JIT: handle adjacent no-GC regions (#115719) by AndyAyersMS · Pull Request #130834 · dotnet/runtime

AndyAyersMS · 2026-07-16T02:08:30Z

Ensure the first instruction of each non-interruptible GC range is excluded from the range. This breaks up adjacent no-GC regions, where previously only the first instruction of the first range was excluded.

Fixes #115719.

Ensure the first instruction of each non-interruptible GC range is excluded from the range. This breaks up adjacent no-GC regions, where previously only the first instruction of the first range was excluded. Fixes dotnet#115719. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

AndyAyersMS · 2026-07-16T02:08:49Z

@EgorBo PTAL
fyi @dotnet/jit-contrib

azure-pipelines · 2026-07-16T02:09:12Z

Azure Pipelines: Successfully started running 5 pipeline(s). 10 pipeline(s) were filtered out due to trigger conditions. There may be pipelines that require an authorized user to comment /azp run to run.

dotnet-policy-service · 2026-07-16T02:09:48Z

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Copilot

Pull request overview

Updates CoreCLR JIT GC info encoding so that each IGF_NOGCINTERRUPT (non-interruptible / no-GC) region consistently has its first instruction treated as interruptible (except for prolog regions), even when two no-GC regions are adjacent. This improves the accuracy of interruptible range reporting used by the runtime/debugger.

Changes:

Adjust InterruptibleRangeReporter to consider igOffs + firstInstrSize (non-prolog) when deciding whether to emit an interruptible range.
This ensures an interruptible “gap” is emitted for the first instruction even when igOffs == m_uninterruptibleEnd, breaking up adjacent no-GC regions.

jakobbotsch · 2026-07-16T10:23:44Z

Fixes #115719.

What happened in the issue? We end up defining a 0-length interruptible GC range and that does something odd?

jakobbotsch · 2026-07-16T10:24:05Z

/azp run runtime-coreclr gcstress0x3-gcstress0xc, runtime-coreclr gcstress-extra

azure-pipelines · 2026-07-16T10:24:24Z

Azure Pipelines: Successfully started running 2 pipeline(s).

AndyAyersMS · 2026-07-16T14:59:15Z

Fixes #115719.

What happened in the issue? We end up defining a 0-length interruptible GC range and that does something odd?

We fused together adjacent no-gc regions into one big region, and the user was unable to run funceval when broken in at the start of the second one (0x64), since it was considered no-gc.

;; interruptible ranges

Before: [0x27, 0x57)              [0x76, 0x89) [0x96, 0x9E)
After:  [0x27, 0x57) [0x64, 0x69) [0x76, 0x89) [0x96, 0x9E)

AndyAyersMS · 2026-07-16T16:08:18Z

GC stress failures are related.

AndyAyersMS · 2026-07-16T16:16:09Z

We can't distinguish between a single no-gc region broken up across IGs vs multiple no-gc regions that happen to be adjacent. Looking instead at adding an explicit NOP to clearly delineate the start of a second, separate no-gc region.

tannergooding · 2026-07-16T16:22:50Z

Is this one worth backporting?

AndyAyersMS · 2026-07-16T16:40:42Z

Is this one worth backporting?

It is fixing an impaired debugging experience. I don't know if that meets the servicing bar. The issue has been open a while now.

tannergooding · 2026-07-16T16:42:39Z

Personally I'd say its worth a bar check. Pretty sure I've hit this a few times and it would be nice to not have a degraded debug experience in the LTS

AndyAyersMS · 2026-07-16T19:30:37Z

Revising this to insert a NOP in lower instead, so that we have a clear gc safe point with appropriate debug info.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

AndyAyersMS · 2026-07-16T20:09:27Z

/azp run runtime-coreclr gcstress0x3-gcstress0xc, runtime-coreclr gcstress-extra

azure-pipelines · 2026-07-16T20:09:39Z

Azure Pipelines: Successfully started running 2 pipeline(s).

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated no new comments.

