irqs++;
414 set_irq_regs(old_regs);
418 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
419 static void smp_timer_broadcast(const struct cpumask *mask)
421 send_ipi_message(mask, IPI_TIMER);
424 static void broadcast_timer_set_mode(enum clock_event_mode mode,
425 struct clock_event_device *evt)
429 static void local_timer_setup(struct clock_event_device *evt)
431 evt->name = "dummy_timer";
432 evt->features = CLOCK_EVT_FEAT_ONESHOT |
433 CLOCK_EVT_FEAT_PERIODIC |
434 CLOCK_EVT_FEAT_DUMMY;
437 evt->set_mode = broadcast_timer_set_mode;
438 evt->broadcast = smp_timer_broadcast;
440 clockevents_register_device(evt);
444 void __cpuinit percpu_timer_setup(void)
446 unsigned int cpu = smp_processor_id();
447 struct clock_event_device *evt = &per_cpu(percpu_clockevent, cpu);
449 evt->cpumask = cpumask_of(cpu);
451 local_timer_setup(evt);
454 static DEFINE_SPINLOCK(stop_lock);
457 * ipi_cpu_stop - handle IPI from smp_send_stop()
459 static void ipi_cpu_stop(unsigned int cpu)
461 spin_lock(&stop_lock);
462 printk(KERN_CRIT "CPU%u: stopping\n", cpu);
464 spin_unlock(&stop_lock);
466 set_cpu_online(cpu, false);
476 * Main handler for inter-processor interrupts
478 * For ARM, the ipimask now only identifies a single
479 * category of IPI (Bit 1 IPIs have been replaced by a
480 * different mechanism):
482 * Bit 0 - Inter-processor function call
484 asmlinkage void __exception do_IPI(struct pt_regs *regs)
486 unsigned int cpu = smp_processor_id();
487 struct ipi_data *ipi = &per_cpu(ipi_data, cpu);
488 struct pt_regs *old_regs = set_irq_regs(regs);
495 spin_lock(&ipi->lock);
498 spin_unlock(&ipi->lock);
506 nextmsg = msgs & -msgs;
508 nextmsg = ffz(~nextmsg);
517 * nothing more to do - eveything is
518 * done on the interrupt return path
523 generic_smp_call_function_interrupt();
526 case IPI_CALL_FUNC_SINGLE:
527 generic_smp_call_function_single_interrupt();
535 printk(KERN_CRIT "CPU%u: Unknown IPI message 0x%x\n",
542 set_irq_regs(old_regs);
545 void smp_send_reschedule(int cpu)
547 send_ipi_message(cpumask_of(cpu), IPI_RESCHEDULE);
550 void smp_send_stop(void)
552 cpumask_t mask = cpu_online_map;
553 cpu_clear(smp_processor_id(), mask);
554 send_ipi_message(&mask, IPI_CPU_STOP);
560 int setup_profiling_timer(unsigned int multiplier)
566 on_each_cpu_mask(void (*func)(void *), void *info, int wait,
567 const struct cpumask *mask)
571 smp_call_function_many(mask, func, info, wait);
572 if (cpumask_test_cpu(smp_processor_id(), mask))
578 /**********************************************************************/
584 struct vm_area_struct *ta_vma;
585 unsigned long ta_start;
586 unsigned long ta_end;
589 /* all SMP configurations have the extended CPUID registers */
590 static inline int tlb_ops_need_broadcast(void)
592 return ((read_cpuid_ext(CPUID_EXT_MMFR3) >> 12) & 0xf) < 2;
595 static inline void ipi_flush_tlb_all(void *ignored)
597 local_flush_tlb_all();
600 static inline void ipi_flush_tlb_mm(void *arg)
602 struct mm_struct *mm = (struct mm_struct *)arg;
604 local_flush_tlb_mm(mm);
607 static inline void ipi_flush_tlb_page(void *arg)
609 struct tlb_args *ta = (struct tlb_args *)arg;
611 local_flush_tlb_page(ta->ta_vma, ta->ta_start);
614 static inline void ipi_flush_tlb_kernel_page(void *arg)
616 struct tlb_args *ta = (struct tlb_args *)arg;
618 local_flush_tlb_kernel_page(ta->ta_start);
621 static inline void ipi_flush_tlb_range(void *arg)
623 struct tlb_args *ta = (struct tlb_args *)arg;
625 local_flush_tlb_range(ta->ta_vma, ta->ta_start, ta->ta_end);
628 static inline void ipi_flush_tlb_kernel_range(void *arg)
630 struct tlb_args *ta = (struct tlb_args *)arg;
632 local_flush_tlb_kernel_range(ta->ta_start, ta->ta_end);
635 void flush_tlb_all(void)
637 if (tlb_ops_need_broadcast())
638 on_each_cpu(ipi_flush_tlb_all, NULL, 1);
640 local_flush_tlb_all();
643 void flush_tlb_mm(struct mm_struct *mm)
645 if (tlb_ops_need_broadcast())
646 on_each_cpu_mask(ipi_flush_tlb_mm, mm, 1, mm_cpumask(mm));
648 local_flush_tlb_mm(mm);
651 void flush_tlb_page(struct vm_area_struct *vma, unsigned long uaddr)
653 if (tlb_ops_need_broadcast()) {
657 on_each_cpu_mask(ipi_flush_tlb_page, &ta, 1, mm_cpumask(vma->vm_mm));
659 local_flush_tlb_page(vma, uaddr);
662 void flush_tlb_kernel_page(unsigned long kaddr)
664 if (tlb_ops_need_broadcast()) {
667 on_each_cpu(ipi_flush_tlb_kernel_page, &ta, 1);
669 local_flush_tlb_kernel_page(kaddr);
672 void flush_tlb_range(struct vm_area_struct *vma,
673 unsigned long start, unsigned long end)
675 if (tlb_ops_need_broadcast()) {
680 on_each_cpu_mask(ipi_flush_tlb_range, &ta, 1, mm_cpumask(vma->vm_mm));
682 local_flush_tlb_range(vma, start, end);
685 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
687 if (tlb_ops_need_broadcast()) {
691 on_each_cpu(ipi_flush_tlb_kernel_range, &ta, 1);
693 local_flush_tlb_kernel_range(start, end);