Commit | Line | Data |
---|---|---|
0cf1bfd2 MT |
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> | |
1da8a77b | 28 | #include <linux/highmem.h> |
096d14a3 | 29 | #include <linux/hardirq.h> |
fd10cde9 GN |
30 | #include <linux/notifier.h> |
31 | #include <linux/reboot.h> | |
631bc487 GN |
32 | #include <linux/hash.h> |
33 | #include <linux/sched.h> | |
34 | #include <linux/slab.h> | |
35 | #include <linux/kprobes.h> | |
a90ede7b | 36 | #include <asm/timer.h> |
fd10cde9 | 37 | #include <asm/cpu.h> |
631bc487 GN |
38 | #include <asm/traps.h> |
39 | #include <asm/desc.h> | |
6c047cd9 | 40 | #include <asm/tlbflush.h> |
096d14a3 | 41 | |
fd10cde9 GN |
42 | static int kvmapf = 1; |
43 | ||
44 | static int parse_no_kvmapf(char *arg) | |
45 | { | |
46 | kvmapf = 0; | |
47 | return 0; | |
48 | } | |
49 | ||
50 | early_param("no-kvmapf", parse_no_kvmapf); | |
51 | ||
d910f5c1 GC |
52 | static int steal_acc = 1; |
53 | static int parse_no_stealacc(char *arg) | |
54 | { | |
55 | steal_acc = 0; | |
56 | return 0; | |
57 | } | |
58 | ||
59 | early_param("no-steal-acc", parse_no_stealacc); | |
60 | ||
fd10cde9 | 61 | static DEFINE_PER_CPU(struct kvm_vcpu_pv_apf_data, apf_reason) __aligned(64); |
d910f5c1 GC |
62 | static DEFINE_PER_CPU(struct kvm_steal_time, steal_time) __aligned(64); |
63 | static int has_steal_clock = 0; | |
096d14a3 | 64 | |
0cf1bfd2 MT |
65 | /* |
66 | * No need for any "IO delay" on KVM | |
67 | */ | |
68 | static void kvm_io_delay(void) | |
69 | { | |
70 | } | |
71 | ||
631bc487 GN |
72 | #define KVM_TASK_SLEEP_HASHBITS 8 |
73 | #define KVM_TASK_SLEEP_HASHSIZE (1<<KVM_TASK_SLEEP_HASHBITS) | |
74 | ||
75 | struct kvm_task_sleep_node { | |
76 | struct hlist_node link; | |
77 | wait_queue_head_t wq; | |
78 | u32 token; | |
79 | int cpu; | |
6c047cd9 GN |
80 | bool halted; |
81 | struct mm_struct *mm; | |
631bc487 GN |
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); | |
6c047cd9 GN |
110 | int cpu, idle; |
111 | ||
112 | cpu = get_cpu(); | |
113 | idle = idle_cpu(cpu); | |
114 | put_cpu(); | |
631bc487 GN |
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(); | |
6c047cd9 GN |
128 | n.mm = current->active_mm; |
129 | n.halted = idle || preempt_count() > 1; | |
130 | atomic_inc(&n.mm->mm_count); | |
631bc487 GN |
131 | init_waitqueue_head(&n.wq); |
132 | hlist_add_head(&n.link, &b->list); | |
133 | spin_unlock(&b->lock); | |
134 | ||
135 | for (;;) { | |
6c047cd9 GN |
136 | if (!n.halted) |
137 | prepare_to_wait(&n.wq, &wait, TASK_UNINTERRUPTIBLE); | |
631bc487 GN |
138 | if (hlist_unhashed(&n.link)) |
139 | break; | |
6c047cd9 GN |
140 | |
141 | if (!n.halted) { | |
142 | local_irq_enable(); | |
143 | schedule(); | |
144 | local_irq_disable(); | |
145 | } else { | |
146 | /* | |
147 | * We cannot reschedule. So halt. | |
148 | */ | |
149 | native_safe_halt(); | |
150 | local_irq_disable(); | |
151 | } | |
631bc487 | 152 | } |
6c047cd9 GN |
153 | if (!n.halted) |
154 | finish_wait(&n.wq, &wait); | |
631bc487 GN |
155 | |
156 | return; | |
157 | } | |
158 | EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait); | |
