Skip to content

Test failure: FindServicePoint_ReturnedServicePointMatchesExpectedValues #56689

New issue
New issue
Closed
#57016
Closed
Test failure: FindServicePoint_ReturnedServicePointMatchesExpectedValues#56689
#57016
Assignees
AndyAyersMS
Labels
area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMItest-run-coreTest failures in .NET Core test runs
Milestone
6.0.0

Description

@karelz
karelz
opened on Aug 1, 2021

Test: System.Net.Tests.ServicePointManagerTest.FindServicePoint_ReturnedServicePointMatchesExpectedValues

Failures 6/1-8/1 (incl. PRs):

Day Run OS
7/25 fullpgo Windows.10.Arm64v8.Open
7/31 PR #56651 - fullpgo Ubuntu.1804.Arm64.Open
7/31 fullpgo Ubuntu.1804.Arm64.Open

Failure:

System.MissingMethodException : .ctor

   at System.RuntimeMethodHandle.IsCAVisibleFromDecoratedType(QCallTypeHandle attrTypeHandle, RuntimeMethodHandleInternal attrCtor, QCallTypeHandle sourceTypeHandle, QCallModule sourceModule)
   at System.Reflection.CustomAttribute.AddCustomAttributes(ListBuilder`1& attributes, RuntimeModule decoratedModule, Int32 decoratedMetadataToken, RuntimeType attributeFilterType, Boolean mustBeInheritable, ListBuilder`1 derivedAttributes) in /_/src/coreclr/System.Private.CoreLib/src/System/Reflection/RuntimeCustomAttributeData.cs:line 1263
   at System.Reflection.CustomAttribute.GetCustomAttributes(RuntimeModule decoratedModule, Int32 decoratedMetadataToken, Int32 pcaCount, RuntimeType attributeFilterType) in /_/src/coreclr/System.Private.CoreLib/src/System/Reflection/RuntimeCustomAttributeData.cs:line 1132
   at System.Reflection.CustomAttribute.GetCustomAttributes(RuntimeAssembly assembly, RuntimeType caType) in /_/src/coreclr/System.Private.CoreLib/src/System/Reflection/RuntimeCustomAttributeData.cs:line 1051
   at System.Reflection.RuntimeAssembly.GetCustomAttributes(Type attributeType, Boolean inherit) in /_/src/coreclr/System.Private.CoreLib/src/System/Reflection/RuntimeAssembly.cs:line 307
   at System.Attribute.GetCustomAttributes(Assembly element, Type attributeType, Boolean inherit) in /_/src/coreclr/System.Private.CoreLib/src/System/Attribute.CoreCLR.cs:line 737
   at System.Attribute.GetCustomAttribute(Assembly element, Type attributeType, Boolean inherit) in /_/src/coreclr/System.Private.CoreLib/src/System/Attribute.CoreCLR.cs:line 782
   at System.Attribute.GetCustomAttribute(Assembly element, Type attributeType) in /_/src/coreclr/System.Private.CoreLib/src/System/Attribute.CoreCLR.cs:line 775
   at System.Reflection.CustomAttributeExtensions.GetCustomAttribute(Assembly element, Type attributeType) in /_/src/libraries/System.Private.CoreLib/src/System/Reflection/CustomAttributeExtensions.cs:line 13
   at System.Reflection.CustomAttributeExtensions.GetCustomAttribute[T](Assembly element) in /_/src/libraries/System.Private.CoreLib/src/System/Reflection/CustomAttributeExtensions.cs:line 30
   at System.Environment.get_Version() in /_/src/libraries/System.Private.CoreLib/src/System/Environment.cs:line 193
   at Microsoft.DotNet.RemoteExecutor.RemoteExecutor.IsNetCore() in /_/src/Microsoft.DotNet.RemoteExecutor/src/RemoteExecutor.cs:line 96
   at Microsoft.DotNet.RemoteExecutor.RemoteExecutor.GetConsoleAppArgs(RemoteInvokeOptions options, IEnumerable`1& toDispose) in /_/src/Microsoft.DotNet.RemoteExecutor/src/RemoteExecutor.cs:line 421
   at Microsoft.DotNet.RemoteExecutor.RemoteExecutor.Invoke(MethodInfo method, String[] args, RemoteInvokeOptions options, Boolean pasteArguments) in /_/src/Microsoft.DotNet.RemoteExecutor/src/RemoteExecutor.cs:line 397
   at Microsoft.DotNet.RemoteExecutor.RemoteExecutor.Invoke(Action method, RemoteInvokeOptions options) in /_/src/Microsoft.DotNet.RemoteExecutor/src/RemoteExecutor.cs:line 118
   at System.Net.Tests.ServicePointManagerTest.FindServicePoint_ReturnedServicePointMatchesExpectedValues() in /_/src/libraries/System.Net.ServicePoint/tests/ServicePointManagerTest.cs:line 307

Activity

  1. karelz
    added
    area-System.Net
    test-run-coreTest failures in .NET Core test runs
    on Aug 1, 2021
  2. ghost added
    untriagedNew issue has not been triaged by the area owner
    on Aug 1, 2021

  4. karelz commented on Aug 1, 2021

    @karelz
    karelz
    on Aug 1, 2021
    MemberAuthor

    @dotnet/jit-contrib this looks like PGO run failure only + missing method ... is it a known problem?

