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 struct xen_common_irq
{
46 static DEFINE_PER_CPU(struct xen_common_irq
, xen_resched_irq
) = { .irq
= -1 };
47 static DEFINE_PER_CPU(struct xen_common_irq
, xen_callfunc_irq
) = { .irq
= -1 };
48 static DEFINE_PER_CPU(struct xen_common_irq
, xen_callfuncsingle_irq
) = { .irq
= -1 };
49 static DEFINE_PER_CPU(struct xen_common_irq
, xen_irq_work
) = { .irq
= -1 };
50 static DEFINE_PER_CPU(struct xen_common_irq
, xen_debug_irq
) = { .irq
= -1 };
52 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
);
53 static irqreturn_t
xen_call_function_single_interrupt(int irq
, void *dev_id
);
54 static irqreturn_t
xen_irq_work_interrupt(int irq
, void *dev_id
);
57 * Reschedule call back.
59 static irqreturn_t
xen_reschedule_interrupt(int irq
, void *dev_id
)
61 inc_irq_stat(irq_resched_count
);
67 static void __cpuinit
cpu_bringup(void)
72 touch_softlockup_watchdog();
75 xen_enable_sysenter();
78 cpu
= smp_processor_id();
79 smp_store_cpu_info(cpu
);
80 cpu_data(cpu
).x86_max_cores
= 1;
81 set_cpu_sibling_map(cpu
);
83 xen_setup_cpu_clockevents();
85 notify_cpu_starting(cpu
);
87 set_cpu_online(cpu
, true);
89 this_cpu_write(cpu_state
, CPU_ONLINE
);
93 /* We can take interrupts now: we're officially "up". */
96 wmb(); /* make sure everything is out */
99 static void __cpuinit
cpu_bringup_and_idle(void)
102 cpu_startup_entry(CPUHP_ONLINE
);
105 static void xen_smp_intr_free(unsigned int cpu
)
107 if (per_cpu(xen_resched_irq
, cpu
).irq
>= 0) {
108 unbind_from_irqhandler(per_cpu(xen_resched_irq
, cpu
).irq
, NULL
);
109 per_cpu(xen_resched_irq
, cpu
).irq
= -1;
111 if (per_cpu(xen_callfunc_irq
, cpu
).irq
>= 0) {
112 unbind_from_irqhandler(per_cpu(xen_callfunc_irq
, cpu
).irq
, NULL
);
113 per_cpu(xen_callfunc_irq
, cpu
).irq
= -1;
115 if (per_cpu(xen_debug_irq
, cpu
).irq
>= 0) {
116 unbind_from_irqhandler(per_cpu(xen_debug_irq
, cpu
).irq
, NULL
);
117 per_cpu(xen_debug_irq
, cpu
).irq
= -1;
119 if (per_cpu(xen_callfuncsingle_irq
, cpu
).irq
>= 0) {
120 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq
, cpu
).irq
,
122 per_cpu(xen_callfuncsingle_irq
, cpu
).irq
= -1;
124 if (xen_hvm_domain())
127 if (per_cpu(xen_irq_work
, cpu
).irq
>= 0) {
128 unbind_from_irqhandler(per_cpu(xen_irq_work
, cpu
).irq
, NULL
);
129 per_cpu(xen_irq_work
, cpu
).irq
= -1;
132 static int xen_smp_intr_init(unsigned int cpu
)
135 const char *resched_name
, *callfunc_name
, *debug_name
;
137 resched_name
= kasprintf(GFP_KERNEL
, "resched%d", cpu
);
138 rc
= bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR
,
140 xen_reschedule_interrupt
,
141 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
146 per_cpu(xen_resched_irq
, cpu
).irq
= rc
;
148 callfunc_name
= kasprintf(GFP_KERNEL
, "callfunc%d", cpu
);
149 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR
,
151 xen_call_function_interrupt
,
152 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
157 per_cpu(xen_callfunc_irq
, cpu
).irq
= rc
;
159 debug_name
= kasprintf(GFP_KERNEL
, "debug%d", cpu
);
160 rc
= bind_virq_to_irqhandler(VIRQ_DEBUG
, cpu
, xen_debug_interrupt
,
161 IRQF_DISABLED
| IRQF_PERCPU
| IRQF_NOBALANCING
,
165 per_cpu(xen_debug_irq
, cpu
).irq
= rc
;
167 callfunc_name
= kasprintf(GFP_KERNEL
, "callfuncsingle%d", cpu
);
168 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR
,
170 xen_call_function_single_interrupt
,
171 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
176 per_cpu(xen_callfuncsingle_irq
, cpu
).irq
= rc
;
179 * The IRQ worker on PVHVM goes through the native path and uses the
182 if (xen_hvm_domain())
185 callfunc_name
= kasprintf(GFP_KERNEL
, "irqwork%d", cpu
);
186 rc
= bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR
,
188 xen_irq_work_interrupt
,
189 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
194 per_cpu(xen_irq_work
, cpu
).irq
= rc
;
199 xen_smp_intr_free(cpu
);
203 static void __init
xen_fill_possible_map(void)
207 if (xen_initial_domain())
210 for (i
= 0; i
< nr_cpu_ids
; i
++) {
211 rc
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, i
, NULL
);
214 set_cpu_possible(i
, true);
219 static void __init
xen_filter_cpu_maps(void)
222 unsigned int subtract
= 0;
224 if (!xen_initial_domain())
229 for (i
= 0; i
< nr_cpu_ids
; i
++) {
230 rc
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, i
, NULL
);
233 set_cpu_possible(i
, true);
235 set_cpu_possible(i
, false);
236 set_cpu_present(i
, false);
240 #ifdef CONFIG_HOTPLUG_CPU
241 /* This is akin to using 'nr_cpus' on the Linux command line.
242 * Which is OK as when we use 'dom0_max_vcpus=X' we can only
243 * have up to X, while nr_cpu_ids is greater than X. This
244 * normally is not a problem, except when CPU hotplugging
245 * is involved and then there might be more than X CPUs
246 * in the guest - which will not work as there is no
247 * hypercall to expand the max number of VCPUs an already
248 * running guest has. So cap it up to X. */
250 nr_cpu_ids
= nr_cpu_ids
- subtract
;
255 static void __init
xen_smp_prepare_boot_cpu(void)
257 BUG_ON(smp_processor_id() != 0);
258 native_smp_prepare_boot_cpu();
260 /* We've switched to the "real" per-cpu gdt, so make sure the
261 old memory can be recycled */
262 make_lowmem_page_readwrite(xen_initial_gdt
);
264 xen_filter_cpu_maps();
265 xen_setup_vcpu_info_placement();
268 static void __init
xen_smp_prepare_cpus(unsigned int max_cpus
)
273 if (skip_ioapic_setup
) {
274 char *m
= (max_cpus
== 0) ?
275 "The nosmp parameter is incompatible with Xen; " \
276 "use Xen dom0_max_vcpus=1 parameter" :
277 "The noapic parameter is incompatible with Xen";
282 xen_init_lock_cpu(0);
284 smp_store_boot_cpu_info();
285 cpu_data(0).x86_max_cores
= 1;
287 for_each_possible_cpu(i
) {
288 zalloc_cpumask_var(&per_cpu(cpu_sibling_map
, i
), GFP_KERNEL
);
289 zalloc_cpumask_var(&per_cpu(cpu_core_map
, i
), GFP_KERNEL
);
290 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map
, i
), GFP_KERNEL
);
292 set_cpu_sibling_map(0);
294 if (xen_smp_intr_init(0))
297 if (!alloc_cpumask_var(&xen_cpu_initialized_map
, GFP_KERNEL
))
298 panic("could not allocate xen_cpu_initialized_map\n");
300 cpumask_copy(xen_cpu_initialized_map
, cpumask_of(0));
302 /* Restrict the possible_map according to max_cpus. */
303 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus
)) {
304 for (cpu
= nr_cpu_ids
- 1; !cpu_possible(cpu
); cpu
--)
306 set_cpu_possible(cpu
, false);
309 for_each_possible_cpu(cpu
)
310 set_cpu_present(cpu
, true);
314 cpu_initialize_context(unsigned int cpu
, struct task_struct
*idle
)
316 struct vcpu_guest_context
*ctxt
;
317 struct desc_struct
*gdt
;
318 unsigned long gdt_mfn
;
320 if (cpumask_test_and_set_cpu(cpu
, xen_cpu_initialized_map
))
323 ctxt
= kzalloc(sizeof(*ctxt
), GFP_KERNEL
);
327 gdt
= get_cpu_gdt_table(cpu
);
329 ctxt
->flags
= VGCF_IN_KERNEL
;
330 ctxt
->user_regs
.ss
= __KERNEL_DS
;
332 ctxt
->user_regs
.fs
= __KERNEL_PERCPU
;
333 ctxt
->user_regs
.gs
= __KERNEL_STACK_CANARY
;
335 ctxt
->gs_base_kernel
= per_cpu_offset(cpu
);
337 ctxt
->user_regs
.eip
= (unsigned long)cpu_bringup_and_idle
;
339 memset(&ctxt
->fpu_ctxt
, 0, sizeof(ctxt
->fpu_ctxt
));
342 ctxt
->user_regs
.eflags
= 0x1000; /* IOPL_RING1 */
343 ctxt
->user_regs
.ds
= __USER_DS
;
344 ctxt
->user_regs
.es
= __USER_DS
;
346 xen_copy_trap_info(ctxt
->trap_ctxt
);
350 BUG_ON((unsigned long)gdt
& ~PAGE_MASK
);
352 gdt_mfn
= arbitrary_virt_to_mfn(gdt
);
353 make_lowmem_page_readonly(gdt
);
354 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn
));
356 ctxt
->gdt_frames
[0] = gdt_mfn
;
357 ctxt
->gdt_ents
= GDT_ENTRIES
;
359 ctxt
->kernel_ss
= __KERNEL_DS
;
360 ctxt
->kernel_sp
= idle
->thread
.sp0
;
363 ctxt
->event_callback_cs
= __KERNEL_CS
;
364 ctxt
->failsafe_callback_cs
= __KERNEL_CS
;
366 ctxt
->event_callback_eip
=
367 (unsigned long)xen_hypervisor_callback
;
368 ctxt
->failsafe_callback_eip
=
369 (unsigned long)xen_failsafe_callback
;
371 ctxt
->user_regs
.cs
= __KERNEL_CS
;
372 ctxt
->user_regs
.esp
= idle
->thread
.sp0
- sizeof(struct pt_regs
);
374 per_cpu(xen_cr3
, cpu
) = __pa(swapper_pg_dir
);
375 ctxt
->ctrlreg
[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir
));
377 if (HYPERVISOR_vcpu_op(VCPUOP_initialise
, cpu
, ctxt
))
384 static int __cpuinit
xen_cpu_up(unsigned int cpu
, struct task_struct
*idle
)
388 per_cpu(current_task
, cpu
) = idle
;
392 clear_tsk_thread_flag(idle
, TIF_FORK
);
393 per_cpu(kernel_stack
, cpu
) =
394 (unsigned long)task_stack_page(idle
) -
395 KERNEL_STACK_OFFSET
+ THREAD_SIZE
;
397 xen_setup_runstate_info(cpu
);
398 xen_setup_timer(cpu
);
399 xen_init_lock_cpu(cpu
);
401 per_cpu(cpu_state
, cpu
) = CPU_UP_PREPARE
;
403 /* make sure interrupts start blocked */
404 per_cpu(xen_vcpu
, cpu
)->evtchn_upcall_mask
= 1;
406 rc
= cpu_initialize_context(cpu
, idle
);
410 if (num_online_cpus() == 1)
411 /* Just in case we booted with a single CPU. */
412 alternatives_enable_smp();
414 rc
= xen_smp_intr_init(cpu
);
418 rc
= HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
);
421 while(per_cpu(cpu_state
, cpu
) != CPU_ONLINE
) {
422 HYPERVISOR_sched_op(SCHEDOP_yield
, NULL
);
429 static void xen_smp_cpus_done(unsigned int max_cpus
)
433 #ifdef CONFIG_HOTPLUG_CPU
434 static int xen_cpu_disable(void)
436 unsigned int cpu
= smp_processor_id();
440 cpu_disable_common();
442 load_cr3(swapper_pg_dir
);
446 static void xen_cpu_die(unsigned int cpu
)
448 while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up
, cpu
, NULL
)) {
449 current
->state
= TASK_UNINTERRUPTIBLE
;
450 schedule_timeout(HZ
/10);
452 xen_smp_intr_free(cpu
);
453 xen_uninit_lock_cpu(cpu
);
454 xen_teardown_timer(cpu
);
457 static void __cpuinit
xen_play_dead(void) /* used only with HOTPLUG_CPU */
460 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
464 #else /* !CONFIG_HOTPLUG_CPU */
465 static int xen_cpu_disable(void)
470 static void xen_cpu_die(unsigned int cpu
)
475 static void xen_play_dead(void)
481 static void stop_self(void *v
)
483 int cpu
= smp_processor_id();
485 /* make sure we're not pinning something down */
486 load_cr3(swapper_pg_dir
);
487 /* should set up a minimal gdt */
489 set_cpu_online(cpu
, false);
491 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
);
495 static void xen_stop_other_cpus(int wait
)
497 smp_call_function(stop_self
, NULL
, wait
);
500 static void xen_smp_send_reschedule(int cpu
)
502 xen_send_IPI_one(cpu
, XEN_RESCHEDULE_VECTOR
);
505 static void __xen_send_IPI_mask(const struct cpumask
*mask
,
510 for_each_cpu_and(cpu
, mask
, cpu_online_mask
)
511 xen_send_IPI_one(cpu
, vector
);
514 static void xen_smp_send_call_function_ipi(const struct cpumask
*mask
)
518 __xen_send_IPI_mask(mask
, XEN_CALL_FUNCTION_VECTOR
);
520 /* Make sure other vcpus get a chance to run if they need to. */
521 for_each_cpu(cpu
, mask
) {
522 if (xen_vcpu_stolen(cpu
)) {
523 HYPERVISOR_sched_op(SCHEDOP_yield
, NULL
);
529 static void xen_smp_send_call_function_single_ipi(int cpu
)
531 __xen_send_IPI_mask(cpumask_of(cpu
