arch/x86/xen/smp.c

   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.
  14  */
  15 #include <linux/sched.h>
  16 #include <linux/err.h>
  17 #include <linux/slab.h>
  18 #include <linux/smp.h>
  19
  20 #include <asm/paravirt.h>
  21 #include <asm/desc.h>
  22 #include <asm/pgtable.h>
  23 #include <asm/cpu.h>
  24
  25 #include <xen/interface/xen.h>
  26 #include <xen/interface/vcpu.h>
  27
  28 #include <asm/xen/interface.h>
  29 #include <asm/xen/hypercall.h>
  30
  31 #include <xen/xen.h>
  32 #include <xen/page.h>
  33 #include <xen/events.h>
  34
  35 #include "xen-ops.h"
  36 #include "mmu.h"
  37
  38 cpumask_var_t xen_cpu_initialized_map;
  39
  40 static DEFINE_PER_CPU(int, xen_resched_irq);
  41 static DEFINE_PER_CPU(int, xen_callfunc_irq);
  42 static DEFINE_PER_CPU(int, xen_callfuncsingle_irq);
  43 static DEFINE_PER_CPU(int, xen_debug_irq) = -1;
  44
  45 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
  46 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
  47
  48 /*
  49  * Reschedule call back. Nothing to do,
  50  * all the work is done automatically when
  51  * we return from the interrupt.
  52  */
  53 static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
  54 {
  55         inc_irq_stat(irq_resched_count);
  56
  57         return IRQ_HANDLED;
  58 }
  59
  60 static __cpuinit void cpu_bringup(void)
  61 {
  62         int cpu = smp_processor_id();
  63
  64         cpu_init();
  65         touch_softlockup_watchdog();
  66         preempt_disable();
  67
  68         xen_enable_sysenter();
  69         xen_enable_syscall();
  70
  71         cpu = smp_processor_id();
  72         smp_store_cpu_info(cpu);
  73         cpu_data(cpu).x86_max_cores = 1;
  74         set_cpu_sibling_map(cpu);
  75
  76         xen_setup_cpu_clockevents();
  77
  78         set_cpu_online(cpu, true);
  79         percpu_write(cpu_state, CPU_ONLINE);
  80         wmb();
  81
  82         /* We can take interrupts now: we're officially "up". */
  83         local_irq_enable();
  84
  85         wmb();                  /* make sure everything is out */
  86 }
  87
  88 static __cpuinit void cpu_bringup_and_idle(void)
  89 {
  90         cpu_bringup();
  91         cpu_idle();
  92 }
  93
  94 static int xen_smp_intr_init(unsigned int cpu)
  95 {
  96         int rc;
  97         const char *resched_name, *callfunc_name, *debug_name;
  98
  99         resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
 100         rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
 101                                     cpu,
 102                                     xen_reschedule_interrupt,
 103                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
 104                                     resched_name,
 105                                     NULL);
 106         if (rc < 0)
 107                 goto fail;
 108         per_cpu(xen_resched_irq, cpu) = rc;
 109
 110         callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu);
 111         rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR,
 112                                     cpu,
 113                                     xen_call_function_interrupt,
 114                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
 115                                     callfunc_name,
 116                                     NULL);
 117         if (rc < 0)
 118                 goto fail;
 119         per_cpu(xen_callfunc_irq, cpu) = rc;
 120
 121         debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu);
 122         rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt,
 123                                      IRQF_DISABLED | IRQF_PERCPU | IRQF_NOBALANCING,
 124                                      debug_name, NULL);
 125         if (rc < 0)
 126                 goto fail;
 127         per_cpu(xen_debug_irq, cpu) = rc;
 128
 129         callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu);
 130         rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR,
 131                                     cpu,
 132                                     xen_call_function_single_interrupt,
 133                                     IRQF_DISABLED|IRQF_PERCPU|IRQF_NOBALANCING,
 134                                     callfunc_name,
 135                                     NULL);
 136         if (rc < 0)
 137                 goto fail;
 138         per_cpu(xen_callfuncsingle_irq, cpu) = rc;
