asm/vdso/gettimeofday.h

0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 /*
0003  * Fast user context implementation of clock_gettime, gettimeofday, and time.
0004  *
0005  * Copyright (C) 2019 ARM Limited.
0006  * Copyright 2006 Andi Kleen, SUSE Labs.
0007  * 32 Bit compat layer by Stefani Seibold <stefani@seibold.net>
0008  *  sponsored by Rohde & Schwarz GmbH & Co. KG Munich/Germany
0009  */
0010 #ifndef __ASM_VDSO_GETTIMEOFDAY_H
0011 #define __ASM_VDSO_GETTIMEOFDAY_H
0012
0013 #ifndef __ASSEMBLY__
0014
0015 #include <uapi/linux/time.h>
0016 #include <asm/vgtod.h>
0017 #include <asm/vvar.h>
0018 #include <asm/unistd.h>
0019 #include <asm/msr.h>
0020 #include <asm/pvclock.h>
0021 #include <clocksource/hyperv_timer.h>
0022
0023 #define __vdso_data (VVAR(_vdso_data))
0024 #define __timens_vdso_data (TIMENS(_vdso_data))
0025
0026 #define VDSO_HAS_TIME 1
0027
0028 #define VDSO_HAS_CLOCK_GETRES 1
0029
0030 /*
0031  * Declare the memory-mapped vclock data pages.  These come from hypervisors.
0032  * If we ever reintroduce something like direct access to an MMIO clock like
0033  * the HPET again, it will go here as well.
0034  *
0035  * A load from any of these pages will segfault if the clock in question is
0036  * disabled, so appropriate compiler barriers and checks need to be used
0037  * to prevent stray loads.
0038  *
0039  * These declarations MUST NOT be const.  The compiler will assume that
0040  * an extern const variable has genuinely constant contents, and the
0041  * resulting code won't work, since the whole point is that these pages
0042  * change over time, possibly while we're accessing them.
0043  */
0044
0045 #ifdef CONFIG_PARAVIRT_CLOCK
0046 /*
0047  * This is the vCPU 0 pvclock page.  We only use pvclock from the vDSO
0048  * if the hypervisor tells us that all vCPUs can get valid data from the
0049  * vCPU 0 page.
0050  */
0051 extern struct pvclock_vsyscall_time_info pvclock_page
0052     __attribute__((visibility("hidden")));
0053 #endif
0054
0055 #ifdef CONFIG_HYPERV_TIMER
0056 extern struct ms_hyperv_tsc_page hvclock_page
0057     __attribute__((visibility("hidden")));
0058 #endif
0059
0060 #ifdef CONFIG_TIME_NS
0061 static __always_inline
0062 const struct vdso_data *__arch_get_timens_vdso_data(const struct vdso_data *vd)
0063 {
0064     return __timens_vdso_data;
0065 }
0066 #endif
0067
0068 #ifndef BUILD_VDSO32
0069
0070 static __always_inline
0071 long clock_gettime_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
0072 {
0073     long ret;
0074
0075     asm ("syscall" : "=a" (ret), "=m" (*_ts) :
0076          "0" (__NR_clock_gettime), "D" (_clkid), "S" (_ts) :
0077          "rcx", "r11");
0078
0079     return ret;
0080 }
0081
0082 static __always_inline
0083 long gettimeofday_fallback(struct __kernel_old_timeval *_tv,
0084                struct timezone *_tz)
0085 {
0086     long ret;
0087
0088     asm("syscall" : "=a" (ret) :
0089         "0" (__NR_gettimeofday), "D" (_tv), "S" (_tz) : "memory");
0090
0091     return ret;
0092 }
0093
0094 static __always_inline
0095 long clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
0096 {
0097     long ret;
0098
0099     asm ("syscall" : "=a" (ret), "=m" (*_ts) :
0100          "0" (__NR_clock_getres), "D" (_clkid), "S" (_ts) :
0101          "rcx", "r11");
0102
0103     return ret;
0104 }
0105
0106 #else
0107
0108 static __always_inline
