]>
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> |
38b8d208 | 21 | #include <linux/nmi.h> |
f87e4cac JF |
22 | |
23 | #include <asm/paravirt.h> | |
24 | #include <asm/desc.h> | |
25 | #include <asm/pgtable.h> | |
26 | #include <asm/cpu.h> | |
27 | ||
28 | #include <xen/interface/xen.h> | |
29 | #include <xen/interface/vcpu.h> | |
65d0cf0b | 30 | #include <xen/interface/xenpmu.h> |
f87e4cac JF |
31 | |
32 | #include <asm/xen/interface.h> | |
33 | #include <asm/xen/hypercall.h> | |
34 | ||
ea5b8f73 | 35 | #include <xen/xen.h> |
f87e4cac JF |
36 | #include <xen/page.h> |
37 | #include <xen/events.h> | |
38 | ||
ed467e69 | 39 | #include <xen/hvc-console.h> |
f87e4cac JF |
40 | #include "xen-ops.h" |
41 | #include "mmu.h" | |
a2ef5dc2 | 42 | #include "smp.h" |
65d0cf0b | 43 | #include "pmu.h" |
f87e4cac | 44 | |
b78936e1 | 45 | cpumask_var_t xen_cpu_initialized_map; |
f87e4cac | 46 | |
9547689f KRW |
47 | struct xen_common_irq { |
48 | int irq; | |
49 | char *name; | |
50 | }; | |
ee336e10 KRW |
51 | static DEFINE_PER_CPU(struct xen_common_irq, xen_resched_irq) = { .irq = -1 }; |
52 | static DEFINE_PER_CPU(struct xen_common_irq, xen_callfunc_irq) = { .irq = -1 }; | |
53 | static DEFINE_PER_CPU(struct xen_common_irq, xen_callfuncsingle_irq) = { .irq = -1 }; | |
54 | static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 }; | |
9547689f | 55 | static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 }; |
65d0cf0b | 56 | static DEFINE_PER_CPU(struct xen_common_irq, xen_pmu_irq) = { .irq = -1 }; |
f87e4cac JF |
57 | |
58 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id); | |
3b16cf87 | 59 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id); |
1ff2b0c3 | 60 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id); |
f87e4cac JF |
61 | |
62 | /* | |
184748cc | 63 | * Reschedule call back. |
f87e4cac JF |
64 | */ |
65 | static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id) | |
66 | { | |
1b437c8c | 67 | inc_irq_stat(irq_resched_count); |
184748cc | 68 | scheduler_ipi(); |
38bb5ab4 | 69 | |
f87e4cac JF |
70 | return IRQ_HANDLED; |
71 | } | |
72 | ||
148f9bb8 | 73 | static void cpu_bringup(void) |
f87e4cac | 74 | { |
e8c9e788 | 75 | int cpu; |
f87e4cac JF |
76 | |
77 | cpu_init(); | |
d68d82af | 78 | touch_softlockup_watchdog(); |
c7b75947 JF |
79 | preempt_disable(); |
80 | ||
5840c84b MR |
81 | /* PVH runs in ring 0 and allows us to do native syscalls. Yay! */ |
82 | if (!xen_feature(XENFEAT_supervisor_mode_kernel)) { | |
83 | xen_enable_sysenter(); | |
84 | xen_enable_syscall(); | |
85 | } | |
c7b75947 JF |
86 | cpu = smp_processor_id(); |
87 | smp_store_cpu_info(cpu); | |
88 | cpu_data(cpu).x86_max_cores = 1; | |
89 | set_cpu_sibling_map(cpu); | |
f87e4cac JF |
90 | |
91 | xen_setup_cpu_clockevents(); | |
92 | ||
106b4438 KRW |
93 | notify_cpu_starting(cpu); |
94 | ||
d7d3756c | 95 | set_cpu_online(cpu, true); |
106b4438 | 96 | |
2a442c9c | 97 | cpu_set_state_online(cpu); /* Implies full memory barrier. */ |
c7b75947 | 98 | |
f87e4cac JF |
99 | /* We can take interrupts now: we're officially "up". */ |
100 | local_irq_enable(); | |
d68d82af AN |
101 | } |
102 | ||
063334f3 | 103 | asmlinkage __visible void cpu_bringup_and_idle(void) |
d68d82af AN |
104 | { |
105 | cpu_bringup(); | |
fc6d73d6 | 106 | cpu_startup_entry(CPUHP_AP_ONLINE_IDLE); |
f87e4cac JF |
107 | } |
108 | ||
5fc509bc | 109 | void xen_smp_intr_free(unsigned int cpu) |
