Commit | Line | Data |
---|---|---|
f87e4cac JF |
1 | /* |
2 | * Xen SMP support | |
3 | * | |
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. | |
7 | * | |
8 | * IPIs are handled through the Xen event mechanism. | |
9 | * | |
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 | |
13 | * single-threaded. | |
f87e4cac JF |
14 | */ |
15 | #include <linux/sched.h> | |
16 | #include <linux/err.h> | |
5a0e3ad6 | 17 | #include <linux/slab.h> |
f87e4cac | 18 | #include <linux/smp.h> |
1ff2b0c3 | 19 | #include <linux/irq_work.h> |
466318a8 | 20 | #include <linux/tick.h> |
f87e4cac JF |
21 | |
22 | #include <asm/paravirt.h> | |
23 | #include <asm/desc.h> | |
24 | #include <asm/pgtable.h> | |
25 | #include <asm/cpu.h> | |
26 | ||
27 | #include <xen/interface/xen.h> | |
28 | #include <xen/interface/vcpu.h> | |
29 | ||
30 | #include <asm/xen/interface.h> | |
31 | #include <asm/xen/hypercall.h> | |
32 | ||
ea5b8f73 | 33 | #include <xen/xen.h> |
f87e4cac JF |
34 | #include <xen/page.h> |
35 | #include <xen/events.h> | |
36 | ||
ed467e69 | 37 | #include <xen/hvc-console.h> |
f87e4cac JF |
38 | #include "xen-ops.h" |
39 | #include "mmu.h" | |
40 | ||
b78936e1 | 41 | cpumask_var_t xen_cpu_initialized_map; |
f87e4cac | 42 | |
c6e22f9e TH |
43 | static DEFINE_PER_CPU(int, xen_resched_irq); |
44 | static DEFINE_PER_CPU(int, xen_callfunc_irq); | |
45 | static DEFINE_PER_CPU(int, xen_callfuncsingle_irq); | |
1ff2b0c3 | 46 | static DEFINE_PER_CPU(int, xen_irq_work); |
c6e22f9e | 47 | static DEFINE_PER_CPU(int, xen_debug_irq) = -1; |
f87e4cac JF |
48 | |
49 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id); | |
3b16cf87 | 50 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id); |
1ff2b0c3 | 51 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id); |
f87e4cac JF |
52 | |
53 | /* | |
184748cc | 54 | * Reschedule call back. |
f87e4cac JF |
55 | */ |
56 | static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id) | |
57 | { | |
1b437c8c | 58 | inc_irq_stat(irq_resched_count); |
184748cc | 59 | scheduler_ipi(); |
38bb5ab4 | 60 | |
f87e4cac JF |
61 | return IRQ_HANDLED; |
62 | } | |
63 | ||
b53cedeb | 64 | static void __cpuinit cpu_bringup(void) |
f87e4cac | 65 | { |
e8c9e788 | 66 | int cpu; |
f87e4cac JF |
67 | |
68 | cpu_init(); | |
d68d82af | 69 | touch_softlockup_watchdog(); |
c7b75947 JF |
70 | preempt_disable(); |
71 | ||
e2a81baf | 72 | xen_enable_sysenter(); |
6fcac6d3 | 73 | xen_enable_syscall(); |
f87e4cac | 74 | |
c7b75947 JF |
75 | cpu = smp_processor_id(); |
76 | smp_store_cpu_info(cpu); | |
77 | cpu_data(cpu).x86_max_cores = 1; | |
78 | set_cpu_sibling_map(cpu); | |
f87e4cac JF |
79 | |
80 | xen_setup_cpu_clockevents(); | |
81 | ||
106b4438 KRW |
82 | notify_cpu_starting(cpu); |
83 | ||
d7d3756c | 84 | set_cpu_online(cpu, true); |
106b4438 | 85 | |
2113f469 | 86 | this_cpu_write(cpu_state, CPU_ONLINE); |
106b4438 | 87 | |
c7b75947 JF |
88 | wmb(); |
89 | ||
f87e4cac JF |
90 | /* We can take interrupts now: we're officially "up". */ |
91 | local_irq_enable(); | |
92 | ||
93 | wmb(); /* make sure everything is out */ | |
d68d82af AN |
94 | } |
95 | ||
