[libvirt PATCH v2 12/17] ch: methods for cgroup mgmt in ch driver

From: Vineeth Pillai <viremana(a)linux.microsoft.com&gt; Signed-off-by: Vineeth Pillai <viremana(a)linux.microsoft.com&gt; Signed-off-by: Praveen K Paladugu <prapal(a)linux.microsoft.com&gt; --- src/ch/ch_conf.c | 2 + src/ch/ch_conf.h | 4 +- src/ch/ch_domain.c | 34 +++++ src/ch/ch_domain.h | 3 +- src/ch/ch_monitor.c | 96 ++++++++++++++ src/ch/ch_monitor.h | 54 +++++++- src/ch/ch_process.c | 306 ++++++++++++++++++++++++++++++++++++++++++-- src/ch/ch_process.h | 3 + 8 files changed, 484 insertions(+), 18 deletions(-) diff --git a/src/ch/ch_conf.c b/src/ch/ch_conf.c index ef6f4b5ba8..cfc1174354 100644 --- a/src/ch/ch_conf.c +++ b/src/ch/ch_conf.c @@ -129,6 +129,8 @@ virCHDriverConfigNew(bool privileged) if (!(cfg = virObjectNew(virCHDriverConfigClass))) return NULL; + cfg->cgroupControllers = -1; /* Auto detect */ + if (privileged) { if (virGetUserID(CH_USER, &cfg->user) < 0) return NULL; diff --git a/src/ch/ch_conf.h b/src/ch/ch_conf.h index 8fe69c8545..1790295ede 100644 --- a/src/ch/ch_conf.h +++ b/src/ch/ch_conf.h @@ -35,11 +35,13 @@ struct _virCHDriverConfig { char *stateDir; char *logDir; - + int cgroupControllers; uid_t user; gid_t group; }; +G_DEFINE_AUTOPTR_CLEANUP_FUNC(virCHDriverConfig, virObjectUnref); + struct _virCHDriver { virMutex lock; diff --git a/src/ch/ch_domain.c b/src/ch/ch_domain.c index 326c3802c3..0a0b1206d6 100644 --- a/src/ch/ch_domain.c +++ b/src/ch/ch_domain.c @@ -322,6 +322,40 @@ chValidateDomainDeviceDef(const virDomainDeviceDef * dev, _("Serial can only be enabled for a PTY")); return -1; } + return 0; +} + +int +virCHDomainRefreshThreadInfo(virDomainObj * vm) +{ + size_t maxvcpus = virDomainDefGetVcpusMax(vm->def); + virCHMonitorThreadInfo *info = NULL; + size_t nthreads, ncpus = 0; + size_t i; + + nthreads = virCHMonitorGetThreadInfo(virCHDomainGetMonitor(vm), + true, &info); + + for (i = 0; i < nthreads; i++) { + virCHDomainVcpuPrivate *vcpupriv; + virDomainVcpuDef *vcpu; + virCHMonitorCPUInfo *vcpuInfo; + + if (info[i].type != virCHThreadTypeVcpu) + continue; + + // TODO: hotplug support + vcpuInfo = &info[i].vcpuInfo; + vcpu = virDomainDefGetVcpu(vm->def, vcpuInfo->cpuid); + vcpupriv = CH_DOMAIN_VCPU_PRIVATE(vcpu); + vcpupriv->tid = vcpuInfo->tid; + ncpus++; + } + + // TODO: Remove the warning when hotplug is implemented. + if (ncpus != maxvcpus) + VIR_WARN("Mismatch in the number of cpus, expected: %ld, actual: %ld", + maxvcpus, ncpus); return 0; } diff --git a/src/ch/ch_domain.h b/src/ch/ch_domain.h index 3ac3421015..2ce3e2cef3 100644 --- a/src/ch/ch_domain.h +++ b/src/ch/ch_domain.h @@ -89,7 +89,8 @@ virCHDomainObjBeginJob(virDomainObj *obj, enum virCHDomainJob job) void virCHDomainObjEndJob(virDomainObj *obj); -int virCHDomainRefreshVcpuInfo(virDomainObj *vm); +int virCHDomainRefreshThreadInfo(virDomainObj *vm); + pid_t virCHDomainGetVcpuPid(virDomainObj *vm, unsigned int vcpuid); bool virCHDomainHasVcpuPids(virDomainObj *vm); diff --git