Loading...
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
27#include <asm/x86_init.h>
28#include <asm/reboot.h>
29
30static int kvmclock = 1;
31static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
32static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
33
34static int parse_no_kvmclock(char *arg)
35{
36 kvmclock = 0;
37 return 0;
38}
39early_param("no-kvmclock", parse_no_kvmclock);
40
41/* The hypervisor will put information about time periodically here */
42static DEFINE_PER_CPU_SHARED_ALIGNED(struct pvclock_vcpu_time_info, hv_clock);
43static struct pvclock_wall_clock wall_clock;
44
45/*
46 * The wallclock is the time of day when we booted. Since then, some time may
47 * have elapsed since the hypervisor wrote the data. So we try to account for
48 * that with system time
49 */
50static unsigned long kvm_get_wallclock(void)
51{
52 struct pvclock_vcpu_time_info *vcpu_time;
53 struct timespec ts;
54 int low, high;
55
56 low = (int)__pa_symbol(&wall_clock);
57 high = ((u64)__pa_symbol(&wall_clock) >> 32);
58
59 native_write_msr(msr_kvm_wall_clock, low, high);
60
61 vcpu_time = &get_cpu_var(hv_clock);
62 pvclock_read_wallclock(&wall_clock, vcpu_time, &ts);
63 put_cpu_var(hv_clock);
64
65 return ts.tv_sec;
66}
67
68static int kvm_set_wallclock(unsigned long now)
69{
70 return -1;
71}
72
73static cycle_t kvm_clock_read(void)
74{
75 struct pvclock_vcpu_time_info *src;
76 cycle_t ret;
77
78 preempt_disable_notrace();
79 src = &__get_cpu_var(hv_clock);
80 ret = pvclock_clocksource_read(src);
81 preempt_enable_notrace();
82 return ret;
83}
84
85static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
86{
87 return kvm_clock_read();
88}
89
90/*
91 * If we don't do that, there is the possibility that the guest
92 * will calibrate under heavy load - thus, getting a lower lpj -
93 * and execute the delays themselves without load. This is wrong,
94 * because no delay loop can finish beforehand.
95 * Any heuristics is subject to fail, because ultimately, a large
96 * poll of guests can be running and trouble each other. So we preset
97 * lpj here
98 */
99static unsigned long kvm_get_tsc_khz(void)
100{
101 struct pvclock_vcpu_time_info *src;
102 src = &per_cpu(hv_clock, 0);
103 return pvclock_tsc_khz(src);
104}
105
106static void kvm_get_preset_lpj(void)
107{
108 unsigned long khz;
109 u64 lpj;
110
111 khz = kvm_get_tsc_khz();
112
113 lpj = ((u64)khz * 1000);
114 do_div(lpj, HZ);
115 preset_lpj = lpj;
116}
117
118bool kvm_check_and_clear_guest_paused(void)
119{
120 bool ret = false;
121 struct pvclock_vcpu_time_info *src;
122
123 src = &__get_cpu_var(hv_clock);
124 if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
125 __this_cpu_and(hv_clock.flags, ~PVCLOCK_GUEST_STOPPED);
126 ret = true;
127 }
128
129 return ret;
130}
131
132static struct clocksource kvm_clock = {
133 .name = "kvm-clock",
134 .read = kvm_clock_get_cycles,
135 .rating = 400,
136 .mask = CLOCKSOURCE_MASK(64),
137 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
138};
139
140int kvm_register_clock(char *txt)
141{
142 int cpu = smp_processor_id();
143 int low, high, ret;
144
145 low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1;
146 high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32);
147 ret = native_write_msr_safe(msr_kvm_system_time, low, high);
148 printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
149 cpu, high, low, txt);
150
151 return ret;
152}
153
154static void kvm_save_sched_clock_state(void)
155{
156}
157
158static void kvm_restore_sched_clock_state(void)
159{
160 kvm_register_clock("primary cpu clock, resume");
161}
162
163#ifdef CONFIG_X86_LOCAL_APIC
164static void __cpuinit kvm_setup_secondary_clock(void)
165{
166 /*
167 * Now that the first cpu already had this clocksource initialized,
168 * we shouldn't fail.
