Commit | Line | Data |
---|---|---|
f87e4cac JF |
1 | /* |
2 | * Xen SMP support | |
3 | * | |
4 | * This file implements the Xen versions of smp_ops. SMP under Xen is | |
5 | * very straightforward. Bringing a CPU up is simply a matter of | |
6 | * loading its initial context and setting it running. | |
7 | * | |
8 | * IPIs are handled through the Xen event mechanism. | |
9 | * | |
10 | * Because virtual CPUs can be scheduled onto any real CPU, there's no | |
11 | * useful topology information for the kernel to make use of. As a | |
12 | * result, all CPUs are treated as if they're single-core and | |
13 | * single-threaded. | |
f87e4cac JF |
14 | */ |
15 | #include <linux/sched.h> | |
16 | #include <linux/err.h> | |
5a0e3ad6 | 17 | #include <linux/slab.h> |
f87e4cac | 18 | #include <linux/smp.h> |
1ff2b0c3 | 19 | #include <linux/irq_work.h> |
466318a8 | 20 | #include <linux/tick.h> |
f87e4cac JF |
21 | |
22 | #include <asm/paravirt.h> | |
23 | #include <asm/desc.h> | |
24 | #include <asm/pgtable.h> | |
25 | #include <asm/cpu.h> | |
26 | ||
27 | #include <xen/interface/xen.h> | |
28 | #include <xen/interface/vcpu.h> | |
29 | ||
30 | #include <asm/xen/interface.h> | |
31 | #include <asm/xen/hypercall.h> | |
32 | ||
ea5b8f73 | 33 | #include <xen/xen.h> |
f87e4cac JF |
34 | #include <xen/page.h> |
35 | #include <xen/events.h> | |
36 | ||
ed467e69 | 37 | #include <xen/hvc-console.h> |
f87e4cac JF |
38 | #include "xen-ops.h" |
39 | #include "mmu.h" | |
40 | ||
b78936e1 | 41 | cpumask_var_t xen_cpu_initialized_map; |
f87e4cac | 42 | |
9547689f KRW |
43 | struct xen_common_irq { |
44 | int irq; | |
45 | char *name; | |
46 | }; | |
ee336e10 KRW |
47 | static DEFINE_PER_CPU(struct xen_common_irq, xen_resched_irq) = { .irq = -1 }; |
48 | static DEFINE_PER_CPU(struct xen_common_irq, xen_callfunc_irq) = { .irq = -1 }; | |
49 | static DEFINE_PER_CPU(struct xen_common_irq, xen_callfuncsingle_irq) = { .irq = -1 }; | |
50 | static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 }; | |
9547689f | 51 | static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 }; |
f87e4cac JF |
52 | |
53 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id); | |
3b16cf87 | 54 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id); |
1ff2b0c3 | 55 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id); |
f87e4cac JF |
56 | |
57 | /* | |
184748cc | 58 | * Reschedule call back. |
f87e4cac JF |
59 | */ |
60 | static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id) | |
61 | { | |
1b437c8c | 62 | inc_irq_stat(irq_resched_count); |
184748cc | 63 | scheduler_ipi(); |
38bb5ab4 | 64 | |
f87e4cac JF |
65 | return IRQ_HANDLED; |
66 | } | |
67 | ||
148f9bb8 | 68 | static void cpu_bringup(void) |
f87e4cac | 69 | { |
e8c9e788 | 70 | int cpu; |
f87e4cac JF |
71 | |
72 | cpu_init(); | |
d68d82af | 73 | touch_softlockup_watchdog(); |
c7b75947 JF |
74 | preempt_disable(); |
75 | ||
e2a81baf | 76 | xen_enable_sysenter(); |
6fcac6d3 | 77 | xen_enable_syscall(); |
f87e4cac | 78 | |
c7b75947 JF |
79 | cpu = smp_processor_id(); |
80 | smp_store_cpu_info(cpu); | |
81 | cpu_data(cpu).x86_max_cores = 1; | |
82 | set_cpu_sibling_map(cpu); | |
f87e4cac JF |
83 | |
84 | xen_setup_cpu_clockevents(); | |
85 | ||
106b4438 KRW |
86 | notify_cpu_starting(cpu); |
87 | ||
d7d3756c | 88 | set_cpu_online(cpu, true); |
106b4438 | 89 | |
2113f469 | 90 | this_cpu_write(cpu_state, CPU_ONLINE); |
106b4438 | 91 | |
c7b75947 JF |
92 | wmb(); |
93 | ||
f87e4cac JF |
