[LoongArch64] Place async resumption info in read-write section which reference #123433

[shushanhf]

[LoongArch64] Place async resumption info in read-write section which reference #123433

[dotnet-policy-service]

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

[shushanhf]

Hi, @jakobbotsch
after add this PR, the async test case failed liking this

Assert failure(PID 3342786 [0x003301c2], Thread: 3342786 [0x3301c2]): Assertion failed '!"Duplicate slots definitions found in GC information!"' in 'AwaitNotAsync:AsyncEntryPoint()' during 'Emit GC+EH tables    ' (IL size 123; hash 0xfcda73b9; MinOpts)

     File: /data/qiao/runtime/src/coreclr/gcinfo/gcinfoencoder.cpp:820

Can you give me some help? thanks

[jakobbotsch]

Hi, @jakobbotsch after add this PR, the async test case failed liking this

Assert failure(PID 3342786 [0x003301c2], Thread: 3342786 [0x3301c2]): Assertion failed '!"Duplicate slots definitions found in GC information!"' in 'AwaitNotAsync:AsyncEntryPoint()' during 'Emit GC+EH tables    ' (IL size 123; hash 0xfcda73b9; MinOpts)

     File: /data/qiao/runtime/src/coreclr/gcinfo/gcinfoencoder.cpp:820

Can you give me some help? thanks

Hmm, I have never seen this assert before.

Did it only start failing with this PR? I notice that genAsyncResumeInfo emits INS_bl, but that seems wrong, isn't that a branch? It is supposed to compute the address of the data.

Note that the read-write section is specifically for NativeAOT (when emitting ELF format). For corerun/coreclr it should work without the relocations.

[shushanhf]

Hmm, I have never seen this assert before.

Did it only start failing with this PR? I notice that genAsyncResumeInfo emits INS_bl, but that seems wrong, isn't that a branch? It is supposed to compute the address of the data.

This is not the finally real instruction name.

For corerun/coreclr it should work without the relocations.

Befor this PR the async test cases are also failed.

[jakobbotsch]

Hmm, I have never seen this assert before.
Did it only start failing with this PR? I notice that genAsyncResumeInfo emits INS_bl, but that seems wrong, isn't that a branch? It is supposed to compute the address of the data.

This is not the finally real instruction name.

For corerun/coreclr it should work without the relocations.

Befor this PR the async test cases are also failed.

You may want to double check the LA64 handling for the runtime async specific nodes (GT_RETURN_SUSPEND, GT_ASYNC_RESUME_INFO, GT_RECORD_ASYNC_RESUME, GT_ASYNC_CONTINUATION) in LSRA and codegen.