linux/drivers/clocksource/arm_arch_timer.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/drivers/clocksource/arm_arch_timer.c
   4 *
   5 *  Copyright (C) 2011 ARM Ltd.
   6 *  All Rights Reserved
   7 */
   8
   9#define pr_fmt(fmt)     "arch_timer: " fmt
  10
  11#include <linux/init.h>
  12#include <linux/kernel.h>
  13#include <linux/device.h>
  14#include <linux/smp.h>
  15#include <linux/cpu.h>
  16#include <linux/cpu_pm.h>
  17#include <linux/clockchips.h>
  18#include <linux/clocksource.h>
  19#include <linux/clocksource_ids.h>
  20#include <linux/interrupt.h>
  21#include <linux/of_irq.h>
  22#include <linux/of_address.h>
  23#include <linux/io.h>
  24#include <linux/slab.h>
  25#include <linux/sched/clock.h>
  26#include <linux/sched_clock.h>
  27#include <linux/acpi.h>
  28#include <linux/arm-smccc.h>
  29#include <linux/ptp_kvm.h>
  30
  31#include <asm/arch_timer.h>
  32#include <asm/virt.h>
  33
  34#include <clocksource/arm_arch_timer.h>
  35
  36#define CNTTIDR         0x08
  37#define CNTTIDR_VIRT(n) (BIT(1) << ((n) * 4))
  38
  39#define CNTACR(n)       (0x40 + ((n) * 4))
  40#define CNTACR_RPCT     BIT(0)
  41#define CNTACR_RVCT     BIT(1)
  42#define CNTACR_RFRQ     BIT(2)
  43#define CNTACR_RVOFF    BIT(3)
  44#define CNTACR_RWVT     BIT(4)
  45#define CNTACR_RWPT     BIT(5)
  46
  47#define CNTVCT_LO       0x08
  48#define CNTVCT_HI       0x0c
  49#define CNTFRQ          0x10
  50#define CNTP_TVAL       0x28
  51#define CNTP_CTL        0x2c
  52#define CNTV_TVAL       0x38
  53#define CNTV_CTL        0x3c
  54
  55static unsigned arch_timers_present __initdata;
  56
  57static void __iomem *arch_counter_base __ro_after_init;
  58
  59struct arch_timer {
  60        void __iomem *base;
  61        struct clock_event_device evt;
  62};
  63
  64#define to_arch_timer(e) container_of(e, struct arch_timer, evt)
  65
  66static u32 arch_timer_rate __ro_after_init;
  67static int arch_timer_ppi[ARCH_TIMER_MAX_TIMER_PPI] __ro_after_init;
  68
  69static const char *arch_timer_ppi_names[ARCH_TIMER_MAX_TIMER_PPI] = {
  70        [ARCH_TIMER_PHYS_SECURE_PPI]    = "sec-phys",
  71        [ARCH_TIMER_PHYS_NONSECURE_PPI] = "phys",
  72        [ARCH_TIMER_VIRT_PPI]           = "virt",
  73        [ARCH_TIMER_HYP_PPI]            = "hyp-phys",
  74        [ARCH_TIMER_HYP_VIRT_PPI]       = "hyp-virt",
  75};
  76
  77static struct clock_event_device __percpu *arch_timer_evt;
  78
  79static enum arch_timer_ppi_nr arch_timer_uses_ppi __ro_after_init = ARCH_TIMER_VIRT_PPI;
  80static bool arch_timer_c3stop __ro_after_init;
  81static bool arch_timer_mem_use_virtual __ro_after_init;
  82static bool arch_counter_suspend_stop __ro_after_init;
  83#ifdef CONFIG_GENERIC_GETTIMEOFDAY
  84static enum vdso_clock_mode vdso_default = VDSO_CLOCKMODE_ARCHTIMER;
  85#else
  86static enum vdso_clock_mode vdso_default = VDSO_CLOCKMODE_NONE;
  87#endif /* CONFIG_GENERIC_GETTIMEOFDAY */
  88
  89static cpumask_t evtstrm_available = CPU_MASK_NONE;
  90static bool evtstrm_enable __ro_after_init = IS_ENABLED(CONFIG_ARM_ARCH_TIMER_EVTSTREAM);
  91
  92static int __init early_evtstrm_cfg(char *buf)
  93{
  94        return strtobool(buf, &evtstrm_enable);
  95}
  96early_param("clocksource.arm_arch_timer.evtstrm", early_evtstrm_cfg);
  97
  98/*
  99 * Architected system timer support.
 100 */
 101
 102static __always_inline
 103void arch_timer_reg_write(int access, enum arch_timer_reg reg, u32 val,
 104                          struct clock_event_device *clk)
 105{
 106        if (access == ARCH_TIMER_MEM_PHYS_ACCESS) {
 107                struct arch_timer *timer = to_arch_timer(clk);
 108                switch (reg) {
 109                case ARCH_TIMER_REG_CTRL:
 110                        writel_relaxed(val, timer->base + CNTP_CTL);
 111                        break;
 112                case ARCH_TIMER_REG_TVAL:
 113                        writel_relaxed(val, timer->base + CNTP_TVAL);
 114                        break;
 115                }
 116        } else if (access == ARCH_TIMER_MEM_VIRT_ACCESS) {
 117                struct arch_timer *timer = to_arch_timer(clk);
 118                switch (reg) {
 119                case ARCH_TIMER_REG_CTRL:
 120                        writel_relaxed(val, timer->base + CNTV_CTL);
 121                        break;
 122                case ARCH_TIMER_REG_TVAL:
 123                        writel_relaxed(val, timer->base + CNTV_TVAL);
 124                        break;
 125                }
 126        } else {
 127                arch_timer_reg_write_cp15(access, reg, val);
 128        }
 129}
 130
 131static __always_inline
 132u32 arch_timer_reg_read(int access, enum arch_timer_reg reg,
 133                        struct clock_event_device *clk)
 134{
 135        u32 val;
 136
 137        if (access == ARCH_TIMER_MEM_PHYS_ACCESS) {
 138                struct arch_timer *timer = to_arch_timer(clk);
 139                switch (reg) {
 140                case ARCH_TIMER_REG_CTRL:
 141                        val = readl_relaxed(timer->base + CNTP_CTL);
 142                        break;
 143                case ARCH_TIMER_REG_TVAL:
 144                        val = readl_relaxed(timer->base + CNTP_TVAL);
 145                        break;
 146                }
 147        } else if (access == ARCH_TIMER_MEM_VIRT_ACCESS) {
 148                struct arch_timer *timer = to_arch_timer(clk);
 149                switch (reg) {
 150                case ARCH_TIMER_REG_CTRL:
 151                        val = readl_relaxed(timer->base + CNTV_CTL);
 152                        break;
 153                case ARCH_TIMER_REG_TVAL:
 154                        val = readl_relaxed(timer->base + CNTV_TVAL);
 155                        break;
 156                }
 157        } else {
 158                val = arch_timer_reg_read_cp15(access, reg);
 159        }
 160
 161        return val;
 162}
 163
 164static notrace u64 arch_counter_get_cntpct_stable(void)
 165{
 166        return __arch_counter_get_cntpct_stable();
 167}
 168
 169static notrace u64 arch_counter_get_cntpct(void)
 170{
 171        return __arch_counter_get_cntpct();
 172}
 173
 174static notrace u64 arch_counter_get_cntvct_stable(void)
 175{
 176        return __arch_counter_get_cntvct_stable();
 177}
 178
 179static notrace u64 arch_counter_get_cntvct(void)
 180{
 181        return __arch_counter_get_cntvct();
 182}
 183
 184/*
 185 * Default to cp15 based access because arm64 uses this function for
 186 * sched_clock() before DT is probed and the cp15 method is guaranteed
 187 * to exist on arm64. arm doesn't use this before DT is probed so even
 188 * if we don't have the cp15 accessors we won't have a problem.
