]>
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> |
f87e4cac JF |
20 | |
21 | #include <asm/paravirt.h> | |
22 | #include <asm/desc.h> | |
23 | #include <asm/pgtable.h> | |
24 | #include <asm/cpu.h> | |
25 | ||
26 | #include <xen/interface/xen.h> | |
27 | #include <xen/interface/vcpu.h> | |
28 | ||
29 | #include <asm/xen/interface.h> | |
30 | #include <asm/xen/hypercall.h> | |
31 | ||
ea5b8f73 | 32 | #include <xen/xen.h> |
f87e4cac JF |
33 | #include <xen/page.h> |
34 | #include <xen/events.h> | |
35 | ||
ed467e69 | 36 | #include <xen/hvc-console.h> |
f87e4cac JF |
37 | #include "xen-ops.h" |
38 | #include "mmu.h" | |
39 | ||
b78936e1 | 40 | cpumask_var_t xen_cpu_initialized_map; |
f87e4cac | 41 | |
9547689f KRW |
42 | struct xen_common_irq { |
43 | int irq; | |
44 | char *name; | |
45 | }; | |
ee336e10 KRW |
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 }; | |
9547689f | 50 | static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 }; |
f87e4cac JF |
51 | |
52 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id); | |
3b16cf87 | 53 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id); |
1ff2b0c3 | 54 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id); |
f87e4cac JF |
55 | |
56 | /* | |
184748cc | 57 | * Reschedule call back. |
f87e4cac JF |
58 | */ |
59 | static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id) | |
60 | { | |
1b437c8c | 61 | inc_irq_stat(irq_resched_count); |
184748cc | 62 | scheduler_ipi(); |
38bb5ab4 | 63 | |
f87e4cac JF |
64 | return IRQ_HANDLED; |
65 | } | |
66 | ||
b53cedeb | 67 | static void __cpuinit cpu_bringup(void) |
f87e4cac | 68 | { |
e8c9e788 | 69 | int cpu; |
f87e4cac JF |
70 | |
71 | cpu_init(); | |
d68d82af | 72 | touch_softlockup_watchdog(); |
c7b75947 JF |
73 | preempt_disable(); |
74 | ||
e2a81baf | 75 | xen_enable_sysenter(); |
6fcac6d3 | 76 | xen_enable_syscall(); |
f87e4cac | 77 | |
c7b75947 JF |
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); | |
f87e4cac JF |
82 | |
83 | xen_setup_cpu_clockevents(); | |
84 | ||
106b4438 KRW |
85 | notify_cpu_starting(cpu); |
86 | ||
d7d3756c | 87 | set_cpu_online(cpu, true); |
106b4438 | 88 | |
2113f469 | 89 | this_cpu_write(cpu_state, CPU_ONLINE); |
106b4438 | 90 | |
c7b75947 JF |
91 | wmb(); |
92 | ||
f87e4cac JF |
93 | /* We can take interrupts now: we're officially "up". */ |
94 | local_irq_enable(); | |
95 | ||
96 | wmb(); /* make sure everything is out */ | |
d68d82af AN |
97 | } |
98 | ||
b53cedeb | 99 | static void __cpuinit cpu_bringup_and_idle(void) |
d68d82af AN |
100 | { |
101 | cpu_bringup(); | |
7d1a9417 | 102 | cpu_startup_entry(CPUHP_ONLINE); |
f87e4cac JF |
103 | } |
104 | ||
53b94fdc KRW |
105 | static void xen_smp_intr_free(unsigned int cpu) |
106 | { | |
ee336e10 | 107 | if (per_cpu(xen_resched_irq, cpu).irq >= 0) { |
9547689f | 108 | unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu).irq, NULL); |
ee336e10 | 109 | per_cpu(xen_resched_irq, cpu).irq = -1; |
b85fffec KRW |
110 | kfree(per_cpu(xen_resched_irq, cpu).name); |
111 | per_cpu(xen_resched_irq, cpu).name = NULL; | |
ee336e10 KRW |
112 | } |
113 | if (per_cpu(xen_callfunc_irq, cpu).irq >= 0) { | |
