[Regions] pCorProfilerInfo->GetObjectGeneration may fail when running under USE_REGIONS

[cshung]

This is a bug found when working on the filtering issue.

Problem:

The crux of the issue is that under USE_REGIONS, the GC may use some unseen memory address since the last GC during allocation. In particular, in soh_try_fit, it may use the get_new_region method to acquire a new region, chain it to the end of gen0 and allocate the object there. At that point of time, the s_currentGenerationTable maintained by the profiler is unaware of those new addresses, therefore during the ObjectAllocated callback, GetObjectGeneration for that newly allocated object would fail.

Repro:
The profiler authored for the gcallocate test can be repurposed to reproduce this bug. Here is the modified code. It does 3 different things:

1. It changes the flags so that it captures SOH as well
2. removed a println, and
3. assert in failure.

// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.

#include "gcallocateprofiler.h"

GUID GCAllocateProfiler::GetClsid()
{
    // {55b9554d-6115-45a2-be1e-c80f7fa35369}
	GUID clsid = { 0x55b9554d, 0x6115, 0x45a2,{ 0xbe, 0x1e, 0xc8, 0x0f, 0x7f, 0xa3, 0x53, 0x69 } };
	return clsid;
}

HRESULT GCAllocateProfiler::Initialize(IUnknown* pICorProfilerInfoUnk)
{
    Profiler::Initialize(pICorProfilerInfoUnk);

    HRESULT hr = S_OK;
    if (FAILED(hr = pCorProfilerInfo->SetEventMask2(COR_PRF_ENABLE_OBJECT_ALLOCATED | COR_PRF_MONITOR_OBJECT_ALLOCATED, COR_PRF_HIGH_BASIC_GC)))
    {
        printf("FAIL: ICorProfilerInfo::SetEventMask2() failed hr=0x%x", hr);
        return hr;
    }

    return S_OK;
}

HRESULT STDMETHODCALLTYPE GCAllocateProfiler::ObjectAllocated(ObjectID objectId, ClassID classId)
{
    COR_PRF_GC_GENERATION_RANGE gen;
    HRESULT hr = pCorProfilerInfo->GetObjectGeneration(objectId, &gen);
    if (FAILED(hr))
    {
        assert (false);
        printf("GetObjectGeneration failed hr=0x%x\n", hr);
        _failures++;
    }
    else if (gen.generation == COR_PRF_GC_LARGE_OBJECT_HEAP)
    {
        _gcLOHAllocations++;
    }
    else if (gen.generation == COR_PRF_GC_PINNED_OBJECT_HEAP)
    {
        _gcPOHAllocations++;
    }
    else
    {
        _failures++;
    }

    return S_OK;
}

HRESULT GCAllocateProfiler::Shutdown()
{
    Profiler::Shutdown();
    if (_gcPOHAllocations == 0)
    {
        printf("There is no POH allocations\n");
    }
    else if (_gcLOHAllocations == 0)
    {
        printf("There is no LOH allocations\n");
    }
    else if (_failures == 0)
    {
        printf("%d LOH objects allocated\n", (int)_gcLOHAllocations);
        printf("%d POH objects allocated\n", (int)_gcPOHAllocations);
        printf("PROFILER TEST PASSES\n");
    }
    fflush(stdout);

    return S_OK;
}

The bug can then be reproduced by setting these environment,

set CORECLR_ENABLE_PROFILING=1
set CORECLR_PROFILER={55b9554d-6115-45a2-be1e-c80f7fa35369}
set CORECLR_PROFILER_PATH=C:\dev\runtime\artifacts\tests\coreclr\windows.x64.Debug\profiler\gc\gcallocate\Profiler.dll

and run GCPerfSim with these parameters:

c:\dev\runtime\artifacts\tests\coreclr\Windows.x64.Debug\Tests\Core_Root\CoreRun.exe C:\dev\performance\artifacts\bin\GCPerfSim\release\netcoreapp5.0\GCPerfSim.dll -tc 6 -tagb 100.0 -tlgb 2.0 -lohar 0 -pohar 0 -sohsi 10 -lohsi 0 -pohsi 0 -sohsr 100-4000 -lohsr 102400-204800 -pohsr 100-4000 -sohpi 10 -lohpi 0 -sohfi 0 -lohfi 0 -pohfi 0 -allocType reference -testKind time

[ghost]

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

Issue Details

---

This is a bug found when working on the filtering issue.

