| FazBrowse GitHub Viewer | Trending | | Home |
| Tools: [Download Repo ZIP] [Original HTTPS Page] |
Sorry, something went wrong.
There was a problem hiding this comment.
lgtm. we will take for consideration in 9.0.x
Sorry, something went wrong.
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
Sorry, something went wrong.
|
@kunalspathak please take a look at the PR failures. we can merge when ready |
Sorry, something went wrong.
|
@jeffschwMSFT - this should be ready to merge. Thanks! |
Sorry, something went wrong.
|
go for it, I just hit the rebase button. the *-stagging branches are open to all committers to merge :) |
Sorry, something went wrong.
This reverts commit 269225f.
…rt/pr-112469-to-release/9.0-staging
|
@jeffschwMSFT - any idea how to make check-labels pass? |
Sorry, something went wrong.
|
not sure everything looks in order, merged |
Sorry, something went wrong.
| Back | FazBrowse Home | New Git URL |
Backport of #112469 to release/9.0-staging
/cc @kunalspathak
Customer Impact
We were wrongly reporting certain registers has gc references which would lead the GC would interpret the values as them as valid object addresses and enumerate over them. This would lead to an undefined behavior or crash. Although it is a rare occurrence and certain events need to line up in place (method's register requirements in given method, trigger of GC at the instruction where wrong registers are tracked, etc.), it is still a potential issue because it can lead to fatal crash and diagnosing that crash would take an area expert (hard to root cause).
Regression
In #97910, we started inlining TLS access for NativeAOT for linux/arm64 and the issue was introduced since then.
Testing
The failing test was reverified and the asmdiffs were studied to make sure that we are reporting the gc references correctly.
Risk
Low. This optimization is present in our mainline for almost a year now and was found from our internal tool just recently.