1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 #include <sys/types.h> 30 #include <sys/clock.h> 31 #include <sys/panic.h> 32 #include <sys/atomic.h> 33 #include <sys/hypervisor.h> 34 35 #include <sys/archsystm.h> 36 37 /* 38 * On the hypervisor, we have a virtualized system time based upon the 39 * information provided for each VCPU, which is updated every time it is 40 * scheduled onto a real CPU. Thus, none of the traditional code in 41 * i86pc/os/timestamp.c applies, our gethrtime() implementation is run through 42 * the PSM, and there is no scaling step to apply. 43 * 44 * However, the platform does not guarantee monotonicity; thus we have to fake 45 * this up, which is a deeply unpleasant thing to have to do. 46 * 47 * Note that the virtualized interface still relies on the current TSC to 48 * calculate the time in nanoseconds since the VCPU was scheduled, and is thus 49 * subject to all the problems with that. For the most part, the hypervisor is 50 * supposed to deal with them. 51 * 52 * Another wrinkle involves suspend/resume/migration. If we come back and time 53 * is apparently less, we may have resumed on a different machine or on the 54 * same machine after a reboot. In this case we need to maintain an addend to 55 * ensure time continues reasonably. Otherwise we could end up taking a very 56 * long time to expire cyclics in the heap. Thus we have two functions: 57 * 58 * xpv_getsystime() 59 * 60 * The unadulterated system time from the hypervisor. This is only to be 61 * used when programming the hypervisor (setting a timer or calculating 62 * the TOD). 63 * 64 * xpv_gethrtime() 65 * 66 * This is the monotonic hrtime counter to be used by everything else such 67 * as the cyclic subsystem. We should never pass an hrtime directly into 68 * a hypervisor interface, as hrtime_addend may well be non-zero. 69 */ 70 71 static volatile hrtime_t hrtime_last; 72 static int hrtime_fake_mt = 1; 73 static hrtime_t hrtime_suspend_time; 74 static hrtime_t hrtime_addend; 75 76 /* 77 * These functions are used in DTrace probe context, and must be removed from 78 * fbt consideration. Currently fbt ignores all weak symbols, so this will 79 * achieve that. 80 */ 81 #pragma weak xpv_gethrtime = dtrace_xpv_gethrtime 82 #pragma weak xpv_getsystime = dtrace_xpv_getsystime 83 #pragma weak dtrace_gethrtime = dtrace_xpv_gethrtime 84 #pragma weak tsc_read = dtrace_xpv_gethrtime 85 86 hrtime_t 87 dtrace_xpv_getsystime(void) 88 { 89 vcpu_time_info_t *src; 90 vcpu_time_info_t __vti, *dst = &__vti; 91 uint64_t tsc_delta; 92 kthread_t *t = curthread; 93 uint64_t tsc; 94 hrtime_t result; 95 96 /* 97 * This stops us from wandering off the virtual cpu. 98 */ 99 t->t_preempt++; 100 101 src = &CPU->cpu_m.mcpu_vcpu_info->time; 102 103 /* 104 * Loop until version has not been changed during our update, and a Xen 105 * update is not under way (lowest bit is set). 106 */ 107 do { 108 dst->version = src->version; 109 110 membar_consumer(); 111 112 dst->tsc_timestamp = src->tsc_timestamp; 113 dst->system_time = src->system_time; 114 dst->tsc_to_system_mul = src->tsc_to_system_mul; 115 dst->tsc_shift = src->tsc_shift; 116 117 /* 118 * Note that this use of the -actual- TSC register 119 * should probably be the SOLE one in the system on this 120 * paravirtualized platform. 121 */ 122 tsc = __rdtsc_insn(); 123 tsc_delta = tsc - dst->tsc_timestamp; 124 125 membar_consumer(); 126 127 } while ((src->version & 1) | (dst->version ^ src->version)); 128 129 if (dst->tsc_shift >= 0) 130 tsc_delta <<= dst->tsc_shift; 131 else if (dst->tsc_shift < 0) 132 tsc_delta >>= -dst->tsc_shift; 133 134 result = dst->system_time + 135 ((uint64_t)(tsc_delta * (uint64_t)dst->tsc_to_system_mul) >> 32); 136 137 t->t_preempt--; 138 139 return (result); 140 } 141 142 hrtime_t 143 dtrace_xpv_gethrtime(void) 144 { 145 hrtime_t result = xpv_getsystime() + hrtime_addend; 146 147 if (hrtime_fake_mt) { 148 hrtime_t last; 149 do { 150 last = hrtime_last; 151 if (result < last) 152 result = last + 1; 153 } while (atomic_cas_64((volatile uint64_t *)&hrtime_last, 154 last, result) != last); 155 } 156 157 return (result); 158 } 159 160 void 161 xpv_time_suspend(void) 162 { 163 hrtime_suspend_time = xpv_getsystime(); 164 } 165 166 void 167 xpv_time_resume(void) 168 { 169 hrtime_t delta = xpv_getsystime() - hrtime_suspend_time; 170 171 if (delta < 0) 172 hrtime_addend += -delta; 173 } 174