Kyle Swenson | 8d8f654 | 2021-03-15 11:02:55 -0600 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright 2010 Tilera Corporation. All Rights Reserved. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU General Public License |
| 6 | * as published by the Free Software Foundation, version 2. |
| 7 | * |
| 8 | * This program is distributed in the hope that it will be useful, but |
| 9 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or |
| 11 | * NON INFRINGEMENT. See the GNU General Public License for |
| 12 | * more details. |
| 13 | * |
| 14 | * Support the cycle counter clocksource and tile timer clock event device. |
| 15 | */ |
| 16 | |
| 17 | #include <linux/time.h> |
| 18 | #include <linux/timex.h> |
| 19 | #include <linux/clocksource.h> |
| 20 | #include <linux/clockchips.h> |
| 21 | #include <linux/hardirq.h> |
| 22 | #include <linux/sched.h> |
| 23 | #include <linux/smp.h> |
| 24 | #include <linux/delay.h> |
| 25 | #include <linux/module.h> |
| 26 | #include <linux/timekeeper_internal.h> |
| 27 | #include <asm/irq_regs.h> |
| 28 | #include <asm/traps.h> |
| 29 | #include <asm/vdso.h> |
| 30 | #include <hv/hypervisor.h> |
| 31 | #include <arch/interrupts.h> |
| 32 | #include <arch/spr_def.h> |
| 33 | |
| 34 | |
| 35 | /* |
| 36 | * Define the cycle counter clock source. |
| 37 | */ |
| 38 | |
| 39 | /* How many cycles per second we are running at. */ |
| 40 | static cycles_t cycles_per_sec __write_once; |
| 41 | |
| 42 | cycles_t get_clock_rate(void) |
| 43 | { |
| 44 | return cycles_per_sec; |
| 45 | } |
| 46 | |
| 47 | #if CHIP_HAS_SPLIT_CYCLE() |
| 48 | cycles_t get_cycles(void) |
| 49 | { |
| 50 | unsigned int high = __insn_mfspr(SPR_CYCLE_HIGH); |
| 51 | unsigned int low = __insn_mfspr(SPR_CYCLE_LOW); |
| 52 | unsigned int high2 = __insn_mfspr(SPR_CYCLE_HIGH); |
| 53 | |
| 54 | while (unlikely(high != high2)) { |
| 55 | low = __insn_mfspr(SPR_CYCLE_LOW); |
| 56 | high = high2; |
| 57 | high2 = __insn_mfspr(SPR_CYCLE_HIGH); |
| 58 | } |
| 59 | |
| 60 | return (((cycles_t)high) << 32) | low; |
| 61 | } |
| 62 | EXPORT_SYMBOL(get_cycles); |
| 63 | #endif |
| 64 | |
| 65 | /* |
| 66 | * We use a relatively small shift value so that sched_clock() |
| 67 | * won't wrap around very often. |
| 68 | */ |
| 69 | #define SCHED_CLOCK_SHIFT 10 |
| 70 | |
| 71 | static unsigned long sched_clock_mult __write_once; |
| 72 | |
| 73 | static cycles_t clocksource_get_cycles(struct clocksource *cs) |
| 74 | { |
| 75 | return get_cycles(); |
| 76 | } |
| 77 | |
| 78 | static struct clocksource cycle_counter_cs = { |
| 79 | .name = "cycle counter", |
| 80 | .rating = 300, |
| 81 | .read = clocksource_get_cycles, |
| 82 | .mask = CLOCKSOURCE_MASK(64), |
| 83 | .flags = CLOCK_SOURCE_IS_CONTINUOUS, |
| 84 | }; |
| 85 | |
| 86 | /* |
| 87 | * Called very early from setup_arch() to set cycles_per_sec. |
| 88 | * We initialize it early so we can use it to set up loops_per_jiffy. |
| 89 | */ |
| 90 | void __init setup_clock(void) |
| 91 | { |
| 92 | cycles_per_sec = hv_sysconf(HV_SYSCONF_CPU_SPEED); |
| 93 | sched_clock_mult = |
| 94 | clocksource_hz2mult(cycles_per_sec, SCHED_CLOCK_SHIFT); |
| 95 | } |
| 96 | |
| 97 | void __init calibrate_delay(void) |
| 98 | { |
| 99 | loops_per_jiffy = get_clock_rate() / HZ; |
| 100 | pr_info("Clock rate yields %lu.%02lu BogoMIPS (lpj=%lu)\n", |
| 101 | loops_per_jiffy / (500000 / HZ), |
| 102 | (loops_per_jiffy / (5000 / HZ)) % 100, loops_per_jiffy); |
| 103 | } |
| 104 | |
| 105 | /* Called fairly late in init/main.c, but before we go smp. */ |
| 106 | void __init time_init(void) |
| 107 | { |
| 108 | /* Initialize and register the clock source. */ |
| 109 | clocksource_register_hz(&cycle_counter_cs, cycles_per_sec); |