 189 */
 190u64 (*arch_timer_read_counter)(void) __ro_after_init = arch_counter_get_cntvct;
 191EXPORT_SYMBOL_GPL(arch_timer_read_counter);
 192
 193static u64 arch_counter_read(struct clocksource *cs)
 194{
 195        return arch_timer_read_counter();
 196}
 197
 198static u64 arch_counter_read_cc(const struct cyclecounter *cc)
 199{
 200        return arch_timer_read_counter();
 201}
 202
 203static struct clocksource clocksource_counter = {
 204        .name   = "arch_sys_counter",
 205        .id     = CSID_ARM_ARCH_COUNTER,
 206        .rating = 400,
 207        .read   = arch_counter_read,
 208        .mask   = CLOCKSOURCE_MASK(56),
 209        .flags  = CLOCK_SOURCE_IS_CONTINUOUS,
 210};
 211
 212static struct cyclecounter cyclecounter __ro_after_init = {
 213        .read   = arch_counter_read_cc,
 214        .mask   = CLOCKSOURCE_MASK(56),
 215};
 216
 217struct ate_acpi_oem_info {
 218        char oem_id[ACPI_OEM_ID_SIZE + 1];
 219        char oem_table_id[ACPI_OEM_TABLE_ID_SIZE + 1];
 220        u32 oem_revision;
 221};
 222
 223#ifdef CONFIG_FSL_ERRATUM_A008585
 224/*
 225 * The number of retries is an arbitrary value well beyond the highest number
 226 * of iterations the loop has been observed to take.
 227 */
 228#define __fsl_a008585_read_reg(reg) ({                  \
 229        u64 _old, _new;                                 \
 230        int _retries = 200;                             \
 231                                                        \
 232        do {                                            \
 233                _old = read_sysreg(reg);                \
 234                _new = read_sysreg(reg);                \
 235                _retries--;                             \
 236        } while (unlikely(_old != _new) && _retries);   \
 237                                                        \
 238        WARN_ON_ONCE(!_retries);                        \
 239        _new;                                           \
 240})
 241
 242static u32 notrace fsl_a008585_read_cntp_tval_el0(void)
 243{
 244        return __fsl_a008585_read_reg(cntp_tval_el0);
 245}
 246
 247static u32 notrace fsl_a008585_read_cntv_tval_el0(void)
 248{
 249        return __fsl_a008585_read_reg(cntv_tval_el0);
 250}
 251
 252static u64 notrace fsl_a008585_read_cntpct_el0(void)
 253{
 254        return __fsl_a008585_read_reg(cntpct_el0);
 255}
 256
 257static u64 notrace fsl_a008585_read_cntvct_el0(void)
 258{
 259        return __fsl_a008585_read_reg(cntvct_el0);
 260}
 261#endif
 262
 263#ifdef CONFIG_HISILICON_ERRATUM_161010101
 264/*
 265 * Verify whether the value of the second read is larger than the first by
 266 * less than 32 is the only way to confirm the value is correct, so clear the
 267 * lower 5 bits to check whether the difference is greater than 32 or not.
 268 * Theoretically the erratum should not occur more than twice in succession
 269 * when reading the system counter, but it is possible that some interrupts
 270 * may lead to more than twice read errors, triggering the warning, so setting
 271 * the number of retries far beyond the number of iterations the loop has been
 272 * observed to take.
 273 */
 274#define __hisi_161010101_read_reg(reg) ({                               \
 275        u64 _old, _new;                                         \
 276        int _retries = 50;                                      \
 277                                                                \
 278        do {                                                    \
 279                _old = read_sysreg(reg);                        \
 280                _new = read_sysreg(reg);                        \
 281                _retries--;                                     \
 282        } while (unlikely((_new - _old) >> 5) && _retries);     \
 283                                                                \
 284        WARN_ON_ONCE(!_retries);                                \
 285        _new;                                                   \
 286})
 287
 288static u32 notrace hisi_161010101_read_cntp_tval_el0(void)
 289{
 290        return __hisi_161010101_read_reg(cntp_tval_el0);
 291}
 292
 293static u32 notrace hisi_161010101_read_cntv_tval_el0(void)
 294{
 295        return __hisi_161010101_read_reg(cntv_tval_el0);
 296}
 297
 298static u64 notrace hisi_161010101_read_cntpct_el0(void)
 299{
 300        return __hisi_161010101_read_reg(cntpct_el0);
 301}
 302
 303static u64 notrace hisi_161010101_read_cntvct_el0(void)
 304{
 305        return __hisi_161010101_read_reg(cntvct_el0);
 306}
 307
 308static struct ate_acpi_oem_info hisi_161010101_oem_info[] = {
 309        /*
 310         * Note that trailing spaces are required to properly match
 311         * the OEM table information.
 312         */
 313        {
 314                .oem_id         = "HISI  ",
 315                .oem_table_id   = "HIP05   ",
 316                .oem_revision   = 0,
 317        },
 318        {
 319                .oem_id         = "HISI  ",
 320                .oem_table_id   = "HIP06   ",
 321                .oem_revision   = 0,
 322        },
 323        {
 324                .oem_id         = "HISI  ",
 325                .oem_table_id   = "HIP07   ",
 326                .oem_revision   = 0,
 327        },
 328        { /* Sentinel indicating the end of the OEM array */ },
 329};
 330#endif
 331
 332#ifdef CONFIG_ARM64_ERRATUM_858921
 333static u64 notrace arm64_858921_read_cntpct_el0(void)
 334{
 335        u64 old, new;
 336
 337        old = read_sysreg(cntpct_el0);
 338        new = read_sysreg(cntpct_el0);
 339        return (((old ^ new) >> 32) & 1) ? old : new;
 340}
 341
 342static u64 notrace arm64_858921_read_cntvct_el0(void)
 343{
 344        u64 old, new;
 345
 346        old = read_sysreg(cntvct_el0);
 347        new = read_sysreg(cntvct_el0);
 348        return (((old ^ new) >> 32) & 1) ? old : new;
 349}
 350#endif
 351
 352#ifdef CONFIG_SUN50I_ERRATUM_UNKNOWN1
 353/*
 354 * The low bits of the counter registers are indeterminate while bit 10 or
 355 * greater is rolling over. Since the counter value can jump both backward
 356 * (7ff -> 000 -> 800) and forward (7ff -> fff -> 800), ignore register values
 357 * with all ones or all zeros in the low bits. Bound the loop by the maximum
 358 * number of CPU cycles in 3 consecutive 24 MHz counter periods.
 359 */
 360#define __sun50i_a64_read_reg(reg) ({                                   \
 361        u64 _val;                                                       \
 362        int _retries = 150;                                             \
 363                                                                        \
 364        do {                                                            \
 365                _val = read_sysreg(reg);                                \
 366                _retries--;                                             \
 367        } while (((_val + 1) & GENMASK(8, 0)) <= 1 && _retries);        \
 368                                                                        \
 369        WARN_ON_ONCE(!_retries);                                        \
 370        _val;                                                           \
 371})
 372
 373static u64 notrace sun50i_a64_read_cntpct_el0(void)
 374{
 375        return __sun50i_a64_read_reg(cntpct_el0);
 376}
 377
 378static u64 notrace sun50i_a64_read_cntvct_el0(void)
 379{
 380        return __sun50i_a64_read_reg(cntvct_el0);
 381}
 382
 383static u32 notrace sun50i_a64_read_cntp_tval_el0(void)
 384{
 385        return read_sysreg(cntp_cval_el0) - sun50i_a64_read_cntpct_el0();
 386}
 387
 388static u32 notrace sun50i_a64_read_cntv_tval_el0(void)
 389{
 390        return read_sysreg(cntv_cval_el0) - sun50i_a64_read_cntvct_el0();
 391}
 392#endif
 393
 394#ifdef CONFIG_ARM_ARCH_TIMER_OOL_WORKAROUND
 395DEFINE_PER_CPU(const struct arch_timer_erratum_workaround *, timer_unstable_counter_workaround);
