]>
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 | ||
a2ef5dc2 MR |
102 | /* |
103 | * Note: cpu parameter is only relevant for PVH. The reason for passing it | |
104 | * is we can't do smp_processor_id until the percpu segments are loaded, for | |
105 | * which we need the cpu number! So we pass it in rdi as first parameter. | |
106 | */ | |
107 | asmlinkage __visible void cpu_bringup_and_idle(int cpu) | |
d68d82af | 108 | { |
a2ef5dc2 | 109 | #ifdef CONFIG_XEN_PVH |
5840c84b MR |
110 | if (xen_feature(XENFEAT_auto_translated_physmap) && |
111 | xen_feature(XENFEAT_supervisor_mode_kernel)) | |
c9f6e997 | 112 | xen_pvh_secondary_vcpu_init(cpu); |
5840c84b | 113 | #endif |
d68d82af | 114 | cpu_bringup(); |
7d1a9417 | 115 | cpu_startup_entry(CPUHP_ONLINE); |
f87e4cac JF |
116 | } |
117 | ||
53b94fdc KRW |
118 | static void xen_smp_intr_free(unsigned int cpu) |
119 | { | |
ee336e10 | 120 | if (per_cpu(xen_resched_irq, cpu).irq >= 0) { |
9547689f | 121 | unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu).irq, NULL); |
ee336e10 | 122 | per_cpu(xen_resched_irq, cpu).irq = -1; |
b85fffec KRW |
123 | kfree(per_cpu(xen_resched_irq, cpu).name); |
124 | per_cpu(xen_resched_irq, cpu).name = NULL; | |
ee336e10 KRW |
125 | } |
126 | if (per_cpu(xen_callfunc_irq, cpu).irq >= 0) { | |
9547689f | 127 | unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu).irq, NULL); |
ee336e10 | 128 | per_cpu(xen_callfunc_irq, cpu).irq = -1; |
b85fffec KRW |
129 | kfree(per_cpu(xen_callfunc_irq, cpu).name); |
130 | per_cpu(xen_callfunc_irq, cpu).name = NULL; | |
ee336e10 KRW |
131 | } |
132 | if (per_cpu(xen_debug_irq, cpu).irq >= 0) { | |
9547689f | 133 | unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu).irq, NULL); |
ee336e10 | 134 | per_cpu(xen_debug_irq, cpu).irq = -1; |
b85fffec KRW |
135 | kfree(per_cpu(xen_debug_irq, cpu).name); |
136 | per_cpu(xen_debug_irq, cpu).name = NULL; | |
ee336e10 KRW |
137 | } |
138 | if (per_cpu(xen_callfuncsingle_irq, cpu).irq >= 0) { | |
9547689f | 139 | unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu).irq, |
53b94fdc | 140 | NULL); |
ee336e10 | 141 | per_cpu(xen_callfuncsingle_irq, cpu).irq = -1; |
b85fffec KRW |
142 | kfree(per_cpu(xen_callfuncsingle_irq, cpu).name); |
143 | per_cpu(xen_callfuncsingle_irq, cpu).name = NULL; | |
ee336e10 | 144 | } |
53b94fdc KRW |
145 | if (xen_hvm_domain()) |
146 | return; | |
147 | ||
ee336e10 | 148 | if (per_cpu(xen_irq_work, cpu).irq >= 0) { |
9547689f | 149 | unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL); |
ee336e10 | 150 | per_cpu(xen_irq_work, cpu).irq = -1; |
b85fffec KRW |
151 | kfree(per_cpu(xen_irq_work, cpu).name); |
152 | per_cpu(xen_irq_work, cpu).name = NULL; | |
ee336e10 | 153 | } |
65d0cf0b BO |
154 | |
155 | if (per_cpu(xen_pmu_irq, cpu).irq >= 0) { | |
156 | unbind_from_irqhandler(per_cpu(xen_pmu_irq, cpu).irq, NULL); | |
157 | per_cpu(xen_pmu_irq, cpu).irq = -1; | |
158 | kfree(per_cpu(xen_pmu_irq, cpu).name); | |
159 | per_cpu(xen_pmu_irq, cpu).name = NULL; | |
160 | } | |
53b94fdc | 161 | }; |
f87e4cac JF |
162 | static int xen_smp_intr_init(unsigned int cpu) |
163 | { | |
164 | int rc; | |
65d0cf0b | 165 | char *resched_name, *callfunc_name, *debug_name, *pmu_name; |
f87e4cac JF |
166 | |
167 | resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu); | |
168 | rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR, | |
169 | cpu, | |
170 | xen_reschedule_interrupt, | |
9d71cee6 | 171 | IRQF_PERCPU|IRQF_NOBALANCING, |
f87e4cac JF |
172 | resched_name, |
173 | NULL); | |
174 | if (rc < 0) | |
175 | goto fail; | |
9547689f | 176 | per_cpu(xen_resched_irq, cpu).irq = rc; |
b85fffec | 177 | per_cpu(xen_resched_irq, cpu).name = resched_name; |
f87e4cac JF |
178 | |
179 | callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu); | |
180 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR, | |
181 | cpu, | |
182 | xen_call_function_interrupt, | |
9d71cee6 | 183 | IRQF_PERCPU|IRQF_NOBALANCING, |
f87e4cac JF |
184 | callfunc_name, |
185 | NULL); | |
186 | if (rc < 0) | |
187 | goto fail; | |
9547689f | 188 | per_cpu(xen_callfunc_irq, cpu).irq = rc; |
b85fffec | 189 | per_cpu(xen_callfunc_irq, cpu).name = callfunc_name; |
f87e4cac | 190 | |
ee523ca1 JF |
191 | debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu); |
192 | rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt, | |
9d71cee6 | 193 | IRQF_PERCPU | IRQF_NOBALANCING, |
ee523ca1 JF |
194 | debug_name, NULL); |
195 | if (rc < 0) | |
196 | goto fail; | |
9547689f | 197 | per_cpu(xen_debug_irq, cpu).irq = rc; |
b85fffec | 198 | per_cpu(xen_debug_irq, cpu).name = debug_name; |
ee523ca1 | 199 | |
3b16cf87 JA |
200 | callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu); |
201 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR, | |
202 | cpu, | |
203 | xen_call_function_single_interrupt, | |
9d71cee6 | 204 | IRQF_PERCPU|IRQF_NOBALANCING, |
3b16cf87 JA |
205 | callfunc_name, |
206 | NULL); | |
207 | if (rc < 0) | |
208 | goto fail; | |
9547689f | 209 | per_cpu(xen_callfuncsingle_irq, cpu).irq = rc; |
b85fffec | 210 | per_cpu(xen_callfuncsingle_irq, cpu).name = callfunc_name; |
3b16cf87 | 211 | |
27d8b207 KRW |
212 | /* |
213 | * The IRQ worker on PVHVM goes through the native path and uses the | |
214 | * IPI mechanism. | |
215 | */ | |
216 | if (xen_hvm_domain()) | |
217 | return 0; | |
218 | ||
1ff2b0c3 LM |
219 | callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu); |
220 | rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR, | |
221 | cpu, | |
222 | xen_irq_work_interrupt, | |
9d71cee6 | 223 | IRQF_PERCPU|IRQF_NOBALANCING, |
1ff2b0c3 LM |
224 | callfunc_name, |
225 | NULL); | |
226 | if (rc < 0) | |
227 | goto fail; | |
9547689f | 228 | per_cpu(xen_irq_work, cpu).irq = rc; |
b85fffec | 229 | per_cpu(xen_irq_work, cpu).name = callfunc_name; |
1ff2b0c3 | 230 | |
65d0cf0b BO |
231 | if (is_xen_pmu(cpu)) { |
232 | pmu_name = kasprintf(GFP_KERNEL, "pmu%d", cpu); | |
233 | rc = bind_virq_to_irqhandler(VIRQ_XENPMU, cpu, | |
234 | xen_pmu_irq_handler, | |
235 | IRQF_PERCPU|IRQF_NOBALANCING, | |
236 | pmu_name, NULL); | |
237 | if (rc < 0) | |
238 | goto fail; | |
239 | per_cpu(xen_pmu_irq, cpu).irq = rc; | |
240 | per_cpu(xen_pmu_irq, cpu).name = pmu_name; | |
241 | } | |
242 | ||
f87e4cac JF |
243 | return 0; |
244 | ||
245 | fail: | |
53b94fdc | 246 | xen_smp_intr_free(cpu); |
f87e4cac JF |
247 | return rc; |
248 | } | |
249 | ||
c7b75947 | 250 | static void __init xen_fill_possible_map(void) |
f87e4cac JF |
251 | { |
252 | int i, rc; | |
253 | ||
ea5b8f73 SS |
254 | if (xen_initial_domain()) |
255 | return; | |
256 | ||
257 | for (i = 0; i < nr_cpu_ids; i++) { | |
