Merge branches 'x86/xen', 'x86/build', 'x86/microcode', 'x86/mm-debug-v2', 'x86/memor...
[linux-2.6] / 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/smp.h>
18
19 #include <asm/paravirt.h>
20 #include <asm/desc.h>
21 #include <asm/pgtable.h>
22 #include <asm/cpu.h>
23
24 #include <xen/interface/xen.h>
25 #include <xen/interface/vcpu.h>
26
27 #include <asm/xen/interface.h>
28 #include <asm/xen/hypercall.h>
29
30 #include <xen/page.h>
31 #include <xen/events.h>
32
33 #include "xen-ops.h"
34 #include "mmu.h"
35
36 cpumask_t xen_cpu_initialized_map;
37
38 static DEFINE_PER_CPU(int, resched_irq);
39 static DEFINE_PER_CPU(int, callfunc_irq);
40 static DEFINE_PER_CPU(int, callfuncsingle_irq);
41 static DEFINE_PER_CPU(int, debug_irq) = -1;
42
43 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
44 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
45
46 /*
47  * Reschedule call back. Nothing to do,
48  * all the work is done automatically when
49  * we return from the interrupt.
50  */
51 static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
52 {
53 #ifdef CONFIG_X86_32
54         __get_cpu_var(irq_stat).irq_resched_count++;
55 #else
56         add_pda(irq_resched_count, 1);
57 #endif
58
59         return IRQ_HANDLED;
60 }
61
62 static __cpuinit void cpu_bringup(void)
63 {
64         int cpu = smp_processor_id();
65
66         cpu_init();
67         touch_softlockup_watchdog();
68         preempt_disable();
69
70         xen_enable_sysenter();
71         xen_enable_syscall();
72
73         cpu = smp_processor_id();
74         smp_store_cpu_info(cpu);
75         cpu_data(cpu).x86_max_cores = 1;
76         set_cpu_sibling_map(cpu);
77
78         xen_setup_cpu_clockevents();
79
80         cpu_set(cpu, cpu_online_map);
81         x86_write_percpu(cpu_state, CPU_ONLINE);
82         wmb();
83
84         /* We can take interrupts now: we're officially "up". */
85         local_irq_enable();
86
87         wmb();                  /* make sure everything is out */
88 }
89
90 static __cpuinit void cpu_bringup_and_idle(void)
91 {
92         cpu_bringup();
93         cpu_idle();
94 }
95
96 static int xen_smp_intr_init(unsigned int cpu)
97 {
98         int rc;
99         const char *resched_name, *callfunc_name, *debug_name;
100
101         resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
102         rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
103                                     cpu,
104                                     xen_reschedule_interrupt,
105                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
106                                     resched_name,
107                                     NULL);
108         if (rc < 0)
109                 goto fail;
110         per_cpu(resched_irq, cpu) = rc;
111
112         callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu);
113         rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR,
114                                     cpu,
115                                     xen_call_function_interrupt,
116                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
117                                     callfunc_name,
118                                     NULL);
119         if (rc < 0)
120                 goto fail;
121         per_cpu(callfunc_irq, cpu) = rc;
122
123         debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu);
124         rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt,
125                                      IRQF_DISABLED | IRQF_PERCPU | IRQF_NOBALANCING,
126                                      debug_name, NULL);
127         if (rc < 0)
128                 goto fail;
129         per_cpu(debug_irq, cpu) = rc;
130
131         callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu);
132         rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR,
133                                     cpu,
134                                     xen_call_function_single_interrupt,
135                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
136                                     callfunc_name,
137                                     NULL);
138         if (rc < 0)
139                 goto fail;
140         per_cpu(callfuncsingle_irq, cpu) = rc;
141
142         return 0;
143
144  fail:
145         if (per_cpu(resched_irq, cpu) >= 0)
146                 unbind_from_irqhandler(per_cpu(resched_irq, cpu), NULL);
147         if (per_cpu(callfunc_irq, cpu) >= 0)
148                 unbind_from_irqhandler(per_cpu(callfunc_irq, cpu), NULL);
149         if (per_cpu(debug_irq, cpu) >= 0)
150                 unbind_from_irqhandler(per_cpu(debug_irq, cpu), NULL);
151         if (per_cpu(callfuncsingle_irq, cpu) >= 0)
152                 unbind_from_irqhandler(per_cpu(callfuncsingle_irq, cpu), NULL);
153
154         return rc;
155 }
156
157 static void __init xen_fill_possible_map(void)
158 {
159         int i, rc;
160
