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