drivers/clocksource/hyperv_timer.c

   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
  26 /*
  27  * If false, we're using the old mechanism for stimer0 interrupts
  28  * where it sends a VMbus message when it expires. The old
  29  * mechanism is used when running on older versions of Hyper-V
  30  * that don't support Direct Mode. While Hyper-V provides
  31  * four stimer's per CPU, Linux uses only stimer0.
  32  */
  33 static bool direct_mode_enabled;
  34
  35 static int stimer0_irq;
  36 static int stimer0_vector;
  37 static int stimer0_message_sint;
  38
  39 /*
  40  * ISR for when stimer0 is operating in Direct Mode.  Direct Mode
  41  * does not use VMbus or any VMbus messages, so process here and not
  42  * in the VMbus driver code.
  43  */
  44 void hv_stimer0_isr(void)
  45 {
  46         struct clock_event_device *ce;
  47
  48         ce = this_cpu_ptr(hv_clock_event);
  49         ce->event_handler(ce);
  50 }
  51 EXPORT_SYMBOL_GPL(hv_stimer0_isr);
  52
  53 static int hv_ce_set_next_event(unsigned long delta,
  54                                 struct clock_event_device *evt)
  55 {
  56         u64 current_tick;
  57
  58         current_tick = hyperv_cs->read(NULL);
  59         current_tick += delta;
  60         hv_init_timer(0, current_tick);
  61         return 0;
  62 }
  63
  64 static int hv_ce_shutdown(struct clock_event_device *evt)
  65 {
  66         hv_init_timer(0, 0);
  67         hv_init_timer_config(0, 0);
  68         if (direct_mode_enabled)
  69                 hv_disable_stimer0_percpu_irq(stimer0_irq);
  70
  71         return 0;
  72 }
  73
  74 static int hv_ce_set_oneshot(struct clock_event_device *evt)
  75 {
  76         union hv_stimer_config timer_cfg;
  77
  78         timer_cfg.as_uint64 = 0;
  79         timer_cfg.enable = 1;
  80         timer_cfg.auto_enable = 1;
  81         if (direct_mode_enabled) {
  82                 /*
  83                  * When it expires, the timer will directly interrupt
  84                  * on the specified hardware vector/IRQ.
  85                  */
  86                 timer_cfg.direct_mode = 1;
  87                 timer_cfg.apic_vector = stimer0_vector;
  88                 hv_enable_stimer0_percpu_irq(stimer0_irq);
  89         } else {
  90                 /*
  91                  * When it expires, the timer will generate a VMbus message,
  92                  * to be handled by the normal VMbus interrupt handler.
  93                  */
  94                 timer_cfg.direct_mode = 0;
  95                 timer_cfg.sintx = stimer0_message_sint;
  96         }
  97         hv_init_timer_config(0, timer_cfg.as_uint64);
  98         return 0;
  99 }
 100
 101 /*
 102  * hv_stimer_init - Per-cpu initialization of the clockevent
 103  */
 104 void hv_stimer_init(unsigned int cpu)
 105 {
 106         struct clock_event_device *ce;
 107
 108         /*
 109          * Synthetic timers are always available except on old versions of
 110          * Hyper-V on x86.  In that case, just return as Linux will use a
 111          * clocksource based on emulated PIT or LAPIC timer hardware.
