1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18#undef DEBUG
19
20#include <linux/kernel.h>
21#include <linux/export.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/device.h>
31#include <linux/cpu.h>
32#include <linux/notifier.h>
33#include <linux/topology.h>
34
35#include <asm/ptrace.h>
36#include <linux/atomic.h>
37#include <asm/irq.h>
38#include <asm/hw_irq.h>
39#include <asm/kvm_ppc.h>
40#include <asm/page.h>
41#include <asm/pgtable.h>
42#include <asm/prom.h>
43#include <asm/smp.h>
44#include <asm/time.h>
45#include <asm/machdep.h>
46#include <asm/cputhreads.h>
47#include <asm/cputable.h>
48#include <asm/mpic.h>
49#include <asm/vdso_datapage.h>
50#ifdef CONFIG_PPC64
51#include <asm/paca.h>
52#endif
53#include <asm/vdso.h>
54#include <asm/debug.h>
55#include <asm/kexec.h>
56
57#ifdef DEBUG
58#include <asm/udbg.h>
59#define DBG(fmt...) udbg_printf(fmt)
60#else
61#define DBG(fmt...)
62#endif
63
64#ifdef CONFIG_HOTPLUG_CPU
65
66static DEFINE_PER_CPU(int, cpu_state) = { 0 };
67#endif
68
69struct thread_info *secondary_ti;
70
71DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
72DEFINE_PER_CPU(cpumask_var_t, cpu_core_map);
73
74EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
75EXPORT_PER_CPU_SYMBOL(cpu_core_map);
76
77
78struct smp_ops_t *smp_ops;
79
80
81volatile unsigned int cpu_callin_map[NR_CPUS];
82
83int smt_enabled_at_boot = 1;
84
85static void (*crash_ipi_function_ptr)(struct pt_regs *) = NULL;
86
87
88
89
90
91
92int smp_generic_cpu_bootable(unsigned int nr)
93{
94
95
96
97 if (system_state == SYSTEM_BOOTING && cpu_has_feature(CPU_FTR_SMT)) {
98 if (!smt_enabled_at_boot && cpu_thread_in_core(nr) != 0)
99 return 0;
100 if (smt_enabled_at_boot
101 && cpu_thread_in_core(nr) >= smt_enabled_at_boot)
102 return 0;
103 }
104
105 return 1;
106}
107
108
109#ifdef CONFIG_PPC64
110int smp_generic_kick_cpu(int nr)
111{
112 BUG_ON(nr < 0 || nr >= NR_CPUS);
113
114
115
116
117
118
119 if (!paca[nr].cpu_start) {
120 paca[nr].cpu_start = 1;
121 smp_mb();
122 return 0;
123 }
124
125#ifdef CONFIG_HOTPLUG_CPU
126
127
128
129
130 generic_set_cpu_up(nr);
131 smp_wmb();
132 smp_send_reschedule(nr);
133#endif
134
135 return 0;
136}
137#endif
138
139static irqreturn_t call_function_action(int irq, void *data)
140{
141 generic_smp_call_function_interrupt();
142 return IRQ_HANDLED;
143}
144
145static irqreturn_t reschedule_action(int irq, void *data)
146{
147 scheduler_ipi();
148 return IRQ_HANDLED;
149}
150
151static irqreturn_t tick_broadcast_ipi_action(int irq, void *data)
152{
153 tick_broadcast_ipi_handler();
154 return IRQ_HANDLED;
155}
156
157static irqreturn_t debug_ipi_action(int irq, void *data)
158{
159 if (crash_ipi_function_ptr) {
160 crash_ipi_function_ptr(get_irq_regs());
161 return IRQ_HANDLED;
162 }
163
164#ifdef CONFIG_DEBUGGER
165 debugger_ipi(get_irq_regs());
166#endif
167
168 return IRQ_HANDLED;
