1
2
3
4
5
6
7
8
9
10
11
12
13
14
15#include <linux/sched.h>
16#include <linux/err.h>
17#include <linux/slab.h>
18#include <linux/smp.h>
19#include <linux/irq_work.h>
20#include <linux/tick.h>
21
22#include <asm/paravirt.h>
23#include <asm/desc.h>
24#include <asm/pgtable.h>
25#include <asm/cpu.h>
26
27#include <xen/interface/xen.h>
28#include <xen/interface/vcpu.h>
29
30#include <asm/xen/interface.h>
31#include <asm/xen/hypercall.h>
32
33#include <xen/xen.h>
34#include <xen/page.h>
35#include <xen/events.h>
36
37#include <xen/hvc-console.h>
38#include "xen-ops.h"
39#include "mmu.h"
40
41cpumask_var_t xen_cpu_initialized_map;
42
43static DEFINE_PER_CPU(int, xen_resched_irq);
44static DEFINE_PER_CPU(int, xen_callfunc_irq);
45static DEFINE_PER_CPU(int, xen_callfuncsingle_irq);
46static DEFINE_PER_CPU(int, xen_irq_work);
47static DEFINE_PER_CPU(int, xen_debug_irq) = -1;
48
49static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
50static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
51static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id);
52
53
54
55
56static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
57{
58 inc_irq_stat(irq_resched_count);
59 scheduler_ipi();
60
61 return IRQ_HANDLED;
62}
63
64static void __cpuinit cpu_bringup(void)
65{
66 int cpu;
67
68 cpu_init();
69 touch_softlockup_watchdog();
70 preempt_disable();
71
72 xen_enable_sysenter();
73 xen_enable_syscall();
74
75 cpu = smp_processor_id();
76 smp_store_cpu_info(cpu);
77 cpu_data(cpu).x86_max_cores = 1;
78 set_cpu_sibling_map(cpu);
79
80 xen_setup_cpu_clockevents();
81
82 notify_cpu_starting(cpu);
83
84 set_cpu_online(cpu, true);
85
86 this_cpu_write(cpu_state, CPU_ONLINE);
87
88 wmb();
89
90
91 local_irq_enable();
92
93 wmb();
94}
95
96static void __cpuinit cpu_bringup_and_idle(void)
97{
98 cpu_bringup();
99 cpu_startup_entry(CPUHP_ONLINE);
100}
101
102static int xen_smp_intr_init(unsigned int cpu)
103{
104 int rc;
105 const char *resched_name, *callfunc_name, *debug_name;
106
107 resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
108 rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
109 cpu,
110 xen_reschedule_interrupt,
111 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
112 resched_name,
113 NULL);
114 if (rc < 0)
115 goto fail;
116 per_cpu(xen_resched_irq, cpu) = rc;
117
118 callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu);
119 rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR,
120 cpu,
121 xen_call_function_interrupt,
122 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
123 callfunc_name,
124 NULL);
125 if (rc < 0)
126 goto fail;
127 per_cpu(xen_callfunc_irq, cpu) = rc;
128
129 debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu);
130 rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt,
131 IRQF_DISABLED | IRQF_PERCPU | IRQF_NOBALANCING,
132 debug_name, NULL);
133 if (rc < 0)
134 goto fail;
135 per_cpu(xen_debug_irq, cpu) = rc;
136
137 callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu);
138 rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR,
139 cpu,
140 xen_call_function_single_interrupt,
141 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
142 callfunc_name,
143 NULL);
144 if (rc < 0)
145 goto fail;
146 per_cpu(xen_callfuncsingle_irq, cpu) = rc;
147
148
149
150
151
152 if (xen_hvm_domain())
153 return 0;
154
155 callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu);
156 rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR,
157 cpu,
158 xen_irq_work_interrupt,
159 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
160 callfunc_name,
161 NULL);
162 if (rc < 0)
163 goto fail;
164 per_cpu(xen_irq_work, cpu) = rc;
165
166 return 0;
167
168 fail:
169 if (per_cpu(xen_resched_irq, cpu) >= 0)
170 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
171 if (per_cpu(xen_callfunc_irq, cpu) >= 0)
172 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
173 if (per_cpu(xen_debug_irq, cpu) >= 0)
174 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
