CS609

SYSTEM PROGRAMMING
93 T0 165
Arranged By Amna

Topic 93: Process Identities

Process identity can be obtained from the PROCESS_INFORMATION structure. The CreateProcess() API
returns the process information in its last parameter. This structure contains the child process handle
and id as well as the primary thread handle and id. Closing the child process handle does not destroys
the process. It only closes the access of the parent process to the child.

Let’s consider these pair of functions for this purpose:

HANDLE GetCurrentProcess (VOID)

DWORD GetCurrentProcessId(VOID)

GetCurrentProcess() is used to create a psuedohandle. It’s not inheritable and desired access attributes
are undefined. The actual handle can be obtained using Process Id. The OpenProcess() function is used
to obtain the process handle using the process ID obtained by GetCurrentProcessId() or otherwise.

HANDLE OpenProcess( DWORD dwDesiredAccess,

dwDesiredAccess provides the desired access attributes for the handle. Some of the commonly required
values can be

SYCHRONIZE : Allows process synchronization using wait()

PROCESS_ALL_ACCESS : All access flags are set.

PROCESS_TERMINATE: It’s possible to terminate the process with TerminateProcess

PROCESS_QUERY_INFORMATION: The handle can be used to query process information

bInheritHandle is used to specify whether the new process handle can be inherited.

The full pathname of the executable file used to run the process can be determined from

GetModuleFileName()

GetModuleFileNameEx()

Topic 94: Duplicating Handles

Parent and Child processes may require different access to an inheritable handle of resources. Parent
Process may also require real handle rather than a pseudohandle. Parent process usually creates a
duplicate handle to take care of this problem. Duplicate handles are created using

API DuplicateHandle()

BOOL DuplicateHandle( HANDLE hSourceProcessHandle,

HANDLE hSourceHandle,

HANDLE hTargetProcessHandle,

LPHANDLE lpTargetHandle,

DWORD dwDesiredAccess,

BOOL bInheritHandle,

DWORD dwOptions );

hSourceProcessHandle is the source process handle. lpTargetHandle receives a pointer to a copy of the
original handle indicated by hSourceHandle. The parent process duplicating the handle must have
PROCESS_DUP_HANDLE access. If the hSourceHandle does not exist in the source process then the
function will fail. The target handle received will only be valid in the process whose handle is specified in
hTargetProcessHandle.

Usually the handle is duplicated by the source or target process. In case it is duplicated by some other
process than a mechanism is required to communicate with the target process DuplicateHandle() can be
used to duplicate any sort of handle. dwOption can be combination of two flags.

DUPLICATE_CLOSE_SOURCE : causes the source handle to be closed

DUPLICATE_SAME_ACCESS: Uses the access rights of the source handle.

dwDesiredAccess is ignored, Reference Count. Windows keep track of all the handles created for a
resource. This count is not accessible to the application. A resource cannot be released unless the last
reference to the resource has been closed. Inherited and duplicated handles reflect on reference count.

Topic 95: Exiting and Terminating a Process

A process can exit using the API ExitProcess(). The function does not return, in fact the process and all its
associated threads terminate. Termination Handlers are ignored. However, a detach call is sent to
DllMain().

VOID ExitProcess(UINT uExitCode) ;

Exit Code

The exit code of the process can be determined using the API GetExitCodeProcess()

BOOL GetExitCodeProcess( HANDLE hProcess, LPDWORD lpExitCode );

The process identified by hProcess must have PROCESS_QUERY_INFORMATION flag set.

lpExitCode points to the dword that will receive the exit code of the given process.

One possible value of Exit code can be STILL_ACTIVE

Terminate Process

One process can terminate another process using the function TerminateProcess()

BOOL TerminateProcess( HANDLE hProcess, UINT uExitCode );

hProcess is the handle to the process to be terminated

uExitCode is the code returned by the process on termination.

Exiting and Terminating Process

Process must be exited and terminated carefully. All the handles need to be closed before exiting the
process. It’s not a good idea to use ExitProcess() within the program as it will not give it chance to
release resources. Use is with Exception handling. TerminateProcess() does not all the process to
execute SHE or DllMain().

Topic No.96
The synchronization can be attained easily by the wait process. Windows provides a general-
purpose wait function. Which can be used to wait for a single object and also multiple objects in
a group. Windows send a signal to the waiting process when the process terminates

We have two API's

i. DWORD waitForSingleObject(HANDLE hHandle, DWORD dwMilliseconds);

ii. DWORD waitForMultipleObjects(DWORD nCount, const HANDLE*lpHandles, BOOL
bWaitAll,DWORD dwMilliseconds);

 hHandle is the single object for which the process waits.

 lpHandle is the array of handles for which the process waits.
 nCount is the number of objects in an array. Should not exceed MAXIMUM WAIT
OBJECTS
 dwMilliseconds is the timeout period for wait. 0 for no wait and INFINITE for indefinite
wait.
 bWaitAll describes if it's necessary to wait for all the objects to get free.

The possible return values are:

 WAIT_OBJECT_0
 WAIT_OBJECT_0+n
 WAIT_TIMEDOUT
 WAIT_FAILED
 WAIT_ABANDONED_0

Topic No.97

 Environment Block string

 An Environment Block is associated with each process.
 The EB contains string regarding the environment of the process of the form Name = Value
 Each string is NULL-terminated
 PATH is an example of an environment string.
 Environment variables can get and set using these API's

DWORD GetEnvironmentVatiable (LPCTSTR lpNAME,LpTSTR lpBuffer, DWORD nSize );

And

BOOL SetEnviromentVariable(LPCTSTR ipName, LPCTSTR lpValue);

To share the parent environment with the child process set lpEnvironment to NULL in the call to
CreateProcess()

Any process can modify the environment variables and make new ones.
lpName is the variable name. On setting the string the value is modified if the variable already exists. If it
does not exist then a new variable is created and assigned the value. IF the value is NULL then the
variable is deleted.

"=" is reserved and cannot be used in the variable name.

GetEnvironmentVariable() return the length of variable string or 0 in case of failure.

If LpValue is not as long as the value specified by the count then the actual length of the string is
returned.

The access rights given in Create Process () are usually Process_All_Access. But there are several options:

 PROCESS_QUERY_INFORMATION
 CREATE_PROCESS
 PROCESS_TERMINATE
 PROCESS_SET_INFORMATION
 DUPLICATE_HANDLE
 CREATE_HANDLE

it can be useful to limit some processes right like giving PROCESS_TERMINATE lpName is the
variable name. On setting the string the value is modified if the variable already exists. If it does not exist
then a new variable is created and assigned the value. IF the value is NULL then the variable is deleted.

"=" is reserved and cannot be used in the variable name.

GetEnvironmentVariable() return the length of variable string or 0 in case of failure.

If LpValue is not as long as the value specified by the count then the actual length of the string is
returned.

The access rights given in Create Process () are usually Process_All_Access. But there are several options:

 PROCESS_QUERY_INFORMATION
 CREATE_PROCESS
 PROCESS_TERMINATE
 PROCESS_SET_INFORMATION
 DUPLICATE_HANDLE
 CREATE_HANDLE

it can be useful to limit some processes right like giving PROCESS_TERMINATE right to parent process
only.

Topic No.98

A pattern searching example:

This example uses the power of windows multitasking to search a specific pattern among numerous
files.

The process takes the specific pattern along with filenames through command line. The standard output
file is specified as inheritable in new process start-up info structure. Wait functions are used for
synchronization. As soon as the search end the results are displayed one at a time. Wait functions are
limited for 64 handles so program works accordingly. The program uses the exit code to identify
whether the process has detected a match or not.

The Example

#include "Everything.h"

int _tmain (int argc, LPTSTR argv[])

/* Create a separate process to search each file on the

command line. Each process is given a temporary file,

in the current directory, to receive the results. */

HANDLE hTempFile;

SECURITY_ATTRIBUTES stdOutSA = /* SA for inheritable handle. */

{sizeof (SECURITY_ATTRIBUTES), NULL, TRUE};

TCHAR commandLine[MAX_PATH + 100];

STARTUPINFO startUpSearch, startUp;

PROCESS_INFORMATION processInfo;

DWORD exitCode, dwCreationFlags = 0;

int iProc;

HANDLE *hProc; /* Pointer to an array of proc handles. */

typedef struct {TCHAR tempFile[MAX_PATH];} PROCFILE;

PROCFILE *procFile; /* Pointer to array of temp file names. */

#ifdef UNICODE

dwCreationFlags = CREATE_UNICODE_ENVIRONMENT;

#endif

if (argc < 3)

ReportError (_T ("Usage: grepMP pattern files."), 1, FALSE);

/* Startup info for each child search process as well as

the child process that will display the results. */

GetStartupInfo (&startUpSearch);

GetStartupInfo (&startUp);

/* Allocate storage for an array of process data structures,

each containing a process handle and a temporary file name. */

procFile = malloc ((argc - 2) * sizeof (PROCFILE));

hProc = malloc ((argc - 2) * sizeof (HANDLE));

/* Create a separate "grep" process for each file on the

command line. Each process also gets a temporary file

name for the results; the handle is communicated through

the STARTUPINFO structure. argv[1] is the search pattern. */

for (iProc = 0; iProc < argc - 2; iProc++) {

/* Create a command line of the form: grep argv[1] argv[iProc + 2] */

/* Allow spaces in the file names. */

_stprintf (commandLine, _T ("grep \"%s\" \"%s\""), argv[1], argv[iProc + 2]);

/* Create the temp file name for std output. */

if (GetTempFileName (_T ("."), _T ("gtm"), 0, procFile[iProc].tempFile) == 0)

ReportError (_T ("Temp file failure."), 2, TRUE);

 /* Set the std output for the search process. */

hTempFile = /* This handle is inheritable */

CreateFile (procFile[iProc].tempFile,

/** GENERIC_READ | Read access not required **/ GENERIC_WRITE,

FILE_SHARE_READ | FILE_SHARE_WRITE, &stdOutSA,

CREATE_ALWAYS, FILE_ATTRIBUTE_NORMAL, NULL);

if (hTempFile == INVALID_HANDLE_VALUE)

ReportError (_T ("Failure opening temp file."), 3, TRUE);

/* Specify that the new process takes its std output

from the temporary file's handles. */

startUpSearch.dwFlags = STARTF_USESTDHANDLES;

startUpSearch.hStdOutput = hTempFile;

startUpSearch.hStdError = hTempFile;

startUpSearch.hStdInput = GetStdHandle (STD_INPUT_HANDLE);

/* Create a process to execute the command line. */

if (!CreateProcess (NULL, commandLine, NULL, NULL,

TRUE, dwCreationFlags, NULL, NULL, &startUpSearch, &processInfo))

ReportError (_T ("ProcCreate failed."), 4, TRUE);

/* Close unwanted handles */

CloseHandle (hTempFile); CloseHandle (processInfo.hThread);

/* Save the process handle. */

hProc[iProc] = processInfo.hProcess;

/* Processes are all running. Wait for them to complete, then output
 the results - in the order of the command line file names. */

for (iProc = 0; iProc < argc-2; iProc += MAXIMUM_WAIT_OBJECTS)

WaitForMultipleObjects (min(MAXIMUM_WAIT_OBJECTS, argc - 2 - iProc),

&hProc[iProc], TRUE, INFINITE);

/* Result files sent to std output using "cat".

Delete each temporary file upon completion. */

for (iProc = 0; iProc < argc - 2; iProc++) {

if (GetExitCodeProcess (hProc[iProc], &exitCode) && exitCode == 0) {

/* Pattern was detected - List results. */

/* List the file name if there is more than one file to search */

if (argc > 3) _tprintf (_T("%s:\n"), argv[iProc+2]);

_stprintf (commandLine, _T ("cat \"%s\""), procFile[iProc].tempFile);

if (!CreateProcess (NULL, commandLine, NULL, NULL,

TRUE, dwCreationFlags, NULL, NULL, &startUp,

&processInfo))

ReportError (_T ("Failure executing cat."), 0, TRUE);

else {

WaitForSingleObject (processInfo.hProcess, INFINITE);

CloseHandle (processInfo.hProcess);

CloseHandle (processInfo.hThread);

CloseHandle (hProc[iProc]);

if (!DeleteFile (procFile[iProc].tempFile))
 ReportError (_T ("Cannot delete temp file."), 6, TRUE);

free (procFile); free (hProc);

return 0;

Topic No.99

Multiprocessor Environment

All the processes run concurrently in windows environment. If the system is uniprocessor the processor
time is multiplexed among multiple processes in an interleaved manner. If the system is multiprocessor,
then windows scheduler can run process threads on separate processors.

There would be substantial performance gain in this case. The performance gain will not be linear
because of dependencies among processes (wait and signal).

From programming point of view, it is essential to understand this potential of windows so that
programs can be designed optimally. Alternately, constraining the processing to a specific processor is
also possible. This is accomplished by processor affinity mask.

Subsequently, it is possible to create independent threads within a process which can be scheduled on
separate processor.

Topic No.100

Process Times

Windows API provides a very simple mechanism for determining the amount of time a process has
consumed.

The function used for this purpose is GetProcessTime()

BOOL GetProcessTimes( HANDLE hProcess,

LPFILETIME lpCreationTime,

LPFILETIME lpExitTime,

LPFILETIME lpKernelTime,

LPFILETIME lpUserTime );
Process handle can be of the current process or a terminated one.

Elapsed time can be computed by subtracting creation time from exit time.

FILETIME is a 64-bit value

GetThreadTime() can be used similarly and requires a hread handle.

Topic No.101

Example of Process Execution Times

This example prints the elapsed, real, and user times. It uses the API GetCommandLine()to get the
command line as a single string. It then uses the SkipArg() function the skip past the executable name.

#include "Everything.h"

int _tmain (int argc, LPTSTR argv[])

STARTUPINFO startUp;

PROCESS_INFORMATION procInfo;

union { /* Structure required for file time arithmetic. */

LONGLONG li;

FILETIME ft;

} createTime, exitTime, elapsedTime;

FILETIME kernelTime, userTime;

SYSTEMTIME elTiSys, keTiSys, usTiSys;

LPTSTR targv, cLine = GetCommandLine ();

OSVERSIONINFO windowsVersion;

HANDLE hProc;
 targv = SkipArg(cLine, 1, argc, argv);

/* Skip past the first blank-space delimited token on the command line */

if (argc <= 1 || NULL == targv)

ReportError (_T("Usage: timep command ..."), 1, FALSE);

/* Determine is this is Windows 2000 or NT. */

windowsVersion.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);

if (!GetVersionEx (&windowsVersion))

ReportError (_T("Can not get OS Version info. %d"), 2, TRUE);

if (windowsVersion.dwPlatformId != VER_PLATFORM_WIN32_NT)

ReportError (_T("This program only runs on Windows NT kernels"), 2, FALSE);

/* Determine is this is Windows 2000 or NT. */

windowsVersion.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);

if (!GetVersionEx (&windowsVersion))

ReportError (_T("Can not get OS Version info. %d"), 2, TRUE);

if (windowsVersion.dwPlatformId != VER_PLATFORM_WIN32_NT)

ReportError (_T("This program only runs on Windows NT kernels"), 2, FALSE);

GetStartupInfo (&startUp);

/* Execute the command line and wait for the process to complete. */

if (!CreateProcess (NULL, targv, NULL, NULL, TRUE,

 NORMAL_PRIORITY_CLASS, NULL, NULL, &startUp, &procInfo))

ReportError (_T ("\nError starting process. %d"), 3, TRUE);

hProc = procInfo.hProcess;

if (WaitForSingleObject (hProc, INFINITE) != WAIT_OBJECT_0)

ReportError (_T("Failed waiting for process termination. %d"), 5, TRUE);;

if (!GetProcessTimes (hProc, &createTime.ft,

&exitTime.ft, &kernelTime, &userTime))

ReportError (_T("Can not get process times. %d"), 6, TRUE);

elapsedTime.li = exitTime.li - createTime.li;

FileTimeToSystemTime (&elapsedTime.ft, &elTiSys);

FileTimeToSystemTime (&kernelTime, &keTiSys);

FileTimeToSystemTime (&userTime, &usTiSys);

_tprintf (_T ("Real Time: %02d:%02d:%02d.%03d\n"),

elTiSys.wHour, elTiSys.wMinute, elTiSys.wSecond,

elTiSys.wMilliseconds);

_tprintf (_T ("User Time: %02d:%02d:%02d.%03d\n"),

usTiSys.wHour, usTiSys.wMinute, usTiSys.wSecond,

usTiSys.wMilliseconds);

_tprintf (_T ("Sys Time: %02d:%02d:%02d.%03d\n"),

keTiSys.wHour, keTiSys.wMinute, keTiSys.wSecond,

 keTiSys.wMilliseconds);

CloseHandle (procInfo.hThread); CloseHandle (procInfo.hProcess);

return 0;

Topic No.102

Example of Process Execution Times

This example prints the elapsed, real and user times. It uses the API GetCommandLine()to get the
command line as a single string. It than uses the SkipArg() function the skip past the executable name.

