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