1/* paravirtual clock -- common code used by kvm/xen 2 3 This program is free software; you can redistribute it and/or modify 4 it under the terms of the GNU General Public License as published by 5 the Free Software Foundation; either version 2 of the License, or 6 (at your option) any later version. 7 8 This program is distributed in the hope that it will be useful, 9 but WITHOUT ANY WARRANTY; without even the implied warranty of 10 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 11 GNU General Public License for more details. 12 13 You should have received a copy of the GNU General Public License 14 along with this program; if not, write to the Free Software 15 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 16*/ 17 18#include <linux/kernel.h> 19#include <linux/percpu.h> 20#include <asm/pvclock.h> 21 22/* 23 * These are perodically updated 24 * xen: magic shared_info page 25 * kvm: gpa registered via msr 26 * and then copied here. 27 */ 28struct pvclock_shadow_time { 29 u64 tsc_timestamp; /* TSC at last update of time vals. */ 30 u64 system_timestamp; /* Time, in nanosecs, since boot. */ 31 u32 tsc_to_nsec_mul; 32 int tsc_shift; 33 u32 version; 34 u8 flags; 35}; 36 37static u8 valid_flags __read_mostly = 0; 38 39void pvclock_set_flags(u8 flags) 40{ 41 valid_flags = flags; 42} 43 44/* 45 * Scale a 64-bit delta by scaling and multiplying by a 32-bit fraction, 46 * yielding a 64-bit result. 47 */ 48static inline u64 scale_delta(u64 delta, u32 mul_frac, int shift) 49{ 50 u64 product; 51#ifdef __i386__ 52 u32 tmp1, tmp2; 53#endif 54 55 if (shift < 0) 56 delta >>= -shift; 57 else 58 delta <<= shift; 59 60#ifdef __i386__ 61 __asm__ ( 62 "mul %5 ; " 63 "mov %4,%%eax ; " 64 "mov %%edx,%4 ; " 65 "mul %5 ; " 66 "xor %5,%5 ; " 67 "add %4,%%eax ; " 68 "adc %5,%%edx ; " 69 : "=A" (product), "=r" (tmp1), "=r" (tmp2) 70 : "a" ((u32)delta), "1" ((u32)(delta >> 32)), "2" (mul_frac) ); 71#elif defined(__x86_64__) 72 __asm__ ( 73 "mul %%rdx ; shrd $32,%%rdx,%%rax" 74 : "=a" (product) : "0" (delta), "d" ((u64)mul_frac) ); 75#else 76#error implement me! 77#endif 78 79 return product; 80} 81 82static u64 pvclock_get_nsec_offset(struct pvclock_shadow_time *shadow) 83{ 84 u64 delta = native_read_tsc() - shadow->tsc_timestamp; 85 return scale_delta(delta, shadow->tsc_to_nsec_mul, shadow->tsc_shift); 86} 87 88/* 89 * Reads a consistent set of time-base values from hypervisor, 90 * into a shadow data area. 91 */ 92static unsigned pvclock_get_time_values(struct pvclock_shadow_time *dst, 93 struct pvclock_vcpu_time_info *src) 94{ 95 do { 96 dst->version = src->version; 97 rmb(); /* fetch version before data */ 98 dst->tsc_timestamp = src->tsc_timestamp; 99 dst->system_timestamp = src->system_time; 100 dst->tsc_to_nsec_mul = src->tsc_to_system_mul; 101 dst->tsc_shift = src->tsc_shift; 102 dst->flags = src->flags; 103 rmb(); /* test version after fetching data */ 104 } while ((src->version & 1) || (dst->version != src->version)); 105 106 return dst->version; 107} 108 109unsigned long pvclock_tsc_khz(struct pvclock_vcpu_time_info *src) 110{ 111 u64 pv_tsc_khz = 1000000ULL << 32; 112 113 do_div(pv_tsc_khz, src->tsc_to_system_mul); 114 if (src->tsc_shift < 0) 115 pv_tsc_khz <<= -src->tsc_shift; 116 else 117 pv_tsc_khz >>= src->tsc_shift; 118 return pv_tsc_khz; 119} 120 121static atomic64_t last_value = ATOMIC64_INIT(0); 122 123cycle_t pvclock_clocksource_read(struct pvclock_vcpu_time_info *src) 124{ 125 struct pvclock_shadow_time shadow; 126 unsigned version; 127 cycle_t ret, offset; 128 u64 last; 129 130 do { 131 version = pvclock_get_time_values(&shadow, src); 132 barrier(); 133 offset = pvclock_get_nsec_offset(&shadow); 134 ret = shadow.system_timestamp + offset; 135 barrier(); 136 } while (version != src->version); 137 138 if ((valid_flags & PVCLOCK_TSC_STABLE_BIT) && 139 (shadow.flags & PVCLOCK_TSC_STABLE_BIT)) 140 return ret; 141 142 /* 143 * Assumption here is that last_value, a global accumulator, always goes 144 * forward. If we are less than that, we should not be much smaller. 145 * We assume there is an error marging we're inside, and then the correction 146 * does not sacrifice accuracy. 147 * 148 * For reads: global may have changed between test and return, 149 * but this means someone else updated poked the clock at a later time. 150 * We just need to make sure we are not seeing a backwards event. 151 * 152 * For updates: last_value = ret is not enough, since two vcpus could be 153 * updating at the same time, and one of them could be slightly behind, 154 * making the assumption that last_value always go forward fail to hold. 155 */ 156 last = atomic64_read(&last_value); 157 do { 158 if (ret < last) 159 return last; 160 last = atomic64_cmpxchg(&last_value, last, ret); 161 } while (unlikely(last != ret)); 162 163 return ret; 164} 165 166void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock, 167 struct pvclock_vcpu_time_info *vcpu_time, 168 struct timespec *ts) 169{ 170 u32 version; 171 u64 delta; 172 struct timespec now; 173 174 /* get wallclock at system boot */ 175 do { 176 version = wall_clock->version; 177 rmb(); /* fetch version before time */ 178 now.tv_sec = wall_clock->sec; 179 now.tv_nsec = wall_clock->nsec; 180 rmb(); /* fetch time before checking version */ 181 } while ((wall_clock->version & 1) || (version != wall_clock->version)); 182 183 delta = pvclock_clocksource_read(vcpu_time); /* time since system boot */ 184 delta += now.tv_sec * (u64)NSEC_PER_SEC + now.tv_nsec; 185 186 now.tv_nsec = do_div(delta, NSEC_PER_SEC); 187 now.tv_sec = delta; 188 189 set_normalized_timespec(ts, now.tv_sec, now.tv_nsec); 190} 191