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;
126 idle = idle_cpu(cpu);
130 e = _find_apf_task(b, token);
132 /* dummy entry exist -> wake up was delivered ahead of PF */
135 spin_unlock(&b->lock);
140 n.cpu = smp_processor_id();
141 n.halted = idle || preempt_count() > 1;
142 init_waitqueue_head(&n.wq);
143 hlist_add_head(&n.link, &b->list);
144 spin_unlock(&b->lock);
148 prepare_to_wait(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
149 if (hlist_unhashed(&n.link))
158 * We cannot reschedule. So halt.
165 finish_wait(&n.wq, &wait);
169 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait);
171 static void apf_task_wake_one(struct kvm_task_sleep_node *n)
173 hlist_del_init(&n->link);
175 smp_send_reschedule(n->cpu);
176 else if (waitqueue_active(&n->wq))
180 static void apf_task_wake_all(void)
184 for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) {
185 struct hlist_node *p, *next;
186 struct kvm_task_sleep_head *b = &async_pf_sleepers[i];
188 hlist_for_each_safe(p, next, &b->list) {
189 struct kvm_task_sleep_node *n =
190 hlist_entry(p, typeof(*n), link);
191 if (n->cpu == smp_processor_id())
192 apf_task_wake_one(n);
194 spin_unlock(&b->lock);
198 void kvm_async_pf_task_wake(u32 token)
200 u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
201 struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
202 struct kvm_task_sleep_node *n;
211 n = _find_apf_task(b, token);
214 * async PF was not yet handled.
215 * Add dummy entry for the token.
217 n = kzalloc(sizeof(*n), GFP_ATOMIC);
220 * Allocation failed! Busy wait while other cpu
223 spin_unlock(&b->lock);
228 n->cpu = smp_processor_id();
229 init_waitqueue_head(&n->wq);
230 hlist_add_head(&n->link, &b->list);
232 apf_task_wake_one(n);
233 spin_unlock(&b->lock);
236 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake);
238 u32 kvm_read_and_reset_pf_reason(void)
242 if (__get_cpu_var(apf_reason).enabled) {
243 reason = __get_cpu_var(apf_reason).reason;
244 __get_cpu_var(apf_reason).reason = 0;
249 EXPORT_SYMBOL_GPL(kvm_read_and_reset_pf_reason);
251 dotraplinkage void __kprobes
252 do_async_page_fault(struct pt_regs *regs, unsigned long error_code)
254 switch (kvm_read_and_reset_pf_reason()) {
256 do_page_fault(regs, error_code);
258 case KVM_PV_REASON_PAGE_NOT_PRESENT:
259 /* page is swapped out by the host. */
262 kvm_async_pf_task_wait((u32)read_cr2());
265 case KVM_PV_REASON_PAGE_READY:
268 kvm_async_pf_task_wake((u32)read_cr2());
274 static void __init paravirt_ops_setup(void)
276 pv_info.name = "KVM";
277 pv_info.paravirt_enabled = 1;
279 if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
280 pv_cpu_ops.io_delay = kvm_io_delay;
282 #ifdef CONFIG_X86_IO_APIC
287 static void kvm_register_steal_time(void)
289 int cpu = smp_processor_id();
290 struct kvm_steal_time *st = &per_cpu(steal_time, cpu);
292 if (!has_steal_clock)
295 memset(st, 0, sizeof(*st));
297 wrmsrl(MSR_KVM_STEAL_TIME, (__pa(st) | KVM_MSR_ENABLED));
298 printk(KERN_INFO "kvm-stealtime: cpu %d, msr %lx\n",
302 static DEFINE_PER_CPU(unsigned long, kvm_apic_eoi) = KVM_PV_EOI_DISABLED;
304 static void kvm_guest_apic_eoi_write(u32 reg, u32 val)
307 * This relies on __test_and_clear_bit to modify the memory
308 * in a way that is atomic with respect to the local CPU.
309 * The hypervisor only accesses this memory from the local CPU so
310 * there's no need for lock or memory barriers.
311 * An optimization barrier is implied in apic write.
