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/apicdef.h>
  32 #include <asm/apic.h>
  33 #include <asm/timer.h>
  34 #include <asm/i8253.h>
  35 #include <asm/irq_vectors.h>
  36
  37 #define VMI_ONESHOT  (VMI_ALARM_IS_ONESHOT  | VMI_CYCLES_REAL | vmi_get_alarm_wiring())
  38 #define VMI_PERIODIC (VMI_ALARM_IS_PERIODIC | VMI_CYCLES_REAL | vmi_get_alarm_wiring())
  39
  40 static DEFINE_PER_CPU(struct clock_event_device, local_events);
  41
  42 static inline u32 vmi_counter(u32 flags)
  43 {
  44         /* Given VMI_ONESHOT or VMI_PERIODIC, return the corresponding
  45          * cycle counter. */
  46         return flags & VMI_ALARM_COUNTER_MASK;
  47 }
  48
  49 /* paravirt_ops.get_wallclock = vmi_get_wallclock */
  50 unsigned long vmi_get_wallclock(void)
  51 {
  52         unsigned long long wallclock;
  53         wallclock = vmi_timer_ops.get_wallclock(); // nsec
  54         (void)do_div(wallclock, 1000000000);       // sec
  55
  56         return wallclock;
  57 }
  58
  59 /* paravirt_ops.set_wallclock = vmi_set_wallclock */
  60 int vmi_set_wallclock(unsigned long now)
  61 {
  62         return 0;
  63 }
  64
  65 /* paravirt_ops.sched_clock = vmi_sched_clock */
  66 unsigned long long vmi_sched_clock(void)
  67 {
  68         return cycles_2_ns(vmi_timer_ops.get_cycle_counter(VMI_CYCLES_AVAILABLE));
  69 }
  70
  71 /* x86_platform.calibrate_tsc = vmi_tsc_khz */
  72 unsigned long vmi_tsc_khz(void)
  73 {
  74         unsigned long long khz;
  75         khz = vmi_timer_ops.get_cycle_frequency();
  76         (void)do_div(khz, 1000);
  77         return khz;
  78 }
  79
  80 static inline unsigned int vmi_get_timer_vector(void)
  81 {
  82         return IRQ0_VECTOR;
  83 }
  84
  85 /** vmi clockchip */
  86 #ifdef CONFIG_X86_LOCAL_APIC
  87 static unsigned int startup_timer_irq(unsigned int irq)
  88 {
  89         unsigned long val = apic_read(APIC_LVTT);
  90         apic_write(APIC_LVTT, vmi_get_timer_vector());
  91
  92         return (val & APIC_SEND_PENDING);
  93 }
  94
  95 static void mask_timer_irq(unsigned int irq)
  96 {
  97         unsigned long val = apic_read(APIC_LVTT);
  98         apic_write(APIC_LVTT, val | APIC_LVT_MASKED);
  99 }
 100
 101 static void unmask_timer_irq(unsigned int irq)
 102 {
 103         unsigned long val = apic_read(APIC_LVTT);
 104         apic_write(APIC_LVTT, val & ~APIC_LVT_MASKED);
 105 }
 106
 107 static void ack_timer_irq(unsigned int irq)
 108 {
 109         ack_APIC_irq();
 110 }
 111
 112 static struct irq_chip vmi_chip __read_mostly = {
 113         .name           = "VMI-LOCAL",
 114         .startup        = startup_timer_irq,
 115         .mask           = mask_timer_irq,
 116         .unmask         = unmask_timer_irq,
 117         .ack            = ack_timer_irq
 118 };
 119 #endif
 120
 121 /** vmi clockevent */
 122 #define VMI_ALARM_WIRED_IRQ0    0x00000000
 123 #define VMI_ALARM_WIRED_LVTT    0x00010000
 124 static int vmi_wiring = VMI_ALARM_WIRED_IRQ0;
 125
 126 static inline int vmi_get_alarm_wiring(void)
 127 {
 128         return vmi_wiring;
 129 }
 130
 131 static void vmi_timer_set_mode(enum clock_event_mode mode,
 132                                struct clock_event_device *evt)
 133 {
 134         cycle_t now, cycles_per_hz;
 135         BUG_ON(!irqs_disabled());
 136
 137         switch (mode) {
 138         case CLOCK_EVT_MODE_ONESHOT:
 139         case CLOCK_EVT_MODE_RESUME:
 140                 break;
 141         case CLOCK_EVT_MODE_PERIODIC:
 142                 cycles_per_hz = vmi_timer_ops.get_cycle_frequency();
 143                 (void)do_div(cycles_per_hz, HZ);
 144                 now = vmi_timer_ops.get_cycle_counter(vmi_counter(VMI_PERIODIC));
 145                 vmi_timer_ops.set_alarm(VMI_PERIODIC, now, cycles_per_hz);
 146                 break;
 147         case CLOCK_EVT_MODE_UNUSED:
 148         case CLOCK_EVT_MODE_SHUTDOWN:
 149                 switch (evt->mode) {
 150                 case CLOCK_EVT_MODE_ONESHOT:
 151                         vmi_timer_ops.cancel_alarm(VMI_ONESHOT);
