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>
46 static int kvmapf
= 1;
48 static int parse_no_kvmapf(char *arg
)
54 early_param("no-kvmapf", parse_no_kvmapf
);
56 static int steal_acc
= 1;
57 static int parse_no_stealacc(char *arg
)
63 early_param("no-steal-acc", parse_no_stealacc
);
65 static DEFINE_PER_CPU(struct kvm_vcpu_pv_apf_data
, apf_reason
) __aligned(64);
66 static DEFINE_PER_CPU(struct kvm_steal_time
, steal_time
) __aligned(64);
67 static int has_steal_clock
= 0;
70 * No need for any "IO delay" on KVM
72 static void kvm_io_delay(void)
76 #define KVM_TASK_SLEEP_HASHBITS 8
77 #define KVM_TASK_SLEEP_HASHSIZE (1<<KVM_TASK_SLEEP_HASHBITS)
79 struct kvm_task_sleep_node
{
80 struct hlist_node link
;
87 static struct kvm_task_sleep_head
{
89 struct hlist_head list
;
90 } async_pf_sleepers
[KVM_TASK_SLEEP_HASHSIZE
];
92 static struct kvm_task_sleep_node
*_find_apf_task(struct kvm_task_sleep_head
*b
,
97 hlist_for_each(p
, &b
->list
) {
98 struct kvm_task_sleep_node
*n
=
99 hlist_entry(p
, typeof(*n
), link
);
100 if (n
->token
== token
)
107 void kvm_async_pf_task_wait(u32 token
)
109 u32 key
= hash_32(token
, KVM_TASK_SLEEP_HASHBITS
);
110 struct kvm_task_sleep_head
*b
= &async_pf_sleepers
[key
];
111 struct kvm_task_sleep_node n
, *e
;
116 idle
= idle_cpu(cpu
);
120 e
= _find_apf_task(b
, token
);
122 /* dummy entry exist -> wake up was delivered ahead of PF */
125 spin_unlock(&b
->lock
);
130 n
.cpu
= smp_processor_id();
131 n
.halted
= idle
|| preempt_count() > 1;
132 init_waitqueue_head(&n
.wq
);
133 hlist_add_head(&n
.link
, &b
->list
);
134 spin_unlock(&b
->lock
);
138 prepare_to_wait(&n
.wq
, &wait
, TASK_UNINTERRUPTIBLE
);
139 if (hlist_unhashed(&n
.link
))
148 * We cannot reschedule. So halt.
155 finish_wait(&n
.wq
, &wait
);
159 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait
);
161 static void apf_task_wake_one(struct kvm_task_sleep_node
*n
)
163 hlist_del_init(&n
->link
);
165 smp_send_reschedule(n
->cpu
);
166 else if (waitqueue_active(&n
->wq
))
170 static void apf_task_wake_all(void)
174 for (i
= 0; i
< KVM_TASK_SLEEP_HASHSIZE
; i
++) {
175 struct hlist_node
*p
, *next
;
176 struct kvm_task_sleep_head
*b
= &async_pf_sleepers
[i
];
178 hlist_for_each_safe(p
, next
, &b
->list
) {
179 struct kvm_task_sleep_node
*n
=
180 hlist_entry(p
, typeof(*n
), link
);
181 if (n
->cpu
== smp_processor_id())
182 apf_task_wake_one(n
);
184 spin_unlock(&b
->lock
);
188 void kvm_async_pf_task_wake(u32 token
)
190 u32 key
= hash_32(token
, KVM_TASK_SLEEP_HASHBITS
);
191 struct kvm_task_sleep_head
*b
= &async_pf_sleepers
[key
];
192 struct kvm_task_sleep_node
*n
;
201 n
= _find_apf_task(b
, token
);
204 * async PF was not yet handled.
205 * Add dummy entry for the token.
207 n
= kzalloc(sizeof(*n
), GFP_ATOMIC
);
210 * Allocation failed! Busy wait while other cpu
213 spin_unlock(&b
->lock
);
218 n
->cpu
= smp_processor_id();
219 init_waitqueue_head(&n
->wq
);
220 hlist_add_head(&n
->link
, &b
->list
);
222 apf_task_wake_one(n
);
223 spin_unlock(&b
->lock
);
226 EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake
);
228 u32
kvm_read_and_reset_pf_reason(void)
232 if (__get_cpu_var(apf_reason
).enabled
) {
233 reason
= __get_cpu_var(apf_reason
).reason
;
234 __get_cpu_var(apf_reason
).reason
= 0;
239 EXPORT_SYMBOL_GPL(kvm_read_and_reset_pf_reason
);
241 dotraplinkage
void __kprobes
242 do_async_page_fault(struct pt_regs
*regs
, unsigned long error_code
)
244 switch (kvm_read_and_reset_pf_reason()) {
246 do_page_fault(regs
, error_code
);
248 case KVM_PV_REASON_PAGE_NOT_PRESENT
:
249 /* page is swapped out by the host. */
252 kvm_async_pf_task_wait((u32
)read_cr2());
255 case KVM_PV_REASON_PAGE_READY
:
258 kvm_async_pf_task_wake((u32
)read_cr2());
264 static void __init
paravirt_ops_setup(void)
266 pv_info
.name
= "KVM";
267 pv_info
.paravirt_enabled
= 1;
269 if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY
))
270 pv_cpu_ops
.io_delay
= kvm_io_delay
;
272 #ifdef CONFIG_X86_IO_APIC
277 static void kvm_register_steal_time(void)
279 int cpu
= smp_processor_id();
280 struct kvm_steal_time
*st
= &per_cpu(steal_time
, cpu
);
282 if (!has_steal_clock
)
285 memset(st
, 0, sizeof(*st
));
287 wrmsrl(MSR_KVM_STEAL_TIME
, (__pa(st
) | KVM_MSR_ENABLED
));
288 printk(KERN_INFO
"kvm-stealtime: cpu %d, msr %lx\n",
292 static DEFINE_PER_CPU(unsigned long, kvm_apic_eoi
) = KVM_PV_EOI_DISABLED
;
294 static void kvm_guest_apic_eoi_write(u32 reg
, u32 val
)
297 * This relies on __test_and_clear_bit to modify the memory
298 * in a way that is atomic with respect to the local CPU.
299 * The hypervisor only accesses this memory from the local CPU so
300 * there's no need for lock or memory barriers.
301 * An optimization barrier is implied in apic write.
303 if (__test_and_clear_bit(KVM_PV_EOI_BIT
, &__get_cpu_var(kvm_apic_eoi
)))
305 apic_write(APIC_EOI
, APIC_EOI_ACK
);
308 void __cpuinit
kvm_guest_cpu_init(void)
310 if (!kvm_para_available())
313 if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF
) && kvmapf
) {
314 u64 pa
= __pa(&__get_cpu_var(apf_reason
));
316 #ifdef CONFIG_PREEMPT
317 pa
|= KVM_ASYNC_PF_SEND_ALWAYS
;
319 wrmsrl(MSR_KVM_ASYNC_PF_EN
, pa
| KVM_ASYNC_PF_ENABLED
);
320 __get_cpu_var(apf_reason
).enabled
= 1;
321 printk(KERN_INFO
"KVM setup async PF for cpu %d\n",
325 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI
)) {
327 /* Size alignment is implied but just to make it explicit. */
328 BUILD_BUG_ON(__alignof__(kvm_apic_eoi
) < 4);
329 __get_cpu_var(kvm_apic_eoi
) = 0;
330 pa
= __pa(&__get_cpu_var(kvm_apic_eoi
)) | KVM_MSR_ENABLED
;
331 wrmsrl(MSR_KVM_PV_EOI_EN
, pa
);
335 kvm_register_steal_time();
338 static void kvm_pv_disable_apf(void)
340 if (!__get_cpu_var(apf_reason
).enabled
)
343 wrmsrl(MSR_KVM_ASYNC_PF_EN
, 0);
344 __get_cpu_var(apf_reason
).enabled
= 0;
346 printk(KERN_INFO
"Unregister pv shared memory for cpu %d\n",
350 static void kvm_pv_guest_cpu_reboot(void *unused
)
353 * We disable PV EOI before we load a new kernel by kexec,
354 * since MSR_KVM_PV_EOI_EN stores a pointer into old kernel's memory.