b53cedeb | 96 | static void __cpuinit cpu_bringup_and_idle(void) |
d68d82af AN |
97 | { |
98 | cpu_bringup(); | |
7d1a9417 | 99 | cpu_startup_entry(CPUHP_ONLINE); |
f87e4cac JF |
100 | } |
101 | ||
102 | static int xen_smp_intr_init(unsigned int cpu) | |
103 | { | |
104 | int rc; | |
ee523ca1 | 105 | const char *resched_name, *callfunc_name, *debug_name; |
f87e4cac JF |
106 | |
107 | resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu); | |
108 | rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR, | |
109 | cpu, | |
110 | xen_reschedule_interrupt, | |
111 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
112 | resched_name, | |
113 | NULL); | |
114 | if (rc < 0) | |
115 | goto fail; | |
c6e22f9e | 116 | per_cpu(xen_resched_irq, cpu) = rc; |
f87e4cac JF |
117 | |
118 | callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu); | |
119 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR, | |
120 | cpu, | |
121 | xen_call_function_interrupt, | |
122 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
123 | callfunc_name, | |
124 | NULL); | |
125 | if (rc < 0) | |
126 | goto fail; | |
c6e22f9e | 127 | per_cpu(xen_callfunc_irq, cpu) = rc; |
f87e4cac | 128 | |
ee523ca1 JF |
129 | debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu); |
130 | rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt, | |
131 | IRQF_DISABLED | IRQF_PERCPU | IRQF_NOBALANCING, | |
132 | debug_name, NULL); | |
133 | if (rc < 0) | |
134 | goto fail; | |
c6e22f9e | 135 | per_cpu(xen_debug_irq, cpu) = rc; |
ee523ca1 | 136 | |
3b16cf87 JA |
137 | callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu); |
138 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR, | |
139 | cpu, | |
140 | xen_call_function_single_interrupt, | |
141 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
142 | callfunc_name, | |
143 | NULL); | |
144 | if (rc < 0) | |
145 | goto fail; | |
c6e22f9e | 146 | per_cpu(xen_callfuncsingle_irq, cpu) = rc; |
3b16cf87 | 147 | |
27d8b207 KRW |
148 | /* |
149 | * The IRQ worker on PVHVM goes through the native path and uses the | |
150 | * IPI mechanism. | |
151 | */ | |
152 | if (xen_hvm_domain()) | |
153 | return 0; | |
154 | ||
1ff2b0c3 LM |
155 | callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu); |
156 | rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR, | |
157 | cpu, | |
158 | xen_irq_work_interrupt, | |
159 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
160 | callfunc_name, | |
161 | NULL); | |
162 | if (rc < 0) | |
163 | goto fail; | |
164 | per_cpu(xen_irq_work, cpu) = rc; | |
165 | ||
f87e4cac JF |
166 | return 0; |
167 | ||
168 | fail: | |
c6e22f9e TH |
169 | if (per_cpu(xen_resched_irq, cpu) >= 0) |
170 | unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL); | |
171 | if (per_cpu(xen_callfunc_irq, cpu) >= 0) | |
172 | unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL); | |
173 | if (per_cpu(xen_debug_irq, cpu) >= 0) | |
174 | unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL); | |
175 | if (per_cpu(xen_callfuncsingle_irq, cpu) >= 0) | |
176 | unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), | |
177 | NULL); | |
27d8b207 KRW |
