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 * This does not handle HOTPLUG_CPU yet.
17 #include <linux/sched.h>
18 #include <linux/err.h>
19 #include <linux/smp.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>
33 #include <xen/events.h>
38 static cpumask_t xen_cpu_initialized_map
;
39 static DEFINE_PER_CPU(int, resched_irq
) = -1;
40 static DEFINE_PER_CPU(int, callfunc_irq
) = -1;
41 static DEFINE_PER_CPU(int, debug_irq
) = -1;
44 * Structure and data for smp_call_function(). This is designed to minimise
45 * static memory requirements. It also looks cleaner.
47 static DEFINE_SPINLOCK(call_lock
);
49 struct call_data_struct
{
50 void (*func
) (void *info
);
57 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
);
59 static struct call_data_struct
*call_data
;
62 * Reschedule call back. Nothing to do,
63 * all the work is done automatically when
64 * we return from the interrupt.
66 static irqreturn_t
xen_reschedule_interrupt(int irq
, void *dev_id
)
71 static __cpuinit
void cpu_bringup_and_idle(void)
73 int cpu
= smp_processor_id();
76 xen_enable_sysenter();
79 per_cpu(cpu_state
, cpu
) = CPU_ONLINE
;
81 xen_setup_cpu_clockevents();
83 /* We can take interrupts now: we're officially "up". */
86 wmb(); /* make sure everything is out */
90 static int xen_smp_intr_init(unsigned int cpu
)
93 const char *resched_name
, *callfunc_name
, *debug_name
;
95 resched_name
= kasprintf(GFP_KERNEL
, "resched%d", cpu
);
96 rc
= bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR
,
98 xen_reschedule_interrupt
,
99 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
104 per_cpu(resched_irq
, cpu
) = rc
;
106 callfunc_name
= kasprintf(GFP_KERNEL
, "callfunc%d", cpu
);
107 rc
= bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR
,
109 xen_call_function_interrupt
,
110 IRQF_DISABLED
|IRQF_PERCPU
|IRQF_NOBALANCING
,
115 per_cpu(callfunc_irq
, cpu
) = rc
;
117 debug_name
= kasprintf(GFP_KERNEL
, "debug%d", cpu
);
118 rc
= bind_virq_to_irqhandler(VIRQ_DEBUG
, cpu
, xen_debug_interrupt
,
119 IRQF_DISABLED
| IRQF_PERCPU
| IRQF_NOBALANCING
,
123 per_cpu(debug_irq
, cpu
) = rc
;
128 if (per_cpu(resched_irq
, cpu
) >= 0)
129 unbind_from_irqhandler(per_cpu(resched_irq
, cpu
), NULL
);
130 if (per_cpu(callfunc_irq
, cpu
) >= 0)
131 unbind_from_irqhandler(per_cpu(callfunc_irq
, cpu
), NULL
);
132 if (per_cpu(debug_irq
, cpu
) >= 0)
133 unbind_from_irqhandler(per_cpu(debug_irq
, cpu
), NULL
);
137 void __init
xen_fill_possible_map(void)
141 for (i
= 0; i
< NR_CPUS
; i
++) {
142 rc
= HYPERVISOR_vcpu_op(VCPUOP_is_up
, i
, NULL
);
144 cpu_set(i
, cpu_possible_map
);
148 void __init
xen_smp_prepare_boot_cpu(void)
152 BUG_ON(smp_processor_id() != 0);
153 native_smp_prepare_boot_cpu();
155 /* We've switched to the "real" per-cpu gdt, so make sure the
156 old memory can be recycled */
157 make_lowmem_page_readwrite(&per_cpu__gdt_page
);
159 for_each_possible_cpu(cpu
) {
160 cpus_clear(per_cpu(cpu_sibling_map
, cpu
));
162 * cpu_core_map lives in a per cpu area that is cleared
163 * when the per cpu array is allocated.
165 * cpus_clear(per_cpu(cpu_core_map, cpu));
169 xen_setup_vcpu_info_placement();
172 void __init
xen_smp_prepare_cpus(unsigned int max_cpus
)
176 for_each_possible_cpu(cpu
) {
177 cpus_clear(per_cpu(cpu_sibling_map
, cpu
));
179 * cpu_core_ map will be zeroed when the per
180 * cpu area is allocated.