94 | /* We can take interrupts now: we're officially "up". */ |
95 | local_irq_enable(); | |
96 | ||
97 | wmb(); /* make sure everything is out */ | |
d68d82af AN |
98 | } |
99 | ||
148f9bb8 | 100 | static void cpu_bringup_and_idle(void) |
d68d82af AN |
101 | { |
102 | cpu_bringup(); | |
7d1a9417 | 103 | cpu_startup_entry(CPUHP_ONLINE); |
f87e4cac JF |
104 | } |
105 | ||
53b94fdc KRW |
106 | static void xen_smp_intr_free(unsigned int cpu) |
107 | { | |
ee336e10 | 108 | if (per_cpu(xen_resched_irq, cpu).irq >= 0) { |
9547689f | 109 | unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu).irq, NULL); |
ee336e10 | 110 | per_cpu(xen_resched_irq, cpu).irq = -1; |
b85fffec KRW |
111 | kfree(per_cpu(xen_resched_irq, cpu).name); |
112 | per_cpu(xen_resched_irq, cpu).name = NULL; | |
ee336e10 KRW |
113 | } |
114 | if (per_cpu(xen_callfunc_irq, cpu).irq >= 0) { | |
9547689f | 115 | unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu).irq, NULL); |
ee336e10 | 116 | per_cpu(xen_callfunc_irq, cpu).irq = -1; |
b85fffec KRW |
117 | kfree(per_cpu(xen_callfunc_irq, cpu).name); |
118 | per_cpu(xen_callfunc_irq, cpu).name = NULL; | |
ee336e10 KRW |
119 | } |
120 | if (per_cpu(xen_debug_irq, cpu).irq >= 0) { | |
9547689f | 121 | unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu).irq, NULL); |
ee336e10 | 122 | per_cpu(xen_debug_irq, cpu).irq = -1; |
b85fffec KRW |
123 | kfree(per_cpu(xen_debug_irq, cpu).name); |
124 | per_cpu(xen_debug_irq, cpu).name = NULL; | |
ee336e10 KRW |
125 | } |
126 | if (per_cpu(xen_callfuncsingle_irq, cpu).irq >= 0) { | |
9547689f | 127 | unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu).irq, |
53b94fdc | 128 | NULL); |
ee336e10 | 129 | per_cpu(xen_callfuncsingle_irq, cpu).irq = -1; |
b85fffec KRW |
130 | kfree(per_cpu(xen_callfuncsingle_irq, cpu).name); |
131 | per_cpu(xen_callfuncsingle_irq, cpu).name = NULL; | |
ee336e10 | 132 | } |
53b94fdc KRW |
133 | if (xen_hvm_domain()) |
134 | return; | |
135 | ||
ee336e10 | 136 | if (per_cpu(xen_irq_work, cpu).irq >= 0) { |
9547689f | 137 | unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL); |
ee336e10 | 138 | per_cpu(xen_irq_work, cpu).irq = -1; |
b85fffec KRW |
139 | kfree(per_cpu(xen_irq_work, cpu).name); |
140 | per_cpu(xen_irq_work, cpu).name = NULL; | |
ee336e10 | 141 | } |
53b94fdc | 142 | }; |
f87e4cac JF |
143 | static int xen_smp_intr_init(unsigned int cpu) |
144 | { | |
145 | int rc; | |
b85fffec | 146 | char *resched_name, *callfunc_name, *debug_name; |
f87e4cac JF |
147 | |
148 | resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu); | |
149 | rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR, | |
150 | cpu, | |
151 | xen_reschedule_interrupt, | |
152 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
153 | resched_name, | |
154 | NULL); | |
155 | if (rc < 0) | |
156 | goto fail; | |
9547689f | 157 | per_cpu(xen_resched_irq, cpu).irq = rc; |
b85fffec | 158 | per_cpu(xen_resched_irq, cpu).name = resched_name; |
f87e4cac JF |
159 | |
160 | callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu); | |
161 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR, | |
162 | cpu, | |
163 | xen_call_function_interrupt, | |
164 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
165 | callfunc_name, | |
166 | NULL); | |
167 | if (rc < 0) | |
168 | goto fail; | |
9547689f | 169 | per_cpu(xen_callfunc_irq, cpu).irq = rc; |
b85fffec | 170 | per_cpu(xen_callfunc_irq, cpu).name = callfunc_name; |
f87e4cac | 171 | |
ee523ca1 JF |
172 | debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu); |