53b94fdc | 110 | { |
ee336e10 | 111 | if (per_cpu(xen_resched_irq, cpu).irq >= 0) { |
9547689f | 112 | unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu).irq, NULL); |
ee336e10 | 113 | per_cpu(xen_resched_irq, cpu).irq = -1; |
b85fffec KRW |
114 | kfree(per_cpu(xen_resched_irq, cpu).name); |
115 | per_cpu(xen_resched_irq, cpu).name = NULL; | |
ee336e10 KRW |
116 | } |
117 | if (per_cpu(xen_callfunc_irq, cpu).irq >= 0) { | |
9547689f | 118 | unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu).irq, NULL); |
ee336e10 | 119 | per_cpu(xen_callfunc_irq, cpu).irq = -1; |
b85fffec KRW |
120 | kfree(per_cpu(xen_callfunc_irq, cpu).name); |
121 | per_cpu(xen_callfunc_irq, cpu).name = NULL; | |
ee336e10 KRW |
122 | } |
123 | if (per_cpu(xen_debug_irq, cpu).irq >= 0) { | |
9547689f | 124 | unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu).irq, NULL); |
ee336e10 | 125 | per_cpu(xen_debug_irq, cpu).irq = -1; |
b85fffec KRW |
126 | kfree(per_cpu(xen_debug_irq, cpu).name); |
127 | per_cpu(xen_debug_irq, cpu).name = NULL; | |
ee336e10 KRW |
128 | } |
129 | if (per_cpu(xen_callfuncsingle_irq, cpu).irq >= 0) { | |
9547689f | 130 | unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu).irq, |
53b94fdc | 131 | NULL); |
ee336e10 | 132 | per_cpu(xen_callfuncsingle_irq, cpu).irq = -1; |
b85fffec KRW |
133 | kfree(per_cpu(xen_callfuncsingle_irq, cpu).name); |
134 | per_cpu(xen_callfuncsingle_irq, cpu).name = NULL; | |
ee336e10 | 135 | } |
53b94fdc KRW |
136 | if (xen_hvm_domain()) |
137 | return; | |
138 | ||
ee336e10 | 139 | if (per_cpu(xen_irq_work, cpu).irq >= 0) { |
9547689f | 140 | unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL); |
ee336e10 | 141 | per_cpu(xen_irq_work, cpu).irq = -1; |
b85fffec KRW |
142 | kfree(per_cpu(xen_irq_work, cpu).name); |
143 | per_cpu(xen_irq_work, cpu).name = NULL; | |
ee336e10 | 144 | } |
65d0cf0b BO |
145 | |
146 | if (per_cpu(xen_pmu_irq, cpu).irq >= 0) { | |
147 | unbind_from_irqhandler(per_cpu(xen_pmu_irq, cpu).irq, NULL); | |
148 | per_cpu(xen_pmu_irq, cpu).irq = -1; | |
149 | kfree(per_cpu(xen_pmu_irq, cpu).name); | |
150 | per_cpu(xen_pmu_irq, cpu).name = NULL; | |
151 | } | |
53b94fdc | 152 | }; |
5fc509bc | 153 | int xen_smp_intr_init(unsigned int cpu) |
f87e4cac JF |
154 | { |
155 | int rc; | |
65d0cf0b | 156 | char *resched_name, *callfunc_name, *debug_name, *pmu_name; |
f87e4cac JF |
157 | |
158 | resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu); | |
159 | rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR, | |
160 | cpu, | |
161 | xen_reschedule_interrupt, | |
9d71cee6 | 162 | IRQF_PERCPU|IRQF_NOBALANCING, |
f87e4cac JF |
163 | resched_name, |
164 | NULL); | |
165 | if (rc < 0) | |
166 | goto fail; | |
9547689f | 167 | per_cpu(xen_resched_irq, cpu).irq = rc; |
b85fffec | 168 | per_cpu(xen_resched_irq, cpu).name = resched_name; |
f87e4cac JF |
169 | |
170 | callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu); | |
171 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR, | |
172 | cpu, | |
173 | xen_call_function_interrupt, | |
9d71cee6 | 174 | IRQF_PERCPU|IRQF_NOBALANCING, |
f87e4cac JF |
175 | callfunc_name, |
176 | NULL); | |
177 | if (rc < 0) | |
178 | goto fail; | |
9547689f | 179 | per_cpu(xen_callfunc_irq, cpu).irq = rc; |
b85fffec | 180 | per_cpu(xen_callfunc_irq, cpu).name = callfunc_name; |
f87e4cac | 181 | |
ee523ca1 JF |