9547689f | 114 | unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu).irq, NULL); |
ee336e10 | 115 | per_cpu(xen_callfunc_irq, cpu).irq = -1; |
b85fffec KRW |
116 | kfree(per_cpu(xen_callfunc_irq, cpu).name); |
117 | per_cpu(xen_callfunc_irq, cpu).name = NULL; | |
ee336e10 KRW |
118 | } |
119 | if (per_cpu(xen_debug_irq, cpu).irq >= 0) { | |
9547689f | 120 | unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu).irq, NULL); |
ee336e10 | 121 | per_cpu(xen_debug_irq, cpu).irq = -1; |
b85fffec KRW |
122 | kfree(per_cpu(xen_debug_irq, cpu).name); |
123 | per_cpu(xen_debug_irq, cpu).name = NULL; | |
ee336e10 KRW |
124 | } |
125 | if (per_cpu(xen_callfuncsingle_irq, cpu).irq >= 0) { | |
9547689f | 126 | unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu).irq, |
53b94fdc | 127 | NULL); |
ee336e10 | 128 | per_cpu(xen_callfuncsingle_irq, cpu).irq = -1; |
b85fffec KRW |
129 | kfree(per_cpu(xen_callfuncsingle_irq, cpu).name); |
130 | per_cpu(xen_callfuncsingle_irq, cpu).name = NULL; | |
ee336e10 | 131 | } |
53b94fdc KRW |
132 | if (xen_hvm_domain()) |
133 | return; | |
134 | ||
ee336e10 | 135 | if (per_cpu(xen_irq_work, cpu).irq >= 0) { |
9547689f | 136 | unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL); |
ee336e10 | 137 | per_cpu(xen_irq_work, cpu).irq = -1; |
b85fffec KRW |
138 | kfree(per_cpu(xen_irq_work, cpu).name); |
139 | per_cpu(xen_irq_work, cpu).name = NULL; | |
ee336e10 | 140 | } |
53b94fdc | 141 | }; |
f87e4cac JF |
142 | static int xen_smp_intr_init(unsigned int cpu) |
143 | { | |
144 | int rc; | |
b85fffec | 145 | char *resched_name, *callfunc_name, *debug_name; |
f87e4cac JF |
146 | |
147 | resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu); | |
148 | rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR, | |
149 | cpu, | |
150 | xen_reschedule_interrupt, | |
151 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
152 | resched_name, | |
153 | NULL); | |
154 | if (rc < 0) | |
155 | goto fail; | |
9547689f | 156 | per_cpu(xen_resched_irq, cpu).irq = rc; |
b85fffec | 157 | per_cpu(xen_resched_irq, cpu).name = resched_name; |
f87e4cac JF |
158 | |
159 | callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu); | |
160 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR, | |
161 | cpu, | |
162 | xen_call_function_interrupt, | |
163 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
164 | callfunc_name, | |
165 | NULL); | |
166 | if (rc < 0) | |
167 | goto fail; | |
9547689f | 168 | per_cpu(xen_callfunc_irq, cpu).irq = rc; |
b85fffec | 169 | per_cpu(xen_callfunc_irq, cpu).name = callfunc_name; |
f87e4cac | 170 | |
ee523ca1 JF |
171 | debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu); |
172 | rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt, | |
173 | IRQF_DISABLED | IRQF_PERCPU | IRQF_NOBALANCING, | |
174 | debug_name, NULL); | |
175 | if (rc < 0) | |
176 | goto fail; | |
9547689f | 177 | per_cpu(xen_debug_irq, cpu).irq = rc; |
b85fffec | 178 | per_cpu(xen_debug_irq, cpu).name = debug_name; |
ee523ca1 | 179 | |
3b16cf87 JA |
180 | callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu); |
