/* $Id: tstTSC.cpp $ */ /** @file * IPRT Testcase - SMP TSC testcase. */ /* * Copyright (C) 2006-2022 Oracle and/or its affiliates. * * This file is part of VirtualBox base platform packages, as * available from https://www.virtualbox.org. * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public License * as published by the Free Software Foundation, in version 3 of the * License. * * This program is distributed in the hope that it will be useful, but * WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, see . * * The contents of this file may alternatively be used under the terms * of the Common Development and Distribution License Version 1.0 * (CDDL), a copy of it is provided in the "COPYING.CDDL" file included * in the VirtualBox 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. * * SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0 */ /********************************************************************************************************************************* * Header Files * *********************************************************************************************************************************/ #include #include #include #include #include #include #include #include #include /********************************************************************************************************************************* * Structures and Typedefs * *********************************************************************************************************************************/ typedef struct TSCDATA { /** The TSC. */ uint64_t volatile TSC; /** The APIC ID. */ uint8_t volatile u8ApicId; /** Did it succeed? */ bool volatile fRead; /** Did it fail? */ bool volatile fFailed; /** The thread handle. */ RTTHREAD Thread; } TSCDATA, *PTSCDATA; /********************************************************************************************************************************* * Global Variables * *********************************************************************************************************************************/ /** The number of CPUs waiting on their user event semaphore. */ static volatile uint32_t g_cWaiting; /** The number of CPUs ready (in spin) to do the TSC read. */ static volatile uint32_t g_cReady; /** The variable the CPUs are spinning on. * 0: Spin. * 1: Go ahead. * 2: You're too late, back to square one. */ static volatile uint32_t g_u32Go; /** The number of CPUs that managed to read the TSC. */ static volatile uint32_t g_cRead; /** The number of CPUs that failed to read the TSC. */ static volatile uint32_t g_cFailed; /** Indicator forcing the threads to quit. */ static volatile bool g_fDone; /********************************************************************************************************************************* * Internal Functions * *********************************************************************************************************************************/ static DECLCALLBACK(int) ThreadFunction(RTTHREAD Thread, void *pvUser); /** * Thread function for catching the other cpus. * * @returns VINF_SUCCESS (we don't care). * @param Thread The thread handle. * @param pvUser PTSCDATA. */ static DECLCALLBACK(int) ThreadFunction(RTTHREAD Thread, void *pvUser) { PTSCDATA pTscData = (PTSCDATA)pvUser; while (!g_fDone) { /* * Wait. */ ASMAtomicIncU32(&g_cWaiting); RTThreadUserWait(Thread, RT_INDEFINITE_WAIT); RTThreadUserReset(Thread); ASMAtomicDecU32(&g_cWaiting); if (g_fDone) break; /* * Spin. */ ASMAtomicIncU32(&g_cReady); while (!g_fDone) { const uint8_t ApicId1 = ASMGetApicId(); const uint64_t TSC1 = ASMReadTSC(); const uint32_t u32Go = g_u32Go; if (u32Go == 0) continue; if (u32Go == 1) { /* do the reading. */ const uint8_t ApicId2 = ASMGetApicId(); const uint64_t TSC2 = ASMReadTSC(); const uint8_t ApicId3 = ASMGetApicId(); const uint64_t TSC3 = ASMReadTSC(); const uint8_t ApicId4 = ASMGetApicId(); if ( ApicId1 == ApicId2 && ApicId1 == ApicId3 && ApicId1 == ApicId4 && TSC3 - TSC1 < 2250 /* WARNING: This is just a guess, increase if it doesn't work for you. */ && TSC2 - TSC1 < TSC3 - TSC1 ) { /* succeeded. */ pTscData->TSC = TSC2; pTscData->u8ApicId = ApicId1; pTscData->fFailed = false; pTscData->fRead = true; ASMAtomicIncU32(&g_cRead); break; } } /* failed */ pTscData->fFailed = true; pTscData->fRead = false; ASMAtomicIncU32(&g_cFailed); break; } } return VINF_SUCCESS; } static int tstTSCCalcDrift(void) { /* * This is only relevant to on SMP systems. */ const unsigned cCpus = RTMpGetOnlineCount(); if (cCpus <= 1) { RTPrintf("tstTSC: SKIPPED - Only relevant on SMP systems\n"); return 0; } /* * Create the threads. */ static TSCDATA s_aData[254]; uint32_t i; if (cCpus > RT_ELEMENTS(s_aData)) { RTPrintf("tstTSC: FAILED - too many CPUs (%u)\n", cCpus); return 1; } /* ourselves. */ s_aData[0].Thread = RTThreadSelf(); /* the others */ for (i = 1; i < cCpus; i++) { int rc = RTThreadCreate(&s_aData[i].Thread, ThreadFunction, &s_aData[i], 0, RTTHREADTYPE_TIMER, RTTHREADFLAGS_WAITABLE, "OTHERCPU"); if (RT_FAILURE(rc)) { RTPrintf("tstTSC: FAILURE - RTThreatCreate failed when creating thread #%u, rc=%Rrc!\n", i, rc); ASMAtomicXchgSize(&g_fDone, true); while (i-- > 1) { RTThreadUserSignal(s_aData[i].Thread); RTThreadWait(s_aData[i].Thread, 5000, NULL); } return 1; } } /* * Retry until we get lucky (or give up). */ for (unsigned cTries = 0; ; cTries++) { if (cTries > 10240) { RTPrintf("tstTSC: FAILURE - %d attempts, giving.\n", cTries); break; } /* * Wait for the other threads to get ready (brute force active wait, I'm lazy). */ i = 0; while (g_cWaiting < cCpus - 1) { if (i++ > _2G32) break; RTThreadSleep(i & 0xf); } if (g_cWaiting != cCpus - 1) { RTPrintf("tstTSC: FAILURE - threads failed to get waiting (%d != %d (i=%d))\n", g_cWaiting + 1, cCpus, i); break; } /* * Send them spinning. */ ASMAtomicXchgU32(&g_cReady, 0); ASMAtomicXchgU32(&g_u32Go, 0); ASMAtomicXchgU32(&g_cRead, 0); ASMAtomicXchgU32(&g_cFailed, 0); for (i = 1; i < cCpus; i++) { ASMAtomicXchgSize(&s_aData[i].fFailed, false); ASMAtomicXchgSize(&s_aData[i].fRead, false); ASMAtomicXchgU8(&s_aData[i].u8ApicId, 0xff); int rc = RTThreadUserSignal(s_aData[i].Thread); if (RT_FAILURE(rc)) RTPrintf("tstTSC: WARNING - RTThreadUserSignal(%#u) -> rc=%Rrc!\n", i, rc); } /* wait for them to get ready. */ i = 0; while (g_cReady < cCpus - 1) { if (i++ > _2G32) break; } if (g_cReady != cCpus - 1) { RTPrintf("tstTSC: FAILURE - threads failed to get ready (%d != %d, i=%d)\n", g_cWaiting + 1, cCpus, i); break; } /* * Flip the "go" switch and do our readings. * We give the other threads the slack it takes to two extra TSC and APIC ID reads. */ const uint8_t ApicId1 = ASMGetApicId(); const uint64_t TSC1 = ASMReadTSC(); ASMAtomicXchgU32(&g_u32Go, 1); const uint8_t ApicId2 = ASMGetApicId(); const uint64_t TSC2 = ASMReadTSC(); const uint8_t ApicId3 = ASMGetApicId(); const uint64_t TSC3 = ASMReadTSC(); const uint8_t ApicId4 = ASMGetApicId(); const uint64_t TSC4 = ASMReadTSC(); ASMAtomicXchgU32(&g_u32Go, 2); const uint8_t ApicId5 = ASMGetApicId(); const uint64_t TSC5 = ASMReadTSC(); const uint8_t ApicId6 = ASMGetApicId(); /* Compose our own result. */ if ( ApicId1 == ApicId2 && ApicId1 == ApicId3 && ApicId1 == ApicId4 && ApicId1 == ApicId5 && ApicId1 == ApicId6 && TSC5 - TSC1 < 2750 /* WARNING: This is just a