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>
19 #include <linux/irq_work.h>
21 #include <asm/paravirt.h>
23 #include <asm/pgtable.h>
26 #include <xen/interface/xen.h>
27 #include <xen/interface/vcpu.h>
29 #include <asm/xen/interface.h>
30 #include <asm/xen/hypercall.h>
34 #include <xen/events.h>
36 #include <xen/hvc-console.h>
40 cpumask_var_t xen_cpu_initialized_map
;
42 static DEFINE_PER_CPU(int, xen_resched_irq
);
43 static DEFINE_PER_CPU(int, xen_callfunc_irq
);
44 static DEFINE_PER_CPU(int, xen_callfuncsingle_irq
);
45 static DEFINE_PER_CPU(int, xen_irq_work
);
46 static DEFINE_PER_CPU(int, xen_debug_irq
) = -1;
48 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
);
49 static irqreturn_t
xen_call_function_single_interrupt(int irq
, void *dev_id
);
50 static irqreturn_t
xen_irq_work_interrupt(int irq
, void *dev_id
);
53 * Reschedule call back.
55 static irqreturn_t
xen_reschedule_interrupt(int irq
, void *dev_id
)
57 inc_irq_stat(irq_resched_count
);
63 static void __cpuinit
cpu_bringup(void)
68 touch_softlockup_watchdog();
71 xen_enable_sysenter();
74 cpu
= smp_processor_id();
75 smp_store_cpu_info(cpu
);
76 cpu_data(cpu
).x86_max_cores
= 1;
77 set_cpu_sibling_map(cpu
);
79 xen_setup_cpu_clockevents();
81 notify_cpu_starting(cpu
);
84 set_cpu_online(cpu
, true);
87 this_cpu_write(cpu_state
, CPU_ONLINE
);
91 /* We can take interrupts now: we're officially "up". */
94 wmb(); /* make sure everything is out */
97 static void __cpuinit
cpu_bringup_and_idle(void)
103 static int xen_smp_intr_init(unsigned int cpu
)
106 const char *resched_name
, *callfunc_name
, *debug_name
;
108 resched_name
= kasprintf(GFP_KERNEL
, "resched%d", cpu
);
109 rc
= bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR
,
111 xen_reschedule_interrupt
,
112 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
117 per_cpu(xen_resched_irq
, cpu
) = rc
;
119 callfunc_name
= kasprintf(GFP_KERNEL
, "callfunc%d", cpu
);
120 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR
,
122 xen_call_function_interrupt
,
123 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
128 per_cpu(xen_callfunc_irq
, cpu
) = rc
;
130 debug_name
= kasprintf(GFP_KERNEL
, "debug%d", cpu
);
131 rc
= bind_virq_to_irqhandler(VIRQ_DEBUG
, cpu
, xen_debug_interrupt
,
132 IRQF_DISABLED
| IRQF_PERCPU
| IRQF_NOBALANCING
,
136 per_cpu(xen_debug_irq
, cpu
) = rc
;
138 callfunc_name
= kasprintf(GFP_KERNEL
, "callfuncsingle%d", cpu
);
139 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR
,
141 xen_call_function_single_interrupt
,
142 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
147 per_cpu(xen_callfuncsingle_irq
, cpu
) = rc
;
149 callfunc_name
= kasprintf(GFP_KERNEL
, "irqwork%d", cpu
);
150 rc
= bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR
,
152 xen_irq_work_interrupt
,
153 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
158 per_cpu(xen_irq_work
, cpu
) = rc
;
163 if (per_cpu(xen_resched_irq
, cpu
) >= 0)
164 unbind_from_irqhandler(per_cpu(xen_resched_irq
, cpu
), NULL
);
165 if (per_cpu(xen_callfunc_irq
, cpu
) >= 0)
166 unbind_from_irqhandler(per_cpu(xen_callfunc_irq
, cpu
), NULL
);
167 if (per_cpu(xen_debug_irq
, cpu
) >= 0)
168 unbind_from_irqhandler(per_cpu(xen_debug_irq
, cpu
), NULL
);
169 if (per_cpu(xen_callfuncsingle_irq
, cpu
) >= 0)
170 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq
, cpu
),
172 if (per_cpu(xen_irq_work
, cpu
) >= 0)
173 unbind_from_irqhandler(per_cpu(xen_irq_work
, cpu
), NULL
);
178 static void __init
xen_fill_possible_map(void)
182 if (xen_initial_domain())
185 for (i
= 0; i
< nr_cpu_ids
; i
++) {
186 rc
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, i
, NULL
);
189 set_cpu_possible(i
, true);
194 static void __init
xen_filter_cpu_maps(void)
197 unsigned int subtract
= 0;
199 if (!xen_initial_domain())
204 for (i
= 0; i
< nr_cpu_ids
; i
++) {
205 rc
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, i
, NULL
);
208 set_cpu_possible(i
, true);
210 set_cpu_possible(i
, false);
211 set_cpu_present(i
, false);
215 #ifdef CONFIG_HOTPLUG_CPU
216 /* This is akin to using 'nr_cpus' on the Linux command line.
217 * Which is OK as when we use 'dom0_max_vcpus=X' we can only
218 * have up to X, while nr_cpu_ids is greater than X. This
219 * normally is not a problem, except when CPU hotplugging
220 * is involved and then there might be more than X CPUs
221 * in the guest - which will not work as there is no
222 * hypercall to expand the max number of VCPUs an already
223 * running guest has. So cap it up to X. */
225 nr_cpu_ids
= nr_cpu_ids
- subtract
;
230 static void __init
xen_smp_prepare_boot_cpu(void)
232 BUG_ON(smp_processor_id() != 0);
233 native_smp_prepare_boot_cpu();
235 /* We've switched to the "real" per-cpu gdt, so make sure the
236 old memory can be recycled */
237 make_lowmem_page_readwrite(xen_initial_gdt
);
239 xen_filter_cpu_maps();
240 xen_setup_vcpu_info_placement();
243 static void __init
xen_smp_prepare_cpus(unsigned int max_cpus
)
248 if (skip_ioapic_setup
) {
249 char *m
= (max_cpus
== 0) ?
250 "The nosmp parameter is incompatible with Xen; " \
251 "use Xen dom0_max_vcpus=1 parameter" :
252 "The noapic parameter is incompatible with Xen";
257 xen_init_lock_cpu(0);
259 smp_store_cpu_info(0);
260 cpu_data(0).x86_max_cores
= 1;
262 for_each_possible_cpu(i
) {
263 zalloc_cpumask_var(&per_cpu(cpu_sibling_map
, i
), GFP_KERNEL
);
264 zalloc_cpumask_var(&per_cpu(cpu_core_map
, i
), GFP_KERNEL
);
265 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map
, i
), GFP_KERNEL
);
267 set_cpu_sibling_map(0);
269 if (xen_smp_intr_init(0))
272 if (!alloc_cpumask_var(&xen_cpu_initialized_map
, GFP_KERNEL
))
273 panic("could not allocate xen_cpu_initialized_map\n");
275 cpumask_copy(xen_cpu_initialized_map
, cpumask_of(0));
277 /* Restrict the possible_map according to max_cpus. */
278 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus
)) {
279 for (cpu
= nr_cpu_ids
- 1; !cpu_possible(cpu
); cpu
--)
281 set_cpu_possible(cpu
, false);
284 for_each_possible_cpu(cpu
)
285 set_cpu_present(cpu
, true);
289 cpu_initialize_context(unsigned int cpu
, struct task_struct
*idle
)
291 struct vcpu_guest_context
*ctxt
;
292 struct desc_struct
*gdt
;
293 unsigned long gdt_mfn
;
295 if (cpumask_test_and_set_cpu(cpu
, xen_cpu_initialized_map
))
298 ctxt
= kzalloc(sizeof(*ctxt
), GFP_KERNEL
);
302 gdt
= get_cpu_gdt_table(cpu
);
304 ctxt
->flags
= VGCF_IN_KERNEL
;
305 ctxt
->user_regs
.ds
= __USER_DS
;
306 ctxt
->user_regs
.es
= __USER_DS
;
307 ctxt
->user_regs
.ss
= __KERNEL_DS
;
309 ctxt
->user_regs
.fs
= __KERNEL_PERCPU
;
310 ctxt
->user_regs
.gs
= __KERNEL_STACK_CANARY
;
312 ctxt
->gs_base_kernel
= per_cpu_offset(cpu
);
314 ctxt
->user_regs
.eip
= (unsigned long)cpu_bringup_and_idle
;
315 ctxt
->user_regs
.eflags
= 0x1000; /* IOPL_RING1 */
317 memset(&ctxt
->fpu_ctxt
, 0, sizeof(ctxt
->fpu_ctxt
));
319 xen_copy_trap_info(ctxt
->trap_ctxt
);
323 BUG_ON((unsigned long)gdt
& ~PAGE_MASK
);
325 gdt_mfn
= arbitrary_virt_to_mfn(gdt
);
326 make_lowmem_page_readonly(gdt
);
327 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn
));
329 ctxt
->gdt_frames
[0] = gdt_mfn
;
330 ctxt
->gdt_ents
= GDT_ENTRIES
;
332 ctxt
->user_regs
.cs
= __KERNEL_CS
;
333 ctxt
->user_regs
.esp
= idle
->thread
.sp0
- sizeof(struct pt_regs
);
335 ctxt
->kernel_ss
= __KERNEL_DS
;
336 ctxt
->kernel_sp
= idle
->thread
.sp0
;
339 ctxt
->event_callback_cs
= __KERNEL_CS
;
340 ctxt
->failsafe_callback_cs
= __KERNEL_CS
;
342 ctxt
->event_callback_eip
= (unsigned long)xen_hypervisor_callback
;
343 ctxt
->failsafe_callback_eip
= (unsigned long)xen_failsafe_callback
;
345 per_cpu(xen_cr3
, cpu
) = __pa(swapper_pg_dir
);
346 ctxt
->ctrlreg
[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir
));
348 if (HYPERVISOR_vcpu_op(VCPUOP_initialise
, cpu
, ctxt
))
355 static int __cpuinit
xen_cpu_up(unsigned int cpu
, struct task_struct
*idle
)
359 per_cpu(current_task
, cpu
) = idle
;
363 clear_tsk_thread_flag(idle
, TIF_FORK
);
364 per_cpu(kernel_stack
, cpu
) =
365 (unsigned long)task_stack_page(idle
) -
366 KERNEL_STACK_OFFSET
+ THREAD_SIZE
;
368 xen_setup_runstate_info(cpu
);
369 xen_setup_timer(cpu
);
370 xen_init_lock_cpu(cpu
);
372 per_cpu(cpu_state
, cpu
) = CPU_UP_PREPARE
;
374 /* make sure interrupts start blocked */
375 per_cpu(xen_vcpu
, cpu
)->evtchn_upcall_mask
= 1;
377 rc
= cpu_initialize_context(cpu
, idle
);
381 if (num_online_cpus() == 1)
382 alternatives_smp_switch(1);
384 rc
= xen_smp_intr_init(cpu
);
388 rc
= HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
);
391 while(per_cpu(cpu_state
, cpu
) != CPU_ONLINE
) {
392 HYPERVISOR_sched_op(SCHEDOP_yield
, NULL
);
399 static void xen_smp_cpus_done(unsigned int max_cpus
)
403 #ifdef CONFIG_HOTPLUG_CPU
404 static int xen_cpu_disable(void)
406 unsigned int cpu
= smp_processor_id();
410 cpu_disable_common();
412 load_cr3(swapper_pg_dir
);
416 static void xen_cpu_die(unsigned int cpu
)
418 while (HYPERVISOR_vcpu_op(VCPUOP_is_up
, cpu
, NULL
)) {
419 current
->state
= TASK_UNINTERRUPTIBLE
;
420 schedule_timeout(HZ
/10);
422 unbind_from_irqhandler(per_cpu(xen_resched_irq
, cpu
), NULL
);
423 unbind_from_irqhandler(per_cpu(xen_callfunc_irq
, cpu
), NULL
);
424 unbind_from_irqhandler(per_cpu(xen_debug_irq
, cpu
), NULL
);
425 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq
, cpu
), NULL
);
426 unbind_from_irqhandler(per_cpu(xen_irq_work
, cpu
), NULL
);
427 xen_uninit_lock_cpu(cpu
);
428 xen_teardown_timer(cpu
);
430 if (num_online_cpus() == 1)
431 alternatives_smp_switch(0);
434 static void __cpuinit
xen_play_dead(void) /* used only with HOTPLUG_CPU */
437 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
440 * Balance out the preempt calls - as we are running in cpu_idle
441 * loop which has been called at bootup from cpu_bringup_and_idle.