169}
170
171static irq_handler_t smp_ipi_action[] = {
172 [PPC_MSG_CALL_FUNCTION] = call_function_action,
173 [PPC_MSG_RESCHEDULE] = reschedule_action,
174 [PPC_MSG_TICK_BROADCAST] = tick_broadcast_ipi_action,
175 [PPC_MSG_DEBUGGER_BREAK] = debug_ipi_action,
176};
177
178const char *smp_ipi_name[] = {
179 [PPC_MSG_CALL_FUNCTION] = "ipi call function",
180 [PPC_MSG_RESCHEDULE] = "ipi reschedule",
181 [PPC_MSG_TICK_BROADCAST] = "ipi tick-broadcast",
182 [PPC_MSG_DEBUGGER_BREAK] = "ipi debugger",
183};
184
185
186int smp_request_message_ipi(int virq, int msg)
187{
188 int err;
189
190 if (msg < 0 || msg > PPC_MSG_DEBUGGER_BREAK) {
191 return -EINVAL;
192 }
193#if !defined(CONFIG_DEBUGGER) && !defined(CONFIG_KEXEC)
194 if (msg == PPC_MSG_DEBUGGER_BREAK) {
195 return 1;
196 }
197#endif
198 err = request_irq(virq, smp_ipi_action[msg],
199 IRQF_PERCPU | IRQF_NO_THREAD | IRQF_NO_SUSPEND,
200 smp_ipi_name[msg], NULL);
201 WARN(err < 0, "unable to request_irq %d for %s (rc %d)\n",
202 virq, smp_ipi_name[msg], err);
203
204 return err;
205}
206
207#ifdef CONFIG_PPC_SMP_MUXED_IPI
208struct cpu_messages {
209 int messages;
210 unsigned long data;
211};
212static DEFINE_PER_CPU_SHARED_ALIGNED(struct cpu_messages, ipi_message);
213
214void smp_muxed_ipi_set_data(int cpu, unsigned long data)
215{
216 struct cpu_messages *info = &per_cpu(ipi_message, cpu);
217
218 info->data = data;
219}
220
221void smp_muxed_ipi_message_pass(int cpu, int msg)
222{
223 struct cpu_messages *info = &per_cpu(ipi_message, cpu);
224 char *message = (char *)&info->messages;
225
226
227
228
229 smp_mb();
230 message[msg] = 1;
231
232
233
234
235 smp_ops->cause_ipi(cpu, info->data);
236}
237
238#ifdef __BIG_ENDIAN__
239#define IPI_MESSAGE(A) (1 << (24 - 8 * (A)))
240#else
241#define IPI_MESSAGE(A) (1 << (8 * (A)))
242#endif
243
244irqreturn_t smp_ipi_demux(void)
245{
246 struct cpu_messages *info = this_cpu_ptr(&ipi_message);
247 unsigned int all;
248
249 mb();
250
251 do {
252 all = xchg(&info->messages, 0);
253 if (all & IPI_MESSAGE(PPC_MSG_CALL_FUNCTION))
254 generic_smp_call_function_interrupt();
255 if (all & IPI_MESSAGE(PPC_MSG_RESCHEDULE))
256 scheduler_ipi();
257 if (all & IPI_MESSAGE(PPC_MSG_TICK_BROADCAST))
258 tick_broadcast_ipi_handler();
259 if (all & IPI_MESSAGE(PPC_MSG_DEBUGGER_BREAK))
260 debug_ipi_action(0, NULL);
261 } while (info->messages);
262
263 return IRQ_HANDLED;
264}
265#endif
266
267static inline void do_message_pass(int cpu, int msg)
268{
269 if (smp_ops->message_pass)
270 smp_ops->message_pass(cpu, msg);
271#ifdef CONFIG_PPC_SMP_MUXED_IPI
272 else
273 smp_muxed_ipi_message_pass(cpu, msg);
274#endif
275}
276
277void smp_send_reschedule(int cpu)
278{
279 if (likely(smp_ops))
280 do_message_pass(cpu, PPC_MSG_RESCHEDULE);
281}
282EXPORT_SYMBOL_GPL(smp_send_reschedule);
283
284void arch_send_call_function_single_ipi(int cpu)