178 | if (xen_hvm_domain()) |
179 | return rc; | |
180 | ||
1ff2b0c3 LM |
181 | if (per_cpu(xen_irq_work, cpu) >= 0) |
182 | unbind_from_irqhandler(per_cpu(xen_irq_work, cpu), NULL); | |
3b16cf87 | 183 | |
f87e4cac JF |
184 | return rc; |
185 | } | |
186 | ||
c7b75947 | 187 | static void __init xen_fill_possible_map(void) |
f87e4cac JF |
188 | { |
189 | int i, rc; | |
190 | ||
ea5b8f73 SS |
191 | if (xen_initial_domain()) |
192 | return; | |
193 | ||
194 | for (i = 0; i < nr_cpu_ids; i++) { | |
195 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); | |
196 | if (rc >= 0) { | |
197 | num_processors++; | |
198 | set_cpu_possible(i, true); | |
199 | } | |
200 | } | |
201 | } | |
202 | ||
203 | static void __init xen_filter_cpu_maps(void) | |
204 | { | |
205 | int i, rc; | |
cf405ae6 | 206 | unsigned int subtract = 0; |
ea5b8f73 SS |
207 | |
208 | if (!xen_initial_domain()) | |
209 | return; | |
210 | ||
801fd14a SS |
211 | num_processors = 0; |
212 | disabled_cpus = 0; | |
e7986739 | 213 | for (i = 0; i < nr_cpu_ids; i++) { |
f87e4cac | 214 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); |
4560a294 JF |
215 | if (rc >= 0) { |
216 | num_processors++; | |
4f062896 | 217 | set_cpu_possible(i, true); |
801fd14a SS |
218 | } else { |
219 | set_cpu_possible(i, false); | |
220 | set_cpu_present(i, false); | |
cf405ae6 | 221 | subtract++; |
4560a294 | 222 | } |
f87e4cac | 223 | } |
cf405ae6 KRW |
224 | #ifdef CONFIG_HOTPLUG_CPU |
225 | /* This is akin to using 'nr_cpus' on the Linux command line. | |
226 | * Which is OK as when we use 'dom0_max_vcpus=X' we can only | |
227 | * have up to X, while nr_cpu_ids is greater than X. This | |
228 | * normally is not a problem, except when CPU hotplugging | |
229 | * is involved and then there might be more than X CPUs | |
230 | * in the guest - which will not work as there is no | |
231 | * hypercall to expand the max number of VCPUs an already | |
232 | * running guest has. So cap it up to X. */ | |
233 | if (subtract) | |
234 | nr_cpu_ids = nr_cpu_ids - subtract; | |
235 | #endif | |
236 | ||
f87e4cac JF |
237 | } |
238 | ||
a9e7062d | 239 | static void __init xen_smp_prepare_boot_cpu(void) |
f87e4cac | 240 | { |
f87e4cac JF |
241 | BUG_ON(smp_processor_id() != 0); |
242 | native_smp_prepare_boot_cpu(); | |
243 | ||
f87e4cac JF |
244 | /* We've switched to the "real" per-cpu gdt, so make sure the |
245 | old memory can be recycled */ | |
38341432 | 246 | make_lowmem_page_readwrite(xen_initial_gdt); |
60223a32 | 247 | |
119c1d25 FZ |
248 | #ifdef CONFIG_X86_32 |
249 | /* | |
250 | * Xen starts us with XEN_FLAT_RING1_DS, but linux code | |
251 | * expects __USER_DS | |
252 | */ | |
253 | loadsegment(ds, __USER_DS); | |
254 | loadsegment(es, __USER_DS); | |
255 | #endif | |
256 | ||
ea5b8f73 | 257 | xen_filter_cpu_maps(); |
60223a32 | 258 | xen_setup_vcpu_info_placement(); |
f87e4cac JF |
259 | } |
260 | ||
a9e7062d | 261 | static void __init xen_smp_prepare_cpus(unsigned int max_cpus) |
f87e4cac JF |
262 | { |
263 | unsigned cpu; | |
900cba88 | 264 | unsigned int i; |
f87e4cac | 265 | |
ed467e69 KRW |
266 | if (skip_ioapic_setup) { |