181 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR, | |
182 | cpu, | |
183 | xen_call_function_single_interrupt, | |
184 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
185 | callfunc_name, | |
186 | NULL); | |
187 | if (rc < 0) | |
188 | goto fail; | |
9547689f | 189 | per_cpu(xen_callfuncsingle_irq, cpu).irq = rc; |
b85fffec | 190 | per_cpu(xen_callfuncsingle_irq, cpu).name = callfunc_name; |
3b16cf87 | 191 | |
27d8b207 KRW |
192 | /* |
193 | * The IRQ worker on PVHVM goes through the native path and uses the | |
194 | * IPI mechanism. | |
195 | */ | |
196 | if (xen_hvm_domain()) | |
197 | return 0; | |
198 | ||
1ff2b0c3 LM |
199 | callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu); |
200 | rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR, | |
201 | cpu, | |
202 | xen_irq_work_interrupt, | |
203 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
204 | callfunc_name, | |
205 | NULL); | |
206 | if (rc < 0) | |
207 | goto fail; | |
9547689f | 208 | per_cpu(xen_irq_work, cpu).irq = rc; |
b85fffec | 209 | per_cpu(xen_irq_work, cpu).name = callfunc_name; |
1ff2b0c3 | 210 | |
f87e4cac JF |
211 | return 0; |
212 | ||
213 | fail: | |
53b94fdc | 214 | xen_smp_intr_free(cpu); |
f87e4cac JF |
215 | return rc; |
216 | } | |
217 | ||
c7b75947 | 218 | static void __init xen_fill_possible_map(void) |
f87e4cac JF |
219 | { |
220 | int i, rc; | |
221 | ||
ea5b8f73 SS |
222 | if (xen_initial_domain()) |
223 | return; | |
224 | ||
225 | for (i = 0; i < nr_cpu_ids; i++) { | |
226 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); | |
227 | if (rc >= 0) { | |
228 | num_processors++; | |
229 | set_cpu_possible(i, true); | |
230 | } | |
231 | } | |
232 | } | |
233 | ||
234 | static void __init xen_filter_cpu_maps(void) | |
235 | { | |
236 | int i, rc; | |
cf405ae6 | 237 | unsigned int subtract = 0; |
ea5b8f73 SS |
238 | |
239 | if (!xen_initial_domain()) | |
240 | return; | |
241 | ||
801fd14a SS |
242 | num_processors = 0; |
243 | disabled_cpus = 0; | |
e7986739 | 244 | for (i = 0; i < nr_cpu_ids; i++) { |
f87e4cac | 245 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); |
4560a294 JF |
246 | if (rc >= 0) { |
247 | num_processors++; | |
4f062896 | 248 | set_cpu_possible(i, true); |
801fd14a SS |
249 | } else { |
250 | set_cpu_possible(i, false); | |
251 | set_cpu_present(i, false); | |
cf405ae6 | 252 | subtract++; |
4560a294 | 253 | } |
f87e4cac | 254 | } |
cf405ae6 KRW |
255 | #ifdef CONFIG_HOTPLUG_CPU |
256 | /* This is akin to using 'nr_cpus' on the Linux command line. | |
257 | * Which is OK as when we use 'dom0_max_vcpus=X' we can only | |
258 | * have up to X, while nr_cpu_ids is greater than X. This | |
259 | * normally is not a problem, except when CPU hotplugging | |
260 | * is involved and then there might be more than X CPUs | |
261 | * in the guest - which will not work as there is no | |
262 | * hypercall to expand the max number of VCPUs an already | |
263 | * running guest has. So cap it up to X. */ | |
264 | if (subtract) | |
265 | nr_cpu_ids = nr_cpu_ids - subtract; | |
266 | #endif | |
267 | ||
f87e4cac JF |
268 | } |
269 | ||
a9e7062d | 270 | static void __init xen_smp_prepare_boot_cpu(void) |
