Report RhpInitialDynamicInterfaceDispatch reference in gfids
#103948
Conversation
Some subset of tests is failing with control flow guard enabled because we're not able to indirectly call `RhpInitialDynamicInterfaceDispatch`. It's not clear why since we address take this method in C++ source files as well, but we should be reporting this from our object file too and doing that fixes it.
|
Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas |
|
/azp run runtime-nativeaot-outerloop |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
/azp run runtime-nativeaot-outerloop |
|
Azure Pipelines successfully started running 1 pipeline(s). |
src/coreclr/jit/hwintrinsicxarch.cpp
Outdated
| @@ -267,17 +267,17 @@ static CORINFO_InstructionSet lookupInstructionSet(const char* className) | |||
| { | |||
| if (strncmp(className + 6, "128", 3) == 0) | |||
| { | |||
| assert((className[9] == '\0') || (strcmp(className + 9, "`1") == 0)); | |||
| //assert((className[9] == '\0') || (strcmp(className + 9, "`1") == 0)); | |||
There was a problem hiding this comment.
See commit title f71b073
Outerloop is on the floor again and I'm getting really tired being the janitor.
There was a problem hiding this comment.
GitHub UI only showed me the first two commits in the list and thus I didn't see the title. Must have been some glitch.
There was a problem hiding this comment.
A fix for this up here: https://github.com/dotnet/runtime/pull/103995/files
However, the fact that the assert triggered looks to be a bug in NAOT, as there are types being marked as [Intrinsic] which should not be marked [Intrinsic] and which are not marked as such in the actual managed implementation.
| @@ -44,6 +44,11 @@ public abstract partial class NodeFactory | |||
| ObjectDataInterner dataInterner) | |||
| { | |||
| _target = context.Target; | |||
|
|
|||
| InitialInterfaceDispatchStub = _target.Architecture == TargetArchitecture.ARM | |||
| ? new InitialInterfaceDispatchStubNode() | |||
There was a problem hiding this comment.
I am wondering whether the special ARM dispatch stub wrapper is still necessary.
There was a problem hiding this comment.
I can try to delete that in a different PR and see what the CI says, but I don't have hardware to test it on.
There was a problem hiding this comment.
Testing it in CI should be good enough if you would like to give it a try.
I think this was only needed for multi-.dll Windows arm32 support that we do not have anymore.
|
/ba-g only deleted old workarounds and #103215 needs this and I'm going to bed |
Some subset of tests is failing with control flow guard enabled because we're not able to indirectly call
RhpInitialDynamicInterfaceDispatch. It's not clear why since we address take this method in C++ source files as well, but we should be reporting this from our object file too and doing that fixes it.Fixes #103798
Cc @dotnet/ilc-contrib