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#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
35static int kvmclock __ro_after_init = 1;
36static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
37static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
38static u64 kvm_sched_clock_offset;
39
40static int parse_no_kvmclock(char *arg)
41{
42 kvmclock = 0;
43 return 0;
44}
45early_param("no-kvmclock", parse_no_kvmclock);
46
47/* The hypervisor will put information about time periodically here */
48static struct pvclock_vsyscall_time_info *hv_clock;
49static struct pvclock_wall_clock *wall_clock;
50
51/*
52 * The wallclock is the time of day when we booted. Since then, some time may
53 * have elapsed since the hypervisor wrote the data. So we try to account for
54 * that with system time
55 */
56static void kvm_get_wallclock(struct timespec *now)
57{
58 struct pvclock_vcpu_time_info *vcpu_time;
59 int low, high;
60 int cpu;
61
62 low = (int)slow_virt_to_phys(wall_clock);
63 high = ((u64)slow_virt_to_phys(wall_clock) >> 32);
64
65 native_write_msr(msr_kvm_wall_clock, low, high);
66
67 cpu = get_cpu();
68
69 vcpu_time = &hv_clock[cpu].pvti;
70 pvclock_read_wallclock(wall_clock, vcpu_time, now);
71
72 put_cpu();
73}
74
75static int kvm_set_wallclock(const struct timespec *now)
76{
77 return -ENODEV;
78}
79
80static u64 kvm_clock_read(void)
81{
82 struct pvclock_vcpu_time_info *src;
83 u64 ret;
84 int cpu;
85
86 preempt_disable_notrace();
87 cpu = smp_processor_id();
88 src = &hv_clock[cpu].pvti;
89 ret = pvclock_clocksource_read(src);
90 preempt_enable_notrace();
91 return ret;
92}
93
94static u64 kvm_clock_get_cycles(struct clocksource *cs)
95{
96 return kvm_clock_read();
97}
98
99static u64 kvm_sched_clock_read(void)
100{
101 return kvm_clock_read() - kvm_sched_clock_offset;
102}
103
104static inline void kvm_sched_clock_init(bool stable)
105{
106 if (!stable) {
107 pv_time_ops.sched_clock = kvm_clock_read;
108 clear_sched_clock_stable();
109 return;
110 }
111
112 kvm_sched_clock_offset = kvm_clock_read();
113 pv_time_ops.sched_clock = kvm_sched_clock_read;
114
115 printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
116 kvm_sched_clock_offset);
117
118 BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
119 sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
120}
121
122/*
123 * If we don't do that, there is the possibility that the guest
124 * will calibrate under heavy load - thus, getting a lower lpj -
125 * and execute the delays themselves without load. This is wrong,
126 * because no delay loop can finish beforehand.
127 * Any heuristics is subject to fail, because ultimately, a large
128 * poll of guests can be running and trouble each other. So we preset
129 * lpj here
130 */
131static unsigned long kvm_get_tsc_khz(void)
132{
133 struct pvclock_vcpu_time_info *src;
134 int cpu;
135 unsigned long tsc_khz;
136
137 cpu = get_cpu();
138 src = &hv_clock[cpu].pvti;
139 tsc_khz = pvclock_tsc_khz(src);
140 put_cpu();
141 return tsc_khz;
142}
143
144static void kvm_get_preset_lpj(void)
145{
146 unsigned long khz;
147 u64 lpj;
148
149 khz = kvm_get_tsc_khz();
150
151 lpj = ((u64)khz * 1000);
152 do_div(lpj, HZ);
153 preset_lpj = lpj;
154}
155
156bool kvm_check_and_clear_guest_paused(void)
157{
158 bool ret = false;
159 struct pvclock_vcpu_time_info *src;
160 int cpu = smp_processor_id();
161
162 if (!hv_clock)
163 return ret;
164
165 src = &hv_clock[cpu].pvti;
166 if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
167 src->flags &= ~PVCLOCK_GUEST_STOPPED;
168 pvclock_touch_watchdogs();
169 ret = true;
170 }
171
172 return ret;
173}
174
175struct clocksource kvm_clock = {
176 .name = "kvm-clock",
177 .read = kvm_clock_get_cycles,