175 if (per_cpu(xen_callfuncsingle_irq, cpu) >= 0)
176 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu),
177 NULL);
178 if (xen_hvm_domain())
179 return rc;
180
181 if (per_cpu(xen_irq_work, cpu) >= 0)
182 unbind_from_irqhandler(per_cpu(xen_irq_work, cpu), NULL);
183
184 return rc;
185}
186
187static void __init xen_fill_possible_map(void)
188{
189 int i, rc;
190
191 if (xen_initial_domain())
192 return;
193
194 for (i = 0; i < nr_cpu_ids; i++) {
195 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
196 if (rc >= 0) {
197 num_processors++;
198 set_cpu_possible(i, true);
199 }
200 }
201}
202
203static void __init xen_filter_cpu_maps(void)
204{
205 int i, rc;
206 unsigned int subtract = 0;
207
208 if (!xen_initial_domain())
209 return;
210
211 num_processors = 0;
212 disabled_cpus = 0;
213 for (i = 0; i < nr_cpu_ids; i++) {
214 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
215 if (rc >= 0) {
216 num_processors++;
217 set_cpu_possible(i, true);
218 } else {
219 set_cpu_possible(i, false);
220 set_cpu_present(i, false);
221 subtract++;
222 }
223 }
224#ifdef CONFIG_HOTPLUG_CPU
225
226
227
228
229
230
231
232
233 if (subtract)
234 nr_cpu_ids = nr_cpu_ids - subtract;
235#endif
236
237}
238
239static void __init xen_smp_prepare_boot_cpu(void)
240{
241 BUG_ON(smp_processor_id() != 0);
242 native_smp_prepare_boot_cpu();
243
244
245
246 make_lowmem_page_readwrite(xen_initial_gdt);
247
248 xen_filter_cpu_maps();
249 xen_setup_vcpu_info_placement();
250}
251
252static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
253{
254 unsigned cpu;
255 unsigned int i;
256
257 if (skip_ioapic_setup) {
258 char *m = (max_cpus == 0) ?
259 "The nosmp parameter is incompatible with Xen; " \
260 "use Xen dom0_max_vcpus=1 parameter" :
261 "The noapic parameter is incompatible with Xen";
262
263 xen_raw_printk(m);
264 panic(m);
265 }
266 xen_init_lock_cpu(0);
267
268 smp_store_boot_cpu_info();
269 cpu_data(0).x86_max_cores = 1;
270
271 for_each_possible_cpu(i) {
272 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
273 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
274 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
275 }
276 set_cpu_sibling_map(0);
277
278 if (xen_smp_intr_init(0))
279 BUG();
280
281 if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
282 panic("could not allocate xen_cpu_initialized_map\n");
283
284 cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
285
286
287 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
288 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
289 continue;
290 set_cpu_possible(cpu, false);
291 }
292
293 for_each_possible_cpu(cpu)
294 set_cpu_present(cpu, true);
295}
296
297static int __cpuinit
298cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
299{
300 struct vcpu_guest_context *ctxt;
301 struct desc_struct *gdt;
302 unsigned long gdt_mfn;
303
304 if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
305 return 0;
306
307 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
308 if (ctxt == NULL)
309 return -ENOMEM;
310
311 gdt = get_cpu_gdt_table(cpu);
312
313 ctxt->flags = VGCF_IN_KERNEL;
314 ctxt->user_regs.ss = __KERNEL_DS;
315#ifdef CONFIG_X86_32
316 ctxt->user_regs.fs = __KERNEL_PERCPU;
317 ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
318#else
319 ctxt->gs_base_kernel = per_cpu_offset(cpu);
320#endif
321 ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
322
323 memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
324
325 {
326 ctxt->user_regs.eflags = 0x1000;
327 ctxt->user_regs.ds = __USER_DS;
328 ctxt->user_regs.es = __USER_DS;
329
330 xen_copy_trap_info(ctxt->trap_ctxt);
331
332 ctxt->ldt_ents = 0;
333
334 BUG_ON((unsigned long)gdt & ~PAGE_MASK);
335
336 gdt_mfn = arbitrary_virt_to_mfn(gdt);
337 make_lowmem_page_readonly(gdt);
338 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
339
340 ctxt->gdt_frames[0] = gdt_mfn;
341 ctxt->gdt_ents = GDT_ENTRIES;
342
343 ctxt->kernel_ss = __KERNEL_DS;
344 ctxt->kernel_sp = idle->thread.sp0;
345
346#ifdef CONFIG_X86_32