159 | ||
160 | static void apf_task_wake_one(struct kvm_task_sleep_node *n) | |
161 | { | |
162 | hlist_del_init(&n->link); | |
6c047cd9 GN |
163 | if (!n->mm) |
164 | return; | |
165 | mmdrop(n->mm); | |
166 | if (n->halted) | |
167 | smp_send_reschedule(n->cpu); | |
168 | else if (waitqueue_active(&n->wq)) | |
631bc487 GN |
169 | wake_up(&n->wq); |
170 | } | |
171 | ||
172 | static void apf_task_wake_all(void) | |
173 | { | |
174 | int i; | |
175 | ||
176 | for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) { | |
177 | struct hlist_node *p, *next; | |
178 | struct kvm_task_sleep_head *b = &async_pf_sleepers[i]; | |
179 | spin_lock(&b->lock); | |
180 | hlist_for_each_safe(p, next, &b->list) { | |
181 | struct kvm_task_sleep_node *n = | |
182 | hlist_entry(p, typeof(*n), link); | |
183 | if (n->cpu == smp_processor_id()) | |
184 | apf_task_wake_one(n); | |
185 | } | |
186 | spin_unlock(&b->lock); | |
187 | } | |
188 | } | |
189 | ||
190 | void kvm_async_pf_task_wake(u32 token) | |
191 | { | |
192 | u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS); | |
193 | struct kvm_task_sleep_head *b = &async_pf_sleepers[key]; | |
194 | struct kvm_task_sleep_node *n; | |
195 | ||
196 | if (token == ~0) { | |
197 | apf_task_wake_all(); | |
198 | return; | |
199 | } | |
200 | ||
201 | again: | |
202 | spin_lock(&b->lock); | |
203 | n = _find_apf_task(b, token); | |
204 | if (!n) { | |
205 | /* | |
206 | * async PF was not yet handled. | |
207 | * Add dummy entry for the token. | |
208 | */ | |
209 | n = kmalloc(sizeof(*n), GFP_ATOMIC); | |
210 | if (!n) { | |
211 | /* | |
212 | * Allocation failed! Busy wait while other cpu | |
213 | * handles async PF. | |
214 | */ | |
215 | spin_unlock(&b->lock); | |
216 | cpu_relax(); | |
217 | goto again; | |
218 | } | |
219 | n->token = token; | |
220 | n->cpu = smp_processor_id(); | |
6c047cd9 | 221 | n->mm = NULL; |
631bc487 GN |
222 | init_waitqueue_head(&n->wq); |
223 | hlist_add_head(&n->link, &b->list); | |
224 | } else | |
225 | apf_task_wake_one(n); | |
226 | spin_unlock(&b->lock); | |
227 | return; | |
228 | } | |
229 | EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake); | |
230 | ||
231 | u32 kvm_read_and_reset_pf_reason(void) | |
232 | { | |
233 | u32 reason = 0; | |
234 | ||
235 | if (__get_cpu_var(apf_reason).enabled) { | |
236 | reason = __get_cpu_var(apf_reason).reason; | |
237 | __get_cpu_var(apf_reason).reason = 0; | |
238 | } | |
239 | ||
240 | return reason; | |
241 | } | |
242 | EXPORT_SYMBOL_GPL(kvm_read_and_reset_pf_reason); | |
243 | ||
244 | dotraplinkage void __kprobes | |
245 | do_async_page_fault(struct pt_regs *regs, unsigned long error_code) | |
246 | { | |
247 | switch (kvm_read_and_reset_pf_reason()) { | |
248 | default: | |
249 | do_page_fault(regs, error_code); | |
250 | break; | |
251 | case KVM_PV_REASON_PAGE_NOT_PRESENT: | |
252 | /* page is swapped out by the host. */ | |
253 | kvm_async_pf_task_wait((u32)read_cr2()); | |
254 | break; | |
255 | case KVM_PV_REASON_PAGE_READY: | |
