arch/x86/kernel/kvm.c

   1 /*
   2  * KVM paravirt_ops implementation
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  *
  18  * Copyright (C) 2007, Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
  19  * Copyright IBM Corporation, 2007
  20  *   Authors: Anthony Liguori <aliguori@us.ibm.com>
  21  */
  22
  23 #include <linux/module.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kvm_para.h>
  26 #include <linux/cpu.h>
  27 #include <linux/mm.h>
  28 #include <linux/highmem.h>
  29 #include <linux/hardirq.h>
  30 #include <linux/notifier.h>
  31 #include <linux/reboot.h>
  32 #include <linux/hash.h>
  33 #include <linux/sched.h>
  34 #include <linux/slab.h>
  35 #include <linux/kprobes.h>
  36 #include <asm/timer.h>
  37 #include <asm/cpu.h>
  38 #include <asm/traps.h>
  39 #include <asm/desc.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/idle.h>
  42
  43 static int kvmapf = 1;
  44
  45 static int parse_no_kvmapf(char *arg)
  46 {
  47         kvmapf = 0;
  48         return 0;
  49 }
  50
  51 early_param("no-kvmapf", parse_no_kvmapf);
  52
  53 static int steal_acc = 1;
  54 static int parse_no_stealacc(char *arg)
  55 {
  56         steal_acc = 0;
  57         return 0;
  58 }
  59
  60 early_param("no-steal-acc", parse_no_stealacc);
  61
  62 static DEFINE_PER_CPU(struct kvm_vcpu_pv_apf_data, apf_reason) __aligned(64);
  63 static DEFINE_PER_CPU(struct kvm_steal_time, steal_time) __aligned(64);
  64 static int has_steal_clock = 0;
  65
  66 /*
  67  * No need for any "IO delay" on KVM
  68  */
  69 static void kvm_io_delay(void)
  70 {
  71 }
  72
  73 #define KVM_TASK_SLEEP_HASHBITS 8
  74 #define KVM_TASK_SLEEP_HASHSIZE (1<<KVM_TASK_SLEEP_HASHBITS)
  75
  76 struct kvm_task_sleep_node {
  77         struct hlist_node link;
  78         wait_queue_head_t wq;
  79         u32 token;
  80         int cpu;
  81         bool halted;
  82 };
  83
  84 static struct kvm_task_sleep_head {
  85         spinlock_t lock;
  86         struct hlist_head list;
  87 } async_pf_sleepers[KVM_TASK_SLEEP_HASHSIZE];
  88
  89 static struct kvm_task_sleep_node *_find_apf_task(struct kvm_task_sleep_head *b,
  90                                                   u32 token)
  91 {
  92         struct hlist_node *p;
  93
  94         hlist_for_each(p, &b->list) {
  95                 struct kvm_task_sleep_node *n =
  96                         hlist_entry(p, typeof(*n), link);
  97                 if (n->token == token)
  98                         return n;
  99         }
 100
 101         return NULL;
 102 }
 103
 104 void kvm_async_pf_task_wait(u32 token)
 105 {
 106         u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
 107         struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
 108         struct kvm_task_sleep_node n, *e;
 109         DEFINE_WAIT(wait);
 110         int cpu, idle;
 111
 112         cpu = get_cpu();
 113         idle = idle_cpu(cpu);
 114         put_cpu();
 115
 116         spin_lock(&b->lock);
 117         e = _find_apf_task(b, token);
 118         if (e) {
 119                 /* dummy entry exist -> wake up was delivered ahead of PF */
 120                 hlist_del(&e->link);
 121                 kfree(e);
 122                 spin_unlock(&b->lock);
 123                 return;
 124         }
 125
 126         n.token = token;
 127         n.cpu = smp_processor_id();
 128         n.halted = idle || preempt_count() > 1;
 129         init_waitqueue_head(&n.wq);
 130         hlist_add_head(&n.link, &b->list);
 131         spin_unlock(&b->lock);
 132
 133         for (;;) {
 134                 if (!n.halted)
 135                         prepare_to_wait(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
 136                 if (hlist_unhashed(&n.link))
 137                         break;
 138
 139                 if (!n.halted) {
 140                         local_irq_enable();
 141                         schedule();
 142                         local_irq_disable();
 143                 } else {
 144                         /*
 145                          * We cannot reschedule. So halt.
 146                          */
 147                         native_safe_halt();
 148                         local_irq_disable();
 149                 }
 150         }
 151         if (!n.halted)
 152                 finish_wait(&n.wq, &wait);
 153
 154         return;
 155 }
 156 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait);
 157
 158 static void apf_task_wake_one(struct kvm_task_sleep_node *n)
 159 {
 160         hlist_del_init(&n->link);
 161         if (n->halted)
 162                 smp_send_reschedule(n->cpu);
 163         else if (waitqueue_active(&n->wq))
 164                 wake_up(&n->wq);
 165 }
 166
 167 static void apf_task_wake_all(void)
 168 {
 169         int i;
 170
 171         for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) {
 172                 struct hlist_node *p, *next;
 173                 struct kvm_task_sleep_head *b = &async_pf_sleepers[i];
 174                 spin_lock(&b->lock);
 175                 hlist_for_each_safe(p, next, &b->list) {
 176                         struct kvm_task_sleep_node *n =
 177                                 hlist_entry(p, typeof(*n), link);
 178                         if (n->cpu == smp_processor_id())
 179                                 apf_task_wake_one(n);
 180                 }
 181                 spin_unlock(&b->lock);
 182         }
 183 }
 184
 185 void kvm_async_pf_task_wake(u32 token)
 186 {
 187         u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
 188         struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
 189         struct kvm_task_sleep_node *n;
 190
 191         if (token == ~0) {
 192                 apf_task_wake_all();
 193                 return;
 194         }
 195
 196 again:
 197         spin_lock(&b->lock);
 198         n = _find_apf_task(b, token);
 199         if (!n) {
 200                 /*
 201                  * async PF was not yet handled.
 202                  * Add dummy entry for the token.
 203                  */
 204                 n = kzalloc(sizeof(*n), GFP_ATOMIC);
 205                 if (!n) {
 206                         /*
 207                          * Allocation failed! Busy wait while other cpu
 208                          * handles async PF.
 209                          */
 210                         spin_unlock(&b->lock);
 211                         cpu_relax();
 212                         goto again;
 213                 }
 214                 n->token = token;
 215                 n->cpu = smp_processor_id();
 216                 init_waitqueue_head(&n->wq);
 217                 hlist_add_head(&n->link, &b->list);
 218         } else
 219                 apf_task_wake_one(n);
 220         spin_unlock(&b->lock);
 221         return;
 222 }
 223 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake);
 224
 225 u32 kvm_read_and_reset_pf_reason(void)
 226 {
 227         u32 reason = 0;
 228
 229         if (__get_cpu_var(apf_reason).enabled) {
 230                 reason = __get_cpu_var(apf_reason).reason;
 231                 __get_cpu_var(apf_reason).reason = 0;
 232         }
 233
 234         return reason;
 235 }
 236 EXPORT_SYMBOL_GPL(kvm_read_and_reset_pf_reason);
 237
 238 dotraplinkage void __kprobes
 239 do_async_page_fault(struct pt_regs *regs, unsigned long error_code)
 240 {
 241         switch (kvm_read_and_reset_pf_reason()) {
 242         default:
 243                 do_page_fault(regs, error_code);
 244                 break;
 245         case KVM_PV_REASON_PAGE_NOT_PRESENT:
 246                 /* page is swapped out by the host. */
 247                 kvm_async_pf_task_wait((u32)read_cr2());
 248                 break;
 249         case KVM_PV_REASON_PAGE_READY:
 250                 rcu_irq_enter();
 251                 exit_idle();
 252                 kvm_async_pf_task_wake((u32)read_cr2());
 253                 rcu_irq_exit();
 254                 break;
 255         }
 256 }
 257
 258 static void __init paravirt_ops_setup(void)
 259 {
 260         pv_info.name = "KVM";
 261         pv_info.paravirt_enabled = 1;
 262
 263         if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
 264                 pv_cpu_ops.io_delay = kvm_io_delay;
 265
 266 #ifdef CONFIG_X86_IO_APIC
 267         no_timer_check = 1;
 268 #endif
 269 }
 270
 271 static void kvm_register_steal_time(void)
 272 {
 273         int cpu = smp_processor_id();
 274         struct kvm_steal_time *st = &per_cpu(steal_time, cpu);
 275
 276         if (!has_steal_clock)
 277                 return;
 278
 279         memset(st, 0, sizeof(*st));
 280
 281         wrmsrl(MSR_KVM_STEAL_TIME, (__pa(st) | KVM_MSR_ENABLED));
 282         printk(KERN_INFO "kvm-stealtime: cpu %d, msr %lx\n",
 283                 cpu, __pa(st));
 284 }
 285
 286 void __cpuinit kvm_guest_cpu_init(void)
 287 {
 288         if (!kvm_para_available())
 289                 return;
 290
