arch/x86/kernel/vmiclock_32.c

   1 /*
   2  * VMI paravirtual timer support routines.
   3  *
   4  * Copyright (C) 2007, VMware, Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  14  * NON INFRINGEMENT.  See the GNU General Public License for more
  15  * details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, write to the Free Software
  19  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  *
  21  */
  22
  23 #include <linux/smp.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/cpumask.h>
  26 #include <linux/clocksource.h>
  27 #include <linux/clockchips.h>
  28
  29 #include <asm/vmi.h>
  30 #include <asm/vmi_time.h>
  31 #include <asm/arch_hooks.h>
  32 #include <asm/apicdef.h>
  33 #include <asm/apic.h>
  34 #include <asm/timer.h>
  35 #include <asm/i8253.h>
  36 #include <asm/irq_vectors.h>
  37
  38 #define VMI_ONESHOT  (VMI_ALARM_IS_ONESHOT  | VMI_CYCLES_REAL | vmi_get_alarm_wiring())
  39 #define VMI_PERIODIC (VMI_ALARM_IS_PERIODIC | VMI_CYCLES_REAL | vmi_get_alarm_wiring())
  40
  41 static DEFINE_PER_CPU(struct clock_event_device, local_events);
  42
  43 static inline u32 vmi_counter(u32 flags)
  44 {
  45         /* Given VMI_ONESHOT or VMI_PERIODIC, return the corresponding
  46          * cycle counter. */
  47         return flags & VMI_ALARM_COUNTER_MASK;
  48 }
  49
  50 /* paravirt_ops.get_wallclock = vmi_get_wallclock */
  51 unsigned long vmi_get_wallclock(void)
  52 {
  53         unsigned long long wallclock;
  54         wallclock = vmi_timer_ops.get_wallclock(); // nsec
  55         (void)do_div(wallclock, 1000000000);       // sec
  56
  57         return wallclock;
  58 }
  59
  60 /* paravirt_ops.set_wallclock = vmi_set_wallclock */
  61 int vmi_set_wallclock(unsigned long now)
  62 {
  63         return 0;
  64 }
  65
  66 /* paravirt_ops.sched_clock = vmi_sched_clock */
  67 unsigned long long vmi_sched_clock(void)
  68 {
  69         return cycles_2_ns(vmi_timer_ops.get_cycle_counter(VMI_CYCLES_AVAILABLE));
  70 }
  71
  72 /* paravirt_ops.get_tsc_khz = vmi_tsc_khz */
  73 unsigned long vmi_tsc_khz(void)
  74 {
  75         unsigned long long khz;
  76         khz = vmi_timer_ops.get_cycle_frequency();
  77         (void)do_div(khz, 1000);
  78         return khz;
  79 }
  80
  81 static inline unsigned int vmi_get_timer_vector(void)
  82 {
  83 #ifdef CONFIG_X86_IO_APIC
  84         return FIRST_DEVICE_VECTOR;
  85 #else
  86         return FIRST_EXTERNAL_VECTOR;
  87 #endif
  88 }
  89
  90 /** vmi clockchip */
  91 #ifdef CONFIG_X86_LOCAL_APIC
  92 static unsigned int startup_timer_irq(unsigned int irq)
  93 {
  94         unsigned long val = apic_read(APIC_LVTT);
  95         apic_write(APIC_LVTT, vmi_get_timer_vector());
  96
  97         return (val & APIC_SEND_PENDING);
  98 }
  99
 100 static void mask_timer_irq(unsigned int irq)
 101 {
 102         unsigned long val = apic_read(APIC_LVTT);
 103         apic_write(APIC_LVTT, val | APIC_LVT_MASKED);
 104 }
 105
 106 static void unmask_timer_irq(unsigned int irq)
 107 {
 108         unsigned long val = apic_read(APIC_LVTT);
 109         apic_write(APIC_LVTT, val & ~APIC_LVT_MASKED);
 110 }
 111
 112 static void ack_timer_irq(unsigned int irq)
 113 {
 114         ack_APIC_irq();
 115 }
 116
 117 static struct irq_chip vmi_chip __read_mostly = {
 118         .name           = "VMI-LOCAL",
 119         .startup        = startup_timer_irq,
 120         .mask           = mask_timer_irq,
 121         .unmask         = unmask_timer_irq,
 122         .ack            = ack_timer_irq
 123 };
 124 #endif
 125
 126 /** vmi clockevent */
 127 #define VMI_ALARM_WIRED_IRQ0    0x00000000
 128 #define VMI_ALARM_WIRED_LVTT    0x00010000
 129 static int vmi_wiring = VMI_ALARM_WIRED_IRQ0;
 130
 131 static inline int vmi_get_alarm_wiring(void)
 132 {
 133         return vmi_wiring;
 134 }
 135
 136 static void vmi_timer_set_mode(enum clock_event_mode mode,
