DPDK 21.11编译指定cascadelake时VM误检测AVX512致崩溃

我使用以下命令编译DPDK 21.11：
因服务器型号不同，必须指定-Dcpu_instruction_set参数。

编译命令：

meson -Dcpu_instruction_set=cascadelake -Dprefix=/ build

但在部分虚拟机（VM）中，尽管实际硬件不支持AVX512指令集，编译时仍会选中AVX512；不指定该参数时编译正常。目前用该参数编译后的应用会崩溃，请问误检测AVX512标志的原因是什么？

相关检测信息 #

编译阶段指令集检测输出 #

Checking if "AVX512 checking" compiles: YES
Fetching value of define "__SSE4_2__" : 1
Fetching value of define "__AES__" : 1
Fetching value of define "__AVX__" : 1
Fetching value of define "__AVX2__" : 1
Fetching value of define "__AVX512BW__" : 1
Fetching value of define "__AVX512CD__" : 1
Fetching value of define "__AVX512DQ__" : 1
Fetching value of define "__AVX512F__" : 1
Fetching value of define "__AVX512VL__" : 1
Fetching value of define "__PCLMUL__" : 1
Fetching value of define "__RDRND__" : 1
Fetching value of define "__RDSEED__" : 1
Fetching value of define "__VPCLMULQDQ__" :

CPUID实际指令集检测结果 #

cpuid | grep AVX512
      AVX512F: AVX-512 foundation instructions = false
      AVX512DQ: double & quadword instructions = false
      AVX512IFMA: fused multiply add           = false
      AVX512PF: prefetch instructions          = false
      AVX512ER: exponent & reciprocal instrs   = false
      AVX512CD: conflict detection instrs      = false
      AVX512BW: byte & word instructions       = false
      AVX512VL: vector length                  = false
      AVX512VBMI: vector byte manipulation     = false
      AVX512_VBMI2: byte VPCOMPRESS, VPEXPAND  = false
      AVX512_VNNI: neural network instructions = false
      AVX512_BITALG: bit count/shiffle         = false
      AVX512: VPOPCNTDQ instruction            = false
      AVX512_4VNNIW: neural network instrs     = false
      AVX512_4FMAPS: multiply acc single prec  = false
      AVX512_VP2INTERSECT: intersect mask regs = false
      # 后续重复检测结果均为false，此处省略

系统CPU信息 #

Architecture:        x86_64
CPU op-mode(s):      32-bit, 64-bit
Byte Order:          Little Endian
CPU(s):              12
On-line CPU(s) list: 0-11
Thread(s) per core:  1
Core(s) per socket:  2
Socket(s):           6
NUMA node(s):        6
Vendor ID:           GenuineIntel
BIOS Vendor ID:      GenuineIntel
CPU family:          6
Model:               45
Model name:          Intel(R) Xeon(R) Platinum 8253 CPU @ 2.20GHz
BIOS Model name:     Intel(R) Xeon(R) Platinum 8253 CPU @ 2.20GHz
Stepping:            2
CPU MHz:             2194.843
BogoMIPS:            4389.68
Hypervisor vendor:   VMware
Virtualization type: full
L1d cache:           32K
L1i cache:           32K
L2 cache:            1024K
L3 cache:            22528K
NUMA node0 CPU(s):   0,1
NUMA node1 CPU(s):   2,3
NUMA node2 CPU(s):   4,5
NUMA node3 CPU(s):   6,7
NUMA node4 CPU(s):   8,9
NUMA node5 CPU(s):   10,11
Flags:               fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts mmx fxsr sse sse2 ss ht syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts nopl xtopology tsc_reliable nonstop_tsc cpuid pni pclmulqdq ssse3 cx16 pcid sse4_1 sse4_2 x2apic popcnt aes xsave avx hypervisor lahf_lm pti arat

原因分析 #

1. cpu_instruction_set参数的硬性指定逻辑：当手动指定该参数为cascadelake时，DPDK会直接套用Cascade Lake架构预设的全量指令集特性，跳过对当前硬件的实时指令集检测。而Cascade Lake原生支持AVX512，因此编译系统会强制启用所有AVX512相关编译宏。
2. 虚拟机CPU标识与实际指令集的 mismatch：虚拟机的CPU型号显示为Cascade Lake系列的Xeon Platinum 8253，但虚拟化层实际屏蔽了AVX512指令集。此时指定cascadelake参数会让DPDK误判硬件支持AVX512，编译出包含该指令集的代码，而虚拟机无法执行这些指令，最终导致应用崩溃。
3. 无参数时的自动适配逻辑：不指定cpu_instruction_set时，DPDK会通过CPUID指令实时检测当前硬件的实际指令集支持情况，能正确识别到AVX512未被启用，因此编译出的代码仅包含虚拟机真实支持的指令集，运行正常。

内容的提问来源于stack exchange，提问作者nmurshed