182 * cpus_clear(per_cpu(cpu_core_map, cpu));
186 smp_store_cpu_info(0);
187 set_cpu_sibling_map(0);
189 if (xen_smp_intr_init(0))
192 xen_cpu_initialized_map
= cpumask_of_cpu(0);
194 /* Restrict the possible_map according to max_cpus. */
195 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus
)) {
196 for (cpu
= NR_CPUS
- 1; !cpu_possible(cpu
); cpu
--)
198 cpu_clear(cpu
, cpu_possible_map
);
201 for_each_possible_cpu (cpu
) {
202 struct task_struct
*idle
;
207 idle
= fork_idle(cpu
);
209 panic("failed fork for CPU %d", cpu
);
211 cpu_set(cpu
, cpu_present_map
);
214 //init_xenbus_allowed_cpumask();
218 cpu_initialize_context(unsigned int cpu
, struct task_struct
*idle
)
220 struct vcpu_guest_context
*ctxt
;
221 struct gdt_page
*gdt
= &per_cpu(gdt_page
, cpu
);
223 if (cpu_test_and_set(cpu
, xen_cpu_initialized_map
))
226 ctxt
= kzalloc(sizeof(*ctxt
), GFP_KERNEL
);
230 ctxt
->flags
= VGCF_IN_KERNEL
;
231 ctxt
->user_regs
.ds
= __USER_DS
;
232 ctxt
->user_regs
.es
= __USER_DS
;
233 ctxt
->user_regs
.fs
= __KERNEL_PERCPU
;
234 ctxt
->user_regs
.gs
= 0;
235 ctxt
->user_regs
.ss
= __KERNEL_DS
;
236 ctxt
->user_regs
.eip
= (unsigned long)cpu_bringup_and_idle
;
237 ctxt
->user_regs
.eflags
= 0x1000; /* IOPL_RING1 */
239 memset(&ctxt
->fpu_ctxt
, 0, sizeof(ctxt
->fpu_ctxt
));
241 xen_copy_trap_info(ctxt
->trap_ctxt
);
245 BUG_ON((unsigned long)gdt
->gdt
& ~PAGE_MASK
);
246 make_lowmem_page_readonly(gdt
->gdt
);
248 ctxt
->gdt_frames
[0] = virt_to_mfn(gdt
->gdt
);
249 ctxt
->gdt_ents
= ARRAY_SIZE(gdt
->gdt
);
251 ctxt
->user_regs
.cs
= __KERNEL_CS
;
252 ctxt
->user_regs
.esp
= idle
->thread
.sp0
- sizeof(struct pt_regs
);
254 ctxt
->kernel_ss
= __KERNEL_DS
;
255 ctxt
->kernel_sp
= idle
->thread
.sp0
;
257 ctxt
->event_callback_cs
= __KERNEL_CS
;
258 ctxt
->event_callback_eip
= (unsigned long)xen_hypervisor_callback
;
259 ctxt
->failsafe_callback_cs
= __KERNEL_CS
;
260 ctxt
->failsafe_callback_eip
= (unsigned long)xen_failsafe_callback
;
262 per_cpu(xen_cr3
, cpu
) = __pa(swapper_pg_dir
);
263 ctxt
->ctrlreg
[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir
));
265 if (HYPERVISOR_vcpu_op(VCPUOP_initialise
, cpu
, ctxt
))
272 int __cpuinit
xen_cpu_up(unsigned int cpu
)
274 struct task_struct
*idle
= idle_task(cpu
);
278 rc
= cpu_up_check(cpu
);
284 per_cpu(current_task
, cpu
) = idle
;
286 xen_setup_timer(cpu
);
288 /* make sure interrupts start blocked */
289 per_cpu(xen_vcpu
, cpu
)->evtchn_upcall_mask
= 1;
291 rc
= cpu_initialize_context(cpu
, idle
);
295 if (num_online_cpus() == 1)
296 alternatives_smp_switch(1);
298 rc
= xen_smp_intr_init(cpu
);
302 smp_store_cpu_info(cpu
);
303 set_cpu_sibling_map(cpu
);
304 /* This must be done before setting cpu_online_map */
307 cpu_set(cpu
, cpu_online_map
);
309 rc
= HYPERVISOR_vcpu_op(VCPUOP_up
, cpu
, NULL
);
315 void xen_smp_cpus_done(unsigned int max_cpus
)
319 static void stop_self(void *v
)
321 int cpu
= smp_processor_id();
323 /* make sure we're not pinning something down */
324 load_cr3(swapper_pg_dir
);
325 /* should set up a minimal gdt */
327 HYPERVISOR_vcpu_op(VCPUOP_down
, cpu
, NULL
);
331 void xen_smp_send_stop(void)
333 smp_call_function(stop_self
, NULL
, 0, 0);
336 void xen_smp_send_reschedule(int cpu
)
338 xen_send_IPI_one(cpu
, XEN_RESCHEDULE_VECTOR
);
342 static void xen_send_IPI_mask(cpumask_t mask
, enum ipi_vector vector
)
346 cpus_and(mask
, mask
, cpu_online_map
);
348 for_each_cpu_mask(cpu
, mask
)
349 xen_send_IPI_one(cpu
, vector
);
352 static irqreturn_t
xen_call_function_interrupt(int irq
, void *dev_id
)
354 void (*func
) (void *info
) = call_data
->func
;
355 void *info
= call_data
->info
;
356 int wait
= call_data
->wait
;
359 * Notify initiating CPU that I've grabbed the data and am
360 * about to execute the function
363 atomic_inc(&call_data
->started
);
365 * At this point the info structure may be out of scope unless wait==1
369 __get_cpu_var(irq_stat
).irq_call_count
++;
373 mb(); /* commit everything before setting finished */
374 atomic_inc(&call_data
->finished
);
380 int xen_smp_call_function_mask(cpumask_t mask
, void (*func
)(void *),
381 void *info
, int wait
)
383 struct call_data_struct data
;
387 /* Holding any lock stops cpus from going down. */
388 spin_lock(&call_lock
);
390 cpu_clear(smp_processor_id(), mask
);
392 cpus
= cpus_weight(mask
);
394 spin_unlock(&call_lock
);
398 /* Can deadlock when called with interrupts disabled */
399 WARN_ON(irqs_disabled());
403 atomic_set(&data
.started
, 0);
406 atomic_set(&data
.finished
, 0);
409 mb(); /* write everything before IPI */
411 /* Send a message to other CPUs and wait for them to respond */
412 xen_send_IPI_mask(mask
, XEN_CALL_FUNCTION_VECTOR
);
414 /* Make sure other vcpus get a chance to run if they need to. */
416 for_each_cpu_mask(cpu
, mask
)
417 if (xen_vcpu_stolen(cpu
))
421 HYPERVISOR_sched_op(SCHEDOP_yield
, 0);
423 /* Wait for response */
424 while (atomic_read(&data
.started
) != cpus
||
425 (wait
&& atomic_read(&data
.finished
) != cpus
))
428 spin_unlock(&call_lock
);