f87e4cac | 271 | { |
f87e4cac JF |
272 | BUG_ON(smp_processor_id() != 0); |
273 | native_smp_prepare_boot_cpu(); | |
274 | ||
f87e4cac JF |
275 | /* We've switched to the "real" per-cpu gdt, so make sure the |
276 | old memory can be recycled */ | |
38341432 | 277 | make_lowmem_page_readwrite(xen_initial_gdt); |
60223a32 | 278 | |
ea5b8f73 | 279 | xen_filter_cpu_maps(); |
60223a32 | 280 | xen_setup_vcpu_info_placement(); |
f87e4cac JF |
281 | } |
282 | ||
a9e7062d | 283 | static void __init xen_smp_prepare_cpus(unsigned int max_cpus) |
f87e4cac JF |
284 | { |
285 | unsigned cpu; | |
900cba88 | 286 | unsigned int i; |
f87e4cac | 287 | |
ed467e69 KRW |
288 | if (skip_ioapic_setup) { |
289 | char *m = (max_cpus == 0) ? | |
290 | "The nosmp parameter is incompatible with Xen; " \ | |
291 | "use Xen dom0_max_vcpus=1 parameter" : | |
292 | "The noapic parameter is incompatible with Xen"; | |
293 | ||
294 | xen_raw_printk(m); | |
295 | panic(m); | |
296 | } | |
2d9e1e2f JF |
297 | xen_init_lock_cpu(0); |
298 | ||
06d0b5d9 | 299 | smp_store_boot_cpu_info(); |
c7b75947 | 300 | cpu_data(0).x86_max_cores = 1; |
900cba88 AJ |
301 | |
302 | for_each_possible_cpu(i) { | |
303 | zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL); | |
304 | zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL); | |
305 | zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL); | |
306 | } | |
f87e4cac JF |
307 | set_cpu_sibling_map(0); |
308 | ||
309 | if (xen_smp_intr_init(0)) | |
310 | BUG(); | |
311 | ||
b78936e1 MT |
312 | if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL)) |
313 | panic("could not allocate xen_cpu_initialized_map\n"); | |
314 | ||
315 | cpumask_copy(xen_cpu_initialized_map, cpumask_of(0)); | |
f87e4cac JF |
316 | |
317 | /* Restrict the possible_map according to max_cpus. */ | |
318 | while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) { | |
e7986739 | 319 | for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--) |
f87e4cac | 320 | continue; |
4f062896 | 321 | set_cpu_possible(cpu, false); |
f87e4cac JF |
322 | } |
323 | ||
7eb43a6d | 324 | for_each_possible_cpu(cpu) |
4f062896 | 325 | set_cpu_present(cpu, true); |
f87e4cac JF |
326 | } |
327 | ||
b53cedeb | 328 | static int __cpuinit |
f87e4cac JF |
329 | cpu_initialize_context(unsigned int cpu, struct task_struct *idle) |
330 | { | |
331 | struct vcpu_guest_context *ctxt; | |
c7b75947 | 332 | struct desc_struct *gdt; |
9976b39b | 333 | unsigned long gdt_mfn; |
f87e4cac | 334 | |
b78936e1 | 335 | if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map)) |
f87e4cac JF |
336 | return 0; |
337 | ||
338 | ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); | |
339 | if (ctxt == NULL) | |
340 | return -ENOMEM; | |
341 | ||
c7b75947 JF |
342 | gdt = get_cpu_gdt_table(cpu); |
343 | ||
f87e4cac | 344 | ctxt->flags = VGCF_IN_KERNEL; |
f87e4cac | 345 | ctxt->user_regs.ss = __KERNEL_DS; |
c7b75947 JF |
346 | #ifdef CONFIG_X86_32 |
347 | ctxt->user_regs.fs = __KERNEL_PERCPU; | |
577eebea | 348 | ctxt->user_regs.gs = __KERNEL_STACK_CANARY; |
795f99b6 JF |
349 | #else |
350 | ctxt->gs_base_kernel = per_cpu_offset(cpu); | |