258 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); | |
259 | if (rc >= 0) { | |
260 | num_processors++; | |
261 | set_cpu_possible(i, true); | |
262 | } | |
263 | } | |
264 | } | |
265 | ||
266 | static void __init xen_filter_cpu_maps(void) | |
267 | { | |
268 | int i, rc; | |
cf405ae6 | 269 | unsigned int subtract = 0; |
ea5b8f73 SS |
270 | |
271 | if (!xen_initial_domain()) | |
272 | return; | |
273 | ||
801fd14a SS |
274 | num_processors = 0; |
275 | disabled_cpus = 0; | |
e7986739 | 276 | for (i = 0; i < nr_cpu_ids; i++) { |
f87e4cac | 277 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); |
4560a294 JF |
278 | if (rc >= 0) { |
279 | num_processors++; | |
4f062896 | 280 | set_cpu_possible(i, true); |
801fd14a SS |
281 | } else { |
282 | set_cpu_possible(i, false); | |
283 | set_cpu_present(i, false); | |
cf405ae6 | 284 | subtract++; |
4560a294 | 285 | } |
f87e4cac | 286 | } |
cf405ae6 KRW |
287 | #ifdef CONFIG_HOTPLUG_CPU |
288 | /* This is akin to using 'nr_cpus' on the Linux command line. | |
289 | * Which is OK as when we use 'dom0_max_vcpus=X' we can only | |
290 | * have up to X, while nr_cpu_ids is greater than X. This | |
291 | * normally is not a problem, except when CPU hotplugging | |
292 | * is involved and then there might be more than X CPUs | |
293 | * in the guest - which will not work as there is no | |
294 | * hypercall to expand the max number of VCPUs an already | |
295 | * running guest has. So cap it up to X. */ | |
296 | if (subtract) | |
297 | nr_cpu_ids = nr_cpu_ids - subtract; | |
298 | #endif | |
299 | ||
f87e4cac JF |
300 | } |
301 | ||
a9e7062d | 302 | static void __init xen_smp_prepare_boot_cpu(void) |
f87e4cac | 303 | { |
f87e4cac JF |
304 | BUG_ON(smp_processor_id() != 0); |
305 | native_smp_prepare_boot_cpu(); | |
306 | ||
26a79995 | 307 | if (xen_pv_domain()) { |
5840c84b MR |
308 | if (!xen_feature(XENFEAT_writable_page_tables)) |
309 | /* We've switched to the "real" per-cpu gdt, so make | |
310 | * sure the old memory can be recycled. */ | |
311 | make_lowmem_page_readwrite(xen_initial_gdt); | |
60223a32 | 312 | |
7cde9b27 FZ |
313 | #ifdef CONFIG_X86_32 |
314 | /* | |
315 | * Xen starts us with XEN_FLAT_RING1_DS, but linux code | |
316 | * expects __USER_DS | |
317 | */ | |
318 | loadsegment(ds, __USER_DS); | |
319 | loadsegment(es, __USER_DS); | |
320 | #endif | |
321 | ||
26a79995 KRW |
322 | xen_filter_cpu_maps(); |
323 | xen_setup_vcpu_info_placement(); | |
324 | } | |
325 | /* | |
326 | * The alternative logic (which patches the unlock/lock) runs before | |
327 | * the smp bootup up code is activated. Hence we need to set this up | |
328 | * the core kernel is being patched. Otherwise we will have only | |
329 | * modules patched but not core code. | |
330 | */ | |
bf7aab3a | 331 | xen_init_spinlocks(); |
f87e4cac JF |
332 | } |
333 | ||
a9e7062d | 334 | static void __init xen_smp_prepare_cpus(unsigned int max_cpus) |
f87e4cac JF |
335 | { |
336 | unsigned cpu; | |
900cba88 | 337 | unsigned int i; |
f87e4cac | 338 | |
ed467e69 KRW |
339 | if (skip_ioapic_setup) { |
340 | char *m = (max_cpus == 0) ? | |
341 | "The nosmp parameter is incompatible with Xen; " \ | |
342 | "use Xen dom0_max_vcpus=1 parameter" : | |
343 | "The noapic parameter is incompatible with Xen"; | |
344 | ||
345 | xen_raw_printk(m); | |