355 * New kernel can re-enable when it boots.
357 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI
))
358 wrmsrl(MSR_KVM_PV_EOI_EN
, 0);
359 kvm_pv_disable_apf();
360 kvm_disable_steal_time();
363 static int kvm_pv_reboot_notify(struct notifier_block
*nb
,
364 unsigned long code
, void *unused
)
366 if (code
== SYS_RESTART
)
367 on_each_cpu(kvm_pv_guest_cpu_reboot
, NULL
, 1);
371 static struct notifier_block kvm_pv_reboot_nb
= {
372 .notifier_call
= kvm_pv_reboot_notify
,
375 static u64
kvm_steal_clock(int cpu
)
378 struct kvm_steal_time
*src
;
381 src
= &per_cpu(steal_time
, cpu
);
383 version
= src
->version
;
387 } while ((version
& 1) || (version
!= src
->version
));
392 void kvm_disable_steal_time(void)
394 if (!has_steal_clock
)
397 wrmsr(MSR_KVM_STEAL_TIME
, 0, 0);
401 static void __init
kvm_smp_prepare_boot_cpu(void)
403 WARN_ON(kvm_register_clock("primary cpu clock"));
404 kvm_guest_cpu_init();
405 native_smp_prepare_boot_cpu();
408 static void __cpuinit
kvm_guest_cpu_online(void *dummy
)
410 kvm_guest_cpu_init();
413 static void kvm_guest_cpu_offline(void *dummy
)
415 kvm_disable_steal_time();
416 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI
))
417 wrmsrl(MSR_KVM_PV_EOI_EN
, 0);
418 kvm_pv_disable_apf();
422 static int __cpuinit
kvm_cpu_notify(struct notifier_block
*self
,
423 unsigned long action
, void *hcpu
)
425 int cpu
= (unsigned long)hcpu
;
428 case CPU_DOWN_FAILED
:
429 case CPU_ONLINE_FROZEN
:
430 smp_call_function_single(cpu
, kvm_guest_cpu_online
, NULL
, 0);
432 case CPU_DOWN_PREPARE
:
433 case CPU_DOWN_PREPARE_FROZEN
:
434 smp_call_function_single(cpu
, kvm_guest_cpu_offline
, NULL
, 1);
442 static struct notifier_block __cpuinitdata kvm_cpu_notifier
= {
443 .notifier_call
= kvm_cpu_notify
,
447 static void __init
kvm_apf_trap_init(void)
449 set_intr_gate(14, &async_page_fault
);
452 void __init
kvm_guest_init(void)
456 if (!kvm_para_available())
459 paravirt_ops_setup();
460 register_reboot_notifier(&kvm_pv_reboot_nb
);
461 for (i
= 0; i
< KVM_TASK_SLEEP_HASHSIZE
; i
++)
462 spin_lock_init(&async_pf_sleepers
[i
].lock
);
463 if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF
))
464 x86_init
.irqs
.trap_init
= kvm_apf_trap_init
;
466 if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME
)) {
468 pv_time_ops
.steal_clock
= kvm_steal_clock
;
471 if (kvm_para_has_feature(KVM_FEATURE_PV_EOI
))
472 apic_set_eoi_write(kvm_guest_apic_eoi_write
);
475 smp_ops
.smp_prepare_boot_cpu
= kvm_smp_prepare_boot_cpu
;
476 register_cpu_notifier(&kvm_cpu_notifier
);
478 kvm_guest_cpu_init();
482 static bool __init
kvm_detect(void)
484 if (!kvm_para_available())
489 const struct hypervisor_x86 x86_hyper_kvm __refconst
= {
491 .detect
= kvm_detect
,
493 EXPORT_SYMBOL_GPL(x86_hyper_kvm
);
495 static __init
int activate_jump_labels(void)
497 if (has_steal_clock
) {
498 static_key_slow_inc(¶virt_steal_enabled
);
500 static_key_slow_inc(¶virt_steal_rq_enabled
);
505 arch_initcall(activate_jump_labels
);