a/src/ch/ch_monitor.c b/src/ch/ch_monitor.c index 6bc7d035ad..d7593d4aa6 100644 --- a/src/ch/ch_monitor.c +++ b/src/ch/ch_monitor.c @@ -41,6 +41,7 @@ VIR_LOG_INIT("ch.ch_monitor"); static virClass *virCHMonitorClass; static void virCHMonitorDispose(void *obj); +static void virCHMonitorThreadInfoFree(virCHMonitor * mon); static int virCHMonitorOnceInit(void) @@ -589,6 +590,7 @@ virCHMonitorDispose(void *opaque) virCHMonitor *mon = opaque; VIR_DEBUG("mon=%p", mon); + virCHMonitorThreadInfoFree(mon); virObjectUnref(mon->vm); } @@ -756,6 +758,100 @@ virCHMonitorGet(virCHMonitor * mon, const char *endpoint, return ret; } +static void +virCHMonitorThreadInfoFree(virCHMonitor * mon) +{ + mon->nthreads = 0; + if (mon->threads) + VIR_FREE(mon->threads); +} + +static size_t +virCHMonitorRefreshThreadInfo(virCHMonitor * mon) +{ + virCHMonitorThreadInfo *info = NULL; + g_autofree pid_t *tids = NULL; + virDomainObj *vm = mon->vm; + size_t ntids = 0; + size_t i; + + + virCHMonitorThreadInfoFree(mon); + if (virProcessGetPids(vm->pid, &ntids, &tids) < 0) { + mon->threads = NULL; + return 0; + } + + info = g_new0(virCHMonitorThreadInfo, ntids); + for (i = 0; i < ntids; i++) { + g_autofree char *proc = NULL; + g_autofree char *data = NULL; + + proc = g_strdup_printf("/proc/%d/task/%d/comm", + (int) vm->pid, (int) tids[i]); + + if (virFileReadAll(proc, (1 << 16), &data) < 0) { + continue; + } + + VIR_DEBUG("VM PID: %d, TID %d, COMM: %s", + (int) vm->pid, (int) tids[i], data); + if (STRPREFIX(data, "vcpu")) { + int cpuid; + char *tmp; + + if (virStrToLong_i(data + 4, &tmp, 0, &cpuid) < 0) { + VIR_WARN("Index is not specified correctly"); + continue; + } + info[i].type = virCHThreadTypeVcpu; + info[i].vcpuInfo.tid = tids[i]; + info[i].vcpuInfo.online = true; + info[i].vcpuInfo.cpuid = cpuid; + VIR_DEBUG("vcpu%d -> tid: %d", cpuid, tids[i]); + } else if (STRPREFIX(data, "_disk") || STRPREFIX(data, "_net") || + STRPREFIX(data, "_rng")) { + /* Prefixes used by cloud-hypervisor for IO Threads are captured at + * https://github.com/cloud-hypervisor/cloud-hypervisor/blob/main/vmm/src/de... */ + info[i].type = virCHThreadTypeIO; + info[i].ioInfo.tid = tids[i]; + virStrcpy(info[i].ioInfo.thrName, data, VIRCH_THREAD_NAME_LEN - 1); + } else { + info[i].type = virCHThreadTypeEmulator; + info[i].emuInfo.tid = tids[i]; + virStrcpy(info[i].emuInfo.thrName, data, VIRCH_THREAD_NAME_LEN - 1); + } + mon->nthreads++; + + } + mon->threads = info; + + return mon->nthreads; +} + +/** + * virCHMonitorGetThreadInfo: + * @mon: Pointer to the monitor + * @refresh: Refresh thread info or not + * + * Retrive thread info and store to @threads + * + * Returns count of threads on success. + */ +size_t +virCHMonitorGetThreadInfo(virCHMonitor * mon, bool refresh, + virCHMonitorThreadInfo ** threads) +{ + int nthreads = 0; + + if (refresh) + nthreads = virCHMonitorRefreshThreadInfo(mon); + + *threads = mon->threads; + + return nthreads; +} + int virCHMonitorShutdownVMM(virCHMonitor * mon) { diff --git a/src/ch/ch_monitor.h b/src/ch/ch_monitor.h index 8ca9e17a9a..f8c3fa75e8 100644 --- a/src/ch/ch_monitor.h +++ b/src/ch/ch_monitor.h @@ -37,6 +37,50 @@ #define URL_VM_RESUME "vm.resume" #define URL_VM_INFO "vm.info" +#define VIRCH_THREAD_NAME_LEN 16 + +typedef enum { + virCHThreadTypeEmulator, + virCHThreadTypeVcpu, + virCHThreadTypeIO, + virCHThreadTypeMax +} virCHThreadType; + +typedef struct _virCHMonitorCPUInfo virCHMonitorCPUInfo; + +struct _virCHMonitorCPUInfo { + int cpuid; + pid_t tid; + + bool online; +}; + +typedef struct _virCHMonitorEmuThreadInfo virCHMonitorEmuThreadInfo; + +struct _virCHMonitorEmuThreadInfo { + char thrName[VIRCH_THREAD_NAME_LEN]; + pid_t tid; +}; + +typedef struct _virCHMonitorIOThreadInfo virCHMonitorIOThreadInfo; + +struct _virCHMonitorIOThreadInfo { + char thrName[VIRCH_THREAD_NAME_LEN]; + pid_t tid; +}; + +typedef struct _virCHMonitorThreadInfo virCHMonitorThreadInfo; + +struct _virCHMonitorThreadInfo { + virCHThreadType type; + + union { + virCHMonitorCPUInfo vcpuInfo; + virCHMonitorEmuThreadInfo emuInfo; + virCHMonitorIOThreadInfo ioInfo; + }; +}; + typedef struct _virCHMonitor virCHMonitor; struct _virCHMonitor { @@ -49,6 +93,9 @@ struct _virCHMonitor { pid_t pid; virDomainObj *vm; + + size_t nthreads; + virCHMonitorThreadInfo *threads; }; virCHMonitor *virCHMonitorNew(virDomainObj *vm, const char *socketdir); @@ -65,12 +112,9 @@ int virCHMonitorSuspendVM(virCHMonitor *mon); int virCHMonitorResumeVM(virCHMonitor *mon); int virCHMonitorGetInfo(virCHMonitor *mon, virJSONValue **info); -typedef struct _virCHMonitorCPUInfo virCHMonitorCPUInfo; -struct _virCHMonitorCPUInfo { - pid_t tid; - bool online; -}; void virCHMonitorCPUInfoFree(virCHMonitorCPUInfo *cpus); int virCHMonitorGetCPUInfo(virCHMonitor *mon, virCHMonitorCPUInfo **vcpus, size_t maxvcpus); +size_t virCHMonitorGetThreadInfo(virCHMonitor *mon, bool refresh, + virCHMonitorThreadInfo **threads); diff --git a/src/ch/ch_process.c b/src/ch/ch_process.c index 3b7f6fcddf..7a51531224 100644 --- a/src/ch/ch_process.c +++ b/src/ch/ch_process.c @@ -26,6 +26,8 @@ #include "ch_domain.h" #include "ch_monitor.h" #include "ch_process.h" +#include "domain_cgroup.h" +#include "virnuma.h" #include "viralloc.h" #include "virerror.h" #include "virjson.h" @@ -133,6 +135,254 @@ virCHProcessUpdateInfo(virDomainObj *vm) return 0; } +static int +virCHProcessGetAllCpuAffinity(virBitmap ** cpumapRet) +{ + *cpumapRet = NULL; + + if (!virHostCPUHasBitmap()) + return 0; + + if (!(*cpumapRet = virHostCPUGetOnlineBitmap())) + return -1; + + return 0; +} + +#if defined(WITH_SCHED_GETAFFINITY) || defined(WITH_BSD_CPU_AFFINITY) +static int +virCHProcessInitCpuAffinity(virDomainObj * vm) +{ + g_autoptr(virBitmap) cpumapToSet = NULL; + virDomainNumatuneMemMode mem_mode; + virCHDomainObjPrivate *priv = vm->privateData; + + if (!vm->pid) { + virReportError(VIR_ERR_INTERNAL_ERROR, "%s", + _("Cannot setup CPU affinity until process is started")); + return -1; + } + + if (virDomainNumaGetNodeCount(vm->def->numa) <= 1 && + virDomainNumatuneGetMode(vm->def->numa, -1, &mem_mode) == 0 && + mem_mode == VIR_DOMAIN_NUMATUNE_MEM_STRICT) { + virBitmap *nodeset = NULL; + + if (virDomainNumatuneMaybeGetNodeset(vm->def->numa, + priv->autoNodeset, + &nodeset, -1) < 0) + return -1; + + if (virNumaNodesetToCPUset(nodeset, &cpumapToSet) < 0) + return -1; + } else if (vm->def->cputune.emulatorpin) { + if (!(cpumapToSet = virBitmapNewCopy(vm->def->cputune.emulatorpin))) + return -1; + } else { + if (virCHProcessGetAllCpuAffinity(&cpumapToSet) < 0) + return -1; + } + + if (cpumapToSet && virProcessSetAffinity(vm->pid, cpumapToSet, false) < 0) { + return -1; + } + + return 0; +} +#else /* !defined(WITH_SCHED_GETAFFINITY) && !defined(WITH_BSD_CPU_AFFINITY) */ +static int +virCHProcessInitCpuAffinity(virDomainObj * vm G_GNUC_UNUSED) +{ + return 0; +} +#endif /* !defined(WITH_SCHED_GETAFFINITY) && !defined(WITH_BSD_CPU_AFFINITY) */ + +/** + * virCHProcessSetupPid: + * + * This function sets resource properties (affinity, cgroups, + * scheduler) for any PID associated with a domain. It should be used + * to set up emulator PIDs as well as vCPU and I/O thread pids to + * ensure they are all handled the same way. + * + * Returns 0 on success, -1 on error. + */ +static int +virCHProcessSetupPid(virDomainObj *vm, + pid_t pid, + virCgroupThreadName nameval, + int id, + virBitmap *cpumask, + unsigned long long period, + long long quota, + virDomainThreadSchedParam *sched) +{ + virCHDomainObjPrivate *priv = vm->privateData; + virDomainNumatuneMemMode mem_mode; + virCgroup *cgroup = NULL; + virBitmap *use_cpumask = NULL; + virBitmap *affinity_cpumask = NULL; + g_autoptr(virBitmap) hostcpumap = NULL; + g_autofree char *mem_mask = NULL; + int ret = -1; + + if ((period || quota) && + !virCgroupHasController(priv->cgroup, VIR_CGROUP_CONTROLLER_CPU)) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("cgroup cpu is required for scheduler tuning")); + goto cleanup; + } + + /* Infer which cpumask shall be used. */ + if (cpumask) { + use_cpumask = cpumask; + } else if (vm->def->placement_mode == VIR_DOMAIN_CPU_PLACEMENT_MODE_AUTO) { + use_cpumask = priv->autoCpuset; + } else if (vm->def->cpumask) { + use_cpumask = vm->def->cpumask; + } else { + /* we can't assume cloud-hypervisor itself is running on all pCPUs, + * so we need to explicitly set the spawned instance to all pCPUs. */ + if (virCHProcessGetAllCpuAffinity(&hostcpumap) < 0) + goto cleanup; + affinity_cpumask = hostcpumap; + } + + /* + * If CPU cgroup controller is not initialized here, then we need + * neither period nor quota settings. And if CPUSET controller is + * not initialized either, then there's nothing to do anyway. + */ + if (virCgroupHasController(priv->cgroup, VIR_CGROUP_CONTROLLER_CPU) || + virCgroupHasController(priv->cgroup, VIR_CGROUP_CONTROLLER_CPUSET)) { + + if (virDomainNumatuneGetMode(vm->def->numa, -1, &mem_mode) == 0 && + mem_mode == VIR_DOMAIN_NUMATUNE_MEM_STRICT && + virDomainNumatuneMaybeFormatNodeset(vm->def->numa, + priv->autoNodeset, + &mem_mask, -1) < 0) + goto cleanup; + + if (virCgroupNewThread(priv->cgroup, nameval, id, true, &cgroup) < 0) + goto cleanup; + + if (virCgroupHasController(priv->cgroup, VIR_CGROUP_CONTROLLER_CPUSET)) { + if (use_cpumask && + virSetupCgroupCpusetCpus(cgroup, use_cpumask) < 0) + goto cleanup; + + if (mem_mask && virCgroupSetCpusetMems(cgroup, mem_mask) < 0) + goto cleanup; + + } + + if ((period || quota) && + virSetupCgroupVcpuBW(cgroup, period, quota) < 0) + goto cleanup; + + /* Move the thread to the sub dir */ + VIR_INFO("Adding pid %d to cgroup", pid); + if (virCgroupAddThread(cgroup, pid) < 0) + goto cleanup; + + } + + if (!affinity_cpumask) + affinity_cpumask = use_cpumask; + + /* Setup legacy affinity. */ + if (affinity_cpumask + && virProcessSetAffinity(pid, affinity_cpumask, false) < 0) + goto cleanup; + + /* Set scheduler type and priority, but not for the main thread. */ + if (sched && + nameval != VIR_CGROUP_THREAD_EMULATOR && + virProcessSetScheduler(pid, sched->policy, sched->priority) < 0) + goto cleanup; + + ret = 0; + cleanup: + if (cgroup) { + if (ret < 0) + virCgroupRemove(cgroup); + virCgroupFree(cgroup); + } + + return ret; +} + +/** + * virCHProcessSetupVcpu: + * @vm: domain object + * @vcpuid: id of VCPU to set defaults + * + * This function sets resource properties (cgroups, affinity, scheduler) for a + * vCPU. This function expects that the vCPU is online and the vCPU pids were + * correctly detected at the point when it's called. + * + * Returns 0 on success, -1 on error. + */ +int +virCHProcessSetupVcpu(virDomainObj * vm, unsigned int vcpuid) +{ + pid_t vcpupid = virCHDomainGetVcpuPid(vm, vcpuid); + virDomainVcpuDef *vcpu = virDomainDefGetVcpu(vm->def, vcpuid); + + return virCHProcessSetupPid(vm, vcpupid, VIR_CGROUP_THREAD_VCPU, + vcpuid, vcpu->cpumask, + vm->def->cputune.period, + vm->def->cputune.quota, &vcpu->sched); +} + +static int +virCHProcessSetupVcpus(virDomainObj * vm) +{ + virDomainVcpuDef *vcpu; + unsigned int maxvcpus = virDomainDefGetVcpusMax(vm->def); + size_t i; + + if ((vm->def->cputune.period || vm->def->cputune.quota) && + !virCgroupHasController(((virCHDomainObjPrivate *) vm->privateData)-> + cgroup, VIR_CGROUP_CONTROLLER_CPU)) { + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s", + _("cgroup cpu is required for scheduler tuning")); + return -1; + } + + if (!virCHDomainHasVcpuPids(vm)) { + /* If any CPU has custom affinity that differs from the + * VM default affinity, we must reject it */ + for (i = 0; i < maxvcpus; i++) { + vcpu = virDomainDefGetVcpu(vm->def, i); + + if (!vcpu->online) + continue; + + if (vcpu->cpumask && + !virBitmapEqual(vm->def->cpumask, vcpu->cpumask)) { + virReportError(VIR_ERR_OPERATION_INVALID, "%s", + _("cpu affinity is not supported")); + return -1; + } + } + + return 0; + } + + for (i = 0; i < maxvcpus; i++) { + vcpu = virDomainDefGetVcpu(vm->def, i); + + if (!vcpu->online) + continue; + + if (virCHProcessSetupVcpu(vm, i) < 0) + return -1; + } + + return 0; +} + /** * virCHProcessStart: * @driver: pointer to driver structure @@ -143,12 +393,13 @@ virCHProcessUpdateInfo(virDomainObj *vm) * * Returns 0 on success or -1 in case of error */ -int virCHProcessStart(virCHDriver *driver, - virDomainObj *vm, - virDomainRunningReason reason) +int +virCHProcessStart(virCHDriver * driver, + virDomainObj * vm, virDomainRunningReason reason) { int ret = -1; virCHDomainObjPrivate *priv = vm->privateData; + g_autoptr(virCHDriverConfig) cfg = virCHDriverGetConfig(priv->driver); g_autofree int *nicindexes = NULL; size_t nnicindexes = 0; @@ -168,18 +419,39 @@ int virCHProcessStart(virCHDriver *driver, } } + vm->pid = priv->monitor->pid; + vm->def->id = vm->pid; + priv->machineName = virCHDomainGetMachineName(vm); + + if (virSetupCgroup("ch", vm, + nnicindexes, nicindexes, + priv->cgroup, + cfg->cgroupControllers, + 0, /*maxThreadsPerProc*/ + priv->driver->privileged, + priv->machineName) < 0) + goto cleanup; + + if (virCHProcessInitCpuAffinity(vm) < 0) + goto cleanup; + if (virCHMonitorBootVM(priv->monitor) < 0) { virReportError(VIR_ERR_INTERNAL_ERROR, "%s", _("failed to boot guest VM")); goto cleanup; } - priv->machineName = virCHDomainGetMachineName(vm); - vm->pid = priv->monitor->pid; - vm->def->id = vm->pid; + virCHDomainRefreshThreadInfo(vm); - virCHProcessUpdateInfo(vm); + VIR_DEBUG("Setting global CPU cgroup (if required)"); + if (virSetupGlobalCpuCgroup(vm, priv->cgroup, priv->autoNodeset) < 0) + goto cleanup; + + VIR_DEBUG("Setting vCPU tuning/settings"); + if (virCHProcessSetupVcpus(vm) < 0) + goto cleanup; + virCHProcessUpdateInfo(vm); virDomainObjSetState(vm, VIR_DOMAIN_RUNNING, reason); return 0; @@ -191,20 +463,32 @@ int virCHProcessStart(virCHDriver *driver, return ret; } -int virCHProcessStop(virCHDriver *driver G_GNUC_UNUSED, - virDomainObj *vm, - virDomainShutoffReason reason) +int +virCHProcessStop(virCHDriver * driver G_GNUC_UNUSED, + virDomainObj * vm, virDomainShutoffReason reason) { + int ret; + int retries = 0; virCHDomainObjPrivate *priv = vm->privateData; VIR_DEBUG("Stopping VM name=%s pid=%d reason=%d", - vm->def->name, (int)vm->pid, (int)reason); + vm->def->name, (int) vm->pid, (int) reason); if (priv->monitor) { virCHMonitorClose(priv->monitor); priv->monitor = NULL; } + retry: + if ((ret = virRemoveCgroup(vm, priv->cgroup, priv->machineName)) < 0) { + if (ret == -EBUSY && (retries++ < 5)) { + g_usleep(200*1000); + goto retry; + } + VIR_WARN("Failed to remove cgroup for %s", + vm->def->name); + } + vm->pid = -1; vm->def->id = -1; diff --git a/src/ch/ch_process.h b/src/ch/ch_process.h index abc4915979..800e3f4e23 100644 --- a/src/ch/ch_process.h +++ b/src/ch/ch_process.h @@ -29,3 +29,6 @@ int virCHProcessStart(virCHDriver *driver, int virCHProcessStop(virCHDriver *driver, virDomainObj *vm, virDomainShutoffReason reason); + +int virCHProcessSetupVcpu(virDomainObj *vm, + unsigned int vcpuid); -- 2.27.0