Merge tag 'pinctrl-v4.8-1' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
[deliverable/linux.git] / arch / x86 / xen / smp.c
CommitLineData
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 44cpumask_var_t xen_cpu_initialized_map;
f87e4cac 45
9547689f
KRW
46struct xen_common_irq {
47 int irq;
48 char *name;
49};
ee336e10
KRW
50static DEFINE_PER_CPU(struct xen_common_irq, xen_resched_irq) = { .irq = -1 };
51static DEFINE_PER_CPU(struct xen_common_irq, xen_callfunc_irq) = { .irq = -1 };
52static DEFINE_PER_CPU(struct xen_common_irq, xen_callfuncsingle_irq) = { .irq = -1 };
53static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 };
9547689f 54static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 };
65d0cf0b 55static DEFINE_PER_CPU(struct xen_common_irq, xen_pmu_irq) = { .irq = -1 };
f87e4cac
JF
56
57static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
3b16cf87 58static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
1ff2b0c3 59static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id);
f87e4cac
JF
60
61/*
184748cc 62 * Reschedule call back.
f87e4cac
JF
63 */
64static 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 72static 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 */
107asmlinkage __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();
fc6d73d6 115 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
f87e4cac
JF
116}
117
53b94fdc
KRW
118static 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
162static 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 250static 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
266static 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 302static 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 }
ee42d665
VK
325
326 /*
327 * Setup vcpu_info for boot CPU.
328 */
329 if (xen_hvm_domain())
330 xen_vcpu_setup(0);
331
26a79995
KRW
332 /*
333 * The alternative logic (which patches the unlock/lock) runs before
334 * the smp bootup up code is activated. Hence we need to set this up
335 * the core kernel is being patched. Otherwise we will have only
336 * modules patched but not core code.
337 */
bf7aab3a 338 xen_init_spinlocks();
f87e4cac
JF
339}
340
a9e7062d 341static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
f87e4cac
JF
342{
343 unsigned cpu;
900cba88 344 unsigned int i;
f87e4cac 345
ed467e69
KRW
346 if (skip_ioapic_setup) {
347 char *m = (max_cpus == 0) ?
348 "The nosmp parameter is incompatible with Xen; " \
349 "use Xen dom0_max_vcpus=1 parameter" :
350 "The noapic parameter is incompatible with Xen";
351
352 xen_raw_printk(m);
353 panic(m);
354 }
2d9e1e2f
JF
355 xen_init_lock_cpu(0);
356
06d0b5d9 357 smp_store_boot_cpu_info();
c7b75947 358 cpu_data(0).x86_max_cores = 1;
900cba88
AJ
359
360 for_each_possible_cpu(i) {
361 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
362 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
363 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
364 }
f87e4cac
JF
365 set_cpu_sibling_map(0);
366
65d0cf0b
BO
367 xen_pmu_init(0);
368
f87e4cac
JF
369 if (xen_smp_intr_init(0))
370 BUG();
371
b78936e1
MT
372 if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
373 panic("could not allocate xen_cpu_initialized_map\n");
374
375 cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
f87e4cac
JF
376
377 /* Restrict the possible_map according to max_cpus. */
378 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
e7986739 379 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
f87e4cac 380 continue;
4f062896 381 set_cpu_possible(cpu, false);
f87e4cac
JF
382 }
383
7eb43a6d 384 for_each_possible_cpu(cpu)