173 | rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt, | |
174 | IRQF_DISABLED | IRQF_PERCPU | IRQF_NOBALANCING, | |
175 | debug_name, NULL); | |
176 | if (rc < 0) | |
177 | goto fail; | |
9547689f | 178 | per_cpu(xen_debug_irq, cpu).irq = rc; |
b85fffec | 179 | per_cpu(xen_debug_irq, cpu).name = debug_name; |
ee523ca1 | 180 | |
3b16cf87 JA |
181 | callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu); |
182 | rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR, | |
183 | cpu, | |
184 | xen_call_function_single_interrupt, | |
185 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
186 | callfunc_name, | |
187 | NULL); | |
188 | if (rc < 0) | |
189 | goto fail; | |
9547689f | 190 | per_cpu(xen_callfuncsingle_irq, cpu).irq = rc; |
b85fffec | 191 | per_cpu(xen_callfuncsingle_irq, cpu).name = callfunc_name; |
3b16cf87 | 192 | |
27d8b207 KRW |
193 | /* |
194 | * The IRQ worker on PVHVM goes through the native path and uses the | |
195 | * IPI mechanism. | |
196 | */ | |
197 | if (xen_hvm_domain()) | |
198 | return 0; | |
199 | ||
1ff2b0c3 LM |
200 | callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu); |
201 | rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR, | |
202 | cpu, | |
203 | xen_irq_work_interrupt, | |
204 | IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING, | |
205 | callfunc_name, | |
206 | NULL); | |
207 | if (rc < 0) | |
208 | goto fail; | |
9547689f | 209 | per_cpu(xen_irq_work, cpu).irq = rc; |
b85fffec | 210 | per_cpu(xen_irq_work, cpu).name = callfunc_name; |
1ff2b0c3 | 211 | |
f87e4cac JF |
212 | return 0; |
213 | ||
214 | fail: | |
53b94fdc | 215 | xen_smp_intr_free(cpu); |
f87e4cac JF |
216 | return rc; |
217 | } | |
218 | ||
c7b75947 | 219 | static void __init xen_fill_possible_map(void) |
f87e4cac JF |
220 | { |
221 | int i, rc; | |
222 | ||
ea5b8f73 SS |
223 | if (xen_initial_domain()) |
224 | return; | |
225 | ||
226 | for (i = 0; i < nr_cpu_ids; i++) { | |
227 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); | |
228 | if (rc >= 0) { | |
229 | num_processors++; | |
230 | set_cpu_possible(i, true); | |
231 | } | |
232 | } | |
233 | } | |
234 | ||
235 | static void __init xen_filter_cpu_maps(void) | |
236 | { | |
237 | int i, rc; | |
cf405ae6 | 238 | unsigned int subtract = 0; |
ea5b8f73 SS |
239 | |
240 | if (!xen_initial_domain()) | |
241 | return; | |
242 | ||
801fd14a SS |
243 | num_processors = 0; |
244 | disabled_cpus = 0; | |
e7986739 | 245 | for (i = 0; i < nr_cpu_ids; i++) { |
f87e4cac | 246 | rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL); |
4560a294 JF |
247 | if (rc >= 0) { |
248 | num_processors++; | |
4f062896 | 249 | set_cpu_possible(i, true); |
801fd14a SS |
250 | } else { |
251 | set_cpu_possible(i, false); | |
252 | set_cpu_present(i, false); | |
cf405ae6 | 253 | subtract++; |
4560a294 | 254 | } |
f87e4cac | 255 | } |
cf405ae6 KRW |
256 | #ifdef CONFIG_HOTPLUG_CPU |
257 | /* This is akin to using 'nr_cpus' on the Linux command line. | |
258 | * Which is OK as when we use 'dom0_max_vcpus=X' we can only | |
259 | * have up to X, while nr_cpu_ids is greater than X. This | |
260 | * normally is not a problem, except when CPU hotplugging | |
261 | * is involved and then there might be more than X CPUs | |
262 | * in the guest - which will not work as there is no | |
263 | * hypercall to expand the max number of VCPUs an already | |
264 | * running guest has. So cap it up to X. */ | |
265 | if (subtract) | |
266 | nr_cpu_ids = nr_cpu_ids - subtract; | |
267 | #endif | |
268 | ||
f87e4cac JF |
269 | } |
270 | ||
a9e7062d | 271 | static void __init xen_smp_prepare_boot_cpu(void) |
f87e4cac | 272 | { |
f87e4cac JF |
273 | BUG_ON(smp_processor_id() != 0); |
274 | native_smp_prepare_boot_cpu(); | |
275 | ||
26a79995 KRW |
276 | if (xen_pv_domain()) { |
277 | /* We've switched to the "real" per-cpu gdt, so make sure the | |
278 | old memory can be recycled */ | |
279 | make_lowmem_page_readwrite(xen_initial_gdt); | |
60223a32 | 280 | |
7cde9b27 FZ |
281 | #ifdef CONFIG_X86_32 |
282 | /* | |
283 | * Xen starts us with XEN_FLAT_RING1_DS, but linux code | |
284 | * expects __USER_DS | |
285 | */ | |
286 | loadsegment(ds, __USER_DS); | |
287 | loadsegment(es, __USER_DS); | |
288 | #endif | |
289 | ||
26a79995 KRW |
290 | xen_filter_cpu_maps(); |
291 | xen_setup_vcpu_info_placement(); | |
292 | } | |
293 | /* | |
294 | * The alternative logic (which patches the unlock/lock) runs before | |
295 | * the smp bootup up code is activated. Hence we need to set this up | |
296 | * the core kernel is being patched. Otherwise we will have only | |
297 | * modules patched but not core code. | |
298 | */ | |
bf7aab3a | 299 | xen_init_spinlocks(); |
f87e4cac JF |
300 | } |
301 | ||
a9e7062d | 302 | static void __init xen_smp_prepare_cpus(unsigned int max_cpus) |
f87e4cac JF |
303 | { |
304 | unsigned cpu; | |
900cba88 | 305 | unsigned int i; |
f87e4cac | 306 | |
ed467e69 KRW |
307 | if (skip_ioapic_setup) { |
308 | char *m = (max_cpus == 0) ? | |
309 | "The nosmp parameter is incompatible with Xen; " \ | |
310 | "use Xen dom0_max_vcpus=1 parameter" : | |
311 | "The noapic parameter is incompatible with Xen"; | |
312 | ||
313 | xen_raw_printk(m); | |
314 | panic(m); | |
315 | } | |
2d9e1e2f JF |
316 | xen_init_lock_cpu(0); |
317 | ||
06d0b5d9 | 318 | smp_store_boot_cpu_info(); |
c7b75947 | 319 | cpu_data(0).x86_max_cores = 1; |
900cba88 AJ |
320 | |
321 | for_each_possible_cpu(i) { | |
322 | zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL); | |
323 | zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL); | |
324 | zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL); | |
325 | } | |
f87e4cac JF |
326 | set_cpu_sibling_map(0); |
327 | ||
328 | if (xen_smp_intr_init(0)) | |
329 | BUG(); | |
330 | ||
b78936e1 MT |
331 | if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL)) |
332 | panic("could not allocate xen_cpu_initialized_map\n"); | |
333 | ||
334 | cpumask_copy(xen_cpu_initialized_map, cpumask_of(0)); | |
f87e4cac JF |
335 | |
336 | /* Restrict the possible_map according to max_cpus. */ | |
337 | while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) { | |
e7986739 | 338 | for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--) |
f87e4cac | 339 | continue; |
4f062896 | 340 | set_cpu_possible(cpu, false); |
f87e4cac JF |
341 | } |
342 | ||
7eb43a6d | 343 | for_each_possible_cpu(cpu) |
4f062896 | 344 | set_cpu_present(cpu, true); |
f87e4cac JF |
345 | } |
346 | ||
148f9bb8 | 347 | static int |
f87e4cac JF |
348 | cpu_initialize_context(unsigned int cpu, struct task_struct *idle) |
349 | { | |
350 | struct vcpu_guest_context *ctxt; | |
c7b75947 | 351 | struct desc_struct *gdt; |
9976b39b | 352 | unsigned long gdt_mfn; |
f87e4cac | 353 | |
b78936e1 | 354 | if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map)) |
f87e4cac JF |
355 | return 0; |
356 | ||
357 | ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); | |
358 | if (ctxt == NULL) | |