347 ctxt->event_callback_cs = __KERNEL_CS;
348 ctxt->failsafe_callback_cs = __KERNEL_CS;
349#endif
350 ctxt->event_callback_eip =
351 (unsigned long)xen_hypervisor_callback;
352 ctxt->failsafe_callback_eip =
353 (unsigned long)xen_failsafe_callback;
354 }
355 ctxt->user_regs.cs = __KERNEL_CS;
356 ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
357
358 per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
359 ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
360
361 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
362 BUG();
363
364 kfree(ctxt);
365 return 0;
366}
367
368static int __cpuinit xen_cpu_up(unsigned int cpu, struct task_struct *idle)
369{
370 int rc;
371
372 per_cpu(current_task, cpu) = idle;
373#ifdef CONFIG_X86_32
374 irq_ctx_init(cpu);
375#else
376 clear_tsk_thread_flag(idle, TIF_FORK);
377 per_cpu(kernel_stack, cpu) =
378 (unsigned long)task_stack_page(idle) -
379 KERNEL_STACK_OFFSET + THREAD_SIZE;
380#endif
381 xen_setup_runstate_info(cpu);
382 xen_setup_timer(cpu);
383 xen_init_lock_cpu(cpu);
384
385 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
386
387
388 per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
389
390 rc = cpu_initialize_context(cpu, idle);
391 if (rc)
392 return rc;
393
394 if (num_online_cpus() == 1)
395
396 alternatives_enable_smp();
397
398 rc = xen_smp_intr_init(cpu);
399 if (rc)
400 return rc;
401
402 rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
403 BUG_ON(rc);
404
405 while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
406 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
407 barrier();
408 }
409
410 return 0;
411}
412
413static void xen_smp_cpus_done(unsigned int max_cpus)
414{
415}
416
417#ifdef CONFIG_HOTPLUG_CPU
418static int xen_cpu_disable(void)
419{
420 unsigned int cpu = smp_processor_id();
421 if (cpu == 0)
422 return -EBUSY;
423
424 cpu_disable_common();
425
426 load_cr3(swapper_pg_dir);
427 return 0;
428}
429
430static void xen_cpu_die(unsigned int cpu)
431{
432 while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
433 current->state = TASK_UNINTERRUPTIBLE;
434 schedule_timeout(HZ/10);
435 }
436 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
437 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
438 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
439 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
440 if (!xen_hvm_domain())
441 unbind_from_irqhandler(per_cpu(xen_irq_work, cpu), NULL);
442 xen_uninit_lock_cpu(cpu);
443 xen_teardown_timer(cpu);
444}
445
446static void __cpuinit xen_play_dead(void)
447{
448 play_dead_common();
449 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
450 cpu_bringup();
451
452
453
454
455
456
457 tick_nohz_idle_enter();
458}
459
460#else
461static int xen_cpu_disable(void)
462{
463 return -ENOSYS;
464}
465
466static void xen_cpu_die(unsigned int cpu)
467{
468 BUG();
469}
470
471static void xen_play_dead(void)
472{
473 BUG();
474}
475
476#endif
477static void stop_self(void *v)
478{
479 int cpu = smp_processor_id();
480
481
482 load_cr3(swapper_pg_dir);
483
484
485 set_cpu_online(cpu, false);
486
487 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL);
488 BUG();
489}
490
491static void xen_stop_other_cpus(int wait)
492{
493 smp_call_function(stop_self, NULL, wait);
494}
495
496static void xen_smp_send_reschedule(int cpu)
497{
498 xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
499}
500
501static void __xen_send_IPI_mask(const struct cpumask *mask,
502 int vector)
503{
504 unsigned cpu;
505
506 for_each_cpu_and(cpu, mask, cpu_online_mask)
507 xen_send_IPI_one(cpu, vector);
508}
509
510static void xen_smp_send_call_function_ipi(const struct cpumask *mask)
511{
512 int cpu;
513
514 __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
515
516
517 for_each_cpu(cpu, mask) {
518 if (xen_vcpu_stolen(cpu)) {
519 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
520 break;
521 }
522 }
523}
524
525static void xen_smp_send_call_function_single_ipi(int cpu)
526{
