2 * KVM paravirt_ops implementation
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.
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.
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.
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>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/kvm_para.h>
26 #include <linux/cpu.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>
38 #include <asm/traps.h>
40 #include <asm/tlbflush.h>
43 #include <asm/apicdef.h>
44 #include <asm/hypervisor.h>
45 #include <asm/kvm_guest.h>
47 static int kvmapf = 1;
49 static int parse_no_kvmapf(char *arg)
55 early_param("no-kvmapf", parse_no_kvmapf);
57 static int steal_acc = 1;
58 static int parse_no_stealacc(char *arg)
64 early_param("no-steal-acc", parse_no_stealacc);
66 static int kvmclock_vsyscall = 1;
67 static int parse_no_kvmclock_vsyscall(char *arg)
69 kvmclock_vsyscall = 0;
73 early_param("no-kvmclock-vsyscall", parse_no_kvmclock_vsyscall);
75 static DEFINE_PER_CPU(struct kvm_vcpu_pv_apf_data, apf_reason) __aligned(64);
76 static DEFINE_PER_CPU(struct kvm_steal_time, steal_time) __aligned(64);
77 static int has_steal_clock = 0;
80 * No need for any "IO delay" on KVM
82 static void kvm_io_delay(void)
86 #define KVM_TASK_SLEEP_HASHBITS 8
87 #define KVM_TASK_SLEEP_HASHSIZE (1<<KVM_TASK_SLEEP_HASHBITS)
89 struct kvm_task_sleep_node {
90 struct hlist_node link;
97 static struct kvm_task_sleep_head {
99 struct hlist_head list;
100 } async_pf_sleepers[KVM_TASK_SLEEP_HASHSIZE];
102 static struct kvm_task_sleep_node *_find_apf_task(struct kvm_task_sleep_head *b,
105 struct hlist_node *p;
107 hlist_for_each(p, &b->list) {
108 struct kvm_task_sleep_node *n =
109 hlist_entry(p, typeof(*n), link);
110 if (n->token == token)
117 void kvm_async_pf_task_wait(u32 token)
119 u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
120 struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
121 struct kvm_task_sleep_node n, *e;
125 e = _find_apf_task(b, token);
127 /* dummy entry exist -> wake up was delivered ahead of PF */
130 spin_unlock(&b->lock);
135 n.cpu = smp_processor_id();
136 n.halted = is_idle_task(current) || preempt_count() > 1;
137 init_waitqueue_head(&n.wq);
138 hlist_add_head(&n.link, &b->list);
139 spin_unlock(&b->lock);
143 prepare_to_wait(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
144 if (hlist_unhashed(&n.link))
153 * We cannot reschedule. So halt.
160 finish_wait(&n.wq, &wait);
164 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait);
166 static void apf_task_wake_one(struct kvm_task_sleep_node *n)
168 hlist_del_init(&n->link);
170 smp_send_reschedule(n->cpu);
171 else if (waitqueue_active(&n->wq))
175 static void apf_task_wake_all(void)
179 for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) {
180 struct hlist_node *p, *next;
181 struct kvm_task_sleep_head *b = &async_pf_sleepers[i];
183 hlist_for_each_safe(p, next, &b->list) {
184 struct kvm_task_sleep_node *n =
185 hlist_entry(p, typeof(*n), link);
186 if (n->cpu == smp_processor_id())
187 apf_task_wake_one(n);
189 spin_unlock(&b->lock);
193 void kvm_async_pf_task_wake(u32 token)
195 u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
196 struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
197 struct kvm_task_sleep_node *n;
206 n = _find_apf_task(b, token);
209 * async PF was not yet handled.
210 * Add dummy entry for the token.
