1 // SPDX-License-Identifier: GPL-2.0
2
3 /*
4 * Clocksource driver for the synthetic counter and timers
5 * provided by the Hyper-V hypervisor to guest VMs, as described
6 * in the Hyper-V Top Level Functional Spec (TLFS). This driver
7 * is instruction set architecture independent.
8 *
9 * Copyright (C) 2019, Microsoft, Inc.
10 *
11 * Author: Michael Kelley <mikelley@microsoft.com>
12 */
13
14 #include <linux/percpu.h>
15 #include <linux/cpumask.h>
16 #include <linux/clockchips.h>
17 #include <linux/clocksource.h>
18 #include <linux/sched_clock.h>
19 #include <linux/mm.h>
20 #include <clocksource/hyperv_timer.h>
21 #include <asm/hyperv-tlfs.h>
22 #include <asm/mshyperv.h>
23
24 static struct clock_event_device __percpu *hv_clock_event;
25 static u64 hv_sched_clock_offset __ro_after_init;
26
27 /*
28 * If false, we're using the old mechanism for stimer0 interrupts
29 * where it sends a VMbus message when it expires. The old
30 * mechanism is used when running on older versions of Hyper-V
31 * that don't support Direct Mode. While Hyper-V provides
32 * four stimer's per CPU, Linux uses only stimer0.
33 */
34 static bool direct_mode_enabled;
35
36 static int stimer0_irq;
37 static int stimer0_vector;
38 static int stimer0_message_sint;
39
40 /*
41 * ISR for when stimer0 is operating in Direct Mode. Direct Mode
42 * does not use VMbus or any VMbus messages, so process here and not
43 * in the VMbus driver code.
44 */
hv_stimer0_isr(void)45 void hv_stimer0_isr(void)
46 {
47 struct clock_event_device *ce;
48
49 ce = this_cpu_ptr(hv_clock_event);
50 ce->event_handler(ce);
51 }
52 EXPORT_SYMBOL_GPL(hv_stimer0_isr);
53
hv_ce_set_next_event(unsigned long delta,struct clock_event_device * evt)54 static int hv_ce_set_next_event(unsigned long delta,
55 struct clock_event_device *evt)
56 {
57 u64 current_tick;
58
59 current_tick = hyperv_cs->read(NULL);
60 current_tick += delta;
61 hv_init_timer(0, current_tick);
62 return 0;
63 }
64
hv_ce_shutdown(struct clock_event_device * evt)65 static int hv_ce_shutdown(struct clock_event_device *evt)
66 {
67 hv_init_timer(0, 0);
68 hv_init_timer_config(0, 0);
69 if (direct_mode_enabled)
70 hv_disable_stimer0_percpu_irq(stimer0_irq);
71
72 return 0;
73 }
74
hv_ce_set_oneshot(struct clock_event_device * evt)75 static int hv_ce_set_oneshot(struct clock_event_device *evt)
76 {
77 union hv_stimer_config timer_cfg;
78
79 timer_cfg.as_uint64 = 0;
80 timer_cfg.enable = 1;
81 timer_cfg.auto_enable = 1;
82 if (direct_mode_enabled) {
83 /*
84 * When it expires, the timer will directly interrupt
85 * on the specified hardware vector/IRQ.
86 */
87 timer_cfg.direct_mode = 1;
88 timer_cfg.apic_vector = stimer0_vector;
89 hv_enable_stimer0_percpu_irq(stimer0_irq);
90 } else {
91 /*
92 * When it expires, the timer will generate a VMbus message,
93 * to be handled by the normal VMbus interrupt handler.
94 */
95 timer_cfg.direct_mode = 0;
96 timer_cfg.sintx = stimer0_message_sint;
97 }
98 hv_init_timer_config(0, timer_cfg.as_uint64);
99 return 0;
100 }
101
102 /*
103 * hv_stimer_init - Per-cpu initialization of the clockevent
104 */
hv_stimer_init(unsigned int cpu)105 void hv_stimer_init(unsigned int cpu)
106 {
107 struct clock_event_device *ce;
108
109 /*
110 * Synthetic timers are always available except on old versions of
111 * Hyper-V on x86. In that case, just return as Linux will use a
112 * clocksource based on emulated PIT or LAPIC timer hardware.