442 * The cpucpu_bringup_and_idle called cpu_bringup which made a
443 * preempt_disable() So this preempt_enable will balance it out.
448 #else /* !CONFIG_HOTPLUG_CPU */
449 static int xen_cpu_disable(void)
454 static void xen_cpu_die(unsigned int cpu
)
459 static void xen_play_dead(void)
465 static void stop_self(void *v
)
467 int cpu
= smp_processor_id();
469 /* make sure we're not pinning something down */
470 load_cr3(swapper_pg_dir
);
471 /* should set up a minimal gdt */
473 set_cpu_online(cpu
, false);
475 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
);
479 static void xen_stop_other_cpus(int wait
)
481 smp_call_function(stop_self
, NULL
, wait
);
484 static void xen_smp_send_reschedule(int cpu
)
486 xen_send_IPI_one(cpu
, XEN_RESCHEDULE_VECTOR
);
489 static void __xen_send_IPI_mask(const struct cpumask
*mask
,
494 for_each_cpu_and(cpu
, mask
, cpu_online_mask
)
495 xen_send_IPI_one(cpu
, vector
);
498 static void xen_smp_send_call_function_ipi(const struct cpumask
*mask
)
502 __xen_send_IPI_mask(mask
, XEN_CALL_FUNCTION_VECTOR
);
504 /* Make sure other vcpus get a chance to run if they need to. */
505 for_each_cpu(cpu
, mask
) {
506 if (xen_vcpu_stolen(cpu
)) {
507 HYPERVISOR_sched_op(SCHEDOP_yield
, NULL
);
513 static void xen_smp_send_call_function_single_ipi(int cpu
)
515 __xen_send_IPI_mask(cpumask_of(cpu
),
516 XEN_CALL_FUNCTION_SINGLE_VECTOR
);
519 static inline int xen_map_vector(int vector
)
524 case RESCHEDULE_VECTOR
:
525 xen_vector
= XEN_RESCHEDULE_VECTOR
;
527 case CALL_FUNCTION_VECTOR
:
528 xen_vector
= XEN_CALL_FUNCTION_VECTOR
;
530 case CALL_FUNCTION_SINGLE_VECTOR
:
531 xen_vector
= XEN_CALL_FUNCTION_SINGLE_VECTOR
;
533 case IRQ_WORK_VECTOR
:
534 xen_vector
= XEN_IRQ_WORK_VECTOR
;
538 printk(KERN_ERR
"xen: vector 0x%x is not implemented\n",
545 void xen_send_IPI_mask(const struct cpumask
*mask
,
548 int xen_vector
= xen_map_vector(vector
);
551 __xen_send_IPI_mask(mask
, xen_vector
);
554 void xen_send_IPI_all(int vector
)
556 int xen_vector
= xen_map_vector(vector
);
559 __xen_send_IPI_mask(cpu_online_mask
, xen_vector
);
562 void xen_send_IPI_self(int vector
)
564 int xen_vector
= xen_map_vector(vector
);
567 xen_send_IPI_one(smp_processor_id(), xen_vector
);
570 void xen_send_IPI_mask_allbutself(const struct cpumask
*mask
,
574 unsigned int this_cpu
= smp_processor_id();
576 if (!(num_online_cpus() > 1))
579 for_each_cpu_and(cpu
, mask
, cpu_online_mask
) {