 396EXPORT_SYMBOL_GPL(timer_unstable_counter_workaround);
 397
 398static atomic_t timer_unstable_counter_workaround_in_use = ATOMIC_INIT(0);
 399
 400static void erratum_set_next_event_tval_generic(const int access, unsigned long evt,
 401                                                struct clock_event_device *clk)
 402{
 403        unsigned long ctrl;
 404        u64 cval;
 405
 406        ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, clk);
 407        ctrl |= ARCH_TIMER_CTRL_ENABLE;
 408        ctrl &= ~ARCH_TIMER_CTRL_IT_MASK;
 409
 410        if (access == ARCH_TIMER_PHYS_ACCESS) {
 411                cval = evt + arch_counter_get_cntpct_stable();
 412                write_sysreg(cval, cntp_cval_el0);
 413        } else {
 414                cval = evt + arch_counter_get_cntvct_stable();
 415                write_sysreg(cval, cntv_cval_el0);
 416        }
 417
 418        arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
 419}
 420
 421static __maybe_unused int erratum_set_next_event_tval_virt(unsigned long evt,
 422                                            struct clock_event_device *clk)
 423{
 424        erratum_set_next_event_tval_generic(ARCH_TIMER_VIRT_ACCESS, evt, clk);
 425        return 0;
 426}
 427
 428static __maybe_unused int erratum_set_next_event_tval_phys(unsigned long evt,
 429                                            struct clock_event_device *clk)
 430{
 431        erratum_set_next_event_tval_generic(ARCH_TIMER_PHYS_ACCESS, evt, clk);
 432        return 0;
 433}
 434
 435static const struct arch_timer_erratum_workaround ool_workarounds[] = {
 436#ifdef CONFIG_FSL_ERRATUM_A008585
 437        {
 438                .match_type = ate_match_dt,
 439                .id = "fsl,erratum-a008585",
 440                .desc = "Freescale erratum a005858",
 441                .read_cntp_tval_el0 = fsl_a008585_read_cntp_tval_el0,
 442                .read_cntv_tval_el0 = fsl_a008585_read_cntv_tval_el0,
 443                .read_cntpct_el0 = fsl_a008585_read_cntpct_el0,
 444                .read_cntvct_el0 = fsl_a008585_read_cntvct_el0,
 445                .set_next_event_phys = erratum_set_next_event_tval_phys,
 446                .set_next_event_virt = erratum_set_next_event_tval_virt,
 447        },
 448#endif
 449#ifdef CONFIG_HISILICON_ERRATUM_161010101
 450        {
 451                .match_type = ate_match_dt,
 452                .id = "hisilicon,erratum-161010101",
 453                .desc = "HiSilicon erratum 161010101",
 454                .read_cntp_tval_el0 = hisi_161010101_read_cntp_tval_el0,
 455                .read_cntv_tval_el0 = hisi_161010101_read_cntv_tval_el0,
 456                .read_cntpct_el0 = hisi_161010101_read_cntpct_el0,
 457                .read_cntvct_el0 = hisi_161010101_read_cntvct_el0,
 458                .set_next_event_phys = erratum_set_next_event_tval_phys,
 459                .set_next_event_virt = erratum_set_next_event_tval_virt,
 460        },
 461        {
 462                .match_type = ate_match_acpi_oem_info,
 463                .id = hisi_161010101_oem_info,
 464                .desc = "HiSilicon erratum 161010101",
 465                .read_cntp_tval_el0 = hisi_161010101_read_cntp_tval_el0,
 466                .read_cntv_tval_el0 = hisi_161010101_read_cntv_tval_el0,
 467                .read_cntpct_el0 = hisi_161010101_read_cntpct_el0,
 468                .read_cntvct_el0 = hisi_161010101_read_cntvct_el0,
 469                .set_next_event_phys = erratum_set_next_event_tval_phys,
 470                .set_next_event_virt = erratum_set_next_event_tval_virt,
 471        },
 472#endif
 473#ifdef CONFIG_ARM64_ERRATUM_858921
 474        {
 475                .match_type = ate_match_local_cap_id,
 476                .id = (void *)ARM64_WORKAROUND_858921,
 477                .desc = "ARM erratum 858921",
 478                .read_cntpct_el0 = arm64_858921_read_cntpct_el0,
 479                .read_cntvct_el0 = arm64_858921_read_cntvct_el0,
 480        },
 481#endif
 482#ifdef CONFIG_SUN50I_ERRATUM_UNKNOWN1
 483        {
 484                .match_type = ate_match_dt,
 485                .id = "allwinner,erratum-unknown1",
 486                .desc = "Allwinner erratum UNKNOWN1",
 487                .read_cntp_tval_el0 = sun50i_a64_read_cntp_tval_el0,
 488                .read_cntv_tval_el0 = sun50i_a64_read_cntv_tval_el0,
 489                .read_cntpct_el0 = sun50i_a64_read_cntpct_el0,
 490                .read_cntvct_el0 = sun50i_a64_read_cntvct_el0,
 491                .set_next_event_phys = erratum_set_next_event_tval_phys,
 492                .set_next_event_virt = erratum_set_next_event_tval_virt,
 493        },
 494#endif
 495#ifdef CONFIG_ARM64_ERRATUM_1418040
 496        {
 497                .match_type = ate_match_local_cap_id,
 498                .id = (void *)ARM64_WORKAROUND_1418040,
 499                .desc = "ARM erratum 1418040",
 500                .disable_compat_vdso = true,
 501        },
 502#endif
 503};
 504
 505typedef bool (*ate_match_fn_t)(const struct arch_timer_erratum_workaround *,
 506                               const void *);
 507
 508static
 509bool arch_timer_check_dt_erratum(const struct arch_timer_erratum_workaround *wa,
 510                                 const void *arg)
 511{
 512        const struct device_node *np = arg;
 513
 514        return of_property_read_bool(np, wa->id);
 515}
 516
 517static
 518bool arch_timer_check_local_cap_erratum(const struct arch_timer_erratum_workaround *wa,
 519                                        const void *arg)
 520{
 521        return this_cpu_has_cap((uintptr_t)wa->id);
 522}
 523
 524
 525static
 526bool arch_timer_check_acpi_oem_erratum(const struct arch_timer_erratum_workaround *wa,
 527                                       const void *arg)
 528{
 529        static const struct ate_acpi_oem_info empty_oem_info = {};
 530        const struct ate_acpi_oem_info *info = wa->id;
 531        const struct acpi_table_header *table = arg;
 532
 533        /* Iterate over the ACPI OEM info array, looking for a match */
 534        while (memcmp(info, &empty_oem_info, sizeof(*info))) {
 535                if (!memcmp(info->oem_id, table->oem_id, ACPI_OEM_ID_SIZE) &&
 536                    !memcmp(info->oem_table_id, table->oem_table_id, ACPI_OEM_TABLE_ID_SIZE) &&
 537                    info->oem_revision == table->oem_revision)
 538                        return true;
 539
 540                info++;
 541        }
 542
 543        return false;
 544}
 545
 546static const struct arch_timer_erratum_workaround *
 547arch_timer_iterate_errata(enum arch_timer_erratum_match_type type,
 548                          ate_match_fn_t match_fn,
 549                          void *arg)
 550{
 551        int i;
 552
 553        for (i = 0; i < ARRAY_SIZE(ool_workarounds); i++) {
 554                if (ool_workarounds[i].match_type != type)
 555                        continue;
 556
 557                if (match_fn(&ool_workarounds[i], arg))
 558                        return &ool_workarounds[i];
 559        }
 560
 561        return NULL;
 562}
 563
 564static
 565void arch_timer_enable_workaround(const struct arch_timer_erratum_workaround *wa,
 566                                  bool local)
 567{
 568        int i;
 569
 570        if (local) {
 571                __this_cpu_write(timer_unstable_counter_workaround, wa);
 572        } else {
 573                for_each_possible_cpu(i)
 574                        per_cpu(timer_unstable_counter_workaround, i) = wa;
 575        }
 576
 577        if (wa->read_cntvct_el0 || wa->read_cntpct_el0)
 578                atomic_set(&timer_unstable_counter_workaround_in_use, 1);
 579
 580        /*
 581         * Don't use the vdso fastpath if errata require using the
 582         * out-of-line counter accessor. We may change our mind pretty
 583         * late in the game (with a per-CPU erratum, for example), so
 584         * change both the default value and the vdso itself.