285{
286 do_message_pass(cpu, PPC_MSG_CALL_FUNCTION);
287}
288
289void arch_send_call_function_ipi_mask(const struct cpumask *mask)
290{
291 unsigned int cpu;
292
293 for_each_cpu(cpu, mask)
294 do_message_pass(cpu, PPC_MSG_CALL_FUNCTION);
295}
296
297#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
298void tick_broadcast(const struct cpumask *mask)
299{
300 unsigned int cpu;
301
302 for_each_cpu(cpu, mask)
303 do_message_pass(cpu, PPC_MSG_TICK_BROADCAST);
304}
305#endif
306
307#if defined(CONFIG_DEBUGGER) || defined(CONFIG_KEXEC)
308void smp_send_debugger_break(void)
309{
310 int cpu;
311 int me = raw_smp_processor_id();
312
313 if (unlikely(!smp_ops))
314 return;
315
316 for_each_online_cpu(cpu)
317 if (cpu != me)
318 do_message_pass(cpu, PPC_MSG_DEBUGGER_BREAK);
319}
320#endif
321
322#ifdef CONFIG_KEXEC
323void crash_send_ipi(void (*crash_ipi_callback)(struct pt_regs *))
324{
325 crash_ipi_function_ptr = crash_ipi_callback;
326 if (crash_ipi_callback) {
327 mb();
328 smp_send_debugger_break();
329 }
330}
331#endif
332
333static void stop_this_cpu(void *dummy)
334{
335
336 set_cpu_online(smp_processor_id(), false);
337
338 local_irq_disable();
339 while (1)
340 ;
341}
342
343void smp_send_stop(void)
344{
345 smp_call_function(stop_this_cpu, NULL, 0);
346}
347
348struct thread_info *current_set[NR_CPUS];
349
350static void smp_store_cpu_info(int id)
351{
352 per_cpu(cpu_pvr, id) = mfspr(SPRN_PVR);
353#ifdef CONFIG_PPC_FSL_BOOK3E
354 per_cpu(next_tlbcam_idx, id)
355 = (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
356#endif
357}
358
359void __init smp_prepare_cpus(unsigned int max_cpus)
360{
361 unsigned int cpu;
362
363 DBG("smp_prepare_cpus\n");
364
365
366
367
368
369 BUG_ON(boot_cpuid != smp_processor_id());
370
371
372 smp_store_cpu_info(boot_cpuid);
373 cpu_callin_map[boot_cpuid] = 1;
374
375 for_each_possible_cpu(cpu) {
376 zalloc_cpumask_var_node(&per_cpu(cpu_sibling_map, cpu),
377 GFP_KERNEL, cpu_to_node(cpu));
378 zalloc_cpumask_var_node(&per_cpu(cpu_core_map, cpu),
379 GFP_KERNEL, cpu_to_node(cpu));
380
381
382
383 if (cpu_present(cpu)) {
384 set_cpu_numa_node(cpu, numa_cpu_lookup_table[cpu]);
385 set_cpu_numa_mem(cpu,
386 local_memory_node(numa_cpu_lookup_table[cpu]));
387 }
388 }
389
390 cpumask_set_cpu(boot_cpuid, cpu_sibling_mask(boot_cpuid));
391 cpumask_set_cpu(boot_cpuid, cpu_core_mask(boot_cpuid));
392
393 if (smp_ops && smp_ops->probe)
394 smp_ops->probe();
395}
396
397void smp_prepare_boot_cpu(void)
398{
399 BUG_ON(smp_processor_id() != boot_cpuid);
400#ifdef CONFIG_PPC64
401 paca[boot_cpuid].__current = current;
402#endif
403 set_numa_node(numa_cpu_lookup_table[boot_cpuid]);
404 current_set[boot_cpuid] = task_thread_info(current);
405}
406
407#ifdef CONFIG_HOTPLUG_CPU
408
409int generic_cpu_disable(void)
410{
411 unsigned int cpu = smp_processor_id();
412
413 if (cpu == boot_cpuid)