 139
 140         return 0;
 141
 142  fail:
 143         if (per_cpu(xen_resched_irq, cpu) >= 0)
 144                 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
 145         if (per_cpu(xen_callfunc_irq, cpu) >= 0)
 146                 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
 147         if (per_cpu(xen_debug_irq, cpu) >= 0)
 148                 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
 149         if (per_cpu(xen_callfuncsingle_irq, cpu) >= 0)
 150                 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu),
 151                                        NULL);
 152
 153         return rc;
 154 }
 155
 156 static void __init xen_fill_possible_map(void)
 157 {
 158         int i, rc;
 159
 160         if (xen_initial_domain())
 161                 return;
 162
 163         for (i = 0; i < nr_cpu_ids; i++) {
 164                 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
 165                 if (rc >= 0) {
 166                         num_processors++;
 167                         set_cpu_possible(i, true);
 168                 }
 169         }
 170 }
 171
 172 static void __init xen_filter_cpu_maps(void)
 173 {
 174         int i, rc;
 175
 176         if (!xen_initial_domain())
 177                 return;
 178
 179         num_processors = 0;
 180         disabled_cpus = 0;
 181         for (i = 0; i < nr_cpu_ids; i++) {
 182                 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
 183                 if (rc >= 0) {
 184                         num_processors++;
 185                         set_cpu_possible(i, true);
 186                 } else {
 187                         set_cpu_possible(i, false);
 188                         set_cpu_present(i, false);
 189                 }
 190         }
 191 }
 192
 193 static void __init xen_smp_prepare_boot_cpu(void)
 194 {
 195         BUG_ON(smp_processor_id() != 0);
 196         native_smp_prepare_boot_cpu();
 197
 198         /* We've switched to the "real" per-cpu gdt, so make sure the
 199            old memory can be recycled */
 200         make_lowmem_page_readwrite(xen_initial_gdt);
 201
 202         xen_filter_cpu_maps();
 203         xen_setup_vcpu_info_placement();
 204 }
 205
 206 static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
 207 {
 208         unsigned cpu;
 209
 210         xen_init_lock_cpu(0);
 211
 212         smp_store_cpu_info(0);
 213         cpu_data(0).x86_max_cores = 1;
 214         set_cpu_sibling_map(0);
 215
 216         if (xen_smp_intr_init(0))
 217                 BUG();
 218
 219         if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
 220                 panic("could not allocate xen_cpu_initialized_map\n");
 221
 222         cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
 223
 224         /* Restrict the possible_map according to max_cpus. */
 225         while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
 226                 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
 227                         continue;
 228                 set_cpu_possible(cpu, false);
 229         }
 230
 231         for_each_possible_cpu (cpu) {
 232                 struct task_struct *idle;
 233
 234                 if (cpu == 0)
 235                         continue;
 236
 237                 idle = fork_idle(cpu);
 238                 if (IS_ERR(idle))
 239                         panic("failed fork for CPU %d", cpu);
 240
 241                 set_cpu_present(cpu, true);
 242         }
 243 }
 244
 245 static __cpuinit int
 246 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
 247 {
 248         struct vcpu_guest_context *ctxt;
 249         struct desc_struct *gdt;
 250         unsigned long gdt_mfn;
 251
 252         if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
 253                 return 0;
 254
 255         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
 256         if (ctxt == NULL)
 257                 return -ENOMEM;
 258
 259         gdt = get_cpu_gdt_table(cpu);
 260
 261         ctxt->flags = VGCF_IN_KERNEL;
 262         ctxt->user_regs.ds = __USER_DS;
 263         ctxt->user_regs.es = __USER_DS;
 264         ctxt->user_regs.ss = __KERNEL_DS;
 265 #ifdef CONFIG_X86_32
 266         ctxt->user_regs.fs = __KERNEL_PERCPU;
 267         ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
 268 #else
 269         ctxt->gs_base_kernel = per_cpu_offset(cpu);
 270 #endif