#include "Everything.h"

int _tmain (int argc, LPTSTR argv[])

STARTUPINFO startUp;

PROCESS_INFORMATION procInfo;

union { /* Structure required for file time arithmetic. */

LONGLONG li;

FILETIME ft;

} createTime, exitTime, elapsedTime;

FILETIME kernelTime, userTime;

SYSTEMTIME elTiSys, keTiSys, usTiSys;

LPTSTR targv, cLine = GetCommandLine ();

 OSVERSIONINFO windowsVersion;

HANDLE hProc;

targv = SkipArg(cLine, 1, argc, argv);

/* Skip past the first blank-space delimited token on the command line */

if (argc <= 1 || NULL == targv)

ReportError (_T("Usage: timep command ..."), 1, FALSE);

/* Determine is this is Windows 2000 or NT. */

windowsVersion.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);

if (!GetVersionEx (&windowsVersion))

ReportError (_T("Can not get OS Version info. %d"), 2, TRUE);

if (windowsVersion.dwPlatformId != VER_PLATFORM_WIN32_NT)

ReportError (_T("This program only runs on Windows NT kernels"), 2, FALSE);

/* Determine is this is Windows 2000 or NT. */

windowsVersion.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);

if (!GetVersionEx (&windowsVersion))

ReportError (_T("Can not get OS Version info. %d"), 2, TRUE);

if (windowsVersion.dwPlatformId != VER_PLATFORM_WIN32_NT)

ReportError (_T("This program only runs on Windows NT kernels"), 2, FALSE);

GetStartupInfo (&startUp);
 /* Execute the command line and wait for the process to complete. */

if (!CreateProcess (NULL, targv, NULL, NULL, TRUE,

NORMAL_PRIORITY_CLASS, NULL, NULL, &startUp, &procInfo))

ReportError (_T ("\nError starting process. %d"), 3, TRUE);

hProc = procInfo.hProcess;

if (WaitForSingleObject (hProc, INFINITE) != WAIT_OBJECT_0)

ReportError (_T("Failed waiting for process termination. %d"), 5, TRUE);;

if (!GetProcessTimes (hProc, &createTime.ft,

&exitTime.ft, &kernelTime, &userTime))

ReportError (_T("Can not get process times. %d"), 6, TRUE);

elapsedTime.li = exitTime.li - createTime.li;

FileTimeToSystemTime (&elapsedTime.ft, &elTiSys);

FileTimeToSystemTime (&kernelTime, &keTiSys);

FileTimeToSystemTime (&userTime, &usTiSys);

_tprintf (_T ("Real Time: %02d:%02d:%02d.%03d\n"),

elTiSys.wHour, elTiSys.wMinute, elTiSys.wSecond,

elTiSys.wMilliseconds);

_tprintf (_T ("User Time: %02d:%02d:%02d.%03d\n"),

usTiSys.wHour, usTiSys.wMinute, usTiSys.wSecond,

usTiSys.wMilliseconds);
 _tprintf (_T ("Sys Time: %02d:%02d:%02d.%03d\n"),

keTiSys.wHour, keTiSys.wMinute, keTiSys.wSecond,

keTiSys.wMilliseconds);

CloseHandle (procInfo.hThread); CloseHandle (procInfo.hProcess);

return 0;

Topic No.103

Generating Console Events

Terminating another process can be problematic as the terminating process does not get a chance to
release resources. SEH does not help either as there is no mechanism that can be used to raise an
exception in another process.

Console control event allows sending a message from one process to another. Usually, a handler is set
up in a process to catch such signals. Subsequently, the handler generates an exception. It is hence
possible for a process to generate an event in another process.

CreateProcess() can be used with the flag CREATE_NEW_PROCESS_GROUP.

This creates a process group in which the created process is the root and all the subsequently created
processes by the parent are in the new group.

One process in a group can generate a CTRL_C_EVENT or CTRL_BREAK_EVENT in a group identified by

the root process ID.

The target process should have the same console as the process generating the event.

More specifically, the calling process cannot have its own console using CREATE_NEW_CONSOLE or
DETACHED_PROCESS flags.

BOOL GenerateConsoleCtrlEvent( DWORD dwCtrlEvent,

DWORD dwProcessGroupId );

dwCtrlEvent can be CTRL_C_EVENT or CTRL_BREAK_EVENT

dwProcessGroupId is the id of the process group

Topic No.104

Simple Job Management Shell

This job shell will allow three commands to run

 jobbg
 jobs
 kill

The job shell parses the command line and then calls the respective function for the given command.

The shell uses a user-specific file keeping track of process ID and other related information

Several shells can run concurrently and use this shared file.

Also, concurrency issues can be encountered as several shells can try to use the same file.

File locking is used to make the file access mutually exclusive.

/* JobObjectShell.c One program combining three

job management commands:

Jobbg - Run a job in the background

jobs - List all background jobs

kill - Terminate a specified job of the job family

There is an option to generate a console

control signal.

This implementation enhances JobShell with a time limit on each process.

There is a time limit on each process, in seconds, in argv[1] (if present)

0 or omitted means no process time limit

*/

#include "Everything.h"
#include "JobManagement.h"

#define MILLION 1000000

static int Jobbg (int, LPTSTR *, LPTSTR);

static int Jobs (int, LPTSTR *, LPTSTR);

static int Kill (int, LPTSTR *, LPTSTR);

HANDLE hJobObject = NULL;

JOBOBJECT_BASIC_LIMIT_INFORMATION basicLimits = {0, 0, JOB_OBJECT_LIMIT_PROCESS_TIME};

int _tmain (int argc, LPTSTR argv[])

LARGE_INTEGER processTimeLimit;

BOOL exitFlag = FALSE;

TCHAR command[MAX_COMMAND_LINE], *pc;

DWORD i, localArgc;

TCHAR argstr[MAX_ARG][MAX_COMMAND_LINE];

LPTSTR pArgs[MAX_ARG];

/* NT Only - due to file locking */

if (!WindowsVersionOK (3, 1))

ReportError (_T("This program requires Windows NT 3.1 or greater"), 1, FALSE);

hJobObject = NULL;

processTimeLimit.QuadPart = 0;
 if (argc >= 2) processTimeLimit.QuadPart = _ttoi(argv[1]);

basicLimits.PerProcessUserTimeLimit.QuadPart = processTimeLimit.QuadPart *

MILLION;

hJobObject = CreateJobObject(NULL, NULL);

if (NULL == hJobObject)

ReportError(_T("Error creating job object."), 1, TRUE);

if (!SetInformationJobObject(hJobObject, JobObjectBasicLimitInformation, &basicLimits,

sizeof(JOBOBJECT_BASIC_LIMIT_INFORMATION)))

ReportError(_T("Error setting job object information."), 2, TRUE);

for (i = 0; i < MAX_ARG; i++)

pArgs[i] = argstr[i];

_tprintf (_T("Windows Job Mangement with Windows Job Object.\n"));

while (!exitFlag) {

_tprintf (_T("%s"), _T("JM$"));

_fgetts (command, MAX_COMMAND_LINE, stdin);

pc = _tcschr (command, _T('\n'));

*pc = _T('\0');

GetArgs (command, &localArgc, pArgs);

CharLower (argstr[0]);
 if (_tcscmp (argstr[0], _T("jobbg")) == 0) {

Jobbg (localArgc, pArgs, command);

else if (_tcscmp (argstr[0], _T("jobs")) == 0) {

Jobs (localArgc, pArgs, command);

else if (_tcscmp (argstr[0], _T("kill")) == 0) {

Kill (localArgc, pArgs, command);

else if (_tcscmp (argstr[0], _T("quit")) == 0) {

exitFlag = TRUE;

else _tprintf (_T("Illegal command. Try again.\n"));

CloseHandle (hJobObject);

return 0;

Topic No.105

Job Listing

The job management function used for the purpose is DisplayJobs()

The function opens the file. Looks up into the file and acquires the status of the processes listed in the
file. Also displays the status of the processes listed and other information.

BOOL DisplayJobs (void)

/* Scan the job database file, reporting on the status of all jobs.

In the process remove all jobs that no longer exist in the system. */

HANDLE hJobData, hProcess;

JM_JOB jobRecord;

DWORD jobNumber = 0, nXfer, exitCode;

TCHAR jobMgtFileName[MAX_PATH];

OVERLAPPED regionStart;

if ( !GetJobMgtFileName (jobMgtFileName) )

return FALSE;

hJobData = CreateFile (jobMgtFileName, GENERIC_READ | GENERIC_WRITE,

FILE_SHARE_READ | FILE_SHARE_WRITE,

NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL);

if (hJobData == INVALID_HANDLE_VALUE)

return FALSE;

/* Read records and report on each job. */

/* First, create an exclusive lock on the entire

 file as entries will be changed. */

regionStart.Offset = 0;

regionStart.OffsetHigh = 0;

regionStart.hEvent = (HANDLE)0;

LockFileEx (hJobData, LOCKFILE_EXCLUSIVE_LOCK, 0, 0, 0, ®ionStart);

__try {

while (ReadFile (hJobData, &jobRecord, SJM_JOB, &nXfer, NULL) && (nXfer > 0)) {

jobNumber++; /* jobNumber won't exceed MAX_JOBS_ALLOWED as the file is

not allowed to grow that long */

hProcess = NULL;

if (jobRecord.ProcessId == 0) continue;

/* Obtain process status. */

hProcess = OpenProcess (PROCESS_ALL_ACCESS, FALSE, jobRecord.ProcessId);

if (hProcess != NULL) {

GetExitCodeProcess (hProcess, &exitCode);

CloseHandle (hProcess);

/* Show the job status. */

_tprintf (_T (" [%d] "), jobNumber);

if (NULL == hProcess)

_tprintf (_T (" Done"));

else if (exitCode != STILL_ACTIVE)

_tprintf (_T ("+ Done"));

else _tprintf (_T (" "));

/* Show the command. */

_tprintf (_T (" %s\n"), jobRecord.CommandLine);

/* Remove processes that are no longer in the system. */

if (NULL == hProcess) { /* Back up one record. */

/* SetFilePoiner is more convenient here than SetFilePointerEx since the

move is a short shift from the current position */

SetFilePointer (hJobData, -(LONG)nXfer, NULL, FILE_CURRENT);

/* Set the job number to 0. */

jobRecord.ProcessId = 0;

if (!WriteFile (hJobData, &jobRecord, SJM_JOB, &nXfer, NULL))

ReportError (_T ("Rewrite error."), 1, TRUE);

} /* End of while. */

} /* End of try. */

__finally { /* Release the lock on the file. */

UnlockFileEx (hJobData, 0, 0, 0, ®ionStart);

if (NULL != hProcess) CloseHandle (hProcess);

}
 CloseHandle (hJobData);

return TRUE;

Topic 106

Finding a Process ID:

The job management function used for the purpose is FindProcessId()

It obtains the process id of the given job number

It simply looks up into the file based on job number and reads the record at the specific location.

DWORD FindProcessId (DWORD jobNumber)

/* Obtain the process ID of the specified job number. */

HANDLE hJobData;

JM_JOB jobRecord;

DWORD nXfer, fileSizeLow;

TCHAR jobMgtFileName[MAX_PATH+1];

OVERLAPPED regionStart;

LARGE_INTEGER fileSize;

if ( !GetJobMgtFileName (jobMgtFileName) ) return 0;

hJobData = CreateFile (jobMgtFileName, GENERIC_READ,

FILE_SHARE_READ | FILE_SHARE_WRITE,

NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL);

if (hJobData == INVALID_HANDLE_VALUE) return 0;

 /* Position to the correct record, but not past the end of file */

/* As a variation, use GetFileSize to demonstrate its operation. */

if (!GetFileSizeEx (hJobData, &fileSize) ||

(fileSize.HighPart != 0 || SJM_JOB * (jobNumber - 1) > fileSize.LowPart

|| fileSize.LowPart > SJM_JOB * MAX_JOBS_ALLOWED))

return 0;

fileSizeLow = fileSize.LowPart;

/* SetFilePoiner is more convenient here than SetFilePointerEx since the the file is known to be
"short" ( < 4 GB). */

SetFilePointer (hJobData, SJM_JOB * (jobNumber - 1), NULL, FILE_BEGIN);

/* Get a shared lock on the record. */

regionStart.Offset = SJM_JOB * (jobNumber - 1);

regionStart.OffsetHigh = 0; /* Assume a "short" file. */

regionStart.hEvent = (HANDLE)0;

LockFileEx (hJobData, 0, 0, SJM_JOB, 0, &regionStart);

if (!ReadFile (hJobData, &jobRecord, SJM_JOB, &nXfer, NULL))

ReportError (_T ("JobData file error"), 0, TRUE);

UnlockFileEx (hJobData, 0, SJM_JOB, 0, &regionStart);

CloseHandle (hJobData);

return jobRecord.ProcessId;

}
Topic 107

Job Objects:

Windows provides the provision of grouping a number of processes into a job object.

Resource limits can be specified for each job object. This also helps maintain accounting
information.

Creating Job Objects

Firstly a job object is created using CreateJobObject(). It uses a name and security attributes.

Also, OpenJobObject() is used with named object and CloseHandle() is used to close the object.

Once the job object is created it can be assigned processes using AssignProcesstoJobObject()

Creating Job Objects

Once a process is assigned to a job object it cannot be assigned to another job object.

Child processes are automatically assigned the same job object as the parent unless the
CREATE_BREAKAWAY_FROM_JOB flag is used at creation.

Control limits are specified to the processes in job object using:

BOOL SetInformationJobObject( HANDLE hJob,

JOBOBJECTINFOCLASS JobObjectInformationClass,

LPVOID lpJobObjectInformation, DWORD cbJobObjectInformationLength );

hJob is handled to the existing job object

JobObjectInformationClass is the class of the limits you wish to set regarding per process time,
working set limit, the limit on active processes, priority, processor affinity, etc.

lpJobObjectInformation is the structure congaing the actual information as it uses a different

structure for each class

Topic 108
Using Job Objects

In this example job objects are used to limit process execution time and obtain user time statistics.

The simple JobShell program discussed previously is modified.

A command line time limit argument is added.

/* JobObjectShell.c One program combining three

job management commands:

Jobbg - Run a job in the background

jobs - List all background jobs

kill - Terminate a specified job of job family

There is an option to generate a console

control signal.

This implemenation enhances JobShell with a time limit on each process.

There is a time limit on each process, in seconds, in argv[1] (if present)

0 or omitted means no process time limit

*/

#include "Everything.h"

#include "JobManagement.h"

#define MILLION 1000000

static int Jobbg (int, LPTSTR *, LPTSTR);

static int Jobs (int, LPTSTR *, LPTSTR);

static int Kill (int, LPTSTR *, LPTSTR);

HANDLE hJobObject = NULL;

JOBOBJECT_BASIC_LIMIT_INFORMATION basicLimits = {0, 0, JOB_OBJECT_LIMIT_PROCESS_TIME};

int _tmain (int argc, LPTSTR argv[])

LARGE_INTEGER processTimeLimit;

BOOL exitFlag = FALSE;

TCHAR command[MAX_COMMAND_LINE], *pc;

DWORD i, localArgc;

TCHAR argstr[MAX_ARG][MAX_COMMAND_LINE];

LPTSTR pArgs[MAX_ARG];

/* NT Only - due to file locking */

if (!WindowsVersionOK (3, 1))

ReportError (_T("This program requires Windows NT 3.1 or greater"), 1,

FALSE);

hJobObject = NULL;

processTimeLimit.QuadPart = 0;

if (argc >= 2) processTimeLimit.QuadPart = _ttoi(argv[1]);

basicLimits.PerProcessUserTimeLimit.QuadPart = processTimeLimit.QuadPart * MILLION;

hJobObject = CreateJobObject(NULL, NULL);

if (NULL == hJobObject)

ReportError(_T("Error creating job object."), 1, TRUE);

if (!SetInformationJobObject(hJobObject, JobObjectBasicLimitInformation, &basicLimits,
sizeof(JOBOBJECT_BASIC_LIMIT_INFORMATION)))

ReportError(_T("Error setting job object information."), 2, TRUE);

for (i = 0; i < MAX_ARG; i++)

pArgs[i] = argstr[i];

_tprintf (_T("Windows Job Mangement with Windows Job Object.\n"));

while (!exitFlag) {

_tprintf (_T("%s"), _T("JM$"));

_fgetts (command, MAX_COMMAND_LINE, stdin);

pc = _tcschr (command, _T('\n'));

*pc = _T('\0');

GetArgs (command, &localArgc, pArgs);

CharLower (argstr[0]);

if (_tcscmp (argstr[0], _T("jobbg")) == 0) {

Jobbg (localArgc, pArgs, command);

else if (_tcscmp (argstr[0], _T("jobs")) == 0) {

Jobs (localArgc, pArgs, command);

else if (_tcscmp (argstr[0], _T("kill")) == 0) {

Kill (localArgc, pArgs, command);

else if (_tcscmp (argstr[0], _T("quit")) == 0) {

exitFlag = TRUE;

else _tprintf (_T("Illegal command. Try again.\n"));

CloseHandle (hJobObject);

return 0;

/* Jobbg: Execute a command line in the background, put

the job identity in the user's job file, and exit.

Related commands (jobs, fg, kill, and suspend) can be used to manage the jobs. */

/* jobbg [options] command-line

-c: Give the new process a console.

-d: The new process is detached, with no console.

These two options are mutually exclusive.

If neither is set, the background process shares the console with jobbg. */

/* These new features this program illustrates:

1. Creating detached and with separate consoles.

2. Maintaining a Jobs/Process list in a shared file.

3. Determining a process status from a process ID.*/

/* Standard includes files. */

/* - - - - - - - - - - - - - - - - - - - - - - - - - */

int Jobbg (int argc, LPTSTR argv[], LPTSTR command)

/* Similar to timep.c to process command line. */

/* - - - - - - - - - - - - - - - - - - - - - - */

/* Execute the command line (targv) and store the job id,

the process id, and the handle in the jobs file. */

DWORD fCreate;

LONG jobNumber;

BOOL flags[2];

STARTUPINFO startUp;

PROCESS_INFORMATION processInfo;

LPTSTR targv = SkipArg (command, 1, argc, argv);

GetStartupInfo (&startUp);

/* Determine the options. */

Options (argc, argv, _T ("cd"), &flags[0], &flags[1], NULL);

/* Skip over the option field as well, if it exists. */

/* Simplifying assumptions: There's only one of -d, -c (they are mutually exclusive.

Also, commands can't start with -. etc. You may want to fix this. */
if (argv[1][0] == _T('-'))

targv = SkipArg (command, 2, argc, argv);

fCreate = flags[0] ? CREATE_NEW_CONSOLE : flags[1] ? DETACHED_PROCESS : 0;

/* Create the job/thread suspended.

Resume it once the job is entered properly. */

if (!CreateProcess (NULL, targv, NULL, NULL, TRUE,

fCreate | CREATE_SUSPENDED | CREATE_NEW_PROCESS_GROUP,

NULL, NULL, &startUp, &processInfo)) {

ReportError (_T ("Error starting process."), 0, TRUE);

if (hJobObject != NULL)

if (!AssignProcessToJobObject(hJobObject, processInfo.hProcess)) {

ReportError(_T("Could not add process to job object. The process will be terminated."), 0, TRUE);

TerminateProcess (processInfo.hProcess, 4);

CloseHandle (processInfo.hThread);

CloseHandle (processInfo.hProcess);

return 4;

/* Create a job number and enter the process Id and handle

into the Job "database" maintained by the

GetJobNumber function (part of the job management library). */

jobNumber = GetJobNumber (&processInfo, targv);

if (jobNumber >= 0)

ResumeThread (processInfo.hThread);

else {

TerminateProcess (processInfo.hProcess, 3);

CloseHandle (processInfo.hThread);

CloseHandle (processInfo.hProcess);

ReportError (_T ("Error: No room in job control list."), 0, FALSE);

return 5;

CloseHandle (processInfo.hThread);

CloseHandle (processInfo.hProcess);

_tprintf (_T (" [%d] %d\n"), jobNumber, processInfo.dwProcessId);

return 0;

/* Jobs: List all running or stopped jobs that have

been created by this user under job management;

that is, have been started with the jobbg command.

Related commands (jobbg and kill) can be used to manage the jobs. */

/* These new features this program illustrates:

1. Determining process status.

2. Maintaining a Jobs/Process list in a shared file.

3. Obtaining job object information

*/

int Jobs (int argc, LPTSTR argv[], LPTSTR command)

JOBOBJECT_BASIC_ACCOUNTING_INFORMATION basicInfo;

if (!DisplayJobs ()) return 1;

/* Dispaly the job information */

if (!QueryInformationJobObject(hJobObject, JobObjectBasicAccountingInformation, &basicInfo,

sizeof(JOBOBJECT_BASIC_ACCOUNTING_INFORMATION), NULL)) {

ReportError(_T("Failed QueryInformationJobObject"), 0, TRUE);

return 0;

_tprintf (_T("Total Processes: %d, Active: %d, Terminated: %d.\n"),

basicInfo.TotalProcesses, basicInfo.ActiveProcesses, basicInfo.TotalTerminatedProcesses);

_tprintf (_T("User time all processes: %d.%03d\n"),

basicInfo.TotalUserTime.QuadPart / MILLION, (basicInfo.TotalUserTime.QuadPart % MILLION) / 10000);

return 0;

/* kill [options] jobNumber

Terminate the process associated with the specified job number. */

/* This new features this program illustrates:

1. Using TerminateProcess

2. Console control events */

/* Options:
-b Generate a Ctrl-Break

-c Generate a Ctrl-C

Otherwise, terminate the process. */

/* The Job Management data structures, error codes,

constants, and so on are in the following file. */

int Kill (int argc, LPTSTR argv[], LPTSTR command)

DWORD processId, jobNumber, iJobNo;

HANDLE hProcess;

BOOL cntrlC, cntrlB, killed;

iJobNo = Options (argc, argv, _T ("bc"), &cntrlB, &cntrlC, NULL);

/* Find the process ID associated with this job. */

jobNumber = _ttoi (argv[iJobNo]);

processId = FindProcessId (jobNumber);

if (processId == 0) {

ReportError (_T ("Job number not found.\n"), 0, FALSE);

return 1;

hProcess = OpenProcess (PROCESS_TERMINATE, FALSE, processId);

if (hProcess == NULL) {
ReportError (_T ("Process already terminated.\n"), 0, FALSE);

return 2;

if (cntrlB)

killed = GenerateConsoleCtrlEvent (CTRL_BREAK_EVENT, 0);

else if (cntrlC)

killed = GenerateConsoleCtrlEvent (CTRL_C_EVENT, 0);

else

killed = TerminateProcess (hProcess, JM_EXIT_CODE);

if (!killed) {

ReportError (_T ("Process termination failed."), 0, TRUE);

return 3;

WaitForSingleObject (hProcess, 5000);

CloseHandle (hProcess);

_tprintf (_T ("Job [%d] terminated or timed out\n"), jobNumber);

return 0;

Topic 109
Threads overview

Thread is an independent unit of execution within a process.

In a multi-threaded system, multiple threads may exist within a process.

Organizing and coordinating these threads is a challenge.

Some programming problems are greatly simplified by the use of threads.

The conventional scenario uses single-threaded processes being run concurrently

This scenario has varying disadvantages

Switching among processes is time-consuming and expensive. Threads can easily allow
concurrent processing of the same function hence reducing overheads

Usually, independent processes are not tightly coupled, and hence sharing resources is difficult.

Synchronization and mutual exclusion in single-threaded processes halt computations.

Threads can perform the asynchronously overlapped functions with less programming effort.

The use of multithreading can benefit more with a multiprocessor environment using efficient
scheduling of threads.

Topic 110
Issues with Threads

Threads share resources within a process. One thread may inadvertently another thread’s data.

In some specific cases, concurrency can greatly degrade performance rather than improve.

In some simple single-threaded solutions using multithreading greatly complicates matters and even
results in poor performance.

Topic 111
Threads Basics

Threads use the space assigned to a process.

In a multi-threaded system, multiple threads might be associated with a process

Thread within a process share data and code.

However, individual threads may have their unique data storage in addition to the shared data.

This data is not altogether exclusive.

Programmer must assure that a thread only uses its own data and does not interfere with
shared data.
Moreover, each thread has its own stack for function calls.

The calling process usually passes an argument to the thread. These arguments are stored in
the thread’s stack

Each thread can allocate its own Thread Local Storage(TLS) and set clear values.

TLS is a small pointer array that is only accessible to the thread.

This also assures that a thread will not modify data of any other thread’s TLS.
Topic 112
Thread Management

Like every other resource threads are also treated as objects

The API used to create a thread is CreateThread()

Threads can also be treated as parent and child. Although the OS is unaware of that.

CreateThread() has many unique requirements

The CreateThread() requires the starting address of the thread within the calling process.

It also requires stack space size for the thread which comes from process address space.

Default memory space IMB

One page is committed in start, this grows with requirement.

Requires a pointer to thread argument. This can be any structure.

HANDLE CreateThread( LPSECURITY_ATTRIBUTES lpsa, SIZE_T dwStackSize, LPTHREAD_START_ROUTINE

lpStartAddress, LPVOID lpParameter, DWORD dwCreationFlags, LPDWORD lpThreadId );

lpsa is security attributes structure used previously.

dwStackSize is the stack size in bytes.

lpStartAddress is the starting address of the thread function within the calling process of the form:

DWORD WINAPI ThreadFunc(LPVOID)

The function returns a DWORD value which is usually an exit code and accepts a single pointer
argument.

lpThreadParam is the pointer passed to the thread and is usually interpreted as a pointer to a structure
containing arguments

dwCreationFlags if 0 would been that thread would start readily. If its CREATE_SUSPENDED then the
thread will be suspended requiring the use of ResumeThread() to start execution.

lpThreadId is a pointer to a DWORD that will receive the thread identifier. If its kept NULL then no
thread identifier will be returned,

Topic 113
Exiting Thread

All threads in a process can exit using a ExitThread() function.

An alternate is that the thread function returns with the exit code.

When a thread exits the thread stack is deallocated and the handle reffering to the thread are
invalidated.

If the thread is linked to some DLL then the DllMain() function is invoked with te reason
DLL_THREAD_DETACH.

VOID ExitThread(DWORD dwExitCode);

When all the threads exit the process terminates.

One thread can terminate another thread using TerminateThread().

In this case Thread resources will not be deallocated, completion handlers do not execute, no
notification is sent to attached DLLs.

Because of these reasons use of TerminateThread() is strongly discouraged.

A thread object will continue to exist even its execution has ended until the last reference to the thread
handle is destroyed with CloseHandle().

Any other running thread can retrieve the exit code.

BOOL GetExitCodeThread(HANDLE hThread, LPDWORD lpdwExitCode);

lpdwExitCode contains a pointer to exit code.

The thread is still running the exit code will be STILL_ACTIVE

Topic 114
Thread Identity

Thread Ids and handles can be obtained using functions quite similar to one used with processes.

GetCurrentThread()

Non-inheritable pseudohandle to the calling thread.

GetCurrentThreadId()
Obtains thread Id rather than the handle.

GetThreadId()

Obtains thread Id from thread handle

OpenThread()

Creates Thread handle from the thread Id

Topic 115
More on Thread Management

The functions of thread management that are discussed above are enough to program any useful
threading application. However, there are some more functions introduced in the later versions of
Windows i.e. Windows XP and Windows 2003 to write a useful and robust program. These functions are
described below.

GetProcessIdOfThread()

This function was not available in the earlier version of windows and requires Windows 2003 or later
versions. When a thread is specified, this function tells us, to which process a specified thread is linked
while returning the id of that process. This function is useful for, mapping thread and process and program
that manages or interacts with threads in another process

GetThreadIOPendingFlag()

This function determines whether the thread, specified by its handle, has any outstanding I/O requests.
For example, the thread might be blocked for some IO operations. The result is the status at the time that
the function is executed; the actual status could change at any time if the target thread completes or
initiates an operation.

Suspending and Resuming Threads

In some cases, we might require to pause any running thread or resume any paused thread. For these
purposes, Windows maintains a suspend count. Every thread has its separate suspend count. A thread
will only run if the suspend count is 0 and if it is not, then the thread is paused and execution of that
thread will be stopped until the suspend count becomes 0. One thread can increment or decrement the
suspend count of another thread using SuspendThread and ResumeThread. Recall that a thread can be
created in the suspended state with a count of 1

DWORD ResumeThread (HANDLE hThread)

DWORD SuspendThread (HANDLE hThread)

Both functions, if successful, return the previous suspend count. 0xFFFFFFFF indicates failure.
Topic 116
Waiting for Threads

One thread can wait for another thread to complete or terminate in the same way that threads wait for
process termination. Threads and actions should be synchronized so that action can be performed after
completing execution. The wait function can be used with the thread handle to check whether a particular
thread has completed its execution or not. Window treats thread as an object, and there are two different
types of wait function in the windows that can be used for threads as well (Since the thread is also an
object), i.e. WaitForSingleObject() or WaitForMultipleObjects(). WaitForSingleObject() is used to wait for
a single specified object and WaitForMultipleObjects can be used for more than 1 object, those objects
are defined in the form of an array.

For WaitForMultpleObjects, there is a defined limit in windows to wait for execution i.e.
MAXIMUM_WAIT_OBJECTS(64), Usually, it is 64 objects. If there are more than 64 objects, we can use
multiple calls as well i.e. if there are 100 objects, we can create 2 arrays of 64 and 36 objects and call
WaitForMultipleObjects() two times.

The wait function waits for the object, indicated by the handle, to become signaled. In the case of threads,
ExitThread() and TerminateThread() set the object to the signaled state, releasing all other threads waiting
on the object, including threads that might wait in the future after the thread terminates. Once a thread
handle is signaled, it never becomes nonsignaled.

Note that multiple threads can wait on the same object. Similarly, the ExitProcess() function sets the
process state and the states of all its threads to signaled.

