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>
38 cpumask_var_t xen_cpu_initialized_map;
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;
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);
49 * Reschedule call back. Nothing to do,
50 * all the work is done automatically when
51 * we return from the interrupt.
53 static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
55 inc_irq_stat(irq_resched_count);
60 static __cpuinit void 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 __cpuinit void 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)
176 if (!xen_initial_domain())
181 for (i = 0; i < nr_cpu_ids; i++) {
182 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
185 set_cpu_possible(i, true);
187 set_cpu_possible(i, false);
188 set_cpu_present(i, false);
193 static void __init xen_smp_prepare_boot_cpu(void)
195 BUG_ON(smp_processor_id() != 0);
196 native_smp_prepare_boot_cpu();
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);
202 xen_filter_cpu_maps();
203 xen_setup_vcpu_info_placement();
206 static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
210 xen_init_lock_cpu(0);
212 smp_store_cpu_info(0);
213 cpu_data(0).x86_max_cores = 1;
214 set_cpu_sibling_map(0);
216 if (xen_smp_intr_init(0))
219 if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
220 panic("could not allocate xen_cpu_initialized_map\n");
222 cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
224 /* Restrict the possible_map according to max_cpus. */
225 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
226 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
228 set_cpu_possible(cpu, false);
231 for_each_possible_cpu (cpu) {
232 struct task_struct *idle;
237 idle = fork_idle(cpu);
239 panic("failed fork for CPU %d", cpu);
241 set_cpu_present(cpu, true);
246 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
248 struct vcpu_guest_context *ctxt;
249 struct desc_struct *gdt;
250 unsigned long gdt_mfn;
252 if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
255 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
259 gdt = get_cpu_gdt_table(cpu);
261 ctxt->flags = VGCF_IN_KERNEL;
262 ctxt->user_regs.ds = __USER_DS;
263 ctxt->user_regs.es = __USER_DS;
264 ctxt->user_regs.ss = __KERNEL_DS;
266 ctxt->user_regs.fs = __KERNEL_PERCPU;
267 ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
269 ctxt->gs_base_kernel = per_cpu_offset(cpu);
271 ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
272 ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
274 memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
276 xen_copy_trap_info(ctxt->trap_ctxt);
280 BUG_ON((unsigned long)gdt & ~PAGE_MASK);
282 gdt_mfn = arbitrary_virt_to_mfn(gdt);
283 make_lowmem_page_readonly(gdt);
284 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
286 ctxt->gdt_frames[0] = gdt_mfn;
287 ctxt->gdt_ents = GDT_ENTRIES;
289 ctxt->user_regs.cs = __KERNEL_CS;
290 ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
292 ctxt->kernel_ss = __KERNEL_DS;
293 ctxt->kernel_sp = idle->thread.sp0;
296 ctxt->event_callback_cs = __KERNEL_CS;
297 ctxt->failsafe_callback_cs = __KERNEL_CS;
299 ctxt->event_callback_eip = (unsigned long)xen_hypervisor_callback;
300 ctxt->failsafe_callback_eip = (unsigned long)xen_failsafe_callback;
302 per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
303 ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
305 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
312 static int __cpuinit xen_cpu_up(unsigned int cpu)
314 struct task_struct *idle = idle_task(cpu);
317 per_cpu(current_task, cpu) = idle;
321 clear_tsk_thread_flag(idle, TIF_FORK);
322 per_cpu(kernel_stack, cpu) =
323 (unsigned long)task_stack_page(idle) -
324 KERNEL_STACK_OFFSET + THREAD_SIZE;
326 xen_setup_runstate_info(cpu);
327 xen_setup_timer(cpu);
328 xen_init_lock_cpu(cpu);
330 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
332 /* make sure interrupts start blocked */
333 per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
335 rc = cpu_initialize_context(cpu, idle);
339 if (num_online_cpus() == 1)
340 alternatives_smp_switch(1);
342 rc = xen_smp_intr_init(cpu);
346 rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
349 while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
350 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
357 static void xen_smp_cpus_done(unsigned int max_cpus)
361 #ifdef CONFIG_HOTPLUG_CPU
362 static int xen_cpu_disable(void)
364 unsigned int cpu = smp_processor_id();
368 cpu_disable_common();
370 load_cr3(swapper_pg_dir);
374 static void xen_cpu_die(unsigned int cpu)
376 while (HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
377 current->state = TASK_UNINTERRUPTIBLE;
378 schedule_timeout(HZ/10);
380 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
381 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
382 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
383 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
384 xen_uninit_lock_cpu(cpu);
385 xen_teardown_timer(cpu);
387 if (num_online_cpus() == 1)
388 alternatives_smp_switch(0);
391 static void __cpuinit xen_play_dead(void) /* used only with HOTPLUG_CPU */
394 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
398 #else /* !CONFIG_HOTPLUG_CPU */
399 static int xen_cpu_disable(void)
404 static void xen_cpu_die(unsigned int cpu)
409 static void xen_play_dead(void)
415 static void stop_self(void *v)
417 int cpu = smp_processor_id();
419 /* make sure we're not pinning something down */
420 load_cr3(swapper_pg_dir);
421 /* should set up a minimal gdt */
423 set_cpu_online(cpu, false);
425 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL);
429 static void xen_smp_send_stop(void)
431 smp_call_function(stop_self, NULL, 0);
434 static void xen_smp_send_reschedule(int cpu)
436 xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
439 static void xen_send_IPI_mask(const struct cpumask *mask,
440 enum ipi_vector vector)
444 for_each_cpu_and(cpu, mask, cpu_online_mask)
445 xen_send_IPI_one(cpu, vector);
448 static void xen_smp_send_call_function_ipi(const struct cpumask *mask)
452 xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
454 /* Make sure other vcpus get a chance to run if they need to. */
455 for_each_cpu(cpu, mask) {
456 if (xen_vcpu_stolen(cpu)) {
457 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
463 static void xen_smp_send_call_function_single_ipi(int cpu)
465 xen_send_IPI_mask(cpumask_of(cpu),
466 XEN_CALL_FUNCTION_SINGLE_VECTOR);
469 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
472 generic_smp_call_function_interrupt();
473 inc_irq_stat(irq_call_count);
479 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
482 generic_smp_call_function_single_interrupt();
483 inc_irq_stat(irq_call_count);
489 static const struct smp_ops xen_smp_ops __initdata = {
490 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
491 .smp_prepare_cpus = xen_smp_prepare_cpus,
492 .smp_cpus_done = xen_smp_cpus_done,
494 .cpu_up = xen_cpu_up,
495 .cpu_die = xen_cpu_die,
496 .cpu_disable = xen_cpu_disable,
497 .play_dead = xen_play_dead,
499 .smp_send_stop = xen_smp_send_stop,
500 .smp_send_reschedule = xen_smp_send_reschedule,
502 .send_call_func_ipi = xen_smp_send_call_function_ipi,
503 .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
506 void __init xen_smp_init(void)
508 smp_ops = xen_smp_ops;
509 xen_fill_possible_map();
510 xen_init_spinlocks();