0109 long clock_gettime_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
0110 {
0111     long ret;
0112
0113     asm (
0114         "mov %%ebx, %%edx \n"
0115         "mov %[clock], %%ebx \n"
0116         "call __kernel_vsyscall \n"
0117         "mov %%edx, %%ebx \n"
0118         : "=a" (ret), "=m" (*_ts)
0119         : "0" (__NR_clock_gettime64), [clock] "g" (_clkid), "c" (_ts)
0120         : "edx");
0121
0122     return ret;
0123 }
0124
0125 static __always_inline
0126 long clock_gettime32_fallback(clockid_t _clkid, struct old_timespec32 *_ts)
0127 {
0128     long ret;
0129
0130     asm (
0131         "mov %%ebx, %%edx \n"
0132         "mov %[clock], %%ebx \n"
0133         "call __kernel_vsyscall \n"
0134         "mov %%edx, %%ebx \n"
0135         : "=a" (ret), "=m" (*_ts)
0136         : "0" (__NR_clock_gettime), [clock] "g" (_clkid), "c" (_ts)
0137         : "edx");
0138
0139     return ret;
0140 }
0141
0142 static __always_inline
0143 long gettimeofday_fallback(struct __kernel_old_timeval *_tv,
0144                struct timezone *_tz)
0145 {
0146     long ret;
0147
0148     asm(
0149         "mov %%ebx, %%edx \n"
0150         "mov %2, %%ebx \n"
0151         "call __kernel_vsyscall \n"
0152         "mov %%edx, %%ebx \n"
0153         : "=a" (ret)
0154         : "0" (__NR_gettimeofday), "g" (_tv), "c" (_tz)
0155         : "memory", "edx");
0156
0157     return ret;
0158 }
0159
0160 static __always_inline long
0161 clock_getres_fallback(clockid_t _clkid, struct __kernel_timespec *_ts)
0162 {
0163     long ret;
0164
0165     asm (
0166         "mov %%ebx, %%edx \n"
0167         "mov %[clock], %%ebx \n"
0168         "call __kernel_vsyscall \n"
0169         "mov %%edx, %%ebx \n"
0170         : "=a" (ret), "=m" (*_ts)
0171         : "0" (__NR_clock_getres_time64), [clock] "g" (_clkid), "c" (_ts)
0172         : "edx");
0173
0174     return ret;
0175 }
0176
0177 static __always_inline
0178 long clock_getres32_fallback(clockid_t _clkid, struct old_timespec32 *_ts)
0179 {
0180     long ret;
0181
0182     asm (
0183         "mov %%ebx, %%edx \n"
0184         "mov %[clock], %%ebx \n"
0185         "call __kernel_vsyscall \n"
0186         "mov %%edx, %%ebx \n"
0187         : "=a" (ret), "=m" (*_ts)
0188         : "0" (__NR_clock_getres), [clock] "g" (_clkid), "c" (_ts)
0189         : "edx");
0190
0191     return ret;
0192 }
0193
0194 #endif
0195
0196 #ifdef CONFIG_PARAVIRT_CLOCK
0197 static u64 vread_pvclock(void)
0198 {
0199     const struct pvclock_vcpu_time_info *pvti = &pvclock_page.pvti;
0200     u32 version;
0201     u64 ret;
0202
0203     /*
0204      * Note: The kernel and hypervisor must guarantee that cpu ID
0205      * number maps 1:1 to per-CPU pvclock time info.
0206      *
0207      * Because the hypervisor is entirely unaware of guest userspace
0208      * preemption, it cannot guarantee that per-CPU pvclock time
0209      * info is updated if the underlying CPU changes or that that
0210      * version is increased whenever underlying CPU changes.
0211      *
0212      * On KVM, we are guaranteed that pvti updates for any vCPU are
0213      * atomic as seen by *all* vCPUs.  This is an even stronger
0214      * guarantee than we get with a normal seqlock.
0215      *
0216      * On Xen, we don't appear to have that guarantee, but Xen still
0217      * supplies a valid seqlock using the version field.