Problem:

The crux of the issue is that under USE_REGIONS, the GC may use some unseen memory address since the last GC during allocation. In particular, in soh_try_fit, it may use the get_new_region method to acquire a new region, chain it to the end of gen0 and allocate the object there. At that point of time, the s_currentGenerationTable maintained by the profiler is unaware of those new addresses, therefore during the ObjectAllocated callback, GetObjectGeneration for that newly allocated object would fail.

Repro:
The profiler authored for the gcallocate test can be repurposed to reproduce this bug. Here is the modified code. It does 3 different things:

1. It changes the flags so that it captures SOH as well
2. removed a println, and
3. assert in failure.

// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.

#include "gcallocateprofiler.h"

GUID GCAllocateProfiler::GetClsid()
{
    // {55b9554d-6115-45a2-be1e-c80f7fa35369}
	GUID clsid = { 0x55b9554d, 0x6115, 0x45a2,{ 0xbe, 0x1e, 0xc8, 0x0f, 0x7f, 0xa3, 0x53, 0x69 } };
	return clsid;
}

HRESULT GCAllocateProfiler::Initialize(IUnknown* pICorProfilerInfoUnk)
{
    Profiler::Initialize(pICorProfilerInfoUnk);

    HRESULT hr = S_OK;
    if (FAILED(hr = pCorProfilerInfo->SetEventMask2(COR_PRF_ENABLE_OBJECT_ALLOCATED | COR_PRF_MONITOR_OBJECT_ALLOCATED, COR_PRF_HIGH_BASIC_GC)))
    {
        printf("FAIL: ICorProfilerInfo::SetEventMask2() failed hr=0x%x", hr);
        return hr;
    }

    return S_OK;
}

HRESULT STDMETHODCALLTYPE GCAllocateProfiler::ObjectAllocated(ObjectID objectId, ClassID classId)
{
    COR_PRF_GC_GENERATION_RANGE gen;
    HRESULT hr = pCorProfilerInfo->GetObjectGeneration(objectId, &gen);
    if (FAILED(hr))
    {
        assert (false);
        printf("GetObjectGeneration failed hr=0x%x\n", hr);
        _failures++;
    }
    else if (gen.generation == COR_PRF_GC_LARGE_OBJECT_HEAP)
    {
        _gcLOHAllocations++;
    }
    else if (gen.generation == COR_PRF_GC_PINNED_OBJECT_HEAP)
    {
        _gcPOHAllocations++;
    }
    else
    {
        _failures++;
    }

    return S_OK;
}

HRESULT GCAllocateProfiler::Shutdown()
{
    Profiler::Shutdown();
    if (_gcPOHAllocations == 0)
    {
        printf("There is no POH allocations\n");
    }
    else if (_gcLOHAllocations == 0)
    {
        printf("There is no LOH allocations\n");
    }
    else if (_failures == 0)
    {
        printf("%d LOH objects allocated\n", (int)_gcLOHAllocations);
        printf("%d POH objects allocated\n", (int)_gcPOHAllocations);
        printf("PROFILER TEST PASSES\n");
    }
    fflush(stdout);

    return S_OK;
}

The bug can then be reproduced by setting these environment,

set CORECLR_ENABLE_PROFILING=1
set CORECLR_PROFILER={55b9554d-6115-45a2-be1e-c80f7fa35369}
set CORECLR_PROFILER_PATH=C:\dev\runtime\artifacts\tests\coreclr\windows.x64.Debug\profiler\gc\gcallocate\Profiler.dll

and run GCPerfSim with these parameters:

c:\dev\runtime\artifacts\tests\coreclr\Windows.x64.Debug\Tests\Core_Root\CoreRun.exe C:\dev\performance\artifacts\bin\GCPerfSim\release\netcoreapp5.0\GCPerfSim.dll -tc 6 -tagb 100.0 -tlgb 2.0 -lohar 0 -pohar 0 -sohsi 10 -lohsi 0 -pohsi 0 -sohsr 100-4000 -lohsr 102400-204800 -pohsr 100-4000 -sohpi 10 -lohpi 0 -sohfi 0 -lohfi 0 -pohfi 0 -allocType reference -testKind time

Author:	cshung
Assignees:	-
Labels:	area-GC-coreclr
Milestone:	-