359 | return -ENOMEM; | |
360 | ||
c7b75947 JF |
361 | gdt = get_cpu_gdt_table(cpu); |
362 | ||
f87e4cac | 363 | ctxt->flags = VGCF_IN_KERNEL; |
f87e4cac | 364 | ctxt->user_regs.ss = __KERNEL_DS; |
c7b75947 JF |
365 | #ifdef CONFIG_X86_32 |
366 | ctxt->user_regs.fs = __KERNEL_PERCPU; | |
577eebea | 367 | ctxt->user_regs.gs = __KERNEL_STACK_CANARY; |
795f99b6 JF |
368 | #else |
369 | ctxt->gs_base_kernel = per_cpu_offset(cpu); | |
c7b75947 | 370 | #endif |
f87e4cac | 371 | ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle; |
f87e4cac JF |
372 | |
373 | memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt)); | |
374 | ||
dacd45f4 KRW |
375 | { |
376 | ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */ | |
377 | ctxt->user_regs.ds = __USER_DS; | |
378 | ctxt->user_regs.es = __USER_DS; | |
f87e4cac | 379 | |
dacd45f4 | 380 | xen_copy_trap_info(ctxt->trap_ctxt); |
f87e4cac | 381 | |
dacd45f4 | 382 | ctxt->ldt_ents = 0; |
9976b39b | 383 | |
dacd45f4 | 384 | BUG_ON((unsigned long)gdt & ~PAGE_MASK); |
f87e4cac | 385 | |
dacd45f4 KRW |
386 | gdt_mfn = arbitrary_virt_to_mfn(gdt); |
387 | make_lowmem_page_readonly(gdt); | |
388 | make_lowmem_page_readonly(mfn_to_virt(gdt_mfn)); | |
f87e4cac | 389 | |
dacd45f4 KRW |
390 | ctxt->gdt_frames[0] = gdt_mfn; |
391 | ctxt->gdt_ents = GDT_ENTRIES; | |
f87e4cac | 392 | |
dacd45f4 KRW |
393 | ctxt->kernel_ss = __KERNEL_DS; |
394 | ctxt->kernel_sp = idle->thread.sp0; | |
f87e4cac | 395 | |
c7b75947 | 396 | #ifdef CONFIG_X86_32 |
dacd45f4 KRW |
397 | ctxt->event_callback_cs = __KERNEL_CS; |
398 | ctxt->failsafe_callback_cs = __KERNEL_CS; | |
c7b75947 | 399 | #endif |
dacd45f4 KRW |
400 | ctxt->event_callback_eip = |
401 | (unsigned long)xen_hypervisor_callback; | |
402 | ctxt->failsafe_callback_eip = | |
403 | (unsigned long)xen_failsafe_callback; | |
404 | } | |
405 | ctxt->user_regs.cs = __KERNEL_CS; | |
406 | ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs); | |
f87e4cac JF |
407 | |
408 | per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir); | |
409 | ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir)); | |
410 | ||
411 | if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt)) | |
412 | BUG(); | |
413 | ||
414 | kfree(ctxt); | |
415 | return 0; | |
416 | } | |
417 | ||
148f9bb8 | 418 | static int xen_cpu_up(unsigned int cpu, struct task_struct *idle) |
f87e4cac | 419 | { |
f87e4cac JF |
420 | int rc; |
421 | ||
c6f5e0ac | 422 | per_cpu(current_task, cpu) = idle; |
c7b75947 | 423 | #ifdef CONFIG_X86_32 |
f87e4cac | 424 | irq_ctx_init(cpu); |
c7b75947 | 425 | #else |
c7b75947 | 426 | clear_tsk_thread_flag(idle, TIF_FORK); |
38341432 JF |
427 | per_cpu(kernel_stack, cpu) = |
428 | (unsigned long)task_stack_page(idle) - | |
429 | KERNEL_STACK_OFFSET + THREAD_SIZE; | |
c7b75947 | 430 | #endif |
02889672 | 431 | xen_setup_runstate_info(cpu); |
f87e4cac | 432 | xen_setup_timer(cpu); |
2d9e1e2f | 433 | xen_init_lock_cpu(cpu); |
f87e4cac | 434 | |
c7b75947 JF |
435 | per_cpu(cpu_state, cpu) = CPU_UP_PREPARE; |
436 | ||
f87e4cac JF |
437 | /* make sure interrupts start blocked */ |
438 | per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1; | |
439 | ||
440 | rc = cpu_initialize_context(cpu, idle); | |
441 | if (rc) | |
442 | return rc; | |
443 | ||
444 | if (num_online_cpus() == 1) | |
816afe4f RR |
445 | /* Just in case we booted with a single CPU. */ |
446 | alternatives_enable_smp(); | |
f87e4cac JF |
447 | |
448 | rc = xen_smp_intr_init(cpu); | |
449 | if (rc) | |
450 | return rc; | |