4f062896 385 set_cpu_present(cpu, true);
f87e4cac
JF
386}
387
148f9bb8 388static int
f87e4cac
JF
389cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
390{
391 struct vcpu_guest_context *ctxt;
c7b75947 392 struct desc_struct *gdt;
9976b39b 393 unsigned long gdt_mfn;
f87e4cac 394
ce4b1b16
IM
395 /* used to tell cpu_init() that it can proceed with initialization */
396 cpumask_set_cpu(cpu, cpu_callout_mask);
b78936e1 397 if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
f87e4cac
JF
398 return 0;
399
400 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
401 if (ctxt == NULL)
402 return -ENOMEM;
403
c7b75947
JF
404 gdt = get_cpu_gdt_table(cpu);
405
c7b75947 406#ifdef CONFIG_X86_32
5840c84b 407 /* Note: PVH is not yet supported on x86_32. */
c7b75947 408 ctxt->user_regs.fs = __KERNEL_PERCPU;
577eebea 409 ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
c7b75947 410#endif
f87e4cac
JF
411 memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
412
5840c84b 413 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
a2ef5dc2 414 ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
5840c84b 415 ctxt->flags = VGCF_IN_KERNEL;
dacd45f4
KRW
416 ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
417 ctxt->user_regs.ds = __USER_DS;
418 ctxt->user_regs.es = __USER_DS;
5840c84b 419 ctxt->user_regs.ss = __KERNEL_DS;
f87e4cac 420
dacd45f4 421 xen_copy_trap_info(ctxt->trap_ctxt);
f87e4cac 422
dacd45f4 423 ctxt->ldt_ents = 0;
9976b39b 424
dacd45f4 425 BUG_ON((unsigned long)gdt & ~PAGE_MASK);
f87e4cac 426
dacd45f4
KRW
427 gdt_mfn = arbitrary_virt_to_mfn(gdt);
428 make_lowmem_page_readonly(gdt);
429 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
f87e4cac 430
dacd45f4
KRW
431 ctxt->gdt_frames[0] = gdt_mfn;
432 ctxt->gdt_ents = GDT_ENTRIES;
f87e4cac 433
dacd45f4
KRW
434 ctxt->kernel_ss = __KERNEL_DS;
435 ctxt->kernel_sp = idle->thread.sp0;
f87e4cac 436
c7b75947 437#ifdef CONFIG_X86_32
dacd45f4
KRW
438 ctxt->event_callback_cs = __KERNEL_CS;
439 ctxt->failsafe_callback_cs = __KERNEL_CS;
5840c84b
MR
440#else
441 ctxt->gs_base_kernel = per_cpu_offset(cpu);
c7b75947 442#endif
dacd45f4
KRW
443 ctxt->event_callback_eip =
444 (unsigned long)xen_hypervisor_callback;
445 ctxt->failsafe_callback_eip =
446 (unsigned long)xen_failsafe_callback;
5840c84b
MR
447 ctxt->user_regs.cs = __KERNEL_CS;
448 per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
dacd45f4 449 }
a2ef5dc2
MR
450#ifdef CONFIG_XEN_PVH
451 else {
452 /*
453 * The vcpu comes on kernel page tables which have the NX pte
454 * bit set. This means before DS/SS is touched, NX in
455 * EFER must be set. Hence the following assembly glue code.
5840c84b 456 */
a2ef5dc2 457 ctxt->user_regs.eip = (unsigned long)xen_pvh_early_cpu_init;
5840c84b 458 ctxt->user_regs.rdi = cpu;
a2ef5dc2
MR
459 ctxt->user_regs.rsi = true; /* entry == true */
460 }
5840c84b 461#endif
dacd45f4 462 ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
0df4f266 463 ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_gfn(swapper_pg_dir));
ad5475f9 464 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, xen_vcpu_nr(cpu), ctxt))
f87e4cac
JF
465 BUG();
466
467 kfree(ctxt);
468 return 0;
469}
470
148f9bb8 471static int xen_cpu_up(unsigned int cpu, struct task_struct *idle)
f87e4cac 472{
f87e4cac
JF
473 int rc;
474
3f85483b 475 common_cpu_up(cpu, idle);
4461bbc0 476
02889672 477 xen_setup_runstate_info(cpu);
f87e4cac 478 xen_setup_timer(cpu);
2d9e1e2f 479 xen_init_lock_cpu(cpu);
f87e4cac 480
2a442c9c
PM
481 /*
482 * PV VCPUs are always successfully taken down (see 'while' loop
483 * in xen_cpu_die()), so -EBUSY is an error.