  5. JulieLeeMSFT
    assigned
    jakobbotsch
    and
    AndyAyersMS
    and unassigned
    AndyAyersMS
    on Aug 2, 2021
  6. JulieLeeMSFT
    removed
    untriagedNew issue has not been triaged by the area owner
    on Aug 2, 2021

  7. JulieLeeMSFT commented on Aug 2, 2021

    @JulieLeeMSFT
    JulieLeeMSFT
    on Aug 2, 2021
    Member

    @jakobbotsch PTAL.

  8. JulieLeeMSFT
    added this to the 6.0.0 milestone on Aug 2, 2021

  9. AndyAyersMS commented on Aug 3, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 3, 2021
    Member

    @jakobbotsch I'll probably be looking at this some too, since this is the last major failure symptom in the libraries-pgo runs, so let's share notes/progress.

  10. AndyAyersMS
    added
    area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
    and removed
    area-System.Net
    on Aug 3, 2021

  11. AndyAyersMS commented on Aug 3, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 3, 2021
    Member

    Assuming this is codegen, so updated area label.

  12. 1 remaining item

  13. AndyAyersMS commented on Aug 4, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 4, 2021
    Member

    Also note this test DLL seems unable to be run via corerun, you need to swap parts into the test host and run that way, as some tests try to re-exec dotnet.

  14. AndyAyersMS commented on Aug 4, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 4, 2021
    Member

    Ok, I can repro if I bog the machine down with stress -c 4 ...

  15. AndyAyersMS commented on Aug 5, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 5, 2021
    Member

    I've isolated this down to something PGO does to System.Reflection.CustomAttribute.AddCustomAttributes(ListBuilder1& attributes, RuntimeModule decoratedModule, Int32 decoratedMetadataToken, RuntimeType attributeFilterType, Boolean mustBeInheritable, ListBuilder1 derivedAttributes).

    Suspect it involves inlining so will try and pin that down next.

  16. jakobbotsch
    assigned
    AndyAyersMS
    and unassigned
    jakobbotsch
    on Aug 6, 2021

  17. AndyAyersMS commented on Aug 6, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 6, 2021
    Member

    Inline analysis indicates it's likely this chain of inlines (plus maybe some others)

    Inlines into 06005215 [via ReplayPolicy] System.Reflection.CustomAttribute:AddCustomAttributes....
  [11 IL=0103 TR=000076 06005216] [accepted by log replay] System.Reflection.CustomAttribute:FilterCustomAttributeRecord(System.Reflection.MetadataToken,byref,System.Reflection.RuntimeModule,System.Reflection.MetadataToken,System.RuntimeType,bool,byref,byref,byref,byref):bool
    [103 IL=0510 TR=000702 06004F1B] [accepted by log replay] System.Runtime.CompilerServices.QCallTypeHandle:.ctor(byref):this
      [105 IL=0014 TR=002046 06000506] [accepted by log replay] System.RuntimeTypeHandle:get_Value():long:this

  18. AndyAyersMS commented on Aug 6, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 6, 2021
    Member

    I don't have a general-purpose inline XML reducer (yet). The method normally does 136 inlines, and if I cut off inlining at 104, the tests pass, and at 105 they fail.

    Inline 105 is RuntimeTypeHandle:get_Value noted above. While it is a small method, it has control flow, and biased PGO data, and inlining it leads to wide scale block reordering, so it's not that easy to diff between the 104 and 105 codegen to see if there are any unexpected changes. Disabling optOptimizeLayout in an attempt to minimize the ordering diffs makes the 105 case pass, so layout plays some kind of role in the bug.

    So I may work on reducing the inline set further; I suspect there's a version that does a much smaller number of inlines that perhaps will have more tractable diffs.

    Other options are to try and debug the failure, try disabling other parts of the jit, find a simpler test case that also fails, or rely on some psychic debugging skills. I've tried the latter two and so far they haven't panned out.

  19. AndyAyersMS commented on Aug 7, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 7, 2021
    Member

    I managed to narrow it down by first noting that the failure required assertion prop and then by limiting the number of allowed assertions. This lead to a one-instruction diff in the generated code:

    ;; good (7 assertions)
IN0146: 000568                    mov     x0, #0
IN0147: 00056C                    mov     x2, x1
IN0148: 000570                    str     x20, [fp,#232]	// [V01 arg1]
IN0149: 000574                    str     x2, [fp,#200]	// [V13 loc7]

;; bad (8 assertions)
IN0146: 000568                    mov     x1, #0
IN0147: 00056C                    str     x20, [fp,#232]	// [V01 arg1]
IN0148: 000570                    str     x1, [fp,#200]	// [V13 loc7]

    and the upshot it looks like there's another invalid peephole:

    Generating: N1107 (  1,  2) [002721] ------------      t2721 =    CNS_INT   ref    null REG x1 $VN.Null
IN0146:                           mov     x1, #0
							GC regs: 100001 {x0 x20} => 100003 {x0 x1 x20}
                                                              /--*  t2721  ref    
Generating: N1109 (  2,  3) [001470] ----G-------              *  KEEPALIVE void   REG NA $2c5
							GC regs: 100003 {x0 x1 x20} => 100001 {x0 x20}
Generating: N1111 (  3,  2) [001471] -c----------      t1471 =    LCL_VAR   ref    V111 tmp83       u:2 NA (last use) REG NA $1ca
                                                              /--*  t1471  ref    
Generating: N1113 (  4,  3) [001472] ----G-------              *  KEEPALIVE void   REG NA $2c6
							Live vars: {V00 V01 V02 V03 V04 V06 V08 V11 V12 V14 V111 V119 V153 V155 V156 V157 V160 V170 V217} => {V00 V01 V02 V03 V04 V06 V08 V11 V12 V14 V119 V153 V155 V156 V157 V160 V170 V217}
							GCvars: {V03 V06 V12 V111 V155 V156} => {V03 V06 V12 V155 V156}
Generating: N1115 (  3,  2) [001543] ------------      t1543 =    LCL_VAR   ref    V119 tmp91       u:1 x0 (last use) REG x0 $5f6
                                                              /--*  t1543  ref    
Generating: N1117 (  3,  3) [000875] DA--G-------              *  STORE_LCL_VAR ref    V13 loc7         d:5 x1 REG x1
							V119 in reg x0 is becoming dead  [001543]
							Live regs: 3F00001 {x0 x20 x21 x22 x23 x24 x25} => 3F00000 {x20 x21 x22 x23 x24 x25}
							Live vars: {V00 V01 V02 V03 V04 V06 V08 V11 V12 V14 V119 V153 V155 V156 V157 V160 V170 V217} => {V00 V01 V02 V03 V04 V06 V08 V11 V12 V14 V153 V155 V156 V157 V160 V170 V217}
							GC regs: 100001 {x0 x20} => 100000 {x20}

 -- suppressing mov because previous instruction already moved from src to dst register.

    wondering why arm64's emitter::IsRedundantMov does not check isValidLastInsFormats for same src/same dest movs.

  20. AndyAyersMS commented on Aug 7, 2021

    @AndyAyersMS
    AndyAyersMS
    on Aug 7, 2021
    Member

    This seems to fix things, as best I can tell:

    diff --git a/src/coreclr/jit/emitarm64.cpp b/src/coreclr/jit/emitarm64.cpp
index 771ce88ec84..11263bd361b 100644
--- a/src/coreclr/jit/emitarm64.cpp
+++ b/src/coreclr/jit/emitarm64.cpp
@@ -15581,17 +15581,19 @@ bool emitter::IsRedundantMov(instruction ins, emitAttr size, regNumber dst, regN
         regNumber prevSrc    = emitLastIns->idReg2();
         insFormat lastInsfmt = emitLastIns->idInsFmt();
 
-        if ((prevDst == dst) && (prevSrc == src))
+
+
+        // Sometimes emitLastIns can be a mov with single register e.g. "mov reg, #imm". So ensure to
+        // optimize formats that does vector-to-vector or scalar-to-scalar register movs.
+        bool isValidLastInsFormats = ((lastInsfmt == IF_DV_3C) || (lastInsfmt == IF_DR_2G) || (lastInsfmt == IF_DR_2E));
+
+        if (isValidLastInsFormats && (prevDst == dst) && (prevSrc == src))
         {
             assert(emitLastIns->idOpSize() == size);
             JITDUMP("\n -- suppressing mov because previous instruction already moved from src to dst register.\n");
             return true;
         }
 
-        // Sometimes emitLastIns can be a mov with single register e.g. "mov reg, #imm". So ensure to
-        // optimize formats that does vector-to-vector or scalar-to-scalar register movs.
-        bool isValidLastInsFormats = ((lastInsfmt == IF_DV_3C) || (lastInsfmt == IF_DR_2G) || (lastInsfmt == IF_DR_2E));
-
         if ((prevDst == src) && (prevSrc == dst) && isValidLastInsFormats)
         {
             // For mov with EA_8BYTE, ensure src/dst are both scalar or both vector.

    Issue was seemingly there from the start, eg #38179.

  21. AndyAyersMS
    added a commit that references this issue on Aug 7, 2021

    JIT: arm64 redundant move peephole must check instruction format

    4c96273
  22. ghost added
    in-prThere is an active PR which will close this issue when it is merged
    on Aug 7, 2021
  23. AndyAyersMS
    closed this as completedin #57016on Aug 8, 2021
  24. ghost removed
    in-prThere is an active PR which will close this issue when it is merged
    on Aug 8, 2021
  25. AndyAyersMS
    added a commit that references this issue on Aug 8, 2021

    JIT: arm64 redundant move peephole must check instruction format (#57016

    22ae541
  26. ghost locked as resolved and limited conversation to collaborators on Sep 7, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMItest-run-coreTest failures in .NET Core test runs

Type

No type

Projects

No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions