Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright 2006 Andi Kleen, SUSE Labs. |
| 3 | * Subject to the GNU Public License, v.2 |
| 4 | * |
| 5 | * Fast user context implementation of clock_gettime, gettimeofday, and time. |
| 6 | * |
| 7 | * The code should have no internal unresolved relocations. |
| 8 | * Check with readelf after changing. |
| 9 | * Also alternative() doesn't work. |
| 10 | */ |
| 11 | /* |
| 12 | * Copyright (c) 2017 Oracle and/or its affiliates. All rights reserved. |
| 13 | */ |
| 14 | |
| 15 | /* Disable profiling for userspace code: */ |
| 16 | #ifndef DISABLE_BRANCH_PROFILING |
| 17 | #define DISABLE_BRANCH_PROFILING |
| 18 | #endif |
| 19 | |
| 20 | #include <linux/kernel.h> |
| 21 | #include <linux/time.h> |
| 22 | #include <linux/string.h> |
| 23 | #include <asm/io.h> |
| 24 | #include <asm/unistd.h> |
| 25 | #include <asm/timex.h> |
| 26 | #include <asm/clocksource.h> |
| 27 | #include <asm/vvar.h> |
| 28 | |
| 29 | #undef TICK_PRIV_BIT |
| 30 | #ifdef CONFIG_SPARC64 |
| 31 | #define TICK_PRIV_BIT (1UL << 63) |
| 32 | #else |
| 33 | #define TICK_PRIV_BIT (1ULL << 63) |
| 34 | #endif |
| 35 | |
| 36 | #ifdef CONFIG_SPARC64 |
| 37 | #define SYSCALL_STRING \ |
| 38 | "ta 0x6d;" \ |
| 39 | "bcs,a 1f;" \ |
| 40 | " sub %%g0, %%o0, %%o0;" \ |
| 41 | "1:" |
| 42 | #else |
| 43 | #define SYSCALL_STRING \ |
| 44 | "ta 0x10;" \ |
| 45 | "bcs,a 1f;" \ |
| 46 | " sub %%g0, %%o0, %%o0;" \ |
| 47 | "1:" |
| 48 | #endif |
| 49 | |
| 50 | #define SYSCALL_CLOBBERS \ |
| 51 | "f0", "f1", "f2", "f3", "f4", "f5", "f6", "f7", \ |
| 52 | "f8", "f9", "f10", "f11", "f12", "f13", "f14", "f15", \ |
| 53 | "f16", "f17", "f18", "f19", "f20", "f21", "f22", "f23", \ |
| 54 | "f24", "f25", "f26", "f27", "f28", "f29", "f30", "f31", \ |
| 55 | "f32", "f34", "f36", "f38", "f40", "f42", "f44", "f46", \ |
| 56 | "f48", "f50", "f52", "f54", "f56", "f58", "f60", "f62", \ |
| 57 | "cc", "memory" |
| 58 | |
| 59 | /* |
| 60 | * Compute the vvar page's address in the process address space, and return it |
| 61 | * as a pointer to the vvar_data. |
| 62 | */ |
| 63 | static notrace noinline struct vvar_data * |
| 64 | get_vvar_data(void) |
| 65 | { |
| 66 | unsigned long ret; |
| 67 | |
| 68 | /* |
| 69 | * vdso data page is the first vDSO page so grab the return address |
| 70 | * and move up a page to get to the data page. |
| 71 | */ |
| 72 | ret = (unsigned long)__builtin_return_address(0); |
| 73 | ret &= ~(8192 - 1); |
| 74 | ret -= 8192; |
| 75 | |
| 76 | return (struct vvar_data *) ret; |
| 77 | } |
| 78 | |
| 79 | static notrace long |
| 80 | vdso_fallback_gettime(long clock, struct timespec *ts) |
| 81 | { |
| 82 | register long num __asm__("g1") = __NR_clock_gettime; |
| 83 | register long o0 __asm__("o0") = clock; |
| 84 | register long o1 __asm__("o1") = (long) ts; |
| 85 | |
| 86 | __asm__ __volatile__(SYSCALL_STRING : "=r" (o0) : "r" (num), |
| 87 | "0" (o0), "r" (o1) : SYSCALL_CLOBBERS); |
| 88 | return o0; |
| 89 | } |
| 90 | |
| 91 | static notrace __always_inline long |
| 92 | vdso_fallback_gettimeofday(struct timeval *tv, struct timezone *tz) |
| 93 | { |
| 94 | register long num __asm__("g1") = __NR_gettimeofday; |
| 95 | register long o0 __asm__("o0") = (long) tv; |
| 96 | register long o1 __asm__("o1") = (long) tz; |
| 97 | |
| 98 | __asm__ __volatile__(SYSCALL_STRING : "=r" (o0) : "r" (num), |
| 99 | "0" (o0), "r" (o1) : SYSCALL_CLOBBERS); |
| 100 | return o0; |
| 101 | } |
| 102 | |
| 103 | #ifdef CONFIG_SPARC64 |
| 104 | static notrace noinline u64 |
| 105 | vread_tick(void) { |
| 106 | u64 ret; |
| 107 | |
| 108 | __asm__ __volatile__("rd %%asr24, %0 \n" |
| 109 | ".section .vread_tick_patch, \"ax\" \n" |
| 110 | "rd %%tick, %0 \n" |
| 111 | ".previous \n" |
| 112 | : "=&r" (ret)); |
| 113 | return ret & ~TICK_PRIV_BIT; |
| 114 | } |
| 115 | #else |
| 116 | static notrace noinline u64 |
| 117 | vread_tick(void) |
| 118 | { |
| 119 | unsigned int lo, hi; |
| 120 | |
| 121 | __asm__ __volatile__("rd %%asr24, %%g1\n\t" |
| 122 | "srlx %%g1, 32, %1\n\t" |
| 123 | "srl %%g1, 0, %0\n" |
| 124 | ".section .vread_tick_patch, \"ax\" \n" |
| 125 | "rd %%tick, %%g1\n" |
| 126 | ".previous \n" |
| 127 | : "=&r" (lo), "=&r" (hi) |
| 128 | : |
| 129 | : "g1"); |
| 130 | return lo | ((u64)hi << 32); |
| 131 | } |
| 132 | #endif |
| 133 | |
| 134 | static notrace inline u64 |
| 135 | vgetsns(struct vvar_data *vvar) |
| 136 | { |
| 137 | u64 v; |
| 138 | u64 cycles; |
| 139 | |
| 140 | cycles = vread_tick(); |
| 141 | v = (cycles - vvar->clock.cycle_last) & vvar->clock.mask; |
| 142 | return v * vvar->clock.mult; |
| 143 | } |
| 144 | |
| 145 | static notrace noinline int |
| 146 | do_realtime(struct vvar_data *vvar, struct timespec *ts) |
| 147 | { |
| 148 | unsigned long seq; |
| 149 | u64 ns; |
| 150 | |
| 151 | ts->tv_nsec = 0; |
| 152 | do { |
| 153 | seq = vvar_read_begin(vvar); |
| 154 | ts->tv_sec = vvar->wall_time_sec; |
| 155 | ns = vvar->wall_time_snsec; |
| 156 | ns += vgetsns(vvar); |
| 157 | ns >>= vvar->clock.shift; |
| 158 | } while (unlikely(vvar_read_retry(vvar, seq))); |
| 159 | |
| 160 | timespec_add_ns(ts, ns); |
| 161 | |
| 162 | return 0; |
| 163 | } |
| 164 | |
| 165 | static notrace noinline int |
| 166 | do_monotonic(struct vvar_data *vvar, struct timespec *ts) |
| 167 | { |
| 168 | unsigned long seq; |
| 169 | u64 ns; |
| 170 | |
| 171 | ts->tv_nsec = 0; |
| 172 | do { |
| 173 | seq = vvar_read_begin(vvar); |
| 174 | ts->tv_sec = vvar->monotonic_time_sec; |
| 175 | ns = vvar->monotonic_time_snsec; |
| 176 | ns += vgetsns(vvar); |
| 177 | ns >>= vvar->clock.shift; |
| 178 | } while (unlikely(vvar_read_retry(vvar, seq))); |
| 179 | |
| 180 | timespec_add_ns(ts, ns); |
| 181 | |
| 182 | return 0; |
| 183 | } |
| 184 | |
| 185 | static notrace noinline int |
| 186 | do_realtime_coarse(struct vvar_data *vvar, struct timespec *ts) |
| 187 | { |
| 188 | unsigned long seq; |
| 189 | |
