Serial Communications in Win32

点击打开链接

Allen Denver
Microsoft Windows Developer Support

December 11, 1995

Applies to:
   Microsoft® Win32®
   Microsoft Windows®

Summary: Learn how serial communications in Microsoft Win32 is significantly different from serial communications in 16-bit Microsoft Windows. This article assumes a familiarity with the fundamentals of multiple threading and synchronization in Win32. In addition, a basic understanding of the Win32 heap functions is useful to fully comprehend the memory management methods used by the Multithreaded TTY (MTTTY) sample included with this article. (35 printed pages)

Download the MTTTY sample (4918.exe) for this technical article.

Contents

Overview
Introduction
Opening a Port
Reading and Writing
Serial Status
Serial Settings
Conclusion
Bibliography

Overview

Serial communications in Microsoft® Win32® is significantly different from serial communications in 16-bit Microsoft Windows®. Those familiar with 16-bit serial communications functions will have to relearn many parts of the system to program serial communications properly. This article will help to accomplish this. Those unfamiliar with serial communications will find this article a helpful foundation for development efforts.

This article assumes you are familiar with the fundamentals of multiple threading and synchronization in Win32. In addition, a basic familiarity of the Win32 heap functions is useful to fully comprehend the memory management methods used by the sample, MTTTY, included with this article.

For more information regarding these functions, consult the Platform SDK documentation, the Microsoft Win32 SDK Knowledge Base, or the Microsoft Developer Network Library. Application programming interfaces (APIs) that control user interface features of windows and dialog boxes, though not discussed here, are useful to know in order to fully comprehend the sample provided with this article. Readers unfamiliar with general Windows programming practices should learn some of the fundamentals of general Windows programming before taking on serial communications. In other words, get your feet wet before diving in head first. (36 printed pages)

Introduction

The focus of this article is on application programming interfaces (APIs) and methods that are compatible with Microsoft Windows NT and Windows 95; therefore, APIs supported on both platforms are the only ones discussed. Windows 95 supports the Win32 Telephony API (TAPI) and Windows NT 3.x does not; therefore, this discussion will not include TAPI. TAPI does deserve mention, however, in that it very nicely implements modem interfacing and call controlling. A production application that works with modems and makes telephone calls should implement these features using the TAPI interface. This will allow seamless integration with the other TAPI-enabled applications that a user may have. Furthermore, this article does not discuss some of the configuration functions in Win32, such as GetCommProperties.

The sample included with this article, MTTTY: Multithreaded TTY (4918.exe), implements many of the features discussed here. It uses three threads in its implementation: a user interface thread that does memory management, a writer thread that controls all writing, and a reader/status thread that reads data and handles status changes on the port. The sample employs a few different data heaps for memory management. It also makes extensive use of synchronization methods to facilitate communication between threads.

Opening a Port

The CreateFile function opens a communications port. There are two ways to call CreateFile to open the communications port: overlapped and nonoverlapped. The following is the proper way to open a communications resource for overlapped operation:

HANDLE hComm;
hComm = CreateFile( gszPort,  
                    GENERIC_READ | GENERIC_WRITE, 
                    0, 
                    0, 
                    OPEN_EXISTING,
                    FILE_FLAG_OVERLAPPED,
                    0);
if (hComm == INVALID_HANDLE_VALUE)
   // error opening port; abort

Removal of the FILE_FLAG_OVERLAPPED flag from the call to CreateFile specifies nonoverlapped operation. The next section discusses overlapped and nonoverlapped operations.

The Platform SDK documentation states that when opening a communications port, the call to CreateFile has the following requirements:

• fdwShareMode must be zero. Communications ports cannot be shared in the same manner that files are shared. Applications using TAPI can use the TAPI functions to facilitate sharing resources between applications. For Win32 applications not using TAPI, handle inheritance or duplication is necessary to share the communications port. Handle duplication is beyond the scope of this article; please refer to the Platform SDK documentation for more information.
• fdwCreate must specify the OPEN_EXISTING flag.
• hTemplateFile parameter must be NULL.

One thing to note about port names is that traditionally they have been COM1, COM2, COM3, or COM4. The Win32 API does not provide any mechanism for determining what ports exist on a system. Windows NT and Windows 95 keep track of installed ports differently from one another, so any one method would not be portable across all Win32 platforms. Some systems even have more ports than the traditional maximum of four. Hardware vendors and serial-device-driver writers are free to name the ports anything they like. For this reason, it is best that users have the ability to specify the port name they want to use. If a port does not exist, an error will occur (ERROR_FILE_NOT_FOUND) after attempting to open the port, and the user should be notified that the port isn't available.

Reading and Writing

Reading from and writing to communications ports in Win32 is very similar to file input/output (I/O) in Win32. In fact, the functions that accomplish file I/O are the same functions used for serial I/O. I/O in Win32 can be done either of two ways: overlapped or nonoverlapped. The Platform SDK documentation uses the terms asynchronous and synchronous to connote these types of I/O operations. This article, however, uses the terms overlapped and nonoverlapped.

Nonoverlapped I/O is familiar to most developers because this is the traditional form of I/O, where an operation is requested and is assumed to be complete when the function returns. In the case of overlapped I/O, the system may return to the caller immediately even when an operation is not finished and will signal the caller when the operation completes. The program may use the time between the I/O request and its completion to perform some "background" work.

Reading and writing in Win32 is significantly different from reading and writing serial communications ports in 16-bit Windows. 16-bit Windows only has the ReadComm andWriteComm functions. Win32 reading and writing can involve many more functions and choices. These issues are discussed below.

Nonoverlapped I/O

Nonoverlapped I/O is very straightforward, though it has limitations. An operation takes place while the calling thread is blocked. Once the operation is complete, the function returns and the thread can continue its work. This type of I/O is useful for multithreaded applications because while one thread is blocked on an I/O operation, other threads can still perform work. It is the responsibility of the application to serialize access to the port correctly. If one thread is blocked waiting for its I/O operation to complete, all other threads that subsequently call a communications API will be blocked until the original operation completes. For instance, if one thread were waiting for a ReadFile function to return, any other thread that issued a WriteFile function would be blocked.

One of the many factors to consider when choosing between nonoverlapped and overlapped operations is portability. Overlapped operation is not a good choice because most operating systems do not support it. Most operating systems support some form of multithreading, however, so multithreaded nonoverlapped I/O may be the best choice for portability reasons.

Overlapped I/O

Overlapped I/O is not as straightforward as nonoverlapped I/O, but allows more flexibility and efficiency. A port open for overlapped operations allows multiple threads to do I/O operations at the same time and perform other work while the operations are pending. Furthermore, the behavior of overlapped operations allows a single thread to issue many different requests and do work in the background while the operations are pending.

In both single-threaded and multithreaded applications, some synchronization must take place between issuing requests and processing the results. One thread will have to be blocked until the result of an operation is available. The advantage is that overlapped I/O allows a thread to do some work between the time of the request and its completion. If no work can be done, then the only case for overlapped I/O is that it allows for better user responsiveness.

Overlapped I/O is the type of operation that the MTTTY sample uses. It creates a thread that is responsible for reading the port's data and reading the port's status. It also performs periodic background work. The program creates another thread exclusively for writing data out the port.

Note   Applications sometimes abuse multithreading systems by creating too many threads. Although using multiple threads can resolve many difficult problems, creating excessive threads is not the most efficient use of them in an application. Threads are less a strain on the system than processes but still require system resources such as CPU time and memory. An application that creates excessive threads may adversely affect the performance of the entire system. A better use of threads is to create a different request queue for each type of job and to have a worker thread issue an I/O request for each entry in the request queue. This method is used by the MTTTY sample provided with this article.

