1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Generic userspace implementations of gettimeofday() and similar.
4 */
5 #include <vdso/datapage.h>
6 #include <vdso/helpers.h>
7
8 #ifndef vdso_calc_ns
9
10 #ifdef VDSO_DELTA_NOMASK
11 # define VDSO_DELTA_MASK(vd) ULLONG_MAX
12 #else
13 # define VDSO_DELTA_MASK(vd) (vd->mask)
14 #endif
15
16 #ifdef CONFIG_GENERIC_VDSO_OVERFLOW_PROTECT
17 static __always_inline bool vdso_delta_ok(const struct vdso_data *vd, u64 delta)
18 {
19 return delta < vd->max_cycles;
20 }
21 #else
22 static __always_inline bool vdso_delta_ok(const struct vdso_data *vd, u64 delta)
23 {
24 return true;
25 }
26 #endif
27
28 #ifndef vdso_shift_ns
29 static __always_inline u64 vdso_shift_ns(u64 ns, u32 shift)
30 {
31 return ns >> shift;
32 }
33 #endif
34
35 /*
36 * Default implementation which works for all sane clocksources. That
37 * obviously excludes x86/TSC.
38 */
39 static __always_inline u64 vdso_calc_ns(const struct vdso_data *vd, u64 cycles, u64 base)
40 {
41 u64 delta = (cycles - vd->cycle_last) & VDSO_DELTA_MASK(vd);
42
43 if (likely(vdso_delta_ok(vd, delta)))
44 return vdso_shift_ns((delta * vd->mult) + base, vd->shift);
45
46 return mul_u64_u32_add_u64_shr(delta, vd->mult, base, vd->shift);
47 }
48 #endif /* vdso_calc_ns */
49
50 #ifndef __arch_vdso_hres_capable
51 static inline bool __arch_vdso_hres_capable(void)
52 {
53 return true;
54 }
55 #endif
56
57 #ifndef vdso_clocksource_ok
58 static inline bool vdso_clocksource_ok(const struct vdso_data *vd)
59 {
60 return vd->clock_mode != VDSO_CLOCKMODE_NONE;
61 }
62 #endif
63
64 #ifndef vdso_cycles_ok
65 static inline bool vdso_cycles_ok(u64 cycles)
66 {
67 return true;
68 }
69 #endif
70
71 #ifdef CONFIG_TIME_NS
72 static __always_inline int do_hres_timens(const struct vdso_data *vdns, clockid_t clk,
73 struct __kernel_timespec *ts)
74 {
75 const struct timens_offset *offs = &vdns->offset[clk];
76 const struct vdso_timestamp *vdso_ts;
77 const struct vdso_data *vd;
78 u64 cycles, ns;
79 u32 seq;
80 s64 sec;
81
82 vd = vdns - (clk == CLOCK_MONOTONIC_RAW ? CS_RAW : CS_HRES_COARSE);
83 vd = __arch_get_timens_vdso_data(vd);
84 if (clk != CLOCK_MONOTONIC_RAW)
85 vd = &vd[CS_HRES_COARSE];
86 else
87 vd = &vd[CS_RAW];
88 vdso_ts = &vd->basetime[clk];
89
90 do {
91 seq = vdso_read_begin(vd);
92
93 if (unlikely(!vdso_clocksource_ok(vd)))
94 return -1;
95
96 cycles = __arch_get_hw_counter(vd->clock_mode, vd);
97 if (unlikely(!vdso_cycles_ok(cycles)))
98 return -1;
99 ns = vdso_calc_ns(vd, cycles, vdso_ts->nsec);
100 sec = vdso_ts->sec;
101 } while (unlikely(vdso_read_retry(vd, seq)));
102
103 /* Add the namespace offset */
104 sec += offs->sec;
105 ns += offs->nsec;
106
107 /*
108 * Do this outside the loop: a race inside the loop could result
109 * in __iter_div_u64_rem() being extremely slow.