212 n = kzalloc(sizeof(*n), GFP_ATOMIC);
215 * Allocation failed! Busy wait while other cpu
218 spin_unlock(&b->lock);
223 n->cpu = smp_processor_id();
224 init_waitqueue_head(&n->wq);
225 hlist_add_head(&n->link, &b->list);
227 apf_task_wake_one(n);
228 spin_unlock(&b->lock);
231 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake);
233 u32 kvm_read_and_reset_pf_reason(void)
237 if (__get_cpu_var(apf_reason).enabled) {
238 reason = __get_cpu_var(apf_reason).reason;
239 __get_cpu_var(apf_reason).reason = 0;
244 EXPORT_SYMBOL_GPL(kvm_read_and_reset_pf_reason);
246 dotraplinkage void __kprobes
247 do_async_page_fault(struct pt_regs *regs, unsigned long error_code)
249 switch (kvm_read_and_reset_pf_reason()) {
251 do_page_fault(regs, error_code);
253 case KVM_PV_REASON_PAGE_NOT_PRESENT:
254 /* page is swapped out by the host. */
257 kvm_async_pf_task_wait((u32)read_cr2());
260 case KVM_PV_REASON_PAGE_READY:
263 kvm_async_pf_task_wake((u32)read_cr2());
269 static void __init paravirt_ops_setup(void)
271 pv_info.name = "KVM";
272 pv_info.paravirt_enabled = 1;
274 if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
275 pv_cpu_ops.io_delay = kvm_io_delay;
277 #ifdef CONFIG_X86_IO_APIC
282 static void kvm_register_steal_time(void)
284 int cpu = smp_processor_id();
285 struct kvm_steal_time *st = &per_cpu(steal_time, cpu);
287 if (!has_steal_clock)
290 memset(st, 0, sizeof(*st));
292 wrmsrl(MSR_KVM_STEAL_TIME, (__pa(st) | KVM_MSR_ENABLED));
293 printk(KERN_INFO "kvm-stealtime: cpu %d, msr %lx\n",
297 static DEFINE_PER_CPU(unsigned long, kvm_apic_eoi) = KVM_PV_EOI_DISABLED;
299 static void kvm_guest_apic_eoi_write(u32 reg, u32 val)
302 * This relies on __test_and_clear_bit to modify the memory
303 * in a way that is atomic with respect to the local CPU.
304 * The hypervisor only accesses this memory from the local CPU so
305 * there's no need for lock or memory barriers.
306 * An optimization barrier is implied in apic write.
308 if (__test_and_clear_bit(KVM_PV_EOI_BIT, &__get_cpu_var(kvm_apic_eoi)))
310 apic_write(APIC_EOI, APIC_EOI_ACK);
313 void __cpuinit kvm_guest_cpu_init(void)
315 if (!kvm_para_available())
318 if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF) && kvmapf) {
319 u64 pa = __pa(&__get_cpu_var(apf_reason));
321 #ifdef CONFIG_PREEMPT
322 pa |= KVM_ASYNC_PF_SEND_ALWAYS;
324 wrmsrl(MSR_KVM_ASYNC_PF_EN, pa | KVM_ASYNC_PF_ENABLED);
325 __get_cpu_var(apf_reason).enabled = 1;
326 printk(KERN_INFO"KVM setup async PF for cpu %d\n",
330 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI)) {
332 /* Size alignment is implied but just to make it explicit. */
333 BUILD_BUG_ON(__alignof__(kvm_apic_eoi) < 4);
334 __get_cpu_var(kvm_apic_eoi) = 0;
335 pa = __pa(&__get_cpu_var(kvm_apic_eoi)) | KVM_MSR_ENABLED;
336 wrmsrl(MSR_KVM_PV_EOI_EN, pa);
340 kvm_register_steal_time();
343 static void kvm_pv_disable_apf(void)
345 if (!__get_cpu_var(apf_reason).enabled)
348 wrmsrl(MSR_KVM_ASYNC_PF_EN, 0);
349 __get_cpu_var(apf_reason).enabled = 0;
351 printk(KERN_INFO"Unregister pv shared memory for cpu %d\n",
355 static void kvm_pv_guest_cpu_reboot(void *unused)
358 * We disable PV EOI before we load a new kernel by kexec,
359 * since MSR_KVM_PV_EOI_EN stores a pointer into old kernel's memory.