 585         */
 586        if (wa->read_cntvct_el0) {
 587                clocksource_counter.vdso_clock_mode = VDSO_CLOCKMODE_NONE;
 588                vdso_default = VDSO_CLOCKMODE_NONE;
 589        } else if (wa->disable_compat_vdso && vdso_default != VDSO_CLOCKMODE_NONE) {
 590                vdso_default = VDSO_CLOCKMODE_ARCHTIMER_NOCOMPAT;
 591                clocksource_counter.vdso_clock_mode = vdso_default;
 592        }
 593}
 594
 595static void arch_timer_check_ool_workaround(enum arch_timer_erratum_match_type type,
 596                                            void *arg)
 597{
 598        const struct arch_timer_erratum_workaround *wa, *__wa;
 599        ate_match_fn_t match_fn = NULL;
 600        bool local = false;
 601
 602        switch (type) {
 603        case ate_match_dt:
 604                match_fn = arch_timer_check_dt_erratum;
 605                break;
 606        case ate_match_local_cap_id:
 607                match_fn = arch_timer_check_local_cap_erratum;
 608                local = true;
 609                break;
 610        case ate_match_acpi_oem_info:
 611                match_fn = arch_timer_check_acpi_oem_erratum;
 612                break;
 613        default:
 614                WARN_ON(1);
 615                return;
 616        }
 617
 618        wa = arch_timer_iterate_errata(type, match_fn, arg);
 619        if (!wa)
 620                return;
 621
 622        __wa = __this_cpu_read(timer_unstable_counter_workaround);
 623        if (__wa && wa != __wa)
 624                pr_warn("Can't enable workaround for %s (clashes with %s\n)",
 625                        wa->desc, __wa->desc);
 626
 627        if (__wa)
 628                return;
 629
 630        arch_timer_enable_workaround(wa, local);
 631        pr_info("Enabling %s workaround for %s\n",
 632                local ? "local" : "global", wa->desc);
 633}
 634
 635static bool arch_timer_this_cpu_has_cntvct_wa(void)
 636{
 637        return has_erratum_handler(read_cntvct_el0);
 638}
 639
 640static bool arch_timer_counter_has_wa(void)
 641{
 642        return atomic_read(&timer_unstable_counter_workaround_in_use);
 643}
 644#else
 645#define arch_timer_check_ool_workaround(t,a)            do { } while(0)
 646#define arch_timer_this_cpu_has_cntvct_wa()             ({false;})
 647#define arch_timer_counter_has_wa()                     ({false;})
 648#endif /* CONFIG_ARM_ARCH_TIMER_OOL_WORKAROUND */
 649
 650static __always_inline irqreturn_t timer_handler(const int access,
 651                                        struct clock_event_device *evt)
 652{
 653        unsigned long ctrl;
 654
 655        ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, evt);
 656        if (ctrl & ARCH_TIMER_CTRL_IT_STAT) {
 657                ctrl |= ARCH_TIMER_CTRL_IT_MASK;
 658                arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, evt);
 659                evt->event_handler(evt);
 660                return IRQ_HANDLED;
 661        }
 662
 663        return IRQ_NONE;
 664}
 665
 666static irqreturn_t arch_timer_handler_virt(int irq, void *dev_id)
 667{
 668        struct clock_event_device *evt = dev_id;
 669
 670        return timer_handler(ARCH_TIMER_VIRT_ACCESS, evt);
 671}
 672
 673static irqreturn_t arch_timer_handler_phys(int irq, void *dev_id)
 674{
 675        struct clock_event_device *evt = dev_id;
 676
 677        return timer_handler(ARCH_TIMER_PHYS_ACCESS, evt);
 678}
 679
 680static irqreturn_t arch_timer_handler_phys_mem(int irq, void *dev_id)
 681{
 682        struct clock_event_device *evt = dev_id;
 683
 684        return timer_handler(ARCH_TIMER_MEM_PHYS_ACCESS, evt);
 685}
 686
 687static irqreturn_t arch_timer_handler_virt_mem(int irq, void *dev_id)
 688{
 689        struct clock_event_device *evt = dev_id;
 690
 691        return timer_handler(ARCH_TIMER_MEM_VIRT_ACCESS, evt);
 692}
 693
 694static __always_inline int timer_shutdown(const int access,
 695                                          struct clock_event_device *clk)
 696{
 697        unsigned long ctrl;
 698
 699        ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, clk);
 700        ctrl &= ~ARCH_TIMER_CTRL_ENABLE;
 701        arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
 702
 703        return 0;
 704}
 705
 706static int arch_timer_shutdown_virt(struct clock_event_device *clk)
 707{
 708        return timer_shutdown(ARCH_TIMER_VIRT_ACCESS, clk);
 709}
 710
 711static int arch_timer_shutdown_phys(struct clock_event_device *clk)
 712{
 713        return timer_shutdown(ARCH_TIMER_PHYS_ACCESS, clk);
 714}
 715
 716static int arch_timer_shutdown_virt_mem(struct clock_event_device *clk)
 717{
 718        return timer_shutdown(ARCH_TIMER_MEM_VIRT_ACCESS, clk);
 719}
 720
 721static int arch_timer_shutdown_phys_mem(struct clock_event_device *clk)
 722{
 723        return timer_shutdown(ARCH_TIMER_MEM_PHYS_ACCESS, clk);
 724}
 725
 726static __always_inline void set_next_event(const int access, unsigned long evt,
 727                                           struct clock_event_device *clk)
 728{
 729        unsigned long ctrl;
 730        ctrl = arch_timer_reg_read(access, ARCH_TIMER_REG_CTRL, clk);
 731        ctrl |= ARCH_TIMER_CTRL_ENABLE;
 732        ctrl &= ~ARCH_TIMER_CTRL_IT_MASK;
 733        arch_timer_reg_write(access, ARCH_TIMER_REG_TVAL, evt, clk);
 734        arch_timer_reg_write(access, ARCH_TIMER_REG_CTRL, ctrl, clk);
 735}
 736
 737static int arch_timer_set_next_event_virt(unsigned long evt,
 738                                          struct clock_event_device *clk)
 739{
 740        set_next_event(ARCH_TIMER_VIRT_ACCESS, evt, clk);
 741        return 0;
 742}
 743
 744static int arch_timer_set_next_event_phys(unsigned long evt,
 745                                          struct clock_event_device *clk)
 746{
 747        set_next_event(ARCH_TIMER_PHYS_ACCESS, evt, clk);
 748        return 0;
 749}
 750
 751static int arch_timer_set_next_event_virt_mem(unsigned long evt,
 752                                              struct clock_event_device *clk)
 753{
 754        set_next_event(ARCH_TIMER_MEM_VIRT_ACCESS, evt, clk);
 755        return 0;
 756}
 757
 758static int arch_timer_set_next_event_phys_mem(unsigned long evt,
 759                                              struct clock_event_device *clk)
 760{
 761        set_next_event(ARCH_TIMER_MEM_PHYS_ACCESS, evt, clk);
 762        return 0;
 763}
 764
 765static void __arch_timer_setup(unsigned type,
 766                               struct clock_event_device *clk)
 767{
 768        clk->features = CLOCK_EVT_FEAT_ONESHOT;
 769
 770        if (type == ARCH_TIMER_TYPE_CP15) {
 771                typeof(clk->set_next_event) sne;
 772
 773                arch_timer_check_ool_workaround(ate_match_local_cap_id, NULL);
 774
 775                if (arch_timer_c3stop)
 776                        clk->features |= CLOCK_EVT_FEAT_C3STOP;
 777                clk->name = "arch_sys_timer";
 778                clk->rating = 450;
 779                clk->cpumask = cpumask_of(smp_processor_id());
 780                clk->irq = arch_timer_ppi[arch_timer_uses_ppi];
 781                switch (arch_timer_uses_ppi) {
 782                case ARCH_TIMER_VIRT_PPI:
 783                        clk->set_state_shutdown = arch_timer_shutdown_virt;
 784                        clk->set_state_oneshot_stopped = arch_timer_shutdown_virt;
 785                        sne = erratum_handler(set_next_event_virt);
 786                        break;
 787                case ARCH_TIMER_PHYS_SECURE_PPI:
 788                case ARCH_TIMER_PHYS_NONSECURE_PPI:
 789                case ARCH_TIMER_HYP_PPI:
 790                        clk->set_state_shutdown = arch_timer_shutdown_phys;
 791                        clk->set_state_oneshot_stopped = arch_timer_shutdown_phys;
 792                        sne = erratum_handler(set_next_event_phys);
 793                        break;
 794                default:
 795                        BUG();
 796                }
 797
 798                clk->set_next_event = sne;
 799        } else {
 800                clk->features |= CLOCK_EVT_FEAT_DYNIRQ;
 801                clk->name = "arch_mem_timer";
 802                clk->rating = 400;
 803                clk->cpumask = cpu_possible_mask;
 804                if (arch_timer_mem_use_virtual) {
 805                        clk->set_state_shutdown = arch_timer_shutdown_virt_mem;
 806                        clk->set_state_oneshot_stopped = arch_timer_shutdown_virt_mem;
 807                        clk->set_next_event =
 808                                arch_timer_set_next_event_virt_mem;
 809                } else {
 810                        clk->set_state_shutdown = arch_timer_shutdown_phys_mem;
 811                        clk->set_state_oneshot_stopped = arch_timer_shutdown_phys_mem;
 812                        clk->set_next_event =
 813                                arch_timer_set_next_event_phys_mem;
 814                }
 815        }
 816
 817        clk->set_state_shutdown(clk);
 818
 819        clockevents_config_and_register(clk, arch_timer_rate, 0xf, 0x7fffffff);
 820}
 821
 822static void arch_timer_evtstrm_enable(int divider)
 823{
 824        u32 cntkctl = arch_timer_get_cntkctl();
 825
 826        cntkctl &= ~ARCH_TIMER_EVT_TRIGGER_MASK;
 827        /* Set the divider and enable virtual event stream */
 828        cntkctl |= (divider << ARCH_TIMER_EVT_TRIGGER_SHIFT)
 829                        | ARCH_TIMER_VIRT_EVT_EN;
 830        arch_timer_set_cntkctl(cntkctl);
 831        arch_timer_set_evtstrm_feature();
 832        cpumask_set_cpu(smp_processor_id(), &evtstrm_available);
 833}
 834
 835static void arch_timer_configure_evtstream(void)
 836{
 837        int evt_stream_div, lsb;
 838
 839        /*
 840         * As the event stream can at most be generated at half the frequency
 841         * of the counter, use half the frequency when computing the divider.