c7b75947 | 351 | #endif |
f87e4cac | 352 | ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle; |
f87e4cac JF |
353 | |
354 | memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt)); | |
355 | ||
dacd45f4 KRW |
356 | { |
357 | ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */ | |
358 | ctxt->user_regs.ds = __USER_DS; | |
359 | ctxt->user_regs.es = __USER_DS; | |
f87e4cac | 360 | |
dacd45f4 | 361 | xen_copy_trap_info(ctxt->trap_ctxt); |
f87e4cac | 362 | |
dacd45f4 | 363 | ctxt->ldt_ents = 0; |
9976b39b | 364 | |
dacd45f4 | 365 | BUG_ON((unsigned long)gdt & ~PAGE_MASK); |
f87e4cac | 366 | |
dacd45f4 KRW |
367 | gdt_mfn = arbitrary_virt_to_mfn(gdt); |
368 | make_lowmem_page_readonly(gdt); | |
369 | make_lowmem_page_readonly(mfn_to_virt(gdt_mfn)); | |
f87e4cac | 370 | |
dacd45f4 KRW |
371 | ctxt->gdt_frames[0] = gdt_mfn; |
372 | ctxt->gdt_ents = GDT_ENTRIES; | |
f87e4cac | 373 | |
dacd45f4 KRW |
374 | ctxt->kernel_ss = __KERNEL_DS; |
375 | ctxt->kernel_sp = idle->thread.sp0; | |
f87e4cac | 376 | |
c7b75947 | 377 | #ifdef CONFIG_X86_32 |
dacd45f4 KRW |
378 | ctxt->event_callback_cs = __KERNEL_CS; |
379 | ctxt->failsafe_callback_cs = __KERNEL_CS; | |
c7b75947 | 380 | #endif |
dacd45f4 KRW |
381 | ctxt->event_callback_eip = |
382 | (unsigned long)xen_hypervisor_callback; | |
383 | ctxt->failsafe_callback_eip = | |
384 | (unsigned long)xen_failsafe_callback; | |
385 | } | |
386 | ctxt->user_regs.cs = __KERNEL_CS; | |
387 | ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs); | |
f87e4cac JF |
388 | |
389 | per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir); | |
390 | ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir)); | |
391 | ||
392 | if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt)) | |
393 | BUG(); | |
394 | ||
395 | kfree(ctxt); | |
396 | return 0; | |
397 | } | |
398 | ||
7eb43a6d | 399 | static int __cpuinit xen_cpu_up(unsigned int cpu, struct task_struct *idle) |
f87e4cac | 400 | { |
f87e4cac JF |
401 | int rc; |
402 | ||
c6f5e0ac | 403 | per_cpu(current_task, cpu) = idle; |
c7b75947 | 404 | #ifdef CONFIG_X86_32 |
f87e4cac | 405 | irq_ctx_init(cpu); |
c7b75947 | 406 | #else |
c7b75947 | 407 | clear_tsk_thread_flag(idle, TIF_FORK); |
38341432 JF |
408 | per_cpu(kernel_stack, cpu) = |
409 | (unsigned long)task_stack_page(idle) - | |
410 | KERNEL_STACK_OFFSET + THREAD_SIZE; | |
c7b75947 | 411 | #endif |
02889672 | 412 | xen_setup_runstate_info(cpu); |
f87e4cac | 413 | xen_setup_timer(cpu); |
2d9e1e2f | 414 | xen_init_lock_cpu(cpu); |
f87e4cac | 415 | |
c7b75947 JF |
416 | per_cpu(cpu_state, cpu) = CPU_UP_PREPARE; |
417 | ||
f87e4cac JF |
418 | /* make sure interrupts start blocked */ |
419 | per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1; | |
420 | ||
421 | rc = cpu_initialize_context(cpu, idle); | |
422 | if (rc) | |
423 | return rc; | |
424 | ||
425 | if (num_online_cpus() == 1) | |
816afe4f RR |
426 | /* Just in case we booted with a single CPU. */ |
427 | alternatives_enable_smp(); | |
f87e4cac JF |
428 | |
429 | rc = xen_smp_intr_init(cpu); | |
430 | if (rc) | |
431 | return rc; | |
432 | ||
f87e4cac JF |