414 return -EBUSY;
415
416 set_cpu_online(cpu, false);
417#ifdef CONFIG_PPC64
418 vdso_data->processorCount--;
419#endif
420 migrate_irqs();
421 return 0;
422}
423
424void generic_cpu_die(unsigned int cpu)
425{
426 int i;
427
428 for (i = 0; i < 100; i++) {
429 smp_rmb();
430 if (per_cpu(cpu_state, cpu) == CPU_DEAD)
431 return;
432 msleep(100);
433 }
434 printk(KERN_ERR "CPU%d didn't die...\n", cpu);
435}
436
437void generic_set_cpu_dead(unsigned int cpu)
438{
439 per_cpu(cpu_state, cpu) = CPU_DEAD;
440}
441
442
443
444
445
446
447void generic_set_cpu_up(unsigned int cpu)
448{
449 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
450}
451
452int generic_check_cpu_restart(unsigned int cpu)
453{
454 return per_cpu(cpu_state, cpu) == CPU_UP_PREPARE;
455}
456
457static bool secondaries_inhibited(void)
458{
459 return kvm_hv_mode_active();
460}
461
462#else
463
464#define secondaries_inhibited() 0
465
466#endif
467
468static void cpu_idle_thread_init(unsigned int cpu, struct task_struct *idle)
469{
470 struct thread_info *ti = task_thread_info(idle);
471
472#ifdef CONFIG_PPC64
473 paca[cpu].__current = idle;
474 paca[cpu].kstack = (unsigned long)ti + THREAD_SIZE - STACK_FRAME_OVERHEAD;
475#endif
476 ti->cpu = cpu;
477 secondary_ti = current_set[cpu] = ti;
478}
479
480int __cpu_up(unsigned int cpu, struct task_struct *tidle)
481{
482 int rc, c;
483
484
485
486
487 if (threads_per_core > 1 && secondaries_inhibited() &&
488 cpu_thread_in_subcore(cpu))
489 return -EBUSY;
490
491 if (smp_ops == NULL ||
492 (smp_ops->cpu_bootable && !smp_ops->cpu_bootable(cpu)))
493 return -EINVAL;
494
495 cpu_idle_thread_init(cpu, tidle);
496
497
498
499
500 cpu_callin_map[cpu] = 0;
501
502
503
504
505
506 smp_mb();
507
508
509 DBG("smp: kicking cpu %d\n", cpu);
510 rc = smp_ops->kick_cpu(cpu);
511 if (rc) {
512 pr_err("smp: failed starting cpu %d (rc %d)\n", cpu, rc);
513 return rc;
514 }
515
516
517
518
519
520
521 if (system_state < SYSTEM_RUNNING)
522 for (c = 50000; c && !cpu_callin_map[cpu]; c--)
523 udelay(100);
524#ifdef CONFIG_HOTPLUG_CPU
525 else
526
527
528
529
530 for (c = 5000; c && !cpu_callin_map[cpu]; c--)
531 msleep(1);
532#endif
533
534 if (!cpu_callin_map[cpu]) {
535 printk(KERN_ERR "Processor %u is stuck.\n", cpu);
536 return -ENOENT;
537 }
538
539 DBG("Processor %u found.\n", cpu);
540
541 if (smp_ops->give_timebase)
542 smp_ops->give_timebase();
543
544
545 while (!cpu_online(cpu) || !cpu_active(cpu))
546 cpu_relax();
547
548 return 0;
549}
550
551
552
553
554int cpu_to_core_id(int cpu)
555{
556 struct device_node *np;
557 const __be32 *reg;
558 int id = -1;
559
560 np = of_get_cpu_node(cpu, NULL);
561 if (!np)
562 goto out;
563
564 reg = of_get_property(np, "reg", NULL);
565 if (!reg)
566 goto out;
567
568 id = be32_to_cpup(reg);
569out:
570 of_node_put(np);
571 return id;
572}
573
574