 842         */
 843        evt_stream_div = arch_timer_rate / ARCH_TIMER_EVT_STREAM_FREQ / 2;
 844
 845        /*
 846         * Find the closest power of two to the divisor. If the adjacent bit
 847         * of lsb (last set bit, starts from 0) is set, then we use (lsb + 1).
 848         */
 849        lsb = fls(evt_stream_div) - 1;
 850        if (lsb > 0 && (evt_stream_div & BIT(lsb - 1)))
 851                lsb++;
 852
 853        /* enable event stream */
 854        arch_timer_evtstrm_enable(max(0, min(lsb, 15)));
 855}
 856
 857static void arch_counter_set_user_access(void)
 858{
 859        u32 cntkctl = arch_timer_get_cntkctl();
 860
 861        /* Disable user access to the timers and both counters */
 862        /* Also disable virtual event stream */
 863        cntkctl &= ~(ARCH_TIMER_USR_PT_ACCESS_EN
 864                        | ARCH_TIMER_USR_VT_ACCESS_EN
 865                        | ARCH_TIMER_USR_VCT_ACCESS_EN
 866                        | ARCH_TIMER_VIRT_EVT_EN
 867                        | ARCH_TIMER_USR_PCT_ACCESS_EN);
 868
 869        /*
 870         * Enable user access to the virtual counter if it doesn't
 871         * need to be workaround. The vdso may have been already
 872         * disabled though.
 873         */
 874        if (arch_timer_this_cpu_has_cntvct_wa())
 875                pr_info("CPU%d: Trapping CNTVCT access\n", smp_processor_id());
 876        else
 877                cntkctl |= ARCH_TIMER_USR_VCT_ACCESS_EN;
 878
 879        arch_timer_set_cntkctl(cntkctl);
 880}
 881
 882static bool arch_timer_has_nonsecure_ppi(void)
 883{
 884        return (arch_timer_uses_ppi == ARCH_TIMER_PHYS_SECURE_PPI &&
 885                arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI]);
 886}
 887
 888static u32 check_ppi_trigger(int irq)
 889{
 890        u32 flags = irq_get_trigger_type(irq);
 891
 892        if (flags != IRQF_TRIGGER_HIGH && flags != IRQF_TRIGGER_LOW) {
 893                pr_warn("WARNING: Invalid trigger for IRQ%d, assuming level low\n", irq);
 894                pr_warn("WARNING: Please fix your firmware\n");
 895                flags = IRQF_TRIGGER_LOW;
 896        }
 897
 898        return flags;
 899}
 900
 901static int arch_timer_starting_cpu(unsigned int cpu)
 902{
 903        struct clock_event_device *clk = this_cpu_ptr(arch_timer_evt);
 904        u32 flags;
 905
 906        __arch_timer_setup(ARCH_TIMER_TYPE_CP15, clk);
 907
 908        flags = check_ppi_trigger(arch_timer_ppi[arch_timer_uses_ppi]);
 909        enable_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi], flags);
 910
 911        if (arch_timer_has_nonsecure_ppi()) {
 912                flags = check_ppi_trigger(arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI]);
 913                enable_percpu_irq(arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI],
 914                                  flags);
 915        }
 916
 917        arch_counter_set_user_access();
 918        if (evtstrm_enable)
 919                arch_timer_configure_evtstream();
 920
 921        return 0;
 922}
 923
 924static int validate_timer_rate(void)
 925{
 926        if (!arch_timer_rate)
 927                return -EINVAL;
 928
 929        /* Arch timer frequency < 1MHz can cause trouble */
 930        WARN_ON(arch_timer_rate < 1000000);
 931
 932        return 0;
 933}
 934
 935/*
 936 * For historical reasons, when probing with DT we use whichever (non-zero)
 937 * rate was probed first, and don't verify that others match. If the first node
 938 * probed has a clock-frequency property, this overrides the HW register.
 939 */
 940static void __init arch_timer_of_configure_rate(u32 rate, struct device_node *np)
 941{
 942        /* Who has more than one independent system counter? */
 943        if (arch_timer_rate)
 944                return;
 945
 946        if (of_property_read_u32(np, "clock-frequency", &arch_timer_rate))
 947                arch_timer_rate = rate;
 948
 949        /* Check the timer frequency. */
 950        if (validate_timer_rate())
 951                pr_warn("frequency not available\n");
 952}
 953
 954static void __init arch_timer_banner(unsigned type)
 955{
 956        pr_info("%s%s%s timer(s) running at %lu.%02luMHz (%s%s%s).\n",
 957                type & ARCH_TIMER_TYPE_CP15 ? "cp15" : "",
 958                type == (ARCH_TIMER_TYPE_CP15 | ARCH_TIMER_TYPE_MEM) ?
 959                        " and " : "",
 960                type & ARCH_TIMER_TYPE_MEM ? "mmio" : "",
 961                (unsigned long)arch_timer_rate / 1000000,
 962                (unsigned long)(arch_timer_rate / 10000) % 100,
 963                type & ARCH_TIMER_TYPE_CP15 ?
 964                        (arch_timer_uses_ppi == ARCH_TIMER_VIRT_PPI) ? "virt" : "phys" :
 965                        "",
 966                type == (ARCH_TIMER_TYPE_CP15 | ARCH_TIMER_TYPE_MEM) ? "/" : "",
 967                type & ARCH_TIMER_TYPE_MEM ?
 968                        arch_timer_mem_use_virtual ? "virt" : "phys" :
 969                        "");
 970}
 971
 972u32 arch_timer_get_rate(void)
 973{
 974        return arch_timer_rate;
 975}
 976
 977bool arch_timer_evtstrm_available(void)
 978{
 979        /*
 980         * We might get called from a preemptible context. This is fine
 981         * because availability of the event stream should be always the same
 982         * for a preemptible context and context where we might resume a task.