 291         if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF) && kvmapf) {
 292                 u64 pa = __pa(&__get_cpu_var(apf_reason));
 293
 294 #ifdef CONFIG_PREEMPT
 295                 pa |= KVM_ASYNC_PF_SEND_ALWAYS;
 296 #endif
 297                 wrmsrl(MSR_KVM_ASYNC_PF_EN, pa | KVM_ASYNC_PF_ENABLED);
 298                 __get_cpu_var(apf_reason).enabled = 1;
 299                 printk(KERN_INFO"KVM setup async PF for cpu %d\n",
 300                        smp_processor_id());
 301         }
 302
 303         if (has_steal_clock)
 304                 kvm_register_steal_time();
 305 }
 306
 307 static void kvm_pv_disable_apf(void *unused)
 308 {
 309         if (!__get_cpu_var(apf_reason).enabled)
 310                 return;
 311
 312         wrmsrl(MSR_KVM_ASYNC_PF_EN, 0);
 313         __get_cpu_var(apf_reason).enabled = 0;
 314
 315         printk(KERN_INFO"Unregister pv shared memory for cpu %d\n",
 316                smp_processor_id());
 317 }
 318
 319 static int kvm_pv_reboot_notify(struct notifier_block *nb,
 320                                 unsigned long code, void *unused)
 321 {
 322         if (code == SYS_RESTART)
 323                 on_each_cpu(kvm_pv_disable_apf, NULL, 1);
 324         return NOTIFY_DONE;
 325 }
 326
 327 static struct notifier_block kvm_pv_reboot_nb = {
 328         .notifier_call = kvm_pv_reboot_notify,
 329 };
 330
 331 static u64 kvm_steal_clock(int cpu)
 332 {
 333         u64 steal;
 334         struct kvm_steal_time *src;
 335         int version;
 336
 337         src = &per_cpu(steal_time, cpu);
 338         do {
 339                 version = src->version;
 340                 rmb();
 341                 steal = src->steal;
 342                 rmb();
 343         } while ((version & 1) || (version != src->version));
 344
 345         return steal;
 346 }
 347
 348 void kvm_disable_steal_time(void)
 349 {
 350         if (!has_steal_clock)
 351                 return;
 352
 353         wrmsr(MSR_KVM_STEAL_TIME, 0, 0);
 354 }
 355
 356 #ifdef CONFIG_SMP
 357 static void __init kvm_smp_prepare_boot_cpu(void)
 358 {
 359 #ifdef CONFIG_KVM_CLOCK
 360         WARN_ON(kvm_register_clock("primary cpu clock"));
 361 #endif
 362         kvm_guest_cpu_init();
 363         native_smp_prepare_boot_cpu();
 364 }
 365
 366 static void __cpuinit kvm_guest_cpu_online(void *dummy)
 367 {
 368         kvm_guest_cpu_init();
 369 }
 370
 371 static void kvm_guest_cpu_offline(void *dummy)
 372 {
 373         kvm_disable_steal_time();
 374         kvm_pv_disable_apf(NULL);
 375         apf_task_wake_all();
 376 }
 377
 378 static int __cpuinit kvm_cpu_notify(struct notifier_block *self,
 379                                     unsigned long action, void *hcpu)
 380 {
 381         int cpu = (unsigned long)hcpu;
 382         switch (action) {
 383         case CPU_ONLINE:
 384         case CPU_DOWN_FAILED:
 385         case CPU_ONLINE_FROZEN:
 386                 smp_call_function_single(cpu, kvm_guest_cpu_online, NULL, 0);
 387                 break;
 388         case CPU_DOWN_PREPARE:
 389         case CPU_DOWN_PREPARE_FROZEN:
 390                 smp_call_function_single(cpu, kvm_guest_cpu_offline, NULL, 1);
 391                 break;
 392         default:
 393                 break;
 394         }
 395         return NOTIFY_OK;
 396 }
 397
 398 static struct notifier_block __cpuinitdata kvm_cpu_notifier = {
 399         .notifier_call  = kvm_cpu_notify,
 400 };
 401 #endif
 402
 403 static void __init kvm_apf_trap_init(void)
 404 {
 405         set_intr_gate(14, &async_page_fault);
 406 }
 407
 408 void __init kvm_guest_init(void)
 409 {
 410         int i;
 411
 412         if (!kvm_para_available())
 413                 return;
 414
 415         paravirt_ops_setup();
 416         register_reboot_notifier(&kvm_pv_reboot_nb);
 417         for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++)
 418                 spin_lock_init(&async_pf_sleepers[i].lock);
 419         if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF))
 420                 x86_init.irqs.trap_init = kvm_apf_trap_init;
 421
 422         if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
 423                 has_steal_clock = 1;
 424                 pv_time_ops.steal_clock = kvm_steal_clock;
 425         }
 426
 427 #ifdef CONFIG_SMP
 428         smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
 429         register_cpu_notifier(&kvm_cpu_notifier);
 430 #else
 431         kvm_guest_cpu_init();
 432 #endif
 433 }
 434
 435 static __init int activate_jump_labels(void)
 436 {
 437         if (has_steal_clock) {
 438                 static_key_slow_inc(&paravirt_steal_enabled);
 439                 if (steal_acc)
 440                         static_key_slow_inc(&paravirt_steal_rq_enabled);
 441         }
 442
 443         return 0;
 444 }
 445 arch_initcall(activate_jump_labels);