/* $Id: VBoxServiceCpuHotPlug.cpp 62521 2016-07-22 19:16:33Z vboxsync $ */ /** @file * VBoxService - Guest Additions CPU Hot-Plugging Service. */ /* * Copyright (C) 2010-2016 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. */ /** @page pg_vgsvc_cpuhotplug VBoxService - CPU Hot-Plugging * * The CPU Hot-Plugging subservice helps execute and coordinate CPU hot-plugging * between the guest OS and the VMM. * * CPU Hot-Plugging is useful for reallocating CPU resources from one VM to * other VMs or/and the host. It talks to the VMM via VMMDev, new hot-plugging * events being signalled with an interrupt (no polling). * * Currently only supported for linux guests. */ /********************************************************************************************************************************* * Header Files * *********************************************************************************************************************************/ #include #include #include #include #include #include #include #include #include "VBoxServiceInternal.h" #ifdef RT_OS_LINUX # include # include /* For the sysfs API */ #endif /********************************************************************************************************************************* * Defined Constants And Macros * *********************************************************************************************************************************/ #ifdef RT_OS_LINUX /** @name Paths to access the CPU device * @{ */ # define SYSFS_ACPI_CPU_PATH "/sys/devices" # define SYSFS_CPU_PATH "/sys/devices/system/cpu" /** @} */ /** Path component for the ACPI CPU path. */ typedef struct SYSFSCPUPATHCOMP { /** Flag whether the name is suffixed with a number */ bool fNumberedSuffix; /** Name of the component */ const char *pcszName; } SYSFSCPUPATHCOMP, *PSYSFSCPUPATHCOMP; /** Pointer to a const component. */ typedef const SYSFSCPUPATHCOMP *PCSYSFSCPUPATHCOMP; /** * Structure which defines how the entries are assembled. */ typedef struct SYSFSCPUPATH { /** Id when probing for the correct path. */ uint32_t uId; /** Array holding the possible components. */ PCSYSFSCPUPATHCOMP aComponentsPossible; /** Number of entries in the array, excluding the terminator. */ unsigned cComponents; /** Directory handle */ PRTDIR pDir; /** Current directory to try. */ char *pszPath; } SYSFSCPUPATH, *PSYSFSCPUPATH; /** Content of uId if the path wasn't probed yet. */ # define ACPI_CPU_PATH_NOT_PROBED UINT32_MAX #endif /* RT_OS_LINUX*/ /********************************************************************************************************************************* * Global Variables * *********************************************************************************************************************************/ #ifdef RT_OS_LINUX /** Possible combinations of all path components for level 1. */ static const SYSFSCPUPATHCOMP g_aAcpiCpuPathLvl1[] = { /** LNXSYSTEM: */ { true, "LNXSYSTM:*" } }; /** Possible combinations of all path components for level 2. */ static const SYSFSCPUPATHCOMP g_aAcpiCpuPathLvl2[] = { /** device: */ { true, "device:*" }, /** LNXSYBUS: */ { true, "LNXSYBUS:*" } }; /** Possible combinations of all path components for level 3 */ static const