| 190 | do { |
| 191 | seq = vvar_read_begin(vvar); |
| 192 | ts->tv_sec = vvar->wall_time_coarse_sec; |
| 193 | ts->tv_nsec = vvar->wall_time_coarse_nsec; |
| 194 | } while (unlikely(vvar_read_retry(vvar, seq))); |
| 195 | return 0; |
| 196 | } |
| 197 | |
| 198 | static notrace noinline int |
| 199 | do_monotonic_coarse(struct vvar_data *vvar, struct timespec *ts) |
| 200 | { |
| 201 | unsigned long seq; |
| 202 | |
| 203 | do { |
| 204 | seq = vvar_read_begin(vvar); |
| 205 | ts->tv_sec = vvar->monotonic_time_coarse_sec; |
| 206 | ts->tv_nsec = vvar->monotonic_time_coarse_nsec; |
| 207 | } while (unlikely(vvar_read_retry(vvar, seq))); |
| 208 | |
| 209 | return 0; |
| 210 | } |
| 211 | |
| 212 | notrace int |
| 213 | __vdso_clock_gettime(clockid_t clock, struct timespec *ts) |
| 214 | { |
| 215 | struct vvar_data *vvd = get_vvar_data(); |
| 216 | |
| 217 | switch (clock) { |
| 218 | case CLOCK_REALTIME: |
| 219 | if (unlikely(vvd->vclock_mode == VCLOCK_NONE)) |
| 220 | break; |
| 221 | return do_realtime(vvd, ts); |
| 222 | case CLOCK_MONOTONIC: |
| 223 | if (unlikely(vvd->vclock_mode == VCLOCK_NONE)) |
| 224 | break; |
| 225 | return do_monotonic(vvd, ts); |
| 226 | case CLOCK_REALTIME_COARSE: |
| 227 | return do_realtime_coarse(vvd, ts); |
| 228 | case CLOCK_MONOTONIC_COARSE: |
| 229 | return do_monotonic_coarse(vvd, ts); |
| 230 | } |
| 231 | /* |
| 232 | * Unknown clock ID ? Fall back to the syscall. |
| 233 | */ |
| 234 | return vdso_fallback_gettime(clock, ts); |
| 235 | } |
| 236 | int |
| 237 | clock_gettime(clockid_t, struct timespec *) |
| 238 | __attribute__((weak, alias("__vdso_clock_gettime"))); |
| 239 | |
| 240 | notrace int |
| 241 | __vdso_gettimeofday(struct timeval *tv, struct timezone *tz) |
| 242 | { |
| 243 | struct vvar_data *vvd = get_vvar_data(); |
| 244 | |
| 245 | if (likely(vvd->vclock_mode != VCLOCK_NONE)) { |
| 246 | if (likely(tv != NULL)) { |
| 247 | union tstv_t { |
| 248 | struct timespec ts; |
| 249 | struct timeval tv; |
| 250 | } *tstv = (union tstv_t *) tv; |
| 251 | do_realtime(vvd, &tstv->ts); |
| 252 | /* |
| 253 | * Assign before dividing to ensure that the division is |
| 254 | * done in the type of tv_usec, not tv_nsec. |
| 255 | * |
| 256 | * There cannot be > 1 billion usec in a second: |
| 257 | * do_realtime() has already distributed such overflow |
| 258 | * into tv_sec. So we can assign it to an int safely. |
| 259 | */ |
| 260 | tstv->tv.tv_usec = tstv->ts.tv_nsec; |
| 261 | tstv->tv.tv_usec /= 1000; |
| 262 | } |
| 263 | if (unlikely(tz != NULL)) { |
| 264 | /* Avoid memcpy. Some old compilers fail to inline it */ |
| 265 | tz->tz_minuteswest = vvd->tz_minuteswest; |
| 266 | tz->tz_dsttime = vvd->tz_dsttime; |
| 267 | } |
| 268 | return 0; |
| 269 | } |
| 270 | return vdso_fallback_gettimeofday(tv, tz); |
| 271 | } |
| 272 | int |
| 273 | gettimeofday(struct timeval *, struct timezone *) |
| 274 | __attribute__((weak, alias("__vdso_gettimeofday"))); |