From nobody Wed May 8 22:48:11 2024 Delivered-To: importer@patchew.org Received-SPF: pass (zoho.com: domain of redhat.com designates 209.132.183.28 as permitted sender) client-ip=209.132.183.28; envelope-from=libvir-list-bounces@redhat.com; helo=mx1.redhat.com; Authentication-Results: mx.zohomail.com; spf=pass (zoho.com: domain of redhat.com designates 209.132.183.28 as permitted sender) smtp.mailfrom=libvir-list-bounces@redhat.com; dmarc=pass(p=none dis=none) header.from=redhat.com Return-Path: Received: from mx1.redhat.com (mx1.redhat.com [209.132.183.28]) by mx.zohomail.com with SMTPS id 1548927269735985.8709693556231; Thu, 31 Jan 2019 01:34:29 -0800 (PST) Received: from smtp.corp.redhat.com (int-mx05.intmail.prod.int.phx2.redhat.com [10.5.11.15]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mx1.redhat.com (Postfix) with ESMTPS id 1612E7AEA2; Thu, 31 Jan 2019 09:34:27 +0000 (UTC) Received: from colo-mx.corp.redhat.com (colo-mx01.intmail.prod.int.phx2.redhat.com [10.5.11.20]) by smtp.corp.redhat.com (Postfix) with ESMTPS id 8928B5D787; Thu, 31 Jan 2019 09:34:25 +0000 (UTC) Received: from lists01.pubmisc.prod.ext.phx2.redhat.com (lists01.pubmisc.prod.ext.phx2.redhat.com [10.5.19.33]) by colo-mx.corp.redhat.com (Postfix) with ESMTP id 827911803391; Thu, 31 Jan 2019 09:34:24 +0000 (UTC) Received: from smtp.corp.redhat.com (int-mx02.intmail.prod.int.phx2.redhat.com [10.5.11.12]) by lists01.pubmisc.prod.ext.phx2.redhat.com (8.13.8/8.13.8) with ESMTP id x0V9YNQ1009591 for ; Thu, 31 Jan 2019 04:34:23 -0500 Received: by smtp.corp.redhat.com (Postfix) id 9D84360C66; Thu, 31 Jan 2019 09:34:23 +0000 (UTC) Received: from moe.brq.redhat.com (unknown [10.43.2.192]) by smtp.corp.redhat.com (Postfix) with ESMTP id 1F9C260244 for ; Thu, 31 Jan 2019 09:34:20 +0000 (UTC) From: Michal Privoznik To: libvir-list@redhat.com Date: Thu, 31 Jan 2019 10:34:17 +0100 Message-Id: <54b1a8b313023f871d9df2ad2cb3cfc62fc100ed.1548927196.git.mprivozn@redhat.com> MIME-Version: 1.0 X-Scanned-By: MIMEDefang 2.79 on 10.5.11.12 X-loop: libvir-list@redhat.com Subject: [libvirt] [PATCH v2] qemu: Rework setting process affinity X-BeenThere: libvir-list@redhat.com X-Mailman-Version: 2.1.12 Precedence: junk List-Id: Development discussions about the libvirt library & tools List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Content-Transfer-Encoding: quoted-printable Sender: libvir-list-bounces@redhat.com Errors-To: libvir-list-bounces@redhat.com X-Scanned-By: MIMEDefang 2.79 on 10.5.11.15 X-Greylist: Sender IP whitelisted, not delayed by milter-greylist-4.5.16 (mx1.redhat.com [10.5.110.25]); Thu, 31 Jan 2019 09:34:28 +0000 (UTC) Content-Type: text/plain; charset="utf-8" https://bugzilla.redhat.com/show_bug.cgi?id=3D1503284 The way we currently start qemu from CPU affinity POV is as follows: 1) the child process is set affinity to all online CPUs (unless some vcpu pinning was given in the domain XML) 2) Once qemu is running, cpuset cgroup is configured taking memory pinning into account Problem is that we let qemu allocate its memory just anywhere in 1) and then rely in 2) to be able to move the memory to configured NUMA nodes. This might not be always possible (e.g. qemu might lock some parts of its memory) and is very suboptimal (copying large memory between NUMA nodes takes significant amount of time). The solution is to set affinity to one of (in priority order): - The CPUs associated with NUMA memory affinity mask - The CPUs associated with emulator pinning - All online host CPUs Later (once QEMU has allocated its memory) we then change this again to (again in priority order): - The CPUs associated with emulator pinning - The CPUs returned by numad - The CPUs associated with vCPU pinning - All online host CPUs Signed-off-by: Michal Privoznik Reviewed-by: Daniel P. Berrang=C3=A9 --- diff to v1 (both points suggested by Dan): - Expanded the commit message - fixed qemuProcessGetAllCpuAffinity so that it returns online CPU map only src/qemu/qemu_process.c | 132 +++++++++++++++++++--------------------- 1 file changed, 63 insertions(+), 69 deletions(-) diff --git a/src/qemu/qemu_process.c b/src/qemu/qemu_process.c index 7592c98b74..dace5aaca1 100644 --- a/src/qemu/qemu_process.c +++ b/src/qemu/qemu_process.c @@ -2435,6 +2435,21 @@ qemuProcessDetectIOThreadPIDs(virQEMUDriverPtr drive= r, } =20 =20 +static int +qemuProcessGetAllCpuAffinity(virBitmapPtr *cpumapRet) +{ + *cpumapRet =3D NULL; + + if (!virHostCPUHasBitmap()) + return 0; + + if (!(*cpumapRet =3D virHostCPUGetOnlineBitmap())) + return -1; + + return 0; +} + + /* * To be run between fork/exec of QEMU only */ @@ -2443,9 +2458,9 @@ static int qemuProcessInitCpuAffinity(virDomainObjPtr vm) { int ret =3D -1; - virBitmapPtr cpumap =3D NULL; virBitmapPtr cpumapToSet =3D NULL; - virBitmapPtr hostcpumap =3D NULL; + VIR_AUTOPTR(virBitmap) hostcpumap =3D NULL; + virDomainNumatuneMemMode mem_mode; qemuDomainObjPrivatePtr priv =3D vm->privateData; =20 if (!vm->pid) { @@ -2454,59 +2469,39 @@ qemuProcessInitCpuAffinity(virDomainObjPtr vm) return -1; } =20 - if (vm->def->placement_mode =3D=3D VIR_DOMAIN_CPU_PLACEMENT_MODE_AUTO)= { - VIR_DEBUG("Set CPU affinity with advisory nodeset from numad"); - cpumapToSet =3D priv->autoCpuset; + /* Here is the deal, we can't set cpuset.mems before qemu is + * started as it clashes with KVM allocation. Therefore, we + * used to let qemu allocate its memory anywhere as we would + * then move the memory to desired NUMA node via CGroups. + * However, that might not be always possible because qemu + * might lock some parts of its memory (e.g. due to VFIO). + * Even if it possible, memory has to be copied between NUMA + * nodes which is suboptimal. + * Solution is to set affinity that matches the best what we + * would have set in CGroups and then fix it later, once qemu + * is already running. */ + if (virDomainNumaGetNodeCount(vm->def->numa) <=3D 1 && + virDomainNumatuneGetMode(vm->def->numa, -1, &mem_mode) =3D=3D 0 && + mem_mode =3D=3D VIR_DOMAIN_NUMATUNE_MEM_STRICT) { + if (virDomainNumatuneMaybeGetNodeset(vm->def->numa, + priv->autoNodeset, + &cpumapToSet, + -1) < 0) + goto cleanup; + } else if (vm->def->cputune.emulatorpin) { + cpumapToSet =3D vm->def->cputune.emulatorpin; } else { - VIR_DEBUG("Set CPU affinity with specified cpuset"); - if (vm->def->cpumask) { - cpumapToSet =3D vm->def->cpumask; - } else { - /* You may think this is redundant, but we can't assume libvir= td - * itself is running on all pCPUs, so we need to explicitly set - * the spawned QEMU instance to all pCPUs if no map is given in - * its config file */ - int hostcpus; - - if (virHostCPUHasBitmap()) { - hostcpumap =3D virHostCPUGetOnlineBitmap(); - cpumap =3D virProcessGetAffinity(vm->pid); - } - - if (hostcpumap && cpumap && virBitmapEqual(hostcpumap, cpumap)= ) { - /* we're using all available CPUs, no reason to set - * mask. If libvirtd is running without explicit - * affinity, we can use hotplugged CPUs for this VM */ - ret =3D 0; - goto cleanup; - } else { - /* setaffinity fails if you set bits for CPUs which - * aren't present, so we have to limit ourselves */ - if ((hostcpus =3D virHostCPUGetCount()) < 0) - goto cleanup; - - if (hostcpus > QEMUD_CPUMASK_LEN) - hostcpus =3D QEMUD_CPUMASK_LEN; - - virBitmapFree(cpumap); - if (!(cpumap =3D virBitmapNew(hostcpus))) - goto cleanup; - - virBitmapSetAll(cpumap); - - cpumapToSet =3D cpumap; - } - } + if (qemuProcessGetAllCpuAffinity(&hostcpumap) < 0) + goto cleanup; + cpumapToSet =3D hostcpumap; } =20 - if (virProcessSetAffinity(vm->pid, cpumapToSet) < 0) + if (cpumapToSet && + virProcessSetAffinity(vm->pid, cpumapToSet) < 0) goto cleanup; =20 ret =3D 0; - cleanup: - virBitmapFree(cpumap); - virBitmapFree(hostcpumap); return ret; } #else /* !defined(HAVE_SCHED_GETAFFINITY) && !defined(HAVE_BSD_CPU_AFFINIT= Y) */ @@ -2586,7 +2581,8 @@ qemuProcessSetupPid(virDomainObjPtr vm, qemuDomainObjPrivatePtr priv =3D vm->privateData; virDomainNumatuneMemMode mem_mode; virCgroupPtr cgroup =3D NULL; - virBitmapPtr use_cpumask; + virBitmapPtr use_cpumask =3D NULL; + VIR_AUTOPTR(virBitmap) hostcpumap =3D NULL; char *mem_mask =3D NULL; int ret =3D -1; =20 @@ -2598,12 +2594,21 @@ qemuProcessSetupPid(virDomainObjPtr vm, } =20 /* Infer which cpumask shall be used. */ - if (cpumask) + if (cpumask) { use_cpumask =3D cpumask; - else if (vm->def->placement_mode =3D=3D VIR_DOMAIN_CPU_PLACEMENT_MODE_= AUTO) + } else if (vm->def->placement_mode =3D=3D VIR_DOMAIN_CPU_PLACEMENT_MOD= E_AUTO) { use_cpumask =3D priv->autoCpuset; - else + } else if (vm->def->cpumask) { use_cpumask =3D vm->def->cpumask; + } else { + /* You may think this is redundant, but we can't assume libvirtd + * itself is running on all pCPUs, so we need to explicitly set + * the spawned QEMU instance to all pCPUs if no map is given in + * its config file */ + if (qemuProcessGetAllCpuAffinity(&hostcpumap) < 0) + goto cleanup; + use_cpumask =3D hostcpumap; + } =20 /* * If CPU cgroup controller is not initialized here, then we need @@ -2628,13 +2633,7 @@ qemuProcessSetupPid(virDomainObjPtr vm, qemuSetupCgroupCpusetCpus(cgroup, use_cpumask) < 0) goto cleanup; =20 - /* - * Don't setup cpuset.mems for the emulator, they need to - * be set up after initialization in order for kvm - * allocations to succeed. - */ - if (nameval !=3D VIR_CGROUP_THREAD_EMULATOR && - mem_mask && virCgroupSetCpusetMems(cgroup, mem_mask) < 0) + if (mem_mask && virCgroupSetCpusetMems(cgroup, mem_mask) < 0) goto cleanup; =20 } @@ -6634,12 +6633,7 @@ qemuProcessLaunch(virConnectPtr conn, =20 /* This must be done after cgroup placement to avoid resetting CPU * affinity */ - if (!vm->def->cputune.emulatorpin && - qemuProcessInitCpuAffinity(vm) < 0) - goto cleanup; - - VIR_DEBUG("Setting emulator tuning/settings"); - if (qemuProcessSetupEmulator(vm) < 0) + if (qemuProcessInitCpuAffinity(vm) < 0) goto cleanup; =20 VIR_DEBUG("Setting cgroup for external devices (if required)"); @@ -6708,10 +6702,6 @@ qemuProcessLaunch(virConnectPtr conn, if (qemuProcessUpdateAndVerifyCPU(driver, vm, asyncJob) < 0) goto cleanup; =20 - VIR_DEBUG("Setting up post-init cgroup restrictions"); - if (qemuSetupCpusetMems(vm) < 0) - goto cleanup; - VIR_DEBUG("setting up hotpluggable cpus"); if (qemuDomainHasHotpluggableStartupVcpus(vm->def)) { if (qemuDomainRefreshVcpuInfo(driver, vm, asyncJob, false) < 0) @@ -6737,6 +6727,10 @@ qemuProcessLaunch(virConnectPtr conn, if (qemuProcessDetectIOThreadPIDs(driver, vm, asyncJob) < 0) goto cleanup; =20 + VIR_DEBUG("Setting emulator tuning/settings"); + if (qemuProcessSetupEmulator(vm) < 0) + goto cleanup; + VIR_DEBUG("Setting global CPU cgroup (if required)"); if (qemuSetupGlobalCpuCgroup(vm) < 0) goto cleanup; --=20 2.19.2 -- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list