169 */
170 WARN_ON(kvm_register_clock("secondary cpu clock"));
171}
172#endif
173
174/*
175 * After the clock is registered, the host will keep writing to the
176 * registered memory location. If the guest happens to shutdown, this memory
177 * won't be valid. In cases like kexec, in which you install a new kernel, this
178 * means a random memory location will be kept being written. So before any
179 * kind of shutdown from our side, we unregister the clock by writting anything
180 * that does not have the 'enable' bit set in the msr
181 */
182#ifdef CONFIG_KEXEC
183static void kvm_crash_shutdown(struct pt_regs *regs)
184{
185 native_write_msr(msr_kvm_system_time, 0, 0);
186 kvm_disable_steal_time();
187 native_machine_crash_shutdown(regs);
188}
189#endif
190
191static void kvm_shutdown(void)
192{
193 native_write_msr(msr_kvm_system_time, 0, 0);
194 kvm_disable_steal_time();
195 native_machine_shutdown();
196}
197
198void __init kvmclock_init(void)
199{
200 if (!kvm_para_available())
201 return;
202
203 if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
204 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
205 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
206 } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
207 return;
208
209 printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
210 msr_kvm_system_time, msr_kvm_wall_clock);
211
212 if (kvm_register_clock("boot clock"))
213 return;
214 pv_time_ops.sched_clock = kvm_clock_read;
215 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
216 x86_platform.get_wallclock = kvm_get_wallclock;
217 x86_platform.set_wallclock = kvm_set_wallclock;
218#ifdef CONFIG_X86_LOCAL_APIC
219 x86_cpuinit.early_percpu_clock_init =
220 kvm_setup_secondary_clock;
221#endif
222 x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
223 x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
224 machine_ops.shutdown = kvm_shutdown;
225#ifdef CONFIG_KEXEC
226 machine_ops.crash_shutdown = kvm_crash_shutdown;
227#endif
228 kvm_get_preset_lpj();
229 clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
230 pv_info.paravirt_enabled = 1;
231 pv_info.name = "KVM";
232
233 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
234 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
235}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* KVM paravirtual clock driver. A clocksource implementation
3 Copyright (C) 2008 Glauber de Oliveira Costa, Red Hat Inc.
4*/
5
6#include <linux/clocksource.h>
7#include <linux/kvm_para.h>
8#include <asm/pvclock.h>
9#include <asm/msr.h>
10#include <asm/apic.h>
11#include <linux/percpu.h>
12#include <linux/hardirq.h>
13#include <linux/cpuhotplug.h>
14#include <linux/sched.h>
15#include <linux/sched/clock.h>
16#include <linux/mm.h>
17#include <linux/slab.h>
18#include <linux/set_memory.h>
19#include <linux/cc_platform.h>
20
21#include <asm/hypervisor.h>
22#include <asm/x86_init.h>
23#include <asm/kvmclock.h>
24
25static int kvmclock __initdata = 1;
26static int kvmclock_vsyscall __initdata = 1;
27static int msr_kvm_system_time __ro_after_init = MSR_KVM_SYSTEM_TIME;
28static int msr_kvm_wall_clock __ro_after_init = MSR_KVM_WALL_CLOCK;
29static u64 kvm_sched_clock_offset __ro_after_init;
30
31static int __init parse_no_kvmclock(char *arg)
32{
33 kvmclock = 0;
34 return 0;
35}
36early_param("no-kvmclock", parse_no_kvmclock);
37
38static int __init parse_no_kvmclock_vsyscall(char *arg)
39{
40 kvmclock_vsyscall = 0;
41 return 0;
42}
43early_param("no-kvmclock-vsyscall", parse_no_kvmclock_vsyscall);
44
45/* Aligned to page sizes to match whats mapped via vsyscalls to userspace */
46#define HVC_BOOT_ARRAY_SIZE \
47 (PAGE_SIZE / sizeof(struct pvclock_vsyscall_time_info))
48
49static struct pvclock_vsyscall_time_info
50 hv_clock_boot[HVC_BOOT_ARRAY_SIZE] __bss_decrypted __aligned(PAGE_SIZE);
51static struct pvclock_wall_clock wall_clock __bss_decrypted;
52static struct pvclock_vsyscall_time_info *hvclock_mem;
53DEFINE_PER_CPU(struct pvclock_vsyscall_time_info *, hv_clock_per_cpu);
54EXPORT_PER_CPU_SYMBOL_GPL(hv_clock_per_cpu);
55
56/*
57 * The wallclock is the time of day when we booted. Since then, some time may
