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/smp.h>
19 #include <asm/paravirt.h>
21 #include <asm/pgtable.h>
24 #include <xen/interface/xen.h>
25 #include <xen/interface/vcpu.h>
27 #include <asm/xen/interface.h>
28 #include <asm/xen/hypercall.h>
31 #include <xen/events.h>
36 cpumask_var_t xen_cpu_initialized_map
;
38 static DEFINE_PER_CPU(int, resched_irq
);
39 static DEFINE_PER_CPU(int, callfunc_irq
);
40 static DEFINE_PER_CPU(int, callfuncsingle_irq
);
41 static DEFINE_PER_CPU(int, debug_irq
) = -1;
43 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
);
44 static irqreturn_t
xen_call_function_single_interrupt(int irq
, void *dev_id
);
47 * Reschedule call back. Nothing to do,
48 * all the work is done automatically when
49 * we return from the interrupt.
51 static irqreturn_t
xen_reschedule_interrupt(int irq
, void *dev_id
)
53 inc_irq_stat(irq_resched_count
);
58 static __cpuinit
void cpu_bringup(void)
60 int cpu
= smp_processor_id();
63 touch_softlockup_watchdog();
66 xen_enable_sysenter();
69 cpu
= smp_processor_id();
70 smp_store_cpu_info(cpu
);
71 cpu_data(cpu
).x86_max_cores
= 1;
72 set_cpu_sibling_map(cpu
);
74 xen_setup_cpu_clockevents();
76 cpu_set(cpu
, cpu_online_map
);
77 percpu_write(cpu_state
, CPU_ONLINE
);
80 /* We can take interrupts now: we're officially "up". */
83 wmb(); /* make sure everything is out */
86 static __cpuinit
void cpu_bringup_and_idle(void)
92 static int xen_smp_intr_init(unsigned int cpu
)
95 const char *resched_name
, *callfunc_name
, *debug_name
;
97 resched_name
= kasprintf(GFP_KERNEL
, "resched%d", cpu
);
98 rc
= bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR
,
100 xen_reschedule_interrupt
,
101 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
106 per_cpu(resched_irq
, cpu
) = rc
;
108 callfunc_name
= kasprintf(GFP_KERNEL
, "callfunc%d", cpu
);
109 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR
,
111 xen_call_function_interrupt
,
112 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
117 per_cpu(callfunc_irq
, cpu
) = rc
;
119 debug_name
= kasprintf(GFP_KERNEL
, "debug%d", cpu
);
120 rc
= bind_virq_to_irqhandler(VIRQ_DEBUG
, cpu
, xen_debug_interrupt
,
121 IRQF_DISABLED
| IRQF_PERCPU
| IRQF_NOBALANCING
,
125 per_cpu(debug_irq
, cpu
) = rc
;
127 callfunc_name
= kasprintf(GFP_KERNEL
, "callfuncsingle%d", cpu
);
128 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR
,
130 xen_call_function_single_interrupt
,
131 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
136 per_cpu(callfuncsingle_irq
, cpu
) = rc
;
141 if (per_cpu(resched_irq
, cpu
) >= 0)
142 unbind_from_irqhandler(per_cpu(resched_irq
, cpu
), NULL
);
143 if (per_cpu(callfunc_irq
, cpu
) >= 0)
144 unbind_from_irqhandler(per_cpu(callfunc_irq
, cpu
), NULL
);
145 if (per_cpu(debug_irq
, cpu
) >= 0)
146 unbind_from_irqhandler(per_cpu(debug_irq
, cpu
), NULL
);
147 if (per_cpu(callfuncsingle_irq
, cpu
) >= 0)
148 unbind_from_irqhandler(per_cpu(callfuncsingle_irq
, cpu
), NULL
);
153 static void __init
xen_fill_possible_map(void)
157 for (i
= 0; i
< nr_cpu_ids
; i
++) {
158 rc
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, i
, NULL
);
161 cpu_set(i
, cpu_possible_map
);
166 static void __init
xen_smp_prepare_boot_cpu(void)
168 BUG_ON(smp_processor_id() != 0);