| 110 | |
| 111 | /* Start up the tile-timer interrupt source on the boot cpu. */ |
| 112 | setup_tile_timer(); |
| 113 | } |
| 114 | |
| 115 | /* |
| 116 | * Define the tile timer clock event device. The timer is driven by |
| 117 | * the TILE_TIMER_CONTROL register, which consists of a 31-bit down |
| 118 | * counter, plus bit 31, which signifies that the counter has wrapped |
| 119 | * from zero to (2**31) - 1. The INT_TILE_TIMER interrupt will be |
| 120 | * raised as long as bit 31 is set. |
| 121 | * |
| 122 | * The TILE_MINSEC value represents the largest range of real-time |
| 123 | * we can possibly cover with the timer, based on MAX_TICK combined |
| 124 | * with the slowest reasonable clock rate we might run at. |
| 125 | */ |
| 126 | |
| 127 | #define MAX_TICK 0x7fffffff /* we have 31 bits of countdown timer */ |
| 128 | #define TILE_MINSEC 5 /* timer covers no more than 5 seconds */ |
| 129 | |
| 130 | static int tile_timer_set_next_event(unsigned long ticks, |
| 131 | struct clock_event_device *evt) |
| 132 | { |
| 133 | BUG_ON(ticks > MAX_TICK); |
| 134 | __insn_mtspr(SPR_TILE_TIMER_CONTROL, ticks); |
| 135 | arch_local_irq_unmask_now(INT_TILE_TIMER); |
| 136 | return 0; |
| 137 | } |
| 138 | |
| 139 | /* |
| 140 | * Whenever anyone tries to change modes, we just mask interrupts |
| 141 | * and wait for the next event to get set. |
| 142 | */ |
| 143 | static int tile_timer_shutdown(struct clock_event_device *evt) |
| 144 | { |
| 145 | arch_local_irq_mask_now(INT_TILE_TIMER); |
| 146 | return 0; |
| 147 | } |
| 148 | |
| 149 | /* |
| 150 | * Set min_delta_ns to 1 microsecond, since it takes about |
| 151 | * that long to fire the interrupt. |
| 152 | */ |
| 153 | static DEFINE_PER_CPU(struct clock_event_device, tile_timer) = { |
| 154 | .name = "tile timer", |
| 155 | .features = CLOCK_EVT_FEAT_ONESHOT, |
| 156 | .min_delta_ns = 1000, |
| 157 | .rating = 100, |
| 158 | .irq = -1, |
| 159 | .set_next_event = tile_timer_set_next_event, |
| 160 | .set_state_shutdown = tile_timer_shutdown, |
| 161 | .set_state_oneshot = tile_timer_shutdown, |
| 162 | .tick_resume = tile_timer_shutdown, |
| 163 | }; |
| 164 | |
| 165 | void setup_tile_timer(void) |
| 166 | { |
| 167 | struct clock_event_device *evt = this_cpu_ptr(&tile_timer); |
| 168 | |
| 169 | /* Fill in fields that are speed-specific. */ |
| 170 | clockevents_calc_mult_shift(evt, cycles_per_sec, TILE_MINSEC); |
| 171 | evt->max_delta_ns = clockevent_delta2ns(MAX_TICK, evt); |
| 172 | |
| 173 | /* Mark as being for this cpu only. */ |
| 174 | evt->cpumask = cpumask_of(smp_processor_id()); |
| 175 | |
| 176 | /* Start out with timer not firing. */ |
| 177 | arch_local_irq_mask_now(INT_TILE_TIMER); |
| 178 | |
| 179 | /* Register tile timer. */ |
| 180 | clockevents_register_device(evt); |
| 181 | } |
| 182 | |
| 183 | /* Called from the interrupt vector. */ |
| 184 | void do_timer_interrupt(struct pt_regs *regs, int fault_num) |
| 185 | { |
| 186 | struct pt_regs *old_regs = set_irq_regs(regs); |
| 187 | struct clock_event_device *evt = this_cpu_ptr(&tile_timer); |
| 188 | |
| 189 | /* |
| 190 | * Mask the timer interrupt here, since we are a oneshot timer |
| 191 | * and there are now by definition no events pending. |
| 192 | */ |
| 193 | arch_local_irq_mask(INT_TILE_TIMER); |
| 194 | |
| 195 | /* Track time spent here in an interrupt context */ |
| 196 | irq_enter(); |
| 197 | |
| 198 | /* Track interrupt count. */ |
| 199 | __this_cpu_inc(irq_stat.irq_timer_count); |
| 200 | |
| 201 | /* Call the generic timer handler */ |
| 202 | evt->event_handler(evt); |
| 203 | |
| 204 | /* |
| 205 | * Track time spent against the current process again and |
