2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version 2
5 * of the License, or (at your option) any later version.
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10 * GNU General Public License for more details.
12 * You should have received a copy of the GNU General Public License
13 * along with this program; if not, write to the Free Software
14 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
16 * Copyright (C) 2000, 2001 Kanoj Sarcar
17 * Copyright (C) 2000, 2001 Ralf Baechle
18 * Copyright (C) 2000, 2001 Silicon Graphics, Inc.
19 * Copyright (C) 2000, 2001, 2003 Broadcom Corporation
21 #include <linux/cache.h>
22 #include <linux/delay.h>
23 #include <linux/init.h>
24 #include <linux/interrupt.h>
25 #include <linux/smp.h>
26 #include <linux/spinlock.h>
27 #include <linux/threads.h>
28 #include <linux/export.h>
29 #include <linux/time.h>
30 #include <linux/timex.h>
31 #include <linux/sched.h>
32 #include <linux/cpumask.h>
33 #include <linux/cpu.h>
34 #include <linux/err.h>
35 #include <linux/ftrace.h>
36 #include <linux/irqdomain.h>
38 #include <linux/of_irq.h>
40 #include <linux/atomic.h>
42 #include <asm/processor.h>
44 #include <asm/r4k-timer.h>
45 #include <asm/mips-cpc.h>
46 #include <asm/mmu_context.h>
48 #include <asm/setup.h>
51 cpumask_t cpu_callin_map; /* Bitmask of started secondaries */
53 int __cpu_number_map[NR_CPUS]; /* Map physical to logical */
54 EXPORT_SYMBOL(__cpu_number_map);
56 int __cpu_logical_map[NR_CPUS]; /* Map logical to physical */
57 EXPORT_SYMBOL(__cpu_logical_map);
59 /* Number of TCs (or siblings in Intel speak) per CPU core */
60 int smp_num_siblings = 1;
61 EXPORT_SYMBOL(smp_num_siblings);
63 /* representing the TCs (or siblings in Intel speak) of each logical CPU */
64 cpumask_t cpu_sibling_map[NR_CPUS] __read_mostly;
65 EXPORT_SYMBOL(cpu_sibling_map);
67 /* representing the core map of multi-core chips of each logical CPU */
68 cpumask_t cpu_core_map[NR_CPUS] __read_mostly;
69 EXPORT_SYMBOL(cpu_core_map);
72 * A logcal cpu mask containing only one VPE per core to
73 * reduce the number of IPIs on large MT systems.
75 cpumask_t cpu_foreign_map[NR_CPUS] __read_mostly;
76 EXPORT_SYMBOL(cpu_foreign_map);
78 /* representing cpus for which sibling maps can be computed */
79 static cpumask_t cpu_sibling_setup_map;
81 /* representing cpus for which core maps can be computed */
82 static cpumask_t cpu_core_setup_map;
84 cpumask_t cpu_coherent_mask;
86 #ifdef CONFIG_GENERIC_IRQ_IPI
87 static struct irq_desc *call_desc;
88 static struct irq_desc *sched_desc;
91 static inline void set_cpu_sibling_map(int cpu)
95 cpumask_set_cpu(cpu, &cpu_sibling_setup_map);
97 if (smp_num_siblings > 1) {
98 for_each_cpu(i, &cpu_sibling_setup_map) {
99 if (cpu_data[cpu].package == cpu_data[i].package &&
100 cpu_data[cpu].core == cpu_data[i].core) {
101 cpumask_set_cpu(i, &cpu_sibling_map[cpu]);
102 cpumask_set_cpu(cpu, &cpu_sibling_map[i]);
106 cpumask_set_cpu(cpu, &cpu_sibling_map[cpu]);
109 static inline void set_cpu_core_map(int cpu)
113 cpumask_set_cpu(cpu, &cpu_core_setup_map);
115 for_each_cpu(i, &cpu_core_setup_map) {
116 if (cpu_data[cpu].package == cpu_data[i].package) {
117 cpumask_set_cpu(i, &cpu_core_map[cpu]);
118 cpumask_set_cpu(cpu, &cpu_core_map[i]);
124 * Calculate a new cpu_foreign_map mask whenever a
125 * new cpu appears or disappears.
