Merge tag 'stable/for-linus-3.4-tag' of git://git.kernel.org/pub/scm/linux/kernel...
[~shefty/rdma-dev.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/hvc-console.h>
36 #include "xen-ops.h"
37 #include "mmu.h"
38
39 cpumask_var_t xen_cpu_initialized_map;
40
41 static DEFINE_PER_CPU(int, xen_resched_irq);
42 static DEFINE_PER_CPU(int, xen_callfunc_irq);
43 static DEFINE_PER_CPU(int, xen_callfuncsingle_irq);
44 static DEFINE_PER_CPU(int, xen_debug_irq) = -1;
45
46 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
47 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
48
49 /*
50  * Reschedule call back.
51  */
52 static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
53 {
54         inc_irq_stat(irq_resched_count);
55         scheduler_ipi();
56
57         return IRQ_HANDLED;
58 }
59
60 static void __cpuinit cpu_bringup(void)
61 {
62         int cpu = smp_processor_id();
63
64         cpu_init();
65         touch_softlockup_watchdog();
66         preempt_disable();
67
68         xen_enable_sysenter();
69         xen_enable_syscall();
70
71         cpu = smp_processor_id();
72         smp_store_cpu_info(cpu);
73         cpu_data(cpu).x86_max_cores = 1;
74         set_cpu_sibling_map(cpu);
75
76         xen_setup_cpu_clockevents();
77
78         set_cpu_online(cpu, true);
79         this_cpu_write(cpu_state, CPU_ONLINE);
80         wmb();
81
82         /* We can take interrupts now: we're officially "up". */
83         local_irq_enable();
84
85         wmb();                  /* make sure everything is out */
86 }
87
88 static void __cpuinit cpu_bringup_and_idle(void)
89 {
90         cpu_bringup();
91         cpu_idle();
92 }
93
94 static int xen_smp_intr_init(unsigned int cpu)
95 {
96         int rc;
97         const char *resched_name, *callfunc_name, *debug_name;
98
99         resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
100         rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
101                                     cpu,
102                                     xen_reschedule_interrupt,
103                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
104                                     resched_name,
105                                     NULL);
106         if (rc < 0)
107                 goto fail;
108         per_cpu(xen_resched_irq, cpu) = rc;
109
110         callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu);
111         rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR,
112                                     cpu,
113                                     xen_call_function_interrupt,
114                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
115                                     callfunc_name,
116                                     NULL);
117         if (rc < 0)
118                 goto fail;
119         per_cpu(xen_callfunc_irq, cpu) = rc;
120
121         debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu);
122         rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt,
123                                      IRQF_DISABLED | IRQF_PERCPU | IRQF_NOBALANCING,
124                                      debug_name, NULL);
125         if (rc < 0)
126                 goto fail;
127         per_cpu(xen_debug_irq, cpu) = rc;
128
129         callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu);
130         rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR,
131                                     cpu,
132                                     xen_call_function_single_interrupt,
133                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
134                                     callfunc_name,
135                                     NULL);
136         if (rc < 0)
137                 goto fail;
138         per_cpu(xen_callfuncsingle_irq, cpu) = rc;
139
140         return 0;
141
142  fail:
143         if (per_cpu(xen_resched_irq, cpu) >= 0)
144                 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
145         if (per_cpu(xen_callfunc_irq, cpu) >= 0)
146                 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
147         if (per_cpu(xen_debug_irq, cpu) >= 0)
148                 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
149         if (per_cpu(xen_callfuncsingle_irq, cpu) >= 0)
150                 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu),
151                                        NULL);
152
153         return rc;
154 }
155
156 static void __init xen_fill_possible_map(void)
157 {
158         int i, rc;
159
160         if (xen_initial_domain())
161                 return;
162
163         for (i = 0; i < nr_cpu_ids; i++) {
164                 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
165                 if (rc >= 0) {
166                         num_processors++;
167                         set_cpu_possible(i, true);
168                 }
169         }
170 }
171
172 static void __init xen_filter_cpu_maps(void)
173 {
174         int i, rc;
175
176         if (!xen_initial_domain())
177                 return;
178
179         num_processors = 0;
180         disabled_cpus = 0;
181         for (i = 0; i < nr_cpu_ids; i++) {
182                 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
183                 if (rc >= 0) {
184                         num_processors++;
185                         set_cpu_possible(i, true);
186                 } else {
187                         set_cpu_possible(i, false);
188                         set_cpu_present(i, false);
189                 }
190         }
191 }
192
193 static void __init xen_smp_prepare_boot_cpu(void)
194 {
195         BUG_ON(smp_processor_id() != 0);
196         native_smp_prepare_boot_cpu();
197
198         /* We've switched to the "real" per-cpu gdt, so make sure the
199            old memory can be recycled */
200         make_lowmem_page_readwrite(xen_initial_gdt);
201
202         xen_filter_cpu_maps();
203         xen_setup_vcpu_info_placement();
204 }
205
206 static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
207 {
208         unsigned cpu;
209         unsigned int i;
210
211         if (skip_ioapic_setup) {
212                 char *m = (max_cpus == 0) ?
213                         "The nosmp parameter is incompatible with Xen; " \
214                         "use Xen dom0_max_vcpus=1 parameter" :
215                         "The noapic parameter is incompatible with Xen";
216
217                 xen_raw_printk(m);
218                 panic(m);
219         }
220         xen_init_lock_cpu(0);
221
222         smp_store_cpu_info(0);
223         cpu_data(0).x86_max_cores = 1;
224
225         for_each_possible_cpu(i) {
226                 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
227                 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
228                 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
229         }
230         set_cpu_sibling_map(0);
231
232         if (xen_smp_intr_init(0))
233                 BUG();
234
235         if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
236                 panic("could not allocate xen_cpu_initialized_map\n");
237
238         cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
239
240         /* Restrict the possible_map according to max_cpus. */
241         while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
242                 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
243                         continue;
244                 set_cpu_possible(cpu, false);
245         }
246
247         for_each_possible_cpu (cpu) {
248                 struct task_struct *idle;
249
250                 if (cpu == 0)
251                         continue;
252
253                 idle = fork_idle(cpu);
254                 if (IS_ERR(idle))
255                         panic("failed fork for CPU %d", cpu);
256
257                 set_cpu_present(cpu, true);
258         }
259 }
260
261 static int __cpuinit
262 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
263 {
264         struct vcpu_guest_context *ctxt;
265         struct desc_struct *gdt;
266         unsigned long gdt_mfn;
267
268         if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
269                 return 0;
270
271         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
272         if (ctxt == NULL)
273                 return -ENOMEM;
274
275         gdt = get_cpu_gdt_table(cpu);
276
277         ctxt->flags = VGCF_IN_KERNEL;
278         ctxt->user_regs.ds = __USER_DS;
279         ctxt->user_regs.es = __USER_DS;
280         ctxt->user_regs.ss = __KERNEL_DS;
281 #ifdef CONFIG_X86_32
282         ctxt->user_regs.fs = __KERNEL_PERCPU;
283         ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
284 #else
285         ctxt->gs_base_kernel = per_cpu_offset(cpu);
286 #endif
287         ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
288         ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
289
290         memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
291
292         xen_copy_trap_info(ctxt->trap_ctxt);
293
294         ctxt->ldt_ents = 0;
295
296         BUG_ON((unsigned long)gdt & ~PAGE_MASK);
297
298         gdt_mfn = arbitrary_virt_to_mfn(gdt);
299         make_lowmem_page_readonly(gdt);
300         make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
301
302         ctxt->gdt_frames[0] = gdt_mfn;
303         ctxt->gdt_ents      = GDT_ENTRIES;
304
305         ctxt->user_regs.cs = __KERNEL_CS;
306         ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
307
308         ctxt->kernel_ss = __KERNEL_DS;
309         ctxt->kernel_sp = idle->thread.sp0;
310
311 #ifdef CONFIG_X86_32
312         ctxt->event_callback_cs     = __KERNEL_CS;
313         ctxt->failsafe_callback_cs  = __KERNEL_CS;
314 #endif
315         ctxt->event_callback_eip    = (unsigned long)xen_hypervisor_callback;
316         ctxt->failsafe_callback_eip = (unsigned long)xen_failsafe_callback;
317
318         per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
319         ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
320
321         if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
322                 BUG();
323
324         kfree(ctxt);
325         return 0;
326 }
327
328 static int __cpuinit xen_cpu_up(unsigned int cpu)
329 {
330         struct task_struct *idle = idle_task(cpu);
331         int rc;
332
333         per_cpu(current_task, cpu) = idle;
334 #ifdef CONFIG_X86_32
335         irq_ctx_init(cpu);
336 #else
337         clear_tsk_thread_flag(idle, TIF_FORK);
338         per_cpu(kernel_stack, cpu) =
339                 (unsigned long)task_stack_page(idle) -
340                 KERNEL_STACK_OFFSET + THREAD_SIZE;
341 #endif
342         xen_setup_runstate_info(cpu);
343         xen_setup_timer(cpu);
344         xen_init_lock_cpu(cpu);
345
346         per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
347
348         /* make sure interrupts start blocked */
349         per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
350
351         rc = cpu_initialize_context(cpu, idle);
352         if (rc)
353                 return rc;
354
355         if (num_online_cpus() == 1)
356                 alternatives_smp_switch(1);
357
358         rc = xen_smp_intr_init(cpu);
359         if (rc)
360                 return rc;
361
362         rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
363         BUG_ON(rc);
364
365         while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
366                 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
367                 barrier();
368         }
369
370         return 0;
371 }
372
373 static void xen_smp_cpus_done(unsigned int max_cpus)
374 {
375 }
376
377 #ifdef CONFIG_HOTPLUG_CPU
378 static int xen_cpu_disable(void)
379 {
380         unsigned int cpu = smp_processor_id();
381         if (cpu == 0)
382                 return -EBUSY;
383
384         cpu_disable_common();
385
386         load_cr3(swapper_pg_dir);
387         return 0;
388 }
389
390 static void xen_cpu_die(unsigned int cpu)
391 {
392         while (HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
393                 current->state = TASK_UNINTERRUPTIBLE;
394                 schedule_timeout(HZ/10);
395         }
396         unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
397         unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
398         unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
399         unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
400         xen_uninit_lock_cpu(cpu);
401         xen_teardown_timer(cpu);
402
403         if (num_online_cpus() == 1)
404                 alternatives_smp_switch(0);
405 }
406
407 static void __cpuinit xen_play_dead(void) /* used only with HOTPLUG_CPU */
408 {
409         play_dead_common();
410         HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
411         cpu_bringup();
412         /*
413          * Balance out the preempt calls - as we are running in cpu_idle
414          * loop which has been called at bootup from cpu_bringup_and_idle.
415          * The cpucpu_bringup_and_idle called cpu_bringup which made a
416          * preempt_disable() So this preempt_enable will balance it out.
417          */
418         preempt_enable();
419 }
420
421 #else /* !CONFIG_HOTPLUG_CPU */
422 static int xen_cpu_disable(void)
423 {
424         return -ENOSYS;
425 }
426
427 static void xen_cpu_die(unsigned int cpu)
428 {
429         BUG();
430 }
431
432 static void xen_play_dead(void)
433 {
434         BUG();
435 }
436
437 #endif
438 static void stop_self(void *v)
439 {
440         int cpu = smp_processor_id();
441
442         /* make sure we're not pinning something down */
443         load_cr3(swapper_pg_dir);
444         /* should set up a minimal gdt */
445
446         set_cpu_online(cpu, false);
447
448         HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL);
449         BUG();
450 }
451
452 static void xen_stop_other_cpus(int wait)
453 {
454         smp_call_function(stop_self, NULL, wait);
455 }
456
457 static void xen_smp_send_reschedule(int cpu)
458 {
459         xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
460 }
461
462 static void xen_send_IPI_mask(const struct cpumask *mask,
463                               enum ipi_vector vector)
464 {
465         unsigned cpu;
466
467         for_each_cpu_and(cpu, mask, cpu_online_mask)
468                 xen_send_IPI_one(cpu, vector);
469 }
470
471 static void xen_smp_send_call_function_ipi(const struct cpumask *mask)
472 {
473         int cpu;
474
475         xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
476
477         /* Make sure other vcpus get a chance to run if they need to. */
478         for_each_cpu(cpu, mask) {
479                 if (xen_vcpu_stolen(cpu)) {
480                         HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
481                         break;
482                 }
483         }
484 }
485
486 static void xen_smp_send_call_function_single_ipi(int cpu)
487 {
488         xen_send_IPI_mask(cpumask_of(cpu),
489                           XEN_CALL_FUNCTION_SINGLE_VECTOR);
490 }
491
492 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
493 {
494         irq_enter();
495         generic_smp_call_function_interrupt();
496         inc_irq_stat(irq_call_count);
497         irq_exit();
498
499         return IRQ_HANDLED;
500 }
501
502 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
503 {
504         irq_enter();
505         generic_smp_call_function_single_interrupt();
506         inc_irq_stat(irq_call_count);
507         irq_exit();
508
509         return IRQ_HANDLED;
510 }
511
512 static const struct smp_ops xen_smp_ops __initconst = {
513         .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
514         .smp_prepare_cpus = xen_smp_prepare_cpus,
515         .smp_cpus_done = xen_smp_cpus_done,
516
517         .cpu_up = xen_cpu_up,
518         .cpu_die = xen_cpu_die,
519         .cpu_disable = xen_cpu_disable,
520         .play_dead = xen_play_dead,
521
522         .stop_other_cpus = xen_stop_other_cpus,
523         .smp_send_reschedule = xen_smp_send_reschedule,
524
525         .send_call_func_ipi = xen_smp_send_call_function_ipi,
526         .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
527 };
528
529 void __init xen_smp_init(void)
530 {
531         smp_ops = xen_smp_ops;
532         xen_fill_possible_map();
533         xen_init_spinlocks();
534 }
535
536 static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus)
537 {
538         native_smp_prepare_cpus(max_cpus);
539         WARN_ON(xen_smp_intr_init(0));
540
541         xen_init_lock_cpu(0);
542 }
543
544 static int __cpuinit xen_hvm_cpu_up(unsigned int cpu)
545 {
546         int rc;
547         rc = native_cpu_up(cpu);
548         WARN_ON (xen_smp_intr_init(cpu));
549         return rc;
550 }
551
552 static void xen_hvm_cpu_die(unsigned int cpu)
553 {
554         unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
555         unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
556         unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
557         unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
558         native_cpu_die(cpu);
559 }
560
561 void __init xen_hvm_smp_init(void)
562 {
563         if (!xen_have_vector_callback)
564                 return;
565         smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
566         smp_ops.smp_send_reschedule = xen_smp_send_reschedule;
567         smp_ops.cpu_up = xen_hvm_cpu_up;
568         smp_ops.cpu_die = xen_hvm_cpu_die;
569         smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
570         smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
571 }