 983         */
 984        return cpumask_test_cpu(raw_smp_processor_id(), &evtstrm_available);
 985}
 986
 987static u64 arch_counter_get_cntvct_mem(void)
 988{
 989        u32 vct_lo, vct_hi, tmp_hi;
 990
 991        do {
 992                vct_hi = readl_relaxed(arch_counter_base + CNTVCT_HI);
 993                vct_lo = readl_relaxed(arch_counter_base + CNTVCT_LO);
 994                tmp_hi = readl_relaxed(arch_counter_base + CNTVCT_HI);
 995        } while (vct_hi != tmp_hi);
 996
 997        return ((u64) vct_hi << 32) | vct_lo;
 998}
 999
1000static struct arch_timer_kvm_info arch_timer_kvm_info;
1001
1002struct arch_timer_kvm_info *arch_timer_get_kvm_info(void)
1003{
1004        return &arch_timer_kvm_info;
1005}
1006
1007static void __init arch_counter_register(unsigned type)
1008{
1009        u64 start_count;
1010
1011        /* Register the CP15 based counter if we have one */
1012        if (type & ARCH_TIMER_TYPE_CP15) {
1013                u64 (*rd)(void);
1014
1015                if ((IS_ENABLED(CONFIG_ARM64) && !is_hyp_mode_available()) ||
1016                    arch_timer_uses_ppi == ARCH_TIMER_VIRT_PPI) {
1017                        if (arch_timer_counter_has_wa())
1018                                rd = arch_counter_get_cntvct_stable;
1019                        else
1020                                rd = arch_counter_get_cntvct;
1021                } else {
1022                        if (arch_timer_counter_has_wa())
1023                                rd = arch_counter_get_cntpct_stable;
1024                        else
1025                                rd = arch_counter_get_cntpct;
1026                }
1027
1028                arch_timer_read_counter = rd;
1029                clocksource_counter.vdso_clock_mode = vdso_default;
1030        } else {
1031                arch_timer_read_counter = arch_counter_get_cntvct_mem;
1032        }
1033
1034        if (!arch_counter_suspend_stop)
1035                clocksource_counter.flags |= CLOCK_SOURCE_SUSPEND_NONSTOP;
1036        start_count = arch_timer_read_counter();
1037        clocksource_register_hz(&clocksource_counter, arch_timer_rate);
1038        cyclecounter.mult = clocksource_counter.mult;
1039        cyclecounter.shift = clocksource_counter.shift;
1040        timecounter_init(&arch_timer_kvm_info.timecounter,
1041                         &cyclecounter, start_count);
1042
1043        /* 56 bits minimum, so we assume worst case rollover */
1044        sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
1045}
1046
1047static void arch_timer_stop(struct clock_event_device *clk)
1048{
1049        pr_debug("disable IRQ%d cpu #%d\n", clk->irq, smp_processor_id());
1050
1051        disable_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi]);
1052        if (arch_timer_has_nonsecure_ppi())
1053                disable_percpu_irq(arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI]);
1054
1055        clk->set_state_shutdown(clk);
1056}
1057
1058static int arch_timer_dying_cpu(unsigned int cpu)
1059{
1060        struct clock_event_device *clk = this_cpu_ptr(arch_timer_evt);
1061
1062        cpumask_clear_cpu(smp_processor_id(), &evtstrm_available);
1063
1064        arch_timer_stop(clk);
1065        return 0;
1066}
1067
1068#ifdef CONFIG_CPU_PM
1069static DEFINE_PER_CPU(unsigned long, saved_cntkctl);
1070static int arch_timer_cpu_pm_notify(struct notifier_block *self,
1071                                    unsigned long action, void *hcpu)
1072{
1073        if (action == CPU_PM_ENTER) {
1074                __this_cpu_write(saved_cntkctl, arch_timer_get_cntkctl());
1075
1076                cpumask_clear_cpu(smp_processor_id(), &evtstrm_available);
1077        } else if (action == CPU_PM_ENTER_FAILED || action == CPU_PM_EXIT) {
1078                arch_timer_set_cntkctl(__this_cpu_read(saved_cntkctl));
1079
1080                if (arch_timer_have_evtstrm_feature())
1081                        cpumask_set_cpu(smp_processor_id(), &evtstrm_available);
1082        }
1083        return NOTIFY_OK;
1084}
1085
1086static struct notifier_block arch_timer_cpu_pm_notifier = {
1087        .notifier_call = arch_timer_cpu_pm_notify,
1088};
1089
1090static int __init arch_timer_cpu_pm_init(void)
1091{
1092        return cpu_pm_register_notifier(&arch_timer_cpu_pm_notifier);
1093}
1094
1095static void __init arch_timer_cpu_pm_deinit(void)
1096{
1097        WARN_ON(cpu_pm_unregister_notifier(&arch_timer_cpu_pm_notifier));
1098}
1099
1100#else
1101static int __init arch_timer_cpu_pm_init(void)
1102{
1103        return 0;
1104}
1105
1106static void __init arch_timer_cpu_pm_deinit(void)
1107{
1108}
1109#endif
1110
1111static int __init arch_timer_register(void)
1112{
1113        int err;
1114        int ppi;
1115
1116        arch_timer_evt = alloc_percpu(struct clock_event_device);
1117        if (!arch_timer_evt) {
1118                err = -ENOMEM;
1119                goto out;
1120        }
1121
1122        ppi = arch_timer_ppi[arch_timer_uses_ppi];
1123        switch (arch_timer_uses_ppi) {
1124        case ARCH_TIMER_VIRT_PPI:
1125                err = request_percpu_irq(ppi, arch_timer_handler_virt,
1126                                         "arch_timer", arch_timer_evt);
1127                break;
1128        case ARCH_TIMER_PHYS_SECURE_PPI:
1129        case ARCH_TIMER_PHYS_NONSECURE_PPI:
1130                err = request_percpu_irq(ppi, arch_timer_handler_phys,
1131                                         "arch_timer", arch_timer_evt);
1132                if (!err && arch_timer_has_nonsecure_ppi()) {
1133                        ppi = arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI];
1134                        err = request_percpu_irq(ppi, arch_timer_handler_phys,
1135                                                 "arch_timer", arch_timer_evt);
1136                        if (err)
1137                                free_percpu_irq(arch_timer_ppi[ARCH_TIMER_PHYS_SECURE_PPI],
1138                                                arch_timer_evt);
1139                }
1140                break;
1141        case ARCH_TIMER_HYP_PPI:
1142                err = request_percpu_irq(ppi, arch_timer_handler_phys,
1143                                         "arch_timer", arch_timer_evt);
1144                break;
1145        default:
1146                BUG();
1147        }
1148
1149        if (err) {
1150                pr_err("can't register interrupt %d (%d)\n", ppi, err);
1151                goto out_free;
1152        }
1153
1154        err = arch_timer_cpu_pm_init();
1155        if (err)
1156                goto out_unreg_notify;
1157
1158        /* Register and immediately configure the timer on the boot CPU */
1159        err = cpuhp_setup_state(CPUHP_AP_ARM_ARCH_TIMER_STARTING,
1160                                "clockevents/arm/arch_timer:starting",
1161                                arch_timer_starting_cpu, arch_timer_dying_cpu);
1162        if (err)
1163                goto out_unreg_cpupm;
1164        return 0;
1165
1166out_unreg_cpupm:
1167        arch_timer_cpu_pm_deinit();
1168
1169out_unreg_notify:
1170        free_percpu_irq(arch_timer_ppi[arch_timer_uses_ppi], arch_timer_evt);
1171        if (arch_timer_has_nonsecure_ppi())
1172                free_percpu_irq(arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI],
1173                                arch_timer_evt);
1174
1175out_free:
1176        free_percpu(arch_timer_evt);
1177out:
1178        return err;
1179}
1180
1181static int __init arch_timer_mem_register(void __iomem *base, unsigned int irq)
1182{
1183        int ret;
1184        irq_handler_t func;
1185        struct arch_timer *t;
1186
1187        t = kzalloc(sizeof(*t), GFP_KERNEL);
1188        if (!t)
1189                return -ENOMEM;
1190
1191        t->base = base;
1192        t->evt.irq = irq;
1193        __arch_timer_setup(ARCH_TIMER_TYPE_MEM, &t->evt);
1194
1195        if (arch_timer_mem_use_virtual)
1196                func = arch_timer_handler_virt_mem;
1197        else
1198                func = arch_timer_handler_phys_mem;
1199
1200        ret = request_irq(irq, func, IRQF_TIMER, "arch_mem_timer", &t->evt);
1201        if (ret) {
1202                pr_err("Failed to request mem timer irq\n");
1203                kfree(t);
1204        }
1205
1206        return ret;
1207}
1208
1209static const struct of_device_id arch_timer_of_match[] __initconst = {
1210        { .compatible   = "arm,armv7-timer",    },
1211        { .compatible   = "arm,armv8-timer",    },
1212        {},
1213};
1214
1215static const struct of_device_id arch_timer_mem_of_match[] __initconst = {
1216        { .compatible   = "arm,armv7-timer-mem", },
1217        {},
1218};
1219
1220static bool __init arch_timer_needs_of_probing(void)
1221{
1222        struct device_node *dn;
1223        bool needs_probing = false;
1224        unsigned int mask = ARCH_TIMER_TYPE_CP15 | ARCH_TIMER_TYPE_MEM;
1225
1226        /* We have two timers, and both device-tree nodes are probed. */
1227        if ((arch_timers_present & mask) == mask)
1228                return false;
1229
1230        /*
1231         * Only one type of timer is probed,
1232         * check if we have another type of timer node in device-tree.
1233         */
1234        if (arch_timers_present & ARCH_TIMER_TYPE_CP15)
1235                dn = of_find_matching_node(NULL, arch_timer_mem_of_match);
1236        else
1237                dn = of_find_matching_node(NULL, arch_timer_of_match);
1238
1239        if (dn && of_device_is_available(dn))
1240                needs_probing = true;
1241
1242        of_node_put(dn);
1243
1244        return needs_probing;
1245}
1246
1247static int __init arch_timer_common_init(void)
1248{
1249        arch_timer_banner(arch_timers_present);
1250        arch_counter_register(arch_timers_present);
1251        return arch_timer_arch_init();
1252}
1253
1254/**
1255 * arch_timer_select_ppi() - Select suitable PPI for the current system.
1256 *
1257 * If HYP mode is available, we know that the physical timer
1258 * has been configured to be accessible from PL1. Use it, so
1259 * that a guest can use the virtual timer instead.
1260 *
1261 * On ARMv8.1 with VH extensions, the kernel runs in HYP. VHE
1262 * accesses to CNTP_*_EL1 registers are silently redirected to
1263 * their CNTHP_*_EL2 counterparts, and use a different PPI
1264 * number.
1265 *
1266 * If no interrupt provided for virtual timer, we'll have to
1267 * stick to the physical timer. It'd better be accessible...
1268 * For arm64 we never use the secure interrupt.
1269 *
1270 * Return: a suitable PPI type for the current system.
1271 */
1272static enum arch_timer_ppi_nr __init arch_timer_select_ppi(void)
1273{
1274        if (is_kernel_in_hyp_mode())
1275                return ARCH_TIMER_HYP_PPI;
1276
1277        if (!is_hyp_mode_available() && arch_timer_ppi[ARCH_TIMER_VIRT_PPI])
1278                return ARCH_TIMER_VIRT_PPI;
1279
1280        if (IS_ENABLED(CONFIG_ARM64))
1281                return ARCH_TIMER_PHYS_NONSECURE_PPI;
1282
1283        return ARCH_TIMER_PHYS_SECURE_PPI;
1284}
1285
1286static void __init arch_timer_populate_kvm_info(void)
1287{
1288        arch_timer_kvm_info.virtual_irq = arch_timer_ppi[ARCH_TIMER_VIRT_PPI];
1289        if (is_kernel_in_hyp_mode())
1290                arch_timer_kvm_info.physical_irq = arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI];
1291}
1292
1293static int __init arch_timer_of_init(struct device_node *np)
1294{
1295        int i, irq, ret;
1296        u32 rate;
1297        bool has_names;
1298
1299        if (arch_timers_present & ARCH_TIMER_TYPE_CP15) {
1300                pr_warn("multiple nodes in dt, skipping\n");
1301                return 0;
1302        }
1303
1304        arch_timers_present |= ARCH_TIMER_TYPE_CP15;
1305
1306        has_names = of_property_read_bool(np, "interrupt-names");
1307
1308        for (i = ARCH_TIMER_PHYS_SECURE_PPI; i < ARCH_TIMER_MAX_TIMER_PPI; i++) {
1309                if (has_names)
1310                        irq = of_irq_get_byname(np, arch_timer_ppi_names[i]);
1311                else
1312                        irq = of_irq_get(np, i);
1313                if (irq > 0)
1314                        arch_timer_ppi[i] = irq;
1315        }
1316
1317        arch_timer_populate_kvm_info();
1318
1319        rate = arch_timer_get_cntfrq();
1320        arch_timer_of_configure_rate(rate, np);
1321
1322        arch_timer_c3stop = !of_property_read_bool(np, "always-on");
1323
1324        /* Check for globally applicable workarounds */
1325        arch_timer_check_ool_workaround(ate_match_dt, np);
1326
1327        /*
1328         * If we cannot rely on firmware initializing the timer registers then
1329         * we should use the physical timers instead.
1330         */
1331        if (IS_ENABLED(CONFIG_ARM) &&
1332            of_property_read_bool(np, "arm,cpu-registers-not-fw-configured"))
1333                arch_timer_uses_ppi = ARCH_TIMER_PHYS_SECURE_PPI;
1334        else
1335                arch_timer_uses_ppi = arch_timer_select_ppi();
1336
1337        if (!arch_timer_ppi[arch_timer_uses_ppi]) {
1338                pr_err("No interrupt available, giving up\n");
1339                return -EINVAL;
1340        }
1341
1342        /* On some systems, the counter stops ticking when in suspend. */
1343        arch_counter_suspend_stop = of_property_read_bool(np,
1344                                                         "arm,no-tick-in-suspend");
1345
1346        ret = arch_timer_register();
1347        if (ret)
1348                return ret;
1349
1350        if (arch_timer_needs_of_probing())
1351                return 0;
1352
1353        return arch_timer_common_init();
1354}
1355TIMER_OF_DECLARE(armv7_arch_timer, "arm,armv7-timer", arch_timer_of_init);
1356TIMER_OF_DECLARE(armv8_arch_timer, "arm,armv8-timer", arch_timer_of_init);
1357
1358static u32 __init
1359arch_timer_mem_frame_get_cntfrq(struct arch_timer_mem_frame *frame)
1360{
1361        void __iomem *base;
1362        u32 rate;
1363
1364        base = ioremap(frame->cntbase, frame->size);
1365        if (!base) {
1366                pr_err("Unable to map frame @ %pa\n", &frame->cntbase);
1367                return 0;
1368        }
1369
1370        rate = readl_relaxed(base + CNTFRQ);
1371
1372        iounmap(base);
1373
1374        return rate;
1375}
1376
1377static struct arch_timer_mem_frame * __init
1378arch_timer_mem_find_best_frame(struct arch_timer_mem *timer_mem)
1379{
1380        struct arch_timer_mem_frame *frame, *best_frame = NULL;
1381        void __iomem *cntctlbase;
1382        u32 cnttidr;
1383        int i;
1384
1385        cntctlbase = ioremap(timer_mem->cntctlbase, timer_mem->size);
1386        if (!cntctlbase) {
1387                pr_err("Can't map CNTCTLBase @ %pa\n",
1388                        &timer_mem->cntctlbase);
1389                return NULL;
1390        }
1391
1392        cnttidr = readl_relaxed(cntctlbase + CNTTIDR);
1393
1394        /*
1395         * Try to find a virtual capable frame. Otherwise fall back to a
1396         * physical capable frame.
1397         */
1398        for (i = 0; i < ARCH_TIMER_MEM_MAX_FRAMES; i++) {
1399                u32 cntacr = CNTACR_RFRQ | CNTACR_RWPT | CNTACR_RPCT |
1400                             CNTACR_RWVT | CNTACR_RVOFF | CNTACR_RVCT;
1401
1402                frame = &timer_mem->frame[i];
1403                if (!frame->valid)
1404                        continue;
1405
1406                /* Try enabling everything, and see what sticks */
1407                writel_relaxed(cntacr, cntctlbase + CNTACR(i));
1408                cntacr = readl_relaxed(cntctlbase + CNTACR(i));
1409
1410                if ((cnttidr & CNTTIDR_VIRT(i)) &&
1411                    !(~cntacr & (CNTACR_RWVT | CNTACR_RVCT))) {
1412                        best_frame = frame;
1413                        arch_timer_mem_use_virtual = true;
1414                        break;
1415                }
1416
1417                if (~cntacr & (CNTACR_RWPT | CNTACR_RPCT))
1418                        continue;
1419
1420                best_frame = frame;
1421        }
1422
1423        iounmap(cntctlbase);
1424
1425        return best_frame;
1426}
1427
1428static int __init
1429arch_timer_mem_frame_register(struct arch_timer_mem_frame *frame)
1430{
1431        void __iomem *base;
1432        int ret, irq = 0;
1433
1434        if (arch_timer_mem_use_virtual)
1435                irq = frame->virt_irq;
1436        else
1437                irq = frame->phys_irq;
1438
1439        if (!irq) {
1440                pr_err("Frame missing %s irq.\n",
1441                       arch_timer_mem_use_virtual ? "virt" : "phys");
1442                return -EINVAL;
1443        }
1444
1445        if (!request_mem_region(frame->cntbase, frame->size,
1446                                "arch_mem_timer"))
1447                return -EBUSY;
1448
1449        base = ioremap(frame->cntbase, frame->size);
1450        if (!base) {
1451                pr_err("Can't map frame's registers\n");
1452                return -ENXIO;
1453        }
1454
1455        ret = arch_timer_mem_register(base, irq);
1456        if (ret) {
1457                iounmap(base);
1458                return ret;
1459        }
1460
1461        arch_counter_base = base;
1462        arch_timers_present |= ARCH_TIMER_TYPE_MEM;
1463
1464        return 0;
1465}
1466
1467static int __init arch_timer_mem_of_init(struct device_node *np)
1468{
1469        struct arch_timer_mem *timer_mem;
1470        struct arch_timer_mem_frame *frame;
1471        struct device_node *frame_node;
1472        struct resource res;
1473        int ret = -EINVAL;
1474        u32 rate;
1475
1476        timer_mem = kzalloc(sizeof(*timer_mem), GFP_KERNEL);
1477        if (!timer_mem)
1478                return -ENOMEM;
1479
1480        if (of_address_to_resource(np, 0, &res))
1481                goto out;
1482        timer_mem->cntctlbase = res.start;
1483        timer_mem->size = resource_size(&res);
1484
1485        for_each_available_child_of_node(np, frame_node) {
1486                u32 n;
1487                struct arch_timer_mem_frame *frame;
1488
1489                if (of_property_read_u32(frame_node, "frame-number", &n)) {
1490                        pr_err(FW_BUG "Missing frame-number.\n");
1491                        of_node_put(frame_node);
1492                        goto out;
1493                }
1494                if (n >= ARCH_TIMER_MEM_MAX_FRAMES) {
1495                        pr_err(FW_BUG "Wrong frame-number, only 0-%u are permitted.\n",
1496                               ARCH_TIMER_MEM_MAX_FRAMES - 1);
1497                        of_node_put(frame_node);
1498                        goto out;
1499                }
1500                frame = &timer_mem->frame[n];
1501
1502                if (frame->valid) {
1503                        pr_err(FW_BUG "Duplicated frame-number.\n");
1504                        of_node_put(frame_node);
1505                        goto out;
1506                }
1507
1508                if (of_address_to_resource(frame_node, 0, &res)) {
1509                        of_node_put(frame_node);
1510                        goto out;
1511                }
1512                frame->cntbase = res.start;
1513                frame->size = resource_size(&res);
1514
1515                frame->virt_irq = irq_of_parse_and_map(frame_node,
1516                                                       ARCH_TIMER_VIRT_SPI);
1517                frame->phys_irq = irq_of_parse_and_map(frame_node,
1518                                                       ARCH_TIMER_PHYS_SPI);
1519
1520                frame->valid = true;
1521        }
1522
1523        frame = arch_timer_mem_find_best_frame(timer_mem);
1524        if (!frame) {
1525                pr_err("Unable to find a suitable frame in timer @ %pa\n",
1526                        &timer_mem->cntctlbase);
1527                ret = -EINVAL;
1528                goto out;
1529        }
1530
1531        rate = arch_timer_mem_frame_get_cntfrq(frame);
1532        arch_timer_of_configure_rate(rate, np);
1533
1534        ret = arch_timer_mem_frame_register(frame);
1535        if (!ret && !arch_timer_needs_of_probing())
1536                ret = arch_timer_common_init();
1537out:
1538        kfree(timer_mem);
1539        return ret;
1540}
1541TIMER_OF_DECLARE(armv7_arch_timer_mem, "arm,armv7-timer-mem",
1542                       arch_timer_mem_of_init);
1543
1544#ifdef CONFIG_ACPI_GTDT
1545static int __init
1546arch_timer_mem_verify_cntfrq(struct arch_timer_mem *timer_mem)
1547{
1548        struct arch_timer_mem_frame *frame;
1549        u32 rate;
1550        int i;
1551
1552        for (i = 0; i < ARCH_TIMER_MEM_MAX_FRAMES; i++) {
1553                frame = &timer_mem->frame[i];
1554
1555                if (!frame->valid)
1556                        continue;
1557
1558                rate = arch_timer_mem_frame_get_cntfrq(frame);
1559                if (rate == arch_timer_rate)
1560                        continue;
1561
1562                pr_err(FW_BUG "CNTFRQ mismatch: frame @ %pa: (0x%08lx), CPU: (0x%08lx)\n",
1563                        &frame->cntbase,
1564                        (unsigned long)rate, (unsigned long)arch_timer_rate);
1565
1566                return -EINVAL;
1567        }
1568
1569        return 0;
1570}
1571
1572static int __init arch_timer_mem_acpi_init(int platform_timer_count)
1573{
1574        struct arch_timer_mem *timers, *timer;
1575        struct arch_timer_mem_frame *frame, *best_frame = NULL;
1576        int timer_count, i, ret = 0;
1577
1578        timers = kcalloc(platform_timer_count, sizeof(*timers),
1579                            GFP_KERNEL);
1580        if (!timers)
1581                return -ENOMEM;
1582
1583        ret = acpi_arch_timer_mem_init(timers, &timer_count);
1584        if (ret || !timer_count)
1585                goto out;
1586
1587        /*
1588         * While unlikely, it's theoretically possible that none of the frames
1589         * in a timer expose the combination of feature we want.
1590         */
1591        for (i = 0; i < timer_count; i++) {
1592                timer = &timers[i];
1593
1594                frame = arch_timer_mem_find_best_frame(timer);
1595                if (!best_frame)
1596                        best_frame = frame;
1597
1598                ret = arch_timer_mem_verify_cntfrq(timer);
1599                if (ret) {
1600                        pr_err("Disabling MMIO timers due to CNTFRQ mismatch\n");
1601                        goto out;
1602                }
1603
1604                if (!best_frame) /* implies !frame */
1605                        /*
1606                         * Only complain about missing suitable frames if we
1607                         * haven't already found one in a previous iteration.
1608                         */
1609                        pr_err("Unable to find a suitable frame in timer @ %pa\n",
1610                                &timer->cntctlbase);
1611        }
1612
1613        if (best_frame)
1614                ret = arch_timer_mem_frame_register(best_frame);
1615out:
1616        kfree(timers);
1617        return ret;
1618}
1619
1620/* Initialize per-processor generic timer and memory-mapped timer(if present) */
1621static int __init arch_timer_acpi_init(struct acpi_table_header *table)
1622{
1623        int ret, platform_timer_count;
1624
1625        if (arch_timers_present & ARCH_TIMER_TYPE_CP15) {
1626                pr_warn("already initialized, skipping\n");
1627                return -EINVAL;
1628        }
1629
1630        arch_timers_present |= ARCH_TIMER_TYPE_CP15;
1631
1632        ret = acpi_gtdt_init(table, &platform_timer_count);
1633        if (ret)
1634                return ret;
1635
1636        arch_timer_ppi[ARCH_TIMER_PHYS_NONSECURE_PPI] =
1637                acpi_gtdt_map_ppi(ARCH_TIMER_PHYS_NONSECURE_PPI);
1638
1639        arch_timer_ppi[ARCH_TIMER_VIRT_PPI] =
1640                acpi_gtdt_map_ppi(ARCH_TIMER_VIRT_PPI);
1641
1642        arch_timer_ppi[ARCH_TIMER_HYP_PPI] =
1643                acpi_gtdt_map_ppi(ARCH_TIMER_HYP_PPI);
1644
1645        arch_timer_populate_kvm_info();
1646
1647        /*
1648         * When probing via ACPI, we have no mechanism to override the sysreg
1649         * CNTFRQ value. This *must* be correct.
1650         */
1651        arch_timer_rate = arch_timer_get_cntfrq();
1652        ret = validate_timer_rate();
1653        if (ret) {
1654                pr_err(FW_BUG "frequency not available.\n");
1655                return ret;
1656        }
1657
1658        arch_timer_uses_ppi = arch_timer_select_ppi();
1659        if (!arch_timer_ppi[arch_timer_uses_ppi]) {
1660                pr_err("No interrupt available, giving up\n");
1661                return -EINVAL;
1662        }
1663
1664        /* Always-on capability */
1665        arch_timer_c3stop = acpi_gtdt_c3stop(arch_timer_uses_ppi);
1666
1667        /* Check for globally applicable workarounds */
1668        arch_timer_check_ool_workaround(ate_match_acpi_oem_info, table);
1669
1670        ret = arch_timer_register();
1671        if (ret)
1672                return ret;
1673
1674        if (platform_timer_count &&
1675            arch_timer_mem_acpi_init(platform_timer_count))
1676                pr_err("Failed to initialize memory-mapped timer.\n");
1677
1678        return arch_timer_common_init();
1679}
1680TIMER_ACPI_DECLARE(arch_timer, ACPI_SIG_GTDT, arch_timer_acpi_init);
1681#endif
1682
1683int kvm_arch_ptp_get_crosststamp(u64 *cycle, struct timespec64 *ts,
1684                                 struct clocksource **cs)
1685{
1686        struct arm_smccc_res hvc_res;
1687        u32 ptp_counter;
1688        ktime_t ktime;
1689
1690        if (!IS_ENABLED(CONFIG_HAVE_ARM_SMCCC_DISCOVERY))
1691                return -EOPNOTSUPP;
1692
1693        if (arch_timer_uses_ppi == ARCH_TIMER_VIRT_PPI)
1694                ptp_counter = KVM_PTP_VIRT_COUNTER;
1695        else
1696                ptp_counter = KVM_PTP_PHYS_COUNTER;
1697
1698        arm_smccc_1_1_invoke(ARM_SMCCC_VENDOR_HYP_KVM_PTP_FUNC_ID,
1699                             ptp_counter, &hvc_res);
1700
1701        if ((int)(hvc_res.a0) < 0)
1702                return -EOPNOTSUPP;
1703
1704        ktime = (u64)hvc_res.a0 << 32 | hvc_res.a1;
1705        *ts = ktime_to_timespec64(ktime);
1706        if (cycle)
1707                *cycle = (u64)hvc_res.a2 << 32 | hvc_res.a3;
1708        if (cs)
1709                *cs = &clocksource_counter;
1710
1711        return 0;
1712}
1713EXPORT_SYMBOL_GPL(kvm_arch_ptp_get_crosststamp);
1714