Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Just running tests with new EH enabled on all platforms #90790

Closed

Conversation

janvorli
Copy link
Member

No description provided.

@janvorli janvorli added NO-MERGE The PR is not ready for merge yet (see discussion for detailed reasons) NO-REVIEW Experimental/testing PR, do NOT review it labels Aug 18, 2023
@ghost ghost assigned janvorli Aug 18, 2023
@janvorli janvorli removed the request for review from MichalStrehovsky August 18, 2023 12:10
@janvorli
Copy link
Member Author

/azp run runtime-coreclr outerloop

@azure-pipelines
Copy link

Azure Pipelines successfully started running 1 pipeline(s).

This change ports NativeAOT exception handling to CoreCLR, resulting in
3.5..4 times speedup in exception handling. Due to time constraints and
various complexities, thread abort and debugger support is not
completed yet, so this change is enabled only when
`DOTNET_EnableNewExceptionHandling` env variable is set. By default,
the old way is used.
This change supports all OSes and targets we support except of x86
Windows. That may be doable too in the future, but the difference in
exception handling on x86 Windows adds quite a lot of complexity into
the picture.

Notes for the PR:
* I have left the `ExceptionHandling.cs` and `StackFrameIterator.cs` in
  the nativeaot folder to simplify the review. I can move it to some
  common location after the change is reviewed. Also it was not clear to
  me where that should be, so advise would be welcome here.
* Naming of the native helpers like `RhpCallCatchFunclet` was left the
  same as in the NativeAOT for now.
* There are still some little things I'd like to eventually clean up,
  like `ExInfo` encapsulation and possibly moving `REGDISPLAY` and
  `CONTEXT` it uses into the `ExInfo` itself or moving debug members of
  `StackFrameIterator` and `REGDISPLAY` to the end of those structures
  so that the `AsmOffsets.cs` can be simplified. It also may be possible
  to unify the exception handling callback that's used for ObjectiveC to
  use the managed version. I've tried and there were some ugly
  complications, so I've left it separated.
* There are two bug fixes for bugs unrelated to this PR and a removal of
  unused parameter in existing code that could be made as separate PRs
  before this PR.
  * `ProfilerEnter` and `ProfilerLeave` for the case of
    `UnmanagedCallersOnly` method were being called in preemptive mode.
  * NativeAOT code for rethrowing exception was incorrectly calling
    `DispatchEx` with last argument set to `activeExInfo._idxCurClause`
    to start at the last clause processed when the rethrown exception
    was originally thrown instead of starting from the first one again.
    I have a accidentally came with a simple test that discovered this
    bug and causes failures in  the original NativeAOT too.
* Changes in  the stackwalk.cpp add support for
  * Usage of `ExInfo` instead of `ExceptionTracker`
  * Handling of case when GC runs while finally funclet is on the stack
    and then again when the code is back in the new exception handling
    code in managed code before other finally or catch funclet is
    called. The NativeAOT solves that by disabling GC for the 2nd pass
    of EH, for this change it would not be reasonable.
  * Handling the GC reporting when funclet is found while walking the
    stack. It needs to scan frames of the managed code that handles the
    exception too,  since it contains live references. The old EH way
    doesn't have this case.
* I needed to add `GCFrame::Remove` method that can remove the `GCFrame`
  from any location in the chain. There is a managed runtime method that
  calls `GCReporting::Unregister` that was popping it with my changes
  out of order due to the exception handling code being managed.

Fix context initialization after rebase
The `UNINSTALL_UNWIND_AND_CONTINUE_HANDLER_NO_PROBE` in the
`EE_TO_JIT_TRANSITION` needs to rethrow an exception (if any) using native
exception handling mechanism instead of calling the new managed
exception handling, as the native exception needs to propagate through
some native code layers from there.
This change adds parameter to the
`UNINSTALL_UNWIND_AND_CONTINUE_HANDLER_NO_PROBE` macro to select whether
to rethrow the exception as native or to invoke the new managed
exception handling.

This problem didn't show up until I ran the coreclr tests with tiered
compilation disabled.
There were three places where the UNINSTALL_UNWIND_AND_CONTINUE_HANDLER
needed to be replaced by
UNINSTALL_UNWIND_AND_CONTINUE_HANDLER_NO_PROBE(true).
@janvorli janvorli force-pushed the port-nativeaot-eh-to-coreclr-final-2-test branch from cd175d8 to 0c3bcfd Compare August 23, 2023 21:01
@janvorli janvorli closed this Sep 1, 2023
@ghost ghost locked as resolved and limited conversation to collaborators Oct 1, 2023
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
area-VM-coreclr NO-MERGE The PR is not ready for merge yet (see discussion for detailed reasons) NO-REVIEW Experimental/testing PR, do NOT review it
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant