Fix unwinder loading on 6.8 kernels #2667
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
kakkoyun
approved these changes
Apr 4, 2024
|
Is the failure of the native unwinder integration tests on arm64 expected? That should have worked, but the rest could fail. |
|
No, it's not expected. I need to look into that, thanks. |
umanwizard
added a commit
to umanwizard/parca-agent
that referenced
this pull request
Apr 25, 2024
See PR parca-dev#2667 for details about why this is needed, as well as the mailing list thread https://lore.kernel.org/bpf/[email protected]/ . The approach tried in that PR caused a regression on older kernels; hopefully, this one will work while also being less confusing.
See comments, and mailing list thread linked therein, for details. Before this change, recent kernels (6.8 and later) cannot verify the native unwinder, because improvements in the precision of register bounds tracking cause an explosion in state space resulting in exceeding the limit of 1 million instructions processed per load. This commit works around that issue, as the comment in the code explains.
This lets us inline the function without letting clang/llvm optimize out the double xor.
umanwizard
commented
Apr 26, 2024
|
OK, this finally. @kakkoyun and @brancz , can you please have another look? We can't use function calls; everything has to be inlined in order to work on old kernels. But if we let clang/llvm inline the function, it will optimize it away... I solved this problem by just using inline volatile asm, which the compiler won't optimize. It's an ugly solution, but it seems to work. |
brancz
pushed a commit
that referenced
this pull request
May 7, 2024
See PR #2667 for details about why this is needed, as well as the mailing list thread https://lore.kernel.org/bpf/[email protected]/ . The approach tried in that PR caused a regression on older kernels; hopefully, this one will work while also being less confusing.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
See comments, and mailing list thread linked therein, for details. Before this change, recent kernels (6.8 and later) cannot verify the native unwinder, because improvements in the precision of register bounds tracking cause an explosion in state space resulting in exceeding the limit of 1 million instructions processed per load.
This commit works around that issue, as the comment in the code explains.