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
);
83 set_cpu_online(cpu
, true);
85 this_cpu_write(cpu_state
, CPU_ONLINE
);
89 /* We can take interrupts now: we're officially "up". */
92 wmb(); /* make sure everything is out */
95 static void __cpuinit
cpu_bringup_and_idle(void)
101 static int xen_smp_intr_init(unsigned int cpu
)
104 const char *resched_name
, *callfunc_name
, *debug_name
;
106 resched_name
= kasprintf(GFP_KERNEL
, "resched%d", cpu
);
107 rc
= bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR
,
109 xen_reschedule_interrupt
,
110 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
115 per_cpu(xen_resched_irq
, cpu
) = rc
;
117 callfunc_name
= kasprintf(GFP_KERNEL
, "callfunc%d", cpu
);
118 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR
,
120 xen_call_function_interrupt
,
121 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
126 per_cpu(xen_callfunc_irq
, cpu
) = rc
;
128 debug_name
= kasprintf(GFP_KERNEL
, "debug%d", cpu
);
129 rc
= bind_virq_to_irqhandler(VIRQ_DEBUG
, cpu
, xen_debug_interrupt
,
130 IRQF_DISABLED
| IRQF_PERCPU
| IRQF_NOBALANCING
,
134 per_cpu(xen_debug_irq
, cpu
) = rc
;
136 callfunc_name
= kasprintf(GFP_KERNEL
, "callfuncsingle%d", cpu
);
137 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR
,
139 xen_call_function_single_interrupt
,
140 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
145 per_cpu(xen_callfuncsingle_irq
, cpu
) = rc
;
147 callfunc_name
= kasprintf(GFP_KERNEL
, "irqwork%d", cpu
);
148 rc
= bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR
,
150 xen_irq_work_interrupt
,
151 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
156 per_cpu(xen_irq_work
, cpu
) = rc
;
161 if (per_cpu(xen_resched_irq
, cpu
) >= 0)
162 unbind_from_irqhandler(per_cpu(xen_resched_irq
, cpu
), NULL
);
163 if (per_cpu(xen_callfunc_irq
, cpu
) >= 0)
164 unbind_from_irqhandler(per_cpu(xen_callfunc_irq
, cpu
), NULL
);
165 if (per_cpu(xen_debug_irq
, cpu
) >= 0)
166 unbind_from_irqhandler(per_cpu(xen_debug_irq
, cpu
), NULL
);
167 if (per_cpu(xen_callfuncsingle_irq
, cpu
) >= 0)
168 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq
, cpu
),
170 if (per_cpu(xen_irq_work
, cpu
) >= 0)
171 unbind_from_irqhandler(per_cpu(xen_irq_work
, cpu
), NULL
);
176 static void __init
xen_fill_possible_map(void)
180 if (xen_initial_domain())
183 for (i
= 0; i
< nr_cpu_ids
; i
++) {
184 rc
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, i
, NULL
);
187 set_cpu_possible(i
, true);
192 static void __init
xen_filter_cpu_maps(void)
195 unsigned int subtract
= 0;
197 if (!xen_initial_domain())
202 for (i
= 0; i
< nr_cpu_ids
; i
++) {
203 rc
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, i
, NULL
);
206 set_cpu_possible(i
, true);
208 set_cpu_possible(i
, false);
209 set_cpu_present(i
, false);
213 #ifdef CONFIG_HOTPLUG_CPU
214 /* This is akin to using 'nr_cpus' on the Linux command line.
215 * Which is OK as when we use 'dom0_max_vcpus=X' we can only
216 * have up to X, while nr_cpu_ids is greater than X. This
217 * normally is not a problem, except when CPU hotplugging
218 * is involved and then there might be more than X CPUs
219 * in the guest - which will not work as there is no
220 * hypercall to expand the max number of VCPUs an already
221 * running guest has. So cap it up to X. */
223 nr_cpu_ids
= nr_cpu_ids
- subtract
;
228 static void __init
xen_smp_prepare_boot_cpu(void)
230 BUG_ON(smp_processor_id() != 0);
231 native_smp_prepare_boot_cpu();
233 /* We've switched to the "real" per-cpu gdt, so make sure the
234 old memory can be recycled */
235 make_lowmem_page_readwrite(xen_initial_gdt
);
237 xen_filter_cpu_maps();
238 xen_setup_vcpu_info_placement();
241 static void __init
xen_smp_prepare_cpus(unsigned int max_cpus
)
246 if (skip_ioapic_setup
) {
247 char *m
= (max_cpus
== 0) ?
248 "The nosmp parameter is incompatible with Xen; " \
249 "use Xen dom0_max_vcpus=1 parameter" :
250 "The noapic parameter is incompatible with Xen";
255 xen_init_lock_cpu(0);
257 smp_store_cpu_info(0);
258 cpu_data(0).x86_max_cores
= 1;
260 for_each_possible_cpu(i
) {
261 zalloc_cpumask_var(&per_cpu(cpu_sibling_map
, i
), GFP_KERNEL
);
262 zalloc_cpumask_var(&per_cpu(cpu_core_map
, i
), GFP_KERNEL
);
263 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map
, i
), GFP_KERNEL
);
265 set_cpu_sibling_map(0);
267 if (xen_smp_intr_init(0))
270 if (!alloc_cpumask_var(&xen_cpu_initialized_map
, GFP_KERNEL
))
271 panic("could not allocate xen_cpu_initialized_map\n");
273 cpumask_copy(xen_cpu_initialized_map
, cpumask_of(0));
275 /* Restrict the possible_map according to max_cpus. */
276 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus
)) {
277 for (cpu
= nr_cpu_ids
- 1; !cpu_possible(cpu
); cpu
--)
279 set_cpu_possible(cpu
, false);
282 for_each_possible_cpu(cpu
)
283 set_cpu_present(cpu
, true);
287 cpu_initialize_context(unsigned int cpu
, struct task_struct
*idle
)
289 struct vcpu_guest_context
*ctxt
;
290 struct desc_struct
*gdt
;
291 unsigned long gdt_mfn
;
293 if (cpumask_test_and_set_cpu(cpu
, xen_cpu_initialized_map
))
296 ctxt
= kzalloc(sizeof(*ctxt
), GFP_KERNEL
);
300 gdt
= get_cpu_gdt_table(cpu
);
302 ctxt
->flags
= VGCF_IN_KERNEL
;
303 ctxt
->user_regs
.ds
= __USER_DS
;
304 ctxt
->user_regs
.es
= __USER_DS
;
305 ctxt
->user_regs
.ss
= __KERNEL_DS
;
307 ctxt
->user_regs
.fs
= __KERNEL_PERCPU
;
308 ctxt
->user_regs
.gs
= __KERNEL_STACK_CANARY
;
310 ctxt
->gs_base_kernel
= per_cpu_offset(cpu
);
312 ctxt
->user_regs
.eip
= (unsigned long)cpu_bringup_and_idle
;
313 ctxt
->user_regs
.eflags
= 0x1000; /* IOPL_RING1 */
315 memset(&ctxt
->fpu_ctxt
, 0, sizeof(ctxt
->fpu_ctxt
));
317 xen_copy_trap_info(ctxt
->trap_ctxt
);
321 BUG_ON((unsigned long)gdt
& ~PAGE_MASK
);
323 gdt_mfn
= arbitrary_virt_to_mfn(gdt
);
324 make_lowmem_page_readonly(gdt
);
325 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn
));
327 ctxt
->gdt_frames
[0] = gdt_mfn
;
328 ctxt
->gdt_ents
= GDT_ENTRIES
;
330 ctxt
->user_regs
.cs
= __KERNEL_CS
;
331 ctxt
->user_regs
.esp
= idle
->thread
.sp0
- sizeof(struct pt_regs
);
333 ctxt
->kernel_ss
= __KERNEL_DS
;
334 ctxt
->kernel_sp
= idle
->thread
.sp0
;
337 ctxt
->event_callback_cs
= __KERNEL_CS
;
338 ctxt
->failsafe_callback_cs
= __KERNEL_CS
;
340 ctxt
->event_callback_eip
= (unsigned long)xen_hypervisor_callback
;
341 ctxt
->failsafe_callback_eip
= (unsigned long)xen_failsafe_callback
;
343 per_cpu(xen_cr3
, cpu
) = __pa(swapper_pg_dir
);
344 ctxt
->ctrlreg
[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir
));
346 if (HYPERVISOR_vcpu_op(VCPUOP_initialise
, cpu
, ctxt
))
353 static int __cpuinit
xen_cpu_up(unsigned int cpu
, struct task_struct
*idle
)
357 per_cpu(current_task
, cpu
) = idle
;
361 clear_tsk_thread_flag(idle
, TIF_FORK
);
362 per_cpu(kernel_stack
, cpu
) =
363 (unsigned long)task_stack_page(idle
) -
364 KERNEL_STACK_OFFSET
+ THREAD_SIZE
;
366 xen_setup_runstate_info(cpu
);
367 xen_setup_timer(cpu
);
368 xen_init_lock_cpu(cpu
);
370 per_cpu(cpu_state
, cpu
) = CPU_UP_PREPARE
;
372 /* make sure interrupts start blocked */