575int cpu_core_index_of_thread(int cpu)
576{
577 return cpu >> threads_shift;
578}
579EXPORT_SYMBOL_GPL(cpu_core_index_of_thread);
580
581int cpu_first_thread_of_core(int core)
582{
583 return core << threads_shift;
584}
585EXPORT_SYMBOL_GPL(cpu_first_thread_of_core);
586
587static void traverse_siblings_chip_id(int cpu, bool add, int chipid)
588{
589 const struct cpumask *mask;
590 struct device_node *np;
591 int i, plen;
592 const __be32 *prop;
593
594 mask = add ? cpu_online_mask : cpu_present_mask;
595 for_each_cpu(i, mask) {
596 np = of_get_cpu_node(i, NULL);
597 if (!np)
598 continue;
599 prop = of_get_property(np, "ibm,chip-id", &plen);
600 if (prop && plen == sizeof(int) &&
601 of_read_number(prop, 1) == chipid) {
602 if (add) {
603 cpumask_set_cpu(cpu, cpu_core_mask(i));
604 cpumask_set_cpu(i, cpu_core_mask(cpu));
605 } else {
606 cpumask_clear_cpu(cpu, cpu_core_mask(i));
607 cpumask_clear_cpu(i, cpu_core_mask(cpu));
608 }
609 }
610 of_node_put(np);
611 }
612}
613
614
615
616
617static struct device_node *cpu_to_l2cache(int cpu)
618{
619 struct device_node *np;
620 struct device_node *cache;
621
622 if (!cpu_present(cpu))
623 return NULL;
624
625 np = of_get_cpu_node(cpu, NULL);
626 if (np == NULL)
627 return NULL;
628
629 cache = of_find_next_cache_node(np);
630
631 of_node_put(np);
632
633 return cache;
634}
635
636static void traverse_core_siblings(int cpu, bool add)
637{
638 struct device_node *l2_cache, *np;
639 const struct cpumask *mask;
640 int i, chip, plen;
641 const __be32 *prop;
642
643
644 np = of_get_cpu_node(cpu, NULL);
645 if (np) {
646 chip = -1;
647 prop = of_get_property(np, "ibm,chip-id", &plen);
648 if (prop && plen == sizeof(int))
649 chip = of_read_number(prop, 1);
650 of_node_put(np);
651 if (chip >= 0) {
652 traverse_siblings_chip_id(cpu, add, chip);
653 return;
654 }
655 }
656
657 l2_cache = cpu_to_l2cache(cpu);
658 mask = add ? cpu_online_mask : cpu_present_mask;
659 for_each_cpu(i, mask) {
660 np = cpu_to_l2cache(i);
661 if (!np)
662 continue;
663 if (np == l2_cache) {
664 if (add) {
665 cpumask_set_cpu(cpu, cpu_core_mask(i));
666 cpumask_set_cpu(i, cpu_core_mask(cpu));
667 } else {
668 cpumask_clear_cpu(cpu, cpu_core_mask(i));
669 cpumask_clear_cpu(i, cpu_core_mask(cpu));
670 }
671 }
672 of_node_put(np);
673 }
674 of_node_put(l2_cache);
675}
676
677
678void start_secondary(void *unused)
679{
680 unsigned int cpu = smp_processor_id();
681 int i, base;
682
683 atomic_inc(&init_mm.mm_count);
684 current->active_mm = &init_mm;
685
686 smp_store_cpu_info(cpu);
687 set_dec(tb_ticks_per_jiffy);
688 preempt_disable();
689 cpu_callin_map[cpu] = 1;
690
691 if (smp_ops->setup_cpu)
692 smp_ops->setup_cpu(cpu);
693 if (smp_ops->take_timebase)
694 smp_ops->take_timebase();
695
696 secondary_cpu_time_init();
697
698#ifdef CONFIG_PPC64
699 if (system_state == SYSTEM_RUNNING)
700 vdso_data->processorCount++;
701
