xref: /linux/arch/x86/kernel/kvmclock.c (revision a356d2ae50790f49858ebed35da9e206336fafee)
1 /*  KVM paravirtual clock driver. A clocksource implementation
2     Copyright (C) 2008 Glauber de Oliveira Costa, Red Hat Inc.
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, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17 */
18 
19 #include <linux/clocksource.h>
20 #include <linux/kvm_para.h>
21 #include <asm/pvclock.h>
22 #include <asm/msr.h>
23 #include <asm/apic.h>
24 #include <linux/percpu.h>
25 #include <linux/hardirq.h>
26 #include <linux/memblock.h>
27 #include <linux/sched.h>
28 #include <linux/sched/clock.h>
29 
30 #include <asm/mem_encrypt.h>
31 #include <asm/x86_init.h>
32 #include <asm/reboot.h>
33 #include <asm/kvmclock.h>
34 
35 static int kvmclock __ro_after_init = 1;
36 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
37 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
38 static u64 kvm_sched_clock_offset;
39 
40 static int parse_no_kvmclock(char *arg)
41 {
42 	kvmclock = 0;
43 	return 0;
44 }
45 early_param("no-kvmclock", parse_no_kvmclock);
46 
47 /* The hypervisor will put information about time periodically here */
48 static struct pvclock_vsyscall_time_info *hv_clock;
49 static struct pvclock_wall_clock *wall_clock;
50 
51 struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void)
52 {
53 	return hv_clock;
54 }
55 EXPORT_SYMBOL_GPL(pvclock_pvti_cpu0_va);
56 
57 /*
58  * The wallclock is the time of day when we booted. Since then, some time may
59  * have elapsed since the hypervisor wrote the data. So we try to account for
60  * that with system time
61  */
62 static void kvm_get_wallclock(struct timespec *now)
63 {
64 	struct pvclock_vcpu_time_info *vcpu_time;
65 	int low, high;
66 	int cpu;
67 
68 	low = (int)slow_virt_to_phys(wall_clock);
69 	high = ((u64)slow_virt_to_phys(wall_clock) >> 32);
70 
71 	native_write_msr(msr_kvm_wall_clock, low, high);
72 
73 	cpu = get_cpu();
74 
75 	vcpu_time = &hv_clock[cpu].pvti;
76 	pvclock_read_wallclock(wall_clock, vcpu_time, now);
77 
78 	put_cpu();
79 }
80 
81 static int kvm_set_wallclock(const struct timespec *now)
82 {
83 	return -ENODEV;
84 }
85 
86 static u64 kvm_clock_read(void)
87 {
88 	struct pvclock_vcpu_time_info *src;
89 	u64 ret;
90 	int cpu;
91 
92 	preempt_disable_notrace();
93 	cpu = smp_processor_id();
94 	src = &hv_clock[cpu].pvti;
95 	ret = pvclock_clocksource_read(src);
96 	preempt_enable_notrace();
97 	return ret;
98 }
99 
100 static u64 kvm_clock_get_cycles(struct clocksource *cs)
101 {
102 	return kvm_clock_read();
103 }
104 
105 static u64 kvm_sched_clock_read(void)
106 {
107 	return kvm_clock_read() - kvm_sched_clock_offset;
108 }
109 
110 static inline void kvm_sched_clock_init(bool stable)
111 {
112 	if (!stable) {
113 		pv_time_ops.sched_clock = kvm_clock_read;
114 		clear_sched_clock_stable();
115 		return;
116 	}
117 
118 	kvm_sched_clock_offset = kvm_clock_read();
119 	pv_time_ops.sched_clock = kvm_sched_clock_read;
120 
121 	printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
122 			kvm_sched_clock_offset);
123 
124 	BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
125 	         sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
126 }
127 
128 /*
129  * If we don't do that, there is the possibility that the guest
130  * will calibrate under heavy load - thus, getting a lower lpj -
131  * and execute the delays themselves without load. This is wrong,
132  * because no delay loop can finish beforehand.