433 | rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL); |
434 | BUG_ON(rc); | |
435 | ||
c7b75947 | 436 | while(per_cpu(cpu_state, cpu) != CPU_ONLINE) { |
1207cf8e | 437 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
c7b75947 JF |
438 | barrier(); |
439 | } | |
440 | ||
f87e4cac JF |
441 | return 0; |
442 | } | |
443 | ||
a9e7062d | 444 | static void xen_smp_cpus_done(unsigned int max_cpus) |
f87e4cac JF |
445 | { |
446 | } | |
447 | ||
2737146b | 448 | #ifdef CONFIG_HOTPLUG_CPU |
26fd1051 | 449 | static int xen_cpu_disable(void) |
d68d82af AN |
450 | { |
451 | unsigned int cpu = smp_processor_id(); | |
452 | if (cpu == 0) | |
453 | return -EBUSY; | |
454 | ||
455 | cpu_disable_common(); | |
456 | ||
457 | load_cr3(swapper_pg_dir); | |
458 | return 0; | |
459 | } | |
460 | ||
26fd1051 | 461 | static void xen_cpu_die(unsigned int cpu) |
d68d82af | 462 | { |
b12abaa1 | 463 | while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) { |
d68d82af AN |
464 | current->state = TASK_UNINTERRUPTIBLE; |
465 | schedule_timeout(HZ/10); | |
466 | } | |
53b94fdc | 467 | xen_smp_intr_free(cpu); |
d68d82af AN |
468 | xen_uninit_lock_cpu(cpu); |
469 | xen_teardown_timer(cpu); | |
d68d82af AN |
470 | } |
471 | ||
71709247 | 472 | static void __cpuinit xen_play_dead(void) /* used only with HOTPLUG_CPU */ |
d68d82af AN |
473 | { |
474 | play_dead_common(); | |
475 | HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); | |
476 | cpu_bringup(); | |
477 | } | |
478 | ||
2737146b | 479 | #else /* !CONFIG_HOTPLUG_CPU */ |
26fd1051 | 480 | static int xen_cpu_disable(void) |
2737146b AN |
481 | { |
482 | return -ENOSYS; | |
483 | } | |
484 | ||
26fd1051 | 485 | static void xen_cpu_die(unsigned int cpu) |
2737146b AN |
486 | { |
487 | BUG(); | |
488 | } | |
489 | ||
26fd1051 | 490 | static void xen_play_dead(void) |
2737146b AN |
491 | { |
492 | BUG(); | |
493 | } | |
494 | ||
495 | #endif | |
f87e4cac JF |
496 | static void stop_self(void *v) |
497 | { | |
498 | int cpu = smp_processor_id(); | |
499 | ||
500 | /* make sure we're not pinning something down */ | |
501 | load_cr3(swapper_pg_dir); | |
502 | /* should set up a minimal gdt */ | |
503 | ||
086748e5 IC |
504 | set_cpu_online(cpu, false); |
505 | ||
f87e4cac JF |
506 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL); |
507 | BUG(); | |
508 | } | |
509 | ||
76fac077 | 510 | static void xen_stop_other_cpus(int wait) |
f87e4cac | 511 | { |
76fac077 | 512 | smp_call_function(stop_self, NULL, wait); |
f87e4cac JF |
513 | } |
514 | ||
a9e7062d | 515 | static void xen_smp_send_reschedule(int cpu) |
f87e4cac JF |
516 | { |
517 | xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR); | |
518 | } | |
519 | ||
f447d56d BG |
520 | static void __xen_send_IPI_mask(const struct cpumask *mask, |
521 | int vector) | |
f87e4cac JF |
522 | { |
523 | unsigned cpu; | |
524 | ||
bcda016e | 525 | for_each_cpu_and(cpu, mask, cpu_online_mask) |
f87e4cac JF |
526 | xen_send_IPI_one(cpu, vector); |
527 | } | |
528 | ||
bcda016e | 529 | static void xen_smp_send_call_function_ipi(const struct cpumask *mask) |
3b16cf87 JA |
530 | { |
531 | int cpu; | |
532 | ||