178 .rating = 400,
179 .mask = CLOCKSOURCE_MASK(64),
180 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
181};
182EXPORT_SYMBOL_GPL(kvm_clock);
183
184int kvm_register_clock(char *txt)
185{
186 int cpu = smp_processor_id();
187 int low, high, ret;
188 struct pvclock_vcpu_time_info *src;
189
190 if (!hv_clock)
191 return 0;
192
193 src = &hv_clock[cpu].pvti;
194 low = (int)slow_virt_to_phys(src) | 1;
195 high = ((u64)slow_virt_to_phys(src) >> 32);
196 ret = native_write_msr_safe(msr_kvm_system_time, low, high);
197 printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
198 cpu, high, low, txt);
199
200 return ret;
201}
202
203static void kvm_save_sched_clock_state(void)
204{
205}
206
207static void kvm_restore_sched_clock_state(void)
208{
209 kvm_register_clock("primary cpu clock, resume");
210}
211
212#ifdef CONFIG_X86_LOCAL_APIC
213static void kvm_setup_secondary_clock(void)
214{
215 /*
216 * Now that the first cpu already had this clocksource initialized,
217 * we shouldn't fail.
218 */
219 WARN_ON(kvm_register_clock("secondary cpu clock"));
220}
221#endif
222
223/*
224 * After the clock is registered, the host will keep writing to the
225 * registered memory location. If the guest happens to shutdown, this memory
226 * won't be valid. In cases like kexec, in which you install a new kernel, this
227 * means a random memory location will be kept being written. So before any
228 * kind of shutdown from our side, we unregister the clock by writing anything
229 * that does not have the 'enable' bit set in the msr
230 */
231#ifdef CONFIG_KEXEC_CORE
232static void kvm_crash_shutdown(struct pt_regs *regs)
233{
234 native_write_msr(msr_kvm_system_time, 0, 0);
235 kvm_disable_steal_time();
236 native_machine_crash_shutdown(regs);
237}
238#endif
239
240static void kvm_shutdown(void)
241{
242 native_write_msr(msr_kvm_system_time, 0, 0);
243 kvm_disable_steal_time();
244 native_machine_shutdown();
245}
246
247static phys_addr_t __init kvm_memblock_alloc(phys_addr_t size,
248 phys_addr_t align)
249{
250 phys_addr_t mem;
251
252 mem = memblock_alloc(size, align);
253 if (!mem)
254 return 0;
255
256 if (sev_active()) {
257 if (early_set_memory_decrypted((unsigned long)__va(mem), size))
258 goto e_free;
259 }
260
261 return mem;
262e_free:
263 memblock_free(mem, size);
264 return 0;
265}
266
267static void __init kvm_memblock_free(phys_addr_t addr, phys_addr_t size)
268{
269 if (sev_active())
270 early_set_memory_encrypted((unsigned long)__va(addr), size);
271
272 memblock_free(addr, size);
273}
274
275void __init kvmclock_init(void)
276{
277 struct pvclock_vcpu_time_info *vcpu_time;
278 unsigned long mem, mem_wall_clock;
279 int size, cpu, wall_clock_size;
280 u8 flags;
281
282 size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
283
284 if (!kvm_para_available())
285 return;
286
287 if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
288 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
289 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
290 } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
291 return;
292
293 wall_clock_size = PAGE_ALIGN(sizeof(struct pvclock_wall_clock));
294 mem_wall_clock = kvm_memblock_alloc(wall_clock_size, PAGE_SIZE);
295 if (!mem_wall_clock)
296 return;
297
298 wall_clock = __va(mem_wall_clock);
299 memset(wall_clock, 0, wall_clock_size);
300
301 mem = kvm_memblock_alloc(size, PAGE_SIZE);
302 if (!mem) {
303 kvm_memblock_free(mem_wall_clock, wall_clock_size);
304 wall_clock = NULL;
305 return;
306 }
307
308 hv_clock = __va(mem);
309 memset(hv_clock, 0, size);
310
311 if (kvm_register_clock("primary cpu clock")) {
312 hv_clock = NULL;
313 kvm_memblock_free(mem, size);