127 void calculate_cpu_foreign_map(void)
129 int i, k, core_present;
130 cpumask_t temp_foreign_map;
132 /* Re-calculate the mask */
133 cpumask_clear(&temp_foreign_map);
134 for_each_online_cpu(i) {
136 for_each_cpu(k, &temp_foreign_map)
137 if (cpu_data[i].package == cpu_data[k].package &&
138 cpu_data[i].core == cpu_data[k].core)
141 cpumask_set_cpu(i, &temp_foreign_map);
144 for_each_online_cpu(i)
145 cpumask_andnot(&cpu_foreign_map[i],
146 &temp_foreign_map, &cpu_sibling_map[i]);
149 struct plat_smp_ops *mp_ops;
150 EXPORT_SYMBOL(mp_ops);
152 void register_smp_ops(struct plat_smp_ops *ops)
155 printk(KERN_WARNING "Overriding previously set SMP ops\n");
160 #ifdef CONFIG_GENERIC_IRQ_IPI
161 void mips_smp_send_ipi_single(int cpu, unsigned int action)
163 mips_smp_send_ipi_mask(cpumask_of(cpu), action);
166 void mips_smp_send_ipi_mask(const struct cpumask *mask, unsigned int action)
172 local_irq_save(flags);
175 case SMP_CALL_FUNCTION:
176 __ipi_send_mask(call_desc, mask);
179 case SMP_RESCHEDULE_YOURSELF:
180 __ipi_send_mask(sched_desc, mask);
187 if (mips_cpc_present()) {
188 for_each_cpu(cpu, mask) {
189 core = cpu_data[cpu].core;
191 if (core == current_cpu_data.core)
194 while (!cpumask_test_cpu(cpu, &cpu_coherent_mask)) {
195 mips_cm_lock_other(core, 0);
196 mips_cpc_lock_other(core);
197 write_cpc_co_cmd(CPC_Cx_CMD_PWRUP);
198 mips_cpc_unlock_other();
199 mips_cm_unlock_other();
204 local_irq_restore(flags);
208 static irqreturn_t ipi_resched_interrupt(int irq, void *dev_id)
215 static irqreturn_t ipi_call_interrupt(int irq, void *dev_id)
217 generic_smp_call_function_interrupt();
222 static struct irqaction irq_resched = {
223 .handler = ipi_resched_interrupt,
224 .flags = IRQF_PERCPU,
225 .name = "IPI resched"
228 static struct irqaction irq_call = {
229 .handler = ipi_call_interrupt,
230 .flags = IRQF_PERCPU,
234 static void smp_ipi_init_one(unsigned int virq,
235 struct irqaction *action)
239 irq_set_handler(virq, handle_percpu_irq);
240 ret = setup_irq(virq, action);
244 static unsigned int call_virq, sched_virq;
246 int mips_smp_ipi_allocate(const struct cpumask *mask)
249 struct irq_domain *ipidomain;
250 struct device_node *node;
252 node = of_irq_find_parent(of_root);
253 ipidomain = irq_find_matching_host(node, DOMAIN_BUS_IPI);
256 * Some platforms have half DT setup. So if we found irq node but
257 * didn't find an ipidomain, try to search for one that is not in the
260 if (node && !ipidomain)
261 ipidomain = irq_find_matching_host(NULL, DOMAIN_BUS_IPI);
264 * There are systems which only use IPI domains some of the time,
265 * depending upon configuration we don't know until runtime. An
266 * example is Malta where we may compile in support for GIC & the
267 * MT ASE, but run on a system which has multiple VPEs in a single
268 * core and doesn't include a GIC. Until all IPI implementations
269 * have been converted to use IPI domains the best we can do here
270 * is to return & hope some other code sets up the IPIs.