f447d56d | 533 | __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR); |
3b16cf87 JA |
534 | |
535 | /* Make sure other vcpus get a chance to run if they need to. */ | |
bcda016e | 536 | for_each_cpu(cpu, mask) { |
3b16cf87 | 537 | if (xen_vcpu_stolen(cpu)) { |
1207cf8e | 538 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
3b16cf87 JA |
539 | break; |
540 | } | |
541 | } | |
542 | } | |
543 | ||
a9e7062d | 544 | static void xen_smp_send_call_function_single_ipi(int cpu) |
3b16cf87 | 545 | { |
f447d56d | 546 | __xen_send_IPI_mask(cpumask_of(cpu), |
e7986739 | 547 | XEN_CALL_FUNCTION_SINGLE_VECTOR); |
3b16cf87 JA |
548 | } |
549 | ||
f447d56d BG |
550 | static inline int xen_map_vector(int vector) |
551 | { | |
552 | int xen_vector; | |
553 | ||
554 | switch (vector) { | |
555 | case RESCHEDULE_VECTOR: | |
556 | xen_vector = XEN_RESCHEDULE_VECTOR; | |
557 | break; | |
558 | case CALL_FUNCTION_VECTOR: | |
559 | xen_vector = XEN_CALL_FUNCTION_VECTOR; | |
560 | break; | |
561 | case CALL_FUNCTION_SINGLE_VECTOR: | |
562 | xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR; | |
563 | break; | |
1ff2b0c3 LM |
564 | case IRQ_WORK_VECTOR: |
565 | xen_vector = XEN_IRQ_WORK_VECTOR; | |
566 | break; | |
f447d56d BG |
567 | default: |
568 | xen_vector = -1; | |
569 | printk(KERN_ERR "xen: vector 0x%x is not implemented\n", | |
570 | vector); | |
571 | } | |
572 | ||
573 | return xen_vector; | |
574 | } | |
575 | ||
576 | void xen_send_IPI_mask(const struct cpumask *mask, | |
577 | int vector) | |
578 | { | |
579 | int xen_vector = xen_map_vector(vector); | |
580 | ||
581 | if (xen_vector >= 0) | |
582 | __xen_send_IPI_mask(mask, xen_vector); | |
583 | } | |
584 | ||
585 | void xen_send_IPI_all(int vector) | |
586 | { | |
587 | int xen_vector = xen_map_vector(vector); | |
588 | ||
589 | if (xen_vector >= 0) | |
590 | __xen_send_IPI_mask(cpu_online_mask, xen_vector); | |
591 | } | |
592 | ||
593 | void xen_send_IPI_self(int vector) | |
594 | { | |
595 | int xen_vector = xen_map_vector(vector); | |
596 | ||
597 | if (xen_vector >= 0) | |
598 | xen_send_IPI_one(smp_processor_id(), xen_vector); | |
599 | } | |
600 | ||
601 | void xen_send_IPI_mask_allbutself(const struct cpumask *mask, | |
602 | int vector) | |
603 | { | |
604 | unsigned cpu; | |
605 | unsigned int this_cpu = smp_processor_id(); | |
1db01b49 | 606 | int xen_vector = xen_map_vector(vector); |
f447d56d | 607 | |
1db01b49 | 608 | if (!(num_online_cpus() > 1) || (xen_vector < 0)) |
f447d56d BG |
609 | return; |
610 | ||
611 | for_each_cpu_and(cpu, mask, cpu_online_mask) { | |
612 | if (this_cpu == cpu) | |
613 | continue; | |
614 | ||
1db01b49 | 615 | xen_send_IPI_one(cpu, xen_vector); |
f447d56d BG |
616 | } |
617 | } | |
618 | ||
619 | void xen_send_IPI_allbutself(int vector) | |
620 | { | |
1db01b49 | 621 | xen_send_IPI_mask_allbutself(cpu_online_mask, vector); |
f447d56d BG |
622 | } |
623 | ||
f87e4cac JF |
624 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id) |
625 | { | |
f87e4cac | 626 | irq_enter(); |
3b16cf87 | 627 | generic_smp_call_function_interrupt(); |
1b437c8c | 628 | inc_irq_stat(irq_call_count); |
f87e4cac JF |