169 native_smp_prepare_boot_cpu();
171 /* We've switched to the "real" per-cpu gdt, so make sure the
172 old memory can be recycled */
173 make_lowmem_page_readwrite(xen_initial_gdt
);
175 xen_setup_vcpu_info_placement();
178 static void __init
xen_smp_prepare_cpus(unsigned int max_cpus
)
182 xen_init_lock_cpu(0);
184 smp_store_cpu_info(0);
185 cpu_data(0).x86_max_cores
= 1;
186 set_cpu_sibling_map(0);
188 if (xen_smp_intr_init(0))
191 if (!alloc_cpumask_var(&xen_cpu_initialized_map
, GFP_KERNEL
))
192 panic("could not allocate xen_cpu_initialized_map\n");
194 cpumask_copy(xen_cpu_initialized_map
, cpumask_of(0));
196 /* Restrict the possible_map according to max_cpus. */
197 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus
)) {
198 for (cpu
= nr_cpu_ids
- 1; !cpu_possible(cpu
); cpu
--)
200 cpu_clear(cpu
, cpu_possible_map
);
203 for_each_possible_cpu (cpu
) {
204 struct task_struct
*idle
;
209 idle
= fork_idle(cpu
);
211 panic("failed fork for CPU %d", cpu
);
213 cpu_set(cpu
, cpu_present_map
);
218 cpu_initialize_context(unsigned int cpu
, struct task_struct
*idle
)
220 struct vcpu_guest_context
*ctxt
;
221 struct desc_struct
*gdt
;
223 if (cpumask_test_and_set_cpu(cpu
, xen_cpu_initialized_map
))
226 ctxt
= kzalloc(sizeof(*ctxt
), GFP_KERNEL
);
230 gdt
= get_cpu_gdt_table(cpu
);
232 ctxt
->flags
= VGCF_IN_KERNEL
;
233 ctxt
->user_regs
.ds
= __USER_DS
;
234 ctxt
->user_regs
.es
= __USER_DS
;
235 ctxt
->user_regs
.ss
= __KERNEL_DS
;
237 ctxt
->user_regs
.fs
= __KERNEL_PERCPU
;
239 ctxt
->gs_base_kernel
= per_cpu_offset(cpu
);
241 ctxt
->user_regs
.eip
= (unsigned long)cpu_bringup_and_idle
;
242 ctxt
->user_regs
.eflags
= 0x1000; /* IOPL_RING1 */
244 memset(&ctxt
->fpu_ctxt
, 0, sizeof(ctxt
->fpu_ctxt
));
246 xen_copy_trap_info(ctxt
->trap_ctxt
);
250 BUG_ON((unsigned long)gdt
& ~PAGE_MASK
);
251 make_lowmem_page_readonly(gdt
);
253 ctxt
->gdt_frames
[0] = virt_to_mfn(gdt
);
254 ctxt
->gdt_ents
= GDT_ENTRIES
;
256 ctxt
->user_regs
.cs
= __KERNEL_CS
;
257 ctxt
->user_regs
.esp
= idle
->thread
.sp0
- sizeof(struct pt_regs
);
259 ctxt
->kernel_ss
= __KERNEL_DS
;
260 ctxt
->kernel_sp
= idle
->thread
.sp0
;
263 ctxt
->event_callback_cs
= __KERNEL_CS
;
264 ctxt
->failsafe_callback_cs
= __KERNEL_CS
;
266 ctxt
->event_callback_eip
= (unsigned long)xen_hypervisor_callback
;
267 ctxt
->failsafe_callback_eip
= (unsigned long)xen_failsafe_callback
;
269 per_cpu(xen_cr3
, cpu
) = __pa(swapper_pg_dir
);
270 ctxt
->ctrlreg
[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir
));
272 if (HYPERVISOR_vcpu_op(VCPUOP_initialise
, cpu
, ctxt
))
279 static int __cpuinit
xen_cpu_up(unsigned int cpu
)
281 struct task_struct
*idle
= idle_task(cpu
);
284 per_cpu(current_task
, cpu
) = idle
;
288 clear_tsk_thread_flag(idle
, TIF_FORK
);
289 per_cpu(kernel_stack
, cpu
) =
290 (unsigned long)task_stack_page(idle
) -
291 KERNEL_STACK_OFFSET
+ THREAD_SIZE
;
293 xen_setup_timer(cpu
);
294 xen_init_lock_cpu(cpu
);
296 per_cpu(cpu_state
, cpu
) = CPU_UP_PREPARE
;
298 /* make sure interrupts start blocked */
299 per_cpu(xen_vcpu
, cpu
)->evtchn_upcall_mask
= 1;
301 rc
= cpu_initialize_context(cpu
, idle
);
305 if (num_online_cpus() == 1)
306 alternatives_smp_switch(1);
308 rc
= xen_smp_intr_init(cpu
);
312 rc
= HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