Topic 117
C Library in Threads

Let’s assume a scenario, where we are using the window threading function with C library functions
concurrently, there might arise a problem of thread safety. For example, strtok() is a C library function,
used to extract the token from the specified string, this function uses the global memory space for its
internal processing, and if we are using the different copies of that string, they all might use global
space. The result achieved from this operation might get compromised and unsatisfactory. In such cases,
these types of problems are resolved by using C Library threading functions rather than Windows
threading functions. For C Library, Windows C provides a thread-safe library named LIBCMT. This library
can be used for thread-relevant functions to program a multithreaded program. So far, we were using
the CreateThread and ExitThread functions to create and exit the threads, this library provides us with
these equivalent functions i.e. _beginthreadex() and endthreadex() respectively. These C library
functions are quite simpler but are not diverse as compared to Windows threading functions. i.e.
_beingthreadex() is a simpler function but it does not allow users to specify the security
attributes. _endthreadex() does not allow return values and does not pass information regarding the
status of the thread. If we are using the windows program and using this C Libray thread function the
return value must be type cast to HANDLE to process it further, since the original return type of
_beginthreadex() is not HANDLE.
Topic 118
Multithreaded Pattern Searching

In this example of pattern searching with multithreading, the program is managing concurrent I/O to
multiple files, and the main thread, or any other thread, can perform additional processing before waiting
for I/O completion. In this way, we can manage several files in a very simple and efficient way. In the
previous examples of pattern searching, we used the multitasking approach, but here we will use the
multithreading technique, that enables us to write an optimal program.

In Synchronous Input-Output, suppose an example of the keyboard, when the user presses any keyboard
button the execution stops until the button is released, but in asynchronous input-output, a Sound card
plays the song at the same time other operations are also performed.

When a read operation is performed, this can be performed concurrently, a file can be read by several
processes at the same time or we can also read several files at the same time. But the problem arises
when several processes attempt to write a single file. For now, we will be limited to read operation only.
This program can be provided several files in which a specific pattern is to be searched. For every file, a
separate thread will be created which will search the pattern in that file. Once the pattern is found in the
file, it will be reported in a temporary file

/* grepMT. */

/* Parallel grep-- multiple thread version. */

#include "Everything.h"

typedef struct { /* grep thread's data structure. */

int argc;

TCHAR targv[4][MAX_PATH];

} GREP_THREAD_ARG;

typedef GREP_THREAD_ARG * PGR_ARGS;

static DWORD WINAPI ThGrep (PGR_ARGS pArgs);

int _tmain(int argc, LPTSTR argv[])

GREP_THREADED_ARG *gArg;
HANDLE *tHandle;

DWORD threadIndex, exitCode;

TCHAR commandLine[MAX_COMMAND_LINE);

int iThrd, threadCount;

STARTUPINFO startUp;

PROCESS_INFORMATION processInfo;

GetStartupInfo(&startUp);

/* Boss Thread: create seperate "grep" thread for each file. */

tHandle = malloc((argc-2) * sizeof(HANDLE));

gArg = malloc((argc-2) * sizeof (GREP_THREAD_ARG));

for(iThrd=0, iThrd<argc-2; iThrd++)

_tcscpy (gArg[iThrd].targv[1], argv[1]); /* Pattern */

_tcscpy (gArg[iThrd].targv[2], argv[iThrd + 2]);

GetTempFileName /* Temp file name. */

(".", "Gre", 0, gArg[iThrd].targv[3];

gArg[iThrd].argc = 4;

/* Create a worker thread to execute the command line. */

tHandle[iThrd] = (HANDLE)_beginthreadex (
 NULL, 0, ThGrep, &gArg[iThrd], 0, NULL);

/* Redirect std Output for file listing process. */

startUp.dwFlags = STARTF_USESTDHANDLES;

startUp.hStdOutput = GetStdHandle (STD_OUTPUT_HANDLE);

/* Worker threads ar eall running . Wait for them to complete. */

threadCount = argc - 2;

while (threadCount>0) {

threadIndex = WaitForMultipleObjects (

threadCount, tHandle, FALSE, INFINITE);

iThrd = (int) threadIndex - int (WAIT_OBJECT_0);

GetExitCodeThread ( tHandle[iThrd], &exitCode);

CloseHandle ( tHandle[iThrd]);

if ( exitCode == 0) { /* Pattern Found */

if ( argc > 3) {

/* Print file name if more than one . */

_tprintf ( _T("\n**Search results - file : %s\n"),

gArg[iThrd].targv[2];

fflush(stdout);

/* Use the "cat" program to list the result files. */

_stprintf ( commandLine, _T("%s%s"), _T("cat "),

 gArg[iThrd].targv[3]);

CreateProcess(NULL, commandLine, NULL, NULL,

TRUE, 0, NULL, NULL, &startUP, &processInfo);

waitForSingleObject ( processInfo.hProcess, INFINITE);

CloseHandle (processInfo.hProcess);

CloseHandle (processInfo.hThread);

DeleteFile ( gArg[iThrd].targv[3]);

/* Adjust thread and file name arrays. */

tHandle[iThrd] = tHandle[threadCount - 1];

_tcscpy(gArg[iThrd].targv[3], gArg[threadCount -1 ].targv[3]);

_tcscpy(gArg[iThrd].targv[2], gArg[threadCount -1 ].targv[2]);

threadCount--;

/* The form of grep thread function code is :

static DWORD WINAPI ThGrep (PGR_ARGS pArgs)

....

} */
 Structure of thread argument is created with argument count (argc) and thread
argument value (targv) which will be passed to thread. A Thread prototype is also
created named as ThGrep which will be used to search the specific patterns in the file.
  STARTUPINFO and PROCESS_INFORMATION structures are created for startup and for
creating process respectively
 stratUp information is placed in GetStratupInfo function
 Files are specified using the command line, in which patterns are to be searched, and
for every file inputted, a separate thread will run.
 Loop will run till argc-2 times, in argc, the first two parameters will be process name and
pattern and 3rd parameter is inputted file names, though, this loop will run, till the
number of files inputted. In this loop for every file, the name of the file is copied and its
temporary file is created
 Further, arguments that are to be passed to the thread are also prepared, 1st argument
will store the pattern which is to be searched, 2nd argument will store the name of the
file in which the pattern is to be searched and 4th argument will store the count i.e. how
many arguments are stored.
 Thread is created using _beginthreadex, whose name is ThGrep and is passed all the
arguments that were created (gArg). With this, the handle of every thread will be stored
in the form of an array (tHandle)
 Standard output files, standard error files, and flags are set for startup information
 Total number of threads are stored in ThreadCount
 Another loop is run, in which the Wait function is called to wait for multiple objects
specified with a number of threads (threadCount) and handles of all threads (tHandle
array). When the wait function is completed, an exitcode will be generated which will
tell us why the wait function is stopped, and further, for garbage collection, the handle
of the thread is also closed with the CloseHandle function.
 When the wait function was in execution, a process is created with help of the
CreateProcess function which has been provided, startup information(startUp), and
process information (processInfo).
 Another wait function is also called but this time WaitForSingleObject is called which
has been provided the handle of the process named hProcess.
 Once the execution of the process is completed, the handle of the process as well as
the thread are closed.

Topic 119
Boss worker and other thread models

In the example of multithreaded pattern searching, there was one main thread, which was running other
threads. Each file was assigned a separate thread, which was finding the pattern in that file. This model is
more like a boss worker model. The boss worker model is one in which there is one boss and many
workers. The Boss assigns work to workers and each worker report result back to the boss. There are
many other models, which are used to write an efficient and more understandable multithreaded
program depending on the scenarios.

The work crew model is one in which the workers cooperate on a single task, each performing a small
piece. They might even divide up the work themselves without direction from the boss. Multithreaded
programs can employ nearly every management arrangement that humans use to manage concurrent
tasks.
The Client-Server model is mostly used worldwide, in which a client requests the server and the server
runs a thread for that client. For every client, a separate thread is run at the server end. In this way, the
work is done concurrently rather than sequentially. Another major model is the pipeline model, where
work moves from one thread to the next

There are many advantages to using these models when designing a multithreaded program, including
the following.

 Most models can be used which makes things simpler and expedites programming and
debugging efforts.
 Models help you obtain the best performance and avoid common mistakes
 Models naturally correspond to the structure of programming language constructs.
 Maintenance is simplified.
 Troubleshooting is simplified when they are analyzed in terms of a specific model.
 Synchronization and coordination are simplified using well-defined models.

Topic 120
MergeSort: Exploiting Multiple Processors

The window is a multiprocessing system i.e. more than 1 process can run simultaneously. This example
shows, how multithreading can be used to achieve the optimal performance gain in a multiprocessing
system, each process runs a different thread to utilize the optimal resources. The main idea behind this is
to subdivide the process into similar tasks so that a separate thread is run for each subtask i.e. A big array
is divided into smaller parts, each part is sorted separately with different threads, and all the parts are
merged. This will allow parallelism and better performance gain. The strategy implemented in this
example is the worker crew model, work is divided into different workers, and all the work is merged in
the end. This strategy could also be used, by using multiprocessing instead of multithreading, but the
result might not be as efficient as it is with multithreading, because switching overhead is low in
multithreading but high in multiprocessing. In the example of MergeSort, each subarray is sorted with
qsort() and merged as in the mergesort algorithm. The program code will run most accurately if a number
of records are divisible by a number of threads and the number of threads is in the power of two. If the
number of processes is equal to the number of threads, this would be the most optimal situation
otherwise less optimal. If a list is subdivided into 4 sub-lists and 4 threads are created for these sublists,
they all must be created at a suspended state and should only be resumed when all the threads are
created. If one thread is completed and the other, which is to be merged, is not created or does not exist,
this will occur in a race condition. To avoid the race condition, all the threads should be created with a
suspended state and resumed to run all concurrently. The following diagrams explain it further
A Large array is divided into smaller 4 subarrays. For each subarray, a separate thread is created i.e. thread
0, thread 1, thread 2, and thread 3. When thread 0 is sorted it will wait for thread 1 to be sorted, once
sorted, they both will be merged. The same happens for thread 2 and thread 3, they are merged when
sorted. These 2 merged subarrays are then sorted and merged to form a large sorted array.

Topic 121
MergeSort: Exploiting Multiple Processors

/* Chapter 7 SortMT. Work crew model

File sorting with multiple threads and merge sort.

sortMT [options] nt file. Work crew model. */

/* This program is based on sortHP.

It allocates a block for the sort file, reads the file,

 then creates an "nt" threads (if 0, use the number of

processors to so sort a piece of the file. It then

merges the results in pairs. */

/* LIMITATIONS:

1. The number of threads must be a power of 2

2. The number of 64-byte records must be a multiple

of the number of threads.

An exercise asks you to remove these limitations. */

#include "Everything.h"

/* Definitions of the record structure in the sort file. */

#define DATALEN 56 /* Correct length for presdnts.txt and monarchs.txt. */

#define KEYLEN 8

typedef struct _RECORD {

TCHAR key[KEYLEN];

TCHAR data[DATALEN];

} RECORD;

#define RECSIZE sizeof (RECORD)

typedef RECORD * LPRECORD;

typedef struct _THREADARG { /* Thread argument */

DWORD iTh; /* Thread number: 0, 1, 3, ... */

LPRECORD lowRecord; /* Low Record */

LPRECORD highRecord; /* High record */

} THREADARG, *PTHREADARG;
static DWORD WINAPI SortThread (PTHREADARG pThArg);

static int KeyCompare (LPCTSTR, LPCTSTR);

static DWORD nRec; /* Total number of records to be sorted. */

static HANDLE * pThreadHandle;

int _tmain (int argc, LPTSTR argv[])

/* The file is the first argument. Sorting is done in place. */

/* Sorting is done in memory heaps. */

HANDLE hFile, mHandle;

LPRECORD pRecords = NULL;

DWORD lowRecordNum, nRecTh,numFiles, iTh;

LARGE_INTEGER fileSize;

BOOL noPrint;

int iFF, iNP;

PTHREADARG threadArg;

LPTSTR stringEnd;

iNP = Options (argc, argv, _T("n"), &noPrint, NULL);

iFF = iNP + 1;

numFiles = _ttoi(argv[iNP]);

if (argc <= iFF)

ReportError (_T ("Usage: sortMT [options] nTh files."), 1, FALSE);

 /* Open the file and map it */

hFile = CreateFile (argv[iFF], GENERIC_READ | GENERIC_WRITE,

0, NULL, OPEN_EXISTING, 0, NULL);

if (hFile == INVALID_HANDLE_VALUE)

ReportError (_T ("Failure to open input file."), 2, TRUE);

// For technical reasons, we need to add bytes to the end.

/* SetFilePointer is convenient as it's a short addition from the file end */

if (!SetFilePointer(hFile, 2, 0, FILE_END) || !SetEndOfFile(hFile))

ReportError (_T ("Failure position extend input file."), 3, TRUE);

mHandle = CreateFileMapping(hFile, NULL, PAGE_READWRITE, 0, 0, NULL);

if (NULL == mHandle)

ReportError (_T ("Failure to create mapping handle on input file."), 4, TRUE);

/* Get the file size. */

if (!GetFileSizeEx (hFile, &fileSize))

ReportError (_T ("Error getting file size."), 5, TRUE);

nRec = (DWORD)fileSize.QuadPart / RECSIZE; /* Total number of records. Note

assumed limit */

nRecTh = nRec / numFiles; /* Records per thread. */

threadArg = malloc (numFiles * sizeof (THREADARG)); /* Array of thread args. */

pThreadHandle = malloc (numFiles * sizeof (HANDLE));

 /* Map the entire file */

pRecords = MapViewOfFile(mHandle, FILE_MAP_ALL_ACCESS, 0, 0, 0);

if (NULL == pRecords)

ReportError (_T ("Failure to map input file."), 6, TRUE);

CloseHandle (mHandle);

/* Create the sorting threads. */

lowRecordNum = 0;

for (iTh = 0; iTh < numFiles; iTh++) {

threadArg[iTh].iTh = iTh;

threadArg[iTh].lowRecord = pRecords + lowRecordNum;

threadArg[iTh].highRecord = pRecords + (lowRecordNum + nRecTh);

lowRecordNum += nRecTh;

pThreadHandle[iTh] = (HANDLE)_beginthreadex (

NULL, 0, SortThread, &threadArg[iTh],

CREATE_SUSPENDED, NULL);

/* Resume all the initially suspened threads. */

for (iTh = 0; iTh < numFiles; iTh++)

ResumeThread (pThreadHandle[iTh]);

/* Wait for the sort-merge threads to complete. */

WaitForSingleObject (pThreadHandle[0], INFINITE);

for (iTh = 0; iTh < numFiles; iTh++)

 CloseHandle (pThreadHandle[iTh]);

/* Print out the entire sorted file. Treat it as one single string. */

stringEnd = (LPTSTR) pRecords + nRec*RECSIZE;

*stringEnd =_T('\0');

if (!noPrint) {

_tprintf (_T("%s"), (LPCTSTR) pRecords);

UnmapViewOfFile(pRecords);

// Restore the file length

/* SetFilePointer is convenient as it's a short addition from the file end */

if (!SetFilePointer(hFile, -2, 0, FILE_END) || !SetEndOfFile(hFile))

ReportError (_T("Failure restore input file lenght."), 7, TRUE);

CloseHandle(hFile);

free (threadArg); free (pThreadHandle);

return 0;

} /* End of _tmain. */

static VOID MergeArrays (LPRECORD, DWORD);

DWORD WINAPI SortThread (PTHREADARG pThArg)

ReportError (_T("Failure restore input file lenght."), 7, TRUE);

 CloseHandle(hFile);

free (threadArg); free (pThreadHandle);

return 0;

} /* End of _tmain. */

static VOID MergeArrays (LPRECORD, DWORD);

DWORD WINAPI SortThread (PTHREADARG pThArg)

DWORD groupSize = 2, myNumber, twoToI = 1;

/* twoToI = 2^i, where i is the merge step number. */

DWORD_PTR numbersInGroup;

LPRECORD first;

myNumber = pThArg->iTh;

first = pThArg->lowRecord;

numbersInGroup = (DWORD)(pThArg->highRecord - first);

/* Sort this portion of the array. */

qsort (first, numbersInGroup, RECSIZE, KeyCompare);

/* Either exit the thread or wait for the adjoining thread. */

while ((myNumber % groupSize) == 0 && numbersInGroup < nRec) {

/* Merge with the adjacent sorted array. */

WaitForSingleObject (pThreadHandle[myNumber + twoToI], INFINITE);

