[NativeAOT] Making thread local storage completely managed.

[VSadov]

Basically implementing m_pThreadLocalModuleStatics as a managed object[][] array

• no need to involve unmanaged allocator and GC handles, thus access to threadstatics has one less indirection
• managing the threadlocal storage happens entirely in managed code
• destruction of the entire threadstatics structure is a noop

The threadstatics root is reported together with other thread-owned roots like locals.
Once the owning thread's lifecycle gets to the stage when it stops participating in root reporting, m_pThreadLocalModuleStatics can be reclaimed.

[ghost]

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

Issue Details

---

Basically implementing m_pThreadLocalModuleStatics as a managed object[][] array

• no need to involve unmanaged allocator and GC handles, thus access to threadstatics has one less indirection
• managing the threadlocal storage happens entirely in managed code
• destruction of the entire threadstatics structure is a noop

Author:	VSadov
Assignees:	VSadov
Labels:	area-NativeAOT-coreclr
Milestone:	-

[VSadov]

closed/reopened to unstuck the CLA check

[VSadov]

/azp run runtime-extra-platforms

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[VSadov]

Assert in ilc (I think unrelated to this change)

 ILC: D:\a\_work\1\s\src\coreclr\jit\lir.cpp:1691
ILC: Assertion failed 'node->IsUnusedValue() && "found an unmarked unused value"' in 'System.Net.Sockets.Tests.ConnectEap+<>c__DisplayClass1_0+<<ConnectAsync_WithData_DataReceived>b__0>d:MoveNext():this' during 'Lowering nodeinfo' (IL size 473; hash 0x4698aa3f; FullOpts)

[jkotas]

Thanks!

[VSadov]

the ilc assert is #84112

[VSadov]

Thanks!

[MichalStrehovsky]

the ilc assert is #81356

That issue is closed. Maybe same assert, but definitely a different root cause.

Could you create an issue and exclude the test from building like I did in #83167 and #83250 (PR that added exclusion when I first noticed it).

We should not be merging PRs that have an entire test leg on the floor because of a build failure.

[VSadov]

Right. It is a slightly different assert. It needs a recent build but happens without my change.
(a testrun with the commit just prior to mine reproes the failure, but it might have been introduced earlier)

The failure is only in Chk builds, so who does not run "extra-platforms" will not see the failure.

[VSadov]

The assert looked very similar at first glance and I thought it was fixed since my PR hit it. I did not notice that the issue is fairly old.