451 | ||
f87e4cac JF |
452 | rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL); |
453 | BUG_ON(rc); | |
454 | ||
c7b75947 | 455 | while(per_cpu(cpu_state, cpu) != CPU_ONLINE) { |
1207cf8e | 456 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
c7b75947 JF |
457 | barrier(); |
458 | } | |
459 | ||
f87e4cac JF |
460 | return 0; |
461 | } | |
462 | ||
a9e7062d | 463 | static void xen_smp_cpus_done(unsigned int max_cpus) |
f87e4cac JF |
464 | { |
465 | } | |
466 | ||
2737146b | 467 | #ifdef CONFIG_HOTPLUG_CPU |
26fd1051 | 468 | static int xen_cpu_disable(void) |
d68d82af AN |
469 | { |
470 | unsigned int cpu = smp_processor_id(); | |
471 | if (cpu == 0) | |
472 | return -EBUSY; | |
473 | ||
474 | cpu_disable_common(); | |
475 | ||
476 | load_cr3(swapper_pg_dir); | |
477 | return 0; | |
478 | } | |
479 | ||
26fd1051 | 480 | static void xen_cpu_die(unsigned int cpu) |
d68d82af | 481 | { |
b12abaa1 | 482 | while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) { |
d68d82af AN |
483 | current->state = TASK_UNINTERRUPTIBLE; |
484 | schedule_timeout(HZ/10); | |
485 | } | |
53b94fdc | 486 | xen_smp_intr_free(cpu); |
d68d82af AN |
487 | xen_uninit_lock_cpu(cpu); |
488 | xen_teardown_timer(cpu); | |
d68d82af AN |
489 | } |
490 | ||
148f9bb8 | 491 | static void xen_play_dead(void) /* used only with HOTPLUG_CPU */ |
d68d82af AN |
492 | { |
493 | play_dead_common(); | |
494 | HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL); | |
495 | cpu_bringup(); | |
466318a8 KRW |
496 | /* |
497 | * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down) | |
498 | * clears certain data that the cpu_idle loop (which called us | |
499 | * and that we return from) expects. The only way to get that | |
500 | * data back is to call: | |
501 | */ | |
502 | tick_nohz_idle_enter(); | |
d68d82af AN |
503 | } |
504 | ||
2737146b | 505 | #else /* !CONFIG_HOTPLUG_CPU */ |
26fd1051 | 506 | static int xen_cpu_disable(void) |
2737146b AN |
507 | { |
508 | return -ENOSYS; | |
509 | } | |
510 | ||
26fd1051 | 511 | static void xen_cpu_die(unsigned int cpu) |
2737146b AN |
512 | { |
513 | BUG(); | |
514 | } | |
515 | ||
26fd1051 | 516 | static void xen_play_dead(void) |
2737146b AN |
517 | { |
518 | BUG(); | |
519 | } | |
520 | ||
521 | #endif | |
f87e4cac JF |
522 | static void stop_self(void *v) |
523 | { | |
524 | int cpu = smp_processor_id(); | |
525 | ||
526 | /* make sure we're not pinning something down */ | |
527 | load_cr3(swapper_pg_dir); | |
528 | /* should set up a minimal gdt */ | |
529 | ||
086748e5 IC |
530 | set_cpu_online(cpu, false); |
531 | ||
f87e4cac JF |
532 | HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL); |
533 | BUG(); | |
534 | } | |
535 | ||
76fac077 | 536 | static void xen_stop_other_cpus(int wait) |
f87e4cac | 537 | { |
76fac077 | 538 | smp_call_function(stop_self, NULL, wait); |
f87e4cac JF |
539 | } |
540 | ||
a9e7062d | 541 | static void xen_smp_send_reschedule(int cpu) |
f87e4cac JF |
542 | { |
543 | xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR); | |
544 | } | |
545 | ||
f447d56d BG |
546 | static void __xen_send_IPI_mask(const struct cpumask *mask, |
547 | int vector) | |
f87e4cac JF |
548 | { |
549 | unsigned cpu; | |
550 | ||
bcda016e | 551 | for_each_cpu_and(cpu, mask, cpu_online_mask) |
f87e4cac JF |
552 | xen_send_IPI_one(cpu, vector); |
553 | } | |
554 | ||
bcda016e | 555 | static void xen_smp_send_call_function_ipi(const struct cpumask *mask) |
3b16cf87 JA |
556 | { |
557 | int cpu; | |
558 | ||