133  * Any heuristics is subject to fail, because ultimately, a large
134  * poll of guests can be running and trouble each other. So we preset
135  * lpj here
136  */
137 static unsigned long kvm_get_tsc_khz(void)
138 {
139 	struct pvclock_vcpu_time_info *src;
140 	int cpu;
141 	unsigned long tsc_khz;
142 
143 	cpu = get_cpu();
144 	src = &hv_clock[cpu].pvti;
145 	tsc_khz = pvclock_tsc_khz(src);
146 	put_cpu();
147 	return tsc_khz;
148 }
149 
150 static void kvm_get_preset_lpj(void)
151 {
152 	unsigned long khz;
153 	u64 lpj;
154 
155 	khz = kvm_get_tsc_khz();
156 
157 	lpj = ((u64)khz * 1000);
158 	do_div(lpj, HZ);
159 	preset_lpj = lpj;
160 }
161 
162 bool kvm_check_and_clear_guest_paused(void)
163 {
164 	bool ret = false;
165 	struct pvclock_vcpu_time_info *src;
166 	int cpu = smp_processor_id();
167 
168 	if (!hv_clock)
169 		return ret;
170 
171 	src = &hv_clock[cpu].pvti;
172 	if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
173 		src->flags &= ~PVCLOCK_GUEST_STOPPED;
174 		pvclock_touch_watchdogs();
175 		ret = true;
176 	}
177 
178 	return ret;
179 }
180 
181 struct clocksource kvm_clock = {
182 	.name = "kvm-clock",
183 	.read = kvm_clock_get_cycles,
184 	.rating = 400,
185 	.mask = CLOCKSOURCE_MASK(64),
186 	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
187 };
188 EXPORT_SYMBOL_GPL(kvm_clock);
189 
190 int kvm_register_clock(char *txt)
191 {
192 	int cpu = smp_processor_id();
193 	int low, high, ret;
194 	struct pvclock_vcpu_time_info *src;
195 
196 	if (!hv_clock)
197 		return 0;
198 
199 	src = &hv_clock[cpu].pvti;
200 	low = (int)slow_virt_to_phys(src) | 1;
201 	high = ((u64)slow_virt_to_phys(src) >> 32);
202 	ret = native_write_msr_safe(msr_kvm_system_time, low, high);
203 	printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
204 	       cpu, high, low, txt);
205 
206 	return ret;
207 }
208 
209 static void kvm_save_sched_clock_state(void)
210 {
211 }
212 
213 static void kvm_restore_sched_clock_state(void)
214 {
215 	kvm_register_clock("primary cpu clock, resume");
216 }
217 
218 #ifdef CONFIG_X86_LOCAL_APIC
219 static void kvm_setup_secondary_clock(void)
220 {
221 	/*
222 	 * Now that the first cpu already had this clocksource initialized,
223 	 * we shouldn't fail.
224 	 */
225 	WARN_ON(kvm_register_clock("secondary cpu clock"));
226 }
227 #endif
228 
229 /*
230  * After the clock is registered, the host will keep writing to the
231  * registered memory location. If the guest happens to shutdown, this memory
232  * won't be valid. In cases like kexec, in which you install a new kernel, this
233  * means a random memory location will be kept being written. So before any
234  * kind of shutdown from our side, we unregister the clock by writing anything
235  * that does not have the 'enable' bit set in the msr
236  */
237 #ifdef CONFIG_KEXEC_CORE
238 static void kvm_crash_shutdown(struct pt_regs *regs)
239 {
240 	native_write_msr(msr_kvm_system_time, 0, 0);
241 	kvm_disable_steal_time();
242 	native_machine_crash_shutdown(regs);
243 }
244 #endif
245 
246 static void kvm_shutdown(void)
247 {
248 	native_write_msr(msr_kvm_system_time, 0, 0);
249 	kvm_disable_steal_time();
250 	native_machine_shutdown();
251 }
252 
253 static phys_addr_t __init kvm_memblock_alloc(phys_addr_t size,
254 					     phys_addr_t align)
255 {
256 	phys_addr_t mem;
257 
258 	mem = memblock_alloc(size, align);
259 	if (!mem)
260 		return 0;
261 
262 	if (sev_active()) {