58 * have elapsed since the hypervisor wrote the data. So we try to account for
59 * that with system time
60 */
61static void kvm_get_wallclock(struct timespec64 *now)
62{
63 wrmsrl(msr_kvm_wall_clock, slow_virt_to_phys(&wall_clock));
64 preempt_disable();
65 pvclock_read_wallclock(&wall_clock, this_cpu_pvti(), now);
66 preempt_enable();
67}
68
69static int kvm_set_wallclock(const struct timespec64 *now)
70{
71 return -ENODEV;
72}
73
74static u64 kvm_clock_read(void)
75{
76 u64 ret;
77
78 preempt_disable_notrace();
79 ret = pvclock_clocksource_read(this_cpu_pvti());
80 preempt_enable_notrace();
81 return ret;
82}
83
84static u64 kvm_clock_get_cycles(struct clocksource *cs)
85{
86 return kvm_clock_read();
87}
88
89static u64 kvm_sched_clock_read(void)
90{
91 return kvm_clock_read() - kvm_sched_clock_offset;
92}
93
94static inline void kvm_sched_clock_init(bool stable)
95{
96 if (!stable)
97 clear_sched_clock_stable();
98 kvm_sched_clock_offset = kvm_clock_read();
99 paravirt_set_sched_clock(kvm_sched_clock_read);
100
101 pr_info("kvm-clock: using sched offset of %llu cycles",
102 kvm_sched_clock_offset);
103
104 BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
105 sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
106}
107
108/*
109 * If we don't do that, there is the possibility that the guest
110 * will calibrate under heavy load - thus, getting a lower lpj -
111 * and execute the delays themselves without load. This is wrong,
112 * because no delay loop can finish beforehand.
113 * Any heuristics is subject to fail, because ultimately, a large
114 * poll of guests can be running and trouble each other. So we preset
115 * lpj here
116 */
117static unsigned long kvm_get_tsc_khz(void)
118{
119 setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
120 return pvclock_tsc_khz(this_cpu_pvti());
121}
122
123static void __init kvm_get_preset_lpj(void)
124{
125 unsigned long khz;
126 u64 lpj;
127
128 khz = kvm_get_tsc_khz();
129
130 lpj = ((u64)khz * 1000);
131 do_div(lpj, HZ);
132 preset_lpj = lpj;
133}
134
135bool kvm_check_and_clear_guest_paused(void)
136{
137 struct pvclock_vsyscall_time_info *src = this_cpu_hvclock();
138 bool ret = false;
139
140 if (!src)
141 return ret;
142
143 if ((src->pvti.flags & PVCLOCK_GUEST_STOPPED) != 0) {
144 src->pvti.flags &= ~PVCLOCK_GUEST_STOPPED;
145 pvclock_touch_watchdogs();
146 ret = true;
147 }
148 return ret;
149}
150
151static int kvm_cs_enable(struct clocksource *cs)
152{
153 vclocks_set_used(VDSO_CLOCKMODE_PVCLOCK);
154 return 0;
155}
156
157struct clocksource kvm_clock = {
158 .name = "kvm-clock",
159 .read = kvm_clock_get_cycles,
160 .rating = 400,
161 .mask = CLOCKSOURCE_MASK(64),
162 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
163 .enable = kvm_cs_enable,
164};
165EXPORT_SYMBOL_GPL(kvm_clock);
166
167static void kvm_register_clock(char *txt)
168{
169 struct pvclock_vsyscall_time_info *src = this_cpu_hvclock();
170 u64 pa;
171
172 if (!src)
173 return;
174
175 pa = slow_virt_to_phys(&src->pvti) | 0x01ULL;
176 wrmsrl(msr_kvm_system_time, pa);
177 pr_debug("kvm-clock: cpu %d, msr %llx, %s", smp_processor_id(), pa, txt);
178}
179
180static void kvm_save_sched_clock_state(void)
181{
182}
183
184static void kvm_restore_sched_clock_state(void)
185{
186 kvm_register_clock("primary cpu clock, resume");
187}
188
189#ifdef CONFIG_X86_LOCAL_APIC
190static void kvm_setup_secondary_clock(void)
191{
192 kvm_register_clock("secondary cpu clock");
193}
194#endif
195
196void kvmclock_disable(void)
197{
198 native_write_msr(msr_kvm_system_time, 0, 0);
199}
200
201static void __init kvmclock_init_mem(void)
202{
203 unsigned long ncpus;
204 unsigned int order;
205 struct page *p;
206 int r;
207
208 if (HVC_BOOT_ARRAY_SIZE >= num_possible_cpus())
209 return;
210
211 ncpus = num_possible_cpus() - HVC_BOOT_ARRAY_SIZE;
212 order = get_order(ncpus * sizeof(*hvclock_mem));
213
214 p = alloc_pages(GFP_KERNEL, order);
215 if (!p) {
216 pr_warn("%s: failed to alloc %d pages", __func__, (1U << order));
217 return;
218 }
219
220 hvclock_mem = page_address(p);
221
222 /*
223 * hvclock is shared between the guest and the hypervisor, must
224 * be mapped decrypted.