484 */
485 rc = cpu_check_up_prepare(cpu);
486 if (rc)
487 return rc;
c7b75947 488
f87e4cac
JF
489 /* make sure interrupts start blocked */
490 per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
491
492 rc = cpu_initialize_context(cpu, idle);
493 if (rc)
494 return rc;
495
65d0cf0b
BO
496 xen_pmu_init(cpu);
497
f87e4cac
JF
498 rc = xen_smp_intr_init(cpu);
499 if (rc)
500 return rc;
501
ad5475f9 502 rc = HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL);
f87e4cac
JF
503 BUG_ON(rc);
504
2a442c9c 505 while (cpu_report_state(cpu) != CPU_ONLINE)
1207cf8e 506 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
c7b75947 507
f87e4cac
JF
508 return 0;
509}
510
a9e7062d 511static void xen_smp_cpus_done(unsigned int max_cpus)
f87e4cac
JF
512{
513}
514
2737146b 515#ifdef CONFIG_HOTPLUG_CPU
26fd1051 516static int xen_cpu_disable(void)
d68d82af
AN
517{
518 unsigned int cpu = smp_processor_id();
519 if (cpu == 0)
520 return -EBUSY;
521
522 cpu_disable_common();
523
524 load_cr3(swapper_pg_dir);
525 return 0;
526}
527
26fd1051 528static void xen_cpu_die(unsigned int cpu)
d68d82af 529{
ad5475f9
VK
530 while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up,
531 xen_vcpu_nr(cpu), NULL)) {
57b6b99b 532 __set_current_state(TASK_UNINTERRUPTIBLE);
d68d82af
AN
533 schedule_timeout(HZ/10);
534 }
54279552 535
2a442c9c
PM
536 if (common_cpu_die(cpu) == 0) {
537 xen_smp_intr_free(cpu);
538 xen_uninit_lock_cpu(cpu);
539 xen_teardown_timer(cpu);
65d0cf0b 540 xen_pmu_finish(cpu);
2a442c9c 541 }
d68d82af
AN
542}
543
148f9bb8 544static void xen_play_dead(void) /* used only with HOTPLUG_CPU */
d68d82af
AN
545{
546 play_dead_common();
ad5475f9 547 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(smp_processor_id()), NULL);
d68d82af 548 cpu_bringup();
466318a8
KRW
549 /*
550 * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down)
551 * clears certain data that the cpu_idle loop (which called us
552 * and that we return from) expects. The only way to get that
553 * data back is to call:
554 */
555 tick_nohz_idle_enter();
dc6416f1
BO
556
557 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
d68d82af
AN
558}
559
2737146b 560#else /* !CONFIG_HOTPLUG_CPU */
26fd1051 561static int xen_cpu_disable(void)
2737146b
AN
562{
563 return -ENOSYS;
564}
565
26fd1051 566static void xen_cpu_die(unsigned int cpu)
2737146b
AN
567{
568 BUG();
569}
570
26fd1051 571static void xen_play_dead(void)
2737146b
AN
572{
573 BUG();
574}
575
576#endif
f87e4cac
JF
577static void stop_self(void *v)
578{
579 int cpu = smp_processor_id();
580
581 /* make sure we're not pinning something down */
582 load_cr3(swapper_pg_dir);
583 /* should set up a minimal gdt */
584
086748e5
IC
585 set_cpu_online(cpu, false);
586
ad5475f9 587 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL);
f87e4cac
JF
588 BUG();
589}
590
76fac077 591static void xen_stop_other_cpus(int wait)
f87e4cac 592{
76fac077 593 smp_call_function(stop_self, NULL, wait);
f87e4cac
JF
594}
595
a9e7062d 596static void xen_smp_send_reschedule(int cpu)
f87e4cac
JF
597{
598 xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
599}
600
f447d56d
BG