373 per_cpu(xen_vcpu
, cpu
)->evtchn_upcall_mask
= 1;
375 rc
= cpu_initialize_context(cpu
, idle
);
379 if (num_online_cpus() == 1)
380 alternatives_smp_switch(1);
382 rc
= xen_smp_intr_init(cpu
);
386 rc
= HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
);
389 while(per_cpu(cpu_state
, cpu
) != CPU_ONLINE
) {
390 HYPERVISOR_sched_op(SCHEDOP_yield
, NULL
);
397 static void xen_smp_cpus_done(unsigned int max_cpus
)
401 #ifdef CONFIG_HOTPLUG_CPU
402 static int xen_cpu_disable(void)
404 unsigned int cpu
= smp_processor_id();
408 cpu_disable_common();
410 load_cr3(swapper_pg_dir
);
414 static void xen_cpu_die(unsigned int cpu
)
416 while (HYPERVISOR_vcpu_op(VCPUOP_is_up
, cpu
, NULL
)) {
417 current
->state
= TASK_UNINTERRUPTIBLE
;
418 schedule_timeout(HZ
/10);
420 unbind_from_irqhandler(per_cpu(xen_resched_irq
, cpu
), NULL
);
421 unbind_from_irqhandler(per_cpu(xen_callfunc_irq
, cpu
), NULL
);
422 unbind_from_irqhandler(per_cpu(xen_debug_irq
, cpu
), NULL
);
423 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq
, cpu
), NULL
);
424 unbind_from_irqhandler(per_cpu(xen_irq_work
, cpu
), NULL
);
425 xen_uninit_lock_cpu(cpu
);
426 xen_teardown_timer(cpu
);
428 if (num_online_cpus() == 1)
429 alternatives_smp_switch(0);
432 static void __cpuinit
xen_play_dead(void) /* used only with HOTPLUG_CPU */
435 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
438 * Balance out the preempt calls - as we are running in cpu_idle
439 * loop which has been called at bootup from cpu_bringup_and_idle.
440 * The cpucpu_bringup_and_idle called cpu_bringup which made a
441 * preempt_disable() So this preempt_enable will balance it out.
446 #else /* !CONFIG_HOTPLUG_CPU */
447 static int xen_cpu_disable(void)
452 static void xen_cpu_die(unsigned int cpu
)
457 static void xen_play_dead(void)
463 static void stop_self(void *v
)
465 int cpu
= smp_processor_id();
467 /* make sure we're not pinning something down */
468 load_cr3(swapper_pg_dir
);
469 /* should set up a minimal gdt */
471 set_cpu_online(cpu
, false);
473 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
);
477 static void xen_stop_other_cpus(int wait
)
479 smp_call_function(stop_self
, NULL
, wait
);
482 static void xen_smp_send_reschedule(int cpu
)
484 xen_send_IPI_one(cpu
, XEN_RESCHEDULE_VECTOR
);
487 static void __xen_send_IPI_mask(const struct cpumask
*mask
,
492 for_each_cpu_and(cpu
, mask
, cpu_online_mask
)
493 xen_send_IPI_one(cpu
, vector
);
496 static void xen_smp_send_call_function_ipi(const struct cpumask
*mask
)
500 __xen_send_IPI_mask(mask
, XEN_CALL_FUNCTION_VECTOR
);
502 /* Make sure other vcpus get a chance to run if they need to. */
503 for_each_cpu(cpu
, mask
) {
504 if (xen_vcpu_stolen(cpu
)) {
505 HYPERVISOR_sched_op(SCHEDOP_yield
, NULL
);
511 static void xen_smp_send_call_function_single_ipi(int cpu
)
513 __xen_send_IPI_mask(cpumask_of(cpu
),
514 XEN_CALL_FUNCTION_SINGLE_VECTOR
);
517 static inline int xen_map_vector(int vector
)
522 case RESCHEDULE_VECTOR
:
523 xen_vector
= XEN_RESCHEDULE_VECTOR
;
525 case CALL_FUNCTION_VECTOR
:
526 xen_vector
= XEN_CALL_FUNCTION_VECTOR
;
528 case CALL_FUNCTION_SINGLE_VECTOR
:
529 xen_vector
= XEN_CALL_FUNCTION_SINGLE_VECTOR
;
531 case IRQ_WORK_VECTOR
:
532 xen_vector
= XEN_IRQ_WORK_VECTOR
;
536 printk(KERN_ERR
"xen: vector 0x%x is not implemented\n",
543 void xen_send_IPI_mask(const struct cpumask
*mask
,
546 int xen_vector
= xen_map_vector(vector
);
549 __xen_send_IPI_mask(mask
, xen_vector
);
552 void xen_send_IPI_all(int vector
)
554 int xen_vector
= xen_map_vector(vector
);
557 __xen_send_IPI_mask(cpu_online_mask
, xen_vector
);
560 void xen_send_IPI_self(int vector
)
562 int xen_vector
= xen_map_vector(vector
);
565 xen_send_IPI_one(smp_processor_id(), xen_vector
);
568 void xen_send_IPI_mask_allbutself(const struct cpumask
*mask
,
572 unsigned int this_cpu
= smp_processor_id();
574 if (!(num_online_cpus() > 1))
577 for_each_cpu_and(cpu
, mask
, cpu_online_mask
) {
581 xen_smp_send_call_function_single_ipi(cpu
);
585 void xen_send_IPI_allbutself(int vector
)
587 int xen_vector
= xen_map_vector(vector
);
590 xen_send_IPI_mask_allbutself(cpu_online_mask
, xen_vector
);
593 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
)
596 generic_smp_call_function_interrupt();
597 inc_irq_stat(irq_call_count
);
603 static irqreturn_t
xen_call_function_single_interrupt(int irq
, void *dev_id
)
606 generic_smp_call_function_single_interrupt();
607 inc_irq_stat(irq_call_count
);
613 static irqreturn_t
xen_irq_work_interrupt(int irq
, void *dev_id
)
617 inc_irq_stat(apic_irq_work_irqs
);
623 static const struct smp_ops xen_smp_ops __initconst
= {
624 .smp_prepare_boot_cpu
= xen_smp_prepare_boot_cpu
,
625 .smp_prepare_cpus
= xen_smp_prepare_cpus
,
626 .smp_cpus_done
= xen_smp_cpus_done
,
628 .cpu_up
= xen_cpu_up
,
629 .cpu_die
= xen_cpu_die
,
630 .cpu_disable
= xen_cpu_disable
,
631 .play_dead
= xen_play_dead
,
633 .stop_other_cpus
= xen_stop_other_cpus
,
634 .smp_send_reschedule
= xen_smp_send_reschedule
,
636 .send_call_func_ipi
= xen_smp_send_call_function_ipi
,
637 .send_call_func_single_ipi
= xen_smp_send_call_function_single_ipi
,
640 void __init
xen_smp_init(void)
642 smp_ops
= xen_smp_ops
;
643 xen_fill_possible_map();
644 xen_init_spinlocks();
647 static void __init
xen_hvm_smp_prepare_cpus(unsigned int max_cpus
)
649 native_smp_prepare_cpus(max_cpus
);
650 WARN_ON(xen_smp_intr_init(0));
652 xen_init_lock_cpu(0);
655 static int __cpuinit
xen_hvm_cpu_up(unsigned int cpu
, struct task_struct
*tidle
)
658 rc
= native_cpu_up(cpu
, tidle
);
659 WARN_ON (xen_smp_intr_init(cpu
));
663 static void xen_hvm_cpu_die(unsigned int cpu
)
665 unbind_from_irqhandler(per_cpu(xen_resched_irq
, cpu
), NULL
);
666 unbind_from_irqhandler(per_cpu(xen_callfunc_irq
, cpu
), NULL
);
667 unbind_from_irqhandler(per_cpu(xen_debug_irq
, cpu
), NULL
);
668 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq
, cpu
), NULL
);
669 unbind_from_irqhandler(per_cpu(xen_irq_work
, cpu
), NULL
);
673 void __init
xen_hvm_smp_init(void)
675 if (!xen_have_vector_callback
)
677 smp_ops
.smp_prepare_cpus
= xen_hvm_smp_prepare_cpus
;
678 smp_ops
.smp_send_reschedule
= xen_smp_send_reschedule
;
679 smp_ops
.cpu_up
= xen_hvm_cpu_up
;
680 smp_ops
.cpu_die
= xen_hvm_cpu_die
;
681 smp_ops
.send_call_func_ipi
= xen_smp_send_call_function_ipi
;
682 smp_ops
.send_call_func_single_ipi
= xen_smp_send_call_function_single_ipi
;