182 | debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu); |
183 | rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt, | |
9d71cee6 | 184 | IRQF_PERCPU | IRQF_NOBALANCING, |
ee523ca1 JF |
185 | debug_name, NULL); |
186 | if (rc < 0) | |
187 | goto fail; | |
9547689f | 188 | per_cpu(xen_debug_irq, cpu).irq = rc; |
b85fffec | 189 | per_cpu(xen_debug_irq, cpu).name = debug_name; |
ee523ca1 | 190 | |
3b16cf87 JA |
191 | callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu); |
192 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR, | |
193 | cpu, | |
194 | xen_call_function_single_interrupt, | |
9d71cee6 | 195 | IRQF_PERCPU|IRQF_NOBALANCING, |
3b16cf87 JA |
196 | callfunc_name, |
197 | NULL); | |
198 | if (rc < 0) | |
199 | goto fail; | |
9547689f | 200 | per_cpu(xen_callfuncsingle_irq, cpu).irq = rc; |
b85fffec | 201 | per_cpu(xen_callfuncsingle_irq, cpu).name = callfunc_name; |
3b16cf87 | 202 | |
27d8b207 KRW |
203 | /* |
204 | * The IRQ worker on PVHVM goes through the native path and uses the | |
205 | * IPI mechanism. | |
206 | */ | |
207 | if (xen_hvm_domain()) | |
208 | return 0; | |
209 | ||
1ff2b0c3 LM |
210 | callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu); |
211 | rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR, | |
212 | cpu, | |
213 | xen_irq_work_interrupt, | |
9d71cee6 | 214 | IRQF_PERCPU|IRQF_NOBALANCING, |
1ff2b0c3 LM |
215 | callfunc_name, |
216 | NULL); | |
217 | if (rc < 0) | |
218 | goto fail; | |
9547689f | 219 | per_cpu(xen_irq_work, cpu).irq = rc; |
b85fffec | 220 | per_cpu(xen_irq_work, cpu).name = callfunc_name; |
1ff2b0c3 | 221 | |
65d0cf0b BO |
222 | if (is_xen_pmu(cpu)) { |
223 | pmu_name = kasprintf(GFP_KERNEL, "pmu%d", cpu); | |
224 | rc = bind_virq_to_irqhandler(VIRQ_XENPMU, cpu, | |
225 | xen_pmu_irq_handler, | |
226 | IRQF_PERCPU|IRQF_NOBALANCING, | |
227 | pmu_name, NULL); | |
228 | if (rc < 0) | |
229 | goto fail; | |
230 | per_cpu(xen_pmu_irq, cpu).irq = rc; | |
231 | per_cpu(xen_pmu_irq, cpu).name = pmu_name; | |
232 | } | |
233 | ||
f87e4cac JF |
234 | return 0; |
235 | ||
236 | fail: | |
53b94fdc | 237 | xen_smp_intr_free(cpu); |
f87e4cac JF |
238 | return rc; |
239 | } | |
240 | ||
c7b75947 | 241 | static void __init xen_fill_possible_map(void) |
f87e4cac JF |
242 | { |
243 | int i, rc; | |
244 | ||
ea5b8f73 SS |
245 | if (xen_initial_domain()) |
246 | return; | |
247 | ||
248 | for (i = 0; i < nr_cpu_ids; i++) { | |
249 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); | |
250 | if (rc >= 0) { | |
251 | num_processors++; | |
252 | set_cpu_possible(i, true); | |
253 | } | |
254 | } | |
255 | } | |
256 | ||
257 | static void __init xen_filter_cpu_maps(void) | |
258 | { | |
259 | int i, rc; | |
cf405ae6 | 260 | unsigned int subtract = 0; |
ea5b8f73 SS |
261 | |
262 | if (!xen_initial_domain()) | |
263 | return; | |
264 | ||
801fd14a SS |
265 | num_processors = 0; |
266 | disabled_cpus = 0; | |
e7986739 | 267 | for (i = 0; i < nr_cpu_ids; i++) { |
f87e4cac | 268 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); |
4560a294 JF |
269 | if (rc >= 0) { |
270 | num_processors++; | |
4f062896 | 271 | set_cpu_possible(i, true); |
801fd14a SS |
272 | } else { |
273 | set_cpu_possible(i, false); | |
274 | set_cpu_present(i, false); | |
cf405ae6 | 275 | subtract++; |
4560a294 | 276 | } |
f87e4cac | 277 | } |
cf405ae6 KRW |
278 | #ifdef CONFIG_HOTPLUG_CPU |
279 | /* This is akin to using 'nr_cpus' on the Linux command line. | |