346 | panic(m); | |
347 | } | |
2d9e1e2f JF |
348 | xen_init_lock_cpu(0); |
349 | ||
06d0b5d9 | 350 | smp_store_boot_cpu_info(); |
c7b75947 | 351 | cpu_data(0).x86_max_cores = 1; |
900cba88 AJ |
352 | |
353 | for_each_possible_cpu(i) { | |
354 | zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL); | |
355 | zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL); | |
356 | zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL); | |
357 | } | |
f87e4cac JF |
358 | set_cpu_sibling_map(0); |
359 | ||
65d0cf0b BO |
360 | xen_pmu_init(0); |
361 | ||
f87e4cac JF |
362 | if (xen_smp_intr_init(0)) |
363 | BUG(); | |
364 | ||
b78936e1 MT |
365 | if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL)) |
366 | panic("could not allocate xen_cpu_initialized_map\n"); | |
367 | ||
368 | cpumask_copy(xen_cpu_initialized_map, cpumask_of(0)); | |
f87e4cac JF |
369 | |
370 | /* Restrict the possible_map according to max_cpus. */ | |
371 | while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) { | |
e7986739 | 372 | for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--) |
f87e4cac | 373 | continue; |
4f062896 | 374 | set_cpu_possible(cpu, false); |
f87e4cac JF |
375 | } |
376 | ||
7eb43a6d | 377 | for_each_possible_cpu(cpu) |
4f062896 | 378 | set_cpu_present(cpu, true); |
f87e4cac JF |
379 | } |
380 | ||
148f9bb8 | 381 | static int |
f87e4cac JF |
382 | cpu_initialize_context(unsigned int cpu, struct task_struct *idle) |
383 | { | |
384 | struct vcpu_guest_context *ctxt; | |
c7b75947 | 385 | struct desc_struct *gdt; |
9976b39b | 386 | unsigned long gdt_mfn; |
f87e4cac | 387 | |
ce4b1b16 IM |
388 | /* used to tell cpu_init() that it can proceed with initialization */ |
389 | cpumask_set_cpu(cpu, cpu_callout_mask); | |
b78936e1 | 390 | if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map)) |
f87e4cac JF |
391 | return 0; |
392 | ||
393 | ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); | |
394 | if (ctxt == NULL) | |
395 | return -ENOMEM; | |
396 | ||
c7b75947 JF |
397 | gdt = get_cpu_gdt_table(cpu); |
398 | ||
c7b75947 | 399 | #ifdef CONFIG_X86_32 |
5840c84b | 400 | /* Note: PVH is not yet supported on x86_32. */ |
c7b75947 | 401 | ctxt->user_regs.fs = __KERNEL_PERCPU; |
577eebea | 402 | ctxt->user_regs.gs = __KERNEL_STACK_CANARY; |
c7b75947 | 403 | #endif |
f87e4cac JF |
404 | memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt)); |
405 | ||
5840c84b | 406 | if (!xen_feature(XENFEAT_auto_translated_physmap)) { |
a2ef5dc2 | 407 | ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle; |
5840c84b | 408 | ctxt->flags = VGCF_IN_KERNEL; |
dacd45f4 KRW |
409 | ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */ |
410 | ctxt->user_regs.ds = __USER_DS; | |
411 | ctxt->user_regs.es = __USER_DS; | |
5840c84b | 412 | ctxt->user_regs.ss = __KERNEL_DS; |
f87e4cac | 413 | |
dacd45f4 | 414 | xen_copy_trap_info(ctxt->trap_ctxt); |
f87e4cac | 415 | |
dacd45f4 | 416 | ctxt->ldt_ents = 0; |
9976b39b | 417 | |
dacd45f4 | 418 | BUG_ON((unsigned long)gdt & ~PAGE_MASK); |
f87e4cac | 419 | |
dacd45f4 KRW |
420 | gdt_mfn = arbitrary_virt_to_mfn(gdt); |
421 | make_lowmem_page_readonly(gdt); | |
422 | make_lowmem_page_readonly(mfn_to_virt(gdt_mfn)); | |
f87e4cac | 423 | |
dacd45f4 KRW |
424 | ctxt->gdt_frames[0] = gdt_mfn; |
425 | ctxt->gdt_ents = GDT_ENTRIES; | |
f87e4cac | 426 | |
