/* $Id: timesupref.h 62477 2016-07-22 18:27:37Z vboxsync $ */ /** @file * IPRT - Time using SUPLib, the C Code Template. */ /* * Copyright (C) 2006-2016 Oracle Corporation * * This file is part of VirtualBox Open Source Edition (OSE), as * available from http://www.virtualbox.org. This file is free software; * you can redistribute it and/or modify it under the terms of the GNU * General Public License (GPL) as published by the Free Software * Foundation, in version 2 as it comes in the "COPYING" file of the * VirtualBox OSE distribution. VirtualBox OSE is distributed in the * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind. * * The contents of this file may alternatively be used under the terms * of the Common Development and Distribution License Version 1.0 * (CDDL) only, as it comes in the "COPYING.CDDL" file of the * VirtualBox OSE distribution, in which case the provisions of the * CDDL are applicable instead of those of the GPL. * * You may elect to license modified versions of this file under the * terms and conditions of either the GPL or the CDDL or both. */ /** * The C reference implementation of the assembly routines. * * Calculate NanoTS using the information in the global information page (GIP) * which the support library (SUPLib) exports. * * This function guarantees that the returned timestamp is later (in time) than * any previous calls in the same thread. * * @remark The way the ever increasing time guarantee is currently implemented means * that if you call this function at a frequency higher than 1GHz you're in for * trouble. We currently assume that no idiot will do that for real life purposes. * * @returns Nanosecond timestamp. * @param pData Pointer to the data structure. */ RTDECL(uint64_t) rtTimeNanoTSInternalRef(PRTTIMENANOTSDATA pData) { #if TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA && defined(IN_RING3) PSUPGIPCPU pGipCpuAttemptedTscRecalibration = NULL; #endif AssertCompile(RT_IS_POWER_OF_TWO(RTCPUSET_MAX_CPUS)); for (;;) { #ifndef IN_RING3 /* This simplifies and improves everything. */ RTCCUINTREG const uFlags = ASMIntDisableFlags(); #endif /* * Check that the GIP is sane and that the premises for this worker function * hasn't changed (CPU onlined with bad delta or missing features). */ PSUPGLOBALINFOPAGE pGip = g_pSUPGlobalInfoPage; if ( RT_LIKELY(pGip) && RT_LIKELY(pGip->u32Magic == SUPGLOBALINFOPAGE_MAGIC) #if TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA && RT_LIKELY(pGip->enmUseTscDelta >= SUPGIPUSETSCDELTA_PRACTICALLY_ZERO) #else && RT_LIKELY(pGip->enmUseTscDelta <= SUPGIPUSETSCDELTA_ROUGHLY_ZERO) #endif #if defined(IN_RING3) && TMPL_GET_CPU_METHOD != 0 && TMPL_GET_CPU_METHOD != SUPGIPGETCPU_APIC_ID && RT_LIKELY(pGip->fGetGipCpu & TMPL_GET_CPU_METHOD) #endif ) { /* * Resolve pGipCpu if needed. If the instruction is serializing, we * read the transaction id first if possible. */ #if TMPL_MODE == TMPL_MODE_ASYNC || TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA # if defined(IN_RING0) uint32_t const iCpuSet = RTMpCurSetIndex(); uint16_t const iGipCpu = iCpuSet < RT_ELEMENTS(pGip->aiCpuFromCpuSetIdx) ? pGip->aiCpuFromCpuSetIdx[iCpuSet] : UINT16_MAX; # elif defined(IN_RC) uint32_t const iCpuSet = VMMGetCpu(&g_VM)->iHostCpuSet; uint16_t const iGipCpu = iCpuSet < RT_ELEMENTS(pGip->aiCpuFromCpuSetIdx) ? pGip->aiCpuFromCpuSetIdx[iCpuSet] : UINT16_MAX; # elif TMPL_GET_CPU_METHOD == SUPGIPGETCPU_APIC_ID # if TMPL_MODE != TMPL_MODE_ASYNC uint32_t const u32TransactionId = pGip->aCPUs[0].u32TransactionId; # endif uint8_t const idApic = ASMGetApicId(); uint16_t const iGipCpu = pGip->aiCpuFromApicId[idApic]; # elif TMPL_GET_CPU_METHOD == SUPGIPGETCPU_RDTSCP_MASK_MAX_SET_CPUS # if TMPL_MODE != TMPL_MODE_ASYNC uint32_t const u32TransactionId = pGip->aCPUs[0].u32TransactionId; # endif uint32_t uAux; ASMReadTscWithAux(&uAux); uint16_t const iCpuSet = uAux & (RTCPUSET_MAX_CPUS - 1); uint16_t const iGipCpu = pGip->aiCpuFromCpuSetIdx[iCpuSet]; # elif TMPL_GET_CPU_METHOD == SUPGIPGETCPU_IDTR_LIMIT_MASK_MAX_SET_CPUS uint16_t const cbLim = ASMGetIdtrLimit(); uint16_t const iCpuSet = (cbLim - 256 * (ARCH_BITS == 64 ? 