 112          */
 113         if (!(ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE))
 114                 return;
 115
 116         ce = per_cpu_ptr(hv_clock_event, cpu);
 117         ce->name = "Hyper-V clockevent";
 118         ce->features = CLOCK_EVT_FEAT_ONESHOT;
 119         ce->cpumask = cpumask_of(cpu);
 120         ce->rating = 1000;
 121         ce->set_state_shutdown = hv_ce_shutdown;
 122         ce->set_state_oneshot = hv_ce_set_oneshot;
 123         ce->set_next_event = hv_ce_set_next_event;
 124
 125         clockevents_config_and_register(ce,
 126                                         HV_CLOCK_HZ,
 127                                         HV_MIN_DELTA_TICKS,
 128                                         HV_MAX_MAX_DELTA_TICKS);
 129 }
 130 EXPORT_SYMBOL_GPL(hv_stimer_init);
 131
 132 /*
 133  * hv_stimer_cleanup - Per-cpu cleanup of the clockevent
 134  */
 135 void hv_stimer_cleanup(unsigned int cpu)
 136 {
 137         struct clock_event_device *ce;
 138
 139         /* Turn off clockevent device */
 140         if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE) {
 141                 ce = per_cpu_ptr(hv_clock_event, cpu);
 142                 hv_ce_shutdown(ce);
 143         }
 144 }
 145 EXPORT_SYMBOL_GPL(hv_stimer_cleanup);
 146
 147 /* hv_stimer_alloc - Global initialization of the clockevent and stimer0 */
 148 int hv_stimer_alloc(int sint)
 149 {
 150         int ret;
 151
 152         hv_clock_event = alloc_percpu(struct clock_event_device);
 153         if (!hv_clock_event)
 154                 return -ENOMEM;
 155
 156         direct_mode_enabled = ms_hyperv.misc_features &
 157                         HV_STIMER_DIRECT_MODE_AVAILABLE;
 158         if (direct_mode_enabled) {
 159                 ret = hv_setup_stimer0_irq(&stimer0_irq, &stimer0_vector,
 160                                 hv_stimer0_isr);
 161                 if (ret) {
 162                         free_percpu(hv_clock_event);
 163                         hv_clock_event = NULL;
 164                         return ret;
 165                 }
 166         }
 167
 168         stimer0_message_sint = sint;
 169         return 0;
 170 }
 171 EXPORT_SYMBOL_GPL(hv_stimer_alloc);
 172
 173 /* hv_stimer_free - Free global resources allocated by hv_stimer_alloc() */
 174 void hv_stimer_free(void)
 175 {
 176         if (direct_mode_enabled && (stimer0_irq != 0)) {
 177                 hv_remove_stimer0_irq(stimer0_irq);
 178                 stimer0_irq = 0;
 179         }
 180         free_percpu(hv_clock_event);
 181         hv_clock_event = NULL;
 182 }
 183 EXPORT_SYMBOL_GPL(hv_stimer_free);
 184
 185 /*
 186  * Do a global cleanup of clockevents for the cases of kexec and
 187  * vmbus exit
 188  */
 189 void hv_stimer_global_cleanup(void)
 190 {
 191         int     cpu;
 192         struct clock_event_device *ce;
 193
 194         if (ms_hyperv.features & HV_MSR_SYNTIMER_AVAILABLE) {
 195                 for_each_present_cpu(cpu) {
 196                         ce = per_cpu_ptr(hv_clock_event, cpu);
 197                         clockevents_unbind_device(ce, cpu);
 198                 }
 199         }
 200         hv_stimer_free();
 201 }
 202 EXPORT_SYMBOL_GPL(hv_stimer_global_cleanup);
 203
 204 /*
 205  * Code and definitions for the Hyper-V clocksources.  Two
 206  * clocksources are defined: one that reads the Hyper-V defined MSR, and
 207  * the other that uses the TSC reference page feature as defined in the
 208  * TLFS.  The MSR version is for compatibility with old versions of
 209  * Hyper-V and 32-bit x86.  The TSC reference page version is preferred.
 210  */
 211
 212 struct clocksource *hyperv_cs;
 213 EXPORT_SYMBOL_GPL(hyperv_cs);
 214
 215 #ifdef CONFIG_HYPERV_TSCPAGE
 216
 217 static struct ms_hyperv_tsc_page *tsc_pg;
 218
 219 struct ms_hyperv_tsc_page *hv_get_tsc_page(void)
 220 {
 221         return tsc_pg;
 222 }
 223 EXPORT_SYMBOL_GPL(hv_get_tsc_page);
 224
 225 static u64 notrace read_hv_sched_clock_tsc(void)
 226 {
 227         u64 current_tick = hv_read_tsc_page(tsc_pg);
 228
 229         if (current_tick == U64_MAX)
 230                 hv_get_time_ref_count(current_tick);
 231
 232         return current_tick;
 233 }
 234
 235 static u64 read_hv_clock_tsc(struct clocksource *arg)
 236 {
 237         return read_hv_sched_clock_tsc();
 238 }
 239
 240 static struct clocksource hyperv_cs_tsc = {
 241         .name   = "hyperv_clocksource_tsc_page",
 242         .rating = 400,
 243         .read   = read_hv_clock_tsc,
 244         .mask   = CLOCKSOURCE_MASK(64),
 245         .flags  = CLOCK_SOURCE_IS_CONTINUOUS,
 246 };
 247 #endif
 248
 249 static u64 notrace read_hv_sched_clock_msr(void)
 250 {
 251         u64 current_tick;
 252         /*
 253          * Read the partition counter to get the current tick count. This count
 254          * is set to 0 when the partition is created and is incremented in
 255          * 100 nanosecond units.
 256          */
 257         hv_get_time_ref_count(current_tick);
 258         return current_tick;
 259 }
 260
 261 static u64 read_hv_clock_msr(struct clocksource *arg)
 262 {
 263         return read_hv_sched_clock_msr();
 264 }
 265
 266 static struct clocksource hyperv_cs_msr = {
 267         .name   = "hyperv_clocksource_msr",
 268         .rating = 400,
 269         .read   = read_hv_clock_msr,
 270         .mask   = CLOCKSOURCE_MASK(64),
 271         .flags  = CLOCK_SOURCE_IS_CONTINUOUS,
 272 };
 273
 274 #ifdef CONFIG_HYPERV_TSCPAGE
 275 static bool __init hv_init_tsc_clocksource(void)
 276 {
 277         u64             tsc_msr;
 278         phys_addr_t     phys_addr;
 279
 280         if (!(ms_hyperv.features & HV_MSR_REFERENCE_TSC_AVAILABLE))
 281                 return false;
 282
 283         tsc_pg = vmalloc(PAGE_SIZE);
 284         if (!tsc_pg)
 285                 return false;
 286
 287         hyperv_cs = &hyperv_cs_tsc;
 288         phys_addr = page_to_phys(vmalloc_to_page(tsc_pg));
 289
 290         /*
 291          * The Hyper-V TLFS specifies to preserve the value of reserved
 292          * bits in registers. So read the existing value, preserve the
 293          * low order 12 bits, and add in the guest physical address
 294          * (which already has at least the low 12 bits set to zero since
 295          * it is page aligned). Also set the "enable" bit, which is bit 0.
 296          */
 297         hv_get_reference_tsc(tsc_msr);
 298         tsc_msr &= GENMASK_ULL(11, 0);
 299         tsc_msr = tsc_msr | 0x1 | (u64)phys_addr;
 300         hv_set_reference_tsc(tsc_msr);
 301
 302         hv_set_clocksource_vdso(hyperv_cs_tsc);
 303         clocksource_register_hz(&hyperv_cs_tsc, NSEC_PER_SEC/100);
 304
 305         /* sched_clock_register is needed on ARM64 but is a no-op on x86 */
 306         sched_clock_register(read_hv_sched_clock_tsc, 64, HV_CLOCK_HZ);
 307         return true;
 308 }
 309 #else
 310 static bool __init hv_init_tsc_clocksource(void)
 311 {
 312         return false;
 313 }
 314 #endif
 315
 316
 317 void __init hv_init_clocksource(void)
 318 {
 319         /*
 320          * Try to set up the TSC page clocksource. If it succeeds, we're
 321          * done. Otherwise, set up the MSR clocksoruce.  At least one of
 322          * these will always be available except on very old versions of
 323          * Hyper-V on x86.  In that case we won't have a Hyper-V
 324          * clocksource, but Linux will still run with a clocksource based
 325          * on the emulated PIT or LAPIC timer.
 326          */
 327         if (hv_init_tsc_clocksource())
 328                 return;
 329
 330         if (!(ms_hyperv.features & HV_MSR_TIME_REF_COUNT_AVAILABLE))
 331                 return;
 332
 333         hyperv_cs = &hyperv_cs_msr;
 334         clocksource_register_hz(&hyperv_cs_msr, NSEC_PER_SEC/100);
 335
 336         /* sched_clock_register is needed on ARM64 but is a no-op on x86 */
 337         sched_clock_register(read_hv_sched_clock_msr, 64, HV_CLOCK_HZ);
 338 }
 339 EXPORT_SYMBOL_GPL(hv_init_clocksource);