 271         ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
 272         ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
 273
 274         memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
 275
 276         xen_copy_trap_info(ctxt->trap_ctxt);
 277
 278         ctxt->ldt_ents = 0;
 279
 280         BUG_ON((unsigned long)gdt & ~PAGE_MASK);
 281
 282         gdt_mfn = arbitrary_virt_to_mfn(gdt);
 283         make_lowmem_page_readonly(gdt);
 284         make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
 285
 286         ctxt->gdt_frames[0] = gdt_mfn;
 287         ctxt->gdt_ents      = GDT_ENTRIES;
 288
 289         ctxt->user_regs.cs = __KERNEL_CS;
 290         ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
 291
 292         ctxt->kernel_ss = __KERNEL_DS;
 293         ctxt->kernel_sp = idle->thread.sp0;
 294
 295 #ifdef CONFIG_X86_32
 296         ctxt->event_callback_cs     = __KERNEL_CS;
 297         ctxt->failsafe_callback_cs  = __KERNEL_CS;
 298 #endif
 299         ctxt->event_callback_eip    = (unsigned long)xen_hypervisor_callback;
 300         ctxt->failsafe_callback_eip = (unsigned long)xen_failsafe_callback;
 301
 302         per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
 303         ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
 304
 305         if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
 306                 BUG();
 307
 308         kfree(ctxt);
 309         return 0;
 310 }
 311
 312 static int __cpuinit xen_cpu_up(unsigned int cpu)
 313 {
 314         struct task_struct *idle = idle_task(cpu);
 315         int rc;
 316
 317         per_cpu(current_task, cpu) = idle;
 318 #ifdef CONFIG_X86_32
 319         irq_ctx_init(cpu);
 320 #else
 321         clear_tsk_thread_flag(idle, TIF_FORK);
 322         per_cpu(kernel_stack, cpu) =
 323                 (unsigned long)task_stack_page(idle) -
 324                 KERNEL_STACK_OFFSET + THREAD_SIZE;
 325 #endif
 326         xen_setup_runstate_info(cpu);
 327         xen_setup_timer(cpu);
 328         xen_init_lock_cpu(cpu);
 329
 330         per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
 331
 332         /* make sure interrupts start blocked */
 333         per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
 334
 335         rc = cpu_initialize_context(cpu, idle);
 336         if (rc)
 337                 return rc;
 338
 339         if (num_online_cpus() == 1)
 340                 alternatives_smp_switch(1);
 341
 342         rc = xen_smp_intr_init(cpu);
 343         if (rc)
 344                 return rc;
 345
 346         rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
 347         BUG_ON(rc);
 348
 349         while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
 350                 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
 351                 barrier();
 352         }
 353
 354         return 0;
 355 }
 356
 357 static void xen_smp_cpus_done(unsigned int max_cpus)
 358 {
 359 }
 360
 361 #ifdef CONFIG_HOTPLUG_CPU
 362 static int xen_cpu_disable(void)
 363 {
 364         unsigned int cpu = smp_processor_id();
 365         if (cpu == 0)
 366                 return -EBUSY;
 367
 368         cpu_disable_common();
 369
 370         load_cr3(swapper_pg_dir);
 371         return 0;
 372 }
 373
 374 static void xen_cpu_die(unsigned int cpu)
 375 {
 376         while (HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
 377                 current->state = TASK_UNINTERRUPTIBLE;
 378                 schedule_timeout(HZ/10);
 379         }
 380         unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
 381         unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
 382         unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
 383         unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
 384         xen_uninit_lock_cpu(cpu);
 385         xen_teardown_timer(cpu);
 386
 387         if (num_online_cpus() == 1)
 388                 alternatives_smp_switch(0);
 389 }
 390
 391 static void __cpuinit xen_play_dead(void) /* used only with HOTPLUG_CPU */
 392 {
 393         play_dead_common();
 394         HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
 395         cpu_bringup();
 396 }
 397
 398 #else /* !CONFIG_HOTPLUG_CPU */
 399 static int xen_cpu_disable(void)
 400 {
 401         return -ENOSYS;
 402 }
 403
 404 static void xen_cpu_die(unsigned int cpu)
 405 {
 406         BUG();
 407 }
 408
