xref: /openbmc/linux/arch/x86/kernel/kvmclock.c (revision 4e1a33b1)
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 
29 #include <asm/x86_init.h>
30 #include <asm/reboot.h>
31 #include <asm/kvmclock.h>
32 
33 static int kvmclock __ro_after_init = 1;
34 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
35 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
36 static u64 kvm_sched_clock_offset;
37 
38 static int parse_no_kvmclock(char *arg)
39 {
40 	kvmclock = 0;
41 	return 0;
42 }
43 early_param("no-kvmclock", parse_no_kvmclock);
44 
45 /* The hypervisor will put information about time periodically here */
46 static struct pvclock_vsyscall_time_info *hv_clock;
47 static struct pvclock_wall_clock wall_clock;
48 
49 struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void)
50 {
51 	return hv_clock;
52 }
53 EXPORT_SYMBOL_GPL(pvclock_pvti_cpu0_va);
54 
55 /*
56  * The wallclock is the time of day when we booted. Since then, some time may
57  * have elapsed since the hypervisor wrote the data. So we try to account for
58  * that with system time
59  */
60 static void kvm_get_wallclock(struct timespec *now)
61 {
62 	struct pvclock_vcpu_time_info *vcpu_time;
63 	int low, high;
64 	int cpu;
65 
66 	low = (int)__pa_symbol(&wall_clock);
67 	high = ((u64)__pa_symbol(&wall_clock) >> 32);
68 
69 	native_write_msr(msr_kvm_wall_clock, low, high);
70 
71 	cpu = get_cpu();
72 
73 	vcpu_time = &hv_clock[cpu].pvti;
74 	pvclock_read_wallclock(&wall_clock, vcpu_time, now);
75 
76 	put_cpu();
77 }
78 
79 static int kvm_set_wallclock(const struct timespec *now)
80 {
81 	return -1;
82 }
83 
84 static u64 kvm_clock_read(void)
85 {
86 	struct pvclock_vcpu_time_info *src;
87 	u64 ret;
88 	int cpu;
89 
90 	preempt_disable_notrace();
91 	cpu = smp_processor_id();
92 	src = &hv_clock[cpu].pvti;
93 	ret = pvclock_clocksource_read(src);
94 	preempt_enable_notrace();
95 	return ret;
96 }
97 
98 static u64 kvm_clock_get_cycles(struct clocksource *cs)
99 {
100 	return kvm_clock_read();
101 }
102 
103 static u64 kvm_sched_clock_read(void)
104 {
105 	return kvm_clock_read() - kvm_sched_clock_offset;
106 }
107 
108 static inline void kvm_sched_clock_init(bool stable)
109 {
110 	if (!stable) {
111 		pv_time_ops.sched_clock = kvm_clock_read;
112 		clear_sched_clock_stable();
113 		return;
114 	}
115 
116 	kvm_sched_clock_offset = kvm_clock_read();
117 	pv_time_ops.sched_clock = kvm_sched_clock_read;
118 
119 	printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
120 			kvm_sched_clock_offset);
121 
122 	BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
123 	         sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
124 }
125 
126 /*
127  * If we don't do that, there is the possibility that the guest
128  * will calibrate under heavy load - thus, getting a lower lpj -
129  * and execute the delays themselves without load. This is wrong,
130  * because no delay loop can finish beforehand.
131  * Any heuristics is subject to fail, because ultimately, a large
132  * poll of guests can be running and trouble each other. So we preset
133  * lpj here
134  */
135 static unsigned long kvm_get_tsc_khz(void)
136 {
137 	struct pvclock_vcpu_time_info *src;
138 	int cpu;
139 	unsigned long tsc_khz;
140 
141 	cpu = get_cpu();
142 	src = &hv_clock[cpu].pvti;
143 	tsc_khz = pvclock_tsc_khz(src);
144 	put_cpu();
145 	return tsc_khz;
146 }
147 
148 static void kvm_get_preset_lpj(void)
149 {
150 	unsigned long khz;
151 	u64 lpj;
152 
153 	khz = kvm_get_tsc_khz();
154 
155 	lpj = ((u64)khz * 1000);
156 	do_div(lpj, HZ);
157 	preset_lpj = lpj;
158 }
159 
160 bool kvm_check_and_clear_guest_paused(void)
161 {
162 	bool ret = false;
163 	struct pvclock_vcpu_time_info *src;
164 	int cpu = smp_processor_id();
165 
166 	if (!hv_clock)
167 		return ret;
168 
169 	src = &hv_clock[cpu].pvti;
170 	if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
171 		src->flags &= ~PVCLOCK_GUEST_STOPPED;
172 		pvclock_touch_watchdogs();
173 		ret = true;
174 	}
175 
176 	return ret;
177 }
178 
179 struct clocksource kvm_clock = {
180 	.name = "kvm-clock",
181 	.read = kvm_clock_get_cycles,
182 	.rating = 400,
183 	.mask = CLOCKSOURCE_MASK(64),
184 	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
185 };
186 EXPORT_SYMBOL_GPL(kvm_clock);
187 
188 int kvm_register_clock(char *txt)
189 {
190 	int cpu = smp_processor_id();
191 	int low, high, ret;
192 	struct pvclock_vcpu_time_info *src;
193 
194 	if (!hv_clock)
195 		return 0;
196 
197 	src = &hv_clock[cpu].pvti;
198 	low = (int)slow_virt_to_phys(src) | 1;
199 	high = ((u64)slow_virt_to_phys(src) >> 32);
200 	ret = native_write_msr_safe(msr_kvm_system_time, low, high);
201 	printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
202 	       cpu, high, low, txt);
203 
204 	return ret;
205 }
206 
207 static void kvm_save_sched_clock_state(void)
208 {
209 }
210 
211 static void kvm_restore_sched_clock_state(void)
212 {
213 	kvm_register_clock("primary cpu clock, resume");
214 }
215 
216 #ifdef CONFIG_X86_LOCAL_APIC
217 static void kvm_setup_secondary_clock(void)
218 {
219 	/*
220 	 * Now that the first cpu already had this clocksource initialized,
221 	 * we shouldn't fail.