280 | * Which is OK as when we use 'dom0_max_vcpus=X' we can only | |
281 | * have up to X, while nr_cpu_ids is greater than X. This | |
282 | * normally is not a problem, except when CPU hotplugging | |
283 | * is involved and then there might be more than X CPUs | |
284 | * in the guest - which will not work as there is no | |
285 | * hypercall to expand the max number of VCPUs an already | |
286 | * running guest has. So cap it up to X. */ | |
287 | if (subtract) | |
288 | nr_cpu_ids = nr_cpu_ids - subtract; | |
289 | #endif | |
290 | ||
f87e4cac JF |
291 | } |
292 | ||
a9e7062d | 293 | static void __init xen_smp_prepare_boot_cpu(void) |
f87e4cac | 294 | { |
f87e4cac JF |
295 | BUG_ON(smp_processor_id() != 0); |
296 | native_smp_prepare_boot_cpu(); | |
297 | ||
26a79995 | 298 | if (xen_pv_domain()) { |
5840c84b MR |
299 | if (!xen_feature(XENFEAT_writable_page_tables)) |
300 | /* We've switched to the "real" per-cpu gdt, so make | |
301 | * sure the old memory can be recycled. */ | |
302 | make_lowmem_page_readwrite(xen_initial_gdt); | |
60223a32 | 303 | |
7cde9b27 FZ |
304 | #ifdef CONFIG_X86_32 |
305 | /* | |
306 | * Xen starts us with XEN_FLAT_RING1_DS, but linux code | |
307 | * expects __USER_DS | |
308 | */ | |
309 | loadsegment(ds, __USER_DS); | |
310 | loadsegment(es, __USER_DS); | |
311 | #endif | |
312 | ||
26a79995 KRW |
313 | xen_filter_cpu_maps(); |
314 | xen_setup_vcpu_info_placement(); | |
315 | } | |
ee42d665 VK |
316 | |
317 | /* | |
318 | * Setup vcpu_info for boot CPU. | |
319 | */ | |
320 | if (xen_hvm_domain()) | |
321 | xen_vcpu_setup(0); | |
322 | ||
26a79995 KRW |
323 | /* |
324 | * The alternative logic (which patches the unlock/lock) runs before | |
325 | * the smp bootup up code is activated. Hence we need to set this up | |
326 | * the core kernel is being patched. Otherwise we will have only | |
327 | * modules patched but not core code. | |
328 | */ | |
bf7aab3a | 329 | xen_init_spinlocks(); |
f87e4cac JF |
330 | } |
331 | ||
a9e7062d | 332 | static void __init xen_smp_prepare_cpus(unsigned int max_cpus) |
f87e4cac JF |
333 | { |
334 | unsigned cpu; | |
900cba88 | 335 | unsigned int i; |
f87e4cac | 336 | |
ed467e69 KRW |
337 | if (skip_ioapic_setup) { |
338 | char *m = (max_cpus == 0) ? | |
339 | "The nosmp parameter is incompatible with Xen; " \ | |
340 | "use Xen dom0_max_vcpus=1 parameter" : | |
341 | "The noapic parameter is incompatible with Xen"; | |
342 | ||
343 | xen_raw_printk(m); | |
344 | panic(m); | |
345 | } | |
2d9e1e2f JF |
346 | xen_init_lock_cpu(0); |
347 | ||
06d0b5d9 | 348 | smp_store_boot_cpu_info(); |
c7b75947 | 349 | cpu_data(0).x86_max_cores = 1; |
900cba88 AJ |
350 | |
351 | for_each_possible_cpu(i) { | |
352 | zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL); | |
353 | zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL); | |
354 | zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL); | |
355 | } | |
f87e4cac JF |
356 | set_cpu_sibling_map(0); |
357 | ||
65d0cf0b BO |
358 | xen_pmu_init(0); |
359 | ||
f87e4cac JF |
360 | if (xen_smp_intr_init(0)) |
361 | BUG(); | |
362 | ||
b78936e1 MT |
363 | if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL)) |
364 | panic("could not allocate xen_cpu_initialized_map\n"); | |
365 | ||
366 | cpumask_copy(xen_cpu_initialized_map, cpumask_of(0)); | |
f87e4cac JF |
367 | |
368 | /* Restrict the possible_map according to max_cpus. */ | |