113 */
114 if (!(ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE))
115 return;
116
117 ce = per_cpu_ptr(hv_clock_event, cpu);
118 ce->name = "Hyper-V clockevent";
119 ce->features = CLOCK_EVT_FEAT_ONESHOT;
120 ce->cpumask = cpumask_of(cpu);
121 ce->rating = 1000;
122 ce->set_state_shutdown = hv_ce_shutdown;
123 ce->set_state_oneshot = hv_ce_set_oneshot;
124 ce->set_next_event = hv_ce_set_next_event;
125
126 clockevents_config_and_register(ce,
127 HV_CLOCK_HZ,
128 HV_MIN_DELTA_TICKS,
129 HV_MAX_MAX_DELTA_TICKS);
130 }
131 EXPORT_SYMBOL_GPL(hv_stimer_init);
132
133 /*
134 * hv_stimer_cleanup - Per-cpu cleanup of the clockevent
135 */
hv_stimer_cleanup(unsigned int cpu)136 void hv_stimer_cleanup(unsigned int cpu)
137 {
138 struct clock_event_device *ce;
139
140 /* Turn off clockevent device */
141 if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE) {
142 ce = per_cpu_ptr(hv_clock_event, cpu);
143 hv_ce_shutdown(ce);
144 }
145 }
146 EXPORT_SYMBOL_GPL(hv_stimer_cleanup);
147
148 /* hv_stimer_alloc - Global initialization of the clockevent and stimer0 */
hv_stimer_alloc(int sint)149 int hv_stimer_alloc(int sint)
150 {
151 int ret;
152
153 hv_clock_event = alloc_percpu(struct clock_event_device);
154 if (!hv_clock_event)
155 return -ENOMEM;
156
157 direct_mode_enabled = ms_hyperv.misc_features &
158 HV_STIMER_DIRECT_MODE_AVAILABLE;
159 if (direct_mode_enabled) {
160 ret = hv_setup_stimer0_irq(&stimer0_irq, &stimer0_vector,
161 hv_stimer0_isr);
162 if (ret) {
163 free_percpu(hv_clock_event);
164 hv_clock_event = NULL;
165 return ret;
166 }
167 }
168
169 stimer0_message_sint = sint;
170 return 0;
171 }
172 EXPORT_SYMBOL_GPL(hv_stimer_alloc);
173
174 /* hv_stimer_free - Free global resources allocated by hv_stimer_alloc() */
hv_stimer_free(void)175 void hv_stimer_free(void)
176 {
177 if (direct_mode_enabled && (stimer0_irq != 0)) {
178 hv_remove_stimer0_irq(stimer0_irq);
179 stimer0_irq = 0;
180 }
181 free_percpu(hv_clock_event);
182 hv_clock_event = NULL;
183 }
184 EXPORT_SYMBOL_GPL(hv_stimer_free);
185
186 /*
187 * Do a global cleanup of clockevents for the cases of kexec and
188 * vmbus exit
189 */
hv_stimer_global_cleanup(void)190 void hv_stimer_global_cleanup(void)
191 {
192 int cpu;
193 struct clock_event_device *ce;
194
195 if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE) {
196 for_each_present_cpu(cpu) {
197 ce = per_cpu_ptr(hv_clock_event, cpu);
198 clockevents_unbind_device(ce, cpu);
199 }
200 }
201 hv_stimer_free();
202 }
203 EXPORT_SYMBOL_GPL(hv_stimer_global_cleanup);
204
205 /*
206 * Code and definitions for the Hyper-V clocksources. Two
207 * clocksources are defined: one that reads the Hyper-V defined MSR, and
208 * the other that uses the TSC reference page feature as defined in the
209 * TLFS. The MSR version is for compatibility with old versions of
210 * Hyper-V and 32-bit x86. The TSC reference page version is preferred.