314 kvm_memblock_free(mem_wall_clock, wall_clock_size);
315 wall_clock = NULL;
316 return;
317 }
318
319 printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
320 msr_kvm_system_time, msr_kvm_wall_clock);
321
322 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
323 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
324
325 cpu = get_cpu();
326 vcpu_time = &hv_clock[cpu].pvti;
327 flags = pvclock_read_flags(vcpu_time);
328
329 kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
330 put_cpu();
331
332 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
333 x86_platform.calibrate_cpu = kvm_get_tsc_khz;
334 x86_platform.get_wallclock = kvm_get_wallclock;
335 x86_platform.set_wallclock = kvm_set_wallclock;
336#ifdef CONFIG_X86_LOCAL_APIC
337 x86_cpuinit.early_percpu_clock_init =
338 kvm_setup_secondary_clock;
339#endif
340 x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
341 x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
342 machine_ops.shutdown = kvm_shutdown;
343#ifdef CONFIG_KEXEC_CORE
344 machine_ops.crash_shutdown = kvm_crash_shutdown;
345#endif
346 kvm_get_preset_lpj();
347 clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
348 pv_info.name = "KVM";
349}
350
351int __init kvm_setup_vsyscall_timeinfo(void)
352{
353#ifdef CONFIG_X86_64
354 int cpu;
355 u8 flags;
356 struct pvclock_vcpu_time_info *vcpu_time;
357 unsigned int size;
358
359 if (!hv_clock)
360 return 0;
361
362 size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
363
364 cpu = get_cpu();
365
366 vcpu_time = &hv_clock[cpu].pvti;
367 flags = pvclock_read_flags(vcpu_time);
368
369 if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
370 put_cpu();
371 return 1;
372 }
373
374 pvclock_set_pvti_cpu0_va(hv_clock);
375 put_cpu();
376
377 kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
378#endif
379 return 0;
380}
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
26#include <asm/x86_init.h>
27#include <asm/reboot.h>
28
29static int kvmclock = 1;
30static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
31static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
32
33static int parse_no_kvmclock(char *arg)
34{
35 kvmclock = 0;
36 return 0;
37}
38early_param("no-kvmclock", parse_no_kvmclock);
39
40/* The hypervisor will put information about time periodically here */
41static DEFINE_PER_CPU_SHARED_ALIGNED(struct pvclock_vcpu_time_info, hv_clock);
42static struct pvclock_wall_clock wall_clock;
43
44/*
45 * The wallclock is the time of day when we booted. Since then, some time may
46 * have elapsed since the hypervisor wrote the data. So we try to account for
47 * that with system time
48 */
49static unsigned long kvm_get_wallclock(void)
50{
51 struct pvclock_vcpu_time_info *vcpu_time;
52 struct timespec ts;
53 int low, high;
54
55 low = (int)__pa_symbol(&wall_clock);
56 high = ((u64)__pa_symbol(&wall_clock) >> 32);
57
58 native_write_msr(msr_kvm_wall_clock, low, high);
59
60 vcpu_time = &get_cpu_var(hv_clock);
61 pvclock_read_wallclock(&wall_clock, vcpu_time, &ts);
62 put_cpu_var(hv_clock);
63
64 return ts.tv_sec;
65}
66
67static int kvm_set_wallclock(unsigned long now)
68{
69 return -1;
70}
71
72static cycle_t kvm_clock_read(void)
73{
74 struct pvclock_vcpu_time_info *src;
75 cycle_t ret;
76
77 src = &get_cpu_var(hv_clock);
78 ret = pvclock_clocksource_read(src);
79 put_cpu_var(hv_clock);
80 return ret;
81}
82
83static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
84{
85 return kvm_clock_read();
86}
87
88/*
89 * If we don't do that, there is the possibility that the guest
90 * will calibrate under heavy load - thus, getting a lower lpj -
91 * and execute the delays themselves without load. This is wrong,
92 * because no delay loop can finish beforehand.