0218      *
0219      * We only do pvclock vdso timing at all if
0220      * PVCLOCK_TSC_STABLE_BIT is set, and we interpret that bit to
0221      * mean that all vCPUs have matching pvti and that the TSC is
0222      * synced, so we can just look at vCPU 0's pvti.
0223      */
0224
0225     do {
0226         version = pvclock_read_begin(pvti);
0227
0228         if (unlikely(!(pvti->flags & PVCLOCK_TSC_STABLE_BIT)))
0229             return U64_MAX;
0230
0231         ret = __pvclock_read_cycles(pvti, rdtsc_ordered());
0232     } while (pvclock_read_retry(pvti, version));
0233
0234     return ret;
0235 }
0236 #endif
0237
0238 #ifdef CONFIG_HYPERV_TIMER
0239 static u64 vread_hvclock(void)
0240 {
0241     return hv_read_tsc_page(&hvclock_page);
0242 }
0243 #endif
0244
0245 static inline u64 __arch_get_hw_counter(s32 clock_mode,
0246                     const struct vdso_data *vd)
0247 {
0248     if (likely(clock_mode == VDSO_CLOCKMODE_TSC))
0249         return (u64)rdtsc_ordered();
0250     /*
0251      * For any memory-mapped vclock type, we need to make sure that gcc
0252      * doesn't cleverly hoist a load before the mode check.  Otherwise we
0253      * might end up touching the memory-mapped page even if the vclock in
0254      * question isn't enabled, which will segfault.  Hence the barriers.
0255      */
0256 #ifdef CONFIG_PARAVIRT_CLOCK
0257     if (clock_mode == VDSO_CLOCKMODE_PVCLOCK) {
0258         barrier();
0259         return vread_pvclock();
0260     }
0261 #endif
0262 #ifdef CONFIG_HYPERV_TIMER
0263     if (clock_mode == VDSO_CLOCKMODE_HVCLOCK) {
0264         barrier();
0265         return vread_hvclock();
0266     }
0267 #endif
0268     return U64_MAX;
0269 }
0270
0271 static __always_inline const struct vdso_data *__arch_get_vdso_data(void)
0272 {
0273     return __vdso_data;
0274 }
0275
0276 static inline bool arch_vdso_clocksource_ok(const struct vdso_data *vd)
0277 {
0278     return true;
0279 }
0280 #define vdso_clocksource_ok arch_vdso_clocksource_ok
0281
0282 /*
0283  * Clocksource read value validation to handle PV and HyperV clocksources
0284  * which can be invalidated asynchronously and indicate invalidation by
0285  * returning U64_MAX, which can be effectively tested by checking for a
0286  * negative value after casting it to s64.
0287  */
0288 static inline bool arch_vdso_cycles_ok(u64 cycles)
0289 {
0290     return (s64)cycles >= 0;
0291 }
0292 #define vdso_cycles_ok arch_vdso_cycles_ok
0293
0294 /*
0295  * x86 specific delta calculation.
0296  *
0297  * The regular implementation assumes that clocksource reads are globally
0298  * monotonic. The TSC can be slightly off across sockets which can cause
0299  * the regular delta calculation (@cycles - @last) to return a huge time
0300  * jump.
0301  *
0302  * Therefore it needs to be verified that @cycles are greater than
0303  * @last. If not then use @last, which is the base time of the current
0304  * conversion period.
0305  *
0306  * This variant also removes the masking of the subtraction because the
0307  * clocksource mask of all VDSO capable clocksources on x86 is U64_MAX
0308  * which would result in a pointless operation. The compiler cannot
0309  * optimize it away as the mask comes from the vdso data and is not compile
0310  * time constant.
0311  */
0312 static __always_inline
0313 u64 vdso_calc_delta(u64 cycles, u64 last, u64 mask, u32 mult)
0314 {
0315     if (cycles > last)
0316         return (cycles - last) * mult;
0317     return 0;
0318 }
0319 #define vdso_calc_delta vdso_calc_delta
0320
0321 #endif /* !__ASSEMBLY__ */
0322
0323 #endif /* __ASM_VDSO_GETTIMEOFDAY_H */