527 __xen_send_IPI_mask(cpumask_of(cpu),
528 XEN_CALL_FUNCTION_SINGLE_VECTOR);
529}
530
531static inline int xen_map_vector(int vector)
532{
533 int xen_vector;
534
535 switch (vector) {
536 case RESCHEDULE_VECTOR:
537 xen_vector = XEN_RESCHEDULE_VECTOR;
538 break;
539 case CALL_FUNCTION_VECTOR:
540 xen_vector = XEN_CALL_FUNCTION_VECTOR;
541 break;
542 case CALL_FUNCTION_SINGLE_VECTOR:
543 xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR;
544 break;
545 case IRQ_WORK_VECTOR:
546 xen_vector = XEN_IRQ_WORK_VECTOR;
547 break;
548 default:
549 xen_vector = -1;
550 printk(KERN_ERR "xen: vector 0x%x is not implemented\n",
551 vector);
552 }
553
554 return xen_vector;
555}
556
557void xen_send_IPI_mask(const struct cpumask *mask,
558 int vector)
559{
560 int xen_vector = xen_map_vector(vector);
561
562 if (xen_vector >= 0)
563 __xen_send_IPI_mask(mask, xen_vector);
564}
565
566void xen_send_IPI_all(int vector)
567{
568 int xen_vector = xen_map_vector(vector);
569
570 if (xen_vector >= 0)
571 __xen_send_IPI_mask(cpu_online_mask, xen_vector);
572}
573
574void xen_send_IPI_self(int vector)
575{
576 int xen_vector = xen_map_vector(vector);
577
578 if (xen_vector >= 0)
579 xen_send_IPI_one(smp_processor_id(), xen_vector);
580}
581
582void xen_send_IPI_mask_allbutself(const struct cpumask *mask,
583 int vector)
584{
585 unsigned cpu;
586 unsigned int this_cpu = smp_processor_id();
587 int xen_vector = xen_map_vector(vector);
588
589 if (!(num_online_cpus() > 1) || (xen_vector < 0))
590 return;
591
592 for_each_cpu_and(cpu, mask, cpu_online_mask) {
593 if (this_cpu == cpu)
594 continue;
595
596 xen_send_IPI_one(cpu, xen_vector);
597 }
598}
599
600void xen_send_IPI_allbutself(int vector)
601{
602 xen_send_IPI_mask_allbutself(cpu_online_mask, vector);
603}
604
605static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
606{
607 irq_enter();
608 generic_smp_call_function_interrupt();
609 inc_irq_stat(irq_call_count);
610 irq_exit();
611
612 return IRQ_HANDLED;
613}
614
615static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
616{
617 irq_enter();
618 generic_smp_call_function_single_interrupt();
619 inc_irq_stat(irq_call_count);
620 irq_exit();
621
622 return IRQ_HANDLED;
623}
624
625static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id)
626{
627 irq_enter();
628 irq_work_run();
629 inc_irq_stat(apic_irq_work_irqs);
630 irq_exit();
631
632 return IRQ_HANDLED;
633}
634
635static const struct smp_ops xen_smp_ops __initconst = {
636 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
637 .smp_prepare_cpus = xen_smp_prepare_cpus,
638 .smp_cpus_done = xen_smp_cpus_done,
639
640 .cpu_up = xen_cpu_up,
641 .cpu_die = xen_cpu_die,
642 .cpu_disable = xen_cpu_disable,
643 .play_dead = xen_play_dead,
644
645 .stop_other_cpus = xen_stop_other_cpus,
646 .smp_send_reschedule = xen_smp_send_reschedule,
647
648 .send_call_func_ipi = xen_smp_send_call_function_ipi,
649 .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
650};
651
652void __init xen_smp_init(void)
653{
654 smp_ops = xen_smp_ops;
655 xen_fill_possible_map();
656 xen_init_spinlocks();
657}
658
659static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus)
660{
661 native_smp_prepare_cpus(max_cpus);
662 WARN_ON(xen_smp_intr_init(0));
663
664 xen_init_lock_cpu(0);
665}
666
667static int __cpuinit xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle)
668{
669 int rc;
670 rc = native_cpu_up(cpu, tidle);
671 WARN_ON (xen_smp_intr_init(cpu));
672 return rc;
673}
674
675static void xen_hvm_cpu_die(unsigned int cpu)
676{
677 xen_cpu_die(cpu);
678 native_cpu_die(cpu);
679}
680
681void __init xen_hvm_smp_init(void)
682{
683 if (!xen_have_vector_callback)
684 return;
685 smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
686 smp_ops.smp_send_reschedule = xen_smp_send_reschedule;
687 smp_ops.cpu_up = xen_hvm_cpu_up;
688 smp_ops.cpu_die = xen_hvm_cpu_die;
689 smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
690 smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
691}
692