369 | while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) { | |
e7986739 | 370 | for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--) |
f87e4cac | 371 | continue; |
4f062896 | 372 | set_cpu_possible(cpu, false); |
f87e4cac JF |
373 | } |
374 | ||
7eb43a6d | 375 | for_each_possible_cpu(cpu) |
4f062896 | 376 | set_cpu_present(cpu, true); |
f87e4cac JF |
377 | } |
378 | ||
148f9bb8 | 379 | static int |
f87e4cac JF |
380 | cpu_initialize_context(unsigned int cpu, struct task_struct *idle) |
381 | { | |
382 | struct vcpu_guest_context *ctxt; | |
c7b75947 | 383 | struct desc_struct *gdt; |
9976b39b | 384 | unsigned long gdt_mfn; |
f87e4cac | 385 | |
ce4b1b16 IM |
386 | /* used to tell cpu_init() that it can proceed with initialization */ |
387 | cpumask_set_cpu(cpu, cpu_callout_mask); | |
b78936e1 | 388 | if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map)) |
f87e4cac JF |
389 | return 0; |
390 | ||
391 | ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); | |
392 | if (ctxt == NULL) | |
393 | return -ENOMEM; | |
394 | ||
c7b75947 JF |
395 | gdt = get_cpu_gdt_table(cpu); |
396 | ||
c7b75947 JF |
397 | #ifdef CONFIG_X86_32 |
398 | ctxt->user_regs.fs = __KERNEL_PERCPU; | |
577eebea | 399 | ctxt->user_regs.gs = __KERNEL_STACK_CANARY; |
c7b75947 | 400 | #endif |
f87e4cac JF |
401 | memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt)); |
402 | ||
063334f3 BO |
403 | ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle; |
404 | ctxt->flags = VGCF_IN_KERNEL; | |
405 | ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */ | |
406 | ctxt->user_regs.ds = __USER_DS; | |
407 | ctxt->user_regs.es = __USER_DS; | |
408 | ctxt->user_regs.ss = __KERNEL_DS; | |
f87e4cac | 409 | |
063334f3 | 410 | xen_copy_trap_info(ctxt->trap_ctxt); |
f87e4cac | 411 | |
063334f3 | 412 | ctxt->ldt_ents = 0; |
9976b39b | 413 | |
063334f3 | 414 | BUG_ON((unsigned long)gdt & ~PAGE_MASK); |
f87e4cac | 415 | |
063334f3 BO |
416 | gdt_mfn = arbitrary_virt_to_mfn(gdt); |
417 | make_lowmem_page_readonly(gdt); | |
418 | make_lowmem_page_readonly(mfn_to_virt(gdt_mfn)); | |
f87e4cac | 419 | |
063334f3 BO |
420 | ctxt->gdt_frames[0] = gdt_mfn; |
421 | ctxt->gdt_ents = GDT_ENTRIES; | |
f87e4cac | 422 | |
063334f3 BO |
423 | ctxt->kernel_ss = __KERNEL_DS; |
424 | ctxt->kernel_sp = idle->thread.sp0; | |
f87e4cac | 425 | |
c7b75947 | 426 | #ifdef CONFIG_X86_32 |
063334f3 BO |
427 | ctxt->event_callback_cs = __KERNEL_CS; |
428 | ctxt->failsafe_callback_cs = __KERNEL_CS; | |
5840c84b | 429 | #else |
063334f3 | 430 | ctxt->gs_base_kernel = per_cpu_offset(cpu); |
5840c84b | 431 | #endif |
063334f3 BO |
432 | ctxt->event_callback_eip = |
433 | (unsigned long)xen_hypervisor_callback; | |
434 | ctxt->failsafe_callback_eip = | |
435 | (unsigned long)xen_failsafe_callback; | |
436 | ctxt->user_regs.cs = __KERNEL_CS; | |
437 | per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir); | |
438 | ||
dacd45f4 | 439 | ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs); |
0df4f266 | 440 | ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_gfn(swapper_pg_dir)); |
ad5475f9 | 441 | if (HYPERVISOR_vcpu_op(VCPUOP_initialise, xen_vcpu_nr(cpu), ctxt)) |
f87e4cac JF |
442 | BUG(); |
443 | ||
444 | kfree(ctxt); | |
445 | return 0; | |
446 | } | |
447 | ||
148f9bb8 | 448 | static int xen_cpu_up(unsigned int cpu, struct task_struct *idle) |
f87e4cac | 449 | { |
