From patchwork Thu Feb 17 03:49:46 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Reiji Watanabe X-Patchwork-Id: 12749293 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id D00FBC433F5 for ; Thu, 17 Feb 2022 03:50:18 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232375AbiBQDu2 (ORCPT ); Wed, 16 Feb 2022 22:50:28 -0500 Received: from mxb-00190b01.gslb.pphosted.com ([23.128.96.19]:52714 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231430AbiBQDuZ (ORCPT ); Wed, 16 Feb 2022 22:50:25 -0500 Received: from mail-pl1-x64a.google.com (mail-pl1-x64a.google.com [IPv6:2607:f8b0:4864:20::64a]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id E23AFC4280 for ; Wed, 16 Feb 2022 19:50:11 -0800 (PST) Received: by mail-pl1-x64a.google.com with SMTP id a13-20020a170902ee8d00b0014f308fed09so1966677pld.12 for ; Wed, 16 Feb 2022 19:50:11 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=20210112; h=date:message-id:mime-version:subject:from:to:cc; bh=vRtM/cROOx5FsqNMttv0pPiPYqWySn+ox7xCpR/R9pQ=; b=VXI/EO5uA+etW0dY+bDcMbof9vcOt9SX4c8MIwKYL2KjJK778sgRC4ONa+S96Tq0kC 2DNAkeg32L87pidTLjIxaGqseJc+3T5vMuB+w+93GZkI2D2TVXTTsK9tLKdq55IVduJd HY6Yt6e7ITaZd1JWGm6SAif/ms7rSjFues9sLIJCzp12Jt5VwgzXhnRJusVMPVi7fx7K EGLTgjyoji9r7qxZ7uhq821QmUs0eEGMKzsvl5mW4BzjI4gNlfYLd+rvssbuo7pt2BsP pXyzTyGwbqarreY8cymIKeiUQtmMLmNlCpb2jOchN184vkaWtt52ENyHw4o+hy9Vex4V c4Xw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:date:message-id:mime-version:subject:from:to:cc; bh=vRtM/cROOx5FsqNMttv0pPiPYqWySn+ox7xCpR/R9pQ=; b=ygdg777sxR7g4Sb+rL0Fv5zarvCaBS5JabB0FbR7eevYTsrohQEdGqx7Swa2sPqPRz W0dSndX81/HFwFB5Il6Vp8D0lqAsIrOGUAr8p4v1kPW5rZ6VH63zayNysr1s8quwRhtX 16i5hL6LLzzp0CWZuA5mIXZocltUmwXWEmsrNBUsrS1IXiUNiOnS4HmaFc6ZzFctzt07 uMwchfVkA4jOYu1KBG5z+0fB9TWeQ48nsSgWTvkyJUztwoFX+ldygmPX+hYq1I3wOq4z WhWAcA5lExyTkc8jVfiEBzVcemFsHQ90YWF11JXT1o1yVAitYqzlSc2Iza5GkQW843fB 5iNQ== X-Gm-Message-State: AOAM532+EmGgvaQOXRF8OcIfirHEAC1dCkzEx7ZTD3Gx6AwZq9u3UIX9 87Chopm9j5b2sSc7YRKoa8qsb/G4+CE= X-Google-Smtp-Source: ABdhPJz43oeS5F0siXWsZh061o21n+qfPtthgPIgTIT8DyUX6v2rSoMU2ZNvcsVNXhRZGwvcQO+2wDnOY7E= X-Received: from reiji-vws-sp.c.googlers.com ([fda3:e722:ac3:cc00:7f:e700:c0a8:3d59]) (user=reijiw job=sendgmr) by 2002:a17:90a:1f4d:b0:1bb:a657:ace5 with SMTP id y13-20020a17090a1f4d00b001bba657ace5mr2283133pjy.39.1645069811410; Wed, 16 Feb 2022 19:50:11 -0800 (PST) Date: Wed, 16 Feb 2022 19:49:46 -0800 Message-Id: <20220217034947.180935-1-reijiw@google.com> Mime-Version: 1.0 X-Mailer: git-send-email 2.35.1.473.g83b2b277ed-goog Subject: [PATCH 1/2] KVM: selftests: kvm_vm_elf_load() and elfhdr_get() should close fd From: Reiji Watanabe To: Marc Zyngier , kvmarm@lists.cs.columbia.edu, kvm@vger.kernel.org Cc: linux-arm-kernel@lists.infradead.org, James Morse , Alexandru Elisei , Suzuki K Poulose , Andrew Jones , Paolo Bonzini , Will Deacon , Peter Shier , Ricardo Koller , Oliver Upton , Jing Zhang , Raghavendra Rao Anata , Reiji Watanabe Precedence: bulk List-ID: X-Mailing-List: kvm@vger.kernel.org kvm_vm_elf_load() and elfhdr_get() open one file each, but they never close the opened file descriptor. If a test repeatedly creates and destroys a VM with vm_create_with_vcpus(), which (directly or indirectly) calls those two functions, the test might end up getting a open failure with EMFILE. Fix those two functions to close the file descriptor. Signed-off-by: Reiji Watanabe Reviewed-by: Oliver Upton Reviewed-by: Andrew Jones --- tools/testing/selftests/kvm/lib/elf.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/tools/testing/selftests/kvm/lib/elf.c b/tools/testing/selftests/kvm/lib/elf.c index 13e8e3dcf984..9b23537a3caa 100644 --- a/tools/testing/selftests/kvm/lib/elf.c +++ b/tools/testing/selftests/kvm/lib/elf.c @@ -91,6 +91,7 @@ static void elfhdr_get(const char *filename, Elf64_Ehdr *hdrp) " hdrp->e_shentsize: %x\n" " expected: %zx", hdrp->e_shentsize, sizeof(Elf64_Shdr)); + close(fd); } /* VM ELF Load @@ -190,4 +191,5 @@ void kvm_vm_elf_load(struct kvm_vm *vm, const char *filename) phdr.p_filesz); } } + close(fd); } From patchwork Thu Feb 17 03:49:47 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Reiji Watanabe X-Patchwork-Id: 12749294 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id A6058C433F5 for ; Thu, 17 Feb 2022 03:50:27 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S233101AbiBQDuh (ORCPT ); Wed, 16 Feb 2022 22:50:37 -0500 Received: from mxb-00190b01.gslb.pphosted.com ([23.128.96.19]:53394 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231430AbiBQDue (ORCPT ); Wed, 16 Feb 2022 22:50:34 -0500 Received: from mail-pl1-x64a.google.com (mail-pl1-x64a.google.com [IPv6:2607:f8b0:4864:20::64a]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 1E131C5D82 for ; Wed, 16 Feb 2022 19:50:20 -0800 (PST) Received: by mail-pl1-x64a.google.com with SMTP id z3-20020a170903408300b0014d891103aaso1975694plc.0 for ; Wed, 16 Feb 2022 19:50:20 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=20210112; h=date:in-reply-to:message-id:mime-version:references:subject:from:to :cc; bh=15h5Cc5am/9/sg4AeH/YbD/6eomivPF7mAEPY+DVvcw=; b=PLgzol8si987Ws/NhcI3lLtsYeO68cXRTu2rW67velqzcdJwHwH82Su3eBLCWUz/pF 4si6XGZzb/WuboiQuMXL5MlOH8U9OIjmQJW3WoFRZKWmrUmzMkO7R9WOn4ErhLE+KuSw ip6rEsWQkSaZzc73JS5Zlbbk5r471XySFD8Vp7iFUeWdHB96p6cYF2iXuNfdpRwG9T1S UkFf2sET0Qm8WPgJWGnpXw6EaaIQznLTsyMJr4NQtKtOuqyV7psWP9iVQNLmnMotNBl5 B6xmNFZf65rWahApnbIthnMy8hD7BrtLm/B6vq1FWuc2ZOpqD4lXNQoKqr4cWYImJbzK qk5w== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:date:in-reply-to:message-id:mime-version :references:subject:from:to:cc; bh=15h5Cc5am/9/sg4AeH/YbD/6eomivPF7mAEPY+DVvcw=; b=zwhXRTH845bRmW8I596ODlNTdnPrPxVFXzBz72UteE+PXTzldlucxPVgggz9RN8nOb Yh2vKEeL5ovE3MLa9W11vdYGsXNKhNo9gyrK0JbzpeGJrVp0fePCBGOur92ZfRLNJQV1 pO35aBqc+3RWviB39eLE/VeGFr7GuhLodHlG+1yumGrjHyXdMKeQNpXIapXoMu9V3dpF uOmIIY7xyextKSguNwIfzQaKx6XExE+JX2EQVG4MsmZbKkxQxDp8vCqOdRYBzzYsIaEm Gckr5gfGuyN0J9H6NovOWSyDxIPKwbwjME5D9X8qRlwew7Zs7T4JU3nw4KSq2RvXUjPg RDtw== X-Gm-Message-State: AOAM530InmZwXzcju5um3NK4YeAmF4D9Ph28Asq96OOlPD2uwrfdXh5p sviyJ7Ino1tRmzUA9hCi1GcC/r/N9qo= X-Google-Smtp-Source: ABdhPJy1O6nYrCsqLhs9diRaHgJ1EGmTNdENlDbQoBBOIYnwD6JQdGW34f5aA9jEEgWeivDVBRht51xus0Q= X-Received: from reiji-vws-sp.c.googlers.com ([fda3:e722:ac3:cc00:7f:e700:c0a8:3d59]) (user=reijiw job=sendgmr) by 2002:a17:90a:3b05:b0:1b8:af52:e77 with SMTP id d5-20020a17090a3b0500b001b8af520e77mr5366514pjc.180.1645069819634; Wed, 16 Feb 2022 19:50:19 -0800 (PST) Date: Wed, 16 Feb 2022 19:49:47 -0800 In-Reply-To: <20220217034947.180935-1-reijiw@google.com> Message-Id: <20220217034947.180935-2-reijiw@google.com> Mime-Version: 1.0 References: <20220217034947.180935-1-reijiw@google.com> X-Mailer: git-send-email 2.35.1.473.g83b2b277ed-goog Subject: [PATCH 2/2] KVM: arm64: selftests: Introduce get_set_regs_perf test From: Reiji Watanabe To: Marc Zyngier , kvmarm@lists.cs.columbia.edu, kvm@vger.kernel.org Cc: linux-arm-kernel@lists.infradead.org, James Morse , Alexandru Elisei , Suzuki K Poulose , Andrew Jones , Paolo Bonzini , Will Deacon , Peter Shier , Ricardo Koller , Oliver Upton , Jing Zhang , Raghavendra Rao Anata , Reiji Watanabe Precedence: bulk List-ID: X-Mailing-List: kvm@vger.kernel.org Introduce a simple performance test of KVM_GET_ONE_REG/KVM_SET_ONE_REG for registers that are returned by KVM_GET_REG_LIST. This is a pseudo process of saving/restoring registers during live migration, and this test quantifies the performance of the process. Signed-off-by: Reiji Watanabe --- tools/testing/selftests/kvm/.gitignore | 1 + tools/testing/selftests/kvm/Makefile | 1 + .../selftests/kvm/aarch64/get_set_regs_perf.c | 456 ++++++++++++++++++ 3 files changed, 458 insertions(+) create mode 100644 tools/testing/selftests/kvm/aarch64/get_set_regs_perf.c diff --git a/tools/testing/selftests/kvm/.gitignore b/tools/testing/selftests/kvm/.gitignore index dce7de7755e6..b6b18aaf9322 100644 --- a/tools/testing/selftests/kvm/.gitignore +++ b/tools/testing/selftests/kvm/.gitignore @@ -2,6 +2,7 @@ /aarch64/arch_timer /aarch64/debug-exceptions /aarch64/get-reg-list +/aarch64/get_set_regs_perf /aarch64/psci_cpu_on_test /aarch64/vgic_init /aarch64/vgic_irq diff --git a/tools/testing/selftests/kvm/Makefile b/tools/testing/selftests/kvm/Makefile index 0e4926bc9a58..46a28ce89002 100644 --- a/tools/testing/selftests/kvm/Makefile +++ b/tools/testing/selftests/kvm/Makefile @@ -103,6 +103,7 @@ TEST_GEN_PROGS_x86_64 += system_counter_offset_test TEST_GEN_PROGS_aarch64 += aarch64/arch_timer TEST_GEN_PROGS_aarch64 += aarch64/debug-exceptions TEST_GEN_PROGS_aarch64 += aarch64/get-reg-list +TEST_GEN_PROGS_aarch64 += aarch64/get_set_regs_perf TEST_GEN_PROGS_aarch64 += aarch64/psci_cpu_on_test TEST_GEN_PROGS_aarch64 += aarch64/vgic_init TEST_GEN_PROGS_aarch64 += aarch64/vgic_irq diff --git a/tools/testing/selftests/kvm/aarch64/get_set_regs_perf.c b/tools/testing/selftests/kvm/aarch64/get_set_regs_perf.c new file mode 100644 index 000000000000..27944ccce4a1 --- /dev/null +++ b/tools/testing/selftests/kvm/aarch64/get_set_regs_perf.c @@ -0,0 +1,456 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * get_set_regs_perf - KVM_GET_ONE_REG/KVM_SET_ONE_REG performance test + * + * The test runs performance test of saving registers that are returned + * by KVM_GET_REG_LIST for vCPUs in a VM and restoring them for vCPUs + * in another VM (a part of behavior during blackout on live migration). + * + * Copyright (c) 2022, Google LLC. + */ + +#define _GNU_SOURCE + +#include +#include +#include +#include +#include +#include + +#include "kvm_util.h" +#include "processor.h" + +#define N2USEC(nsec) ((nsec)/1000) +#define REG_STR_LEN 64 + +struct reg_val { + /* Big enough buffer even for the longest vector register */ + uint64_t val[32]; +}; + +struct thread_data { + uint32_t vcpuid; + bool is_set_reg; + volatile bool *start; + struct kvm_vm *vm; + struct timespec time; + struct kvm_reg_list *reg_list; + + /* + * Pointer to an array of reg_val, which will hold a value of each + * register. The number of elements in it is indicated in reg_list.n. + */ + struct reg_val *reg_vals; + + /* + * Pointer to an array of timespec, which will hold per register + * latency of KVM_GET_ONE_REG or KVM_SET_ONE_REG. The number of + * elements in it is indicated in reg_list.n. + */ + struct timespec *reg_time; +}; + +pthread_t threads[KVM_MAX_VCPUS]; +struct thread_data thread_args[KVM_MAX_VCPUS]; + +struct sys_reg_params { + u8 Op0; + u8 Op1; + u8 CRn; + u8 CRm; + u8 Op2; +}; + +static int regid_to_params(u64 id, struct sys_reg_params *params) +{ + switch (id & KVM_REG_SIZE_MASK) { + case KVM_REG_SIZE_U64: + /* Any unused index bits means it's not valid. */ + if (id & ~(KVM_REG_ARCH_MASK | KVM_REG_SIZE_MASK + | KVM_REG_ARM_COPROC_MASK + | KVM_REG_ARM64_SYSREG_OP0_MASK + | KVM_REG_ARM64_SYSREG_OP1_MASK + | KVM_REG_ARM64_SYSREG_CRN_MASK + | KVM_REG_ARM64_SYSREG_CRM_MASK + | KVM_REG_ARM64_SYSREG_OP2_MASK)) + return -1; + params->Op0 = ((id & KVM_REG_ARM64_SYSREG_OP0_MASK) + >> KVM_REG_ARM64_SYSREG_OP0_SHIFT); + params->Op1 = ((id & KVM_REG_ARM64_SYSREG_OP1_MASK) + >> KVM_REG_ARM64_SYSREG_OP1_SHIFT); + params->CRn = ((id & KVM_REG_ARM64_SYSREG_CRN_MASK) + >> KVM_REG_ARM64_SYSREG_CRN_SHIFT); + params->CRm = ((id & KVM_REG_ARM64_SYSREG_CRM_MASK) + >> KVM_REG_ARM64_SYSREG_CRM_SHIFT); + params->Op2 = ((id & KVM_REG_ARM64_SYSREG_OP2_MASK) + >> KVM_REG_ARM64_SYSREG_OP2_SHIFT); + return 0; + default: + return -1; + } +} + +static void reg_id_to_str(uint64_t id, char *str, int len) +{ + struct sys_reg_params p; + + TEST_ASSERT((id & KVM_REG_ARCH_MASK) == KVM_REG_ARM64, + "Not KVM_REG_ARM64 register (0x%lx)", id); + + switch (id & KVM_REG_ARM_COPROC_MASK) { + case KVM_REG_ARM_CORE: + snprintf(str, len, "CORE(id:0x%lx)", id); + break; + case KVM_REG_ARM64_SVE: + snprintf(str, len, "SVE(id:0x%lx)", id); + break; + case KVM_REG_ARM_DEMUX: + snprintf(str, len, "DEMUX(id:0x%lx)", id); + break; + case KVM_REG_ARM_FW: + snprintf(str, len, "FW_REG_%ld", id & 0xffff); + break; + case KVM_REG_ARM64_SYSREG: + if (regid_to_params(id, &p) == 0) + snprintf(str, len, "SYSREG(%d,%d,%d,%d,%d)", + p.Op0, p.Op1, p.CRn, p.CRm, p.Op2); + else + snprintf(str, len, "SYSREG(id:0x%lx)", id); + break; + default: + snprintf(str, len, "KVM_REG_ARM64(id:0x%lx)", id); + } +} + +/* + * Save or restore registers that were saved by get_regs_vcpu_thread with + * KVM_GET_ONE_REG or KVM_SET_ONE_REG. Measure time taken by this process. + */ +static void *vcpu_thread(void *data) +{ + int i, ret; + struct thread_data *arg = data; + struct kvm_reg_list *reg_list = arg->reg_list; + bool is_set_reg = arg->is_set_reg; + unsigned int cmd = is_set_reg ? KVM_SET_ONE_REG : KVM_GET_ONE_REG; + char *cmd_str = is_set_reg ? "SET_ONE_REG" : "GET_ONE_REG"; + struct timespec ts_reg_diff, ts_reg_start; + struct kvm_one_reg reg; + + memset(&arg->time, 0, sizeof(arg->time)); + + /* Wait until start is set */ + while (!(*arg->start)) + ; + + /* Warm up the cache. */ + reg.id = reg_list->reg[0]; + reg.addr = (uint64_t)&arg->reg_vals[0]; + ret = _vcpu_ioctl(arg->vm, arg->vcpuid, cmd, ®); + TEST_ASSERT(!ret, "%s(id:0x%llx) failed\n", cmd_str, reg.id); + + pr_debug("%s[%d] start\n", __func__, arg->vcpuid); + + /* Run KVM_SET_ONE_REG or KVM_GET_ONE_REG for registers in reg_list */ + for (i = 0; i < reg_list->n; i++) { + clock_gettime(CLOCK_MONOTONIC, &ts_reg_start); + + reg.id = reg_list->reg[i]; + reg.addr = (uint64_t)&arg->reg_vals[i]; + ret = _vcpu_ioctl(arg->vm, arg->vcpuid, cmd, ®); + + ts_reg_diff = timespec_elapsed(ts_reg_start); + arg->time = timespec_add(arg->time, ts_reg_diff); + if (arg->reg_time) + /* Need per register latency separately */ + arg->reg_time[i] = ts_reg_diff; + + TEST_ASSERT(!ret, "%s(id:0x%llx) failed\n", cmd_str, reg.id); + pr_debug("%s: id:0x%lx, val:0x%lx\n", cmd_str, (uint64_t)reg.id, + arg->reg_vals[i].val[0]); + } + + return NULL; +} + +void aggregate_time(uint64_t cnt, struct timespec *total, struct timespec *data) +{ + uint64_t i; + + for (i = 0; i < cnt; i++) + total[i] = timespec_add(total[i], data[i]); +} + +struct result_timespec { + struct timespec src; + struct timespec dst; + struct timespec *src_per_reg; + struct timespec *dst_per_reg; +}; + +/* + * Run KVM_GET_ONE_REG/KVM_SET_ONE_REG performance test. + * Create a VM with the given number of vCPUs, create a thread for each + * vCPU in the VM, and wait until all the threads complete KVM_GET_ONE_REG + * for all the registers KVM_GET_REG_LIST returns. Then repeat the same thing + * with KVM_SET_ONE_REG. Aggregate the time spent by each thread on + * KVM_GET_ONE_REG and KVM_SET_ONE_REG, and return the vCPU average value to + * the caller via @res. + */ +void run_test_one(int nvcpus, struct kvm_reg_list *reg_list, + struct result_timespec *res) +{ + bool per_reg_data = res->src_per_reg ? true : false; + uint64_t nregs = reg_list->n; + volatile bool start = false; + int i, ret; + struct kvm_vm *src_vm, *dst_vm; + struct thread_data *targ; + struct timespec src_sum = (struct timespec){0}; + struct timespec dst_sum = (struct timespec){0}; + + if (per_reg_data) { + /* Reset the buffers to save the per register result */ + memset(res->src_per_reg, 0, nregs * sizeof(struct timespec)); + memset(res->dst_per_reg, 0, nregs * sizeof(struct timespec)); + } + + /* Create VMs to save/restore registers */ + src_vm = vm_create_default_with_vcpus(nvcpus, 0, 0, NULL, NULL); + dst_vm = vm_create_default_with_vcpus(nvcpus, 0, 0, NULL, NULL); + + /* Start saving registers process for src_vm */ + + /* Create source VM's vCPU threads */ + for (i = 0; i < nvcpus; i++) { + targ = &thread_args[i]; + targ->vm = src_vm; + targ->vcpuid = i; + targ->is_set_reg = false; + targ->start = &start; + targ->reg_list = reg_list; + targ->reg_vals = calloc(nregs, sizeof(struct reg_val)); + TEST_ASSERT(targ->reg_vals, "Failed to allocate reg_vals"); + if (per_reg_data) { + /* Per register result buffer for the vCPU */ + targ->reg_time = calloc(nregs, sizeof(struct timespec)); + TEST_ASSERT(targ->reg_time, "Failed to allocate reg_time"); + } + ret = pthread_create(&threads[i], NULL, vcpu_thread, targ); + TEST_ASSERT(!ret, "pthread_create failed: %d\n", ret); + } + + /* + * Let threads start saving registers for vCPUs, and wait for all + * threads to complete restoring registers. + */ + start = true; + for (i = 0; i < nvcpus; i++) { + targ = &thread_args[i]; + ret = pthread_join(threads[i], NULL); + TEST_ASSERT(!ret, "pthread_join failed: %d\n", ret); + src_sum = timespec_add(src_sum, targ->time); + if (per_reg_data) + aggregate_time(nregs, res->src_per_reg, targ->reg_time); + } + + start = false; + pr_debug("%s Saving registers completed.\n", __func__); + + /* Start restoring registers process for dst_vm */ + + /* Create destination VM's vCPU threads */ + for (i = 0; i < nvcpus; i++) { + targ = &thread_args[i]; + /* Update fields that are different from the src case */ + targ->vm = dst_vm; + targ->is_set_reg = true; + ret = pthread_create(&threads[i], NULL, vcpu_thread, targ); + TEST_ASSERT(!ret, "pthread_create failed: %d\n", ret); + } + + /* + * Let threads start saving registers for vCPUs, and wait for all + * threads to complete restoring registers. + */ + start = true; + for (i = 0; i < nvcpus; i++) { + targ = &thread_args[i]; + ret = pthread_join(threads[i], NULL); + TEST_ASSERT(!ret, "pthread_join failed: %d\n", ret); + + free(targ->reg_vals); + dst_sum = timespec_add(dst_sum, targ->time); + if (per_reg_data) { + aggregate_time(nregs, res->dst_per_reg, targ->reg_time); + free(targ->reg_time); + } + } + + kvm_vm_free(src_vm); + kvm_vm_free(dst_vm); + + pr_debug("%s Restoring registers completed.\n", __func__); + + /* Calculate the vCPU average */ + res->src = timespec_div(src_sum, nvcpus); + res->dst = timespec_div(dst_sum, nvcpus); + if (!per_reg_data) + return; + + /* Calculate the vCPU average for each register */ + for (i = 0; i < nregs; i++) { + res->src_per_reg[i] = timespec_div(res->src_per_reg[i], nvcpus); + res->dst_per_reg[i] = timespec_div(res->dst_per_reg[i], nvcpus); + } +} + +/* + * Run saving/restoring vCPU registers (KVM_GET_ONE_REG/KVM_SET_ONE_REG) + * performance test with the given number of vCPUs (@nvcpus) for the given + * number (@iterations) of times. + * Each iteration of the test returns the latency of saving/restoring all + * the registers that KVM_GET_REG_LIST returns (See comments for + * run_test_one() for more details of each iteration of the test). + * Print the average latency of each iteration, and if @per_reg_result is set, + * print the average latency for each register as well. + */ +void run_test(int nvcpus, int iterations, bool per_reg_result) +{ + int i, j; + uint64_t nregs; + struct result_timespec res = {0}; + struct timespec src_sum = {0}; + struct timespec dst_sum = {0}; + struct timespec avg, src_avg, dst_avg; + struct timespec *src_per_reg_sum, *dst_per_reg_sum; + struct kvm_vm *vm; + struct kvm_reg_list *reg_list; + char reg_str[REG_STR_LEN]; + + /* Get kvm_reg_list */ + vm = vm_create_default(0, 0, NULL); + reg_list = vcpu_get_reg_list(vm, 0); + kvm_vm_free(vm); + nregs = reg_list->n; + + if (per_reg_result) { + /* + * Allocate buffers to save latency of KVM_GET_ONE_REG + * and KVM_SET_ONE_REG respectively. + */ + + /* Buffers for each iteration */ + res.src_per_reg = calloc(nregs, sizeof(struct timespec)); + res.dst_per_reg = calloc(nregs, sizeof(struct timespec)); + + /* Buffers for the sum */ + src_per_reg_sum = calloc(nregs, sizeof(struct timespec)); + dst_per_reg_sum = calloc(nregs, sizeof(struct timespec)); + + TEST_ASSERT(res.src_per_reg && res.dst_per_reg && + src_per_reg_sum && dst_per_reg_sum, + "Failed to allocate per register time buffer"); + } + + pr_info("Iterations %d (vCPUs %d, regs %lld): ", + iterations, nvcpus, reg_list->n); + + for (i = 0; i < iterations; i++) { + /* Run the test */ + run_test_one(nvcpus, reg_list, &res); + + src_sum = timespec_add(src_sum, res.src); + dst_sum = timespec_add(dst_sum, res.dst); + + if (!per_reg_result) + continue; + + /* Aggregate per register result */ + for (j = 0; j < nregs; j++) { + src_per_reg_sum[j] = timespec_add(src_per_reg_sum[j], + res.src_per_reg[j]); + dst_per_reg_sum[j] = timespec_add(dst_per_reg_sum[j], + res.dst_per_reg[j]); + } + } + + /* Calculate the iteration average */ + src_avg = timespec_div(src_sum, iterations); + dst_avg = timespec_div(dst_sum, iterations); + avg = timespec_add(src_avg, dst_avg); + + /* Print the average */ + if (avg.tv_sec > 0) { + pr_info("get+set %ld.%.9lds (get %ld.%.9lds, set %ld.%.9lds)\n", + avg.tv_sec, avg.tv_nsec, + src_avg.tv_sec, src_avg.tv_nsec, + dst_avg.tv_sec, dst_avg.tv_nsec); + } else { + pr_info("get+set %ldus (get %ldus, set %ldus)\n", + N2USEC(avg.tv_nsec), N2USEC(src_avg.tv_nsec), + N2USEC(dst_avg.tv_nsec)); + } + + /* Print per register result when requested */ + if (per_reg_result) { + pr_info("Per Register Result:\n"); + pr_info(" %-32s %13s %9s %9s\n", "Register", "get+set(ns)", "get(ns)", "set(ns)"); + for (i = 0; i < nregs; i++) { + /* Calculate the average for the register */ + src_avg = timespec_div(src_per_reg_sum[i], iterations); + dst_avg = timespec_div(dst_per_reg_sum[i], iterations); + avg = timespec_add(src_avg, dst_avg); + + assert(avg.tv_sec == 0); + reg_id_to_str(reg_list->reg[i], reg_str, REG_STR_LEN); + pr_info(" %-32s %13ld %9ld %9ld\n", reg_str, + avg.tv_nsec, src_avg.tv_nsec, dst_avg.tv_nsec); + } + + free(res.src_per_reg); + free(res.dst_per_reg); + free(src_per_reg_sum); + free(dst_per_reg_sum); + } + + free(reg_list); +} + +int main(int argc, char *argv[]) +{ + int opt; + int nvcpus = 1; + int max_vcpus = kvm_check_cap(KVM_CAP_MAX_VCPUS); + int iterations = 1000; + bool per_reg_result = false; + + while ((opt = getopt(argc, argv, "hri:v:")) != -1) { + switch (opt) { + case 'i': + iterations = atoi(optarg); + break; + case 'v': + nvcpus = atoi(optarg); + TEST_ASSERT(nvcpus > 0 && nvcpus <= max_vcpus, + "Number of vcpus, must be between 1 and %d", + max_vcpus); + break; + case 'r': + per_reg_result = true; + break; + case 'h': + default: + printf("usage: %s [-h] [-v vcpus] [-i iterations]\n", + argv[0]); + exit(0); + } + } + + setbuf(stdout, NULL); + + run_test(nvcpus, iterations, per_reg_result); + + return 0; +}