16 : 8)) & (RTCPUSET_MAX_CPUS - 1); uint16_t const iGipCpu = pGip->aiCpuFromCpuSetIdx[iCpuSet]; # else # error "What?" # endif if (RT_LIKELY(iGipCpu < pGip->cCpus)) { PSUPGIPCPU pGipCpu = &pGip->aCPUs[iGipCpu]; #else { #endif /* * Get the transaction ID if necessary and we haven't already * read it before a serializing instruction above. We can skip * this for ASYNC_TSC mode in ring-0 and raw-mode context since * we disable interrupts. */ #if TMPL_MODE == TMPL_MODE_ASYNC && defined(IN_RING3) uint32_t const u32TransactionId = pGipCpu->u32TransactionId; ASMCompilerBarrier(); TMPL_READ_FENCE(); #elif TMPL_MODE != TMPL_MODE_ASYNC \ && TMPL_GET_CPU_METHOD != SUPGIPGETCPU_APIC_ID \ && TMPL_GET_CPU_METHOD != SUPGIPGETCPU_RDTSCP_MASK_MAX_SET_CPUS uint32_t const u32TransactionId = pGip->aCPUs[0].u32TransactionId; ASMCompilerBarrier(); TMPL_READ_FENCE(); #endif /* * Gather all the data we need. The mess at the end is to make * sure all loads are done before we recheck the transaction ID * without triggering serializing twice. */ uint32_t u32NanoTSFactor0 = pGip->u32UpdateIntervalNS; #if TMPL_MODE == TMPL_MODE_ASYNC uint32_t u32UpdateIntervalTSC = pGipCpu->u32UpdateIntervalTSC; uint64_t u64NanoTS = pGipCpu->u64NanoTS; uint64_t u64TSC = pGipCpu->u64TSC; #else uint32_t u32UpdateIntervalTSC = pGip->aCPUs[0].u32UpdateIntervalTSC; uint64_t u64NanoTS = pGip->aCPUs[0].u64NanoTS; uint64_t u64TSC = pGip->aCPUs[0].u64TSC; # if TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA int64_t i64TscDelta = pGipCpu->i64TSCDelta; # endif #endif uint64_t u64PrevNanoTS = ASMAtomicUoReadU64(pData->pu64Prev); #if TMPL_GET_CPU_METHOD == SUPGIPGETCPU_RDTSCP_MASK_MAX_SET_CPUS ASMCompilerBarrier(); uint32_t uAux2; uint64_t u64Delta = ASMReadTscWithAux(&uAux2); /* serializing */ #else uint64_t u64Delta = ASMReadTSC(); ASMCompilerBarrier(); # if TMPL_GET_CPU_METHOD != SUPGIPGETCPU_APIC_ID /* getting APIC will serialize */ \ && (defined(IN_RING3) || TMPL_MODE != TMPL_MODE_ASYNC) TMPL_READ_FENCE(); /* Expensive (~30 ticks). Would like convincing argumentation that let us remove it. */ # endif #endif /* * Check that we didn't change CPU. */ #if defined(IN_RING3) && ( TMPL_MODE == TMPL_MODE_ASYNC || TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA ) # if TMPL_GET_CPU_METHOD == SUPGIPGETCPU_APIC_ID if (RT_LIKELY(ASMGetApicId() == idApic)) # elif TMPL_GET_CPU_METHOD == SUPGIPGETCPU_RDTSCP_MASK_MAX_SET_CPUS if (RT_LIKELY(uAux2 == uAux)) # elif TMPL_GET_CPU_METHOD == SUPGIPGETCPU_IDTR_LIMIT_MASK_MAX_SET_CPUS if (RT_LIKELY(ASMGetIdtrLimit() == cbLim)) # endif #endif { /* * Check the transaction ID (see above for R0/RC + ASYNC). */ #if defined(IN_RING3) || TMPL_MODE != TMPL_MODE_ASYNC # if TMPL_MODE == TMPL_MODE_ASYNC if (RT_LIKELY(pGipCpu->u32TransactionId == u32TransactionId && !(u32TransactionId & 1) )) # else if (RT_LIKELY(pGip->aCPUs[0].u32TransactionId == u32TransactionId && !(u32TransactionId & 1) )) # endif #endif { /* * Apply the TSC delta. If the delta is invalid and the * execution allows it, try trigger delta recalibration. */ #if TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA && defined(IN_RING3) if (RT_LIKELY( i64TscDelta != INT64_MAX || pGipCpu == pGipCpuAttemptedTscRecalibration)) #endif { #if TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA # ifndef IN_RING3 if (RT_LIKELY(i64TscDelta != INT64_MAX)) # endif u64Delta -= i64TscDelta; #endif /* * Bingo! We've got a consistent set of data. */ #ifndef IN_RING3 ASMSetFlags(uFlags); #endif /* * Calc NanoTS delta. */ u64Delta -= u64TSC; if (RT_LIKELY(u64Delta <= u32UpdateIntervalTSC)) { /* MSVC branch hint, probably pointless. */ } else { /* * We've expired the interval, cap it. If we're here for the 2nd * time without any GIP update in-between, the checks against * *pu64Prev below will force 1ns stepping. */ ASMAtomicIncU32(&pData->cExpired); u64Delta = u32UpdateIntervalTSC; } #if !defined(_MSC_VER) || !defined(RT_ARCH_X86) /* GCC makes very pretty code from these two inline calls, while MSC cannot. */ u64Delta = ASMMult2xU32RetU64((uint32_t)u64Delta, u32NanoTSFactor0); u64Delta = ASMDivU64ByU32RetU32(u64Delta, u32UpdateIntervalTSC); #else __asm { mov eax, dword ptr [u64Delta] mul dword ptr [u32NanoTSFactor0] div dword ptr [u32UpdateIntervalTSC] mov dword ptr [u64Delta], eax xor edx, edx mov dword ptr [u64Delta + 4], edx } #endif /* * Calculate the time and compare it with the previously returned value. */ u64NanoTS += u64Delta; uint64_t u64DeltaPrev = u64NanoTS - u64PrevNanoTS; if (RT_LIKELY( u64DeltaPrev > 0 && u64DeltaPrev < UINT64_C(86000000000000) /* 24h */)) { /* Frequent - less than 24h since last call. */ } else if (RT_LIKELY( (int64_t)u64DeltaPrev <= 0 && (int64_t)u64DeltaPrev + u32NanoTSFactor0 * 2 >= 0)) { /* Occasional - u64NanoTS is in the recent 'past' relative the previous call. */ ASMAtomicIncU32(&pData->c1nsSteps); u64NanoTS = u64PrevNanoTS + 1; } else if (!u64PrevNanoTS) /* We're resuming (see TMVirtualResume). */; else { /* Something has gone bust, if negative offset it's real bad. */ ASMAtomicIncU32(&pData->cBadPrev); pData->pfnBad(pData, u64NanoTS, u64DeltaPrev, u64PrevNanoTS); } /* * Attempt updating the previous value, provided we're still ahead of it. * * There is no point in recalculating u64NanoTS because we got preempted or if * we raced somebody while the GIP was updated, since these are events * that might occur at any point in the return path as well. */ if (RT_LIKELY(ASMAtomicCmpXchgU64(pData->pu64Prev, u64NanoTS, u64PrevNanoTS))) return u64NanoTS; ASMAtomicIncU32(&pData->cUpdateRaces); for (int cTries = 25; cTries > 0; cTries--) { u64PrevNanoTS = ASMAtomicReadU64(pData->pu64Prev); if (u64PrevNanoTS >= u64NanoTS) break; if (ASMAtomicCmpXchgU64(pData->pu64Prev, u64NanoTS, u64PrevNanoTS)) break; ASMNopPause(); } return u64NanoTS; } #if TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA && defined(IN_RING3) /* * Call into the support driver to try make it recalculate the delta. We * remember which GIP CPU structure we're probably working on so we won't * end up in a loop if the driver for some reason cannot get the job done. */ else /* else is unecessary, but helps checking the preprocessor spaghetti. */ { pGipCpuAttemptedTscRecalibration = pGipCpu; uint64_t u64TscTmp; uint16_t idApicUpdate; int rc = SUPR3ReadTsc(&u64TscTmp, &idApicUpdate); if (RT_SUCCESS(rc) && idApicUpdate < RT_ELEMENTS(pGip->aiCpuFromApicId)) { uint32_t iUpdateGipCpu = pGip->aiCpuFromApicId[idApicUpdate]; if (iUpdateGipCpu < pGip->cCpus) pGipCpuAttemptedTscRecalibration = &pGip->aCPUs[iUpdateGipCpu]; } } #endif } } /* * No joy must try again. */ #ifdef _MSC_VER # pragma warning(disable: 4702) #endif #ifndef IN_RING3 ASMSetFlags(uFlags); #endif ASMNopPause(); continue; } #if TMPL_MODE == TMPL_MODE_ASYNC || TMPL_MODE == TMPL_MODE_SYNC_INVAR_WITH_DELTA /* * We've got a bad CPU or APIC index of some kind. */ else /* else is unecessary, but helps checking the preprocessor spaghetti. */ { # ifndef IN_RING3 ASMSetFlags(uFlags); # endif # if defined(IN_RING0) || defined(IN_RC) || TMPL_GET_CPU_METHOD != SUPGIPGETCPU_APIC_ID return pData->pfnBadCpuIndex(pData, UINT16_MAX-1, iCpuSet, iGipCpu); # else return pData->pfnBadCpuIndex(pData, idApic, UINT16_MAX-1, iGipCpu); # endif } #endif } /* * Something changed in the GIP config or it was unmapped, figure out * the right worker function to use now. */ #ifndef IN_RING3 ASMSetFlags(uFlags); #endif return pData->pfnRediscover(pData); } }