dacd45f4 KRW |
427 | ctxt->kernel_ss = __KERNEL_DS; |
428 | ctxt->kernel_sp = idle->thread.sp0; | |
f87e4cac | 429 | |
c7b75947 | 430 | #ifdef CONFIG_X86_32 |
dacd45f4 KRW |
431 | ctxt->event_callback_cs = __KERNEL_CS; |
432 | ctxt->failsafe_callback_cs = __KERNEL_CS; | |
5840c84b MR |
433 | #else |
434 | ctxt->gs_base_kernel = per_cpu_offset(cpu); | |
c7b75947 | 435 | #endif |
dacd45f4 KRW |
436 | ctxt->event_callback_eip = |
437 | (unsigned long)xen_hypervisor_callback; | |
438 | ctxt->failsafe_callback_eip = | |
439 | (unsigned long)xen_failsafe_callback; | |
5840c84b MR |
440 | ctxt->user_regs.cs = __KERNEL_CS; |
441 | per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir); | |
dacd45f4 | 442 | } |
a2ef5dc2 MR |
443 | #ifdef CONFIG_XEN_PVH |
444 | else { | |
445 | /* | |
446 | * The vcpu comes on kernel page tables which have the NX pte | |
447 | * bit set. This means before DS/SS is touched, NX in | |
448 | * EFER must be set. Hence the following assembly glue code. | |
5840c84b | 449 | */ |
a2ef5dc2 | 450 | ctxt->user_regs.eip = (unsigned long)xen_pvh_early_cpu_init; |
5840c84b | 451 | ctxt->user_regs.rdi = cpu; |
a2ef5dc2 MR |
452 | ctxt->user_regs.rsi = true; /* entry == true */ |
453 | } | |
5840c84b | 454 | #endif |
dacd45f4 | 455 | ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs); |
0df4f266 | 456 | ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_gfn(swapper_pg_dir)); |
f87e4cac JF |
457 | if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt)) |
458 | BUG(); | |
459 | ||
460 | kfree(ctxt); | |
461 | return 0; | |
462 | } | |
463 | ||
148f9bb8 | 464 | static int xen_cpu_up(unsigned int cpu, struct task_struct *idle) |
f87e4cac | 465 | { |
f87e4cac JF |
466 | int rc; |
467 | ||
3f85483b | 468 | common_cpu_up(cpu, idle); |
4461bbc0 | 469 | |
02889672 | 470 | xen_setup_runstate_info(cpu); |
f87e4cac | 471 | xen_setup_timer(cpu); |
2d9e1e2f | 472 | xen_init_lock_cpu(cpu); |
f87e4cac | 473 | |
2a442c9c PM |
474 | /* |
475 | * PV VCPUs are always successfully taken down (see 'while' loop | |
476 | * in xen_cpu_die()), so -EBUSY is an error. | |
477 | */ | |
478 | rc = cpu_check_up_prepare(cpu); | |
479 | if (rc) | |
480 | return rc; | |
c7b75947 | 481 | |
f87e4cac JF |
482 | /* make sure interrupts start blocked */ |
483 | per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1; | |
484 | ||
485 | rc = cpu_initialize_context(cpu, idle); | |
486 | if (rc) | |
487 | return rc; | |
488 | ||
65d0cf0b BO |
489 | xen_pmu_init(cpu); |
490 | ||
f87e4cac JF |
491 | rc = xen_smp_intr_init(cpu); |
492 | if (rc) | |
493 | return rc; | |
494 | ||
f87e4cac JF |
495 | rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL); |
496 | BUG_ON(rc); | |
497 | ||
2a442c9c | 498 | while (cpu_report_state(cpu) != CPU_ONLINE) |
1207cf8e | 499 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
c7b75947 | 500 | |
f87e4cac JF |
501 | return 0; |
502 | } | |
503 | ||
a9e7062d | 504 | static void xen_smp_cpus_done(unsigned int max_cpus) |
f87e4cac JF |
505 | { |
506 | } | |
507 | ||
2737146b | 508 | #ifdef CONFIG_HOTPLUG_CPU |
26fd1051 | 509 | static int xen_cpu_disable(void) |
d68d82af AN |
510 | { |
511 | unsigned int cpu = smp_processor_id(); | |
512 | if (cpu == 0) | |
513 | return -EBUSY; | |
514 | ||
515 | cpu_disable_common(); | |
516 | ||
517 | load_cr3(swapper_pg_dir); | |