360 * New kernel can re-enable when it boots.
362 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
363 wrmsrl(MSR_KVM_PV_EOI_EN, 0);
364 kvm_pv_disable_apf();
365 kvm_disable_steal_time();
368 static int kvm_pv_reboot_notify(struct notifier_block *nb,
369 unsigned long code, void *unused)
371 if (code == SYS_RESTART)
372 on_each_cpu(kvm_pv_guest_cpu_reboot, NULL, 1);
376 static struct notifier_block kvm_pv_reboot_nb = {
377 .notifier_call = kvm_pv_reboot_notify,
380 static u64 kvm_steal_clock(int cpu)
383 struct kvm_steal_time *src;
386 src = &per_cpu(steal_time, cpu);
388 version = src->version;
392 } while ((version & 1) || (version != src->version));
397 void kvm_disable_steal_time(void)
399 if (!has_steal_clock)
402 wrmsr(MSR_KVM_STEAL_TIME, 0, 0);
406 static void __init kvm_smp_prepare_boot_cpu(void)
408 WARN_ON(kvm_register_clock("primary cpu clock"));
409 kvm_guest_cpu_init();
410 native_smp_prepare_boot_cpu();
413 static void __cpuinit kvm_guest_cpu_online(void *dummy)
415 kvm_guest_cpu_init();
418 static void kvm_guest_cpu_offline(void *dummy)
420 kvm_disable_steal_time();
421 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
422 wrmsrl(MSR_KVM_PV_EOI_EN, 0);
423 kvm_pv_disable_apf();
427 static int __cpuinit kvm_cpu_notify(struct notifier_block *self,
428 unsigned long action, void *hcpu)
430 int cpu = (unsigned long)hcpu;
433 case CPU_DOWN_FAILED:
434 case CPU_ONLINE_FROZEN:
435 smp_call_function_single(cpu, kvm_guest_cpu_online, NULL, 0);
437 case CPU_DOWN_PREPARE:
438 case CPU_DOWN_PREPARE_FROZEN:
439 smp_call_function_single(cpu, kvm_guest_cpu_offline, NULL, 1);
447 static struct notifier_block __cpuinitdata kvm_cpu_notifier = {
448 .notifier_call = kvm_cpu_notify,
452 static void __init kvm_apf_trap_init(void)
454 set_intr_gate(14, &async_page_fault);
457 void __init kvm_guest_init(void)
461 if (!kvm_para_available())
464 paravirt_ops_setup();
465 register_reboot_notifier(&kvm_pv_reboot_nb);
466 for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++)
467 spin_lock_init(&async_pf_sleepers[i].lock);
468 if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF))
469 x86_init.irqs.trap_init = kvm_apf_trap_init;
471 if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
473 pv_time_ops.steal_clock = kvm_steal_clock;
476 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
477 apic_set_eoi_write(kvm_guest_apic_eoi_write);
479 if (kvmclock_vsyscall)
480 kvm_setup_vsyscall_timeinfo();
483 smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
484 register_cpu_notifier(&kvm_cpu_notifier);
486 kvm_guest_cpu_init();
490 static bool __init kvm_detect(void)
492 if (!kvm_para_available())
497 const struct hypervisor_x86 x86_hyper_kvm __refconst = {
499 .detect = kvm_detect,
501 EXPORT_SYMBOL_GPL(x86_hyper_kvm);
503 static __init int activate_jump_labels(void)
505 if (has_steal_clock) {
506 static_key_slow_inc(¶virt_steal_enabled);
508 static_key_slow_inc(¶virt_steal_rq_enabled);
513 arch_initcall(activate_jump_labels);