161         for (i = 0; i < NR_CPUS; i++) {
162                 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
163                 if (rc >= 0) {
164                         num_processors++;
165                         cpu_set(i, cpu_possible_map);
166                 }
167         }
168 }
169
170 static void __init xen_smp_prepare_boot_cpu(void)
171 {
172         BUG_ON(smp_processor_id() != 0);
173         native_smp_prepare_boot_cpu();
174
175         /* We've switched to the "real" per-cpu gdt, so make sure the
176            old memory can be recycled */
177         make_lowmem_page_readwrite(&per_cpu_var(gdt_page));
178
179         xen_setup_vcpu_info_placement();
180 }
181
182 static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
183 {
184         unsigned cpu;
185
186         xen_init_lock_cpu(0);
187
188         smp_store_cpu_info(0);
189         cpu_data(0).x86_max_cores = 1;
190         set_cpu_sibling_map(0);
191
192         if (xen_smp_intr_init(0))
193                 BUG();
194
195         xen_cpu_initialized_map = cpumask_of_cpu(0);
196
197         /* Restrict the possible_map according to max_cpus. */
198         while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
199                 for (cpu = NR_CPUS - 1; !cpu_possible(cpu); cpu--)
200                         continue;
201                 cpu_clear(cpu, cpu_possible_map);
202         }
203
204         for_each_possible_cpu (cpu) {
205                 struct task_struct *idle;
206
207                 if (cpu == 0)
208                         continue;
209
210                 idle = fork_idle(cpu);
211                 if (IS_ERR(idle))
212                         panic("failed fork for CPU %d", cpu);
213
214                 cpu_set(cpu, cpu_present_map);
215         }
216 }
217
218 static __cpuinit int
219 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
220 {
221         struct vcpu_guest_context *ctxt;
222         struct desc_struct *gdt;
223
224         if (cpu_test_and_set(cpu, xen_cpu_initialized_map))
225                 return 0;
226
227         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
228         if (ctxt == NULL)
229                 return -ENOMEM;
230
231         gdt = get_cpu_gdt_table(cpu);
232
233         ctxt->flags = VGCF_IN_KERNEL;
234         ctxt->user_regs.ds = __USER_DS;
235         ctxt->user_regs.es = __USER_DS;
236         ctxt->user_regs.ss = __KERNEL_DS;
237 #ifdef CONFIG_X86_32
238         ctxt->user_regs.fs = __KERNEL_PERCPU;
239 #endif
240         ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
241         ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
242
243         memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
244
245         xen_copy_trap_info(ctxt->trap_ctxt);
246
247         ctxt->ldt_ents = 0;
248
249         BUG_ON((unsigned long)gdt & ~PAGE_MASK);
250         make_lowmem_page_readonly(gdt);
251
252         ctxt->gdt_frames[0] = virt_to_mfn(gdt);
253         ctxt->gdt_ents      = GDT_ENTRIES;
254
255         ctxt->user_regs.cs = __KERNEL_CS;
256         ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
257
258         ctxt->kernel_ss = __KERNEL_DS;
259         ctxt->kernel_sp = idle->thread.sp0;
260
261 #ifdef CONFIG_X86_32
262         ctxt->event_callback_cs     = __KERNEL_CS;
263         ctxt->failsafe_callback_cs  = __KERNEL_CS;
264 #endif
265         ctxt->event_callback_eip    = (unsigned long)xen_hypervisor_callback;
266         ctxt->failsafe_callback_eip = (unsigned long)xen_failsafe_callback;
267
268         per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
269         ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
270
271         if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
272                 BUG();
273
274         kfree(ctxt);
275         return 0;
276 }
277
278 static int __cpuinit xen_cpu_up(unsigned int cpu)
279 {
280         struct task_struct *idle = idle_task(cpu);
281         int rc;
282
283 #ifdef CONFIG_X86_64
284         /* Allocate node local memory for AP pdas */
285         WARN_ON(cpu == 0);
286         if (cpu > 0) {
287                 rc = get_local_pda(cpu);
288                 if (rc)
289                         return rc;
290         }
291 #endif
292
293 #ifdef CONFIG_X86_32
294         init_gdt(cpu);
295         per_cpu(current_task, cpu) = idle;
296         irq_ctx_init(cpu);
297 #else
298         cpu_pda(cpu)->pcurrent = idle;
299         clear_tsk_thread_flag(idle, TIF_FORK);
300 #endif
301         xen_setup_timer(cpu);
302         xen_init_lock_cpu(cpu);
303
304         per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
305
306         /* make sure interrupts start blocked */
307         per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
308
309         rc = cpu_initialize_context(cpu, idle);
310         if (rc)
311                 return rc;
312
313         if (num_online_cpus() == 1)
314                 alternatives_smp_switch(1);
