JIT: Avoid TLS access when restoring async contexts#127889
Draft
jakobbotsch wants to merge 3 commits intodotnet:mainfrom
Draft
JIT: Avoid TLS access when restoring async contexts#127889jakobbotsch wants to merge 3 commits intodotnet:mainfrom
jakobbotsch wants to merge 3 commits intodotnet:mainfrom
Conversation
github-actions
Bot
added
the
area-CodeGen-coreclr
Contributor
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR updates CoreCLR’s async context save/restore plumbing so the synchronous restore path can use a captured Thread instance instead of re-reading Thread.CurrentThreadAssumedInitialized (TLS) at the end of runtime async functions.
Changes:
- Extend
AsyncHelpers.CaptureContexts/AsyncHelpers.RestoreContextsto include the currentThreadas an explicit argument for the synchronous restore path. - Teach the JIT to allocate/track a new async
Threadlocal (including OSR/patchpoint offset plumbing) and pass it through to the restore helper. - Update the interpreter to allocate the extra local and adjust its async helper call setup (but there are currently critical argument-passing issues; see stored PR comments).
Reviewed changes
Copilot reviewed 14 out of 14 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| src/coreclr/vm/interpexec.cpp | Adjusts interpreter execution of async restore-on-suspend opcode to account for an additional local (currently mis-passing args to the on-suspension helper). |
| src/coreclr/System.Private.CoreLib/src/System/Runtime/CompilerServices/AsyncHelpers.CoreCLR.cs | Adds Thread to CaptureContexts outputs and RestoreContexts inputs to avoid TLS access on synchronous restore. |
| src/coreclr/jit/lclvars.cpp | Allocates stack home for a new async thread local and excludes it from certain frame offset assignments. |
| src/coreclr/jit/compiler.h | Introduces lvaAsyncThreadObjectVar local tracking in the JIT. |
| src/coreclr/jit/compiler.cpp | Adds OSR patchpoint offset reporting/mapping for the async thread local. |
| src/coreclr/jit/codegenxarch.cpp | Updates preserved-area size accounting/assertions for the new async thread local. |
| src/coreclr/jit/codegencommon.cpp | Updates tracked stack pointer contiguity logic to treat the new async thread local as a special case. |
| src/coreclr/jit/codegenarmarch.cpp | Updates preserved-area size accounting/assertions for the new async thread local. |
| src/coreclr/jit/codegenarm64.cpp | Updates funclet prolog/epilog size capture to account for the new async thread local. |
| src/coreclr/jit/codegenarm.cpp | Updates funclet prolog/epilog size capture to account for the new async thread local. |
| src/coreclr/jit/async.cpp | Creates/passes the new async thread local through capture/restore calls and excludes it from suspension live sets. |
| src/coreclr/interpreter/compiler.h | Tracks a new interpreter local index for the async thread local. |
| src/coreclr/interpreter/compiler.cpp | Allocates the new interpreter local and wires it into async helper calls (currently has critical call-arg construction and on-suspend base-local issues). |
| src/coreclr/inc/patchpointinfo.h | Extends patchpoint metadata to include async thread local FP-relative offset and associated accessors. |
Comment on lines
+8277
to
8283
| int32_t numArgs = 3; | ||
| int32_t *callArgs = getAllocator(IMK_CallInfo).allocate<int32_t>(numArgs + 1); | ||
| callArgs[0] = execContextAddressVar; | ||
| callArgs[1] = syncContextAddressVar; | ||
| callArgs[2] = CALL_ARGS_TERMINATOR; | ||
| callArgs[1] = threadAddressVar; | ||
| callArgs[1] = execContextAddressVar; | ||
| callArgs[2] = syncContextAddressVar; | ||
| callArgs[3] = CALL_ARGS_TERMINATOR; | ||
| m_pLastNewIns->info.pCallInfo->pCallArgs = callArgs; |
| AddIns(INTOP_RESTORE_CONTEXTS_ON_SUSPEND); | ||
| m_pLastNewIns->data[0] = suspendDataIndex; | ||
| m_pLastNewIns->SetSVars3(varAllocatedContinuation, m_continuationArgIndex, m_execContextVarIndex /* We know the sync context immediately follows */); | ||
| m_pLastNewIns->SetSVars3(varAllocatedContinuation, m_continuationArgIndex, m_threadObjVarIndex /* We know the contexts immediately follow */); |
Comment on lines
+4452
to
4466
| OBJECTREF thread = LOCAL_VAR(ip[4], OBJECTREF); | ||
| OBJECTREF executionContext = LOCAL_VAR(ip[4] + INTERP_STACK_SLOT_SIZE, OBJECTREF); | ||
| OBJECTREF syncContext = LOCAL_VAR(ip[4] + 2 * INTERP_STACK_SLOT_SIZE, OBJECTREF); | ||
|
|
||
| InterpAsyncSuspendData *pAsyncSuspendData = (InterpAsyncSuspendData*)pMethod->pDataItems[ip[5]]; | ||
| MethodDesc *restoreContextsMethod = pAsyncSuspendData->restoreContextsOnSuspensionMethod; | ||
|
|
||
| returnOffset = ip[1]; | ||
| callArgsOffset = pMethod->allocaSize; | ||
| // Pass argument to the target method | ||
| LOCAL_VAR(callArgsOffset, int32_t) = resumed; | ||
| LOCAL_VAR(callArgsOffset + INTERP_STACK_SLOT_SIZE, OBJECTREF) = executionContext; | ||
| LOCAL_VAR(callArgsOffset + INTERP_STACK_SLOT_SIZE * 2, OBJECTREF) = syncContext; | ||
| LOCAL_VAR(callArgsOffset + INTERP_STACK_SLOT_SIZE * 1, OBJECTREF) = thread; | ||
| LOCAL_VAR(callArgsOffset + INTERP_STACK_SLOT_SIZE * 2, OBJECTREF) = executionContext; | ||
| LOCAL_VAR(callArgsOffset + INTERP_STACK_SLOT_SIZE * 3, OBJECTREF) = syncContext; | ||
| targetMethod = restoreContextsMethod; |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Avoid TLS access when restoring async contexts in the synchronous case at the end of runtime async functions by saving the
Threadobject used when we captured them in a local.