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