SYSFSCPUPATHCOMP g_aAcpiCpuPathLvl3[] = { /** ACPI0004: */ { true, "ACPI0004:*" } }; /** Possible combinations of all path components for level 4 */ static const SYSFSCPUPATHCOMP g_aAcpiCpuPathLvl4[] = { /** LNXCPU: */ { true, "LNXCPU:*" }, /** ACPI_CPU: */ { true, "ACPI_CPU:*" } }; /** All possible combinations. */ static SYSFSCPUPATH g_aAcpiCpuPath[] = { /** Level 1 */ { ACPI_CPU_PATH_NOT_PROBED, g_aAcpiCpuPathLvl1, RT_ELEMENTS(g_aAcpiCpuPathLvl1), NULL, NULL }, /** Level 2 */ { ACPI_CPU_PATH_NOT_PROBED, g_aAcpiCpuPathLvl2, RT_ELEMENTS(g_aAcpiCpuPathLvl2), NULL, NULL }, /** Level 3 */ { ACPI_CPU_PATH_NOT_PROBED, g_aAcpiCpuPathLvl3, RT_ELEMENTS(g_aAcpiCpuPathLvl3), NULL, NULL }, /** Level 4 */ { ACPI_CPU_PATH_NOT_PROBED, g_aAcpiCpuPathLvl4, RT_ELEMENTS(g_aAcpiCpuPathLvl4), NULL, NULL }, }; /** * Possible directories to get to the topology directory for reading core and package id. * * @remark: This is not part of the path above because the eject file is not in one of the directories * below and would make the hot unplug code fail. */ static const char *g_apszTopologyPath[] = { "sysdev", "physical_node" }; #endif /* RT_OS_LINUX*/ #ifdef RT_OS_LINUX /** * Probes for the correct path to the ACPI CPU object in sysfs for the * various different kernel versions and distro's. * * @returns VBox status code. */ static int vgsvcCpuHotPlugProbePath(void) { int rc = VINF_SUCCESS; /* Probe for the correct path if we didn't already. */ if (RT_UNLIKELY(g_aAcpiCpuPath[0].uId == ACPI_CPU_PATH_NOT_PROBED)) { char *pszPath = NULL; /** < Current path, increasing while we dig deeper. */ pszPath = RTStrDup(SYSFS_ACPI_CPU_PATH); if (!pszPath) return VERR_NO_MEMORY; /* * Simple algorithm to find the path. * Performance is not a real problem because it is * only executed once. */ for (unsigned iLvlCurr = 0; iLvlCurr < RT_ELEMENTS(g_aAcpiCpuPath); iLvlCurr++) { PSYSFSCPUPATH pAcpiCpuPathLvl = &g_aAcpiCpuPath[iLvlCurr]; for (unsigned iCompCurr = 0; iCompCurr < pAcpiCpuPathLvl->cComponents; iCompCurr++) { PCSYSFSCPUPATHCOMP pPathComponent = &pAcpiCpuPathLvl->aComponentsPossible[iCompCurr]; /* Open the directory */ PRTDIR pDirCurr = NULL; char *pszPathTmp = RTPathJoinA(pszPath, pPathComponent->pcszName); if (pszPathTmp) { rc = RTDirOpenFiltered(&pDirCurr, pszPathTmp, RTDIRFILTER_WINNT, 0); RTStrFree(pszPathTmp); } else rc = VERR_NO_STR_MEMORY; if (RT_FAILURE(rc)) break; /* Search if the current directory contains one of the possible parts. */ size_t cchName = strlen(pPathComponent->pcszName); RTDIRENTRY DirFolderContent; bool fFound = false; /* Get rid of the * filter which is in the path component. */ if (pPathComponent->fNumberedSuffix) cchName--; while (RT_SUCCESS(RTDirRead(pDirCurr, &DirFolderContent, NULL))) /* Assumption that szName has always enough space */ { if ( DirFolderContent.cbName >= cchName && !strncmp(DirFolderContent.szName, pPathComponent->pcszName, cchName)) { /* Found, use the complete name to dig deeper. */ fFound = true; pAcpiCpuPathLvl->uId = iCompCurr; char *pszPathLvl = RTPathJoinA(pszPath, DirFolderContent.szName); if (pszPathLvl) { RTStrFree(pszPath); pszPath = pszPathLvl; } else rc = VERR_NO_STR_MEMORY; break; } } RTDirClose(pDirCurr); if (fFound) break; } /* For every possible component. */ /* No matching component for this part, no need to continue */ if (RT_FAILURE(rc)) break; } /* For every level */ VGSvcVerbose(1, "Final path after probing %s rc=%Rrc\n", pszPath, rc); RTStrFree(pszPath); } return rc; } /** * Returns the path of the ACPI CPU device with the given core and package ID. * * @returns VBox status code. * @param ppszPath Where to store the path. * @param idCpuCore The core ID of the CPU. * @param idCpuPackage The package ID of the CPU. */ static int vgsvcCpuHotPlugGetACPIDevicePath(char **ppszPath, uint32_t idCpuCore, uint32_t idCpuPackage) { int rc = VINF_SUCCESS; AssertPtrReturn(ppszPath, VERR_INVALID_PARAMETER); rc = vgsvcCpuHotPlugProbePath(); if (RT_SUCCESS(rc)) { /* Build the path from all components. */ bool fFound = false; unsigned iLvlCurr = 0; char *pszPath = NULL; char *pszPathDir = NULL; PSYSFSCPUPATH pAcpiCpuPathLvl = &g_aAcpiCpuPath[iLvlCurr]; /* Init everything. */ Assert(pAcpiCpuPathLvl->uId != ACPI_CPU_PATH_NOT_PROBED); pszPath = RTPathJoinA(SYSFS_ACPI_CPU_PATH, pAcpiCpuPathLvl->aComponentsPossible[pAcpiCpuPathLvl->uId].pcszName); if (!pszPath) return VERR_NO_STR_MEMORY; pAcpiCpuPathLvl->pszPath = RTStrDup(SYSFS_ACPI_CPU_PATH); if (!pAcpiCpuPathLvl->pszPath) { RTStrFree(pszPath); return VERR_NO_STR_MEMORY; } /* Open the directory */ rc = RTDirOpenFiltered(&pAcpiCpuPathLvl->pDir, pszPath, RTDIRFILTER_WINNT, 0); if (RT_SUCCESS(rc)) { RTStrFree(pszPath); /* Search for CPU */ while (!fFound) { /* Get the next directory. */ RTDIRENTRY DirFolderContent; rc = RTDirRead(pAcpiCpuPathLvl->pDir, &DirFolderContent, NULL); if (RT_SUCCESS(rc)) { /* Create the new path. */ char *pszPathCurr = RTPathJoinA(pAcpiCpuPathLvl->pszPath, DirFolderContent.szName); if (!pszPathCurr) { rc = VERR_NO_STR_MEMORY; break; } /* If this is the last level check for the given core and package id. */ if (iLvlCurr == RT_ELEMENTS(g_aAcpiCpuPath) - 1) { /* Get the sysdev */ uint32_t idCore = 0; uint32_t idPackage = 0; for (unsigned i = 0; i < RT_ELEMENTS(g_apszTopologyPath); i++) { int64_t i64Core = 0; int64_t i64Package = 0; int rc2 = RTLinuxSysFsReadIntFile(10, &i64Core, "%s/%s/topology/core_id", pszPathCurr, g_apszTopologyPath[i]); if (RT_SUCCESS(rc2)) rc2 = RTLinuxSysFsReadIntFile(10, &i64Package, "%s/%s/topology/physical_package_id", pszPathCurr, g_apszTopologyPath[i]); if (RT_SUCCESS(rc2)) { idCore = (uint32_t)i64Core; idPackage = (uint32_t)i64Package; break; } } if ( idCore == idCpuCore && idPackage == idCpuPackage) { /* Return the path */ pszPath = pszPathCurr; fFound = true; VGSvcVerbose(3, "CPU found\n"); break; } else { /* Get the next directory. */ RTStrFree(pszPathCurr); VGSvcVerbose(3, "CPU doesn't match, next directory\n"); } } else { /* Go deeper */ iLvlCurr++; VGSvcVerbose(3, "Going deeper (iLvlCurr=%u)\n", iLvlCurr); pAcpiCpuPathLvl = &g_aAcpiCpuPath[iLvlCurr]; Assert(!pAcpiCpuPathLvl->pDir); Assert(!pAcpiCpuPathLvl->pszPath); pAcpiCpuPathLvl->pszPath = pszPathCurr; PCSYSFSCPUPATHCOMP pPathComponent = &pAcpiCpuPathLvl->aComponentsPossible[pAcpiCpuPathLvl->uId]; Assert(pAcpiCpuPathLvl->uId != ACPI_CPU_PATH_NOT_PROBED); pszPathDir = RTPathJoinA(pszPathCurr, pPathComponent->pcszName); if (!pszPathDir) { rc = VERR_NO_STR_MEMORY; break; } VGSvcVerbose(3, "New path %s\n", pszPathDir); /* Open the directory */ rc = RTDirOpenFiltered(&pAcpiCpuPathLvl->pDir, pszPathDir, RTDIRFILTER_WINNT, 0); if (RT_FAILURE(rc)) break; } } else { /* Go back one level and try to get the next entry. */ Assert(iLvlCurr > 0); RTDirClose(pAcpiCpuPathLvl->pDir); RTStrFree(pAcpiCpuPathLvl->pszPath); pAcpiCpuPathLvl->pDir = NULL; pAcpiCpuPathLvl->pszPath = NULL; iLvlCurr--; pAcpiCpuPathLvl = &g_aAcpiCpuPath[iLvlCurr]; VGSvcVerbose(3, "Directory not found, going back (iLvlCurr=%u)\n", iLvlCurr); } } /* while not found */ } /* Successful init */ /* Cleanup */ for (unsigned i = 0; i < RT_ELEMENTS(g_aAcpiCpuPath); i++) { if (g_aAcpiCpuPath[i].pDir) RTDirClose(g_aAcpiCpuPath[i].pDir); if (g_aAcpiCpuPath[i].pszPath) RTStrFree(g_aAcpiCpuPath[i].pszPath); g_aAcpiCpuPath[i].pDir = NULL; g_aAcpiCpuPath[i].pszPath = NULL; } if (pszPathDir) RTStrFree(pszPathDir); if (RT_FAILURE(rc) && pszPath) RTStrFree(pszPath); if (RT_SUCCESS(rc)) *ppszPath = pszPath; } return rc; } #endif /* RT_OS_LINUX */ /** * Handles VMMDevCpuEventType_Plug. * * @param idCpuCore The CPU core ID. * @param idCpuPackage The CPU package ID. */ static void vgsvcCpuHotPlugHandlePlugEvent(uint32_t idCpuCore, uint32_t idCpuPackage) { #ifdef RT_OS_LINUX /* * The topology directory (containing the physical and core id properties) * is not available until the CPU is online. So we just iterate over all directories * and enable every CPU which is not online already. * Because the directory might not be available immediately we try a few times. * */ /** @todo Maybe use udev to monitor hot-add events from the kernel */ bool fCpuOnline = false; unsigned cTries = 5; do { PRTDIR pDirDevices = NULL; int rc = RTDirOpen(&pDirDevices, SYSFS_CPU_PATH); if (RT_SUCCESS(rc)) { RTDIRENTRY DirFolderContent; while (RT_SUCCESS(RTDirRead(pDirDevices, &DirFolderContent, NULL))) /* Assumption that szName has always enough space */ { /** @todo r-bird: This code is bringing all CPUs online; the idCpuCore and * idCpuPackage parameters are unused! * aeichner: These files are not available at this point unfortunately. (see comment above) * bird: Yes, but isn't that easily dealt with by doing: * if (matching_topology() || !have_topology_directory()) * bring_cpu_online() * That could save you the cpu0 and cpuidle checks to. */ /* * Check if this is a CPU object. * cpu0 is excluded because it is not possible to change the state * of the first CPU on Linux (it doesn't even have an online file) * and cpuidle is no CPU device. Prevents error messages later. */ if( !strncmp(DirFolderContent.szName, "cpu", 3) && strncmp(DirFolderContent.szName, "cpu0", 4) && strncmp(DirFolderContent.szName, "cpuidle", 7)) { /* Get the sysdev */ RTFILE hFileCpuOnline = NIL_RTFILE; rc = RTFileOpenF(&hFileCpuOnline, RTFILE_O_WRITE | RTFILE_O_OPEN | RTFILE_O_DENY_NONE, "%s/%s/online", SYSFS_CPU_PATH, DirFolderContent.szName); if (RT_SUCCESS(rc)) { /* Write a 1 to online the CPU */ rc = RTFileWrite(hFileCpuOnline, "1", 1, NULL); RTFileClose(hFileCpuOnline); if (RT_SUCCESS(rc)) { VGSvcVerbose(1, "CpuHotPlug: CPU %u/%u was brought online\n", idCpuPackage, idCpuCore); fCpuOnline = true; break; } /* Error means CPU not present or online already */ } else VGSvcError("CpuHotPlug: Failed to open '%s/%s/online' rc=%Rrc\n", SYSFS_CPU_PATH, DirFolderContent.szName, rc); } } } else VGSvcError("CpuHotPlug: Failed to open path %s rc=%Rrc\n", SYSFS_CPU_PATH, rc); /* Sleep a bit */ if (!fCpuOnline) RTThreadSleep(10); } while ( !fCpuOnline && cTries-- > 0); #else # error "Port me" #endif } /** * Handles VMMDevCpuEventType_Unplug. * * @param idCpuCore The CPU core ID. * @param idCpuPackage The CPU package ID. */ static void vgsvcCpuHotPlugHandleUnplugEvent(uint32_t idCpuCore, uint32_t idCpuPackage) { #ifdef RT_OS_LINUX char *pszCpuDevicePath = NULL; int rc = vgsvcCpuHotPlugGetACPIDevicePath(&pszCpuDevicePath, idCpuCore, idCpuPackage); if (RT_SUCCESS(rc)) { RTFILE hFileCpuEject; rc = RTFileOpenF(&hFileCpuEject, RTFILE_O_WRITE | RTFILE_O_OPEN | RTFILE_O_DENY_NONE, "%s/eject", pszCpuDevicePath); if (RT_SUCCESS(rc)) { /* Write a 1 to eject the CPU */ rc = RTFileWrite(hFileCpuEject, "1", 1, NULL); if (RT_SUCCESS(rc)) VGSvcVerbose(1, "CpuHotPlug: CPU %u/%u was ejected\n", idCpuPackage, idCpuCore); else VGSvcError("CpuHotPlug: Failed to eject CPU %u/%u rc=%Rrc\n", idCpuPackage, idCpuCore, rc); RTFileClose(hFileCpuEject); } else VGSvcError("CpuHotPlug: Failed to open '%s/eject' rc=%Rrc\n", pszCpuDevicePath, rc); RTStrFree(pszCpuDevicePath); } else VGSvcError("CpuHotPlug: Failed to get CPU device path rc=%Rrc\n", rc); #else # error "Port me" #endif } /** @interface_method_impl{VBOXSERVICE,pfnWorker} */ static DECLCALLBACK(int) vgsvcCpuHotPlugWorker(bool volatile *pfShutdown) { /* * Tell the control thread that it can continue spawning services. */ RTThreadUserSignal(RTThreadSelf()); /* * Enable the CPU hotplug notifier. */ int rc = VbglR3CpuHotPlugInit(); if (RT_FAILURE(rc)) return rc; /* * The Work Loop. */ for (;;) { /* Wait for CPU hot-plugging event. */ uint32_t idCpuCore; uint32_t idCpuPackage; VMMDevCpuEventType enmEventType; rc = VbglR3CpuHotPlugWaitForEvent(&enmEventType, &idCpuCore, &idCpuPackage); if (RT_SUCCESS(rc)) { VGSvcVerbose(3, "CpuHotPlug: Event happened idCpuCore=%u idCpuPackage=%u enmEventType=%d\n", idCpuCore, idCpuPackage, enmEventType); switch (enmEventType) { case VMMDevCpuEventType_Plug: vgsvcCpuHotPlugHandlePlugEvent(idCpuCore, idCpuPackage); break; case VMMDevCpuEventType_Unplug: vgsvcCpuHotPlugHandleUnplugEvent(idCpuCore, idCpuPackage); break; default: { static uint32_t s_iErrors = 0; if (s_iErrors++ < 10) VGSvcError("CpuHotPlug: Unknown event: idCpuCore=%u idCpuPackage=%u enmEventType=%d\n", idCpuCore, idCpuPackage, enmEventType); break; } } } else if (rc != VERR_INTERRUPTED && rc != VERR_TRY_AGAIN) { VGSvcError("CpuHotPlug: VbglR3CpuHotPlugWaitForEvent returned %Rrc\n", rc); break; } if (*pfShutdown) break; } VbglR3CpuHotPlugTerm(); return rc; } /** @interface_method_impl{VBOXSERVICE,pfnStop} */ static DECLCALLBACK(void) vgsvcCpuHotPlugStop(void) { VbglR3InterruptEventWaits(); return; } /** * The 'CpuHotPlug' service description. */ VBOXSERVICE g_CpuHotPlug = { /* pszName. */ "cpuhotplug", /* pszDescription. */ "CPU hot-plugging monitor", /* pszUsage. */ NULL, /* pszOptions. */ NULL, /* methods */ VGSvcDefaultPreInit, VGSvcDefaultOption, VGSvcDefaultInit, vgsvcCpuHotPlugWorker, vgsvcCpuHotPlugStop, VGSvcDefaultTerm };