110 */
111 ts->tv_sec = sec + __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
112 ts->tv_nsec = ns;
113
114 return 0;
115 }
116 #else
117 static __always_inline
118 const struct vdso_data *__arch_get_timens_vdso_data(const struct vdso_data *vd)
119 {
120 return NULL;
121 }
122
123 static __always_inline int do_hres_timens(const struct vdso_data *vdns, clockid_t clk,
124 struct __kernel_timespec *ts)
125 {
126 return -EINVAL;
127 }
128 #endif
129
130 static __always_inline int do_hres(const struct vdso_data *vd, clockid_t clk,
131 struct __kernel_timespec *ts)
132 {
133 const struct vdso_timestamp *vdso_ts = &vd->basetime[clk];
134 u64 cycles, sec, ns;
135 u32 seq;
136
137 /* Allows to compile the high resolution parts out */
138 if (!__arch_vdso_hres_capable())
139 return -1;
140
141 do {
142 /*
143 * Open coded function vdso_read_begin() to handle
144 * VDSO_CLOCKMODE_TIMENS. Time namespace enabled tasks have a
145 * special VVAR page installed which has vd->seq set to 1 and
146 * vd->clock_mode set to VDSO_CLOCKMODE_TIMENS. For non time
147 * namespace affected tasks this does not affect performance
148 * because if vd->seq is odd, i.e. a concurrent update is in
149 * progress the extra check for vd->clock_mode is just a few
150 * extra instructions while spin waiting for vd->seq to become
151 * even again.
152 */
153 while (unlikely((seq = READ_ONCE(vd->seq)) & 1)) {
154 if (IS_ENABLED(CONFIG_TIME_NS) &&
155 vd->clock_mode == VDSO_CLOCKMODE_TIMENS)
156 return do_hres_timens(vd, clk, ts);
157 cpu_relax();
158 }
159 smp_rmb();
160
161 if (unlikely(!vdso_clocksource_ok(vd)))
162 return -1;
163
164 cycles = __arch_get_hw_counter(vd->clock_mode, vd);
165 if (unlikely(!vdso_cycles_ok(cycles)))
166 return -1;
167 ns = vdso_calc_ns(vd, cycles, vdso_ts->nsec);
168 sec = vdso_ts->sec;
169 } while (unlikely(vdso_read_retry(vd, seq)));
170
171 /*
172 * Do this outside the loop: a race inside the loop could result
173 * in __iter_div_u64_rem() being extremely slow.
174 */
175 ts->tv_sec = sec + __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
176 ts->tv_nsec = ns;
177
178 return 0;
179 }
180
181 #ifdef CONFIG_TIME_NS
182 static __always_inline int do_coarse_timens(const struct vdso_data *vdns, clockid_t clk,
183 struct __kernel_timespec *ts)
184 {
185 const struct vdso_data *vd = __arch_get_timens_vdso_data(vdns);
186 const struct vdso_timestamp *vdso_ts = &vd->basetime[clk];
187 const struct timens_offset *offs = &vdns->offset[clk];
188 u64 nsec;
189 s64 sec;
190 s32 seq;
191
192 do {
193 seq = vdso_read_begin(vd);
194 sec = vdso_ts->sec;
195 nsec = vdso_ts->nsec;
196 } while (unlikely(vdso_read_retry(vd, seq)));
197
198 /* Add the namespace offset */
199 sec += offs->sec;
200 nsec += offs->nsec;
201
202 /*
203 * Do this outside the loop: a race inside the loop could result
204 * in __iter_div_u64_rem() being extremely slow.
205 */
206 ts->tv_sec = sec + __iter_div_u64_rem(nsec, NSEC_PER_SEC, &nsec);
207 ts->tv_nsec = nsec;
208 return 0;
209 }
210 #else
211 static __always_inline int do_coarse_timens(const struct vdso_data *vdns, clockid_t clk,
212 struct __kernel_timespec *ts)
213 {
214 return -1;
215 }
216 #endif
217
218 static __always_inline int do_coarse(const struct vdso_data *vd, clockid_t clk,
219 struct __kernel_timespec *ts)
220 {
221 const struct vdso_timestamp *vdso_ts = &vd->basetime[clk];
222 u32 seq;
223
224 do {
225 /*
226 * Open coded function vdso_read_begin() to handle
227 * VDSO_CLOCK_TIMENS. See comment in do_hres().
228 */
229 while ((seq = READ_ONCE(vd->seq)) & 1) {
230 if (IS_ENABLED(CONFIG_TIME_NS) &&
231 vd->clock_mode == VDSO_CLOCKMODE_TIMENS)
232 return do_coarse_timens(vd, clk, ts);
233 cpu_relax();
234 }
235 smp_rmb();
236
237 ts->tv_sec = vdso_ts->sec;
238 ts->tv_nsec = vdso_ts->nsec;
239 } while (unlikely(vdso_read_retry(vd, seq)));
240
241 return 0;
242 }
243
244 static __always_inline int
245 __cvdso_clock_gettime_common(const struct vdso_data *vd, clockid_t clock,
246 struct __kernel_timespec *ts)
247 {
248 u32 msk;
249
250 /* Check for negative values or invalid clocks */
251 if (unlikely((u32) clock >= MAX_CLOCKS))
252 return -1;
253
254 /*
255 * Convert the clockid to a bitmask and use it to check which
256 * clocks are handled in the VDSO directly.
257 */
258 msk = 1U << clock;
259 if (likely(msk & VDSO_HRES))
260 vd = &vd[CS_HRES_COARSE];
261 else if (msk & VDSO_COARSE)
262 return do_coarse(&vd[CS_HRES_COARSE], clock, ts);
263 else if (msk & VDSO_RAW)
264 vd = &vd[CS_RAW];
265 else
266 return -1;
267
268 return do_hres(vd, clock, ts);
269 }
270
271 static __maybe_unused int
272 __cvdso_clock_gettime_data(const struct vdso_data *vd, clockid_t clock,
273 struct __kernel_timespec *ts)
274 {
275 int ret = __cvdso_clock_gettime_common(vd, clock, ts);
276
277 if (unlikely(ret))
278 return clock_gettime_fallback(clock, ts);
279 return 0;
280 }
281
282 static __maybe_unused int
283 __cvdso_clock_gettime(clockid_t clock, struct __kernel_timespec *ts)
284 {
285 return __cvdso_clock_gettime_data(__arch_get_vdso_data(), clock, ts);
286 }
287
288 #ifdef BUILD_VDSO32
289 static __maybe_unused int
290 __cvdso_clock_gettime32_data(const struct vdso_data *vd, clockid_t clock,
291 struct old_timespec32 *res)
292 {
293 struct __kernel_timespec ts;
294 int ret;
295
296 ret = __cvdso_clock_gettime_common(vd, clock, &ts);
297
298 if (unlikely(ret))
299 return clock_gettime32_fallback(clock, res);
300
301 /* For ret == 0 */
302 res->tv_sec = ts.tv_sec;
303 res->tv_nsec = ts.tv_nsec;
304
305 return ret;
306 }
307
308 static __maybe_unused int
309 __cvdso_clock_gettime32(clockid_t clock, struct old_timespec32 *res)
310 {
311 return __cvdso_clock_gettime32_data(__arch_get_vdso_data(), clock, res);
312 }
313 #endif /* BUILD_VDSO32 */
314
315 static __maybe_unused int
316 __cvdso_gettimeofday_data(const struct vdso_data *vd,
317 struct __kernel_old_timeval *tv, struct timezone *tz)
318 {
319
320 if (likely(tv != NULL)) {
321 struct __kernel_timespec ts;
322
323 if (do_hres(&vd[CS_HRES_COARSE], CLOCK_REALTIME, &ts))
324 return gettimeofday_fallback(tv, tz);
325
326 tv->tv_sec = ts.tv_sec;
327 tv->tv_usec = (u32)ts.tv_nsec / NSEC_PER_USEC;
328 }
329
330 if (unlikely(tz != NULL)) {
331 if (IS_ENABLED(CONFIG_TIME_NS) &&
332 vd->clock_mode == VDSO_CLOCKMODE_TIMENS)
333 vd = __arch_get_timens_vdso_data(vd);
334
335 tz->tz_minuteswest = vd[CS_HRES_COARSE].tz_minuteswest;
336 tz->tz_dsttime = vd[CS_HRES_COARSE].tz_dsttime;
337 }
338
339 return 0;
340 }
341
342 static __maybe_unused int
343 __cvdso_gettimeofday(struct __kernel_old_timeval *tv, struct timezone *tz)
344 {
345 return __cvdso_gettimeofday_data(__arch_get_vdso_data(), tv, tz);
346 }
347
348 #ifdef VDSO_HAS_TIME
349 static __maybe_unused __kernel_old_time_t
350 __cvdso_time_data(const struct vdso_data *vd, __kernel_old_time_t *time)
351 {
352 __kernel_old_time_t t;
353
354 if (IS_ENABLED(CONFIG_TIME_NS) &&
355 vd->clock_mode == VDSO_CLOCKMODE_TIMENS)
356 vd = __arch_get_timens_vdso_data(vd);
357
358 t = READ_ONCE(vd[CS_HRES_COARSE].basetime[CLOCK_REALTIME].sec);
359
360 if (time)
361 *time = t;
362
363 return t;
364 }
365
366 static __maybe_unused __kernel_old_time_t __cvdso_time(__kernel_old_time_t *time)
367 {
368 return __cvdso_time_data(__arch_get_vdso_data(), time);
369 }
370 #endif /* VDSO_HAS_TIME */
371
372 #ifdef VDSO_HAS_CLOCK_GETRES
373 static __maybe_unused
374 int __cvdso_clock_getres_common(const struct vdso_data *vd, clockid_t clock,
375 struct __kernel_timespec *res)
376 {
377 u32 msk;
378 u64 ns;
379
380 /* Check for negative values or invalid clocks */
381 if (unlikely((u32) clock >= MAX_CLOCKS))
382 return -1;
383
384 if (IS_ENABLED(CONFIG_TIME_NS) &&
385 vd->clock_mode == VDSO_CLOCKMODE_TIMENS)
386 vd = __arch_get_timens_vdso_data(vd);
387
388 /*
389 * Convert the clockid to a bitmask and use it to check which
390 * clocks are handled in the VDSO directly.
391 */
392 msk = 1U << clock;
393 if (msk & (VDSO_HRES | VDSO_RAW)) {
394 /*
395 * Preserves the behaviour of posix_get_hrtimer_res().
396 */
397 ns = READ_ONCE(vd[CS_HRES_COARSE].hrtimer_res);
398 } else if (msk & VDSO_COARSE) {
399 /*
400 * Preserves the behaviour of posix_get_coarse_res().
401 */
402 ns = LOW_RES_NSEC;
403 } else {
404 return -1;
405 }
406
407 if (likely(res)) {
408 res->tv_sec = 0;
409 res->tv_nsec = ns;
410 }
411 return 0;
412 }
413
414 static __maybe_unused
415 int __cvdso_clock_getres_data(const struct vdso_data *vd, clockid_t clock,
416 struct __kernel_timespec *res)
417 {
418 int ret = __cvdso_clock_getres_common(vd, clock, res);
419
420 if (unlikely(ret))
421 return clock_getres_fallback(clock, res);
422 return 0;
423 }
424
425 static __maybe_unused
426 int __cvdso_clock_getres(clockid_t clock, struct __kernel_timespec *res)
427 {
428 return __cvdso_clock_getres_data(__arch_get_vdso_data(), clock, res);
429 }
430
431 #ifdef BUILD_VDSO32
432 static __maybe_unused int
433 __cvdso_clock_getres_time32_data(const struct vdso_data *vd, clockid_t clock,
434 struct old_timespec32 *res)
435 {
436 struct __kernel_timespec ts;
437 int ret;
438
439 ret = __cvdso_clock_getres_common(vd, clock, &ts);
440
441 if (unlikely(ret))
442 return clock_getres32_fallback(clock, res);
443
444 if (likely(res)) {
445 res->tv_sec = ts.tv_sec;
446 res->tv_nsec = ts.tv_nsec;
447 }
448 return ret;
449 }
450
451 static __maybe_unused int
452 __cvdso_clock_getres_time32(clockid_t clock, struct old_timespec32 *res)
453 {
454 return __cvdso_clock_getres_time32_data(__arch_get_vdso_data(),
455 clock, res);
456 }
457 #endif /* BUILD_VDSO32 */
458 #endif /* VDSO_HAS_CLOCK_GETRES */
459
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.