Merge branches 'sched-core-for-linus' and 'sched-urgent-for-linus' of git://git.kerne...
[linux-2.6.git] / arch / x86 / xen / smp.c
1 /*
2  * Xen SMP support
3  *
4  * This file implements the Xen versions of smp_ops.  SMP under Xen is
5  * very straightforward.  Bringing a CPU up is simply a matter of
6  * loading its initial context and setting it running.
7  *
8  * IPIs are handled through the Xen event mechanism.
9  *
10  * Because virtual CPUs can be scheduled onto any real CPU, there's no
11  * useful topology information for the kernel to make use of.  As a
12  * result, all CPUs are treated as if they're single-core and
13  * single-threaded.
14  */
15 #include <linux/sched.h>
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/smp.h>
19
20 #include <asm/paravirt.h>
21 #include <asm/desc.h>
22 #include <asm/pgtable.h>
23 #include <asm/cpu.h>
24
25 #include <xen/interface/xen.h>
26 #include <xen/interface/vcpu.h>
27
28 #include <asm/xen/interface.h>
29 #include <asm/xen/hypercall.h>
30
31 #include <xen/xen.h>
32 #include <xen/page.h>
33 #include <xen/events.h>
34
35 #include "xen-ops.h"
36 #include "mmu.h"
37
38 cpumask_var_t xen_cpu_initialized_map;
39
40 static DEFINE_PER_CPU(int, xen_resched_irq);
41 static DEFINE_PER_CPU(int, xen_callfunc_irq);
42 static DEFINE_PER_CPU(int, xen_callfuncsingle_irq);
43 static DEFINE_PER_CPU(int, xen_debug_irq) = -1;
44
45 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
46 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
47
48 /*
49  * Reschedule call back.
50  */
51 static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
52 {
53         inc_irq_stat(irq_resched_count);
54         scheduler_ipi();
55
56         return IRQ_HANDLED;
57 }
58
59 static void __cpuinit cpu_bringup(void)
60 {
61         int cpu = smp_processor_id();
62
63         cpu_init();
64         touch_softlockup_watchdog();
65         preempt_disable();
66
67         xen_enable_sysenter();
68         xen_enable_syscall();
69
70         cpu = smp_processor_id();
71         smp_store_cpu_info(cpu);
72         cpu_data(cpu).x86_max_cores = 1;
73         set_cpu_sibling_map(cpu);
74
75         xen_setup_cpu_clockevents();
76
77         set_cpu_online(cpu, true);
78         percpu_write(cpu_state, CPU_ONLINE);
79         wmb();
80
81         /* We can take interrupts now: we're officially "up". */
82         local_irq_enable();
83
84         wmb();                  /* make sure everything is out */
85 }
86
87 static void __cpuinit cpu_bringup_and_idle(void)
88 {
89         cpu_bringup();
90         cpu_idle();
91 }
92
93 static int xen_smp_intr_init(unsigned int cpu)
94 {
95         int rc;
96         const char *resched_name, *callfunc_name, *debug_name;
97
98         resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
99         rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
100                                     cpu,
101                                     xen_reschedule_interrupt,
102                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
103                                     resched_name,
104                                     NULL);
105         if (rc < 0)
106                 goto fail;
107         per_cpu(xen_resched_irq, cpu) = rc;
108
109         callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu);
110         rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR,
111                                     cpu,
112                                     xen_call_function_interrupt,
113                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
114                                     callfunc_name,
115                                     NULL);
116         if (rc < 0)
117                 goto fail;
118         per_cpu(xen_callfunc_irq, cpu) = rc;
119
120         debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu);
121         rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt,
122                                      IRQF_DISABLED | IRQF_PERCPU | IRQF_NOBALANCING,
123                                      debug_name, NULL);
124         if (rc < 0)
125                 goto fail;
126         per_cpu(xen_debug_irq, cpu) = rc;
127
128         callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu);
129         rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR,
130                                     cpu,
131                                     xen_call_function_single_interrupt,
132                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
133                                     callfunc_name,
134                                     NULL);
135         if (rc < 0)
136                 goto fail;
137         per_cpu(xen_callfuncsingle_irq, cpu) = rc;
138
139         return 0;
140
141  fail:
142         if (per_cpu(xen_resched_irq, cpu) >= 0)
143                 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
144         if (per_cpu(xen_callfunc_irq, cpu) >= 0)
145                 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
146         if (per_cpu(xen_debug_irq, cpu) >= 0)
147                 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
148         if (per_cpu(xen_callfuncsingle_irq, cpu) >= 0)
149                 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu),
150                                        NULL);
151
152         return rc;
153 }
154
155 static void __init xen_fill_possible_map(void)
156 {
157         int i, rc;
158
159         if (xen_initial_domain())
160                 return;
161
162         for (i = 0; i < nr_cpu_ids; i++) {
163                 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
164                 if (rc >= 0) {
165                         num_processors++;
166                         set_cpu_possible(i, true);
167                 }
168         }
169 }
170
171 static void __init xen_filter_cpu_maps(void)
172 {
173         int i, rc;
174
175         if (!xen_initial_domain())
176                 return;
177
178         num_processors = 0;
179         disabled_cpus = 0;
180         for (i = 0; i < nr_cpu_ids; i++) {
181                 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
182                 if (rc >= 0) {
183                         num_processors++;
184                         set_cpu_possible(i, true);
185                 } else {
186                         set_cpu_possible(i, false);
187                         set_cpu_present(i, false);
188                 }
189         }
190 }
191
192 static void __init xen_smp_prepare_boot_cpu(void)
193 {
194         BUG_ON(smp_processor_id() != 0);
195         native_smp_prepare_boot_cpu();
196
197         /* We've switched to the "real" per-cpu gdt, so make sure the
198            old memory can be recycled */
199         make_lowmem_page_readwrite(xen_initial_gdt);
200
201         xen_filter_cpu_maps();
202         xen_setup_vcpu_info_placement();
203 }
204
205 static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
206 {
207         unsigned cpu;
208
209         xen_init_lock_cpu(0);
210
211         smp_store_cpu_info(0);
212         cpu_data(0).x86_max_cores = 1;
213         set_cpu_sibling_map(0);
214
215         if (xen_smp_intr_init(0))
216                 BUG();
217
218         if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
219                 panic("could not allocate xen_cpu_initialized_map\n");
220
221         cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
222
223         /* Restrict the possible_map according to max_cpus. */
224         while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
225                 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
226                         continue;
227                 set_cpu_possible(cpu, false);
228         }
229
230         for_each_possible_cpu (cpu) {
231                 struct task_struct *idle;
232
233                 if (cpu == 0)
234                         continue;
235
236                 idle = fork_idle(cpu);
237                 if (IS_ERR(idle))
238                         panic("failed fork for CPU %d", cpu);
239
240                 set_cpu_present(cpu, true);
241         }
242 }
243
244 static int __cpuinit
245 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
246 {
247         struct vcpu_guest_context *ctxt;
248         struct desc_struct *gdt;
249         unsigned long gdt_mfn;
250
251         if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
252                 return 0;
253
254         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
255         if (ctxt == NULL)
256                 return -ENOMEM;
257
258         gdt = get_cpu_gdt_table(cpu);
259
260         ctxt->flags = VGCF_IN_KERNEL;
261         ctxt->user_regs.ds = __USER_DS;
262         ctxt->user_regs.es = __USER_DS;
263         ctxt->user_regs.ss = __KERNEL_DS;
264 #ifdef CONFIG_X86_32
265         ctxt->user_regs.fs = __KERNEL_PERCPU;
266         ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
267 #else
268         ctxt->gs_base_kernel = per_cpu_offset(cpu);
269 #endif
270         ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
271         ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
272
273         memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
274
275         xen_copy_trap_info(ctxt->trap_ctxt);
276
277         ctxt->ldt_ents = 0;
278
279         BUG_ON((unsigned long)gdt & ~PAGE_MASK);
280
281         gdt_mfn = arbitrary_virt_to_mfn(gdt);
282         make_lowmem_page_readonly(gdt);
283         make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
284
285         ctxt->gdt_frames[0] = gdt_mfn;
286         ctxt->gdt_ents      = GDT_ENTRIES;
287
288         ctxt->user_regs.cs = __KERNEL_CS;
289         ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
290
291         ctxt->kernel_ss = __KERNEL_DS;
292         ctxt->kernel_sp = idle->thread.sp0;
293
294 #ifdef CONFIG_X86_32
295         ctxt->event_callback_cs     = __KERNEL_CS;
296         ctxt->failsafe_callback_cs  = __KERNEL_CS;
297 #endif
298         ctxt->event_callback_eip    = (unsigned long)xen_hypervisor_callback;
299         ctxt->failsafe_callback_eip = (unsigned long)xen_failsafe_callback;
300
301         per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
302         ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
303
304         if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
305                 BUG();
306
307         kfree(ctxt);
308         return 0;
309 }
310
311 static int __cpuinit xen_cpu_up(unsigned int cpu)
312 {
313         struct task_struct *idle = idle_task(cpu);
314         int rc;
315
316         per_cpu(current_task, cpu) = idle;
317 #ifdef CONFIG_X86_32
318         irq_ctx_init(cpu);
319 #else
320         clear_tsk_thread_flag(idle, TIF_FORK);
321         per_cpu(kernel_stack, cpu) =
322                 (unsigned long)task_stack_page(idle) -
323                 KERNEL_STACK_OFFSET + THREAD_SIZE;
324 #endif
325         xen_setup_runstate_info(cpu);
326         xen_setup_timer(cpu);
327         xen_init_lock_cpu(cpu);
328
329         per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
330
331         /* make sure interrupts start blocked */
332         per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
333
334         rc = cpu_initialize_context(cpu, idle);
335         if (rc)
336                 return rc;
337
338         if (num_online_cpus() == 1)
339                 alternatives_smp_switch(1);
340
341         rc = xen_smp_intr_init(cpu);
342         if (rc)
343                 return rc;
344
345         rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
346         BUG_ON(rc);
347
348         while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
349                 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
350                 barrier();
351         }
352
353         return 0;
354 }
355
356 static void xen_smp_cpus_done(unsigned int max_cpus)
357 {
358 }
359
360 #ifdef CONFIG_HOTPLUG_CPU
361 static int xen_cpu_disable(void)
362 {
363         unsigned int cpu = smp_processor_id();
364         if (cpu == 0)
365                 return -EBUSY;
366
367         cpu_disable_common();
368
369         load_cr3(swapper_pg_dir);
370         return 0;
371 }
372
373 static void xen_cpu_die(unsigned int cpu)
374 {
375         while (HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
376                 current->state = TASK_UNINTERRUPTIBLE;
377                 schedule_timeout(HZ/10);
378         }
379         unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
380         unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
381         unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
382         unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
383         xen_uninit_lock_cpu(cpu);
384         xen_teardown_timer(cpu);
385
386         if (num_online_cpus() == 1)
387                 alternatives_smp_switch(0);
388 }
389
390 static void __cpuinit xen_play_dead(void) /* used only with HOTPLUG_CPU */
391 {
392         play_dead_common();
393         HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
394         cpu_bringup();
395 }
396
397 #else /* !CONFIG_HOTPLUG_CPU */
398 static int xen_cpu_disable(void)
399 {
400         return -ENOSYS;
401 }
402
403 static void xen_cpu_die(unsigned int cpu)
404 {
405         BUG();
406 }
407
408 static void xen_play_dead(void)
409 {
410         BUG();
411 }
412
413 #endif
414 static void stop_self(void *v)
415 {
416         int cpu = smp_processor_id();
417
418         /* make sure we're not pinning something down */
419         load_cr3(swapper_pg_dir);
420         /* should set up a minimal gdt */
421
422         set_cpu_online(cpu, false);
423
424         HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL);
425         BUG();
426 }
427
428 static void xen_stop_other_cpus(int wait)
429 {
430         smp_call_function(stop_self, NULL, wait);
431 }
432
433 static void xen_smp_send_reschedule(int cpu)
434 {
435         xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
436 }
437
438 static void xen_send_IPI_mask(const struct cpumask *mask,
439                               enum ipi_vector vector)
440 {
441         unsigned cpu;
442
443         for_each_cpu_and(cpu, mask, cpu_online_mask)
444                 xen_send_IPI_one(cpu, vector);
445 }
446
447 static void xen_smp_send_call_function_ipi(const struct cpumask *mask)
448 {
449         int cpu;
450
451         xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
452
453         /* Make sure other vcpus get a chance to run if they need to. */
454         for_each_cpu(cpu, mask) {
455                 if (xen_vcpu_stolen(cpu)) {
456                         HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
457                         break;
458                 }
459         }
460 }
461
462 static void xen_smp_send_call_function_single_ipi(int cpu)
463 {
464         xen_send_IPI_mask(cpumask_of(cpu),
465                           XEN_CALL_FUNCTION_SINGLE_VECTOR);
466 }
467
468 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
469 {
470         irq_enter();
471         generic_smp_call_function_interrupt();
472         inc_irq_stat(irq_call_count);
473         irq_exit();
474
475         return IRQ_HANDLED;
476 }
477
478 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
479 {
480         irq_enter();
481         generic_smp_call_function_single_interrupt();
482         inc_irq_stat(irq_call_count);
483         irq_exit();
484
485         return IRQ_HANDLED;
486 }
487
488 static const struct smp_ops xen_smp_ops __initconst = {
489         .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
490         .smp_prepare_cpus = xen_smp_prepare_cpus,
491         .smp_cpus_done = xen_smp_cpus_done,
492
493         .cpu_up = xen_cpu_up,
494         .cpu_die = xen_cpu_die,
495         .cpu_disable = xen_cpu_disable,
496         .play_dead = xen_play_dead,
497
498         .stop_other_cpus = xen_stop_other_cpus,
499         .smp_send_reschedule = xen_smp_send_reschedule,
500
501         .send_call_func_ipi = xen_smp_send_call_function_ipi,
502         .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
503 };
504
505 void __init xen_smp_init(void)
506 {
507         smp_ops = xen_smp_ops;
508         xen_fill_possible_map();
509         xen_init_spinlocks();
510 }
511
512 static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus)
513 {
514         native_smp_prepare_cpus(max_cpus);
515         WARN_ON(xen_smp_intr_init(0));
516
517         if (!xen_have_vector_callback)
518                 return;
519         xen_init_lock_cpu(0);
520         xen_init_spinlocks();
521 }
522
523 static int __cpuinit xen_hvm_cpu_up(unsigned int cpu)
524 {
525         int rc;
526         rc = native_cpu_up(cpu);
527         WARN_ON (xen_smp_intr_init(cpu));
528         return rc;
529 }
530
531 static void xen_hvm_cpu_die(unsigned int cpu)
532 {
533         unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
534         unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
535         unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
536         unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
537         native_cpu_die(cpu);
538 }
539
540 void __init xen_hvm_smp_init(void)
541 {
542         smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
543         smp_ops.smp_send_reschedule = xen_smp_send_reschedule;
544         smp_ops.cpu_up = xen_hvm_cpu_up;
545         smp_ops.cpu_die = xen_hvm_cpu_die;
546         smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
547         smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
548 }