313 if (__test_and_clear_bit(KVM_PV_EOI_BIT, &__get_cpu_var(kvm_apic_eoi)))
315 apic_write(APIC_EOI, APIC_EOI_ACK);
318 void __cpuinit kvm_guest_cpu_init(void)
320 if (!kvm_para_available())
323 if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF) && kvmapf) {
324 u64 pa = __pa(&__get_cpu_var(apf_reason));
326 #ifdef CONFIG_PREEMPT
327 pa |= KVM_ASYNC_PF_SEND_ALWAYS;
329 wrmsrl(MSR_KVM_ASYNC_PF_EN, pa | KVM_ASYNC_PF_ENABLED);
330 __get_cpu_var(apf_reason).enabled = 1;
331 printk(KERN_INFO"KVM setup async PF for cpu %d\n",
335 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI)) {
337 /* Size alignment is implied but just to make it explicit. */
338 BUILD_BUG_ON(__alignof__(kvm_apic_eoi) < 4);
339 __get_cpu_var(kvm_apic_eoi) = 0;
340 pa = __pa(&__get_cpu_var(kvm_apic_eoi)) | KVM_MSR_ENABLED;
341 wrmsrl(MSR_KVM_PV_EOI_EN, pa);
345 kvm_register_steal_time();
348 static void kvm_pv_disable_apf(void)
350 if (!__get_cpu_var(apf_reason).enabled)
353 wrmsrl(MSR_KVM_ASYNC_PF_EN, 0);
354 __get_cpu_var(apf_reason).enabled = 0;
356 printk(KERN_INFO"Unregister pv shared memory for cpu %d\n",
360 static void kvm_pv_guest_cpu_reboot(void *unused)
363 * We disable PV EOI before we load a new kernel by kexec,
364 * since MSR_KVM_PV_EOI_EN stores a pointer into old kernel's memory.
365 * New kernel can re-enable when it boots.
367 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
368 wrmsrl(MSR_KVM_PV_EOI_EN, 0);
369 kvm_pv_disable_apf();
370 kvm_disable_steal_time();
373 static int kvm_pv_reboot_notify(struct notifier_block *nb,
374 unsigned long code, void *unused)
376 if (code == SYS_RESTART)
377 on_each_cpu(kvm_pv_guest_cpu_reboot, NULL, 1);
381 static struct notifier_block kvm_pv_reboot_nb = {
382 .notifier_call = kvm_pv_reboot_notify,
385 static u64 kvm_steal_clock(int cpu)
388 struct kvm_steal_time *src;
391 src = &per_cpu(steal_time, cpu);
393 version = src->version;
397 } while ((version & 1) || (version != src->version));
402 void kvm_disable_steal_time(void)
404 if (!has_steal_clock)
407 wrmsr(MSR_KVM_STEAL_TIME, 0, 0);
411 static void __init kvm_smp_prepare_boot_cpu(void)
413 WARN_ON(kvm_register_clock("primary cpu clock"));
414 kvm_guest_cpu_init();
415 native_smp_prepare_boot_cpu();
418 static void __cpuinit kvm_guest_cpu_online(void *dummy)
420 kvm_guest_cpu_init();
423 static void kvm_guest_cpu_offline(void *dummy)
425 kvm_disable_steal_time();
426 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
427 wrmsrl(MSR_KVM_PV_EOI_EN, 0);
428 kvm_pv_disable_apf();
432 static int __cpuinit kvm_cpu_notify(struct notifier_block *self,
433 unsigned long action, void *hcpu)
435 int cpu = (unsigned long)hcpu;
438 case CPU_DOWN_FAILED:
439 case CPU_ONLINE_FROZEN:
440 smp_call_function_single(cpu, kvm_guest_cpu_online, NULL, 0);
442 case CPU_DOWN_PREPARE:
443 case CPU_DOWN_PREPARE_FROZEN:
444 smp_call_function_single(cpu, kvm_guest_cpu_offline, NULL, 1);
452 static struct notifier_block __cpuinitdata kvm_cpu_notifier = {
453 .notifier_call = kvm_cpu_notify,
457 static void __init kvm_apf_trap_init(void)
459 set_intr_gate(14, &async_page_fault);
462 void __init kvm_guest_init(void)
466 if (!kvm_para_available())
469 paravirt_ops_setup();
470 register_reboot_notifier(&kvm_pv_reboot_nb);
471 for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++)
472 spin_lock_init(&async_pf_sleepers[i].lock);
473 if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF))
474 x86_init.irqs.trap_init = kvm_apf_trap_init;
476 if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
478 pv_time_ops.steal_clock = kvm_steal_clock;
481 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
482 apic_set_eoi_write(kvm_guest_apic_eoi_write);
484 if (kvmclock_vsyscall)
485 kvm_setup_vsyscall_timeinfo();
488 smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
489 register_cpu_notifier(&kvm_cpu_notifier);
491 kvm_guest_cpu_init();
495 static bool __init kvm_detect(void)
497 if (!kvm_para_available())
502 const struct hypervisor_x86 x86_hyper_kvm __refconst = {
504 .detect = kvm_detect,
506 EXPORT_SYMBOL_GPL(x86_hyper_kvm);
508 static __init int activate_jump_labels(void)
510 if (has_steal_clock) {
511 static_key_slow_inc(¶virt_steal_enabled);
513 static_key_slow_inc(¶virt_steal_rq_enabled);
518 arch_initcall(activate_jump_labels);