MergeArrays (first, numbersInGroup);

 numbersInGroup *= 2;

groupSize *= 2;

twoToI *=2;

return 0;

static VOID MergeArrays (LPRECORD p1, DWORD nRecs)

/* Merge two adjacent arrays, each with nRecs records. p1 identifies the first */

DWORD iRec = 0, i1 = 0, i2 = 0;

LPRECORD pDest, p1Hold, pDestHold, p2 = p1 + nRecs;

pDest = pDestHold = malloc (2 * nRecs * RECSIZE);

p1Hold = p1;

while (i1 < nRecs && i2 < nRecs) {

if (KeyCompare ((LPCTSTR)p1, (LPCTSTR)p2) <= 0) {

memcpy (pDest, p1, RECSIZE);

i1++; p1++; pDest++;

else {

memcpy (pDest, p2, RECSIZE);

i2++; p2++; pDest++;

 }

if (i1 >= nRecs)

memcpy (pDest, p2, RECSIZE * (nRecs - i2));

else memcpy (pDest, p1, RECSIZE * (nRecs - i1));

memcpy (p1Hold, pDestHold, 2 * nRecs * RECSIZE);

free (pDestHold);

return;

int KeyCompare (LPCTSTR pRec1, LPCTSTR pRec2)

DWORD i;

TCHAR b1, b2;

LPRECORD p1, p2;

int Result = 0;

p1 = (LPRECORD)pRec1;

p2 = (LPRECORD)pRec2;

for (i = 0; i < KEYLEN && Result == 0; i++) {

b1 = p1->key[i];

b2 = p2->key[i];

if (b1 < b2) Result = -1;

if (b1 > b2) Result = +1;

}
 return Result;

 A thread structure is defined which is passed to every thread as an argument, this

structure has the ‘iTh’ variable which specifies the thread number, ‘lowRecord’ and
‘highRecord’ which are used to define the starting and ending index of the sub list.
 ‘SortThread’ is the thread name that is to be created
 ‘nRec’ is the total number of threads created
 ‘pThreadHandle' is the handle to the thread
 The file, which has records in it, is opened to read and write. This file is mapped
(‘mHandle’) to access that file as memory i.e. in the form of an array.
 Thread argument, which is to be passed as an argument to create a thread, is allocated the
memory, that has all three fields that are to be passed
 ‘nRec’ is the total number of records
 ‘nRec/numFiles’ defines the number of records per file
 ‘pRecord’ is the address of the mapped file
 Different threads are created and run to divided the file into chunks and each thread sorts
the file. To divide the file, a loop is run till ‘numFiles’ times, and specifies the thread
numbers (‘ith’), starting index of a chunk (‘lowRecord’), and ending index of a chunk
(‘highRecord’)
 The thread is created with ‘beginThread’ functions which are passed ‘SortThread’ and
‘ThreadArgument’ in a suspended state.
 When all the threads are created, they are resumed with help of a loop.
 For all created threads, we wait till the execution completes. Once the execution is
completed handle is closed and the file is unmapped.
 When a thread is run, it is passed ‘QSort’ to sort the records, and then it waits for its
adjacent thread to be sorted and completed. Once they are completed, the results are
merged. Every thread waits for its adjacent thread to be completed and merges after
completion

Performance

 Multithreading gives the best results when the number of processors is the same as the
number of threads.
 Additional threads beyond processor count slows down the program
 Performance degrades if there is one processor and memory is low and the array is very
large because most of the time threads will be contending for physical memory. But the
problem alleviates when the memory is sufficient

Topic 122
Introduction to Parallelism

Multiprocessing and multithreading, both are responsible for multiple flows of execution, but the
multithreading is optimal. Windows is not only multithreading or multiprocessing it also supports
multiprocessors as well. If we have a system with multiprocessors, we should learn to program to use the
potential of multiprocessors because the processor’s speed has reached its bottleneck i.e. after a certain
limit, its speed cannot be enhanced. Parallelization is the key to future performance improvement since
we can no longer depend on increased CPU clock rates and since multicore and multiprocessor systems
are increasingly common. Previously, various programs have been discussed that unleash the power of
parallelism. The properties that enabled parallelism include the following:

 Major task is divided into subtasks and many worker threads were run. Subtasks were divided
into worker threads that perform their work. These worker subtasks run independently, without
any interaction between them.
 As subtasks are complete, a master program can merge the results of divided subtasks into a
single result.
 The programs do not require mutual exclusion of any sort. Only the master worker is synchronized
with each worker and waits for them to complete.
 Every worker will work as a separate thread on a separate processor. it is the most optimal
situation
 Program performance scales automatically, up to some limit, as you run on systems with more
processors; the programs themselves do not, in general, determine the processor count on the
host computer. Instead, the Windows kernel assigns worker subtasks to available processors.
 Output remains undisturbed even if the program is serialized.
 If you “serialize” the program the results should get precisely the same as the parallel program.
The serialized program is, moreover, much easier to debug.
 The maximum performance improvement is limited by the program’s “parallelism,” thread
management overhead, and computations that cannot be parallelized.

Topic 123
Thread Local Storage

A thread is an execution unit. In a multithreading program, one procedure can have several threads. Every
thread needs data, that it doesn’t want to share with other threads and which is unique i.e. it varies from
thread to thread. One technique is to have the creating thread call CreateThread (or beginThreadex) with
lpvThreadParm pointing to a data structure that is unique for each thread. The thread can then allocate
additional data structures and access them through lpvThreadParm. Windows also provides Thread Local
Storage (TLS), which gives each thread its array of pointers. The following figure shows this TLS
arrangement.
A function can have many threads i.e. thread 1, thread 2, etc. Every column in the TLS arrangement
corresponds to thread numbers and every thread needs variables i.e. TLS index 0,1,2,3 etc. Initially, no
TLS indexes (rows) are allocated, but new rows can be allocated and deallocated at any time. Once the
row is allocated, it will be allocated to all rows. The primary thread would be a logical choice for TLS space
management, however, every thread can access TLS.

 DWORD TlsAlloc(VOID): This API is used to allocate the index and it returns the TLS index in the
form of the double word. Otherwise returns -1 in case of failure.
 BOOL TlsFree(DWORD dwIndex): Frees the specified index.

 LPVOID TlsGetValue (DWORD dwTlsIndex ) and BOOL TlsSetValue (DWORD dwTlsIndex,

LPVOID lpTlsValue): Provided valid indexes are used. The programmer can access TLS space
using these simple GET/SET APIs

Some Cautions

 TLS provides a convenient mechanism for accessing memory that is global within a thread but
inaccessible to other threads.
 Global storage of a program is accessible by all threads

 TLS provides a convenient mechanism for accessing memory that is global within a thread but
inaccessible to other threads.
 Global storage of a program is accessible by all threads

Topic 124
Processes and Thread priorities
In a multitasking or multithreading system, there are a number of processes running, each of which
competes for resources like processor, memory, etc. The operating system is responsible for managing
the resources. The Windows kernel always runs the highest-priority thread that is ready for execution. A
thread is not ready if it is waiting, suspended, or blocked for some reason. Since the threads are
dependents on the processes and they receive priority relative to their process priority classes. Process
priority classes are set initially when they are created using CreateProcess, and each has a base priority,
with values including

 IDLE_PRIORTY_CLASS for threads that will run only when the system is idle. This is the lowest
priority process.
 NORMAL_PRIORTY_CLASS indicating no special scheduling requirements.
 HIGH_PRIORTY_CLASS indicating time-critical tasks that should be executed immediately.
 REALTIME_PRIORTY_CLASS, the highest possible priority

The priority class of a process can be set and got using BOOL SetPriorityClass(HANDLE hProcess,
DWORD dwPriority) and DWORD GetPriorityClass(HANDLE hProcess) respectively.

There are some enhances variants of priority levels i.e. ABOVE_NORMAL_PRIORTY_CLASS (which is
below HIGH_PRIORTY_CLASS) and BELOW_NORMAL_PRIORTY_CLASS (which is above
IDLE_PRIOTY_CLASS).

PROCESS_MODE_BACKGROUND_BEGIN, which lowers the priority of the process and its threads for
background work without affecting the responsiveness of foreground processes and
threads. PROCESS_MODE_BACKGROUND_END restores the process priority to the value before it was
set with PROCESS_MODE_BACKGROUND_BEGIN.

Thread priorities are either absolute or are set relative to the process base priority. At thread creation
time, the priority is set to that of the process. The relative thread priorities are in a range of ±2 “points”
from the process’s base. The symbolic names of the resulting common thread priorities, starting with
the five relative priorities, are:

 THREAD_PRIORTY_LOWEST
 THREAD_PRIORTY_BELOW_NORMAL
 THREAD_PRIORTY_NORMAL
 THREAD_PRIORTY_ABOVE_NORMAL
 THREAD_PRIORTY_HIGHEST
 THREAD_PRIORTY_TIME_CRITICAL is 15, or 31 if the process class is REAL_TIME_PRIORTY_CLASS
 THREAD_PRIORTY_IDLE is 1, or 16 for REAL_TIME_PRIORTY_CLASSES processes

THREAD_MODE_BACKGROUND_BEGIN and THREAD_MODE_BACKGROUND_END are similar

to PROCESS_MODE_BACKGROUND_BEGIN and PROCESS_MODE_BACKGROUND_END.

Thread priorities can be set within the range of +-2 of the corresponding process priority using these BOOL
SetThreadPriority(HANDLE hThread, int nPriority) AND Int GetThreadPriority(HANDLE hThread).

Thread priorities are dynamic. They change with the priority of process or windows may also boost
thread priority as per need. This feature can be enabled disabled using SetThreadPriorityBoost()
Topic 125
Thread States

The following figure shows how the executive manages threads and shows the possible thread states.
This figure also shows the effect of program actions. Such state diagrams are common to all multitasking
OSs and help clarify how a thread is scheduled for execution and how a thread moves from one state to
another.
 Figure 1 Thread States and Transitions

(From Inside Windows NT, by Helen Custer. Copyright © 1993, Microsoft Press.
Reproduced by permission of Micro-soft Press. All rights reserved.)

 A thread is in the running state when it is running on a processor. More than one thread can be
in the running state on a multiprocessor computer
 The executive places a running thread in the wait state when the thread performs a wait on a non-
signaled handle, such as a thread or process handle. I/O operations will also wait for the
completion of a disk or other data transfer, and numerous other functions can cause waiting. It is
common to say that a thread is blocked, or sleeping, when in the wait state
 A thread is ready if it could be running. The executive’s scheduler could put it in the running state
at any time. The scheduler will run the highest-priority ready thread when a processor becomes
available, and it will run the one that has been in the ready state for the longest time if several
threads have the same high priority. The thread moves through the standby state before entering
the ready state.
 The executive will move a running thread to the ready state if the thread’s time slice expires
without the thread waiting. Executing will also move a thread from the running state to the ready
state.
 The executive will place awaiting thread in the ready state as soon as the appropriate handles are
signaled, although the thread goes through an intermediate transition state. It is common to say
that the thread wakes up.
 A thread, regardless of its state, can be suspended, and a ready thread will not be run if it is
suspended. If a running thread is suspended, either by itself or by a thread on a different
processor, it is placed in the ready state.
 A thread is in the terminated state after it terminates and remains there as long as there are any
open handles on the thread. This arrangement allows other threads to interrogate the thread’s
state and exit code.
 Normally, the scheduler will place a ready thread on any available processor. The programmer
can specify a thread’s processor, which will limit the processors that can run that specific thread.
In this way, the programmer can allocate processors to threads and prevent other threads from
using these processors, helping to assure responsiveness for some threads. The appropriate
functions are SetProcessAffinityMask and GetProcessAffinityMas. SetThreadIdealProcessor can
specify a preferred processor that the scheduler will use whenever possible; this is less restrictive
than assigning a thread to a single processor with the affinity mask.

Topic 126
 Make no assumptions about the order in which the parent and child threads execute.
 It is possible for a child thread to run to completion before the parent, or, conversely, the child
thread may not run at all for a considerable period.
 On a multiprocessor computer, the parent and one or more children may even run concurrently.
 Make sure all the initializations required by a child thread have been performed before calling
CreateThread()
 In case a thread has been run but initialization is required then use some technique like thread
suspension until data is initialized.
 Failure by the parent to initialize data required by the child is a common cause of “race conditions”
wherein the parent “races” the child to initialize data before the child needs it.
  Any thread, at any time, can be preempted, and any thread, at any time, may resume execution
 Do not confuse synchronization and priority. These both are different concepts. Threads are
defined as their priorities when created, However Threads are synchronized in such a way that
one thread completes its specific purpose, and only then another thread may run its process.
 Even more so than with single-threaded programs, testing is necessary, but not sufficient, to
ensure program correctness. It is common for a multithreaded program to pass extensive tests
despite code defects. There is no substitute for careful design, implementation, and code
inspection.
 Threaded program behavior varies widely with processor speed, number of processors, OS
version, and more. Testing on a variety of systems can isolate numerous defects, but the
preceding precaution still applies.
 The default stack size for a thread is 1MB. Make sure the size is sufficient as per thread needs.
 Threads should be used only as appropriate. Thus, if there are activities that are naturally
concurrent, each such activity can be represented by a thread. If, on the other hand, the activities
are naturally sequential, threads only add complexity and performance overhead.
 If you use a large number of threads, be careful, as the numerous stacks will consume virtual
memory space and thread context switching may become expensive. In other cases, it could mean
more threads than the number of processors.
 Fortunately, correct programs are frequently the simplest and have the most elegant designs.
Avoid complexity wherever possible.

Topic 127
Timed Waits

In threading, there are some functions that can be used to wait for threads. The Sleep function allows a
thread to give up the processor and move from the running to the wait state for a specified period of time.
After the completion of the specified time, the thread will move to the ready state and will wait to move
on running state. A thread can perform a task periodically by sleeping after carrying out the task.

VOID Sleep (DWord dwMilliseconds)

The time period is specified in milliseconds and can even be INFINITE, in which case the thread will never
resume. A 0 value will cause the thread to relinquish the remainder of the time slice; the kernel moves
the thread from the running state to the ready state.

The function SwitchToThread() provides another way for a thread to yield its processor to another ready
thread if there is one that is ready to run.

Topic 128
Fibers

A fiber, as the name implies, is a piece of a thread. More precisely, fiber is a unit of execution that can be
scheduled by the application rather than by the operating system. An application can create numerous
fibers, and the fibers themselves determine which fiber will execute next. The fibers have independent
stacks but otherwise run entirely in the context of the thread on which they are scheduled, having access,
for example, to the thread’s TLS and any mutexes owned by the thread. Furthermore, fiber management
occurs entirely in user space outside the kernel. Fibers can be thought of as lightweight threads, although
there are numerous differences. A fiber can execute on any thread, but never on two at one time. Fiber
that is meant to run on different threads at different instances should not access thread-specific data from
TLS.

Fiber Uses

 Fibers make portability easy.

 Fibers need not wait/block on file locks, mutexes, and pipes. They can pool resources and in case
resources are not available they can switch to another fiber.
 Fiber exhibits flexibility. They exist as part of threads but they are not bound to any specific thread.
They can run on any thread but only one at a time.
 Fibers are not pre-emptively scheduled. Windows OS is unaware of fibers. They are controlled
and managed by fiber DLL.
 Fibers can be used as co-routines. Using this an application can switch between related tasks.
Whereas, in the case of threads applications has no control over which thread executes next.
 Major software platform offers the use of fibers and claims its performance advantages.

Topic 129
Fiber APIs

A set of different API functions are provided that can help create and manage fibers. These functions are

A thread must enable fiber operation by calling

ConvertThreadToFiber() or

ConvertThreadToFiberEx()

After calling this API the thread will now contain a fiber. It will provide a pointer to the fiber data
more or less like thread data. This can be used accordingly.

Subsequently, new fibers can be created in this thread using

CreateFiber()

Each new fiber has a start address, stack size, and a parameter. Each fiber is identified by
address and not a handle.

Individual fiber can obtain their data by calling

GetFiberData()

Similarly, a fiber can obtain its identity by calling

 GetCurrentFiber()

A running fiber gives control to another fiber using

SwitchToFiber()

It uses the address of the other fiber. The context of the current fiber is saved and the context of
the other fiber is restored. Fibers must explicitly indicate the next fiber that is to run in the
application.

A running fiber gives control to another fiber using

SwitchToFiber()

It uses the address of the other fiber. The context of the current fiber is saved and the context of
the other fiber is restored. Fibers must explicitly indicate the next fiber that is to run in the
application.

An existing fiber can be deleted using

DeleteFiber()

Topic 130
Using Fibers

Previously we discussed several APIs used to manage the fibers, here we will develop a scheme with these
APIs to use the fibers. Fiber enables us to control the switching of threads. This scheme firstly converts a
thread to fiber and then uses it as the primary fiber. The primary fiber creates other fibers and switching
among these fibers is managed by the application.
In the center top, a primary thread is created in which ConvertThreadToFiber is used to convert into fiber
and then with the help of a loop number of fibers are created. To convert the execution to a certain fiber
SwitchToFiber is used. Primary fiber is switched to Fiber 0 which gets fiber data and then switches to Fiber
1 which also performs the same work i.e. gets data and this fiber switches to primary fiber. The primary
fiber starts execution from the point where it was switched before i.e. now it switches to fiber 2 which
gets data and switches to fiber 0 and then back to fiber 2 and at the end Thread is closed.

There are two policies

 Master-Slave Scheduling: One fiber decides which fiber to run. Each fiber transfers back the
execution to the primary fiber. (Fiber 1)
 Peer to Peer Scheduling: A fiber determine which fiber should be next to execute based on some
policy (Fiber 0 and 2)

Topic 131
Need for Thread Synchronization
 Threads enable concurrent processing and parallelism in multiple processors. However, there are some
pros and cons because of this concurrent processing. When many threads may run concurrently. They
may need to synchronize for tasks in numerous instances. i.e. in the Boss-worker model, there are one
boss thread and many workers threads, the Boss thread waits for all workers to complete execution and
compiles all the data. In case if boss executes before the worker completes its execution, this may affect
the output we are required to ensure that the Boss will not access the worker’s memory unless the worker
completes it. Alternately, the workers do not start working unless the boss has created all the workers so
that a worker may not try to access the data of another worker which has not been created as yet. When
multiple threads are using share data they may require coordination i.e. when a thread is using the data
other threads should wait for it. Also, programmers need to ensure that two or more threads are not
modifying the same data item simultaneously. In case many threads are modifying elements of the queue,
the programmer needs to assure that two or more threads do not attempt to remove an element at the
same time. Several programming flaws can leave such vulnerabilities in the code.

Example

There are two threads i.e. Thread 1 and Thread 2, operation of both of these is the same i.e.

Thread 1 Thread 2

{ {

M=N M=N

M=M+1 M=M+1

N=N N=N

} }

Let’s suppose that the initial value of N is 4 though the final values after execution of thread 1 will be M=5,
N=5. After completion of thread 1 when thread 2 starts executing, the initial value of N is 5 (because of
thread 1 execution) thus final values become M=6 and N=6.
When these same threads are run concurrently as shown in the following
figure.

When the thread is executing, the value of N is 4 initially. After two instructions thread is switched. Thus
the value of M becomes 5 and the value of N is not changed because 3rd instruction is not executed yet.
Since the thread is switched therefore thread 2 starts execution and value N is 4 here. It executes all
three instructions and values become M=5 and N=5. Again thread is switched and the remaining 1
instruction of thread 1 executes which affects the N variable and because of this final value of N
becomes 5.

Final Values in normal processing

M=6 and N=6

Final Values in concurrent processing

M=5 and N=5

Thus we can see in many situations the final output of concurrent processing may not be same as it is in
the normal processing

Topic 132
A Simple Solution to Critical Section Problem

When problems and errors arise because of parallelism or concurrent processing these types of errors
are of critical section problem i.e. A critical resource which is used by number of processes or threads.
This critical section problem can be handled by the following ways:

Using Global Variables

This solution uses a global variable named Flag to indicate to all threads that a thread is modifying
a variable. A thread turns it TRUE before modifying a variable turns it to FALSE after modifying it.
Each thread would check this Flag before modifying the variable. If the Flag is TRUE then if
indicates that the variable is being used by some other thread.

Even in this case, the thread could be preempted between the time FLAG is tested and the time FLAG is
set to TRUE; the first two statements form a critical code region that is not properly protected from
concurrent access by two or more threads. Another attempted solution to the critical code region
synchronization problem might be to give each thread its own copy of the variable, as follows:
Topic 133

Volatile Storage

The volatile stage is a Windows level or Compiler level facility provided when using incrementing
operations which change the shared variables to reduce the conflicts that arise because of switching.

Latent Defects

Even if the synchronization problem is somehow resolved there still may remain some latent problems. A
thread code may switch to another thread while a variable value has been modified in a register without
writing it back. The use of registers for intermediate operations is a compiler optimization technique.

Turning off optimization may adversely affect the performance of the whole program but not always
sometimes it may slow the program. ANSI C provides a qualifier volatile for this purpose. A variable with
a volatile qualifier will always be accessed from memory for operations and will always be stored in
memory after any operation. A volatile qualifier also means that the variable can be accessed at any
instance. A volatile qualifier should only be used where necessary as it degrades the performance.

Using volatile

 Use volatile only if a variable is being modified by two or more threads.

 Even if it’s read only for two or more threads but the outcome of the threads depends on its
new value.

Topic 134

Memory Architecture and Memory Barriers

Cache Coherency:
The volatile qualifier does not assure that the changes are visible to processors in a desired order.
Processor usually hold the values in cache before writing them back to memory. This may alter
the sequence in which different processors see the values.

Memory Barriers
To assure that the memory is accesses in the desired order use memory barrier or memory
fences. The interlocked function provides memory barrier. Further, the concept is clarified the
diagram showing 4 processors in a system on two dual core chips.

The diagram depicts 4 processor core on two chips. Each core has its own register with
intermediate values of variables. Each core has separate Level-1 cache for instruction and data. A
common larger L2 cache for cores on each chip. Memory is shared among cores.

Volatile qualifiers only assure that the new data values will be updated in L1. There is no guarantee
that the values will be visible to other thread running on different processors. Memory barriers
assure that the main memory is updated and cache of all processors is coherent.

For example, if core 0 updates a variable N using a memory barrier. If core 3 L1 cache also contains
the values of N then the value in its cache is either updated or removed so that core 3 could access
the new value coherent value of N. This certainly incurs a huge cost. Moving data within the core
registers costs less than a cycle whereas more data from one core to another via main memory can
cost 100s of cycles

Topic 135
Interlocked Functions:

Interlocked functions are most suited if variable with volatile scope only need to be incremented,
decremented and exchanges. Interlocked functions are simpler and faster and easy to use.
However, they do pose the performance drawback as they do generate a memory barrier.

The most basic of these functions are InterlockedIncrement() and InterlockedDecrement().

Long InterlockedIncrement(LONG volatile *Addend)

Long InterlockedDecrement(LONG volatile *Addend)

They both use a 32-bit signed variable as parameter that should be stored at 4-byte boundary in
memory. They should be used whenever possible to improve performance.

For 64-bit systems InterlockedIncrement64() and InterlockedDecrement64() can be used with

Addend placed at 8-byte boundary.

The function can be used in following way:

InterlockedIncrement(&N);

N should be volatile integer placed on appropriate memory boundary. Function return the new
value of N. However, some other thread may preempt before the value is returned and change the
value of N. Do not call it twice to increment the value by two as the thread may be preempted
between both calls. Rather use InterlockedExchangeAdd().

Topic 136
Local and Global Storage:

Another criterion for correct thread code is that global storage should not be used for local storage
purpose. If a global variable is used to store thread specific data, then it will be used by other
threads as well. This will result in incorrect behavior no matter how the rest of code is written. The
following example is incorrect usage of the global variables.

DWORD N;

....

DWORD WINAPI ThreadFunc (TH_ARGS pArgs)

...

N = 2 * pArgs->Count; ...

/* Use N; value is specific to this call to ThreadFunc*/

N is kept global but is used to store thread specific information from its parameters structure. For
example, if many thread functions are running at the same time and each function modifies N with
its own parameters, no function will be able to use N properly because N is global and all threads
are running at the same time and affecting N's value. Due to this problem, the final results will be
incorrect.
It's important to know when to use local and when to use global variables while dealing with these
variables. If you know a variable contains thread-specific information, make it local. You can
make a variable global if you know the information in it will be used by other threads.

Adhering to such practices can become even more convoluted when single threaded programs are
converted to run as multi-threaded programs have threads being run in parallel. For example:

DWORD N=0;

....

for (int i=0; i < MAX; ++i) {

...Allocate and initialize pArgs data ...

N += ThreadFunc(pArgs);

...

DWORD WINAPI ThreadFunc (ARGS pArgs)

DWORD result;

...

result=...;

return result;

The code above is for a single-threaded program. The program executes sequentially. The thread
function is called repeatedly in a for loop. When a thread is called, the information it returns is
stored in N, and the process is repeated for the next thread and so on. Because this is a single-
threaded program, only one thread will be executed at a time. If you alter this code to a multi-
threaded program, you will face issues.

Topic 137
How to write thread Safe Code?
 Guide lines for writing thread safe code to ensure that the code runs smoothly in a threaded
environment. When more than one thread can run the same code without introducing
synchronization issues, it is said to be thread safe.

Variables that are required locally should not be accessible globally. They should be placed
on Stack or in Data Structure passed to thread or the thread TLS. If a function is being used
by several thread and it contains a variable that is thread specific such as a counter than
store it in TLS or thread dedicated DS. Do not store it in global memory.

Avoid race conditions. If some required variables are uninitialized then create suspended
threads until variables are initialized. If some condition needs to be met before a code block
is executed, then make sure the condition is met by waiting on synchronization objects.

Thread should not change the process environment. Changing environment by one thread
will affect other threads. Thread should not change standard input or output devices. Also
it should not change environment variables. A primary thread may change process
environment as an exception. In that case the same environment used by rest of threads.

As a principle primary thread ensures that no other thread changes the process
environment. All variable that are meant to be shared among threads are either kept global
or static. They are protected by synchronization or interlocked mechanism using memory
barriers.

Topic 138

Thread Synchronization Objects

Windows support different types of synchronization objects. The objects can be used to
enforce synchronization and mutual exclusion.

Synchronization Mechanism

One method that has been used is by WaitForSingleObject() or WaitForMultipleObjects().

Using this method, a thread waits for other threads to terminate. Other method used File
Locks when multiple processes tries to access a file. Windows provides four different
synchronization objects i.e. Mutexes, Semaphores, Events and Critical Section.

Risks in Using Synchronization Objects

There are always inherent risks involved in the use of such objects such as deadlocks. Care
is required while using these objects. In higher version of Windows other objects like SRW
locks and Condition variables are also available. Other advanced objects are waitable
timers and IO completion ports.

Topic 139
CRITICAL_SECTION Objects

Critical Section

Critical section is the part of the code that can only be executed by one thread at a time. Windows
provides the CRITICAL_SECTION objects as a simple lock mechanism for solving critical
section problem. CRITICAL_SECTION Objects are initialized and deleted but do not have
handles and are not shared among processes. Only one thread at a time can be in the CS variable,
although many threads may enter and leave CS at numerous instances.

The function used to initialize and delete CRITICAL_SECTION are

VOID InitializeCriticalSection(LPCRITICAL_SECTION lpCriticalSection)

VOID DeleteCriticalSection(LPCRITICAL_SECTION lpCriticalSection)

When a thread wants to enter into the critical section it must call EnterCriticalSection() and when
it leaves it should call LeaveCriticalSection().

A call to LeaveCriticalSection() must match an EnterCriticalSection(). Multiple threads may call

EnterCriticalSection() but only one thread is allowed to enter the critical section while the rest are
blocked. A blocked thread can enter the critical section when LeaveCriticalSection() is invoked.

VOID EnterCriticalSection(LPCRITICAL_SECTION lpCriticalSection)

VOID LeaveCriticalSection(LPCRITICAL_SECTION lpCriticalSection)

Recursive Critical Section

If one thread has entered a CS it can enter again. Windows maintain a count. Thread will have to
leave as many times as it enters. This is implemented to support recursive functions and to make
shared library functions thread safe. There is no timeout to EnterCriticalSection(). A thread will
remain blocked forever if a matching Leave call is not received. A thread will remain blocked
forever if a matching Leave call is not received.

However, as thread can always poll to see whether another thread owns a CS using the function:

BOOL TryEnterCriticalSection(LPCRITICAL_SECTION lpCriticalSection)

If the function returns TRUE, then it indicates that the current thread now owns the CS. If it returns
false, then it indicates that the CS is owned by some other thread and it is not safe to enter CS.

Since CRITICAL_SECTION is a user space object therefore it has apparent advantages over
other kernel space objects.
Topic 140

CRITICAL_SECTION for Protecting Shared Variables

In this module, we will study how to use protected shared variables in critical section and
how critical section can assist us use protected shared variables.

Using the critical section construct is easy and intuitive. Consider an example of a server
that maintains status related information in shared variables, information like number of
requests, number of responses and requests currently under process. Since such count
variables are shared therefore only one thread can be allowed to modify it at a time.

CRITICAL_SECTION construct can be used easily to ensure this. In this solution also an
intermediate variable is used to emphasize the role of CRITICAL_SECTION.
Topic 141

Protect a Variable with a Single Synchronization Object

In this module, we will discuss a guideline how to protect shared variables, how to use
synchronization object, and what kind of mapping is there between synchronization object and
variables like one to one or one to many.

Single Synchronization Object

All the variables within the critical section must be guarded by a single object. Using different
objects within the same thread or using different objects across numerous threads sharing same
data would be incorrect. Shared variables must be protected by a single object across all threads
for mutual exclusion to work.

Below is given an example of incorrect use of synchronization object. In this example, two
different objects are using the same variable N. Such code may generate incorrect results therefore
share all variables using a single object.
Topic 142
Producer-Consumer Problem

In this module, we will discuss a version of Producer-Consumer problem.

Producer-Consumer Threads

Producer consumer problem is a classical problem in mutual exclusion. It has many versions. Here
we describe a simplistic version. This version clearly shows how to build and protect data
structures for storing objects. Also we discuss how to establish invariant properties of variables
which are always TRUE outside CS. In addition to primary threads there are two more threads: a
producer and a consumer thread. The producer periodically creates a message. The message is
contained in a table. The consumer on request of the user displays the message. The displayed
data must be most recent and no data should be displayed twice. Do not display data while it is
being updated by the producer. Do not display old data. Some of the produced messages may never
be used and maybe lost. This is also like the pipeline model in which a message moves from one
thread to another. The producer also computes a simple checksum of the message. Consumer
makes sure by checking the checksum. If the consumer accesses the table while it is being updated
the table will be invalid. CS ensures that this does not happen. The invariant is that the checksum
is correct for current message contents.