| 206 | * process any softirqs if they are waiting. |
| 207 | */ |
| 208 | irq_exit(); |
| 209 | |
| 210 | set_irq_regs(old_regs); |
| 211 | } |
| 212 | |
| 213 | /* |
| 214 | * Scheduler clock - returns current time in nanosec units. |
| 215 | * Note that with LOCKDEP, this is called during lockdep_init(), and |
| 216 | * we will claim that sched_clock() is zero for a little while, until |
| 217 | * we run setup_clock(), above. |
| 218 | */ |
| 219 | unsigned long long sched_clock(void) |
| 220 | { |
| 221 | return mult_frac(get_cycles(), |
| 222 | sched_clock_mult, 1ULL << SCHED_CLOCK_SHIFT); |
| 223 | } |
| 224 | |
| 225 | int setup_profiling_timer(unsigned int multiplier) |
| 226 | { |
| 227 | return -EINVAL; |
| 228 | } |
| 229 | |
| 230 | /* |
| 231 | * Use the tile timer to convert nsecs to core clock cycles, relying |
| 232 | * on it having the same frequency as SPR_CYCLE. |
| 233 | */ |
| 234 | cycles_t ns2cycles(unsigned long nsecs) |
| 235 | { |
| 236 | /* |
| 237 | * We do not have to disable preemption here as each core has the same |
| 238 | * clock frequency. |
| 239 | */ |
| 240 | struct clock_event_device *dev = raw_cpu_ptr(&tile_timer); |
| 241 | |
| 242 | /* |
| 243 | * as in clocksource.h and x86's timer.h, we split the calculation |
| 244 | * into 2 parts to avoid unecessary overflow of the intermediate |
| 245 | * value. This will not lead to any loss of precision. |
| 246 | */ |
| 247 | u64 quot = (u64)nsecs >> dev->shift; |
| 248 | u64 rem = (u64)nsecs & ((1ULL << dev->shift) - 1); |
| 249 | return quot * dev->mult + ((rem * dev->mult) >> dev->shift); |
| 250 | } |
| 251 | |
| 252 | void update_vsyscall_tz(void) |
| 253 | { |
| 254 | write_seqcount_begin(&vdso_data->tz_seq); |
| 255 | vdso_data->tz_minuteswest = sys_tz.tz_minuteswest; |
| 256 | vdso_data->tz_dsttime = sys_tz.tz_dsttime; |
| 257 | write_seqcount_end(&vdso_data->tz_seq); |
| 258 | } |
| 259 | |
| 260 | void update_vsyscall(struct timekeeper *tk) |
| 261 | { |
| 262 | if (tk->tkr_mono.clock != &cycle_counter_cs) |
| 263 | return; |
| 264 | |
| 265 | write_seqcount_begin(&vdso_data->tb_seq); |
| 266 | |
| 267 | vdso_data->cycle_last = tk->tkr_mono.cycle_last; |
| 268 | vdso_data->mask = tk->tkr_mono.mask; |
| 269 | vdso_data->mult = tk->tkr_mono.mult; |
| 270 | vdso_data->shift = tk->tkr_mono.shift; |
| 271 | |
| 272 | vdso_data->wall_time_sec = tk->xtime_sec; |
| 273 | vdso_data->wall_time_snsec = tk->tkr_mono.xtime_nsec; |
| 274 | |
| 275 | vdso_data->monotonic_time_sec = tk->xtime_sec |
| 276 | + tk->wall_to_monotonic.tv_sec; |
| 277 | vdso_data->monotonic_time_snsec = tk->tkr_mono.xtime_nsec |
| 278 | + ((u64)tk->wall_to_monotonic.tv_nsec |
| 279 | << tk->tkr_mono.shift); |
| 280 | while (vdso_data->monotonic_time_snsec >= |
| 281 | (((u64)NSEC_PER_SEC) << tk->tkr_mono.shift)) { |
| 282 | vdso_data->monotonic_time_snsec -= |
| 283 | ((u64)NSEC_PER_SEC) << tk->tkr_mono.shift; |
| 284 | vdso_data->monotonic_time_sec++; |
| 285 | } |
| 286 | |
| 287 | vdso_data->wall_time_coarse_sec = tk->xtime_sec; |
| 288 | vdso_data->wall_time_coarse_nsec = (long)(tk->tkr_mono.xtime_nsec >> |
| 289 | tk->tkr_mono.shift); |
| 290 | |
| 291 | vdso_data->monotonic_time_coarse_sec = |
| 292 | vdso_data->wall_time_coarse_sec + tk->wall_to_monotonic.tv_sec; |
| 293 | vdso_data->monotonic_time_coarse_nsec = |
| 294 | vdso_data->wall_time_coarse_nsec + tk->wall_to_monotonic.tv_nsec; |
| 295 | |
| 296 | while (vdso_data->monotonic_time_coarse_nsec >= NSEC_PER_SEC) { |
| 297 | vdso_data->monotonic_time_coarse_nsec -= NSEC_PER_SEC; |
| 298 | vdso_data->monotonic_time_coarse_sec++; |
| 299 | } |
| 300 | |
| 301 | write_seqcount_end(&vdso_data->tb_seq); |
| 302 | } |