601static void __xen_send_IPI_mask(const struct cpumask *mask,
602 int vector)
f87e4cac
JF
603{
604 unsigned cpu;
605
bcda016e 606 for_each_cpu_and(cpu, mask, cpu_online_mask)
f87e4cac
JF
607 xen_send_IPI_one(cpu, vector);
608}
609
bcda016e 610static void xen_smp_send_call_function_ipi(const struct cpumask *mask)
3b16cf87
JA
611{
612 int cpu;
613
f447d56d 614 __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
3b16cf87
JA
615
616 /* Make sure other vcpus get a chance to run if they need to. */
bcda016e 617 for_each_cpu(cpu, mask) {
3b16cf87 618 if (xen_vcpu_stolen(cpu)) {
1207cf8e 619 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
3b16cf87
JA
620 break;
621 }
622 }
623}
624
a9e7062d 625static void xen_smp_send_call_function_single_ipi(int cpu)
3b16cf87 626{
f447d56d 627 __xen_send_IPI_mask(cpumask_of(cpu),
e7986739 628 XEN_CALL_FUNCTION_SINGLE_VECTOR);
3b16cf87
JA
629}
630
f447d56d
BG
631static inline int xen_map_vector(int vector)
632{
633 int xen_vector;
634
635 switch (vector) {
636 case RESCHEDULE_VECTOR:
637 xen_vector = XEN_RESCHEDULE_VECTOR;
638 break;
639 case CALL_FUNCTION_VECTOR:
640 xen_vector = XEN_CALL_FUNCTION_VECTOR;
641 break;
642 case CALL_FUNCTION_SINGLE_VECTOR:
643 xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR;
644 break;
1ff2b0c3
LM
645 case IRQ_WORK_VECTOR:
646 xen_vector = XEN_IRQ_WORK_VECTOR;
647 break;
6efa20e4
KRW
648#ifdef CONFIG_X86_64
649 case NMI_VECTOR:
650 case APIC_DM_NMI: /* Some use that instead of NMI_VECTOR */
651 xen_vector = XEN_NMI_VECTOR;
652 break;
653#endif
f447d56d
BG
654 default:
655 xen_vector = -1;
656 printk(KERN_ERR "xen: vector 0x%x is not implemented\n",
657 vector);
658 }
659
660 return xen_vector;
661}
662
663void xen_send_IPI_mask(const struct cpumask *mask,
664 int vector)
665{
666 int xen_vector = xen_map_vector(vector);
667
668 if (xen_vector >= 0)
669 __xen_send_IPI_mask(mask, xen_vector);
670}
671
672void xen_send_IPI_all(int vector)
673{
674 int xen_vector = xen_map_vector(vector);
675
676 if (xen_vector >= 0)
677 __xen_send_IPI_mask(cpu_online_mask, xen_vector);
678}
679
680void xen_send_IPI_self(int vector)
681{
682 int xen_vector = xen_map_vector(vector);
683
684 if (xen_vector >= 0)
685 xen_send_IPI_one(smp_processor_id(), xen_vector);
686}
687
688void xen_send_IPI_mask_allbutself(const struct cpumask *mask,
689 int vector)
690{
691 unsigned cpu;
692 unsigned int this_cpu = smp_processor_id();
1db01b49 693 int xen_vector = xen_map_vector(vector);
f447d56d 694
1db01b49 695 if (!(num_online_cpus() > 1) || (xen_vector < 0))
f447d56d
BG
696 return;
697
698 for_each_cpu_and(cpu, mask, cpu_online_mask) {
699 if (this_cpu == cpu)
700 continue;
701
1db01b49 702 xen_send_IPI_one(cpu, xen_vector);
f447d56d
BG
703 }
704}
705
706void xen_send_IPI_allbutself(int vector)
707{
1db01b49 708 xen_send_IPI_mask_allbutself(cpu_online_mask, vector);
f447d56d
BG
709}
710
f87e4cac
JF
711static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
712{
f87e4cac 713 irq_enter();
3b16cf87 714 generic_smp_call_function_interrupt();
1b437c8c 715 inc_irq_stat(irq_call_count);
f87e4cac
JF
716 irq_exit();
717
f87e4cac
JF
718 return IRQ_HANDLED;
719}
720
3b16cf87 721static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
f87e4cac 722{
3b16cf87
JA
723 irq_enter();
724 generic_smp_call_function_single_interrupt();
1b437c8c 725 inc_irq_stat(irq_call_count);
3b16cf87 726 irq_exit();
f87e4cac 727
3b16cf87 728 return IRQ_HANDLED;
f87e4cac 729}
a9e7062d 730
1ff2b0c3
LM
731static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id)
732{
733 irq_enter();
734 irq_work_run();
735 inc_irq_stat(apic_irq_work_irqs);
736 irq_exit();
737
738 return IRQ_HANDLED;
739}
740
b53cedeb 741static const struct smp_ops xen_smp_ops __initconst = {
a9e7062d
JF
742 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
743 .smp_prepare_cpus = xen_smp_prepare_cpus,
a9e7062d
JF
744 .smp_cpus_done = xen_smp_cpus_done,
745
d68d82af
AN
746 .cpu_up = xen_cpu_up,
747 .cpu_die = xen_cpu_die,
748 .cpu_disable = xen_cpu_disable,
749 .play_dead = xen_play_dead,
750
76fac077 751 .stop_other_cpus = xen_stop_other_cpus,
a9e7062d
JF
752 .smp_send_reschedule = xen_smp_send_reschedule,
753
754 .send_call_func_ipi = xen_smp_send_call_function_ipi,
755 .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
756};
757
758void __init xen_smp_init(void)
759{
760 smp_ops = xen_smp_ops;
c7b75947 761 xen_fill_possible_map();
a9e7062d 762}
99bbb3a8
SS
763
764static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus)
765{
766 native_smp_prepare_cpus(max_cpus);
767 WARN_ON(xen_smp_intr_init(0));
768
99bbb3a8 769 xen_init_lock_cpu(0);
99bbb3a8
SS
770}
771
148f9bb8 772static int xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle)
99bbb3a8
SS
773{
774 int rc;
2a442c9c
PM
775
776 /*
777 * This can happen if CPU was offlined earlier and
778 * offlining timed out in common_cpu_die().
779 */
780 if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
781 xen_smp_intr_free(cpu);
782 xen_uninit_lock_cpu(cpu);
783 }
784
fc78d343
CA
785 /*
786 * xen_smp_intr_init() needs to run before native_cpu_up()
787 * so that IPI vectors are set up on the booting CPU before
788 * it is marked online in native_cpu_up().
789 */
790 rc = xen_smp_intr_init(cpu);
791 WARN_ON(rc);
792 if (!rc)
793 rc = native_cpu_up(cpu, tidle);
1fb3a8b2
KRW
794
795 /*
796 * We must initialize the slowpath CPU kicker _after_ the native
797 * path has executed. If we initialized it before none of the
798 * unlocker IPI kicks would reach the booting CPU as the booting
799 * CPU had not set itself 'online' in cpu_online_mask. That mask
800 * is checked when IPIs are sent (on HVM at least).
801 */
802 xen_init_lock_cpu(cpu);
99bbb3a8
SS
803 return rc;
804}
805
99bbb3a8
SS
806void __init xen_hvm_smp_init(void)
807{
3c05c4be
SS
808 if (!xen_have_vector_callback)
809 return;
99bbb3a8
SS
810 smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
811 smp_ops.smp_send_reschedule = xen_smp_send_reschedule;
812 smp_ops.cpu_up = xen_hvm_cpu_up;
2a442c9c 813 smp_ops.cpu_die = xen_cpu_die;
99bbb3a8
SS
814 smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
815 smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
26a79995 816 smp_ops.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu;
99bbb3a8 817}
This page took 0.524031 seconds and 5 git commands to generate.