f87e4cac JF |
450 | int rc; |
451 | ||
3f85483b | 452 | common_cpu_up(cpu, idle); |
4461bbc0 | 453 | |
02889672 | 454 | xen_setup_runstate_info(cpu); |
f87e4cac | 455 | |
2a442c9c PM |
456 | /* |
457 | * PV VCPUs are always successfully taken down (see 'while' loop | |
458 | * in xen_cpu_die()), so -EBUSY is an error. | |
459 | */ | |
460 | rc = cpu_check_up_prepare(cpu); | |
461 | if (rc) | |
462 | return rc; | |
c7b75947 | 463 | |
f87e4cac JF |
464 | /* make sure interrupts start blocked */ |
465 | per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1; | |
466 | ||
467 | rc = cpu_initialize_context(cpu, idle); | |
468 | if (rc) | |
469 | return rc; | |
470 | ||
65d0cf0b BO |
471 | xen_pmu_init(cpu); |
472 | ||
ad5475f9 | 473 | rc = HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL); |
f87e4cac JF |
474 | BUG_ON(rc); |
475 | ||
2a442c9c | 476 | while (cpu_report_state(cpu) != CPU_ONLINE) |
1207cf8e | 477 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
c7b75947 | 478 | |
f87e4cac JF |
479 | return 0; |
480 | } | |
481 | ||
a9e7062d | 482 | static void xen_smp_cpus_done(unsigned int max_cpus) |
f87e4cac JF |
483 | { |
484 | } | |
485 | ||
2737146b | 486 | #ifdef CONFIG_HOTPLUG_CPU |
26fd1051 | 487 | static int xen_cpu_disable(void) |
d68d82af AN |
488 | { |
489 | unsigned int cpu = smp_processor_id(); | |
490 | if (cpu == 0) | |
491 | return -EBUSY; | |
492 | ||
493 | cpu_disable_common(); | |
494 | ||
495 | load_cr3(swapper_pg_dir); | |
496 | return 0; | |
497 | } | |
498 | ||
26fd1051 | 499 | static void xen_cpu_die(unsigned int cpu) |
d68d82af | 500 | { |
ad5475f9 VK |
501 | while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up, |
502 | xen_vcpu_nr(cpu), NULL)) { | |
57b6b99b | 503 | __set_current_state(TASK_UNINTERRUPTIBLE); |
d68d82af AN |
504 | schedule_timeout(HZ/10); |
505 | } | |
54279552 | 506 | |
2a442c9c PM |
507 | if (common_cpu_die(cpu) == 0) { |
508 | xen_smp_intr_free(cpu); | |
509 | xen_uninit_lock_cpu(cpu); | |
510 | xen_teardown_timer(cpu); | |
65d0cf0b | 511 | xen_pmu_finish(cpu); |
2a442c9c | 512 | } |
d68d82af AN |
513 | } |
514 | ||
148f9bb8 | 515 | static void xen_play_dead(void) /* used only with HOTPLUG_CPU */ |
d68d82af AN |
516 | { |
517 | play_dead_common(); | |
ad5475f9 | 518 | HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(smp_processor_id()), NULL); |
d68d82af | 519 | cpu_bringup(); |
466318a8 KRW |
520 | /* |
521 | * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down) | |
522 | * clears certain data that the cpu_idle loop (which called us | |
523 | * and that we return from) expects. The only way to get that | |
524 | * data back is to call: | |
525 | */ | |
526 | tick_nohz_idle_enter(); | |
dc6416f1 BO |
527 | |
528 | cpu_startup_entry(CPUHP_AP_ONLINE_IDLE); | |
d68d82af AN |
529 | } |
530 | ||
2737146b | 531 | #else /* !CONFIG_HOTPLUG_CPU */ |
26fd1051 | 532 | static int xen_cpu_disable(void) |
2737146b AN |
533 | { |
534 | return -ENOSYS; | |
535 | } | |
536 | ||
26fd1051 | 537 | static void xen_cpu_die(unsigned int cpu) |
2737146b AN |
538 | { |
539 | BUG(); | |
540 | } | |
541 | ||
26fd1051 | 542 | static void xen_play_dead(void) |
2737146b AN |
543 | { |
544 | BUG(); | |
545 | } | |
546 | ||
547 | #endif | |
f87e4cac JF |
548 | static void stop_self(void *v) |
549 | { | |
550 | int cpu = smp_processor_id(); | |
551 | ||
552 | /* make sure we're not pinning something down */ | |