256 | kvm_async_pf_task_wake((u32)read_cr2()); | |
257 | break; | |
258 | } | |
259 | } | |
260 | ||
d3ac8815 | 261 | static void __init paravirt_ops_setup(void) |
0cf1bfd2 MT |
262 | { |
263 | pv_info.name = "KVM"; | |
264 | pv_info.paravirt_enabled = 1; | |
265 | ||
266 | if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY)) | |
267 | pv_cpu_ops.io_delay = kvm_io_delay; | |
268 | ||
a90ede7b MT |
269 | #ifdef CONFIG_X86_IO_APIC |
270 | no_timer_check = 1; | |
271 | #endif | |
0cf1bfd2 MT |
272 | } |
273 | ||
d910f5c1 GC |
274 | static void kvm_register_steal_time(void) |
275 | { | |
276 | int cpu = smp_processor_id(); | |
277 | struct kvm_steal_time *st = &per_cpu(steal_time, cpu); | |
278 | ||
279 | if (!has_steal_clock) | |
280 | return; | |
281 | ||
282 | memset(st, 0, sizeof(*st)); | |
283 | ||
284 | wrmsrl(MSR_KVM_STEAL_TIME, (__pa(st) | KVM_MSR_ENABLED)); | |
285 | printk(KERN_INFO "kvm-stealtime: cpu %d, msr %lx\n", | |
286 | cpu, __pa(st)); | |
287 | } | |
288 | ||
fd10cde9 GN |
289 | void __cpuinit kvm_guest_cpu_init(void) |
290 | { | |
291 | if (!kvm_para_available()) | |
292 | return; | |
293 | ||
294 | if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF) && kvmapf) { | |
295 | u64 pa = __pa(&__get_cpu_var(apf_reason)); | |
296 | ||
6adba527 GN |
297 | #ifdef CONFIG_PREEMPT |
298 | pa |= KVM_ASYNC_PF_SEND_ALWAYS; | |
299 | #endif | |
fd10cde9 GN |
300 | wrmsrl(MSR_KVM_ASYNC_PF_EN, pa | KVM_ASYNC_PF_ENABLED); |
301 | __get_cpu_var(apf_reason).enabled = 1; | |
302 | printk(KERN_INFO"KVM setup async PF for cpu %d\n", | |
303 | smp_processor_id()); | |
304 | } | |
d910f5c1 GC |
305 | |
306 | if (has_steal_clock) | |
307 | kvm_register_steal_time(); | |
fd10cde9 GN |
308 | } |
309 | ||
310 | static void kvm_pv_disable_apf(void *unused) | |
311 | { | |
312 | if (!__get_cpu_var(apf_reason).enabled) | |
313 | return; | |
314 | ||
315 | wrmsrl(MSR_KVM_ASYNC_PF_EN, 0); | |
316 | __get_cpu_var(apf_reason).enabled = 0; | |
317 | ||
318 | printk(KERN_INFO"Unregister pv shared memory for cpu %d\n", | |
319 | smp_processor_id()); | |
320 | } | |
321 | ||
322 | static int kvm_pv_reboot_notify(struct notifier_block *nb, | |
323 | unsigned long code, void *unused) | |
324 | { | |
325 | if (code == SYS_RESTART) | |
326 | on_each_cpu(kvm_pv_disable_apf, NULL, 1); | |
327 | return NOTIFY_DONE; | |
328 | } | |
329 | ||
330 | static struct notifier_block kvm_pv_reboot_nb = { | |
331 | .notifier_call = kvm_pv_reboot_notify, | |
332 | }; | |
333 | ||
d910f5c1 GC |
334 | static u64 kvm_steal_clock(int cpu) |
335 | { | |
336 | u64 steal; | |
337 | struct kvm_steal_time *src; | |
338 | int version; | |
339 | ||
340 | src = &per_cpu(steal_time, cpu); | |
341 | do { | |
342 | version = src->version; | |
343 | rmb(); | |
344 | steal = src->steal; | |
345 | rmb(); | |
346 | } while ((version & 1) || (version != src->version)); | |
347 | ||
348 | return steal; | |
349 | } | |
350 | ||
351 | void kvm_disable_steal_time(void) | |
352 | { | |
353 | if (!has_steal_clock) | |
354 | return; | |
355 | ||
356 | wrmsr(MSR_KVM_STEAL_TIME, 0, 0); | |
357 | } | |
358 | ||
ca3f1017 GN |
359 | #ifdef CONFIG_SMP |
360 | static void __init kvm_smp_prepare_boot_cpu(void) | |
361 | { | |
a63512a4 | 362 | #ifdef CONFIG_KVM_CLOCK |
ca3f1017 | 363 | WARN_ON(kvm_register_clock("primary cpu clock")); |
a63512a4 | 364 | #endif |
fd10cde9 | 365 | kvm_guest_cpu_init(); |
ca3f1017 GN |
366 | native_smp_prepare_boot_cpu(); |
367 | } | |
fd10cde9 | 368 | |
775077a0 | 369 | static void __cpuinit kvm_guest_cpu_online(void *dummy) |
fd10cde9 GN |
370 | { |
371 | kvm_guest_cpu_init(); | |
372 | } | |
373 | ||
374 | static void kvm_guest_cpu_offline(void *dummy) | |
375 | { | |
d910f5c1 | 376 | kvm_disable_steal_time(); |
fd10cde9 | 377 | kvm_pv_disable_apf(NULL); |
631bc487 | 378 | apf_task_wake_all(); |
fd10cde9 GN |
379 | } |
380 | ||
381 | static int __cpuinit kvm_cpu_notify(struct notifier_block *self, | |
382 | unsigned long action, void *hcpu) | |
383 | { | |
384 | int cpu = (unsigned long)hcpu; | |
385 | switch (action) { | |
386 | case CPU_ONLINE: | |
387 | case CPU_DOWN_FAILED: | |
388 | case CPU_ONLINE_FROZEN: | |
389 | smp_call_function_single(cpu, kvm_guest_cpu_online, NULL, 0); | |
390 | break; | |
391 | case CPU_DOWN_PREPARE: | |
392 | case CPU_DOWN_PREPARE_FROZEN: | |
393 | smp_call_function_single(cpu, kvm_guest_cpu_offline, NULL, 1); | |
394 | break; | |
395 | default: | |
396 | break; | |
397 | } | |
398 | return NOTIFY_OK; | |
399 | } | |
400 | ||
401 | static struct notifier_block __cpuinitdata kvm_cpu_notifier = { | |
402 | .notifier_call = kvm_cpu_notify, | |
403 | }; | |
ca3f1017 GN |
404 | #endif |
405 | ||
631bc487 GN |
406 | static void __init kvm_apf_trap_init(void) |
407 | { | |
408 | set_intr_gate(14, &async_page_fault); | |
409 | } | |
410 | ||
0cf1bfd2 MT |
411 | void __init kvm_guest_init(void) |
412 | { | |
631bc487 GN |
413 | int i; |
414 | ||
0cf1bfd2 MT |
415 | if (!kvm_para_available()) |
416 | return; | |
417 | ||
418 | paravirt_ops_setup(); | |
fd10cde9 | 419 | register_reboot_notifier(&kvm_pv_reboot_nb); |
631bc487 GN |
420 | for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) |
421 | spin_lock_init(&async_pf_sleepers[i].lock); | |
422 | if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF)) | |
423 | x86_init.irqs.trap_init = kvm_apf_trap_init; | |
424 | ||
d910f5c1 GC |
425 | if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) { |
426 | has_steal_clock = 1; | |
427 | pv_time_ops.steal_clock = kvm_steal_clock; | |
428 | } | |
429 | ||
ca3f1017 GN |
430 | #ifdef CONFIG_SMP |
431 | smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu; | |
fd10cde9 GN |
432 | register_cpu_notifier(&kvm_cpu_notifier); |
433 | #else | |
434 | kvm_guest_cpu_init(); | |
ca3f1017 | 435 | #endif |
0cf1bfd2 | 436 | } |
d910f5c1 GC |
437 | |
438 | static __init int activate_jump_labels(void) | |
439 | { | |
440 | if (has_steal_clock) { | |
441 | jump_label_inc(¶virt_steal_enabled); | |
442 | if (steal_acc) | |
443 | jump_label_inc(¶virt_steal_rq_enabled); | |
444 | } | |
445 | ||
446 | return 0; | |
447 | } | |
448 | arch_initcall(activate_jump_labels); |