Topic 143
Producer-Consumer

Producer consumer problem is a classical problem in mutual exclusion. Based on limitations and
function described the following program is developed:
/* Chapter 9. simplePC.c */

/* Maintain two threads, a Producer and a Consumer */

/* The Producer periodically creates checksummed mData buffers, */

/* or "message block" which the Consumer displays when prompted */

/* by the user. The consumer reads the most recent complete */

/* set of mData and validates it before display */

#include "Everything.h"

#include <time.h>

#define DATA_SIZE 256

typedef struct MSG_BLOCK_TAG { /* Message block */

CRITICAL_SECTION mGuard; /* Guard the message block structure */

DWORD fReady, fStop;

/* ready state flag, stop flag */

volatile DWORD nCons, mSequence; /* Message block mSequence number */

DWORD nLost;

time_t mTimestamp;

DWORD mChecksum; /* Message contents mChecksum */

DWORD mData[DATA_SIZE]; /* Message Contents */

} MSG_BLOCK;

/* One of the following conditions holds for the message block */

/* 1) !fReady || fStop */

/* nothing is assured about the mData OR */

/* 2) fReady && mData is valid */

/* && mChecksum and mTimestamp are valid */

/* Also, at all times, 0 <= nLost + nCons <= mSequence */

/* Single message block, ready to fill with a new message */

MSG_BLOCK mBlock = { 0, 0, 0, 0, 0 };

DWORD WINAPI Produce (void *);

DWORD WINAPI Consume (void *);

void MessageFill (MSG_BLOCK *);

void MessageDisplay (MSG_BLOCK *);

int _tmain (int argc, LPTSTR argv[])

DWORD status;

HANDLE hProduce, hConsume;

/* Initialize the message block CRITICAL SECTION */

InitializeCriticalSection (&mBlock.mGuard);

/* Create the two threads */

hProduce = (HANDLE)_beginthreadex (NULL, 0, Produce, NULL, 0, NULL);

if (hProduce == NULL)

ReportError (_T("Cannot create Producer thread"), 1, TRUE);

hConsume = (HANDLE)_beginthreadex (NULL, 0, Consume, NULL, 0, NULL);

if (hConsume == NULL)

ReportError (_T("Cannot create Consumer thread"), 2, TRUE);

 /* Wait for the Producer and Consumer to complete */

status = WaitForSingleObject (hConsume, INFINITE);

if (status != WAIT_OBJECT_0)

ReportError (_T("Failed waiting for Consumer thread"), 3, TRUE);

status = WaitForSingleObject (hProduce, INFINITE);

if (status != WAIT_OBJECT_0)

ReportError (__T("Failed waiting for Producer thread"), 4, TRUE);

DeleteCriticalSection (&mBlock.mGuard);

_tprintf (_T("Producer and Consumer threads have terminated\n"));

_tprintf (_T("Messages Produced: %d, Consumed: %d, Lost: %d.\n"),

mBlock.mSequence, mBlock.nCons, mBlock.mSequence - mBlock.nCons);

return 0;

DWORD WINAPI Produce (void *arg)

/* Producer thread - Create new messages at random intervals */

srand ((DWORD)time(NULL)); /* Seed the random # generator */

while (!mBlock.fStop) {

/* Random Delay */

Sleep(rand()/100);
 /* Get the buffer, fill it */

EnterCriticalSection (&mBlock.mGuard);

__try {

if (!mBlock.fStop) {

mBlock.fReady = 0;

MessageFill (&mBlock);

mBlock.fReady = 1;

InterlockedIncrement (&mBlock.mSequence);

__finally { LeaveCriticalSection (&mBlock.mGuard); }

return 0;

DWORD WINAPI Consume (void *arg)

CHAR command, extra;

/* Consume the NEXT message when prompted by the user */

while (!mBlock.fStop) { /* This is the only thread accessing stdin, stdout */

_tprintf (_T("\n**Enter 'c' for Consume; 's' to stop: "));

_tscanf (_T("%c%c"), &command, &extra);

if (command == _T('s')) {
 /* ES not needed here. This is not a read/modify/write.

* The Producer will see the new value after the Consumer returns */

mBlock.fStop = 1;

} else if (command == _T('c')) { /* Get a new buffer to Consume */

EnterCriticalSection (&mBlock.mGuard);

__try {

if (mBlock.fReady == 0)

_tprintf (_T("No new messages. Try again later\n"));

else {

MessageDisplay (&mBlock);

mBlock.nLost = mBlock.mSequence - mBlock.nCons + 1;

mBlock.fReady = 0; /* No new messages are ready */

InterlockedIncrement(&mBlock.nCons);

__finally { LeaveCriticalSection (&mBlock.mGuard); }

} else {

_tprintf (_T("Illegal command. Try again.\n"));

return 0;

void MessageFill (MSG_BLOCK *msgBlock)

{

/* Fill the message buffer, and include mChecksum and mTimestamp */

/* This function is called from the Producer thread while it */

/* owns the message block mutex*/

DWORD i;

msgBlock->mChecksum = 0;

for (i = 0; i < DATA_SIZE; i++) {

msgBlock->mData[i] = rand();

msgBlock->mChecksum ^= msgBlock->mData[i];

msgBlock->mTimestamp = time(NULL);

return;

void MessageDisplay (MSG_BLOCK *msgBlock)

/* Display message buffer, mTimestamp, and validate mChecksum */

/* This function is called from the Consumer thread while it */

/* owns the message block mutex */

DWORD i, tcheck = 0;

for (i = 0; i < DATA_SIZE; i++)

 tcheck ^= msgBlock->mData[i];

_tprintf (_T("\nMessage number %d generated at: %s"),

msgBlock->mSequence, _tctime (&(msgBlock->mTimestamp)));

_tprintf (_T("First and last entries: %x %x\n"),

msgBlock->mData[0], msgBlock->mData[DATA_SIZE-1]);

if (tcheck == msgBlock->mChecksum)

_tprintf (_T("GOOD ->mChecksum was validated.\n"));

else

_tprintf (_T("BAD ->mChecksum failed. message was corrupted\n"));

return;

Topic 144
Mutexes

In this module we will discuss mutexes. Mutex, a short form of Mutual Exclusion. Windows
provides an object called a mutex. Using this object, we can enforce mutual exclusion.

Mutexes have some advantages beyond CRITICAL_SECTION. Mutexes can be named and have
handles. They can also be used for interprocess communication between threads in different
processes. For example, if two processes share files through memory maps, then mutexes can be
used for protection. Mutexes also allow timeout values. Mutexes can automatically become
signaled once abandoned by the terminating thread. A thread gains ownership to mutex by waiting
successfully on mutex handle using WaitForSingleObject() or WaitForMultipleObject().
Ownership is released using ReleaseMutex(). A thread should be careful about releasing a thread
as soon as possible. A thread can acquire a single mutex several times. A thread will not be blocked
if it already has ownership. The recursive property holds for Mutexes also.

Windows functions used to manage mutexes are

• CreateMutex()

• ReleaseMutex()
• OpenMutex()

If bInitialOwner is TRUE, handover ownership to the calling thread immediately. However, it is

ignored if the mutex already exists as per its name.

lpMutexName is the name of the mutex. The name is assigned as per rules of the Windows
namespace.

If NULL is returned it indicates failure.

OpenMutex() is used to open and existing named mutex. The Open operation is followed by
Create. The main thread would usually create a mutex while other threads can open it. This
construct allows to synchronize threads from different processes.

Similarly ReleaseMutex() releases the ownership of the Mutex for a thread. It fails if the Mutex
is not already owned by the thread.

Topic 145
Mutexes, CRITICAL_SECTIONS and Deadlocks

In this module we will discuss deadlocks. Deadlocks may arise when Mutexes or CRITICAL
SECTIONS are used.

Using Concurrency Objects

Concurrency objects must be used carefully, otherwise it can lead to a deadlock situation.
Deadlocks are a byproduct of concurrency control. It occurs when two or more objects try to lock
a resource at the same time.

An Example
For instance, there are two lists (A and B) with the same structure maintained by different worker
threads. In one situation, it can be possible that an operation is only allowed if a certain element is
either present in both the lists or none. An operation is invalid if an element is present in just one.

In another situation, an element in one list cannot be in the other. Based on these situations,
concurrency objects are required for both lists. Using a single mutex for both lists will degrade
performance by restricting concurrent updates.

Next slide shows an implementation for this situation.

Avoiding deadlock situation

One method to avoid this situation is to maintain a hierarchy. All threads should follow the same
hierarchy. They should acquire and release mutexes in the same order. In the example, the situation
can be easily avoided if both the threads acquire mutexes in the same order.

Another technique that can be used is an array of mutexes and using WaitForMultipleObjects()
with the fWaitAll flag. In this case, the thread will either acquire both the mutexes or none. This
technique is not possible with CRITICAL_SECTION.

Topic 146
Mutexes vs CS

This module is about Mutexes VS CRITICAL_SECTIONs. Mutexes and CS are similar in the
sense that they are used to achieve the same goal. Both can be owned by a single thread, and other
threads trying to gain access to a resource shall be denied access until the resource is released.
However, there are few advantages of using mutexes against some performance drawbacks.

Advantages of Mutexes

Mutexes that are abandoned due to the abrupt termination of a thread are automatically signaled
so that waiting threads are not blocked permanently. However, abrupt termination of a thread
indicates a serious programming flaw. Mutex waits can time out, whereas CS wait does not. As
they are named, mutexes are shareable over numerous threads in different processes. Threads that
create the mutexes can acquire immediate ownership (slight convenience). However, in most
cases, CSs will work considerably faster.

Topic 147
Semaphores

This module is about semaphores. Semaphore is simply a data structure that is used for
concurrency control.

Semaphore Count

Semaphore maintains a count. A semaphore is in a signaled state when the count is greater than 0.
Semaphore is unsignaled when the count is 0. Threads use the wait function on semaphore. The
count is decremented when a waiting thread is released. The following functions are used:
CreateSemaphore(), CreateSemaphoreEx(), OpenSemaphore(), and ReleaseSemaphore().
lSemMax must be 1 or greater, which is the maximum value of semaphore count.

lSemInitial is the initial value.

0<= lSemInitial<= lSemMax.

A NULL returns in case of failure.

lpPreviousCount gives the count preceding the release.

cReleaseCount gives the count after the release and must be greater than 0. The call will fail and
return FALSE if it would cause the count to exceed the maximum, and the count will remain
unchanged. Also, in this case, the release count will not be valid.

Any thread can release the semaphore, not just the one that acquired its ownership. Hence, there
is no concept of abandonment.

Topic 148
Using Semaphores

This module is about using semaphores. We have already discussed semaphores. Now we will
discuss how to use the structure and functions in a program to enforce mutual exclusion.

Semaphore Concepts

Classically, a semaphore count represents the number of available resources. Sem Maximum
represents the number of resources. In a producer-consumer scenario, the producer would place an
element in the queue and call ReleaseSemaphore(). The consumer will wait on the semaphore and
decrease the count after consuming the item.
Thread Creation

Previously, in many examples, numerous threads were created in a suspended state and were not
activated until all the threads were created. This problem can be handled by semaphores without
suspending threads. A newly created thread will wait on a semaphore with the semaphore
initialized by 0. The boss thread will call ReleaseSemaphore with the count set to the number of
threads.

Topic 149
Semaphore Limitation

In this module we will discuss limitations of semaphore.

Limitations

Some limitations are encountered while using windows semaphore. How can a thread request to
decrease the count by 2 or more? The thread will have to wait twice on the semaphore. Calling
wait twice will not be atomic and execution may switch in between.

Below is given a code of two threads, thread1 on the left side and thread2 on the right side. The
problem with the below code is that the wait functions are not atomic; they are separate wait
functions. Because they are not single functions, switching between these two wait functions (after
completion of first wait function) in thread1 may occur, and execution control may reach the
thread2 wait function, resulting in a deadlock-like situation. So this is one of the limitations of
semaphores.

How can we deal with this limitation? Make the two wait functions atomic by using a
CriticalSection to avoid switching during the wait, as shown in the below code.
Other Solutions

Another solution one can suggest is the use of WaitForMultipleObjects() with an array holding
multiple references to the same semaphore. However, this solution will fail readily as the call to
WaitForMultipleObjects() fails when it detects duplicate objects in the array. Secondly, all the
handles may get signals even if the count is 1.

Topic 150
Events

In this module, we will discuss Events. Events are synchronization objects like Mutexes,
Semaphores and Critical Section that can be used for concurrency control.

Events can indicate to other threads that a specific condition now holds, such as some message
being available. Multiple threads can be released from wait simultaneously when an event is
triggered. Events are classified as either manual-reset or auto-reset. Event property is set using
CreateEvent(). A manual-reset event can signal several waiting threads simultaneously and can be
reset. An auto-reset event signals a single waiting thread, and the event is reset automatically. The
functions used to handle events are CreateEvent(), CreateEventEx(), OpenEvent(), SetEvent(),
ResetEvent() and PulseEvent().

The function creates an event. bManualReset is set to TRUE to set a manual reset event. If
bInitialState is TRUE, the event is set to a signaled state. A named event can be opened using
OpenEvent() from any process.
The above three functions are used to control events. A thread can signal the event by using
SetEvent(). In the case of auto-reset, a single thread is released out of many. Event automatically
returns to a non-signaled state. If no threads are waiting, the event remains in a signaled state until
a thread waits on it. In this case, the thread will be immediately released.

If the event is manual-reset, it remains signaled until a thread explicitly calls ResetEvent().
Meanwhile, all the waiting threads are released. Consequently, other threads may also wait to be
released immediately before the reset.

PulseEvent() releases all threads currently waiting for a manual reset. The event is automatically
reset. In the case of an auto-reset event, PulseEvent() releases a single waiting thread, if any.

Topic 151
Event Usage Models

In this module, we will discuss Event Usage Models. Event can be implemented using different
types of models. These models are designed according to the usage.

Combinations

There can be four distinct ways to use events. These four models culminate from combinations of
SetEvent(), PulseEvent() and use of auto and manual events. Each combination proves useful
depending on the situation. The combinations are highlighted in the table.
Understanding Events

An auto-reset event can be conceptualized as a door that automatically shuts when opened.
Whereas the manual reset does not shut automatically when opened. Hence, PulseEvent() can be
considered as a door that is open which shuts when a single thread passes through in the case of
auto-reset. While in the case of a manual reset, multiple waiting threads can pass through.
SetEvent(), on the other hand, simply opens the door and releases a thread.

Topic 152
Producer-Consumer Solution Using Events

In this module, we will try to solve Producer-Consumer problem using Events.

Producer Consumer

Here is another example that provides a solution to our Producer-Consumer problem using Events.
The solution uses Mutexes rather than CSs. A combination of auto-reset and SetEvent() is used in
the producer to ensure only one thread is released. Mutexes in the program make access to the
message data structure mutually exclusive while an event is used to signal the availability of new
messages.