222 	 */
223 	WARN_ON(kvm_register_clock("secondary cpu clock"));
224 }
225 #endif
226 
227 /*
228  * After the clock is registered, the host will keep writing to the
229  * registered memory location. If the guest happens to shutdown, this memory
230  * won't be valid. In cases like kexec, in which you install a new kernel, this
231  * means a random memory location will be kept being written. So before any
232  * kind of shutdown from our side, we unregister the clock by writing anything
233  * that does not have the 'enable' bit set in the msr
234  */
235 #ifdef CONFIG_KEXEC_CORE
236 static void kvm_crash_shutdown(struct pt_regs *regs)
237 {
238 	native_write_msr(msr_kvm_system_time, 0, 0);
239 	kvm_disable_steal_time();
240 	native_machine_crash_shutdown(regs);
241 }
242 #endif
243 
244 static void kvm_shutdown(void)
245 {
246 	native_write_msr(msr_kvm_system_time, 0, 0);
247 	kvm_disable_steal_time();
248 	native_machine_shutdown();
249 }
250 
251 void __init kvmclock_init(void)
252 {
253 	struct pvclock_vcpu_time_info *vcpu_time;
254 	unsigned long mem;
255 	int size, cpu;
256 	u8 flags;
257 
258 	size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
259 
260 	if (!kvm_para_available())
261 		return;
262 
263 	if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
264 		msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
265 		msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
266 	} else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
267 		return;
268 
269 	printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
270 		msr_kvm_system_time, msr_kvm_wall_clock);
271 
272 	mem = memblock_alloc(size, PAGE_SIZE);
273 	if (!mem)
274 		return;
275 	hv_clock = __va(mem);
276 	memset(hv_clock, 0, size);
277 
278 	if (kvm_register_clock("primary cpu clock")) {
279 		hv_clock = NULL;
280 		memblock_free(mem, size);
281 		return;
282 	}
283 
284 	if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
285 		pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
286 
287 	cpu = get_cpu();
288 	vcpu_time = &hv_clock[cpu].pvti;
289 	flags = pvclock_read_flags(vcpu_time);
290 
291 	kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
292 	put_cpu();
293 
294 	x86_platform.calibrate_tsc = kvm_get_tsc_khz;
295 	x86_platform.calibrate_cpu = kvm_get_tsc_khz;
296 	x86_platform.get_wallclock = kvm_get_wallclock;
297 	x86_platform.set_wallclock = kvm_set_wallclock;
298 #ifdef CONFIG_X86_LOCAL_APIC
299 	x86_cpuinit.early_percpu_clock_init =
300 		kvm_setup_secondary_clock;
301 #endif
302 	x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
303 	x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
304 	machine_ops.shutdown  = kvm_shutdown;
305 #ifdef CONFIG_KEXEC_CORE
306 	machine_ops.crash_shutdown  = kvm_crash_shutdown;
307 #endif
308 	kvm_get_preset_lpj();
309 	clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
310 	pv_info.name = "KVM";
311 }
312 
313 int __init kvm_setup_vsyscall_timeinfo(void)
314 {
315 #ifdef CONFIG_X86_64
316 	int cpu;
317 	u8 flags;
318 	struct pvclock_vcpu_time_info *vcpu_time;
319 	unsigned int size;
320 
321 	if (!hv_clock)
322 		return 0;
323 
324 	size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
325 
326 	cpu = get_cpu();
327 
328 	vcpu_time = &hv_clock[cpu].pvti;
329 	flags = pvclock_read_flags(vcpu_time);
330 
331 	if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
332 		put_cpu();
333 		return 1;
334 	}
335 
336 	put_cpu();
337 
338 	kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
339 #endif
340 	return 0;
341 }
342