An overlapped I/O operation has two parts: the creation of the operation and the detection of its completion. Creating the operation entails setting up an OVERLAPPED structure, creating a manual-reset event for synchronization, and calling the appropriate function (ReadFile or WriteFile). The I/O operation may or may not be completed immediately. It is an error for an application to assume that a request for an overlapped operation always yields an overlapped operation. If an operation is completed immediately, an application needs to be ready to continue processing normally. The second part of an overlapped operation is to detect its completion. Detecting completion of the operation involves waiting for the event handle, checking the overlapped result, and then handling the data. The reason that there is more work involved with an overlapped operation is that there are more points of failure. If a nonoverlapped operation fails, the function just returns an error-return result. If an overlapped operation fails, it can fail in the creation of the operation or while the operation is pending. You may also have a time-out of the operation or a time-out waiting for the signal that the operation is complete.

Reading

The ReadFile function issues a read operation. ReadFileEx also issues a read operation, but since it is not available on Windows 95, it is not discussed in this article. Here is a code snippet that details how to issue a read request. Notice that the function calls a function to process the data if the ReadFile returns TRUE. This is the same function called if the operation becomes overlapped. Note the fWaitingOnRead flag that is defined by the code; it indicates whether or not a read operation is overlapped. It is used to prevent the creation of a new read operation if one is outstanding.

DWORD dwRead;
BOOL fWaitingOnRead = FALSE;
OVERLAPPED osReader = {0};

// Create the overlapped event. Must be closed before exiting
// to avoid a handle leak.
osReader.hEvent = CreateEvent(NULL, TRUE, FALSE, NULL);

if (osReader.hEvent == NULL)
   // Error creating overlapped event; abort.

if (!fWaitingOnRead) {
   // Issue read operation.
   if (!ReadFile(hComm, lpBuf, READ_BUF_SIZE, &dwRead, &osReader)) {
      if (GetLastError() != ERROR_IO_PENDING)     // read not delayed?
         // Error in communications; report it.
      else
         fWaitingOnRead = TRUE;
   }
   else {    
      // read completed immediately
      HandleASuccessfulRead(lpBuf, dwRead);
    }
}

The second part of the overlapped operation is the detection of its completion. The event handle in the OVERLAPPED structure is passed to the WaitForSingleObject function, which will wait until the object is signaled. Once the event is signaled, the operation is complete. This does not mean that it was completed successfully, just that it was completed. The GetOverlappedResult function reports the result of the operation. If an error occurred, GetOverlappedResult returns FALSE and GetLastError returns the error code. If the operation was completed successfully, GetOverlappedResult will return TRUE.

Note   GetOverlappedResult can detect completion of the operation, as well as return the operation's failure status.  GetOverlappedResult returns FALSE and GetLastError returns ERROR_IO_INCOMPLETE when the operation is not completed. In addition,  GetOverlappedResult can be made to block until the operation completes. This effectively turns the overlapped operation into a nonoverlapped operation and is accomplished by passing TRUE as the  bWait parameter.

Here is a code snippet that shows one way to detect the completion of an overlapped read operation. Note that the code calls the same function to process the data that was called when the operation completed immediately. Also note the use of the fWaitingOnRead flag. Here it controls entry into the detection code, since it should be called only when an operation is outstanding.

#define READ_TIMEOUT      500      // milliseconds

DWORD dwRes;

if (fWaitingOnRead) {
   dwRes = WaitForSingleObject(osReader.hEvent, READ_TIMEOUT);
   switch(dwRes)
   {
      // Read completed.
      case WAIT_OBJECT_0:
          if (!GetOverlappedResult(hComm, &osReader, &dwRead, FALSE))
             // Error in communications; report it.
          else
             // Read completed successfully.
             HandleASuccessfulRead(lpBuf, dwRead);

          //  Reset flag so that another opertion can be issued.
          fWaitingOnRead = FALSE;
          break;