583 xen_smp_send_call_function_single_ipi(cpu
);
587 void xen_send_IPI_allbutself(int vector
)
589 int xen_vector
= xen_map_vector(vector
);
592 xen_send_IPI_mask_allbutself(cpu_online_mask
, xen_vector
);
595 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
)
598 generic_smp_call_function_interrupt();
599 inc_irq_stat(irq_call_count
);
605 static irqreturn_t
xen_call_function_single_interrupt(int irq
, void *dev_id
)
608 generic_smp_call_function_single_interrupt();
609 inc_irq_stat(irq_call_count
);
615 static irqreturn_t
xen_irq_work_interrupt(int irq
, void *dev_id
)
619 inc_irq_stat(apic_irq_work_irqs
);
625 static const struct smp_ops xen_smp_ops __initconst
= {
626 .smp_prepare_boot_cpu
= xen_smp_prepare_boot_cpu
,
627 .smp_prepare_cpus
= xen_smp_prepare_cpus
,
628 .smp_cpus_done
= xen_smp_cpus_done
,
630 .cpu_up
= xen_cpu_up
,
631 .cpu_die
= xen_cpu_die
,
632 .cpu_disable
= xen_cpu_disable
,
633 .play_dead
= xen_play_dead
,
635 .stop_other_cpus
= xen_stop_other_cpus
,
636 .smp_send_reschedule
= xen_smp_send_reschedule
,
638 .send_call_func_ipi
= xen_smp_send_call_function_ipi
,
639 .send_call_func_single_ipi
= xen_smp_send_call_function_single_ipi
,
642 void __init
xen_smp_init(void)
644 smp_ops
= xen_smp_ops
;
645 xen_fill_possible_map();
646 xen_init_spinlocks();
649 static void __init
xen_hvm_smp_prepare_cpus(unsigned int max_cpus
)
651 native_smp_prepare_cpus(max_cpus
);
652 WARN_ON(xen_smp_intr_init(0));
654 xen_init_lock_cpu(0);
657 static int __cpuinit
xen_hvm_cpu_up(unsigned int cpu
, struct task_struct
*tidle
)
660 rc
= native_cpu_up(cpu
, tidle
);
661 WARN_ON (xen_smp_intr_init(cpu
));
665 static void xen_hvm_cpu_die(unsigned int cpu
)
667 unbind_from_irqhandler(per_cpu(xen_resched_irq
, cpu
), NULL
);
668 unbind_from_irqhandler(per_cpu(xen_callfunc_irq
, cpu
), NULL
);
669 unbind_from_irqhandler(per_cpu(xen_debug_irq
, cpu
), NULL
);
670 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq
, cpu
), NULL
);
671 unbind_from_irqhandler(per_cpu(xen_irq_work
, cpu
), NULL
);
675 void __init
xen_hvm_smp_init(void)
677 if (!xen_have_vector_callback
)
679 smp_ops
.smp_prepare_cpus
= xen_hvm_smp_prepare_cpus
;
680 smp_ops
.smp_send_reschedule
= xen_smp_send_reschedule
;
681 smp_ops
.cpu_up
= xen_hvm_cpu_up
;
682 smp_ops
.cpu_die
= xen_hvm_cpu_die
;
683 smp_ops
.send_call_func_ipi
= xen_smp_send_call_function_ipi
;
684 smp_ops
.send_call_func_single_ipi
= xen_smp_send_call_function_single_ipi
;