f447d56d | 559 | __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR); |
3b16cf87 JA |
560 | |
561 | /* Make sure other vcpus get a chance to run if they need to. */ | |
bcda016e | 562 | for_each_cpu(cpu, mask) { |
3b16cf87 | 563 | if (xen_vcpu_stolen(cpu)) { |
1207cf8e | 564 | HYPERVISOR_sched_op(SCHEDOP_yield, NULL); |
3b16cf87 JA |
565 | break; |
566 | } | |
567 | } | |
568 | } | |
569 | ||
a9e7062d | 570 | static void xen_smp_send_call_function_single_ipi(int cpu) |
3b16cf87 | 571 | { |
f447d56d | 572 | __xen_send_IPI_mask(cpumask_of(cpu), |
e7986739 | 573 | XEN_CALL_FUNCTION_SINGLE_VECTOR); |
3b16cf87 JA |
574 | } |
575 | ||
f447d56d BG |
576 | static inline int xen_map_vector(int vector) |
577 | { | |
578 | int xen_vector; | |
579 | ||
580 | switch (vector) { | |
581 | case RESCHEDULE_VECTOR: | |
582 | xen_vector = XEN_RESCHEDULE_VECTOR; | |
583 | break; | |
584 | case CALL_FUNCTION_VECTOR: | |
585 | xen_vector = XEN_CALL_FUNCTION_VECTOR; | |
586 | break; | |
587 | case CALL_FUNCTION_SINGLE_VECTOR: | |
588 | xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR; | |
589 | break; | |
1ff2b0c3 LM |
590 | case IRQ_WORK_VECTOR: |
591 | xen_vector = XEN_IRQ_WORK_VECTOR; | |
592 | break; | |
6efa20e4 KRW |
593 | #ifdef CONFIG_X86_64 |
594 | case NMI_VECTOR: | |
595 | case APIC_DM_NMI: /* Some use that instead of NMI_VECTOR */ | |
596 | xen_vector = XEN_NMI_VECTOR; | |
597 | break; | |
598 | #endif | |
f447d56d BG |
599 | default: |
600 | xen_vector = -1; | |
601 | printk(KERN_ERR "xen: vector 0x%x is not implemented\n", | |
602 | vector); | |
603 | } | |
604 | ||
605 | return xen_vector; | |
606 | } | |
607 | ||
608 | void xen_send_IPI_mask(const struct cpumask *mask, | |
609 | int vector) | |
610 | { | |
611 | int xen_vector = xen_map_vector(vector); | |
612 | ||
613 | if (xen_vector >= 0) | |
614 | __xen_send_IPI_mask(mask, xen_vector); | |
615 | } | |
616 | ||
617 | void xen_send_IPI_all(int vector) | |
618 | { | |
619 | int xen_vector = xen_map_vector(vector); | |
620 | ||
621 | if (xen_vector >= 0) | |
622 | __xen_send_IPI_mask(cpu_online_mask, xen_vector); | |
623 | } | |
624 | ||
625 | void xen_send_IPI_self(int vector) | |
626 | { | |
627 | int xen_vector = xen_map_vector(vector); | |
628 | ||
629 | if (xen_vector >= 0) | |
630 | xen_send_IPI_one(smp_processor_id(), xen_vector); | |
631 | } | |
632 | ||
633 | void xen_send_IPI_mask_allbutself(const struct cpumask *mask, | |
634 | int vector) | |
635 | { | |
636 | unsigned cpu; | |
637 | unsigned int this_cpu = smp_processor_id(); | |
1db01b49 | 638 | int xen_vector = xen_map_vector(vector); |
f447d56d | 639 | |
1db01b49 | 640 | if (!(num_online_cpus() > 1) || (xen_vector < 0)) |
f447d56d BG |
641 | return; |
642 | ||
643 | for_each_cpu_and(cpu, mask, cpu_online_mask) { | |
644 | if (this_cpu == cpu) | |
645 | continue; | |
646 | ||
1db01b49 | 647 | xen_send_IPI_one(cpu, xen_vector); |
f447d56d BG |
648 | } |
649 | } | |
650 | ||
651 | void xen_send_IPI_allbutself(int vector) | |
652 | { | |
1db01b49 | 653 | xen_send_IPI_mask_allbutself(cpu_online_mask, vector); |
f447d56d BG |
654 | } |
655 | ||
f87e4cac JF |
656 | static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id) |
657 | { | |
f87e4cac | 658 | irq_enter(); |
3b16cf87 | 659 | generic_smp_call_function_interrupt(); |
1b437c8c | 660 | inc_irq_stat(irq_call_count); |
f87e4cac JF |
661 | irq_exit(); |
662 | ||
f87e4cac JF |
663 | return IRQ_HANDLED; |
664 | } | |
665 | ||
