Skip to content

Walk YJIT JIT frames via frame pointers for full stack unwinding#26

Draft
dalehamel wants to merge 5 commits intoruby-jit-upstreamfrom
jit-fp-unwinding
Draft

Walk YJIT JIT frames via frame pointers for full stack unwinding#26
dalehamel wants to merge 5 commits intoruby-jit-upstreamfrom
jit-fp-unwinding

Conversation

@dalehamel
Copy link
Copy Markdown
Member

@dalehamel dalehamel commented Feb 10, 2026

Screenshot 2026-02-10 at 3 41 54 PM

Replace the jit_detected flag approach with V8-style frame pointer unwinding through Ruby JIT frames. When YJIT emits frame pointers (always on arm64, with --yjit-perf on x86_64), the Ruby eBPF unwinder walks the native FP chain through JIT frames, pushes each as a RUBY_FRAME_TYPE_JIT frame, then resolves the post-JIT mapping so native unwinding can continue below the Ruby VM stack.

Three bugs fixed along the way:

  • parseMappings discarded prctl-labeled [anon:...] mappings, so the YJIT JIT region was never visible to interpreter handlers.
  • IsAnonymous() returned false for labeled anonymous mappings, so SynchronizeMappings would have skipped them even if parsing preserved them.
  • After walking JIT frames via FP, get_next_unwinder_after_native_frame must be called to re-resolve the mapping for the new PC, otherwise the native unwinder fails with stale text_section_id/offset.

@dalehamel dalehamel force-pushed the jit-fp-unwinding branch 3 times, most recently from 015b62c to 3ac1562 Compare February 10, 2026 22:58
@dalehamel dalehamel force-pushed the jit-fp-unwinding branch 2 times, most recently from 2b462cc to aabba13 Compare February 11, 2026 17:12
dalehamel added 5 commits March 3, 2026 12:29
@dalehamel
Handle JIT frames
c1086a5
@dalehamel
Rebuild BPF blobs for JIT frame handling
e4d6d9a
@dalehamel
Walk YJIT JIT frames via frame pointers for full stack unwinding
3ebe4cb 
Replace the jit_detected flag approach with V8-style frame pointer
unwinding through Ruby JIT frames. When YJIT emits frame pointers
(always on arm64, with --yjit-perf on x86_64), the Ruby eBPF unwinder
walks the native FP chain through JIT frames, pushes each as a
RUBY_FRAME_TYPE_JIT frame, then resolves the post-JIT mapping so native
unwinding can continue below the Ruby VM stack.

When frame pointers are not available, the original behavior is
preserved: a single dummy JIT frame is pushed, cfuncs are pushed inline,
and native unwinding is stopped at the end of the Ruby stack.

Also fixes parseMappings discarding prctl-labeled [anon:...] mappings,
which prevented the YJIT JIT region from being visible to interpreter
handlers.
@dalehamel
Refactor JIT FP walk to per-iteration unwind and fix JIT region detec…
cdada86 
…tion

Restructure the eBPF JIT frame pointer walk: instead of a separate loop
before the main CFP walk, advance the native FP chain by one frame per
main loop iteration. This keeps the native unwind state in lockstep with
the Ruby VM stack, supporting both YJIT (1 JIT frame, exits after first
iteration) and ZJIT (1 JIT frame per iseq, 1:1 with CFPs).

Fix JIT region detection in SynchronizeMappings to scan all mappings
(including non-executable ---p reservations) for the prctl-labeled JIT
region, then only register LPM prefixes for executable pages. This
ensures jit_start/jit_end cover the full reserved address range even
when the r-xp committed pages don't carry the label.

Also fix IsAnonymous() to recognize [anon:...] labeled mappings, and
remove debug log spam from parseMappings.
@dalehamel
Rebuild BPF blobs for JIT FP unwinding
5702d63
@dalehamel dalehamel changed the base branch from jit-rebase to ruby-jit-upstream March 3, 2026 17:36
@dalehamel dalehamel mentioned this pull request Mar 17, 2026
Open
korniltsev-grafanista
korniltsev-grafanista reviewed Mar 18, 2026
View reviewed changes
Comment thread interpreter/ruby/ruby.go
for _, prefix := range prefixes {
_, exists := r.prefixes[prefix]
if !exists {
err := ebpf.UpdatePidInterpreterMapping(pid, prefix, support.ProgUnwindRuby, 0, 0)
Copy link
Copy Markdown

@korniltsev-grafanista korniltsev-grafanista Mar 18, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same as with otel repo, we probably need to call DeletePidInterpreterMapping at some point

This was referenced Apr 7, 2026
Draft
Draft
dalehamel added a commit that referenced this pull request Apr 7, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
c378267 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
dalehamel added a commit that referenced this pull request Apr 7, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
97623a7 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
dalehamel added a commit that referenced this pull request Apr 7, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
e58fa0a 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
dalehamel added a commit that referenced this pull request Apr 8, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
444d984 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
dalehamel added a commit that referenced this pull request Apr 8, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
07dc74d 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
dalehamel added a commit that referenced this pull request Apr 8, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
154051a 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
dalehamel added a commit that referenced this pull request Apr 8, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
8230425 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
@dalehamel dalehamel force-pushed the ruby-jit-upstream branch 2 times, most recently from 5831ed4 to e151da4 Compare April 8, 2026 14:42
dalehamel added a commit that referenced this pull request Apr 8, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
d26eabb 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
dalehamel added a commit that referenced this pull request Apr 8, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
6bc5c9a 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
dalehamel added a commit that referenced this pull request Apr 8, 2026
@dalehamel
Address review feedback: use findJITRegion, restore Detach cleanup
cad68bf 
- Restore findJITRegion() from base (open-telemetry#1102) which was incorrectly removed
  during rebase - PR #26 now reuses the shared function instead of inline code
- Remove redundant inline first-pass JIT detection (now handled by findJITRegion)
- Detach cleanup is inherited from the updated base (open-telemetry#1102)
@dalehamel
Copy link
Copy Markdown
Member Author

dalehamel commented Apr 8, 2026

FYI #36 will replace this branch

@dalehamel dalehamel force-pushed the ruby-jit-upstream branch from 0c685ac to edbf402 Compare April 9, 2026 13:40
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

1 more reviewer

@korniltsev-grafanista korniltsev-grafanista korniltsev-grafanista left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@dalehamel @korniltsev-grafanista