93 * Any heuristics is subject to fail, because ultimately, a large
94 * poll of guests can be running and trouble each other. So we preset
95 * lpj here
96 */
97static unsigned long kvm_get_tsc_khz(void)
98{
99 struct pvclock_vcpu_time_info *src;
100 src = &per_cpu(hv_clock, 0);
101 return pvclock_tsc_khz(src);
102}
103
104static void kvm_get_preset_lpj(void)
105{
106 unsigned long khz;
107 u64 lpj;
108
109 khz = kvm_get_tsc_khz();
110
111 lpj = ((u64)khz * 1000);
112 do_div(lpj, HZ);
113 preset_lpj = lpj;
114}
115
116static struct clocksource kvm_clock = {
117 .name = "kvm-clock",
118 .read = kvm_clock_get_cycles,
119 .rating = 400,
120 .mask = CLOCKSOURCE_MASK(64),
121 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
122};
123
124int kvm_register_clock(char *txt)
125{
126 int cpu = smp_processor_id();
127 int low, high, ret;
128
129 low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1;
130 high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32);
131 ret = native_write_msr_safe(msr_kvm_system_time, low, high);
132 printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
133 cpu, high, low, txt);
134
135 return ret;
136}
137
138#ifdef CONFIG_X86_LOCAL_APIC
139static void __cpuinit kvm_setup_secondary_clock(void)
140{
141 /*
142 * Now that the first cpu already had this clocksource initialized,
143 * we shouldn't fail.
144 */
145 WARN_ON(kvm_register_clock("secondary cpu clock"));
146 /* ok, done with our trickery, call native */
147 setup_secondary_APIC_clock();
148}
149#endif
150
151/*
152 * After the clock is registered, the host will keep writing to the
153 * registered memory location. If the guest happens to shutdown, this memory
154 * won't be valid. In cases like kexec, in which you install a new kernel, this
155 * means a random memory location will be kept being written. So before any
156 * kind of shutdown from our side, we unregister the clock by writting anything
157 * that does not have the 'enable' bit set in the msr
158 */
159#ifdef CONFIG_KEXEC
160static void kvm_crash_shutdown(struct pt_regs *regs)
161{
162 native_write_msr(msr_kvm_system_time, 0, 0);
163 kvm_disable_steal_time();
164 native_machine_crash_shutdown(regs);
165}
166#endif
167
168static void kvm_shutdown(void)
169{
170 native_write_msr(msr_kvm_system_time, 0, 0);
171 kvm_disable_steal_time();
172 native_machine_shutdown();
173}
174
175void __init kvmclock_init(void)
176{
177 if (!kvm_para_available())
178 return;
179
180 if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
181 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
182 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
183 } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
184 return;
185
186 printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
187 msr_kvm_system_time, msr_kvm_wall_clock);
188
189 if (kvm_register_clock("boot clock"))
190 return;
191 pv_time_ops.sched_clock = kvm_clock_read;
192 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
193 x86_platform.get_wallclock = kvm_get_wallclock;
194 x86_platform.set_wallclock = kvm_set_wallclock;
195#ifdef CONFIG_X86_LOCAL_APIC
196 x86_cpuinit.setup_percpu_clockev =
197 kvm_setup_secondary_clock;
198#endif
199 machine_ops.shutdown = kvm_shutdown;
200#ifdef CONFIG_KEXEC
201 machine_ops.crash_shutdown = kvm_crash_shutdown;
202#endif
203 kvm_get_preset_lpj();
204 clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
205 pv_info.paravirt_enabled = 1;
206 pv_info.name = "KVM";
207
208 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
209 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
210}