3b16cf87 | 666 | static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id) |
f87e4cac | 667 | { |
3b16cf87 JA |
668 | irq_enter(); |
669 | generic_smp_call_function_single_interrupt(); | |
1b437c8c | 670 | inc_irq_stat(irq_call_count); |
3b16cf87 | 671 | irq_exit(); |
f87e4cac | 672 | |
3b16cf87 | 673 | return IRQ_HANDLED; |
f87e4cac | 674 | } |
a9e7062d | 675 | |
1ff2b0c3 LM |
676 | static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id) |
677 | { | |
678 | irq_enter(); | |
679 | irq_work_run(); | |
680 | inc_irq_stat(apic_irq_work_irqs); | |
681 | irq_exit(); | |
682 | ||
683 | return IRQ_HANDLED; | |
684 | } | |
685 | ||
b53cedeb | 686 | static const struct smp_ops xen_smp_ops __initconst = { |
a9e7062d JF |
687 | .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu, |
688 | .smp_prepare_cpus = xen_smp_prepare_cpus, | |
a9e7062d JF |
689 | .smp_cpus_done = xen_smp_cpus_done, |
690 | ||
d68d82af AN |
691 | .cpu_up = xen_cpu_up, |
692 | .cpu_die = xen_cpu_die, | |
693 | .cpu_disable = xen_cpu_disable, | |
694 | .play_dead = xen_play_dead, | |
695 | ||
76fac077 | 696 | .stop_other_cpus = xen_stop_other_cpus, |
a9e7062d JF |
697 | .smp_send_reschedule = xen_smp_send_reschedule, |
698 | ||
699 | .send_call_func_ipi = xen_smp_send_call_function_ipi, | |
700 | .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi, | |
701 | }; | |
702 | ||
703 | void __init xen_smp_init(void) | |
704 | { | |
705 | smp_ops = xen_smp_ops; | |
c7b75947 | 706 | xen_fill_possible_map(); |
a9e7062d | 707 | } |
99bbb3a8 SS |
708 | |
709 | static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus) | |
710 | { | |
711 | native_smp_prepare_cpus(max_cpus); | |
712 | WARN_ON(xen_smp_intr_init(0)); | |
713 | ||
99bbb3a8 | 714 | xen_init_lock_cpu(0); |
99bbb3a8 SS |
715 | } |
716 | ||
148f9bb8 | 717 | static int xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle) |
99bbb3a8 SS |
718 | { |
719 | int rc; | |
fc78d343 CA |
720 | /* |
721 | * xen_smp_intr_init() needs to run before native_cpu_up() | |
722 | * so that IPI vectors are set up on the booting CPU before | |
723 | * it is marked online in native_cpu_up(). | |
724 | */ | |
725 | rc = xen_smp_intr_init(cpu); | |
726 | WARN_ON(rc); | |
727 | if (!rc) | |
728 | rc = native_cpu_up(cpu, tidle); | |
1fb3a8b2 KRW |
729 | |
730 | /* | |
731 | * We must initialize the slowpath CPU kicker _after_ the native | |
732 | * path has executed. If we initialized it before none of the | |
733 | * unlocker IPI kicks would reach the booting CPU as the booting | |
734 | * CPU had not set itself 'online' in cpu_online_mask. That mask | |
735 | * is checked when IPIs are sent (on HVM at least). | |
736 | */ | |
737 | xen_init_lock_cpu(cpu); | |
99bbb3a8 SS |
738 | return rc; |
739 | } | |
740 | ||
741 | static void xen_hvm_cpu_die(unsigned int cpu) | |
742 | { | |
b12abaa1 | 743 | xen_cpu_die(cpu); |
99bbb3a8 SS |
744 | native_cpu_die(cpu); |
745 | } | |
746 | ||
747 | void __init xen_hvm_smp_init(void) | |
748 | { | |
3c05c4be SS |
749 | if (!xen_have_vector_callback) |
750 | return; | |
99bbb3a8 SS |
751 | smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus; |
752 | smp_ops.smp_send_reschedule = xen_smp_send_reschedule; | |
753 | smp_ops.cpu_up = xen_hvm_cpu_up; | |
754 | smp_ops.cpu_die = xen_hvm_cpu_die; | |
755 | smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi; | |
756 | smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi; | |
26a79995 | 757 | smp_ops.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu; |
99bbb3a8 | 758 | } |