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.
8 * IPIs are handled through the Xen event mechanism.
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
15 #include <linux/sched.h>
16 #include <linux/err.h>
17 #include <linux/slab.h>
18 #include <linux/smp.h>
20 #include <asm/paravirt.h>
22 #include <asm/pgtable.h>
25 #include <xen/interface/xen.h>
26 #include <xen/interface/vcpu.h>
28 #include <asm/xen/interface.h>
29 #include <asm/xen/hypercall.h>
33 #include <xen/events.h>
35 #include <xen/hvc-console.h>
39 cpumask_var_t xen_cpu_initialized_map;
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;
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);
50 * Reschedule call back.
52 static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
54 inc_irq_stat(irq_resched_count);
60 static void __cpuinit cpu_bringup(void)
62 int cpu = smp_processor_id();
65 touch_softlockup_watchdog();
68 xen_enable_sysenter();
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);
76 xen_setup_cpu_clockevents();
78 set_cpu_online(cpu, true);
79 percpu_write(cpu_state, CPU_ONLINE);
82 /* We can take interrupts now: we're officially "up". */
85 wmb(); /* make sure everything is out */
88 static void __cpuinit cpu_bringup_and_idle(void)
94 static int xen_smp_intr_init(unsigned int cpu)
97 const char *resched_name, *callfunc_name, *debug_name;
99 resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
100 rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
102 xen_reschedule_interrupt,
103 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
108 per_cpu(xen_resched_irq, cpu) = rc;
110 callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu);
111 rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR,
113 xen_call_function_interrupt,
114 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
119 per_cpu(xen_callfunc_irq, cpu) = rc;
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,
127 per_cpu(xen_debug_irq, cpu) = rc;
129 callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu);
130 rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR,
132 xen_call_function_single_interrupt,
133 IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
138 per_cpu(xen_callfuncsingle_irq, cpu) = rc;
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),
156 static void __init xen_fill_possible_map(void)
160 if (xen_initial_domain())
163 for (i = 0; i < nr_cpu_ids; i++) {
164 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
167 set_cpu_possible(i, true);
172 static void __init xen_filter_cpu_maps(void)
175 unsigned int subtract = 0;
177 if (!xen_initial_domain())
182 for (i = 0; i < nr_cpu_ids; i++) {
183 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
186 set_cpu_possible(i, true);
188 set_cpu_possible(i, false);
189 set_cpu_present(i, false);
193 #ifdef CONFIG_HOTPLUG_CPU
194 /* This is akin to using 'nr_cpus' on the Linux command line.
195 * Which is OK as when we use 'dom0_max_vcpus=X' we can only
196 * have up to X, while nr_cpu_ids is greater than X. This
197 * normally is not a problem, except when CPU hotplugging
198 * is involved and then there might be more than X CPUs
199 * in the guest - which will not work as there is no
200 * hypercall to expand the max number of VCPUs an already
201 * running guest has. So cap it up to X. */
203 nr_cpu_ids = nr_cpu_ids - subtract;
208 static void __init xen_smp_prepare_boot_cpu(void)
210 BUG_ON(smp_processor_id() != 0);
211 native_smp_prepare_boot_cpu();
213 /* We've switched to the "real" per-cpu gdt, so make sure the
214 old memory can be recycled */
215 make_lowmem_page_readwrite(xen_initial_gdt);
217 xen_filter_cpu_maps();
218 xen_setup_vcpu_info_placement();
221 static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
226 if (skip_ioapic_setup) {
227 char *m = (max_cpus == 0) ?
228 "The nosmp parameter is incompatible with Xen; " \
229 "use Xen dom0_max_vcpus=1 parameter" :
230 "The noapic parameter is incompatible with Xen";
235 xen_init_lock_cpu(0);
237 smp_store_cpu_info(0);
238 cpu_data(0).x86_max_cores = 1;
240 for_each_possible_cpu(i) {
241 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
242 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
243 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
245 set_cpu_sibling_map(0);
247 if (xen_smp_intr_init(0))
250 if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
251 panic("could not allocate xen_cpu_initialized_map\n");
253 cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
255 /* Restrict the possible_map according to max_cpus. */
256 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
257 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
259 set_cpu_possible(cpu, false);
262 for_each_possible_cpu (cpu) {
263 struct task_struct *idle;
268 idle = fork_idle(cpu);
270 panic("failed fork for CPU %d", cpu);
272 set_cpu_present(cpu, true);
277 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
279 struct vcpu_guest_context *ctxt;
280 struct desc_struct *gdt;
281 unsigned long gdt_mfn;
283 if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
286 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
290 gdt = get_cpu_gdt_table(cpu);
292 ctxt->flags = VGCF_IN_KERNEL;
293 ctxt->user_regs.ds = __USER_DS;
294 ctxt->user_regs.es = __USER_DS;
295 ctxt->user_regs.ss = __KERNEL_DS;
297 ctxt->user_regs.fs = __KERNEL_PERCPU;
298 ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
300 ctxt->gs_base_kernel = per_cpu_offset(cpu);
302 ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
303 ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
305 memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
307 xen_copy_trap_info(ctxt->trap_ctxt);
311 BUG_ON((unsigned long)gdt & ~PAGE_MASK);
