[release/8.0-rc1] Make DacValidateMD and DacValidateMethodTable more resilient #90797
Conversation
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Make both methods more resilient to the case of invalid MethodDesc and MethodTable with value -1. Close #90691
ghost
added
the
|
Tagging subscribers to this area: @tommcdon Issue DetailsBackport of #90794 to release/8.0-rc1 /cc @janvorli Customer ImpactTestingRiskIMPORTANT: If this backport is for a servicing release, please verify that:
|
|
@janvorli please fill out the template and send an email to Tactics requesting approval. Today's the last day for rc1 backports. |
carlossanlop
added
the
Servicing-consider
|
@janvorli how common is this? Would RC2 be sufficient? As Carlos points out we are right at the end of taking RC1 fixes. |
jeffschwMSFT
added
Servicing-approved
|
CI failure unrelated, a remote process in a test failed. Merging now. |
Backport of #90794 to release/8.0-rc1
/cc @janvorli
Customer Impact
BenchmarkDotnetbenchmarks withDisassemblyDiagnoserenabled can randomly crash in case the generated assembly code loads constant 0xffffffffffffffff into a register. It doesn't have to be an explicit constant in the code, it might be coming from an indirection slot. The disassembler tries to handle some of these constants as MethodDesc / MethodTable pointers in order to display class / method names instead of hex constants. The fix checks for -1 (0xffffffffffffffff) and NULL only as these are the only values that are not translated between the debuggee and debugger address spaces.Testing
BenchmarkDotnet test that was failing due to the issue.
Risk
Low, the added early out in the methods dismisses values that would otherwise cause crash.
IMPORTANT: If this backport is for a servicing release, please verify that:
The PR target branch is
release/X.0-staging, notrelease/X.0.If the change touches code that ships in a NuGet package, you have added the necessary package authoring and gotten it explicitly reviewed.