 152                         break;
 153                 case CLOCK_EVT_MODE_PERIODIC:
 154                         vmi_timer_ops.cancel_alarm(VMI_PERIODIC);
 155                         break;
 156                 default:
 157                         break;
 158                 }
 159                 break;
 160         default:
 161                 break;
 162         }
 163 }
 164
 165 static int vmi_timer_next_event(unsigned long delta,
 166                                 struct clock_event_device *evt)
 167 {
 168         /* Unfortunately, set_next_event interface only passes relative
 169          * expiry, but we want absolute expiry.  It'd be better if were
 170          * were passed an absolute expiry, since a bunch of time may
 171          * have been stolen between the time the delta is computed and
 172          * when we set the alarm below. */
 173         cycle_t now = vmi_timer_ops.get_cycle_counter(vmi_counter(VMI_ONESHOT));
 174
 175         BUG_ON(evt->mode != CLOCK_EVT_MODE_ONESHOT);
 176         vmi_timer_ops.set_alarm(VMI_ONESHOT, now + delta, 0);
 177         return 0;
 178 }
 179
 180 static struct clock_event_device vmi_clockevent = {
 181         .name           = "vmi-timer",
 182         .features       = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT,
 183         .shift          = 22,
 184         .set_mode       = vmi_timer_set_mode,
 185         .set_next_event = vmi_timer_next_event,
 186         .rating         = 1000,
 187         .irq            = 0,
 188 };
 189
 190 static irqreturn_t vmi_timer_interrupt(int irq, void *dev_id)
 191 {
 192         struct clock_event_device *evt = &__get_cpu_var(local_events);
 193         evt->event_handler(evt);
 194         return IRQ_HANDLED;
 195 }
 196
 197 static struct irqaction vmi_clock_action  = {
 198         .name           = "vmi-timer",
 199         .handler        = vmi_timer_interrupt,
 200         .flags          = IRQF_DISABLED | IRQF_NOBALANCING | IRQF_TIMER,
 201 };
 202
 203 static void __devinit vmi_time_init_clockevent(void)
 204 {
 205         cycle_t cycles_per_msec;
 206         struct clock_event_device *evt;
 207
 208         int cpu = smp_processor_id();
 209         evt = &__get_cpu_var(local_events);
 210
 211         /* Use cycles_per_msec since div_sc params are 32-bits. */
 212         cycles_per_msec = vmi_timer_ops.get_cycle_frequency();
 213         (void)do_div(cycles_per_msec, 1000);
 214
 215         memcpy(evt, &vmi_clockevent, sizeof(*evt));
 216         /* Must pick .shift such that .mult fits in 32-bits.  Choosing
 217          * .shift to be 22 allows 2^(32-22) cycles per nano-seconds
 218          * before overflow. */
 219         evt->mult = div_sc(cycles_per_msec, NSEC_PER_MSEC, evt->shift);
 220         /* Upper bound is clockevent's use of ulong for cycle deltas. */
 221         evt->max_delta_ns = clockevent_delta2ns(ULONG_MAX, evt);
 222         evt->min_delta_ns = clockevent_delta2ns(1, evt);
 223         evt->cpumask = cpumask_of(cpu);
 224
 225         printk(KERN_WARNING "vmi: registering clock event %s. mult=%u shift=%u\n",
 226                evt->name, evt->mult, evt->shift);
 227         clockevents_register_device(evt);
 228 }
 229
 230 void __init vmi_time_init(void)
 231 {
 232         unsigned int cpu;
 233         /* Disable PIT: BIOSes start PIT CH0 with 18.2hz peridic. */
 234         outb_pit(0x3a, PIT_MODE); /* binary, mode 5, LSB/MSB, ch 0 */
 235
 236         vmi_time_init_clockevent();
 237         setup_irq(0, &vmi_clock_action);
 238         for_each_possible_cpu(cpu)
 239                 per_cpu(vector_irq, cpu)[vmi_get_timer_vector()] = 0;
 240 }
 241
 242 #ifdef CONFIG_X86_LOCAL_APIC
 243 void __devinit vmi_time_bsp_init(void)
 244 {
 245         /*
 246          * On APIC systems, we want local timers to fire on each cpu.  We do
 247          * this by programming LVTT to deliver timer events to the IRQ handler
 248          * for IRQ-0, since we can't re-use the APIC local timer handler
 249          * without interfering with that code.
 250          */
 251         clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
 252         local_irq_disable();
 253 #ifdef CONFIG_SMP
 254         /*
 255          * XXX handle_percpu_irq only defined for SMP; we need to switch over
 256          * to using it, since this is a local interrupt, which each CPU must
 257          * handle individually without locking out or dropping simultaneous
 258          * local timers on other CPUs.  We also don't want to trigger the
 259          * quirk workaround code for interrupts which gets invoked from
 260          * handle_percpu_irq via eoi, so we use our own IRQ chip.
 261          */
 262         set_irq_chip_and_handler_name(0, &vmi_chip, handle_percpu_irq, "lvtt");
 263 #else
 264         set_irq_chip_and_handler_name(0, &vmi_chip, handle_edge_irq, "lvtt");
 265 #endif
 266         vmi_wiring = VMI_ALARM_WIRED_LVTT;
 267         apic_write(APIC_LVTT, vmi_get_timer_vector());
 268         local_irq_enable();
 269         clockevents_notify(CLOCK_EVT_NOTIFY_RESUME, NULL);
 270 }
 271
 272 void __devinit vmi_time_ap_init(void)
 273 {
 274         vmi_time_init_clockevent();
 275         apic_write(APIC_LVTT, vmi_get_timer_vector());
 276 }
 277 #endif
 278
 279 /** vmi clocksource */
 280 static struct clocksource clocksource_vmi;
 281
 282 static cycle_t read_real_cycles(struct clocksource *cs)
 283 {
 284         cycle_t ret = (cycle_t)vmi_timer_ops.get_cycle_counter(VMI_CYCLES_REAL);
 285         return max(ret, clocksource_vmi.cycle_last);
 286 }
 287
 288 static struct clocksource clocksource_vmi = {
 289         .name                   = "vmi-timer",
 290         .rating                 = 450,
 291         .read                   = read_real_cycles,
 292         .mask                   = CLOCKSOURCE_MASK(64),
 293         .mult                   = 0, /* to be set */
 294         .shift                  = 22,
 295         .flags                  = CLOCK_SOURCE_IS_CONTINUOUS,
 296 };
 297
 298 static int __init init_vmi_clocksource(void)
 299 {
 300         cycle_t cycles_per_msec;
 301
 302         if (!vmi_timer_ops.get_cycle_frequency)
 303                 return 0;
 304         /* Use khz2mult rather than hz2mult since hz arg is only 32-bits. */
 305         cycles_per_msec = vmi_timer_ops.get_cycle_frequency();
 306         (void)do_div(cycles_per_msec, 1000);
 307
 308         /* Note that clocksource.{mult, shift} converts in the opposite direction
 309          * as clockevents.  */
 310         clocksource_vmi.mult = clocksource_khz2mult(cycles_per_msec,
 311                                                     clocksource_vmi.shift);
 312
 313         printk(KERN_WARNING "vmi: registering clock source khz=%lld\n", cycles_per_msec);
 314         return clocksource_register(&clocksource_vmi);
 315
 316 }
 317 module_init(init_vmi_clocksource);