--- qemu-2.4.1/vl.c 2015-11-04 01:31:39.000000000 +0530 +++ qemu-2.4.1_aff/vl.c 2016-01-21 21:36:21.824950348 +0530 @@ -126,6 +126,7 @@ #define MAX_VIRTIO_CONSOLES 1 #define MAX_SCLP_CONSOLES 1 +#define MAX_VCPUS 32 static const char *data_dir[16]; static int data_dir_idx; const char *bios_name = NULL; @@ -162,6 +163,8 @@ int max_cpus = 0; int smp_cores = 1; int smp_threads = 1; +int vcpu_affinity[MAX_VCPUS]; +int num_affinity = 0; int acpi_enabled = 1; int no_hpet = 0; int fd_bootchk = 1; @@ -1194,6 +1197,61 @@ }, }; +static QemuOptsList qemu_vcpu_opts = { + .name = "vcpu-opts", + .implied_opt_name = "vcpunum", + .head = QTAILQ_HEAD_INITIALIZER(qemu_vcpu_opts.head), + .desc = { + { + .name = "vcpunum", + .type = QEMU_OPT_NUMBER, + }, { + .name = "affinity", + .type = QEMU_OPT_NUMBER, + }, + { /*End of list */ } + }, +}; + +static int parse_vcpu(void *opaque, QemuOpts *opts, Error **errp) +{ + if (opts){ + unsigned vcpu = qemu_opt_get_number(opts,"vcpunum",0); + unsigned affinity = qemu_opt_get_number(opts,"affinity",0); + + if (vcpu < smp_cores) { + if (vcpu_affinity[vcpu] == -1){ + vcpu_affinity[vcpu] = affinity; + } + else { + error_report("Duplicate affinity statement for vcpu %d\n",vcpu); + return -1; + } + num_affinity += 1; + } + else { + error_report("VCPU %d is more than allowed %d VCPUs in the system\n",vcpu,smp_cores); + return -1; + } + } + return 0; +} + +static void parse_vcpu_opts(MachineClass *mc) +{ + int i; + for (i = 0; i < MAX_VCPUS; i++) + vcpu_affinity[i] = -1; + + if (qemu_opts_foreach(qemu_find_opts("vcpu-opts"), parse_vcpu, NULL, NULL)) { + exit(1); + } + + if (num_affinity != smp_cores) { + fprintf(stderr,"Warning: affinity is not specified for some (%d) of the %d smp cores. Will be using default mask\n",(smp_cores-num_affinity), smp_cores); + } +} + static void smp_parse(QemuOpts *opts) { if (opts) { @@ -2954,6 +3012,9 @@ Error *main_loop_err = NULL; Error *err = NULL; + printf("*****************************************************************\n"); + printf("** QEMU - MODIFIED FOR CPU AFFINITY **\n"); + printf("*****************************************************************\n"); qemu_init_cpu_loop(); qemu_mutex_lock_iothread(); @@ -2979,6 +3040,7 @@ qemu_add_opts(&qemu_machine_opts); qemu_add_opts(&qemu_mem_opts); qemu_add_opts(&qemu_smp_opts); + qemu_add_opts(&qemu_vcpu_opts); qemu_add_opts(&qemu_boot_opts); qemu_add_opts(&qemu_sandbox_opts); qemu_add_opts(&qemu_add_fd_opts); @@ -3683,6 +3745,12 @@ exit(1); } break; + case QEMU_OPTION_vcpu: + if (!qemu_opts_parse_noisily(qemu_find_opts("vcpu-opts"), + optarg, true)) { + exit(1); + } + break; case QEMU_OPTION_vnc: { #ifdef CONFIG_VNC @@ -4097,6 +4165,7 @@ exit(1); } + parse_vcpu_opts(machine_class); /* * Get the default machine options from the machine if it is not already * specified either by the configuration file or by the command line. --- qemu-2.4.1/cpus.c 2015-11-04 01:31:31.000000000 +0530 +++ qemu-2.4.1_aff/cpus.c 2016-01-21 21:36:21.976950357 +0530 @@ -50,6 +50,9 @@ #ifdef CONFIG_LINUX #include +#include +#include +#include #ifndef PR_MCE_KILL #define PR_MCE_KILL 33 @@ -1277,10 +1280,39 @@ } } +#ifdef QEMU_AFFINITY_DEBUG +static void print_cpu_mask(pthread_t t,int cpuindex,int flag) +{ + int i = 0; + cpu_set_t cpuset; + uint64_t mask_low = 0; + uint64_t mask_high = 0; + CPU_ZERO(&cpuset); + /* Get the current CPU Mask */ + pthread_getaffinity_np(t, sizeof(cpu_set_t), &cpuset); + + for (i = 0; i< 128; i++){ + if (CPU_ISSET(i, &cpuset)){ + if (i < 64) + mask_low = mask_low | ((uint64_t)1 << i); + else + mask_high = mask_high | ((uint64_t)1 << (i-64)); + } + } + + if (flag == 0) + printf ("QEMU DEBUG:: Original CPU mask 0x%016"PRIx64"%016"PRIx64" for VCPU %d\n",mask_high,mask_low,cpuindex); + else + printf ("QEMU DEBUG:: New CPU mask 0x%016"PRIx64"%016"PRIx64" for VCPU %d\n",mask_high,mask_low,cpuindex); +} +#endif + +extern int vcpu_affinity[]; static void qemu_kvm_start_vcpu(CPUState *cpu) { char thread_name[VCPU_THREAD_NAME_SIZE]; - + cpu_set_t cpuset; + cpu->thread = g_malloc0(sizeof(QemuThread)); cpu->halt_cond = g_malloc0(sizeof(QemuCond)); qemu_cond_init(cpu->halt_cond); @@ -1288,6 +1320,20 @@ cpu->cpu_index); qemu_thread_create(cpu->thread, thread_name, qemu_kvm_cpu_thread_fn, cpu, QEMU_THREAD_JOINABLE); + +#ifdef QEMU_AFFINITY_DEBUG + print_cpu_mask((cpu->thread)->thread,cpu->cpu_index,0); +#endif + if (vcpu_affinity[cpu->cpu_index] != -1) { + printf("QEMU INFO:: Pinning vcpu %d to Physical CPU %d\n",cpu->cpu_index, vcpu_affinity[cpu->cpu_index]); + CPU_ZERO(&cpuset); + CPU_SET(vcpu_affinity[cpu->cpu_index], &cpuset); + pthread_setaffinity_np((cpu->thread)->thread, sizeof(cpu_set_t), &cpuset); +#ifdef QEMU_AFFINITY_DEBUG + print_cpu_mask((cpu->thread)->thread,cpu->cpu_index,1); +#endif + } + while (!cpu->created) { qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex); } --- qemu-2.4.1/qemu-options.hx 2015-11-04 01:31:35.000000000 +0530 +++ qemu-2.4.1_aff/qemu-options.hx 2016-01-21 21:36:22.369950380 +0530 @@ -113,6 +113,16 @@ specifies the maximum number of hotpluggable CPUs. ETEXI +DEF("vcpu", HAS_ARG, QEMU_OPTION_vcpu, + "-vcpu [vcpunum=]n[,affinity=affinity]\n" + "-vcpu [vcpunum=]n[,affinity=affinity]\n", QEMU_ARCH_ALL) +STEXI +@item -vcpu [vcpunum=]@var{n}[,affinity=@var{affinity}] +@itemx -vcpu [vcpunum=]@var{n}[,affinity=@var{affinity}] +@findex -vcpu +VCPU Affinity. If specified, specify for all the CPUs. +ETEXI + DEF("numa", HAS_ARG, QEMU_OPTION_numa, "-numa node[,mem=size][,cpus=cpu[-cpu]][,nodeid=node]\n" "-numa node[,memdev=id][,cpus=cpu[-cpu]][,nodeid=node]\n", QEMU_ARCH_ALL)