313 gdt_mfn = arbitrary_virt_to_mfn(gdt);
314 make_lowmem_page_readonly(gdt);
315 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
317 ctxt->gdt_frames[0] = gdt_mfn;
318 ctxt->gdt_ents = GDT_ENTRIES;
320 ctxt->user_regs.cs = __KERNEL_CS;
321 ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
323 ctxt->kernel_ss = __KERNEL_DS;
324 ctxt->kernel_sp = idle->thread.sp0;
327 ctxt->event_callback_cs = __KERNEL_CS;
328 ctxt->failsafe_callback_cs = __KERNEL_CS;
330 ctxt->event_callback_eip = (unsigned long)xen_hypervisor_callback;
331 ctxt->failsafe_callback_eip = (unsigned long)xen_failsafe_callback;
333 per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
334 ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
336 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
343 static int __cpuinit xen_cpu_up(unsigned int cpu)
345 struct task_struct *idle = idle_task(cpu);
348 per_cpu(current_task, cpu) = idle;
352 clear_tsk_thread_flag(idle, TIF_FORK);
353 per_cpu(kernel_stack, cpu) =
354 (unsigned long)task_stack_page(idle) -
355 KERNEL_STACK_OFFSET + THREAD_SIZE;
357 xen_setup_runstate_info(cpu);
358 xen_setup_timer(cpu);
359 xen_init_lock_cpu(cpu);
361 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
363 /* make sure interrupts start blocked */
364 per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
366 rc = cpu_initialize_context(cpu, idle);
370 if (num_online_cpus() == 1)
371 alternatives_smp_switch(1);
373 rc = xen_smp_intr_init(cpu);
377 rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
380 while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
381 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
388 static void xen_smp_cpus_done(unsigned int max_cpus)
392 #ifdef CONFIG_HOTPLUG_CPU
393 static int xen_cpu_disable(void)
395 unsigned int cpu = smp_processor_id();
399 cpu_disable_common();
401 load_cr3(swapper_pg_dir);
405 static void xen_cpu_die(unsigned int cpu)
407 while (HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
408 current->state = TASK_UNINTERRUPTIBLE;
409 schedule_timeout(HZ/10);
411 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
412 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
413 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
414 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
415 xen_uninit_lock_cpu(cpu);
416 xen_teardown_timer(cpu);
418 if (num_online_cpus() == 1)
419 alternatives_smp_switch(0);
422 static void __cpuinit xen_play_dead(void) /* used only with HOTPLUG_CPU */
425 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
429 #else /* !CONFIG_HOTPLUG_CPU */
430 static int xen_cpu_disable(void)
435 static void xen_cpu_die(unsigned int cpu)
440 static void xen_play_dead(void)
446 static void stop_self(void *v)
448 int cpu = smp_processor_id();
450 /* make sure we're not pinning something down */
451 load_cr3(swapper_pg_dir);
452 /* should set up a minimal gdt */
454 set_cpu_online(cpu, false);
456 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL);
460 static void xen_stop_other_cpus(int wait)
462 smp_call_function(stop_self, NULL, wait);
465 static void xen_smp_send_reschedule(int cpu)
467 xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
470 static void xen_send_IPI_mask(const struct cpumask *mask,
471 enum ipi_vector vector)
475 for_each_cpu_and(cpu, mask, cpu_online_mask)
476 xen_send_IPI_one(cpu, vector);
479 static void xen_smp_send_call_function_ipi(const struct cpumask *mask)
483 xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
485 /* Make sure other vcpus get a chance to run if they need to. */
486 for_each_cpu(cpu, mask) {
487 if (xen_vcpu_stolen(cpu)) {
488 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
494 static void xen_smp_send_call_function_single_ipi(int cpu)
496 xen_send_IPI_mask(cpumask_of(cpu),
497 XEN_CALL_FUNCTION_SINGLE_VECTOR);
500 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
503 generic_smp_call_function_interrupt();
504 inc_irq_stat(irq_call_count);
510 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
513 generic_smp_call_function_single_interrupt();
514 inc_irq_stat(irq_call_count);
520 static const struct smp_ops xen_smp_ops __initconst = {
521 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
522 .smp_prepare_cpus = xen_smp_prepare_cpus,
523 .smp_cpus_done = xen_smp_cpus_done,
525 .cpu_up = xen_cpu_up,
526 .cpu_die = xen_cpu_die,
527 .cpu_disable = xen_cpu_disable,
528 .play_dead = xen_play_dead,
530 .stop_other_cpus = xen_stop_other_cpus,
531 .smp_send_reschedule = xen_smp_send_reschedule,
533 .send_call_func_ipi = xen_smp_send_call_function_ipi,
534 .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
537 void __init xen_smp_init(void)
539 smp_ops = xen_smp_ops;
540 xen_fill_possible_map();
541 xen_init_spinlocks();
544 static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus)
546 native_smp_prepare_cpus(max_cpus);
547 WARN_ON(xen_smp_intr_init(0));
549 xen_init_lock_cpu(0);
552 static int __cpuinit xen_hvm_cpu_up(unsigned int cpu)
555 rc = native_cpu_up(cpu);
556 WARN_ON (xen_smp_intr_init(cpu));
560 static void xen_hvm_cpu_die(unsigned int cpu)
562 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
563 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
564 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
565 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
569 void __init xen_hvm_smp_init(void)
571 if (!xen_have_vector_callback)
573 smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
574 smp_ops.smp_send_reschedule = xen_smp_send_reschedule;
575 smp_ops.cpu_up = xen_hvm_cpu_up;
576 smp_ops.cpu_die = xen_hvm_cpu_die;
577 smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
578 smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;