518 | return 0; | |
519 | } | |
520 | ||
26fd1051 | 521 | static void xen_cpu_die(unsigned int cpu) |
d68d82af | 522 | { |
b12abaa1 | 523 | while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) { |
57b6b99b | 524 | __set_current_state(TASK_UNINTERRUPTIBLE); |
d68d82af AN |
525 | schedule_timeout(HZ/10); |
526 | } | |
54279552 | 527 | |
2a442c9c PM |
528 | if (common_cpu_die(cpu) == 0) { |
529 | xen_smp_intr_free(cpu); | |
530 | xen_uninit_lock_cpu(cpu); | |
531 | xen_teardown_timer(cpu); | |
65d0cf0b | 532 | xen_pmu_finish(cpu); |
2a442c9c | 533 | } |
d68d82af AN |
534 | } |
535 | ||
148f9bb8 | 536 | static void xen_play_dead(void) /* used only with HOTPLUG_CPU */ |
d68d82af AN |
537 | { |
538 | play_dead_common(); | |
539 | HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); | |
540 | cpu_bringup(); | |
466318a8 KRW |
541 | /* |
542 | * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down) | |
543 | * clears certain data that the cpu_idle loop (which called us | |
544 | * and that we return from) expects. The only way to get that | |
545 | * data back is to call: | |
546 | */ | |
547 | tick_nohz_idle_enter(); | |
d68d82af AN |
548 | } |
549 | ||
2737146b | 550 | #else /* !CONFIG_HOTPLUG_CPU */ |
26fd1051 | 551 | static int xen_cpu_disable(void) |
2737146b AN |
552 | { |
553 | return -ENOSYS; | |
554 | } | |
555 | ||
26fd1051 | 556 | static void xen_cpu_die(unsigned int cpu) |
2737146b AN |
557 | { |
558 | BUG(); | |
559 | } | |
560 | ||
26fd1051 | 561 | static void xen_play_dead(void) |
2737146b AN |
562 | { |
563 | BUG(); | |
564 | } | |
565 | ||
566 | #endif | |
f87e4cac JF |
567 | static void stop_self(void *v) |
568 | { | |
569 | int cpu = smp_processor_id(); | |
570 | ||
571 | /* make sure we're not pinning something down */ | |
572 | load_cr3(swapper_pg_dir); | |
573 | /* should set up a minimal gdt */ | |
574 | ||
086748e5 IC |
575 | set_cpu_online(cpu, false); |
576 | ||
f87e4cac JF |
577 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL); |
578 | BUG(); | |
579 | } | |
580 | ||
76fac077 | 581 | static void xen_stop_other_cpus(int wait) |
f87e4cac | 582 | { |
76fac077 | 583 | smp_call_function(stop_self, NULL, wait); |
f87e4cac JF |
584 | } |
585 | ||
a9e7062d | 586 | static void xen_smp_send_reschedule(int cpu) |
f87e4cac JF |
587 | { |
588 | xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR); | |
589 | } | |
590 | ||
f447d56d BG |
591 | static void __xen_send_IPI_mask(const struct cpumask *mask, |
592 | int vector) | |
f87e4cac JF |
593 | { |
594 | unsigned cpu; | |
595 | ||
bcda016e | 596 | for_each_cpu_and(cpu, mask, cpu_online_mask) |
f87e4cac JF |
597 | xen_send_IPI_one(cpu, vector); |
598 | } | |
599 | ||
bcda016e | 600 | static void xen_smp_send_call_function_ipi(const struct cpumask *mask) |
3b16cf87 JA |
601 | { |
602 | int cpu; | |
603 | ||
f447d56d | 604 | __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR); |
3b16cf87 JA |
605 | |
606 | /* Make sure other vcpus get a chance to run if they need to. */ | |
bcda016e | 607 | for_each_cpu(cpu, mask) { |
3b16cf87 | 608 | if (xen_vcpu_stolen(cpu)) { |
1207cf8e | 609 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
3b16cf87 JA |
610 | break; |
611 | } | |
612 | } | |
613 | } | |
614 | ||
a9e7062d | 615 | static void xen_smp_send_call_function_single_ipi(int cpu) |
3b16cf87 | 616 | { |
f447d56d | 617 | __xen_send_IPI_mask(cpumask_of(cpu), |
e7986739 | 618 | XEN_CALL_FUNCTION_SINGLE_VECTOR); |
3b16cf87 JA |
619 | } |
620 | ||
f447d56d BG |
621 | static inline int xen_map_vector(int vector) |
622 | { | |
623 | int xen_vector; | |
624 | ||
625 | switch (vector) { | |
626 | case RESCHEDULE_VECTOR: | |
627 | xen_vector = XEN_RESCHEDULE_VECTOR; | |
628 | break; | |
629 | case CALL_FUNCTION_VECTOR: | |
630 | xen_vector = XEN_CALL_FUNCTION_VECTOR; | |
631 | break; | |
632 | case CALL_FUNCTION_SINGLE_VECTOR: | |
633 | xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR; | |
634 | break; | |
1ff2b0c3 LM |
635 | case IRQ_WORK_VECTOR: |
636 | xen_vector = XEN_IRQ_WORK_VECTOR; | |
637 | break; | |
6efa20e4 KRW |
638 | #ifdef CONFIG_X86_64 |
639 | case NMI_VECTOR: | |
640 | case APIC_DM_NMI: /* Some use that instead of NMI_VECTOR */ | |
641 | xen_vector = XEN_NMI_VECTOR; | |
642 | break; | |
643 | #endif | |
f447d56d BG |
644 | default: |
645 | xen_vector = -1; | |
646 | printk(KERN_ERR "xen: vector 0x%x is not implemented\n", | |
647 | vector); | |
648 | } | |
649 | ||
650 | return xen_vector; | |
651 | } | |
652 | ||
653 | void xen_send_IPI_mask(const struct cpumask *mask, | |
654 | int vector) | |
655 | { | |
656 | int xen_vector = xen_map_vector(vector); | |
657 | ||
658 | if (xen_vector >= 0) | |
659 | __xen_send_IPI_mask(mask, xen_vector); | |
660 | } | |
661 | ||
662 | void xen_send_IPI_all(int vector) | |
663 | { | |
664 | int xen_vector = xen_map_vector(vector); | |
665 | ||
666 | if (xen_vector >= 0) | |
667 | __xen_send_IPI_mask(cpu_online_mask, xen_vector); | |
668 | } | |
669 | ||
670 | void xen_send_IPI_self(int vector) | |
671 | { | |
672 | int xen_vector = xen_map_vector(vector); | |
673 | ||
674 | if (xen_vector >= 0) | |
675 | xen_send_IPI_one(smp_processor_id(), xen_vector); | |
676 | } | |
677 | ||
678 | void xen_send_IPI_mask_allbutself(const struct cpumask *mask, | |
679 | int vector) | |
680 | { | |
681 | unsigned cpu; | |
682 | unsigned int this_cpu = smp_processor_id(); | |
1db01b49 | 683 | int xen_vector = xen_map_vector(vector); |
f447d56d | 684 | |
1db01b49 | 685 | if (!(num_online_cpus() > 1) || (xen_vector < 0)) |
f447d56d BG |
686 | return; |
687 | ||
688 | for_each_cpu_and(cpu, mask, cpu_online_mask) { | |
689 | if (this_cpu == cpu) | |
690 | continue; | |
691 | ||
1db01b49 | 692 | xen_send_IPI_one(cpu, xen_vector); |
f447d56d BG |
693 | } |
694 | } | |
695 | ||
696 | void xen_send_IPI_allbutself(int vector) | |
697 | { | |
1db01b49 | 698 | xen_send_IPI_mask_allbutself(cpu_online_mask, vector); |
f447d56d BG |
699 | } |
700 | ||
f87e4cac JF |
701 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id) |
702 | { | |
f87e4cac | 703 | irq_enter(); |
3b16cf87 | 704 | generic_smp_call_function_interrupt(); |
1b437c8c | 705 | inc_irq_stat(irq_call_count); |
f87e4cac JF |
706 | irq_exit(); |
707 | ||
f87e4cac JF |
708 | return IRQ_HANDLED; |
709 | } | |
710 | ||
3b16cf87 | 711 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id) |
f87e4cac | 712 | { |
3b16cf87 JA |
713 | irq_enter(); |
714 | generic_smp_call_function_single_interrupt(); | |
1b437c8c | 715 | inc_irq_stat(irq_call_count); |
3b16cf87 | 716 | irq_exit(); |
f87e4cac | 717 | |
3b16cf87 | 718 | return IRQ_HANDLED; |
f87e4cac | 719 | } |
a9e7062d | 720 | |