AndyAyersMS · 2026-07-17T19:53:33Z

/azp run runtime-coreclr gcstress0x3-gcstress0xc, runtime-coreclr gcstress-extra

azure-pipelines · 2026-07-17T19:53:46Z

Azure Pipelines: Successfully started running 2 pipeline(s).

AndyAyersMS · 2026-07-18T15:46:09Z

GC stress failures are the tests fixed by #130974 and the known issue #130993

github-actions · 2026-07-19T13:02:41Z

Workflow state for the Holistic Review Orchestrator.

{
  "version": 5,
  "last_dispatched_commit": "ec4034b1140ce1658a1cc7e6cf7941cfd93af5ab",
  "last_dispatched_base_ref": "main",
  "last_dispatched_base_sha": "49110af56a95223e88f9e540503c0f8a16e64e15",
  "last_reviewed_commit": "ec4034b1140ce1658a1cc7e6cf7941cfd93af5ab",
  "last_reviewed_base_ref": "main",
  "last_reviewed_base_sha": "49110af56a95223e88f9e540503c0f8a16e64e15",
  "last_recorded_worker_run_id": "29688127987",
  "review_attempt_commit": "",
  "review_attempt_base_ref": "",
  "review_attempt_count": 0,
  "max_review_attempts": 5,
  "review_history_format": "holistic-review-disclosure-v1",
  "review_history": [
    {
      "commit": "ec4034b1140ce1658a1cc7e6cf7941cfd93af5ab",
      "review_id": 4730810109
    }
  ]
}

github-actions

Holistic Review

Motivation: Justified. Issue #115719 shows a real debugging failure where evaluation is disallowed because the debugger stops at an unsafe GC point in unoptimized code. The root cause is that adjacent no-GC regions get merged so only the very first instruction of the first region is excluded from no-GC, leaving stack-empty points between logical regions non-interruptible.

Approach: Sound. For debuggable codegen only, the change detects when a new no-GC region is being opened immediately after existing no-GC code at a stack-empty Normal IP mapping matching the current location, and inserts an interruptible INS_nop to break up the adjacent regions. It also introduces emitLastSavedIGWasNoGC tracking and renames the previously-unused emitGCDisabled() to emitLastCodeIsNoGC(), which correctly reports interruptibility of the current IG (if non-empty) or the last saved non-empty IG.

Summary: ✅ LGTM. The change is narrowly gated to compDbgCode, so codegen for optimized code is unaffected. Bookkeeping for emitLastSavedIGWasNoGC is updated consistently in emitSavIG (only for non-empty IGs), emitCreatePlaceholderIG (prolog/epilog placeholders, always no-GC), and initialized in emitBegFN. The renamed helper had no external callers, so the rename is safe. Already approved by two JIT area experts. No blocking issues found.

Detailed Findings

✅ Correctness — No-GC region bookkeeping is consistent

emitLastSavedIGWasNoGC is set from IGF_NOGCINTERRUPT only when the saved IG is non-empty (ig->igSize != 0), matching the emitCurIGsize != 0 guard in emitLastCodeIsNoGC. Placeholder IGs (prolog/epilog) are unconditionally marked no-GC, which is correct since those regions are non-interruptible. Initialization in emitBegFN prevents stale cross-method state.

✅ Scope — Debug-only, low risk

The NOP insertion is guarded by m_compiler->opts.compDbgCode, emitNoGCRequestCount == 0, a non-empty genIPmappings, and precise checks that the last mapping is a Normal, STACK_EMPTY, current-location mapping. This tightly limits the behavioral change to the intended debuggable scenario and avoids emitting spurious NOPs.

Note

This review was generated by this repository's Holistic Review agentic workflow to complement the built-in Copilot review.

Generated by Holistic Review · 62.3 AIC · ⌖ 14.7 AIC · ⊞ 10K

AndyAyersMS · 2026-07-21T18:13:56Z

@EgorBo @jakobbotsch reworked this, take another look.

GC stress in main has some known failures.

JulieLeeMSFT

LGTM.
In emitDisableGC(), conditions are tight and look correct.

jakobbotsch · 2026-07-22T21:28:01Z

I'm worried about this case here:

runtime/src/coreclr/jit/codegenxarch.cpp

Lines 166 to 171 in 78fd36e

    
           // Because of the way the flowgraph is connected, the liveness info for this one instruction 
        
           // after the call is not (can not be) correct in cases where a variable has a last use in the 
        
           // handler.  So turn off GC reporting once we execute the call and reenable after the jmp/nop 
        
           GetEmitter()->emitDisableGC(); 
        
           GetEmitter()->emitIns_J(INS_call, block->GetTarget());

With this PR do we potentially end up allowing a GC in this position that has the wrong GC info described in the comment?

AndyAyersMS · 2026-07-22T22:38:22Z

I'm worried about this case here:

runtime/src/coreclr/jit/codegenxarch.cpp

Lines 166 to 171 in 78fd36e

// Because of the way the flowgraph is connected, the liveness info for this one instruction

// after the call is not (can not be) correct in cases where a variable has a last use in the

// handler. So turn off GC reporting once we execute the call and reenable after the jmp/nop

GetEmitter()->emitDisableGC();

GetEmitter()->emitIns_J(INS_call, block->GetTarget());

With this PR do we potentially end up allowing a GC in this position that has the wrong GC info described in the comment?

I don't think so? The concern there is that in a call finally; jmp continuation GC info can be wrong between the two instructions, so the jmp is a no-GC region.

That doesn't change with this PR.

jakobbotsch · 2026-07-23T05:13:24Z

I'm worried about this case here:

runtime/src/coreclr/jit/codegenxarch.cpp

Lines 166 to 171 in 78fd36e

// Because of the way the flowgraph is connected, the liveness info for this one instruction

// after the call is not (can not be) correct in cases where a variable has a last use in the

// handler. So turn off GC reporting once we execute the call and reenable after the jmp/nop

GetEmitter()->emitDisableGC();

GetEmitter()->emitIns_J(INS_call, block->GetTarget());

With this PR do we potentially end up allowing a GC in this position that has the wrong GC info described in the comment?

I don't think so? The concern there is that in a call finally; jmp continuation GC info can be wrong between the two instructions, so the jmp is a no-GC region.

That doesn't change with this PR.

Ah yes, I see that now

AndyAyersMS · 2026-07-24T15:07:48Z

/azp run runtime-coreclr gcstress0x3-gcstress0xc, runtime-coreclr gcstress-extra

azure-pipelines · 2026-07-24T15:08:03Z

Azure Pipelines: Successfully started running 2 pipeline(s).

Copilot

Pull request overview

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

AndyAyersMS · 2026-07-24T21:38:45Z

Merged up and GC stress is all green.

Copilot AI review requested due to automatic review settings July 16, 2026 02:08

github-actions Bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Jul 16, 2026

dotnet-policy-service Bot assigned AndyAyersMS Jul 16, 2026

AndyAyersMS requested a review from EgorBo July 16, 2026 02:08

Copilot started reviewing on behalf of AndyAyersMS July 16, 2026 02:09 View session

Copilot AI reviewed Jul 16, 2026

View reviewed changes

Comment thread src/coreclr/jit/gcencode.cpp Outdated

EgorBo approved these changes Jul 16, 2026

View reviewed changes

build-analysis Bot mentioned this pull request Jul 16, 2026

The Operation will be canceled. The next steps may not contain expected logs. dotnet/dnceng#3008

Open

3 tasks

tannergooding approved these changes Jul 16, 2026

View reviewed changes

build-analysis Bot mentioned this pull request Jul 16, 2026

Assertion failed: CREATE_CHECK_STRING(bSmallObjectHeapPtr || bLargeObjectHeapPtr) #119140

Open

JIT: insert NOPs instead

e1e41f5

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings July 16, 2026 19:35

Copilot started reviewing on behalf of AndyAyersMS July 16, 2026 19:36 View session

Copilot AI reviewed Jul 16, 2026

jakobbotsch reviewed Jul 17, 2026

View reviewed changes

Comment thread src/coreclr/jit/lower.cpp Outdated

revise, handle in emitDisableGC

ec4034b

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings July 17, 2026 19:28

Copilot started reviewing on behalf of AndyAyersMS July 17, 2026 19:29 View session

Copilot AI reviewed Jul 17, 2026

View reviewed changes

build-analysis Bot mentioned this pull request Jul 18, 2026

[ci-scan] Test failure: System.Security.Cryptography.X509Certificates.Tests.RevocationTests.AiaTests.AiaCompletionHasLimits #130947

Closed

github-actions Bot reviewed Jul 19, 2026

View reviewed changes

AndyAyersMS requested review from EgorBo and jakobbotsch July 21, 2026 18:13

AndyAyersMS added the Priority:1 Work that is critical for the release, but we could probably ship without label Jul 21, 2026

JulieLeeMSFT approved these changes Jul 22, 2026

View reviewed changes

jakobbotsch approved these changes Jul 23, 2026

View reviewed changes

Merge branch 'main' into issue-115719

d9ecffb

Copilot AI review requested due to automatic review settings July 24, 2026 15:06

Copilot started reviewing on behalf of AndyAyersMS July 24, 2026 15:07 View session

Copilot AI reviewed Jul 24, 2026

View reviewed changes

Comment thread src/coreclr/jit/emit.h

build-analysis Bot mentioned this pull request Jul 24, 2026

Unable to pull image from mcr.microsoft.com #117164

Open

AndyAyersMS merged commit 199b1a0 into dotnet:main Jul 24, 2026
166 of 168 checks passed

dotnet-maestro Bot mentioned this pull request Jul 25, 2026

[main] Source code updates from dotnet/runtime dotnet/dotnet#7965

Open

Uh oh!

Conversation

AndyAyersMS commented Jul 16, 2026

Uh oh!

AndyAyersMS commented Jul 16, 2026

Uh oh!

azure-pipelines Bot commented Jul 16, 2026

Uh oh!

dotnet-policy-service Bot commented Jul 16, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

jakobbotsch commented Jul 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jakobbotsch commented Jul 16, 2026

Uh oh!

azure-pipelines Bot commented Jul 16, 2026

Uh oh!

AndyAyersMS commented Jul 16, 2026

Uh oh!

AndyAyersMS commented Jul 16, 2026

Uh oh!

AndyAyersMS commented Jul 16, 2026

Uh oh!

tannergooding commented Jul 16, 2026

Uh oh!

AndyAyersMS commented Jul 16, 2026

Uh oh!

tannergooding commented Jul 16, 2026

Uh oh!

AndyAyersMS commented Jul 16, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

AndyAyersMS commented Jul 16, 2026

Uh oh!

azure-pipelines Bot commented Jul 16, 2026

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

AndyAyersMS commented Jul 17, 2026

Uh oh!

azure-pipelines Bot commented Jul 17, 2026

Uh oh!

AndyAyersMS commented Jul 18, 2026

Uh oh!

github-actions Bot commented Jul 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Holistic Review

Detailed Findings

✅ Correctness — No-GC region bookkeeping is consistent

✅ Scope — Debug-only, low risk

Uh oh!

AndyAyersMS commented Jul 21, 2026

Uh oh!

JulieLeeMSFT left a comment

Choose a reason for hiding this comment

Uh oh!

jakobbotsch commented Jul 22, 2026

Uh oh!

AndyAyersMS commented Jul 22, 2026

Uh oh!

jakobbotsch commented Jul 23, 2026

Uh oh!

AndyAyersMS commented Jul 24, 2026

Uh oh!

azure-pipelines Bot commented Jul 24, 2026

Uh oh!

Copilot AI left a comment

jakobbotsch commented Jul 16, 2026 •

edited

Loading

github-actions Bot commented Jul 19, 2026 •

edited

Loading