如何查找物理cpu，cpu核心和逻辑cpu的数量

环境

• Red Hat Enterprise Linux 4
• Red Hat Enterprise Linux 5
• Red Hat Enterprise Linux 6
• Red Hat Enterprise Linux 7

问题

• x86 / x86_64系统上的物理CPU，CPU内核和逻辑CPU之间有什么区别？
• 如何从RHEL操作系统获取可用处理器列表?

决议

• 物理CPU数量

  • 下面的命令将显示一个系统有多少活动的物理处理器。例如:如果这个数字是2，则可以打开系统机箱，用手删除2个物理处理器。

     

    $ grep physical.id /proc/cpuinfo | sort -u | wc -l
    2

• 每个CPU的核心数

  • 在具有多核处理器的系统上，下面的命令应该报告每个物理处理器的CPU内核数量(尽管在极少数情况下可能不会)。例如:如果这个数字是4，而物理CPU是2，那么这2个物理处理器中的每个都有4个CPU内核，总共有8个内核。

     

    $ grep cpu.cores /proc/cpuinfo | sort -u
    cpu cores : 4

• 逻辑处理器数量

  • 最后这个命令将显示Linux内核所看到的“逻辑”处理器的总数。这个数字通常是三个统计数据中最重要的。它是处理器的有效数量——就操作系统而言，这是能够在任何给定微秒内工作的不同cpu的数量。示例:继续上面的示例场景，下面看到的数字可以是16而不是8。简单地说，如果这个命令显示的数字与CPU内核的总数不同，这是因为在CPU上启用了超线程，从而进一步划分了每个内核(在本例中，分为两个可用的“线程”)。

     

    $ grep processor /proc/cpuinfo | wc -l
    16

许可?

• 为了确定RHEL许可原因的cpu数量，上面的第一个命令就足够了;但是，要进行更多的讨论，请咨询 如何确定系统上的CPU插槽数

• 如果购买的许可根据插槽（已填充或未填充），处理器或内核的数量而变化的第三方软件，请与软件供应商联系，以确切了解它们如何计算CPU数量。

工具

• 注意，Red Hat Enterprise Linux 6和7附带了lscpu命令，该命令可以打印出系统处理器的简单可读摘要。此外，在RHEL6中，hwloc包是可用的(RHEL7附带它)——hwloc包括lstopo命令，以及各种hwloc-*命令。

• 在RHEL6之前，直接检查/proc/cpuinfo(如上所述)或使用x86info或dmidecode命令(同名rpm)都能查到相同的信息。

• 其他第三方选项没有提供担保:xsos(由本文的原始作者开发)和lshw(可在EPEL中获得)。

• 对于任何版本的Red Hat Enterprise Linux:本文附带的BASH脚本都会解析/proc/cpuinfo，以打印如下所示的易于阅读的摘要。(注意，它没有提供任何保证或官方支持。)
  
  Raw

  $ cpu
  128 logical processors (64 CPU cores)
  8 Intel Xeon CPU X7560 @ 2.27GHz (flags: constant_tsc,ht,lm,pae,vmx) 
  └─16 threads / 8 cores each

  #!/bin/bash
  #
  # This simple script uses /proc/cpuinfo (or filename of your choosing) to print
  # a succinct summary about a system's processors.
  # Other useful utilities (some only available in RHEL6 or EPEL):
  #   x86info, dmidecode, lscpu, cpuid, lshw, lstopo, xsos
  #
  # Originally uploaded to redhat.com by Ryan Sawhill <rsaw@redhat.com>, Sep 2012; Updated Jan 2013
  # This code is from xsos, which can do so much more <http://github.com/ryran/xsos>
  #
  
  # Get input
  if [[ -r $1 && -f $1 ]]; then
    # If passed a readable file, use that
    cpuinfo=$1
  else
    # Otherwise, use /proc/cpuinfo
    cpuinfo=/proc/cpuinfo
  fi
  
  # Get model of cpu
  model_cpu=$(awk -F: '/^model name/{print $2; exit}' <"$cpuinfo")
  
  # If no model detected (e.g. on Itanium), try to use vendor+family
  [[ -z $model_cpu ]] && {
    vendor=$(awk -F: '/^vendor /{print $2; exit}' <"$cpuinfo")
    family=$(awk -F: '/^family /{print $2; exit}' <"$cpuinfo")
    model_cpu="$vendor$family"
  }
  
  # Clean up cpu model string
  model_cpu=$(sed -e 's,(R),,g' -e 's,(TM),,g' -e 's,  *, ,g' -e 's,^ ,,' <<<"$model_cpu")
  
  # Get number of logical processors
  num_cpu=$(awk '/^processor/{n++} END{print n}' <"$cpuinfo")
  
  # Get number of physical processors
  num_cpu_phys=$(grep '^physical id' <"$cpuinfo" | sort -u | wc -l)
  
  # If "physical id" not found, we cannot make any assumptions (Virtualization--)
  # But still, multiplying by 0 in some crazy corner case is bad, so set it to 1
  # If num of physical *was* detected, add it to the beginning of the model string
  [[ $num_cpu_phys == 0 ]] && num_cpu_phys=1 || model_cpu="$num_cpu_phys $model_cpu"
  
  # If number of logical != number of physical, try to get info on cores & threads
  if [[ $num_cpu != $num_cpu_phys ]]; then
    
    # Detect number of threads (logical) per cpu
    num_threads_per_cpu=$(awk '/^siblings/{print $3; exit}' <"$cpuinfo")
    
    # Two possibile ways to detect number of cores
    cpu_cores=$(awk '/^cpu cores/{print $4; exit}' <"$cpuinfo")
    core_id=$(grep '^core id' <"$cpuinfo" | sort -u | wc -l)
    
    # The first is the most accurate, if it works
    if [[ -n $cpu_cores ]]; then
      num_cores_per_cpu=$cpu_cores
    
    # If "cpu cores" doesn't work, "core id" method might (e.g. Itanium)
    elif [[ $core_id -gt 0 ]]; then
      num_cores_per_cpu=$core_id
    fi
    
    # If found info on cores, setup core variables for printing
    if [[ -n $num_cores_per_cpu ]]; then
      cores1="($((num_cpu_phys*num_cores_per_cpu)) CPU cores)"
      cores2=" / $num_cores_per_cpu cores"
    # If didn't find info on cores, assume single-core cpu(s)
    else
      cores2=" / 1 core"
    fi
    
    # If found siblings (threads), setup the variable for the final line
    [[ -n $num_threads_per_cpu ]] &&
      coresNthreads="
  └─$num_threads_per_cpu threads${cores2} each"
  fi
  
  # Check important cpu flags
  # pae=physical address extensions  *  lm=64-bit  *  vmx=Intel hw-virt  *  svm=AMD hw-virt
  # ht=hyper-threading  *  aes=AES-NI  *  constant_tsc=Constant Time Stamp Counter
  cpu_flags=$(egrep -o "pae|lm|vmx|svm|ht|aes|constant_tsc" <"$cpuinfo" | sort -u | sed ':a;N;$!ba;s/
  /,/g')
  [[ -n $cpu_flags ]] && cpu_flags="(flags: $cpu_flags)"
  
  # Check kernel version; print warning if Xen
  [[ $(uname -r) =~ xen ]] && {
    echo "Warning: kernel for localhost detected as $(uname -r)"
    echo "With Xen, CPU layout in /proc/cpuinfo will be inaccurate; consult dmidecode"
  }
  
  # Print out the deets
  echo -e "${num_cpu} logical processors ${cores1}"
  echo -e "${model_cpu} ${cpu_flags} ${coresNthreads}"