 137                                struct clock_event_device *evt)
 138 {
 139         cycle_t now, cycles_per_hz;
 140         BUG_ON(!irqs_disabled());
 141
 142         switch (mode) {
 143         case CLOCK_EVT_MODE_ONESHOT:
 144         case CLOCK_EVT_MODE_RESUME:
 145                 break;
 146         case CLOCK_EVT_MODE_PERIODIC:
 147                 cycles_per_hz = vmi_timer_ops.get_cycle_frequency();
 148                 (void)do_div(cycles_per_hz, HZ);
 149                 now = vmi_timer_ops.get_cycle_counter(vmi_counter(VMI_PERIODIC));
 150                 vmi_timer_ops.set_alarm(VMI_PERIODIC, now, cycles_per_hz);
 151                 break;
 152         case CLOCK_EVT_MODE_UNUSED:
 153         case CLOCK_EVT_MODE_SHUTDOWN:
 154                 switch (evt->mode) {
 155                 case CLOCK_EVT_MODE_ONESHOT:
 156                         vmi_timer_ops.cancel_alarm(VMI_ONESHOT);
 157                         break;
 158                 case CLOCK_EVT_MODE_PERIODIC:
 159                         vmi_timer_ops.cancel_alarm(VMI_PERIODIC);
 160                         break;
 161                 default:
 162                         break;
 163                 }
 164                 break;
 165         default:
 166                 break;
 167         }
 168 }
 169
 170 static int vmi_timer_next_event(unsigned long delta,
 171                                 struct clock_event_device *evt)
 172 {
 173         /* Unfortunately, set_next_event interface only passes relative
 174          * expiry, but we want absolute expiry.  It'd be better if were
 175          * were passed an aboslute expiry, since a bunch of time may
 176          * have been stolen between the time the delta is computed and
 177          * when we set the alarm below. */
 178         cycle_t now = vmi_timer_ops.get_cycle_counter(vmi_counter(VMI_ONESHOT));
 179
 180         BUG_ON(evt->mode != CLOCK_EVT_MODE_ONESHOT);
 181         vmi_timer_ops.set_alarm(VMI_ONESHOT, now + delta, 0);
 182         return 0;
 183 }
 184
 185 static struct clock_event_device vmi_clockevent = {
 186         .name           = "vmi-timer",
 187         .features       = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
 188         .shift          = 22,
 189         .set_mode       = vmi_timer_set_mode,
 190         .set_next_event = vmi_timer_next_event,
 191         .rating         = 1000,
 192         .irq            = 0,
 193 };
 194
 195 static irqreturn_t vmi_timer_interrupt(int irq, void *dev_id)
 196 {
 197         struct clock_event_device *evt = &__get_cpu_var(local_events);
 198         evt->event_handler(evt);
 199         return IRQ_HANDLED;
 200 }
 201
 202 static struct irqaction vmi_clock_action  = {
 203         .name           = "vmi-timer",
 204         .handler        = vmi_timer_interrupt,
 205         .flags          = IRQF_DISABLED | IRQF_NOBALANCING,
 206         .mask           = CPU_MASK_ALL,
 207 };
 208
 209 static void __devinit vmi_time_init_clockevent(void)
 210 {
 211         cycle_t cycles_per_msec;
 212         struct clock_event_device *evt;
 213
 214         int cpu = smp_processor_id();
 215         evt = &__get_cpu_var(local_events);
 216
 217         /* Use cycles_per_msec since div_sc params are 32-bits. */
 218         cycles_per_msec = vmi_timer_ops.get_cycle_frequency();
 219         (void)do_div(cycles_per_msec, 1000);
 220
 221         memcpy(evt, &vmi_clockevent, sizeof(*evt));
 222         /* Must pick .shift such that .mult fits in 32-bits.  Choosing
 223          * .shift to be 22 allows 2^(32-22) cycles per nano-seconds
 224          * before overflow. */
 225         evt->mult = div_sc(cycles_per_msec, NSEC_PER_MSEC, evt->shift);
 226         /* Upper bound is clockevent's use of ulong for cycle deltas. */
 227         evt->max_delta_ns = clockevent_delta2ns(ULONG_MAX, evt);