guess, increase if it doesn't work for you. */ && TSC4 - TSC1 < TSC5 - TSC1 && TSC3 - TSC1 < TSC4 - TSC1 && TSC2 - TSC1 < TSC3 - TSC1 ) { /* succeeded. */ s_aData[0].TSC = TSC2; s_aData[0].u8ApicId = ApicId1; s_aData[0].fFailed = false; s_aData[0].fRead = true; ASMAtomicIncU32(&g_cRead); } else { /* failed */ s_aData[0].fFailed = true; s_aData[0].fRead = false; ASMAtomicIncU32(&g_cFailed); } /* * Wait a little while to let the other ones to finish. */ i = 0; while (g_cRead + g_cFailed < cCpus) { if (i++ > _2G32) break; if (i > _1M) RTThreadSleep(i & 0xf); } if (g_cRead + g_cFailed != cCpus) { RTPrintf("tstTSC: FAILURE - threads failed to complete reading (%d + %d != %d)\n", g_cRead, g_cFailed, cCpus); break; } /* * If everone succeeded, print the results. */ if (!g_cFailed) { /* sort it by apic id first. */ bool fDone; do { for (i = 1, fDone = true; i < cCpus; i++) if (s_aData[i - 1].u8ApicId > s_aData[i].u8ApicId) { TSCDATA Tmp = s_aData[i - 1]; s_aData[i - 1] = s_aData[i]; s_aData[i] = Tmp; fDone = false; } } while (!fDone); RTPrintf(" # ID TSC delta0 (decimal)\n" "-----------------------------------------\n"); RTPrintf("%2d %02x %RX64\n", 0, s_aData[0].u8ApicId, s_aData[0].TSC); for (i = 1; i < cCpus; i++) RTPrintf("%2d %02x %RX64 %s%lld\n", i, s_aData[i].u8ApicId, s_aData[i].TSC, s_aData[i].TSC > s_aData[0].TSC ? "+" : "", s_aData[i].TSC - s_aData[0].TSC); RTPrintf("(Needed %u attempt%s.)\n", cTries + 1, cTries ? "s" : ""); break; } } /* * Destroy the threads. */ ASMAtomicXchgSize(&g_fDone, true); for (i = 0; i < cCpus; i++) if (s_aData[i].Thread != RTThreadSelf()) { int rc = RTThreadUserSignal(s_aData[i].Thread); if (RT_FAILURE(rc)) RTPrintf("tstTSC: WARNING - RTThreadUserSignal(%#u) -> rc=%Rrc! (2)\n", i, rc); } for (i = 0; i < cCpus; i++) if (s_aData[i].Thread != RTThreadSelf()) { int rc = RTThreadWait(s_aData[i].Thread, 5000, NULL); if (RT_FAILURE(rc)) RTPrintf("tstTSC: WARNING - RTThreadWait(%#u) -> rc=%Rrc!\n", i, rc); } return g_cFailed != 0 || g_cRead != cCpus; } static int tstTSCCalcFrequency(uint32_t cMsDuration) { /* * Sample the TSC and time, sleep the requested time and calc the deltas. */ uint64_t uNanoTS = RTTimeSystemNanoTS(); uint64_t uTSC = ASMReadTSC(); RTThreadSleep(cMsDuration); uNanoTS = RTTimeSystemNanoTS() - uNanoTS; uTSC = ASMReadTSC() - uTSC; /* * Calc the frequency. */ RTPrintf("tstTSC: %RU64 ticks in %RU64 ns\n", uTSC, uNanoTS); uint64_t cHz = (uint64_t)((long double)uTSC / ((long double)uNanoTS / (long double)1000000000)); RTPrintf("tstTSC: Frequency %RU64 Hz", cHz); if (cHz > _1G) { cHz += _1G / 20; RTPrintf(" %RU64.%RU64 GHz", cHz / _1G, (cHz % _1G) / (_1G / 10)); } else if (cHz > _1M) { cHz += _1M / 20; RTPrintf(" %RU64.%RU64 MHz", cHz / _1M, (cHz % _1M) / (_1M / 10)); } RTPrintf("\n"); return 0; } int main(int argc, char **argv) { RTR3InitExe(argc, &argv, 0); /* * Parse arguments. */ bool fCalcFrequency = false; uint32_t cMsDuration = 1000; /* 1 sec */ static const RTGETOPTDEF s_aOptions[] = { { "--duration", 'd', RTGETOPT_REQ_UINT32 }, { "--calc-frequency", 'f', RTGETOPT_REQ_NOTHING }, }; int ch; RTGETOPTUNION Value; RTGETOPTSTATE GetState; RTGetOptInit(&GetState, argc, argv, s_aOptions, RT_ELEMENTS(s_aOptions), 1, 0 /* fFlags */); while ((ch = RTGetOpt(&GetState, &Value))) switch (ch) { case 'd': cMsDuration = Value.u32; break; case 'f': fCalcFrequency = true; break; case 'h': RTPrintf("usage: tstTSC\n" " or: tstTSC <-f|--calc-frequency> [--duration|-d ms]\n"); return 1; case 'V': RTPrintf("$Revision: 153224 $\n"); return 0; default: return RTGetOptPrintError(ch, &Value); } if (fCalcFrequency) return tstTSCCalcFrequency(cMsDuration); return tstTSCCalcDrift(); }