Expose a Lock type in preview mode

[kouvel]

• For now, the Lock type requires preview features to be enabled
• Ported CoreCLR's AwareLock implementation to C# with a bit of refactoring, folded in a couple of ideas from NativeAOT's previous Lock implementation, and fixed a couple of issues.
• Added an adaptive spin strategy to reduce CPU time from spin-waiting when spin-waits are not effective
• This implementation replaces NativeAOT's Lock implementation. The performance of acquiring a lock under contention is improved in NativeAOT.

API review: #34812
Tracking issue: #87673

[ghost]

Note regarding the new-api-needs-documentation label:

This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change.

[ghost]

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

Issue Details

---

• For now, the Lock type requires preview features to be enabled
• Ported CoreCLR's AwareLock implementation to C# with a bit of refactoring, folded in a couple of ideas from NativeAOT's previous Lock implementation, and fixed a couple of issues.
• Added an adaptive spin strategy to reduce CPU time from spin-waiting when spin-waits are not effective
• This implementation replaces NativeAOT's Lock implementation. The performance of acquiring a lock under contention is improved in NativeAOT.

API review: #34812

Author:	kouvel
Assignees:	kouvel
Labels:	area-System.Threading
Milestone:	8.0.0

[vargaz]

The mono changes look ok.

[kouvel]

Here are some perf numbers on NativeAOT. I used a test that has multiple threads enter a lock, perform a short delay, exit the lock, and perform another short delay. Throughput is the number of enter+delay+exit+delay rounds performed per ms. The delay is a recursive calculation of a random Fibonacci number between 4 and 9. The numbers are from my Intel x64 6-core 12-thread machine on Windows.

Using Monitor:

Threads	Throughput before	CPU usage before (%)	Throughput after	CPU usage after (%)
1	6800	9.3	6800	9.1
2	5700	16.9	5700	16.3
3	5100	23.8	5300	23.5
6	3900	44.9	5400	27.5
12	2500	56.3	5300	24.5
24	2600	64.8	5300	24.9

Using Lock:

Threads	Throughput	CPU usage (%)
1	6800	8.4
2	6400	16.0
3	5800	23.9
6	5500	28.5
12	5500	26.1
24	5500	24.8

[danmoseley]

I don't think I'm following how this gets thrown. It would mean _recursionCount was -1. I don't see a test in LockTests for LockRecursionException.

[danmoseley]

Oh - you're using 32 bit overflow like Native AOT .. since this is a new type, how about picking something smaller that is still always going to mean a bug in the app, but small enough that it won't cause a hang in the app (and also becomes feasibly unit testable). Eg., 65K

[danmoseley]

in fact that's what the comment suggests "is expected to be high enough that it would typically not be reached when the lock is used properly."

[kouvel]

Rebased to fix conflicts

[VSadov]

/azp run runtime-extra-platforms

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[kouvel]

Trying a rebase due to some unrelated build failures.

[kouvel]

The remaining issues in the CI appear to be unrelated.

[radical]

This was merged on Red and broke Wasm.Build.Tests on main. #94212

[radical]

Also one of the failures is:

Unhandled exception. System.InvalidOperationException: There is no currently active test.
   at Xunit.Sdk.TestOutputHelper.QueueTestOutput(String output) in /_/src/xunit.execution/Sdk/Frameworks/TestOutputHelper.cs:line 60
   at System.Diagnostics.AsyncStreamReader.FlushMessageQueue(Boolean rethrowInNewThread)
--- End of stack trace from previous location ---
   at System.Threading.ThreadPoolWorkQueue.Dispatch()
   at System.Threading.PortableThreadPool.WorkerThread.WorkerThreadStart()
./RunTests.sh: line 105:    83 Aborted                 (core dumped) dotnet exec xunit.console.dll Wasm.Build.Tests.dll -xml $XHARNESS_OUT/testResults.xml $HELIX_XUNIT_ARGS -nocolor -verbose -notrait category=IgnoreForCI -notrait category=failing $XUnitTraitArg $RSP_FILE

Could this be related to the change here? I have never seen this happen on the tests before.

cc @lewing

[jkotas]

Also introduced #94227

[kouvel]

This was merged on Red and broke Wasm.Build.Tests on main. #94212
Could this be related to the change here? I have never seen this happen on the tests before.

I have no idea how this PR could cause those failures, apologies if there was a mishap. There are only a few lines of code in this PR that would affect wasm, do you see anything that could be relevant? I'll see if I can get a repro, but otherwise I don't see any link between the failure and this PR.

Also introduced #94227

#94241 should fix the issue.

[kouvel]

Could this be related to the change here? I have never seen this happen on the tests before.

Only thing I can think of at the moment is that there is a pragma warning disable for using preview features when we should probably be enabling preview features for the corelib project. I don't know what kind of effect that would have, but I think that's something that was missed after local testing and unaddressed, so I'll fix that anyway.