315
316         rc = xen_smp_intr_init(cpu);
317         if (rc)
318                 return rc;
319
320         rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
321         BUG_ON(rc);
322
323         while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
324                 HYPERVISOR_sched_op(SCHEDOP_yield, 0);
325                 barrier();
326         }
327
328         return 0;
329 }
330
331 static void xen_smp_cpus_done(unsigned int max_cpus)
332 {
333 }
334
335 #ifdef CONFIG_HOTPLUG_CPU
336 static int xen_cpu_disable(void)
337 {
338         unsigned int cpu = smp_processor_id();
339         if (cpu == 0)
340                 return -EBUSY;
341
342         cpu_disable_common();
343
344         load_cr3(swapper_pg_dir);
345         return 0;
346 }
347
348 static void xen_cpu_die(unsigned int cpu)
349 {
350         while (HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
351                 current->state = TASK_UNINTERRUPTIBLE;
352                 schedule_timeout(HZ/10);
353         }
354         unbind_from_irqhandler(per_cpu(resched_irq, cpu), NULL);
355         unbind_from_irqhandler(per_cpu(callfunc_irq, cpu), NULL);
356         unbind_from_irqhandler(per_cpu(debug_irq, cpu), NULL);
357         unbind_from_irqhandler(per_cpu(callfuncsingle_irq, cpu), NULL);
358         xen_uninit_lock_cpu(cpu);
359         xen_teardown_timer(cpu);
360
361         if (num_online_cpus() == 1)
362                 alternatives_smp_switch(0);
363 }
364
365 static void xen_play_dead(void)
366 {
367         play_dead_common();
368         HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
369         cpu_bringup();
370 }
371
372 #else /* !CONFIG_HOTPLUG_CPU */
373 static int xen_cpu_disable(void)
374 {
375         return -ENOSYS;
376 }
377
378 static void xen_cpu_die(unsigned int cpu)
379 {
380         BUG();
381 }
382
383 static void xen_play_dead(void)
384 {
385         BUG();
386 }
387
388 #endif
389 static void stop_self(void *v)
390 {
391         int cpu = smp_processor_id();
392
393         /* make sure we're not pinning something down */
394         load_cr3(swapper_pg_dir);
395         /* should set up a minimal gdt */
396
397         HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL);
398         BUG();
399 }
400
401 static void xen_smp_send_stop(void)
402 {
403         smp_call_function(stop_self, NULL, 0);
404 }
405
406 static void xen_smp_send_reschedule(int cpu)
407 {
408         xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
409 }
410
411 static void xen_send_IPI_mask(cpumask_t mask, enum ipi_vector vector)
412 {
413         unsigned cpu;
414
415         cpus_and(mask, mask, cpu_online_map);
416
417         for_each_cpu_mask_nr(cpu, mask)
418                 xen_send_IPI_one(cpu, vector);
419 }
420
421 static void xen_smp_send_call_function_ipi(cpumask_t mask)
422 {
423         int cpu;
424
425         xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
426
427         /* Make sure other vcpus get a chance to run if they need to. */
428         for_each_cpu_mask_nr(cpu, mask) {
429                 if (xen_vcpu_stolen(cpu)) {
430                         HYPERVISOR_sched_op(SCHEDOP_yield, 0);
431                         break;
432                 }
433         }
434 }
435
436 static void xen_smp_send_call_function_single_ipi(int cpu)
437 {
438         xen_send_IPI_mask(cpumask_of_cpu(cpu), XEN_CALL_FUNCTION_SINGLE_VECTOR);
439 }
440
441 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
442 {
443         irq_enter();
444         generic_smp_call_function_interrupt();
445 #ifdef CONFIG_X86_32
446         __get_cpu_var(irq_stat).irq_call_count++;
447 #else
448         add_pda(irq_call_count, 1);
449 #endif
450         irq_exit();
451
452         return IRQ_HANDLED;
453 }
454
455 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
456 {
457         irq_enter();
458         generic_smp_call_function_single_interrupt();
459 #ifdef CONFIG_X86_32
460         __get_cpu_var(irq_stat).irq_call_count++;
461 #else
462         add_pda(irq_call_count, 1);
463 #endif
464         irq_exit();
465
466         return IRQ_HANDLED;
467 }
468
469 static const struct smp_ops xen_smp_ops __initdata = {
470         .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
471         .smp_prepare_cpus = xen_smp_prepare_cpus,
472         .smp_cpus_done = xen_smp_cpus_done,
473
474         .cpu_up = xen_cpu_up,
475         .cpu_die = xen_cpu_die,
476         .cpu_disable = xen_cpu_disable,
477         .play_dead = xen_play_dead,
478
479         .smp_send_stop = xen_smp_send_stop,
480         .smp_send_reschedule = xen_smp_send_reschedule,
481
482         .send_call_func_ipi = xen_smp_send_call_function_ipi,
483         .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
484 };
485
486 void __init xen_smp_init(void)
487 {
488         smp_ops = xen_smp_ops;
489         xen_fill_possible_map();
490         xen_init_spinlocks();
491 }