),
532 XEN_CALL_FUNCTION_SINGLE_VECTOR
);
535 static inline int xen_map_vector(int vector
)
540 case RESCHEDULE_VECTOR
:
541 xen_vector
= XEN_RESCHEDULE_VECTOR
;
543 case CALL_FUNCTION_VECTOR
:
544 xen_vector
= XEN_CALL_FUNCTION_VECTOR
;
546 case CALL_FUNCTION_SINGLE_VECTOR
:
547 xen_vector
= XEN_CALL_FUNCTION_SINGLE_VECTOR
;
549 case IRQ_WORK_VECTOR
:
550 xen_vector
= XEN_IRQ_WORK_VECTOR
;
554 printk(KERN_ERR
"xen: vector 0x%x is not implemented\n",
561 void xen_send_IPI_mask(const struct cpumask
*mask
,
564 int xen_vector
= xen_map_vector(vector
);
567 __xen_send_IPI_mask(mask
, xen_vector
);
570 void xen_send_IPI_all(int vector
)
572 int xen_vector
= xen_map_vector(vector
);
575 __xen_send_IPI_mask(cpu_online_mask
, xen_vector
);
578 void xen_send_IPI_self(int vector
)
580 int xen_vector
= xen_map_vector(vector
);
583 xen_send_IPI_one(smp_processor_id(), xen_vector
);
586 void xen_send_IPI_mask_allbutself(const struct cpumask
*mask
,
590 unsigned int this_cpu
= smp_processor_id();
591 int xen_vector
= xen_map_vector(vector
);
593 if (!(num_online_cpus() > 1) || (xen_vector
< 0))
596 for_each_cpu_and(cpu
, mask
, cpu_online_mask
) {
600 xen_send_IPI_one(cpu
, xen_vector
);
604 void xen_send_IPI_allbutself(int vector
)
606 xen_send_IPI_mask_allbutself(cpu_online_mask
, vector
);
609 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
)
612 generic_smp_call_function_interrupt();
613 inc_irq_stat(irq_call_count
);
619 static irqreturn_t
xen_call_function_single_interrupt(int irq
, void *dev_id
)
622 generic_smp_call_function_single_interrupt();
623 inc_irq_stat(irq_call_count
);
629 static irqreturn_t
xen_irq_work_interrupt(int irq
, void *dev_id
)
633 inc_irq_stat(apic_irq_work_irqs
);
639 static const struct smp_ops xen_smp_ops __initconst
= {
640 .smp_prepare_boot_cpu
= xen_smp_prepare_boot_cpu
,
641 .smp_prepare_cpus
= xen_smp_prepare_cpus
,
642 .smp_cpus_done
= xen_smp_cpus_done
,
644 .cpu_up
= xen_cpu_up
,
645 .cpu_die
= xen_cpu_die
,
646 .cpu_disable
= xen_cpu_disable
,
647 .play_dead
= xen_play_dead
,
649 .stop_other_cpus
= xen_stop_other_cpus
,
650 .smp_send_reschedule
= xen_smp_send_reschedule
,
652 .send_call_func_ipi
= xen_smp_send_call_function_ipi
,
653 .send_call_func_single_ipi
= xen_smp_send_call_function_single_ipi
,
656 void __init
xen_smp_init(void)
658 smp_ops
= xen_smp_ops
;
659 xen_fill_possible_map();
660 xen_init_spinlocks();
663 static void __init
xen_hvm_smp_prepare_cpus(unsigned int max_cpus
)
665 native_smp_prepare_cpus(max_cpus
);
666 WARN_ON(xen_smp_intr_init(0));
668 xen_init_lock_cpu(0);
671 static int __cpuinit
xen_hvm_cpu_up(unsigned int cpu
, struct task_struct
*tidle
)
674 rc
= native_cpu_up(cpu
, tidle
);
675 WARN_ON (xen_smp_intr_init(cpu
));
679 static void xen_hvm_cpu_die(unsigned int cpu
)
685 void __init
xen_hvm_smp_init(void)
687 if (!xen_have_vector_callback
)
689 smp_ops
.smp_prepare_cpus
= xen_hvm_smp_prepare_cpus
;
690 smp_ops
.smp_send_reschedule
= xen_smp_send_reschedule
;
691 smp_ops
.cpu_up
= xen_hvm_cpu_up
;
692 smp_ops
.cpu_die
= xen_hvm_cpu_die
;
693 smp_ops
.send_call_func_ipi
= xen_smp_send_call_function_ipi
;
694 smp_ops
.send_call_func_single_ipi
= xen_smp_send_call_function_single_ipi
;