211 */
212
213 struct clocksource *hyperv_cs;
214 EXPORT_SYMBOL_GPL(hyperv_cs);
215
216 static struct ms_hyperv_tsc_page tsc_pg __aligned(PAGE_SIZE);
217
hv_get_tsc_page(void)218 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
219 {
220 return &tsc_pg;
221 }
222 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
223
read_hv_clock_tsc(struct clocksource * arg)224 static u64 notrace read_hv_clock_tsc(struct clocksource *arg)
225 {
226 u64 current_tick = hv_read_tsc_page(&tsc_pg);
227
228 if (current_tick == U64_MAX)
229 hv_get_time_ref_count(current_tick);
230
231 return current_tick;
232 }
233
read_hv_sched_clock_tsc(void)234 static u64 read_hv_sched_clock_tsc(void)
235 {
236 return read_hv_clock_tsc(NULL) - hv_sched_clock_offset;
237 }
238
239 static struct clocksource hyperv_cs_tsc = {
240 .name = "hyperv_clocksource_tsc_page",
241 .rating = 400,
242 .read = read_hv_clock_tsc,
243 .mask = CLOCKSOURCE_MASK(64),
244 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
245 };
246
read_hv_clock_msr(struct clocksource * arg)247 static u64 notrace read_hv_clock_msr(struct clocksource *arg)
248 {
249 u64 current_tick;
250 /*
251 * Read the partition counter to get the current tick count. This count
252 * is set to 0 when the partition is created and is incremented in
253 * 100 nanosecond units.
254 */
255 hv_get_time_ref_count(current_tick);
256 return current_tick;
257 }
258
read_hv_sched_clock_msr(void)259 static u64 read_hv_sched_clock_msr(void)
260 {
261 return read_hv_clock_msr(NULL) - hv_sched_clock_offset;
262 }
263
264 static struct clocksource hyperv_cs_msr = {
265 .name = "hyperv_clocksource_msr",
266 .rating = 400,
267 .read = read_hv_clock_msr,
268 .mask = CLOCKSOURCE_MASK(64),
269 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
270 };
271
hv_init_tsc_clocksource(void)272 static bool __init hv_init_tsc_clocksource(void)
273 {
274 u64 tsc_msr;
275 phys_addr_t phys_addr;
276
277 if (!(ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE))
278 return false;
279
280 hyperv_cs = &hyperv_cs_tsc;
281 phys_addr = virt_to_phys(&tsc_pg);
282
283 /*
284 * The Hyper-V TLFS specifies to preserve the value of reserved
285 * bits in registers. So read the existing value, preserve the
286 * low order 12 bits, and add in the guest physical address
287 * (which already has at least the low 12 bits set to zero since
288 * it is page aligned). Also set the "enable" bit, which is bit 0.
289 */
290 hv_get_reference_tsc(tsc_msr);
291 tsc_msr &= GENMASK_ULL(11, 0);
292 tsc_msr = tsc_msr | 0x1 | (u64)phys_addr;
293 hv_set_reference_tsc(tsc_msr);
294
295 hv_set_clocksource_vdso(hyperv_cs_tsc);
296 clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
297
298 hv_sched_clock_offset = hyperv_cs->read(hyperv_cs);
299 hv_setup_sched_clock(read_hv_sched_clock_tsc);
300
301 return true;
302 }
303
hv_init_clocksource(void)304 void __init hv_init_clocksource(void)
305 {
306 /*
307 * Try to set up the TSC page clocksource. If it succeeds, we're
308 * done. Otherwise, set up the MSR clocksoruce. At least one of
309 * these will always be available except on very old versions of
310 * Hyper-V on x86. In that case we won't have a Hyper-V
311 * clocksource, but Linux will still run with a clocksource based
312 * on the emulated PIT or LAPIC timer.
313 */
314 if (hv_init_tsc_clocksource())
315 return;
316
317 if (!(ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE))
318 return;
319
320 hyperv_cs = &hyperv_cs_msr;
321 clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
322
323 hv_sched_clock_offset = hyperv_cs->read(hyperv_cs);
324 hv_setup_sched_clock(read_hv_sched_clock_msr);
325 }
326 EXPORT_SYMBOL_GPL(hv_init_clocksource);
327