 228         evt->min_delta_ns = clockevent_delta2ns(1, evt);
 229         evt->cpumask = cpumask_of_cpu(cpu);
 230
 231         printk(KERN_WARNING "vmi: registering clock event %s. mult=%lu shift=%u\n",
 232                evt->name, evt->mult, evt->shift);
 233         clockevents_register_device(evt);
 234 }
 235
 236 void __init vmi_time_init(void)
 237 {
 238         unsigned int cpu;
 239         /* Disable PIT: BIOSes start PIT CH0 with 18.2hz peridic. */
 240         outb_pit(0x3a, PIT_MODE); /* binary, mode 5, LSB/MSB, ch 0 */
 241
 242         vmi_time_init_clockevent();
 243         setup_irq(0, &vmi_clock_action);
 244         for_each_possible_cpu(cpu)
 245                 per_cpu(vector_irq, cpu)[vmi_get_timer_vector()] = 0;
 246 }
 247
 248 #ifdef CONFIG_X86_LOCAL_APIC
 249 void __devinit vmi_time_bsp_init(void)
 250 {
 251         /*
 252          * On APIC systems, we want local timers to fire on each cpu.  We do
 253          * this by programming LVTT to deliver timer events to the IRQ handler
 254          * for IRQ-0, since we can't re-use the APIC local timer handler
 255          * without interfering with that code.
 256          */
 257         clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
 258         local_irq_disable();
 259 #ifdef CONFIG_X86_SMP
 260         /*
 261          * XXX handle_percpu_irq only defined for SMP; we need to switch over
 262          * to using it, since this is a local interrupt, which each CPU must
 263          * handle individually without locking out or dropping simultaneous
 264          * local timers on other CPUs.  We also don't want to trigger the
 265          * quirk workaround code for interrupts which gets invoked from
 266          * handle_percpu_irq via eoi, so we use our own IRQ chip.
 267          */
 268         set_irq_chip_and_handler_name(0, &vmi_chip, handle_percpu_irq, "lvtt");
 269 #else
 270         set_irq_chip_and_handler_name(0, &vmi_chip, handle_edge_irq, "lvtt");
 271 #endif
 272         vmi_wiring = VMI_ALARM_WIRED_LVTT;
 273         apic_write(APIC_LVTT, vmi_get_timer_vector());
 274         local_irq_enable();
 275         clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
 276 }
 277
 278 void __devinit vmi_time_ap_init(void)
 279 {
 280         vmi_time_init_clockevent();
 281         apic_write(APIC_LVTT, vmi_get_timer_vector());
 282 }
 283 #endif
 284
 285 /** vmi clocksource */
 286
 287 static cycle_t read_real_cycles(void)
 288 {
 289         return vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
 290 }
 291
 292 static struct clocksource clocksource_vmi = {
 293         .name                   = "vmi-timer",
 294         .rating                 = 450,
 295         .read                   = read_real_cycles,
 296         .mask                   = CLOCKSOURCE_MASK(64),
 297         .mult                   = 0, /* to be set */
 298         .shift                  = 22,
 299         .flags                  = CLOCK_SOURCE_IS_CONTINUOUS,
 300 };
 301
 302 static int __init init_vmi_clocksource(void)
 303 {
 304         cycle_t cycles_per_msec;
 305
 306         if (!vmi_timer_ops.get_cycle_frequency)
 307                 return 0;
 308         /* Use khz2mult rather than hz2mult since hz arg is only 32-bits. */
 309         cycles_per_msec = vmi_timer_ops.get_cycle_frequency();
 310         (void)do_div(cycles_per_msec, 1000);
 311
 312         /* Note that clocksource.{mult, shift} converts in the opposite direction
 313          * as clockevents.  */
 314         clocksource_vmi.mult = clocksource_khz2mult(cycles_per_msec,
 315                                                     clocksource_vmi.shift);
 316
 317         printk(KERN_WARNING "vmi: registering clock source khz=%lld\n", cycles_per_msec);
 318         return clocksource_register(&clocksource_vmi);
 319
 320 }
 321 module_init(init_vmi_clocksource);