225 */
226 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
227 r = set_memory_decrypted((unsigned long) hvclock_mem,
228 1UL << order);
229 if (r) {
230 __free_pages(p, order);
231 hvclock_mem = NULL;
232 pr_warn("kvmclock: set_memory_decrypted() failed. Disabling\n");
233 return;
234 }
235 }
236
237 memset(hvclock_mem, 0, PAGE_SIZE << order);
238}
239
240static int __init kvm_setup_vsyscall_timeinfo(void)
241{
242 if (!kvm_para_available() || !kvmclock || nopv)
243 return 0;
244
245 kvmclock_init_mem();
246
247#ifdef CONFIG_X86_64
248 if (per_cpu(hv_clock_per_cpu, 0) && kvmclock_vsyscall) {
249 u8 flags;
250
251 flags = pvclock_read_flags(&hv_clock_boot[0].pvti);
252 if (!(flags & PVCLOCK_TSC_STABLE_BIT))
253 return 0;
254
255 kvm_clock.vdso_clock_mode = VDSO_CLOCKMODE_PVCLOCK;
256 }
257#endif
258
259 return 0;
260}
261early_initcall(kvm_setup_vsyscall_timeinfo);
262
263static int kvmclock_setup_percpu(unsigned int cpu)
264{
265 struct pvclock_vsyscall_time_info *p = per_cpu(hv_clock_per_cpu, cpu);
266
267 /*
268 * The per cpu area setup replicates CPU0 data to all cpu
269 * pointers. So carefully check. CPU0 has been set up in init
270 * already.
271 */
272 if (!cpu || (p && p != per_cpu(hv_clock_per_cpu, 0)))
273 return 0;
274
275 /* Use the static page for the first CPUs, allocate otherwise */
276 if (cpu < HVC_BOOT_ARRAY_SIZE)
277 p = &hv_clock_boot[cpu];
278 else if (hvclock_mem)
279 p = hvclock_mem + cpu - HVC_BOOT_ARRAY_SIZE;
280 else
281 return -ENOMEM;
282
283 per_cpu(hv_clock_per_cpu, cpu) = p;
284 return p ? 0 : -ENOMEM;
285}
286
287void __init kvmclock_init(void)
288{
289 u8 flags;
290
291 if (!kvm_para_available() || !kvmclock)
292 return;
293
294 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
295 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
296 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
297 } else if (!kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)) {
298 return;
299 }
300
301 if (cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "kvmclock:setup_percpu",
302 kvmclock_setup_percpu, NULL) < 0) {
303 return;
304 }
305
306 pr_info("kvm-clock: Using msrs %x and %x",
307 msr_kvm_system_time, msr_kvm_wall_clock);
308
309 this_cpu_write(hv_clock_per_cpu, &hv_clock_boot[0]);
310 kvm_register_clock("primary cpu clock");
311 pvclock_set_pvti_cpu0_va(hv_clock_boot);
312
313 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
314 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
315
316 flags = pvclock_read_flags(&hv_clock_boot[0].pvti);
317 kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
318
319 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
320 x86_platform.calibrate_cpu = kvm_get_tsc_khz;
321 x86_platform.get_wallclock = kvm_get_wallclock;
322 x86_platform.set_wallclock = kvm_set_wallclock;
323#ifdef CONFIG_X86_LOCAL_APIC
324 x86_cpuinit.early_percpu_clock_init = kvm_setup_secondary_clock;
325#endif
326 x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
327 x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
328 kvm_get_preset_lpj();
329
330 /*
331 * X86_FEATURE_NONSTOP_TSC is TSC runs at constant rate
332 * with P/T states and does not stop in deep C-states.
333 *
334 * Invariant TSC exposed by host means kvmclock is not necessary:
335 * can use TSC as clocksource.
336 *
337 */
338 if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) &&
339 boot_cpu_has(X86_FEATURE_NONSTOP_TSC) &&
340 !check_tsc_unstable())
341 kvm_clock.rating = 299;
342
343 clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
344 pv_info.name = "KVM";
345}