702 vdso_getcpu_init();
703#endif
704
705 base = cpu_first_thread_sibling(cpu);
706 for (i = 0; i < threads_per_core; i++) {
707 if (cpu_is_offline(base + i) && (cpu != base + i))
708 continue;
709 cpumask_set_cpu(cpu, cpu_sibling_mask(base + i));
710 cpumask_set_cpu(base + i, cpu_sibling_mask(cpu));
711
712
713
714
715
716 cpumask_set_cpu(cpu, cpu_core_mask(base + i));
717 cpumask_set_cpu(base + i, cpu_core_mask(cpu));
718 }
719 traverse_core_siblings(cpu, true);
720
721 set_numa_node(numa_cpu_lookup_table[cpu]);
722 set_numa_mem(local_memory_node(numa_cpu_lookup_table[cpu]));
723
724 smp_wmb();
725 notify_cpu_starting(cpu);
726 set_cpu_online(cpu, true);
727
728 local_irq_enable();
729
730 cpu_startup_entry(CPUHP_ONLINE);
731
732 BUG();
733}
734
735int setup_profiling_timer(unsigned int multiplier)
736{
737 return 0;
738}
739
740#ifdef CONFIG_SCHED_SMT
741
742static int powerpc_smt_flags(void)
743{
744 int flags = SD_SHARE_CPUCAPACITY | SD_SHARE_PKG_RESOURCES;
745
746 if (cpu_has_feature(CPU_FTR_ASYM_SMT)) {
747 printk_once(KERN_INFO "Enabling Asymmetric SMT scheduling\n");
748 flags |= SD_ASYM_PACKING;
749 }
750 return flags;
751}
752#endif
753
754static struct sched_domain_topology_level powerpc_topology[] = {
755#ifdef CONFIG_SCHED_SMT
756 { cpu_smt_mask, powerpc_smt_flags, SD_INIT_NAME(SMT) },
757#endif
758 { cpu_cpu_mask, SD_INIT_NAME(DIE) },
759 { NULL, },
760};
761
762void __init smp_cpus_done(unsigned int max_cpus)
763{
764 cpumask_var_t old_mask;
765
766
767
768
769
770 alloc_cpumask_var(&old_mask, GFP_NOWAIT);
771 cpumask_copy(old_mask, tsk_cpus_allowed(current));
772 set_cpus_allowed_ptr(current, cpumask_of(boot_cpuid));
773
774 if (smp_ops && smp_ops->setup_cpu)
775 smp_ops->setup_cpu(boot_cpuid);
776
777 set_cpus_allowed_ptr(current, old_mask);
778
779 free_cpumask_var(old_mask);
780
781 if (smp_ops && smp_ops->bringup_done)
782 smp_ops->bringup_done();
783
784 dump_numa_cpu_topology();
785
786 set_sched_topology(powerpc_topology);
787
788}
789
790#ifdef CONFIG_HOTPLUG_CPU
791int __cpu_disable(void)
792{
793 int cpu = smp_processor_id();
794 int base, i;
795 int err;
796
797 if (!smp_ops->cpu_disable)
798 return -ENOSYS;
799
800 err = smp_ops->cpu_disable();
801 if (err)
802 return err;
803
804
805 base = cpu_first_thread_sibling(cpu);
806 for (i = 0; i < threads_per_core; i++) {
807 cpumask_clear_cpu(cpu, cpu_sibling_mask(base + i));
808 cpumask_clear_cpu(base + i, cpu_sibling_mask(cpu));
809 cpumask_clear_cpu(cpu, cpu_core_mask(base + i));
810 cpumask_clear_cpu(base + i, cpu_core_mask(cpu));
811 }
812 traverse_core_siblings(cpu, false);
813
814 return 0;
815}
816
817void __cpu_die(unsigned int cpu)
818{
819 if (smp_ops->cpu_die)
820 smp_ops->cpu_die(cpu);
821}
822
823void cpu_die(void)
824{
825 if (ppc_md.cpu_die)
826 ppc_md.cpu_die();
827
828
829 start_secondary_resume();
830}
831
832#endif
833