267 | char *m = (max_cpus == 0) ? | |
268 | "The nosmp parameter is incompatible with Xen; " \ | |
269 | "use Xen dom0_max_vcpus=1 parameter" : | |
270 | "The noapic parameter is incompatible with Xen"; | |
271 | ||
272 | xen_raw_printk(m); | |
273 | panic(m); | |
274 | } | |
2d9e1e2f JF |
275 | xen_init_lock_cpu(0); |
276 | ||
06d0b5d9 | 277 | smp_store_boot_cpu_info(); |
c7b75947 | 278 | cpu_data(0).x86_max_cores = 1; |
900cba88 AJ |
279 | |
280 | for_each_possible_cpu(i) { | |
281 | zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL); | |
282 | zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL); | |
283 | zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL); | |
284 | } | |
f87e4cac JF |
285 | set_cpu_sibling_map(0); |
286 | ||
287 | if (xen_smp_intr_init(0)) | |
288 | BUG(); | |
289 | ||
b78936e1 MT |
290 | if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL)) |
291 | panic("could not allocate xen_cpu_initialized_map\n"); | |
292 | ||
293 | cpumask_copy(xen_cpu_initialized_map, cpumask_of(0)); | |
f87e4cac JF |
294 | |
295 | /* Restrict the possible_map according to max_cpus. */ | |
296 | while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) { | |
e7986739 | 297 | for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--) |
f87e4cac | 298 | continue; |
4f062896 | 299 | set_cpu_possible(cpu, false); |
f87e4cac JF |
300 | } |
301 | ||
7eb43a6d | 302 | for_each_possible_cpu(cpu) |
4f062896 | 303 | set_cpu_present(cpu, true); |
f87e4cac JF |
304 | } |
305 | ||
b53cedeb | 306 | static int __cpuinit |
f87e4cac JF |
307 | cpu_initialize_context(unsigned int cpu, struct task_struct *idle) |
308 | { | |
309 | struct vcpu_guest_context *ctxt; | |
c7b75947 | 310 | struct desc_struct *gdt; |
9976b39b | 311 | unsigned long gdt_mfn; |
f87e4cac | 312 | |
b78936e1 | 313 | if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map)) |
f87e4cac JF |
314 | return 0; |
315 | ||
316 | ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); | |
317 | if (ctxt == NULL) | |
318 | return -ENOMEM; | |
319 | ||
c7b75947 JF |
320 | gdt = get_cpu_gdt_table(cpu); |
321 | ||
f87e4cac | 322 | ctxt->flags = VGCF_IN_KERNEL; |
f87e4cac | 323 | ctxt->user_regs.ss = __KERNEL_DS; |
c7b75947 JF |
324 | #ifdef CONFIG_X86_32 |
325 | ctxt->user_regs.fs = __KERNEL_PERCPU; | |
577eebea | 326 | ctxt->user_regs.gs = __KERNEL_STACK_CANARY; |
795f99b6 JF |
327 | #else |
328 | ctxt->gs_base_kernel = per_cpu_offset(cpu); | |
c7b75947 | 329 | #endif |
f87e4cac | 330 | ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle; |
f87e4cac JF |
331 | |
332 | memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt)); | |
333 | ||
dacd45f4 KRW |
334 | { |
335 | ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */ | |
336 | ctxt->user_regs.ds = __USER_DS; | |
337 | ctxt->user_regs.es = __USER_DS; | |
f87e4cac | 338 | |
dacd45f4 | 339 | xen_copy_trap_info(ctxt->trap_ctxt); |
f87e4cac | 340 | |
dacd45f4 | 341 | ctxt->ldt_ents = 0; |
9976b39b | 342 | |
dacd45f4 | 343 | BUG_ON((unsigned long)gdt & ~PAGE_MASK); |
f87e4cac | 344 | |
dacd45f4 KRW |
345 | gdt_mfn = arbitrary_virt_to_mfn(gdt); |
346 | make_lowmem_page_readonly(gdt); | |