);
315 while(per_cpu(cpu_state
, cpu
) != CPU_ONLINE
) {
316 HYPERVISOR_sched_op(SCHEDOP_yield
, 0);
323 static void xen_smp_cpus_done(unsigned int max_cpus
)
327 #ifdef CONFIG_HOTPLUG_CPU
328 static int xen_cpu_disable(void)
330 unsigned int cpu
= smp_processor_id();
334 cpu_disable_common();
336 load_cr3(swapper_pg_dir
);
340 static void xen_cpu_die(unsigned int cpu
)
342 while (HYPERVISOR_vcpu_op(VCPUOP_is_up
, cpu
, NULL
)) {
343 current
->state
= TASK_UNINTERRUPTIBLE
;
344 schedule_timeout(HZ
/10);
346 unbind_from_irqhandler(per_cpu(resched_irq
, cpu
), NULL
);
347 unbind_from_irqhandler(per_cpu(callfunc_irq
, cpu
), NULL
);
348 unbind_from_irqhandler(per_cpu(debug_irq
, cpu
), NULL
);
349 unbind_from_irqhandler(per_cpu(callfuncsingle_irq
, cpu
), NULL
);
350 xen_uninit_lock_cpu(cpu
);
351 xen_teardown_timer(cpu
);
353 if (num_online_cpus() == 1)
354 alternatives_smp_switch(0);
357 static void __cpuinit
xen_play_dead(void) /* used only with CPU_HOTPLUG */
360 HYPERVISOR_vcpu_op(VCPUOP_down
, smp_processor_id(), NULL
);
364 #else /* !CONFIG_HOTPLUG_CPU */
365 static int xen_cpu_disable(void)
370 static void xen_cpu_die(unsigned int cpu
)
375 static void xen_play_dead(void)
381 static void stop_self(void *v
)
383 int cpu
= smp_processor_id();
385 /* make sure we're not pinning something down */
386 load_cr3(swapper_pg_dir
);
387 /* should set up a minimal gdt */
389 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
);
393 static void xen_smp_send_stop(void)
395 smp_call_function(stop_self
, NULL
, 0);
398 static void xen_smp_send_reschedule(int cpu
)
400 xen_send_IPI_one(cpu
, XEN_RESCHEDULE_VECTOR
);
403 static void xen_send_IPI_mask(const struct cpumask
*mask
,
404 enum ipi_vector vector
)
408 for_each_cpu_and(cpu
, mask
, cpu_online_mask
)
409 xen_send_IPI_one(cpu
, vector
);
412 static void xen_smp_send_call_function_ipi(const struct cpumask
*mask
)
416 xen_send_IPI_mask(mask
, XEN_CALL_FUNCTION_VECTOR
);
418 /* Make sure other vcpus get a chance to run if they need to. */
419 for_each_cpu(cpu
, mask
) {
420 if (xen_vcpu_stolen(cpu
)) {
421 HYPERVISOR_sched_op(SCHEDOP_yield
, 0);
427 static void xen_smp_send_call_function_single_ipi(int cpu
)
429 xen_send_IPI_mask(cpumask_of(cpu
),
430 XEN_CALL_FUNCTION_SINGLE_VECTOR
);
433 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
)
436 generic_smp_call_function_interrupt();
437 inc_irq_stat(irq_call_count
);
443 static irqreturn_t
xen_call_function_single_interrupt(int irq
, void *dev_id
)
446 generic_smp_call_function_single_interrupt();
447 inc_irq_stat(irq_call_count
);
453 static const struct smp_ops xen_smp_ops __initdata
= {
454 .smp_prepare_boot_cpu
= xen_smp_prepare_boot_cpu
,
455 .smp_prepare_cpus
= xen_smp_prepare_cpus
,
456 .smp_cpus_done
= xen_smp_cpus_done
,
458 .cpu_up
= xen_cpu_up
,
459 .cpu_die
= xen_cpu_die
,
460 .cpu_disable
= xen_cpu_disable
,
461 .play_dead
= xen_play_dead
,
463 .smp_send_stop
= xen_smp_send_stop
,
464 .smp_send_reschedule
= xen_smp_send_reschedule
,
466 .send_call_func_ipi
= xen_smp_send_call_function_ipi
,
467 .send_call_func_single_ipi
= xen_smp_send_call_function_single_ipi
,
470 void __init
xen_smp_init(void)
472 smp_ops
= xen_smp_ops
;
473 xen_fill_possible_map();
474 xen_init_spinlocks();