553 | load_cr3(swapper_pg_dir); | |
554 | /* should set up a minimal gdt */ | |
555 | ||
086748e5 IC |
556 | set_cpu_online(cpu, false); |
557 | ||
ad5475f9 | 558 | HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL); |
f87e4cac JF |
559 | BUG(); |
560 | } | |
561 | ||
76fac077 | 562 | static void xen_stop_other_cpus(int wait) |
f87e4cac | 563 | { |
76fac077 | 564 | smp_call_function(stop_self, NULL, wait); |
f87e4cac JF |
565 | } |
566 | ||
a9e7062d | 567 | static void xen_smp_send_reschedule(int cpu) |
f87e4cac JF |
568 | { |
569 | xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR); | |
570 | } | |
571 | ||
f447d56d BG |
572 | static void __xen_send_IPI_mask(const struct cpumask *mask, |
573 | int vector) | |
f87e4cac JF |
574 | { |
575 | unsigned cpu; | |
576 | ||
bcda016e | 577 | for_each_cpu_and(cpu, mask, cpu_online_mask) |
f87e4cac JF |
578 | xen_send_IPI_one(cpu, vector); |
579 | } | |
580 | ||
bcda016e | 581 | static void xen_smp_send_call_function_ipi(const struct cpumask *mask) |
3b16cf87 JA |
582 | { |
583 | int cpu; | |
584 | ||
f447d56d | 585 | __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR); |
3b16cf87 JA |
586 | |
587 | /* Make sure other vcpus get a chance to run if they need to. */ | |
bcda016e | 588 | for_each_cpu(cpu, mask) { |
3b16cf87 | 589 | if (xen_vcpu_stolen(cpu)) { |
1207cf8e | 590 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
3b16cf87 JA |
591 | break; |
592 | } | |
593 | } | |
594 | } | |
595 | ||
a9e7062d | 596 | static void xen_smp_send_call_function_single_ipi(int cpu) |
3b16cf87 | 597 | { |
f447d56d | 598 | __xen_send_IPI_mask(cpumask_of(cpu), |
e7986739 | 599 | XEN_CALL_FUNCTION_SINGLE_VECTOR); |
3b16cf87 JA |
600 | } |
601 | ||
f447d56d BG |
602 | static inline int xen_map_vector(int vector) |
603 | { | |
604 | int xen_vector; | |
605 | ||
606 | switch (vector) { | |
607 | case RESCHEDULE_VECTOR: | |
608 | xen_vector = XEN_RESCHEDULE_VECTOR; | |
609 | break; | |
610 | case CALL_FUNCTION_VECTOR: | |
611 | xen_vector = XEN_CALL_FUNCTION_VECTOR; | |
612 | break; | |
613 | case CALL_FUNCTION_SINGLE_VECTOR: | |
614 | xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR; | |
615 | break; | |
1ff2b0c3 LM |
616 | case IRQ_WORK_VECTOR: |
617 | xen_vector = XEN_IRQ_WORK_VECTOR; | |
618 | break; | |
6efa20e4 KRW |
619 | #ifdef CONFIG_X86_64 |
620 | case NMI_VECTOR: | |
621 | case APIC_DM_NMI: /* Some use that instead of NMI_VECTOR */ | |
622 | xen_vector = XEN_NMI_VECTOR; | |
623 | break; | |
624 | #endif | |
f447d56d BG |
625 | default: |
626 | xen_vector = -1; | |
627 | printk(KERN_ERR "xen: vector 0x%x is not implemented\n", | |
628 | vector); | |
629 | } | |
630 | ||
631 | return xen_vector; | |
632 | } | |
633 | ||
634 | void xen_send_IPI_mask(const struct cpumask *mask, | |
635 | int vector) | |
636 | { | |
637 | int xen_vector = xen_map_vector(vector); | |
638 | ||
639 | if (xen_vector >= 0) | |
640 | __xen_send_IPI_mask(mask, xen_vector); | |
641 | } | |
642 | ||
643 | void xen_send_IPI_all(int vector) | |
644 | { | |
645 | int xen_vector = xen_map_vector(vector); | |
646 | ||
647 | if (xen_vector >= 0) | |
648 | __xen_send_IPI_mask(cpu_online_mask, xen_vector); | |
649 | } | |
650 | ||
651 | void xen_send_IPI_self(int vector) | |
652 | { | |
653 | int xen_vector = xen_map_vector(vector); | |
654 | ||
655 | if (xen_vector >= 0) | |
656 | xen_send_IPI_one(smp_processor_id(), xen_vector); | |