#include "Everything.h"

#include <time.h>

#define DATA_SIZE 256

typedef struct MSG_BLOCK_TAG { /* Message block */

HANDLE mGuard; /* Mutex to uard the message block structure*/

 HANDLE mReady; /* "Message ready" auto-reset event */

DWORD fReady, fStop; /* ready state flag, stop flag */

volatile DWORD mSequence; /* Message block mSequence number*/

volatile DWORD nCons, nLost;

time_t mTimestamp;

DWORD mChecksum; /* Message contents mChecksum*/

DWORD mData[DATA_SIZE]; /* Message Contents*/

} MSG_BLOCK;

/*One of the following conditions holds for the message block */

/*1) !fReady || fStop*/

/*nothing is assured about the mData OR */

/*2) fReady && mData is valid*/

/* && mChecksum and mTimestamp are valid*/

/* Also, at all times, 0 <= nLost + nCons <= mSequence*/

/* Single message block, ready to fill with a new message */

MSG_BLOCK mBlock = { 0, 0, 0, 0, 0 };

DWORD WINAPI Produce (void *);

DWORD WINAPI Consume (void *);

void MessageFill (MSG_BLOCK *);

void MessageDisplay (MSG_BLOCK *);

int _tmain (int argc, LPTSTR argv[])

DWORD status;

HANDLE hProduce, hConsume;

 /* Initialize the message block mutex and event (auto-reset) */

mBlock.mGuard = CreateMutex (NULL, FALSE, NULL);

mBlock.mReady = CreateEvent (NULL, FALSE, FALSE, NULL);

/* Create the two threads */

hProduce = (HANDLE)_beginthreadex (NULL, 0, Produce, NULL, 0, NULL);

if (hProduce == NULL)

ReportError (_T("Cannot create producer thread"), 1, TRUE);

hConsume = (HANDLE)_beginthreadex (NULL, 0, Consume, NULL, 0, NULL);

if (hConsume == NULL)

ReportError (_T("Cannot create consumer thread"), 2, TRUE);

/* Wait for the producer and consumer to complete */

status = WaitForSingleObject (hConsume, INFINITE);

if (status != WAIT_OBJECT_0)

ReportError (_T("Failed waiting for consumer thread"), 3, TRUE);

status = WaitForSingleObject (hProduce, INFINITE);

if (status != WAIT_OBJECT_0)

ReportError (__T("Failed waiting for producer thread"), 4, TRUE);

CloseHandle (mBlock.mGuard);

CloseHandle (mBlock.mReady);

_tprintf (_T("Producer and consumer threads have terminated\n"));

_tprintf (_T("Messages produced: %d, Consumed: %d, Known Lost: %d\n"),

mBlock.mSequence, mBlock.nCons, mBlock.mSequence - mBlock.nCons);

 return 0;

DWORD WINAPI Produce (void *arg)

/* Producer thread - Create new messages at random intervals */

srand ((DWORD)time(NULL)); /* Seed the random # generator */

while (!mBlock.fStop) {

/* Random Delay */

Sleep(rand()/5); /* wait a long period for the next message */

/* Adjust the divisor to change message generation rate */

/* Get the buffer, fill it */

WaitForSingleObject (mBlock.mGuard, INFINITE);

__try {

if (!mBlock.fStop) {

mBlock.fReady = 0;

MessageFill (&mBlock);

mBlock.fReady = 1;

InterlockedIncrement(&mBlock.mSequence);

SetEvent(mBlock.mReady); /* Signal that a message is ready. */

__finally { ReleaseMutex (mBlock.mGuard); }

 }

return 0;

DWORD WINAPI Consume (void *arg)

DWORD ShutDown = 0;

CHAR command[10];

/* Consume the NEXT message when prompted by the user */

while (!ShutDown) { /* This is the only thread accessing stdin, stdout */

_tprintf (_T ("\n**Enter 'c' for Consume; 's' to stop: "));

_tscanf_s (_T("%9s"), command, sizeof(command)-1);

if (command[0] == _T('s')) {

WaitForSingleObject (mBlock.mGuard, INFINITE);

ShutDown = mBlock.fStop = 1;

ReleaseMutex (mBlock.mGuard);

} else if (command[0] == _T('c')) { /* Get a new buffer to consume */

WaitForSingleObject (mBlock.mReady, INFINITE);

WaitForSingleObject (mBlock.mGuard, INFINITE);

__try {

if (!mBlock.fReady) _leave; /* Don't process a message twice */

/* Wait for the event indicating a message is ready */

MessageDisplay (&mBlock);

InterlockedIncrement(&mBlock.nCons);
 mBlock.nLost = mBlock.mSequence - mBlock.nCons;

mBlock.fReady = 0; /* No new messages are ready */

__finally { ReleaseMutex (mBlock.mGuard); }

} else {

_tprintf (_T("Illegal command. Try again.\n"));

return 0;

void MessageFill (MSG_BLOCK *msgBlock)

/* Fill the message buffer, and include mChecksum and mTimestamp */

/* This function is called from the producer thread while it */

/* owns the message block mutex*/

DWORD i;

msgBlock->mChecksum = 0;

for (i = 0; i < DATA_SIZE; i++) {

msgBlock->mData[i] = rand();

msgBlock->mChecksum ^= msgBlock->mData[i];

}
 msgBlock->mTimestamp = time(NULL);

return;

void MessageDisplay (MSG_BLOCK *msgBlock)

/* Display message buffer, mTimestamp, and validate mChecksum */

/* This function is called from the consumer thread while it */

/* owns the message block mutex */

DWORD i, tcheck = 0;

TCHAR timeValue[26];

for (i = 0; i < DATA_SIZE; i++)

tcheck ^= msgBlock->mData[i];

_tctime_s (timeValue, 26, &(msgBlock->mTimestamp));

_tprintf (_T("\nMessage number %d generated at: %s"),

msgBlock->mSequence, timeValue);

_tprintf (_T("First and last entries: %x %x\n"),

msgBlock->mData[0], msgBlock->mData[DATA_SIZE-1]);

if (tcheck == msgBlock->mChecksum)

_tprintf (_T("GOOD ->mChecksum was validated.\n"));

else

_tprintf (_T("BAD ->mChecksum failed. message was corrupted\n"));

return;
}

Topic 153
Windows operating system is a multithreaded kernel that provides support for
real time applications and multiprocessors.

Program design and performance can be simplified and improved by threads,

but it requires much care to use threads in case of shared resources at the
same time.

Process Synchronization is a way to coordinate processes that use shared

data

To synchronize access to a resource we use the following synchronization

objects.

a)-Critical Section

b)-Semaphore

c)-Mutex

d)-Event

Topic 154
To minimize the concurrency related issues and errors following are the main
guidelines

1)-The first issue is timeout when you call a wait on any object then you must
specify the timeout otherwise the thread will be blocked for an indefinite
period, so a programmer should take care of it.

2)- When a thread is in the critical section (CS) and it terminates without
releasing the critical section then it will be impossible for the other thread in
the queue to complete its execution and be blocked.

3)-Mutexes have a way out for the above problem in critical section, if a
thread or process owns a mutex and it terminates, then mutex has an
abandonment property, abandoned mutex handles are signaled, it is a useful
feature not available with CSs. It abandons the process and other threads are
not blocked and fail as in CS case. Mutex abandonment of a process shows a
flaw in the programming.
 4)- If you used the construct of WaitForSingleObject() and also specified the
timeout then you must program in such a way that it checks if the wait ends
due to timeout then it must release the critical resource.

5)-Exactly one of the waiting threads at a time should be given the ownership
of mutex. Only the OS Scheduler decides which thread has the priority
according to its scheduling policy. Program should not assume the priority of
any particular thread over the other.

6)- A single mutex can be used to define several critical regions for several
threads which are competing for the resource.

7)-To ensure optimization critical Region size should be minimum.

Programmers must ensure that critical regions cannot be used concurrently.

8)-Performance decrease because of locks so minimize the use of locks

inside the critical region

9)-The same data structure that stores the resource should also be used to
store the mutexes because mutexes correspond to the resources.

10)- Invariant is the property that assures whether you have enforced the
concurrency correctly or not.

12)- Complex conditions and decision structures should be avoided for

entering into the critical region. Each critical region must have one entry and
one exit.

13)- Must ensure that mutex is locked on entry and unlocked on exit.

14)- Avoid premature exits from the critical region such

as break, return or goto statements, termination handlers are useful for
protecting against such problems.

15)- If the critical code region becomes too lengthy (longer than one page,
perhaps), but all the logic is required, consider putting the code in a function
so that the synchronization logic will be easy to read and comprehend.

Topic 155
The interlocked functions provide a simple mechanism for synchronizing
access to a variable that is shared by multiple threads. They also perform
operations on variables in an atomic manner. The threads of different
processes can use these functions if the variable is in shared memory.
 The InterlockedIncrement and InterlockedDecrement functions combine
the steps involved in incrementing or decrementing a variable into an atomic
operation.

Interlocked functions are as useful as they are efficient; they are implemented
using atomic machine instructions (for this reason, they are sometimes called
“compiler intrinsic statements”).

Some Interlocked Functions and their details:

InterlockedExchange stores one variable into another.

======================================

Long InterlockedExchange(

LONG volatile *Target

LONG Value

It returns the previous value of *Target and sets its value to Value.

===========================================

Long InterlockedExchangeAdd(

LONG volatile *Addend

LONG Increment

Increment is added to the *addend and the original value of *addend is

returned.

Topic 156
When numbers of threads are running concurrently, there is a competition for
resources, memory is also a resource and threads compete for the memory.
So, we have to devise a way to reduce or optimize this competition to improve
 the performance. When we allocate a memory dynamically it is allocated from
the heap. Threads allocate memory and free memory using malloc ()
and free () functions respectively.

Managing Access to Heap

a)-Each thread that performs memory management can create a Handle to its
own heap using HeapCreate(). Memory allocation is then performed using
using HeapAlloc() and HeapFree() rather than using malloc() and free()

b)- Consider third party APIs for concurrent memory management.

Topic 157
Synchronization can impact on the performance. Following are the main
reasons:

a)- Locking, waiting and even interlocked operations are time consuming

b)- Locking requires kernel operation and waiting is expensive

c)- Only one thread at a time can execute a critical code region, reducing
concurrency and produces almost serialization execution

d)- Many processes when competing for memory and cache can produce
unexpected effects

Topic 158
To enforce synchronization, we have many constructs like Critical Section,
Mutexes, Semaphore. When we implement synchronization with these
constructs there are some trade-offs performance wise.

We arrange data in the memory in such a way to avoid cache conflicts and
allocation are aligned over optimal cache boundaries

Following are the main modifiers used for minimal memory contention

__declpec(align), with this modifier memory is aligned in such a way to

minimize contention

Similarly __align_malloc() and __align_free() functions are used for

allocation/deallocation.

We developed different programs in different seniors

1)-Without using synchronization – NS

2)-Using Critical Section -CS

3)-Using Mutexes-MX

4)-Using Interlocked Function-IN

Following are the main inferences we drew by running the programs using
these scenarios.

We basically compare the performance on the basis of following three

aspects

a)-Real Time

b)-User Time

c)-System Time

Inferences

1)- NS (no synchronization) and IN (interlocked functions) exhibit all most the
same time for this example.

2)-CS (Critical Section) version costs 2 times more than IN

3)-MX(Mutexes) version costs more than 2 to 30 times than IN

4)-Any type of locking (even IN) is more expensive than no locking but we
cannot discard it because of the significance of synchronization.

#Processes StatsMX statsCS

Clock Rate
1, 1.4GHz 55.03 14.15

2, 2.0GHz 93.11 5.30

4, 2.4GHz 118.3 4.34

8, 1.7GHz 262.2 20.2

 5- Mutexes are very slow, and unlike the behavior with CSs, performance
degrades rapidly as the processor count increases. For instance, Table 9–1
shows the elapsed and times (seconds) for 64 threads and 256,000 work
units on 1-, 2-, 4-, and 8-processor systems. CS performance, however,
improves with processor count and clock rate.

Topic 159
Performance Analysis:

Gauging Performance:

Previously different programs were developed that performed a simple task using different concurrency
control objects.

Also, previously a program was developed that reported the real time, user time and system time a
process takes to run. Now we use this program to analyze the running time of our programs.
Inferences:

 NS (no Synchronization) and IN (interlocked functions) exhibit almost the same time for this
example.
 CS (critical section) version costs 2 times more than IN.
 MX (Mutex) version can cost more than 2 to 30 times IN.
 For older version of windows CS was not scalable. For recent versions its very scalable.
 Any type of locking (even IN) is more expensive than no locking. However, synchronization
control has logical significance which cannot be discarded.
  Mutexes are slow and performance degrades as number of processors increase.
 The Table shows the performance of mutexes as the number of processors increase.

Topic 160
While working with synchronization we face the problem of false sharing
contention

Cache has cache lines having different capacities, if you have a large array
with different elements on a single line and if we access any element on that
line then the whole line will be locked.

If on one line we store more than one element then we have to face the
consequences of false sharing contention. If any one element is locked on a
single line then other elements will also be locked and can only be accessed
serially.

False sharing contention occurs when two or more threads on different

processes place variables on the same address line and a change in one
variable will lock the whole cache line.

Solution:
It will be better to align each variable in such a way that each variable is
placed on a different cache line. We have to use this type of alignment in
order to avoid false sharing. Of course when we use this type of alignment it
will be expensive memory wise.
Topic 161
Different synchronization objects have different overheads. Critical Section
overheads are moderate but we can future optimize these overheads by fine
tuning the spin counts.

Critical Section works in user space; it does not work in kernel space and
does not require complicated and/or convoluted system calls while mutexes
invoke kernel calls for locking and unlocking like ReleaseMutex() requires
system calls.

Lock bit is on

===============

If any thread invoke CS with the help of EnterCriticalSection() it test its lock
bit, if lock bit is off ,it mean no other thread is entered in the CS yet, so lock bit
atomically set and operation proceed without waiting hence locking and
unlocking of CS is very efficient and it just require a couple of machine level
instructions

The ID of the thread is stored in the CS data structure and it also stores the
status of the recursive calls, so recursion overheads are also performed in the
critical section.

Lock bit is off

=========

If CS is already locked then after invoking EnterCriticalSection(), the

thread will enter into the tight loop, thread calls tight loop because it involves
all the processes of the multiprocessor system and tight loop will repetitively
check the lock bit whether it is off or on.

Spin count will tell how much time the tight loop repetitively checks the lock
bit. If the lock bit repetitively checked without yielding the processor then it
gives up calling WaitForSingleObject()

-Spin count determines the number of times the loop repeats and it is used in
multiprocessor system

LeaveCriticalSection() turns off the lock bit and also informs the kernel
by ReleaseSemphore() call in case there are any waiting threads.

Topic 162
Spin count is important due to the fact that you can optimize the Critical Section with the help of
spin count. How to set the spin count it will depend on many factors

We can specify spin count when we initialize the Critical Section using the function

InitializeCriticalSectionAndSpinCount(). The main advantage of optimizing the spin count is

to avoid the system calls, when there are less system calls then overheads will be reduced.

We can set spin count initially and as well as dynamically on runtime using the
function SetCriticalSectionSpinCount()

According to MSDN 4000 is the good spin count for heap related functions but it depends, like if
you have a short critical section then small spincount will work optimally.

Spin count should be set as per the dynamics of the application and the number of processes.

Topic 163
Slim reader writer locks are used to optimize the Synchronization and minimizing the
overheads due to synchronization and mutual exclusion.

Lightweight Reader Writer Locks has two modes

a)- Exclusive Mode

b)- Shared Mode

-In Exclusive mode they can be locked and writable

-In shared mode reading is possible and different threads can read the data at a time in
shared mode

SRWs can be used either in Exclusive Mode or in Shared mode but you can not
upgrade or downgrade the Exclusive Mode or Shared Mode so Exclusive mode cannot
be converted into Shared mode and Shared mode cannot be converted to Exclusive
mode once acquired. Threads have to decide the mode before using SRWs whether it
is exclusive or shared.

SRWs are light weight and slim and the size of the associated pointers are 32 bit or 64
bits only. No Kernel objects are associated with SRWs and hence SRWs locks require
minimal resources.

SRWs do not support recursion unlike Critical Section, so SRW is simpler and faster.

The spin count value of SRW is preset optimally and cannot be set manually
It is not imperative to delete the SRW

Nonblocking calls are not associated with SRW.

Topic 164
We use APIs to manage SRWs, Slim Read Write Constructs are light weight.

Following are the SRW APIs

In order to Initialize the SRWs Lock we use

InitializeSRWLock()

Similarly, there are two APIs which can access SRWs in shared mode

AcquiredSRWLockShared() and ReleaseSRWLockShared().

There are also two APIs which are used to access SRWs in Exclusive mode

-AcquireSRWLockExclusive

-ReleaseSRWLockExclusive

Using SRWs

If any thread needs to read the shared data then we get the SRW in shared mode but
the thread to write that data will use SRW in exclusive mode.

SRWs are used just like CS OR MUTEX in exclusive mode and are used in shared
mode if the guarded variable is not changed by thread

Topic 165
The main advantage of the SRWs is the shared mode, in shared mode you can read the
locked variable. The thread which reads the locked variable need not to wait for the
other threads to release it.

- Similarly, SRWs are light weight in implementation and resource requirement.

- The don’t allow recursion because in recursion you have to maintain the stack and it
will require time and more execution

Evaluation SRWs performance by using threads shows that these are almost twice fast
than critical Section