347 | make_lowmem_page_readonly(mfn_to_virt(gdt_mfn)); | |
f87e4cac | 348 | |
dacd45f4 KRW |
349 | ctxt->gdt_frames[0] = gdt_mfn; |
350 | ctxt->gdt_ents = GDT_ENTRIES; | |
f87e4cac | 351 | |
dacd45f4 KRW |
352 | ctxt->kernel_ss = __KERNEL_DS; |
353 | ctxt->kernel_sp = idle->thread.sp0; | |
f87e4cac | 354 | |
c7b75947 | 355 | #ifdef CONFIG_X86_32 |
dacd45f4 KRW |
356 | ctxt->event_callback_cs = __KERNEL_CS; |
357 | ctxt->failsafe_callback_cs = __KERNEL_CS; | |
c7b75947 | 358 | #endif |
dacd45f4 KRW |
359 | ctxt->event_callback_eip = |
360 | (unsigned long)xen_hypervisor_callback; | |
361 | ctxt->failsafe_callback_eip = | |
362 | (unsigned long)xen_failsafe_callback; | |
363 | } | |
364 | ctxt->user_regs.cs = __KERNEL_CS; | |
365 | ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs); | |
f87e4cac JF |
366 | |
367 | per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir); | |
368 | ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir)); | |
369 | ||
370 | if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt)) | |
371 | BUG(); | |
372 | ||
373 | kfree(ctxt); | |
374 | return 0; | |
375 | } | |
376 | ||
7eb43a6d | 377 | static int __cpuinit xen_cpu_up(unsigned int cpu, struct task_struct *idle) |
f87e4cac | 378 | { |
f87e4cac JF |
379 | int rc; |
380 | ||
c6f5e0ac | 381 | per_cpu(current_task, cpu) = idle; |
c7b75947 | 382 | #ifdef CONFIG_X86_32 |
f87e4cac | 383 | irq_ctx_init(cpu); |
c7b75947 | 384 | #else |
c7b75947 | 385 | clear_tsk_thread_flag(idle, TIF_FORK); |
38341432 JF |
386 | per_cpu(kernel_stack, cpu) = |
387 | (unsigned long)task_stack_page(idle) - | |
388 | KERNEL_STACK_OFFSET + THREAD_SIZE; | |
c7b75947 | 389 | #endif |
02889672 | 390 | xen_setup_runstate_info(cpu); |
f87e4cac | 391 | xen_setup_timer(cpu); |
2d9e1e2f | 392 | xen_init_lock_cpu(cpu); |
f87e4cac | 393 | |
c7b75947 JF |
394 | per_cpu(cpu_state, cpu) = CPU_UP_PREPARE; |
395 | ||
f87e4cac JF |
396 | /* make sure interrupts start blocked */ |
397 | per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1; | |
398 | ||
399 | rc = cpu_initialize_context(cpu, idle); | |
400 | if (rc) | |
401 | return rc; | |
402 | ||
403 | if (num_online_cpus() == 1) | |
816afe4f RR |
404 | /* Just in case we booted with a single CPU. */ |
405 | alternatives_enable_smp(); | |
f87e4cac JF |
406 | |
407 | rc = xen_smp_intr_init(cpu); | |
408 | if (rc) | |
409 | return rc; | |
410 | ||
f87e4cac JF |
411 | rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL); |
412 | BUG_ON(rc); | |
413 | ||
c7b75947 | 414 | while(per_cpu(cpu_state, cpu) != CPU_ONLINE) { |
1207cf8e | 415 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
c7b75947 JF |
416 | barrier(); |
417 | } | |
418 | ||
f87e4cac JF |
419 | return 0; |
420 | } | |
421 | ||
a9e7062d | 422 | static void xen_smp_cpus_done(unsigned int max_cpus) |
f87e4cac JF |
423 | { |
424 | } | |
425 | ||
2737146b | 426 | #ifdef CONFIG_HOTPLUG_CPU |
26fd1051 | 427 | static int xen_cpu_disable(void) |
d68d82af AN |
428 | { |
429 | unsigned int cpu = smp_processor_id(); | |
430 | if (cpu == 0) | |
431 | return -EBUSY; | |
432 | ||
433 | cpu_disable_common(); | |
434 | ||
435 | load_cr3(swapper_pg_dir); | |
436 | return 0; | |
437 | } | |
438 | ||
26fd1051 | 439 | static void xen_cpu_die(unsigned int cpu) |
d68d82af | 440 | { |
b12abaa1 | 441 | while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) { |
d68d82af AN |
442 | current->state = TASK_UNINTERRUPTIBLE; |
443 | schedule_timeout(HZ/10); | |
444 | } | |
c6e22f9e TH |
445 | unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL); |
446 | unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL); | |
447 | unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL); | |
448 | unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL); | |
b12abaa1 KRW |
449 | if (!xen_hvm_domain()) |
450 | unbind_from_irqhandler(per_cpu(xen_irq_work, cpu), NULL); | |
d68d82af AN |
451 | xen_uninit_lock_cpu(cpu); |
452 | xen_teardown_timer(cpu); | |
d68d82af AN |
453 | } |
454 | ||
71709247 | 455 | static void __cpuinit xen_play_dead(void) /* used only with HOTPLUG_CPU */ |
d68d82af AN |
456 | { |
457 | play_dead_common(); | |
458 | HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); | |
459 | cpu_bringup(); | |
466318a8 KRW |
460 | /* |
461 | * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down) | |
462 | * clears certain data that the cpu_idle loop (which called us | |
463 | * and that we return from) expects. The only way to get that | |
464 | * data back is to call: | |
465 | */ | |
466 | tick_nohz_idle_enter(); | |
d68d82af AN |
467 | } |
468 | ||
2737146b | 469 | #else /* !CONFIG_HOTPLUG_CPU */ |
26fd1051 | 470 | static int xen_cpu_disable(void) |
2737146b AN |
471 | { |
472 | return -ENOSYS; | |
473 | } | |
474 | ||
26fd1051 | 475 | static void xen_cpu_die(unsigned int cpu) |
2737146b AN |
476 | { |
477 | BUG(); | |
478 | } | |
479 | ||
26fd1051 | 480 | static void xen_play_dead(void) |
2737146b AN |
481 | { |
482 | BUG(); | |
483 | } | |
484 | ||
485 | #endif | |
f87e4cac JF |
486 | static void stop_self(void *v) |
487 | { | |
488 | int cpu = smp_processor_id(); | |
489 | ||
490 | /* make sure we're not pinning something down */ | |
491 | load_cr3(swapper_pg_dir); | |
492 | /* should set up a minimal gdt */ | |
493 | ||
086748e5 IC |
494 | set_cpu_online(cpu, false); |
495 | ||
f87e4cac JF |
496 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL); |
497 | BUG(); | |
498 | } | |
499 | ||
76fac077 | 500 | static void xen_stop_other_cpus(int wait) |
f87e4cac | 501 | { |
76fac077 | 502 | smp_call_function(stop_self, NULL, wait); |
f87e4cac JF |
503 | } |
504 | ||
a9e7062d | 505 | static void xen_smp_send_reschedule(int cpu) |
f87e4cac JF |
506 | { |
507 | xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR); | |
508 | } | |
509 | ||
f447d56d BG |
510 | static void __xen_send_IPI_mask(const struct cpumask *mask, |
511 | int vector) | |
f87e4cac JF |
512 | { |
513 | unsigned cpu; | |
514 | ||
bcda016e | 515 | for_each_cpu_and(cpu, mask, cpu_online_mask) |
f87e4cac JF |
516 | xen_send_IPI_one(cpu, vector); |
517 | } | |
518 | ||
bcda016e | 519 | static void xen_smp_send_call_function_ipi(const struct cpumask *mask) |
3b16cf87 JA |
520 | { |
521 | int cpu; | |
522 | ||
f447d56d | 523 | __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR); |
3b16cf87 JA |
524 | |
525 | /* Make sure other vcpus get a chance to run if they need to. */ | |
bcda016e | 526 | for_each_cpu(cpu, mask) { |
3b16cf87 | 527 | if (xen_vcpu_stolen(cpu)) { |
1207cf8e | 528 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
3b16cf87 JA |
529 | break; |
530 | } | |
531 | } | |
532 | } | |
533 | ||
a9e7062d | 534 | static void xen_smp_send_call_function_single_ipi(int cpu) |
3b16cf87 | 535 | { |
f447d56d | 536 | __xen_send_IPI_mask(cpumask_of(cpu), |
e7986739 | 537 | XEN_CALL_FUNCTION_SINGLE_VECTOR); |
3b16cf87 JA |
538 | } |
539 | ||
f447d56d BG |
540 | static inline int xen_map_vector(int vector) |
541 | { | |
542 | int xen_vector; | |
543 | ||
544 | switch (vector) { | |
545 | case RESCHEDULE_VECTOR: | |
546 | xen_vector = XEN_RESCHEDULE_VECTOR; | |
547 | break; | |
548 | case CALL_FUNCTION_VECTOR: | |
549 | xen_vector = XEN_CALL_FUNCTION_VECTOR; | |
550 | break; | |
551 | case CALL_FUNCTION_SINGLE_VECTOR: | |
552 | xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR; | |
553 | break; | |
1ff2b0c3 LM |
554 | case IRQ_WORK_VECTOR: |
555 | xen_vector = XEN_IRQ_WORK_VECTOR; | |
556 | break; | |
f447d56d BG |
557 | default: |
558 | xen_vector = -1; | |
559 | printk(KERN_ERR "xen: vector 0x%x is not implemented\n", | |
560 | vector); | |
561 | } | |
562 | ||
563 | return xen_vector; | |
564 | } | |
565 | ||
566 | void xen_send_IPI_mask(const struct cpumask *mask, | |
567 | int vector) | |
568 | { | |
569 | int xen_vector = xen_map_vector(vector); | |
570 | ||
571 | if (xen_vector >= 0) | |
572 | __xen_send_IPI_mask(mask, xen_vector); | |
573 | } | |
574 | ||
575 | void xen_send_IPI_all(int vector) | |
576 | { | |
577 | int xen_vector = xen_map_vector(vector); | |
578 | ||
579 | if (xen_vector >= 0) | |
580 | __xen_send_IPI_mask(cpu_online_mask, xen_vector); | |
581 | } | |
582 | ||
583 | void xen_send_IPI_self(int vector) | |
584 | { | |
585 | int xen_vector = xen_map_vector(vector); | |
586 | ||
587 | if (xen_vector >= 0) | |
588 | xen_send_IPI_one(smp_processor_id(), xen_vector); | |
589 | } | |
590 | ||
591 | void xen_send_IPI_mask_allbutself(const struct cpumask *mask, | |
592 | int vector) | |
593 | { | |
594 | unsigned cpu; | |
595 | unsigned int this_cpu = smp_processor_id(); | |
1db01b49 | 596 | int xen_vector = xen_map_vector(vector); |
f447d56d | 597 | |
1db01b49 | 598 | if (!(num_online_cpus() > 1) || (xen_vector < 0)) |
f447d56d BG |
599 | return; |
600 | ||
601 | for_each_cpu_and(cpu, mask, cpu_online_mask) { | |
602 | if (this_cpu == cpu) | |
603 | continue; | |
604 | ||
1db01b49 | 605 | xen_send_IPI_one(cpu, xen_vector); |
f447d56d BG |
606 | } |
607 | } | |
608 | ||
609 | void xen_send_IPI_allbutself(int vector) | |
610 | { | |
1db01b49 | 611 | xen_send_IPI_mask_allbutself(cpu_online_mask, vector); |
f447d56d BG |
612 | } |
613 | ||
f87e4cac JF |
614 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id) |
615 | { | |
f87e4cac | 616 | irq_enter(); |
3b16cf87 | 617 | generic_smp_call_function_interrupt(); |
1b437c8c | 618 | inc_irq_stat(irq_call_count); |
f87e4cac JF |
619 | irq_exit(); |
620 | ||
f87e4cac JF |
621 | return IRQ_HANDLED; |
622 | } | |
623 | ||
3b16cf87 | 624 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id) |
f87e4cac | 625 | { |
3b16cf87 JA |
626 | irq_enter(); |
627 | generic_smp_call_function_single_interrupt(); | |
1b437c8c | 628 | inc_irq_stat(irq_call_count); |
3b16cf87 | 629 | irq_exit(); |
f87e4cac | 630 | |
3b16cf87 | 631 | return IRQ_HANDLED; |
f87e4cac | 632 | } |
a9e7062d | 633 | |
1ff2b0c3 LM |
634 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id) |
635 | { | |
636 | irq_enter(); | |
637 | irq_work_run(); | |
638 | inc_irq_stat(apic_irq_work_irqs); | |
639 | irq_exit(); | |
640 | ||
641 | return IRQ_HANDLED; | |
642 | } | |
643 | ||
b53cedeb | 644 | static const struct smp_ops xen_smp_ops __initconst = { |
a9e7062d JF |
645 | .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, |
646 | .smp_prepare_cpus = xen_smp_prepare_cpus, | |
a9e7062d JF |
647 | .smp_cpus_done = xen_smp_cpus_done, |
648 | ||
d68d82af AN |
649 | .cpu_up = xen_cpu_up, |
650 | .cpu_die = xen_cpu_die, | |
651 | .cpu_disable = xen_cpu_disable, | |
652 | .play_dead = xen_play_dead, | |
653 | ||
76fac077 | 654 | .stop_other_cpus = xen_stop_other_cpus, |
a9e7062d JF |
655 | .smp_send_reschedule = xen_smp_send_reschedule, |
656 | ||
657 | .send_call_func_ipi = xen_smp_send_call_function_ipi, | |
658 | .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi, | |
659 | }; | |
660 | ||
661 | void __init xen_smp_init(void) | |
662 | { | |
663 | smp_ops = xen_smp_ops; | |
c7b75947 | 664 | xen_fill_possible_map(); |
2d9e1e2f | 665 | xen_init_spinlocks(); |
a9e7062d | 666 | } |
99bbb3a8 SS |
667 | |
668 | static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus) | |
669 | { | |
670 | native_smp_prepare_cpus(max_cpus); | |
671 | WARN_ON(xen_smp_intr_init(0)); | |
672 | ||
99bbb3a8 | 673 | xen_init_lock_cpu(0); |
99bbb3a8 SS |
674 | } |
675 | ||
5cdaf183 | 676 | static int __cpuinit xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle) |
99bbb3a8 SS |
677 | { |
678 | int rc; | |
51f50872 CA |
679 | /* |
680 | * xen_smp_intr_init() needs to run before native_cpu_up() | |
681 | * so that IPI vectors are set up on the booting CPU before | |
682 | * it is marked online in native_cpu_up(). | |
683 | */ | |
684 | rc = xen_smp_intr_init(cpu); | |
685 | WARN_ON(rc); | |
686 | if (!rc) | |
687 | rc = native_cpu_up(cpu, tidle); | |
99bbb3a8 SS |
688 | return rc; |
689 | } | |
690 | ||
691 | static void xen_hvm_cpu_die(unsigned int cpu) | |
692 | { | |
b12abaa1 | 693 | xen_cpu_die(cpu); |
99bbb3a8 SS |
694 | native_cpu_die(cpu); |
695 | } | |
696 | ||
697 | void __init xen_hvm_smp_init(void) | |
698 | { | |
3c05c4be SS |
699 | if (!xen_have_vector_callback) |
700 | return; | |
99bbb3a8 SS |
701 | smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus; |
702 | smp_ops.smp_send_reschedule = xen_smp_send_reschedule; | |
703 | smp_ops.cpu_up = xen_hvm_cpu_up; | |
704 | smp_ops.cpu_die = xen_hvm_cpu_die; | |
705 | smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi; | |
706 | smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi; | |
707 | } |