629 | irq_exit(); |
630 | ||
f87e4cac JF |
631 | return IRQ_HANDLED; |
632 | } | |
633 | ||
3b16cf87 | 634 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id) |
f87e4cac | 635 | { |
3b16cf87 JA |
636 | irq_enter(); |
637 | generic_smp_call_function_single_interrupt(); | |
1b437c8c | 638 | inc_irq_stat(irq_call_count); |
3b16cf87 | 639 | irq_exit(); |
f87e4cac | 640 | |
3b16cf87 | 641 | return IRQ_HANDLED; |
f87e4cac | 642 | } |
a9e7062d | 643 | |
1ff2b0c3 LM |
644 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id) |
645 | { | |
646 | irq_enter(); | |
647 | irq_work_run(); | |
648 | inc_irq_stat(apic_irq_work_irqs); | |
649 | irq_exit(); | |
650 | ||
651 | return IRQ_HANDLED; | |
652 | } | |
653 | ||
b53cedeb | 654 | static const struct smp_ops xen_smp_ops __initconst = { |
a9e7062d JF |
655 | .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, |
656 | .smp_prepare_cpus = xen_smp_prepare_cpus, | |
a9e7062d JF |
657 | .smp_cpus_done = xen_smp_cpus_done, |
658 | ||
d68d82af AN |
659 | .cpu_up = xen_cpu_up, |
660 | .cpu_die = xen_cpu_die, | |
661 | .cpu_disable = xen_cpu_disable, | |
662 | .play_dead = xen_play_dead, | |
663 | ||
76fac077 | 664 | .stop_other_cpus = xen_stop_other_cpus, |
a9e7062d JF |
665 | .smp_send_reschedule = xen_smp_send_reschedule, |
666 | ||
667 | .send_call_func_ipi = xen_smp_send_call_function_ipi, | |
668 | .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi, | |
669 | }; | |
670 | ||
671 | void __init xen_smp_init(void) | |
672 | { | |
673 | smp_ops = xen_smp_ops; | |
c7b75947 | 674 | xen_fill_possible_map(); |
2d9e1e2f | 675 | xen_init_spinlocks(); |
a9e7062d | 676 | } |
99bbb3a8 SS |
677 | |
678 | static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus) | |
679 | { | |
680 | native_smp_prepare_cpus(max_cpus); | |
681 | WARN_ON(xen_smp_intr_init(0)); | |
682 | ||
99bbb3a8 | 683 | xen_init_lock_cpu(0); |
99bbb3a8 SS |
684 | } |
685 | ||
5cdaf183 | 686 | static int __cpuinit xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle) |
99bbb3a8 SS |
687 | { |
688 | int rc; | |
fc78d343 CA |
689 | /* |
690 | * xen_smp_intr_init() needs to run before native_cpu_up() | |
691 | * so that IPI vectors are set up on the booting CPU before | |
692 | * it is marked online in native_cpu_up(). | |
693 | */ | |
694 | rc = xen_smp_intr_init(cpu); | |
695 | WARN_ON(rc); | |
696 | if (!rc) | |
697 | rc = native_cpu_up(cpu, tidle); | |
99bbb3a8 SS |
698 | return rc; |
699 | } | |
700 | ||
701 | static void xen_hvm_cpu_die(unsigned int cpu) | |
702 | { | |
b12abaa1 | 703 | xen_cpu_die(cpu); |
99bbb3a8 SS |
704 | native_cpu_die(cpu); |
705 | } | |
706 | ||
707 | void __init xen_hvm_smp_init(void) | |
708 | { | |
3c05c4be SS |
709 | if (!xen_have_vector_callback) |
710 | return; | |
99bbb3a8 SS |
711 | smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus; |
712 | smp_ops.smp_send_reschedule = xen_smp_send_reschedule; | |
713 | smp_ops.cpu_up = xen_hvm_cpu_up; | |
714 | smp_ops.cpu_die = xen_hvm_cpu_die; | |
715 | smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi; | |
716 | smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi; | |
717 | } |