657 | } | |
658 | ||
659 | void xen_send_IPI_mask_allbutself(const struct cpumask *mask, | |
660 | int vector) | |
661 | { | |
662 | unsigned cpu; | |
663 | unsigned int this_cpu = smp_processor_id(); | |
1db01b49 | 664 | int xen_vector = xen_map_vector(vector); |
f447d56d | 665 | |
1db01b49 | 666 | if (!(num_online_cpus() > 1) || (xen_vector < 0)) |
f447d56d BG |
667 | return; |
668 | ||
669 | for_each_cpu_and(cpu, mask, cpu_online_mask) { | |
670 | if (this_cpu == cpu) | |
671 | continue; | |
672 | ||
1db01b49 | 673 | xen_send_IPI_one(cpu, xen_vector); |
f447d56d BG |
674 | } |
675 | } | |
676 | ||
677 | void xen_send_IPI_allbutself(int vector) | |
678 | { | |
1db01b49 | 679 | xen_send_IPI_mask_allbutself(cpu_online_mask, vector); |
f447d56d BG |
680 | } |
681 | ||
f87e4cac JF |
682 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id) |
683 | { | |
f87e4cac | 684 | irq_enter(); |
3b16cf87 | 685 | generic_smp_call_function_interrupt(); |
1b437c8c | 686 | inc_irq_stat(irq_call_count); |
f87e4cac JF |
687 | irq_exit(); |
688 | ||
f87e4cac JF |
689 | return IRQ_HANDLED; |
690 | } | |
691 | ||
3b16cf87 | 692 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id) |
f87e4cac | 693 | { |
3b16cf87 JA |
694 | irq_enter(); |
695 | generic_smp_call_function_single_interrupt(); | |
1b437c8c | 696 | inc_irq_stat(irq_call_count); |
3b16cf87 | 697 | irq_exit(); |
f87e4cac | 698 | |
3b16cf87 | 699 | return IRQ_HANDLED; |
f87e4cac | 700 | } |
a9e7062d | 701 | |
1ff2b0c3 LM |
702 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id) |
703 | { | |
704 | irq_enter(); | |
705 | irq_work_run(); | |
706 | inc_irq_stat(apic_irq_work_irqs); | |
707 | irq_exit(); | |
708 | ||
709 | return IRQ_HANDLED; | |
710 | } | |
711 | ||
b53cedeb | 712 | static const struct smp_ops xen_smp_ops __initconst = { |
a9e7062d JF |
713 | .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, |
714 | .smp_prepare_cpus = xen_smp_prepare_cpus, | |
a9e7062d JF |
715 | .smp_cpus_done = xen_smp_cpus_done, |
716 | ||
d68d82af AN |
717 | .cpu_up = xen_cpu_up, |
718 | .cpu_die = xen_cpu_die, | |
719 | .cpu_disable = xen_cpu_disable, | |
720 | .play_dead = xen_play_dead, | |
721 | ||
76fac077 | 722 | .stop_other_cpus = xen_stop_other_cpus, |
a9e7062d JF |
723 | .smp_send_reschedule = xen_smp_send_reschedule, |
724 | ||
725 | .send_call_func_ipi = xen_smp_send_call_function_ipi, | |
726 | .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi, | |
727 | }; | |
728 | ||
729 | void __init xen_smp_init(void) | |
730 | { | |
731 | smp_ops = xen_smp_ops; | |
c7b75947 | 732 | xen_fill_possible_map(); |
a9e7062d | 733 | } |
99bbb3a8 SS |
734 | |
735 | static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus) | |
736 | { | |
737 | native_smp_prepare_cpus(max_cpus); | |
738 | WARN_ON(xen_smp_intr_init(0)); | |
739 | ||
99bbb3a8 | 740 | xen_init_lock_cpu(0); |
99bbb3a8 SS |
741 | } |
742 | ||
99bbb3a8 SS |
743 | void __init xen_hvm_smp_init(void) |
744 | { | |
745 | smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus; | |
746 | smp_ops.smp_send_reschedule = xen_smp_send_reschedule; | |
2a442c9c | 747 | smp_ops.cpu_die = xen_cpu_die; |
99bbb3a8 SS |
748 | smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi; |
749 | smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi; | |
26a79995 | 750 | smp_ops.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu; |
99bbb3a8 | 751 | } |