IXCLRDataMethodInstance.GetILAddressMap is broken in .NET 6 RC1

[gregg-miskelly]

Description

In .NET 6 rc1, mscordaccore.dll!ClrDataMethodInstance::GetILAddressMap will always fail with E_INVALIDARG. The problem is that GetMethodNativeMap always provides 0 as the second argument (address) to ClrDataAccess::GetMethodNativeMap. This is passed to ExecutionManager::GetNativeCodeVersion, which will always fail.

Call stack:

 	mscordaccore.dll!ExecutionManager::GetNativeCodeVersion(unsigned __int64 currentPC) Line 4287	C++
 	mscordaccore.dll!ClrDataAccess::GetMethodNativeMap(MethodDesc * methodDesc, unsigned __int64 address, unsigned int * numMap, DebuggerILToNativeMap * * map, bool * mapAllocated, unsigned __int64 * codeStart, unsigned int * codeOffset) Line 6133	C++
>	mscordaccore.dll!ClrDataMethodInstance::GetILAddressMap(unsigned int mapLen, unsigned int * mapNeeded, __MIDL___MIDL_itf_xclrdata_0000_0012_0003 * maps) Line 4250	C++

Configuration

• Which version of .NET is the code running on? .NET 6 rc1
• What OS and version, and what distro if applicable? Windows
• What is the architecture (x64, x86, ARM, ARM64)? x64
• Do you know whether it is specific to that configuration? I don't look like it

Regression?

This worked in preview 7 of .NET 6 as well as all previous .NET releases

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[gregg-miskelly]

CC @tommcdon

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

Issue Details

---

Description

In .NET 6 rc1, mscordaccore.dll!ClrDataMethodInstance::GetILAddressMap will always fail with E_INVALIDARG. The problem is that GetMethodNativeMap always provides 0 as the second argument (address) to ClrDataAccess::GetMethodNativeMap. This is passed to ExecutionManager::GetNativeCodeVersion, which will always fail.

Call stack:

 	mscordaccore.dll!ExecutionManager::GetNativeCodeVersion(unsigned __int64 currentPC) Line 4287	C++
 	mscordaccore.dll!ClrDataAccess::GetMethodNativeMap(MethodDesc * methodDesc, unsigned __int64 address, unsigned int * numMap, DebuggerILToNativeMap * * map, bool * mapAllocated, unsigned __int64 * codeStart, unsigned int * codeOffset) Line 6133	C++
>	mscordaccore.dll!ClrDataMethodInstance::GetILAddressMap(unsigned int mapLen, unsigned int * mapNeeded, __MIDL___MIDL_itf_xclrdata_0000_0012_0003 * maps) Line 4250	C++

Configuration

• Which version of .NET is the code running on? .NET 6 rc1
• What OS and version, and what distro if applicable? Windows
• What is the architecture (x64, x86, ARM, ARM64)? x64
• Do you know whether it is specific to that configuration? I don't look like it

Regression?

This worked in preview 7 of .NET 6 as well as all previous .NET releases

Author:	gregg-miskelly
Assignees:	-
Labels:	area-Diagnostics-coreclr, untriaged
Milestone:	6.0.0

[tommcdon]

@hoyosjs @mikem8361

[hoyosjs]

@gregg-miskelly I see the issue - the main issue is ClrDataMethodInstance is not tiering aware. It only stores the desc. It's easy to give you the latest map - would that work for you?

[gregg-miskelly]

Ideally I think we would want it to work with all versions, though the latest is definitely better than nothing. Is there a different API we could use that would be better? I believe we use this code in two cases:

1. For disassembly -- we have a native instruction address, and we are trying to decode what is there
2. For execution control in time travel debugging -- we have an IL address (module, token, offset) and want to translate that to all the native addresses that we should watch

CC @delmyers

[delmyers]

2. For execution control in time travel debugging -- we have an IL address (module, token, offset) and want to translate that to all the native addresses that we should watch

Yes, for time travel debugging, it is absolutely necessary that we can get all of the native addresses for all versions, because we need to be able to set breakpoints in the native addresses that represent the jitted IL so that we can stop at the correct location if there are multiple versions of the method.

[hoyosjs]

I'll took a look in the DAC, but that method is the only thing I know that has access to such maps and is exposed publicly. Is Cordb something that is possible too to try to see what it would take to get you the right info?

Until recently we didn't realize this method wasn't tiering aware (The fix attempt was what broke this). I just opened a PR that will revert to the old behavior for that case. Might be worth adding a check to see if for 1 you are still disassembling the method in question by calling GetRepresentativeEntryAddress and seeing if you are within the expected range.

For #2, I am not aware of anything that will give you all versions of a given IL token + mod + offset (maybe @davmason knows as it sounds like a profiler type of question. ICordb usually sets the patch in IL and rebinds it as necessary ).