1 /* KVM paravirtual clock driver. A clocksource implementation
2 Copyright (C) 2008 Glauber de Oliveira Costa, Red Hat Inc.
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.
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.
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
19 #include <linux/clocksource.h>
20 #include <linux/kvm_para.h>
21 #include <asm/pvclock.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>
30 #include <asm/mem_encrypt.h>
31 #include <asm/x86_init.h>
32 #include <asm/reboot.h>
33 #include <asm/kvmclock.h>
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
;
40 static int parse_no_kvmclock(char *arg
)
45 early_param("no-kvmclock", parse_no_kvmclock
);
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
;
51 struct pvclock_vsyscall_time_info
*pvclock_pvti_cpu0_va(void)
55 EXPORT_SYMBOL_GPL(pvclock_pvti_cpu0_va
);
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
62 static void kvm_get_wallclock(struct timespec
*now
)
64 struct pvclock_vcpu_time_info
*vcpu_time
;
68 low
= (int)slow_virt_to_phys(wall_clock
);
69 high
= ((u64
)slow_virt_to_phys(wall_clock
) >> 32);
71 native_write_msr(msr_kvm_wall_clock
, low
, high
);
75 vcpu_time
= &hv_clock
[cpu
].pvti
;
76 pvclock_read_wallclock(wall_clock
, vcpu_time
, now
);
81 static int kvm_set_wallclock(const struct timespec
*now
)
86 static u64
kvm_clock_read(void)
88 struct pvclock_vcpu_time_info
*src
;
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();
100 static u64
kvm_clock_get_cycles(struct clocksource
*cs
)
102 return kvm_clock_read();
105 static u64
kvm_sched_clock_read(void)
107 return kvm_clock_read() - kvm_sched_clock_offset
;
110 static inline void kvm_sched_clock_init(bool stable
)
113 pv_time_ops
.sched_clock
= kvm_clock_read
;
114 clear_sched_clock_stable();
118 kvm_sched_clock_offset
= kvm_clock_read();
119 pv_time_ops
.sched_clock
= kvm_sched_clock_read
;
121 printk(KERN_INFO
"kvm-clock: using sched offset of %llu cycles\n",
122 kvm_sched_clock_offset
);
124 BUILD_BUG_ON(sizeof(kvm_sched_clock_offset
) >
125 sizeof(((struct pvclock_vcpu_time_info
*)NULL
)->system_time
));
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
137 static unsigned long kvm_get_tsc_khz(void)
139 struct pvclock_vcpu_time_info
*src
;
141 unsigned long tsc_khz
;
144 src
= &hv_clock
[cpu
].pvti
;
145 tsc_khz
= pvclock_tsc_khz(src
);
150 static void kvm_get_preset_lpj(void)
155 khz
= kvm_get_tsc_khz();
157 lpj
= ((u64
)khz
* 1000);
162 bool kvm_check_and_clear_guest_paused(void)
165 struct pvclock_vcpu_time_info
*src
;
166 int cpu
= smp_processor_id();
171 src
= &hv_clock
[cpu
].pvti
;
172 if ((src
->flags
& PVCLOCK_GUEST_STOPPED
) != 0) {
173 src
->flags
&= ~PVCLOCK_GUEST_STOPPED
;
174 pvclock_touch_watchdogs();
181 struct clocksource kvm_clock
= {
183 .read
= kvm_clock_get_cycles
,
185 .mask
= CLOCKSOURCE_MASK(64),
186 .flags
= CLOCK_SOURCE_IS_CONTINUOUS
,
188 EXPORT_SYMBOL_GPL(kvm_clock
);
190 int kvm_register_clock(char *txt
)
192 int cpu
= smp_processor_id();
194 struct pvclock_vcpu_time_info
*src
;
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
);
209 static void kvm_save_sched_clock_state(void)
213 static void kvm_restore_sched_clock_state(void)
215 kvm_register_clock("primary cpu clock, resume");
218 #ifdef CONFIG_X86_LOCAL_APIC
219 static void kvm_setup_secondary_clock(void)
222 * Now that the first cpu already had this clocksource initialized,
225 WARN_ON(kvm_register_clock("secondary cpu clock"));
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
237 #ifdef CONFIG_KEXEC_CORE
238 static void kvm_crash_shutdown(struct pt_regs
*regs
)
240 native_write_msr(msr_kvm_system_time
, 0, 0);
241 kvm_disable_steal_time();
242 native_machine_crash_shutdown(regs
);
246 static void kvm_shutdown(void)
248 native_write_msr(msr_kvm_system_time
, 0, 0);
249 kvm_disable_steal_time();
250 native_machine_shutdown();
253 static phys_addr_t __init
kvm_memblock_alloc(phys_addr_t size
,
258 mem
= memblock_alloc(size
, align
);
263 if (early_set_memory_decrypted((unsigned long)__va(mem
), size
))
269 memblock_free(mem
, size
);
273 static void __init
kvm_memblock_free(phys_addr_t addr
, phys_addr_t size
)
276 early_set_memory_encrypted((unsigned long)__va(addr
), size
);
278 memblock_free(addr
, size
);
281 void __init
kvmclock_init(void)
283 struct pvclock_vcpu_time_info
*vcpu_time
;
284 unsigned long mem
, mem_wall_clock
;
285 int size
, cpu
, wall_clock_size
;
288 size
= PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info
)*NR_CPUS
);
290 if (!kvm_para_available())
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
)))
299 wall_clock_size
= PAGE_ALIGN(sizeof(struct pvclock_wall_clock
));
300 mem_wall_clock
= kvm_memblock_alloc(wall_clock_size
, PAGE_SIZE
);
304 wall_clock
= __va(mem_wall_clock
);
305 memset(wall_clock
, 0, wall_clock_size
);
307 mem
= kvm_memblock_alloc(size
, PAGE_SIZE
);
309 kvm_memblock_free(mem_wall_clock
, wall_clock_size
);
314 hv_clock
= __va(mem
);
315 memset(hv_clock
, 0, size
);
317 if (kvm_register_clock("primary cpu clock")) {
319 kvm_memblock_free(mem
, size
);
320 kvm_memblock_free(mem_wall_clock
, wall_clock_size
);
325 printk(KERN_INFO
"kvm-clock: Using msrs %x and %x",
326 msr_kvm_system_time
, msr_kvm_wall_clock
);
328 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT
))
329 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT
);
332 vcpu_time
= &hv_clock
[cpu
].pvti
;
333 flags
= pvclock_read_flags(vcpu_time
);
335 kvm_sched_clock_init(flags
& PVCLOCK_TSC_STABLE_BIT
);
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
;
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
;
352 kvm_get_preset_lpj();
353 clocksource_register_hz(&kvm_clock
, NSEC_PER_SEC
);
354 pv_info
.name
= "KVM";
357 int __init
kvm_setup_vsyscall_timeinfo(void)
362 struct pvclock_vcpu_time_info
*vcpu_time
;
368 size
= PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info
)*NR_CPUS
);
372 vcpu_time
= &hv_clock
[cpu
].pvti
;
373 flags
= pvclock_read_flags(vcpu_time
);
375 if (!(flags
& PVCLOCK_TSC_STABLE_BIT
)) {
382 kvm_clock
.archdata
.vclock_mode
= VCLOCK_PVCLOCK
;