GC stress fix for reverse pinvoke stubs #37432
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Reverse PInvoke stubs change the thread to preemptive mode before they exit. Often the epilog immediately follows this helper call, but in some cases there may be a region with a few instructions before we reach the epilog. If there is a normal GC triggered while the method is in this region then the runtime will not report GC references from the stub frame. If the stub method is fully interruptible we will put GC coverage interrupts in this region when doing GC stress. When these trigger a GC then the runtime will report GC references from the stub frame. So if while executing in this region we happen to do a normal GC and then a stress invoked GC, the stress mode GC can find invalid GC references in the frame, as the objects referenced could have been collected/moved by the normal GC that happened just before. There are a couple of avenues for a fix, but since this is a stress mode only problem, I've modified GC stress avoid invoking a GC when the thread is already in preemptive mode and the method is a reverse PINvoke stub. Fixes dotnet#13720.
|
Tagging subscribers to this area: @ViktorHofer |
|
More notes / logs / etc in the linked issue. |
jkotas
reviewed
Jun 4, 2020
src/coreclr/src/vm/gccover.cpp
Outdated
| // So, suppress GC for the reverse PInvoke stubs. | ||
| if (!pThread->PreemptiveGCDisabled()) | ||
| { | ||
| _ASSERTE(pMD->IsILStub()); |
There was a problem hiding this comment.
Suggested change
| _ASSERTE(pMD->IsILStub()); | |
| _ASSERTE(pMD->HasUnmanagedCallersOnlyAttribute()); |
The method does not have to be IL stub.
There was a problem hiding this comment.
Scratch that. It can be any method. PInvoke stub with an inlined frame can appear in any method.
There was a problem hiding this comment.
Good point. That was just some extra paranoia.
|
Added some GC stress runs. |
ghost
locked as resolved and limited conversation to collaborators
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Reverse PInvoke stubs change the thread to preemptive mode before
they exit. Often the epilog immediately follows this helper call,
but in some cases there may be a region with a few instructions
before we reach the epilog.
If there is a normal GC triggered while the method is in this region
then the runtime will not report GC references from the stub frame.
If the stub method is fully interruptible we will put GC coverage
interrupts in this region when doing GC stress. When these trigger a GC
then the runtime will report GC references from the stub frame.
So if while executing in this region we happen to do a normal GC and then
a stress invoked GC, the stress mode GC can find invalid GC references
in the frame, as the objects referenced could have been collected/moved
by the normal GC that happened just before.
There are a couple of avenues for a fix, but since this is a stress mode
only problem, I've modified GC stress avoid invoking a GC when the thread
is already in preemptive mode and the method is a reverse PInvoke stub.
Fixes #13720.