275 virq = irq_reserve_ipi(ipidomain, mask);
280 virq = irq_reserve_ipi(ipidomain, mask);
285 if (irq_domain_is_ipi_per_cpu(ipidomain)) {
288 for_each_cpu(cpu, mask) {
289 smp_ipi_init_one(call_virq + cpu, &irq_call);
290 smp_ipi_init_one(sched_virq + cpu, &irq_resched);
293 smp_ipi_init_one(call_virq, &irq_call);
294 smp_ipi_init_one(sched_virq, &irq_resched);
300 int mips_smp_ipi_free(const struct cpumask *mask)
302 struct irq_domain *ipidomain;
303 struct device_node *node;
305 node = of_irq_find_parent(of_root);
306 ipidomain = irq_find_matching_host(node, DOMAIN_BUS_IPI);
309 * Some platforms have half DT setup. So if we found irq node but
310 * didn't find an ipidomain, try to search for one that is not in the
313 if (node && !ipidomain)
314 ipidomain = irq_find_matching_host(NULL, DOMAIN_BUS_IPI);
318 if (irq_domain_is_ipi_per_cpu(ipidomain)) {
321 for_each_cpu(cpu, mask) {
322 remove_irq(call_virq + cpu, &irq_call);
323 remove_irq(sched_virq + cpu, &irq_resched);
326 irq_destroy_ipi(call_virq, mask);
327 irq_destroy_ipi(sched_virq, mask);
332 static int __init mips_smp_ipi_init(void)
334 mips_smp_ipi_allocate(cpu_possible_mask);
336 call_desc = irq_to_desc(call_virq);
337 sched_desc = irq_to_desc(sched_virq);
341 early_initcall(mips_smp_ipi_init);
345 * First C code run on the secondary CPUs after being started up by
348 asmlinkage void start_secondary(void)
353 per_cpu_trap_init(false);
354 mips_clockevent_init();
355 mp_ops->init_secondary();
360 * XXX parity protection should be folded in here when it's converted
361 * to an option instead of something based on .cputype
366 cpu = smp_processor_id();
367 cpu_data[cpu].udelay_val = loops_per_jiffy;
369 cpumask_set_cpu(cpu, &cpu_coherent_mask);
370 notify_cpu_starting(cpu);
372 cpumask_set_cpu(cpu, &cpu_callin_map);
373 synchronise_count_slave(cpu);
375 set_cpu_online(cpu, true);
377 set_cpu_sibling_map(cpu);
378 set_cpu_core_map(cpu);
380 calculate_cpu_foreign_map();
383 * irq will be enabled in ->smp_finish(), enabling it too early
386 WARN_ON_ONCE(!irqs_disabled());
387 mp_ops->smp_finish();
389 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
392 static void stop_this_cpu(void *dummy)
398 set_cpu_online(smp_processor_id(), false);
399 calculate_cpu_foreign_map();
404 void smp_send_stop(void)
406 smp_call_function(stop_this_cpu, NULL, 0);
409 void __init smp_cpus_done(unsigned int max_cpus)
413 /* called from main before smp_init() */
414 void __init smp_prepare_cpus(unsigned int max_cpus)
416 init_new_context(current, &init_mm);
417 current_thread_info()->cpu = 0;
418 mp_ops->prepare_cpus(max_cpus);
419 set_cpu_sibling_map(0);
421 calculate_cpu_foreign_map();
422 #ifndef CONFIG_HOTPLUG_CPU
423 init_cpu_present(cpu_possible_mask);
425 cpumask_copy(&cpu_coherent_mask, cpu_possible_mask);
428 /* preload SMP state for boot cpu */
429 void smp_prepare_boot_cpu(void)
431 set_cpu_possible(0, true);
432 set_cpu_online(0, true);
433 cpumask_set_cpu(0, &cpu_callin_map);
436 int __cpu_up(unsigned int cpu, struct task_struct *tidle)
438 mp_ops->boot_secondary(cpu, tidle);
441 * Trust is futile. We should really have timeouts ...
443 while (!cpumask_test_cpu(cpu, &cpu_callin_map)) {
448 synchronise_count_master(cpu);
452 /* Not really SMP stuff ... */
453 int setup_profiling_timer(unsigned int multiplier)
458 static void flush_tlb_all_ipi(void *info)
460 local_flush_tlb_all();
463 void flush_tlb_all(void)
465 on_each_cpu(flush_tlb_all_ipi, NULL, 1);
468 static void flush_tlb_mm_ipi(void *mm)
470 local_flush_tlb_mm((struct mm_struct *)mm);
474 * Special Variant of smp_call_function for use by TLB functions:
477 * o collapses to normal function call on UP kernels
478 * o collapses to normal function call on systems with a single shared
481 static inline void smp_on_other_tlbs(void (*func) (void *info), void *info)
483 smp_call_function(func, info, 1);
486 static inline void smp_on_each_tlb(void (*func) (void *info), void *info)
490 smp_on_other_tlbs(func, info);
497 * The following tlb flush calls are invoked when old translations are
498 * being torn down, or pte attributes are changing. For single threaded
499 * address spaces, a new context is obtained on the current cpu, and tlb
500 * context on other cpus are invalidated to force a new context allocation
501 * at switch_mm time, should the mm ever be used on other cpus. For
502 * multithreaded address spaces, intercpu interrupts have to be sent.
503 * Another case where intercpu interrupts are required is when the target
504 * mm might be active on another cpu (eg debuggers doing the flushes on
505 * behalf of debugees, kswapd stealing pages from another process etc).
509 void flush_tlb_mm(struct mm_struct *mm)
513 if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
514 smp_on_other_tlbs(flush_tlb_mm_ipi, mm);
518 for_each_online_cpu(cpu) {
519 if (cpu != smp_processor_id() && cpu_context(cpu, mm))
520 cpu_context(cpu, mm) = 0;
523 local_flush_tlb_mm(mm);
528 struct flush_tlb_data {
529 struct vm_area_struct *vma;
534 static void flush_tlb_range_ipi(void *info)
536 struct flush_tlb_data *fd = info;
538 local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
541 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
543 struct mm_struct *mm = vma->vm_mm;
546 if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
547 struct flush_tlb_data fd = {
553 smp_on_other_tlbs(flush_tlb_range_ipi, &fd);
556 int exec = vma->vm_flags & VM_EXEC;
558 for_each_online_cpu(cpu) {
560 * flush_cache_range() will only fully flush icache if
561 * the VMA is executable, otherwise we must invalidate
562 * ASID without it appearing to has_valid_asid() as if
563 * mm has been completely unused by that CPU.
565 if (cpu != smp_processor_id() && cpu_context(cpu, mm))
566 cpu_context(cpu, mm) = !exec;
569 local_flush_tlb_range(vma, start, end);
573 static void flush_tlb_kernel_range_ipi(void *info)
575 struct flush_tlb_data *fd = info;
577 local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
580 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
582 struct flush_tlb_data fd = {
587 on_each_cpu(flush_tlb_kernel_range_ipi, &fd, 1);
590 static void flush_tlb_page_ipi(void *info)
592 struct flush_tlb_data *fd = info;
594 local_flush_tlb_page(fd->vma, fd->addr1);
597 void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
600 if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) {
601 struct flush_tlb_data fd = {
606 smp_on_other_tlbs(flush_tlb_page_ipi, &fd);
610 for_each_online_cpu(cpu) {
612 * flush_cache_page() only does partial flushes, so
613 * invalidate ASID without it appearing to
614 * has_valid_asid() as if mm has been completely unused
617 if (cpu != smp_processor_id() && cpu_context(cpu, vma->vm_mm))
618 cpu_context(cpu, vma->vm_mm) = 1;
621 local_flush_tlb_page(vma, page);
625 static void flush_tlb_one_ipi(void *info)
627 unsigned long vaddr = (unsigned long) info;
629 local_flush_tlb_one(vaddr);
632 void flush_tlb_one(unsigned long vaddr)
634 smp_on_each_tlb(flush_tlb_one_ipi, (void *) vaddr);
637 EXPORT_SYMBOL(flush_tlb_page);
638 EXPORT_SYMBOL(flush_tlb_one);
640 #if defined(CONFIG_KEXEC)
641 void (*dump_ipi_function_ptr)(void *) = NULL;
642 void dump_send_ipi(void (*dump_ipi_callback)(void *))
645 int cpu = smp_processor_id();
647 dump_ipi_function_ptr = dump_ipi_callback;
649 for_each_online_cpu(i)
651 mp_ops->send_ipi_single(i, SMP_DUMP);
654 EXPORT_SYMBOL(dump_send_ipi);
657 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
659 static DEFINE_PER_CPU(atomic_t, tick_broadcast_count);
660 static DEFINE_PER_CPU(struct call_single_data, tick_broadcast_csd);
662 void tick_broadcast(const struct cpumask *mask)
665 struct call_single_data *csd;
668 for_each_cpu(cpu, mask) {
669 count = &per_cpu(tick_broadcast_count, cpu);
670 csd = &per_cpu(tick_broadcast_csd, cpu);
672 if (atomic_inc_return(count) == 1)
673 smp_call_function_single_async(cpu, csd);
677 static void tick_broadcast_callee(void *info)
679 int cpu = smp_processor_id();
680 tick_receive_broadcast();
681 atomic_set(&per_cpu(tick_broadcast_count, cpu), 0);
684 static int __init tick_broadcast_init(void)
686 struct call_single_data *csd;
689 for (cpu = 0; cpu < NR_CPUS; cpu++) {
690 csd = &per_cpu(tick_broadcast_csd, cpu);
691 csd->func = tick_broadcast_callee;
696 early_initcall(tick_broadcast_init);
698 #endif /* CONFIG_GENERIC_CLOCKEVENTS_BROADCAST */