1ff2b0c3 LM |
721 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id) |
722 | { | |
723 | irq_enter(); | |
724 | irq_work_run(); | |
725 | inc_irq_stat(apic_irq_work_irqs); | |
726 | irq_exit(); | |
727 | ||
728 | return IRQ_HANDLED; | |
729 | } | |
730 | ||
b53cedeb | 731 | static const struct smp_ops xen_smp_ops __initconst = { |
a9e7062d JF |
732 | .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, |
733 | .smp_prepare_cpus = xen_smp_prepare_cpus, | |
a9e7062d JF |
734 | .smp_cpus_done = xen_smp_cpus_done, |
735 | ||
d68d82af AN |
736 | .cpu_up = xen_cpu_up, |
737 | .cpu_die = xen_cpu_die, | |
738 | .cpu_disable = xen_cpu_disable, | |
739 | .play_dead = xen_play_dead, | |
740 | ||
76fac077 | 741 | .stop_other_cpus = xen_stop_other_cpus, |
a9e7062d JF |
742 | .smp_send_reschedule = xen_smp_send_reschedule, |
743 | ||
744 | .send_call_func_ipi = xen_smp_send_call_function_ipi, | |
745 | .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi, | |
746 | }; | |
747 | ||
748 | void __init xen_smp_init(void) | |
749 | { | |
750 | smp_ops = xen_smp_ops; | |
c7b75947 | 751 | xen_fill_possible_map(); |
a9e7062d | 752 | } |
99bbb3a8 SS |
753 | |
754 | static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus) | |
755 | { | |
756 | native_smp_prepare_cpus(max_cpus); | |
757 | WARN_ON(xen_smp_intr_init(0)); | |
758 | ||
99bbb3a8 | 759 | xen_init_lock_cpu(0); |
99bbb3a8 SS |
760 | } |
761 | ||
148f9bb8 | 762 | static int xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle) |
99bbb3a8 SS |
763 | { |
764 | int rc; | |
2a442c9c PM |
765 | |
766 | /* | |
767 | * This can happen if CPU was offlined earlier and | |
768 | * offlining timed out in common_cpu_die(). | |
769 | */ | |
770 | if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) { | |
771 | xen_smp_intr_free(cpu); | |
772 | xen_uninit_lock_cpu(cpu); | |
773 | } | |
774 | ||
fc78d343 CA |
775 | /* |
776 | * xen_smp_intr_init() needs to run before native_cpu_up() | |
777 | * so that IPI vectors are set up on the booting CPU before | |
778 | * it is marked online in native_cpu_up(). | |
779 | */ | |
780 | rc = xen_smp_intr_init(cpu); | |
781 | WARN_ON(rc); | |
782 | if (!rc) | |
783 | rc = native_cpu_up(cpu, tidle); | |
1fb3a8b2 KRW |
784 | |
785 | /* | |
786 | * We must initialize the slowpath CPU kicker _after_ the native | |
787 | * path has executed. If we initialized it before none of the | |
788 | * unlocker IPI kicks would reach the booting CPU as the booting | |
789 | * CPU had not set itself 'online' in cpu_online_mask. That mask | |
790 | * is checked when IPIs are sent (on HVM at least). | |
791 | */ | |
792 | xen_init_lock_cpu(cpu); | |
99bbb3a8 SS |
793 | return rc; |
794 | } | |
795 | ||
99bbb3a8 SS |
796 | void __init xen_hvm_smp_init(void) |
797 | { | |
3c05c4be SS |
798 | if (!xen_have_vector_callback) |
799 | return; | |
99bbb3a8 SS |
800 | smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus; |
801 | smp_ops.smp_send_reschedule = xen_smp_send_reschedule; | |
802 | smp_ops.cpu_up = xen_hvm_cpu_up; | |
2a442c9c | 803 | smp_ops.cpu_die = xen_cpu_die; |
99bbb3a8 SS |
804 | smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi; |
805 | smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi; | |
26a79995 | 806 | smp_ops.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu; |
99bbb3a8 | 807 | } |