 409 static void xen_play_dead(void)
 410 {
 411         BUG();
 412 }
 413
 414 #endif
 415 static void stop_self(void *v)
 416 {
 417         int cpu = smp_processor_id();
 418
 419         /* make sure we're not pinning something down */
 420         load_cr3(swapper_pg_dir);
 421         /* should set up a minimal gdt */
 422
 423         set_cpu_online(cpu, false);
 424
 425         HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL);
 426         BUG();
 427 }
 428
 429 static void xen_stop_other_cpus(int wait)
 430 {
 431         smp_call_function(stop_self, NULL, wait);
 432 }
 433
 434 static void xen_smp_send_reschedule(int cpu)
 435 {
 436         xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
 437 }
 438
 439 static void xen_send_IPI_mask(const struct cpumask *mask,
 440                               enum ipi_vector vector)
 441 {
 442         unsigned cpu;
 443
 444         for_each_cpu_and(cpu, mask, cpu_online_mask)
 445                 xen_send_IPI_one(cpu, vector);
 446 }
 447
 448 static void xen_smp_send_call_function_ipi(const struct cpumask *mask)
 449 {
 450         int cpu;
 451
 452         xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
 453
 454         /* Make sure other vcpus get a chance to run if they need to. */
 455         for_each_cpu(cpu, mask) {
 456                 if (xen_vcpu_stolen(cpu)) {
 457                         HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
 458                         break;
 459                 }
 460         }
 461 }
 462
 463 static void xen_smp_send_call_function_single_ipi(int cpu)
 464 {
 465         xen_send_IPI_mask(cpumask_of(cpu),
 466                           XEN_CALL_FUNCTION_SINGLE_VECTOR);
 467 }
 468
 469 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
 470 {
 471         irq_enter();
 472         generic_smp_call_function_interrupt();
 473         inc_irq_stat(irq_call_count);
 474         irq_exit();
 475
 476         return IRQ_HANDLED;
 477 }
 478
 479 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
 480 {
 481         irq_enter();
 482         generic_smp_call_function_single_interrupt();
 483         inc_irq_stat(irq_call_count);
 484         irq_exit();
 485
 486         return IRQ_HANDLED;
 487 }
 488
 489 static const struct smp_ops xen_smp_ops __initdata = {
 490         .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
 491         .smp_prepare_cpus = xen_smp_prepare_cpus,
 492         .smp_cpus_done = xen_smp_cpus_done,
 493
 494         .cpu_up = xen_cpu_up,
 495         .cpu_die = xen_cpu_die,
 496         .cpu_disable = xen_cpu_disable,
 497         .play_dead = xen_play_dead,
 498
 499         .stop_other_cpus = xen_stop_other_cpus,
 500         .smp_send_reschedule = xen_smp_send_reschedule,
 501
 502         .send_call_func_ipi = xen_smp_send_call_function_ipi,
 503         .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
 504 };
 505
 506 void __init xen_smp_init(void)
 507 {
 508         smp_ops = xen_smp_ops;
 509         xen_fill_possible_map();
 510         xen_init_spinlocks();
 511 }
 512
 513 static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus)
 514 {
 515         native_smp_prepare_cpus(max_cpus);
 516         WARN_ON(xen_smp_intr_init(0));
 517
 518         if (!xen_have_vector_callback)
 519                 return;
 520         xen_init_lock_cpu(0);
 521         xen_init_spinlocks();
 522 }
 523
 524 static int __cpuinit xen_hvm_cpu_up(unsigned int cpu)
 525 {
 526         int rc;
 527         rc = native_cpu_up(cpu);
 528         WARN_ON (xen_smp_intr_init(cpu));
 529         return rc;
 530 }
 531
 532 static void xen_hvm_cpu_die(unsigned int cpu)
 533 {
 534         unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu), NULL);
 535         unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu), NULL);
 536         unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu), NULL);
 537         unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu), NULL);
 538         native_cpu_die(cpu);
 539 }
 540
 541 void __init xen_hvm_smp_init(void)
 542 {
 543         smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
 544         smp_ops.smp_send_reschedule = xen_smp_send_reschedule;
 545         smp_ops.cpu_up = xen_hvm_cpu_up;
 546         smp_ops.cpu_die = xen_hvm_cpu_die;
 547         smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
 548         smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
 549 }