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 */ 28 struct 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 37 static u8 valid_flags __read_mostly = 0; 38 39 void 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 */ 48 static 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 82 static u64 pvclock_get_nsec_offset(struct pvclock_shadow_time *shadow) 83 { 84 u64 delta = native_read_tsc() - shadow->tsc_timestamp; 85 return pvclock_scale_delta(delta, shadow->tsc_to_nsec_mul, 86 shadow->tsc_shift); 87 } 88 89 /* 90 * Reads a consistent set of time-base values from hypervisor, 91 * into a shadow data area. 92 */ 93 static unsigned pvclock_get_time_values(struct pvclock_shadow_time *dst, 94 struct pvclock_vcpu_time_info *src) 95 { 96 do { 97 dst->version = src->version; 98 rmb(); /* fetch version before data */ 99 dst->tsc_timestamp = src->tsc_timestamp; 100 dst->system_timestamp = src->system_time; 101 dst->tsc_to_nsec_mul = src->tsc_to_system_mul; 102 dst->tsc_shift = src->tsc_shift; 103 dst->flags = src->flags; 104 rmb(); /* test version after fetching data */ 105 } while ((src->version & 1) || (dst->version != src->version)); 106 107 return dst->version; 108 } 109 110 unsigned long pvclock_tsc_khz(struct pvclock_vcpu_time_info *src) 111 { 112 u64 pv_tsc_khz = 1000000ULL << 32; 113 114 do_div(pv_tsc_khz, src->tsc_to_system_mul); 115 if (src->tsc_shift < 0) 116 pv_tsc_khz <<= -src->tsc_shift; 117 else 118 pv_tsc_khz >>= src->tsc_shift; 119 return pv_tsc_khz; 120 } 121 122 static atomic64_t last_value = ATOMIC64_INIT(0); 123 124 cycle_t pvclock_clocksource_read(struct pvclock_vcpu_time_info *src) 125 { 126 struct pvclock_shadow_time shadow; 127 unsigned version; 128 cycle_t ret, offset; 129 u64 last; 130 131 do { 132 version = pvclock_get_time_values(&shadow, src); 133 barrier(); 134 offset = pvclock_get_nsec_offset(&shadow); 135 ret = shadow.system_timestamp + offset; 136 barrier(); 137 } while (version != src->version); 138 139 if ((valid_flags & PVCLOCK_TSC_STABLE_BIT) && 140 (shadow.flags & PVCLOCK_TSC_STABLE_BIT)) 141 return ret; 142 143 /* 144 * Assumption here is that last_value, a global accumulator, always goes 145 * forward. If we are less than that, we should not be much smaller. 146 * We assume there is an error marging we're inside, and then the correction 147 * does not sacrifice accuracy. 148 * 149 * For reads: global may have changed between test and return, 150 * but this means someone else updated poked the clock at a later time. 151 * We just need to make sure we are not seeing a backwards event. 152 * 153 * For updates: last_value = ret is not enough, since two vcpus could be 154 * updating at the same time, and one of them could be slightly behind, 155 * making the assumption that last_value always go forward fail to hold. 156 */ 157 last = atomic64_read(&last_value); 158 do { 159 if (ret < last) 160 return last; 161 last = atomic64_cmpxchg(&last_value, last, ret); 162 } while (unlikely(last != ret)); 163 164 return ret; 165 } 166 167 void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock, 168 struct pvclock_vcpu_time_info *vcpu_time, 169 struct timespec *ts) 170 { 171 u32 version; 172 u64 delta; 173 struct timespec now; 174 175 /* get wallclock at system boot */ 176 do { 177 version = wall_clock->version; 178 rmb(); /* fetch version before time */ 179 now.tv_sec = wall_clock->sec; 180 now.tv_nsec = wall_clock->nsec; 181 rmb(); /* fetch time before checking version */ 182 } while ((wall_clock->version & 1) || (version != wall_clock->version)); 183 184 delta = pvclock_clocksource_read(vcpu_time); /* time since system boot */ 185 delta += now.tv_sec * (u64)NSEC_PER_SEC + now.tv_nsec; 186 187 now.tv_nsec = do_div(delta, NSEC_PER_SEC); 188 now.tv_sec = delta; 189 190 set_normalized_timespec(ts, now.tv_sec, now.tv_nsec); 191 } 192