Hitting assert in NibbleMapDeleteUnlocked

[kunalspathak]

Introduced in https://github.com/dotnet/runtime/pull/108939

Assert failure(PID 21700 [0x000054c4], Thread: 47520 [0xb9a0]): ((*pMap) & ~mask) && !IsPointer(*pMap)

CORECLR! EEJitManager::NibbleMapDeleteUnlocked + 0x158 (0x00007ff9`161db138)
CORECLR! EEJitManager::RemoveJitData + 0x1AA (0x00007ff9`161dc9ea)
CORECLR! CEEJitInfo::BackoutJitData + 0xD9 (0x00007ff9`15a60109)
CORECLR! invokeCompileMethod + 0x195 (0x00007ff9`15a99d95)
CORECLR! UnsafeJitFunction + 0x83C (0x00007ff9`15a6e08c)
CORECLR! MethodDesc::JitCompileCodeLocked + 0x1D2 (0x00007ff9`15b45882)
CORECLR! MethodDesc::JitCompileCodeLockedEventWrapper + 0x50D (0x00007ff9`15b4610d)
CORECLR! MethodDesc::JitCompileCode + 0x5AB (0x00007ff9`15b455ab)
CORECLR! MethodDesc::PrepareILBasedCode + 0x50C (0x00007ff9`15b47f4c)
CORECLR! MethodDesc::PrepareCode + 0xF3 (0x00007ff9`15b479e3)
    File: Q:\git\runtime2\src\coreclr\vm\codeman.cpp:4148
    Image: q:\git\runtime2\artifacts\tests\coreclr\windows.x64.Checked\tests\Core_Root\corerun.exe

It always repros with AltJit.

@steveisok @max-charlamb

[steveisok]

@kunalspathak is AltJit doing something it shouldn't as a result of hitting this assert?

[kunalspathak]

@kunalspathak is AltJit doing something it shouldn't as a result of hitting this assert?

It is not doing anything differently than what it should. The easiest way to repro would be to run any hello world app with following environment variables set:

DOTNET_AltJit=*
DOTNET_AltJitName=clrjit_universal_arm64_x64.dll

[dotnet-policy-service[bot]]

Tagging subscribers to this area: @mangod9 See info in area-owners.md if you want to be subscribed.

[max-charlamb]

108939 changed the assert condition to disallow "no-op" deletes. Looking at this example and other uses of CEEJitInfo::BackoutJitData, it seems expected to sometimes delete empty nibbles.

I have created #109976 to relax the assertion and added a comment explaining that no-op deletes can occur.