microsoft / vscode

Visual Studio Code
https://code.visualstudio.com
MIT License
163.22k stars 28.86k forks source link

Node unconditionally uses AVX512 instructions (illegal instruction starting Extension Host agent) #214630

Open mgaudet opened 4 months ago

mgaudet commented 4 months ago

Does this issue occur when all extensions are disabled?: Yes/No

Release:

Version: 1.90.0 (Universal) Commit: 89de5a8d4d6205e5b11647eb6a74844ca23d2573

Inisders:

Version: 1.91.0-insider (Universal) Commit: 6facfe23b300eb448c35d07291cad7ad98625302

Steps to Reproduce:

  1. Use Remote Extension to open window on Ubuntu 24, on a AMD machine with AVX-512 disabled as a boot parameter

Result: Extension host crashes immediately:

[10:51:37.003] [server] [10:51:36] Extension host agent started.
[10:51:37.032] [server] [10:51:37] [<unknown>][d99d6c11][ManagementConnection] New connection established.
[10:51:37.039] [server] [10:51:37] [<unknown>][0be45ac4][ExtensionHostConnection] New connection established.
[10:51:37.081] [server] [10:51:37] [<unknown>][0be45ac4][ExtensionHostConnection] <3493> Launched Extension Host Process.
[10:51:37.874] [server] Illegal instruction (core dumped)

The node backtrace here is:

(gdb) bt 
#0  0x0000000001f9f093 in base64_stream_encode_avx512 ()
#1  0x0000000001f92614 in base64_encode ()
#2  0x0000000000dc2965 in node::StringBytes::Encode(v8::Isolate*, char const*, unsigned long, node::encoding, v8::Local<v8::Value>*) [clone .part.0] ()
#3  0x0000000000c778bf in void node::Buffer::(anonymous namespace)::StringSlice<(node::encoding)2>(v8::FunctionCallbackInfo<v8::Value> const&) ()
#4  0x0000000000f328df in v8::internal::FunctionCallbackArguments::Call(v8::internal::CallHandlerInfo) ()
#5  0x0000000000f3314d in v8::internal::MaybeHandle<v8::internal::Object> v8::internal::(anonymous namespace)::HandleApiCallHelper<false>(v8::internal::Isolate*, v8::internal::Handle<v8::internal::HeapObject>, v8::internal::Handle<v8::internal::FunctionTemplateInfo>, v8::internal::Handle<v8::internal::Object>, unsigned long*, int) ()
#6  0x0000000000f33615 in v8::internal::Builtin_HandleApiCall(int, unsigned long*, v8::internal::Isolate*) ()
#7  0x000000000193cdf6 in Builtins_CEntry_Return1_ArgvOnStack_BuiltinExit ()
#8  0x00000000018aed1c in Builtins_InterpreterEntryTrampoline ()
#9  0x00000e00281804e9 in ?? ()
#10 0x00003d3db84f8191 in ?? ()
#11 0x0000000700000000 in ?? ()
#12 0x00000e00281805b9 in ?? ()
#13 0x00003e25859eefa9 in ?? ()
#14 0x0000000000000000 in ?? ()

This machine is AMD Ryzen Threadripper PRO 7975WX, but for compatibility with https://pernos.co/, I've explicitly disabled AVX-512 by adding clearcpuid=304 to GRUB_CMDLINE_LINUX_DEFAULT in /etc/default/grub

As you can see, AVX512 isn't listed in the /proc/cpuinfo:

flags       : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good amd_lbr_v2 nopl nonstop_tsc cpuid extd_apicid aperfmperf rapl pni pclmulqdq monitor ssse3 fma cx16 pcid sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs skinit wdt tce topoext perfctr_core perfctr_nb bpext perfctr_llc mwaitx cpb cat_l3 cdp_l3 hw_pstate ssbd mba perfmon_v2 ibrs ibpb stibp ibrs_enhanced vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local clzero irperf xsaveerptr rdpru wbnoinvd amd_ppin cppc arat npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold avic v_vmsave_vmload vgif x2avic v_spec_ctrl vnmi umip pku ospke rdpid overflow_recov succor smca fsrm flush_l1d debug_swap
bugs        : sysret_ss_attrs spectre_v1 spectre_v2 spec_store_bypass srso

This bug is similar to https://github.com/microsoft/vscode/issues/180843

mgaudet commented 4 months ago

Note: This is a regression as previous versions of vscode didn't do this, as I was using this configuration yesterday.

mgaudet commented 3 months ago

I suspect the regressing commit will be https://github.com/nodejs/node/pull/52428

mgaudet commented 3 months ago

Bisected down to https://github.com/nodejs/node/commit/f45bb801b69c7105756b971ab076c415d85a9e10 will open a Node bug.

deepak1556 commented 3 months ago

Thanks for tracking down the root cause. As a targeted fix I can backport https://github.com/aklomp/base64/pull/143 when it is merged.