Module Overview. CLR Initialization

Transcription

1 CLR Initialization

2 Module Overview CLR Initialization Getting to Main Method PE Layout & CLR Headers Process Initialization & EE Shim EE Startup CLR Artifacts & Loader Heaps The Managed Object Common Slow Startups Review 2

3 Getting to Main Method OS Loader EE Shim CLR JIT Main method Process initialization phase OS loader finds and load the Execution Engine Shim EE Shim determines version, bitness, path etc. Loads CLR (Mscorwks.dll / clr.dll) CLR starts up Prepares to run managed code JIT gets loaded and compiles mscorlib.dll etc. Main method finally runs 3

4 Portable Executable Layout (EXE/DLL) PE Header CLR Header IL Metadata Header Metadata PE Header: Info about exe/dll type Machine type (x86/x64) CUI/GUI? Timestamp etc., CLR Header: Info used by CLR Version of CLR required MetaDataToken of entry Method Location and size of metadata, Resources etc., IL: Code compile time IL is a higher-level, CPU independent language Metadata Header: Contains version of the assembly Flags and headers that describe location, length of tables & heaps within metadata Metadata: Every managed module contains binary metadata tables 4

5 Physical Layout of a Managed Assembly PE Header CLR Header IL Metadata Header Metadata Whether it is an exe, dll, ocx etc. Not of much importance to.net CLR version (metadata header) Localization Size, resource, module s entry point e.g., main() Intermediate Language CPU independent language (similar to ByteCode) Code that the compiler produced when the Metadata Version of assembly source is header compiled. IL Contains is a higher-level, version of CPU the assembly independent language Contains other flags and headers which Info about describe classes location defined & and classes length referenced of Metadata tables Info about & module heaps and within resources the metadata etc. Every managed module contains binary metadata tables

6 PE Layout Raw Output dumpbin.exe /headers <EXE> FILE HEADER VALUES 14C machine (x86) 3 number of sections 4C26E437 time date stamp Sun Jun 27 01:40: file pointer to symbol table 0 number of symbols E0 size of optional header 102 characteristics Executable 32 bit word machine OPTIONAL HEADER VALUES 10B magic # (PE32) 8.00 linker version size of code 2000 size of initialized data 0 size of uninitialized data 254BE entry point (110254BE) 2000 base of code base of data 2A000 size of image 1000 size of headers 0 checksum 2 subsystem (Windows GUI) HEX disassembly of <EXE> 6

7 CLR Header Within PE Layout How to find Main method? Raw output from ildasm.exe CLR Header: Header size: 0x Major runtime version: 0x0002 Minor runtime version: 0x0005 0x00021de0 [0x ] address [size] of Metadata Directory: Flags: 0x Entry point token: 0x TypeDef #10 ( b) TypDefName: NetDebugDemos.Startup ( B) Flags : [Public] [AutoLayout] [Class] [AnsiClass] [BeforeFieldInit] ( ) Extends : [TypeRef] System.Object Method #2 ( ) [ENTRYPOINT] MethodName: Main ( ) Flags : [Private] [Static] [HideBySig] [ReuseSlot] ( ) RVA : 0x f ImplFlags : [IL] [Managed] ( ) CallCnvntn: [DEFAULT] ReturnType: Void No arguments. 7

8 Demo: Inspect PE Layout PE Headers CLR Headers

9 Process Initialization Loading EE Shim OS Loader finds process entry point Transfer control to Execution Engine (EE) shim loader (mscoree.dll) EE Shim determines version, bitness, path of CLR Uses default Concurrent GC unless config file or CorBindToRuntimeHost is used e.g. ASP.NET (Background GC/ CLRCreateInstance for 4.0) 0:000> kl Child-SP RetAddr Call Site clr.dll / `0012f fe`f KERNELBASE!LoadLibraryExW+0x19c `0012f8b fe`f9123cf8 mscoreei!runtimedesc::loadlibrary+0x82 mscorwks.dll `0012fa fe`f9123ca9 mscoreei!runtimedesc::loadmainruntimemodulehelper+0x4a `0012fa fe`f9123a85 mscoreei!runtimedesc::loadmainruntimemodule+0x4a `0012fae fe`f mscoreei!runtimedesc::ensureloaded+0xa `0012fb fe`f mscoreei!runtimedesc::getprocaddressinternal+0x `0012fba fe`f mscoreei!clrruntimeinfoimpl::getprocaddress+0x5c `0012fc fe`f91232f1 mscoreei!getcorexemainentrypoint+0x `0012fcf fe`f91b5b21 mscoreei!_corexemain+0x `0012fd `76aff56d MSCOREE!_CorExeMain_Exported+0x `0012fd `770d2cc1 KERNEL32!BaseThreadInitThunk+0xd `0012fd ` ntdll!rtluserthreadstart+0x1d 9

10 EE Startup Majority of CLR initialization done here Initializes ETW tracing, GC & Loader Heaps Creates System, Shared & Default domain(s) Loads mscorlib.dll & creates fundamental types e.g. System.Object 0:000> kl Child-SP RetAddr Call Site `0022fac fe`f89b629e clr!eestartup+0x `0022fb fe`f89cd62e clr!ensureeestarted+0xe `0022fb fe`f896ba45 clr!_corexemaininternal+0x7f `0022fbe fe`f clr!_corexemain+0x `0022fc fe`f91b5b21 mscoreei!_corexemain+0x `0022fc `76aff56d MSCOREE!_CorExeMain_Exported+0x `0022fc `770d2cc1 KERNEL32!BaseThreadInitThunk+0xd `0022fcb ` ntdll!rtluserthreadstart+0x1d 10

11 Demo: Break on EE Startup clr!eestartup

12 CLR Artifacts Internal data structures to track objects EEClass - Mold for every unique type - Describes a type s field & assembly from etc. MethodTable - EEClass + MethodTable = CLR s internal representation of objects MethodDesc - CLR s internal structure to track methods in a type - E.g. tracks state of a method if Jitted or not Note Class Type = design time = runtime 12

13 AppDomain What is it? Can be thought of as process within process Sandboxing code for isolation purposes Cost is cheaper than creating a new process Types of AppDomains - System Domain Responsible creating shared & default domain - Shared Domain Domain neutral code lives here - Default Domain Application code gets loaded & run here 13

14 The.NET Runtime in a Process (CLR) e.g. w3wp.exe System Domain AppDomain (!dumpdomain) Shared Domain AppDomain #1 AppDomain #2 14

15 CLR: Assembly & Module w3wp.exe System Domain Shared Domain AppDomain #1 AppDomain (!dumpdomain) Assembly (Foo1.dll) Assembly (Foo2.dll) Assembly/Module (!dumpassembly/!dumpmodule) AppDomain #2 Assembly (Foo3.dll) 15

16 CLR: EEClass w3wp.exe System Domain Shared Domain AppDomain (!dumpdomain) Assembly (Foo1.dll) Assembly/Module (!dumpassembly/!dumpmodule) EEClass (Class1) EEClass (!dumpclass) AppDomain #1 AppDomain #2 Assembly (Foo2.dll) Assembly (Foo3.dll) EEClass (Class2) EEClass (Class3) EEClass (Class4) 16

17 CLR: MethodTable w3wp.exe System Domain Shared Domain AppDomain #1 AppDomain #2 AppDomain (!dumpdomain) Assembly (Foo1.dll) Assembly (Foo2.dll) Assembly (Foo3.dll) Assembly/Module (!dumpassembly/!dumpmodule) EEClass (Class1) EEClass (Class2) EEClass (Class3) EEClass (Class4) EEClass (!dumpclass) MethodTable (!dumpmt) 17

18 CLR: MethodDesc w3wp.exe System Domain Shared Domain AppDomain #1 AppDomain #2 AppDomain (!dumpdomain) Assembly (Foo1.dll) Assembly (Foo2.dll) Assembly (Foo3.dll) Assembly/Module (!dumpassembly/!dumpmodule) EEClass (Class1) EEClass (Class2) EEClass (Class3) EEClass (Class4) EEClass (!dumpclass) MethodTable (!dumpmt) MethodDesc (FunctionOne) MethodDesc (FunctionTwo) MethodDesc (FunctionThree) 18

19 Multi-Domain Applications LoaderOptimizationAttribute or AppDomainSetup.LoaderOptimization SingleDomain No Assembly sharing between Domains Ngen ed Image gets used only for the first Domain Increased Memory Usage, increased Application Load Time, unexpected Jitting Default behavior safest way MultiDomain As much as possible will be loaded Domain Neutral. LoaderOptimizationAttribute on Main() No real control over loading possible anymore Better: AppDomainSetup usage MultiDomainHost Only Assemblies within the GAC will be shared Eg: ASP.NET

20 Loader Heaps Store CLR Artifacts What are CLR artifacts? - E.g. EEClass, MethodTable & MethodDesc Common Loader Heaps - HighFrequencyHeap MethodTable, MethodDesc & FieldDesc - LowFrequencyHeap EEClass 0:005>!eeheap -loader Loader Heap: System Domain: fef LowFrequencyHeap: ff (2000:1000) Size: 0x1000 (4096) bytes. HighFrequencyHeap: ff (8000:1000) Size: 0x1000 (4096) bytes. StubHeap: ff0001a000(2000:2000) Size: 0x2000 (8192) bytes. 20

21 Demo: Loader Heap & AppDomain Load SOS System, Shared & Default domain Loader Heap Sizes

22 The Managed Object Tying it all together! Chunk of memory with instance data Lives in GC Heap i.e. separate from all other CLR artifacts CLR identifies its type through EEClass, MethodTable etc. System.Object = basic building block for all other managed objects An object in memory needs to keep info such as: - Type - Instance contents (fields) - Other data structures bound to objects like hash codes & locks 22

23 Demo: Managed objects MethodTable - dd <objadd> L1!DumpObj String Array Object Properties

24 Slow Startup Common Scenarios Cold Startup - Slow disk I/O Warm Startup - Many dynamic assemblies e.g. ASP.NET & Serializers - Large cache initialization - Contentions during warm startup 24

25 Identify Slow Startups 1 st Run after boot? No Yes Re-run app to ensure not cold startup issue Yes Slow Startup? No Collect ETW traces on I/O performance Collect perf counters &/or profiler traces to further isolate issue 25

26 Performance Counters Which counters to look at?.net CLR Jit - # of IL Bytes Jitted - # of Methods Jitted - % Time in Jit.NET CLR Loading - Bytes in Loader Heap LogicalDisk - % Idle Time - Current Disk Queue Length 26

27 ETW Trace Disk I/O Performance 1. xperf -on PROC_THREAD+LOADER+CSWITCH+FILENAME+FILE_IO+FILE_IO _INIT+DRIVERS -f kernel.etl -stackwalk CSwitch+DiskReadInit+DiskWriteInit+DiskFlushInit+FileCreate+FileCle anup+fileclose+fileread+filewrite -BufferSize MaxBuffers MaxFile FileMode Circular 2. <Reproduce issue> 3. xperf -d ColdStartupIssue.etl 27

28 Review 1. Name the different parts within the layout of a managed assembly. 2. What are AppDomains, MethodTables, and MethodDescs? 3. Name the SOS command that obtains information from each of these. 28

29 Reference Microsoft PE and COFF Specification - An In-Depth Look into the Win32 Portable Executable File Format Part Drill Into.NET Framework Internals to See How the CLR Creates Runtime Objects - A Model for cold startup time of an application on Windows - NGen: Measuring Warm Startup Performance with Xperf

30 Reference (Cont.) Windows Performance Analysis Tools

31 2008 Microsoft Corporation. All rights reserved. Microsoft, Windows, Windows Vista and other product names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries. The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION. 31