[release/10.0] Fix LoaderHeap's free list growing more than expected

[eduardo-vp]

Backport of #129203 to release/10.0

Customer Impact

• Customer reported
• Found internally

This is a regression that impacts apps running test suites, running sometimes even 6x to 8x slower on Linux. Reported in #123124, in particular in this comment #123124 (comment).

Regression

• Yes
• No

Testing

Tested performance with the benchmark provided by the customer and another one in #129203 (comment). The issue happens under a high volume of method instantiations, we don't have such test. We could consider adding one.

Risk

Medium/Low. Actually the change is not that big - SatisfiesMethodConstraints is refactored and calls SatisfiesMethodConstraintsForInstantiation now. Also, inside InstantiatedMethodDesc::NewInstantiatedMethodDesc we previously took a lock, released it and took it again. Now we just don't release it in the middle and the call to SatisfiesMethodConstraints was moved to the beginning of the method. Unfortunately the github diff is way bigger than it should.

I also let the change sit in main for one week and no issues were linked to the original PR.

IMPORTANT: If this backport is for a servicing release, please verify that:

• For .NET 8 and .NET 9: The PR target branch is release/X.0-staging, not release/X.0.
• For .NET 10+: The PR target branch is release/X.0 (no -staging suffix).

Package authoring no longer needed in .NET 9

IMPORTANT: Starting with .NET 9, you no longer need to edit a NuGet package's csproj to enable building and bump the version.
Keep in mind that we still need package authoring in .NET 8 and older versions.

[dotnet-policy-service]

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

[Copilot]

Pull request overview

This PR backports the .NET 10 fix for a severe performance regression caused by UnlockedLoaderHeap free-list growth during high-concurrency generic method instantiation. It prevents multiple threads from racing to allocate and then discard redundant InstantiatedMethodDesc allocations, which previously inflated the free list and made subsequent allocations increasingly expensive.

Changes:

• Added a shared helper (SatisfiesMethodConstraintsForInstantiation) to evaluate generic method constraints given an explicit instantiation.
• Changed InstantiatedMethodDesc::NewInstantiatedMethodDesc to hold m_InstMethodHashTableCrst across lookup and creation/insertion, ensuring only one thread allocates/registers a MethodDesc per instantiation.
• Refactored MethodDesc::SatisfiesMethodConstraints to delegate to the new helper.

[svick]

Hi,

the code complete date for 10.0.10 (the July 2026 release) is 24 June. Make sure to merge this PR on that date at the latest (or explicitly let me know that I should merge it), or it won't make it into that release.

As a reminder, if this is a product change, you also need Tactics approval before merging this PR (test-only or infra-only changes don't require Tactics approval).

[eduardo-vp]

/ba-g Ran the pipeline a second time, the only failure it's because of that (Artifact CoreCLRCrossDacArtifacts already exists).

[JulieLeeMSFT]

We have a tactics approval, but given the risk, we should run it for a few more weeks before merging.

[github-actions]

Workflow state for the Holistic Review Orchestrator.

{
  "version": 5,
  "last_dispatched_commit": "5c44c9de9b295e106d3f80b88f04991e3ae58cdb",
  "last_dispatched_base_ref": "release/10.0",
  "last_dispatched_base_sha": "19775d9fe43decd5704c97e05647b986278381fb",
  "last_reviewed_commit": "5c44c9de9b295e106d3f80b88f04991e3ae58cdb",
  "last_reviewed_base_ref": "release/10.0",
  "last_reviewed_base_sha": "19775d9fe43decd5704c97e05647b986278381fb",
  "last_recorded_worker_run_id": "29677695217",
  "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": "5c44c9de9b295e106d3f80b88f04991e3ae58cdb",
      "review_id": 4730522271
    }
  ]
}

[github-actions]

Holistic Review

Motivation: This is a backport of the merged #129203 to release/10.0. It addresses a customer-reported performance regression where apps running test suites under a high volume of generic method instantiations were 6x-8x slower on Linux (issue #123124). The root cause was InstantiatedMethodDesc::NewInstantiatedMethodDesc acquiring the m_InstMethodHashTableCrst, releasing it, building the MethodDesc, and then re-acquiring it. This create/publish window let multiple threads race to build the same instantiation, discarding the losers' allocations into the LoaderHeap free list, which grew far beyond expectation.

Approach: Two coordinated changes in genmeth.cpp. (1) The lock is now held across both the initial lookup and the MethodDesc creation/insertion, so only one thread allocates and publishes a given instantiation, eliminating the wasted allocations. (2) The generic-method constraint check is moved to the beginning of the function (before any allocation) by extracting the constraint-validation logic from MethodDesc::SatisfiesMethodConstraints into a new free function SatisfiesMethodConstraintsForInstantiation, which operates on the generic method def, parent type handle, and instantiation directly. SatisfiesMethodConstraints now delegates to it. The bulk of the visible diff is re-indentation from restructuring the lock scope; the actual behavioral change is small.

Summary: I verified this backport is content-identical to the merged and validated original PR #129203 (git show 33dc710): the two patches differ only in diff line offsets and blob index hashes, with zero source-content divergence. The refactor preserves semantics: SatisfiesMethodConstraintsForInstantiation reproduces the original constraint loop verbatim (including the typicalInstMatchesMethodInst short-circuit for the InstantiationContext), and the canonical-instantiation exemption is preserved — the exempt check's second operand changed from pNewMD->GetClassInstantiation() to the equivalent pExactMT->GetInstantiation(), which is required since the check now runs before pNewMD exists. Moving the constraint check earlier is safe because it does not depend on the created MethodDesc. CheckRestore() is now called unconditionally at the end for both the freshly-created and the race-lost (found-existing) paths, matching the prior behavior where both paths called it. The original change soaked in main for a week without regressions. This is a low-risk, mechanically clean backport. No actionable findings.

One non-blocking observation carried over from the original (not introduced by this backport, so not a new finding): the PR notes no regression test was added because reproducing requires a high volume of instantiations. That is a reasonable tradeoff for a servicing backport.

Note

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

Generated by Holistic Review · 65.5 AIC · ⌖ 10.2 AIC · ⊞ 10K

[korchak-aleksandr]

We have a tactics approval, but given the risk, we should run it for a few more weeks before merging.

@JulieLeeMSFT @eduardo-vp
When approximately do you plan to merge this PR ?
Thank you in advance.

[JulieLeeMSFT]

@korchak-aleksandr, merging now.

[JulieLeeMSFT]

@rbhanda, please change the milestone.