263 		if (early_set_memory_decrypted((unsigned long)__va(mem), size))
264 			goto e_free;
265 	}
266 
267 	return mem;
268 e_free:
269 	memblock_free(mem, size);
270 	return 0;
271 }
272 
273 static void __init kvm_memblock_free(phys_addr_t addr, phys_addr_t size)
274 {
275 	if (sev_active())
276 		early_set_memory_encrypted((unsigned long)__va(addr), size);
277 
278 	memblock_free(addr, size);
279 }
280 
281 void __init kvmclock_init(void)
282 {
283 	struct pvclock_vcpu_time_info *vcpu_time;
284 	unsigned long mem, mem_wall_clock;
285 	int size, cpu, wall_clock_size;
286 	u8 flags;
287 
288 	size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
289 
290 	if (!kvm_para_available())
291 		return;
292 
293 	if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
294 		msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
295 		msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
296 	} else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
297 		return;
298 
299 	wall_clock_size = PAGE_ALIGN(sizeof(struct pvclock_wall_clock));
300 	mem_wall_clock = kvm_memblock_alloc(wall_clock_size, PAGE_SIZE);
301 	if (!mem_wall_clock)
302 		return;
303 
304 	wall_clock = __va(mem_wall_clock);
305 	memset(wall_clock, 0, wall_clock_size);
306 
307 	mem = kvm_memblock_alloc(size, PAGE_SIZE);
308 	if (!mem) {
309 		kvm_memblock_free(mem_wall_clock, wall_clock_size);
310 		wall_clock = NULL;
311 		return;
312 	}
313 
314 	hv_clock = __va(mem);
315 	memset(hv_clock, 0, size);
316 
317 	if (kvm_register_clock("primary cpu clock")) {
318 		hv_clock = NULL;
319 		kvm_memblock_free(mem, size);
320 		kvm_memblock_free(mem_wall_clock, wall_clock_size);
321 		wall_clock = NULL;
322 		return;
323 	}
324 
325 	printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
326 		msr_kvm_system_time, msr_kvm_wall_clock);
327 
328 	if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
329 		pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
330 
331 	cpu = get_cpu();
332 	vcpu_time = &hv_clock[cpu].pvti;
333 	flags = pvclock_read_flags(vcpu_time);
334 
335 	kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
336 	put_cpu();
337 
338 	x86_platform.calibrate_tsc = kvm_get_tsc_khz;
339 	x86_platform.calibrate_cpu = kvm_get_tsc_khz;
340 	x86_platform.get_wallclock = kvm_get_wallclock;
341 	x86_platform.set_wallclock = kvm_set_wallclock;
342 #ifdef CONFIG_X86_LOCAL_APIC
343 	x86_cpuinit.early_percpu_clock_init =
344 		kvm_setup_secondary_clock;
345 #endif
346 	x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
347 	x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
348 	machine_ops.shutdown  = kvm_shutdown;
349 #ifdef CONFIG_KEXEC_CORE
350 	machine_ops.crash_shutdown  = kvm_crash_shutdown;
351 #endif
352 	kvm_get_preset_lpj();
353 	clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
354 	pv_info.name = "KVM";
355 }
356 
357 int __init kvm_setup_vsyscall_timeinfo(void)
358 {
359 #ifdef CONFIG_X86_64
360 	int cpu;
361 	u8 flags;
362 	struct pvclock_vcpu_time_info *vcpu_time;
363 	unsigned int size;
364 
365 	if (!hv_clock)
366 		return 0;
367 
368 	size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
369 
370 	cpu = get_cpu();
371 
372 	vcpu_time = &hv_clock[cpu].pvti;
373 	flags = pvclock_read_flags(vcpu_time);
374 
375 	if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
376 		put_cpu();
377 		return 1;
378 	}
379 
380 	put_cpu();
381 
382 	kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
383 #endif
384 	return 0;
385 }
386