DOCERR: AllocSelector & FreeSelector Documentation Incomplete

Article translations Article translations
Article ID: 132005 - View products that this article applies to.
This article was previously published under Q132005
Expand all | Collapse all

On This Page


The documentation in the Windows SDK for the AllocSelector and FreeSelector functions does not mention that they can allocate and free multiple selectors. Each function uses the limit (length) of the selector passed to it to determine how many selectors to allocate or free. This article describes how AllocSelector allocates one or more selectors and how FreeSelector frees selectors. It also describes what you must know to use these two functions correctly. The documentation should include all the information given in this article.


AllocSelector(UINT uSelector)

AllocSelector(UINT uSelector) allocates a new selector or array of tiled selectors and copies the base address, limit (length), and access rights of uSelector to the new selector(s). If the limit of uSelector is less than or equal to 64K, a single selector is allocated. If the limit of uSelector is larger than 64K, an array of tiled selectors is allocated such that each selector points to a 64K portion of the limit of uSelector.

If uSelector is NULL, AllocSelector allocates a single, completely uninitialized selector. The selector cannot be used until its base address, limit (length), and access rights have been set. To set the access rights of an unitialized selector, you must call the DOS Protected Mode Interface (DPMI) Set Descriptor Access Rights function (0x09). Furthermore, you must set the selector's table bit and privilege bits by using a bitwise-OR mask of 0x0007 to set the lowest-order three bits of the selector.

You should copy an existing selector instead of allocating an uninitialized selector because it is easier to allow AllocSelector set the newly created selector's access control bits. After creating a copy of an existing selector, use SetSelectorLimit and SetSelectorBase to make the copy point to the desired location.

Selectors allocated by AllocSelector must be freed by FreeSelector because the system does not track or manage them automatically.

FreeSelector(UINT uSelector)

FreeSelector(UINT uSelector) frees either a single selector or an array of tiled selectors depending on the limit of uSelector. Frees one selector for each 64K portion of the limit of uSelector. The selector or array of tiled selectors being freed must have been allocated previously by AllocSelector. Furthermore, the limit of uSelector must be the same as the selector used as a parameter in the call to AllocSelector.

Example Array of Tiled Selectors

An example array of tiled selectors looks like this (assuming that the selector passed to AllocSelector has a limit of 256K):
  Sel 1     Sel 2      Sel 3      Sel 4

   0         64K       128K       192K       256K
   |          |          |          |          |
   |          |          |          |--------->| (Limit = 64K)
   |          |          |-------------------->| (Limit = 128K)
   |          |------------------------------->| (Limit = 192K)
   |------------------------------------------>| (Limit = 256K)
Notice that each successive selector's base address starts 64K from the previous selector's base address and has a limit that is 64K less than the previous selector. What really makes these selectors tiled is that they are contiguous in the local descriptor table (LDT). For example, if Sel 1 has a value of 0x97, then Sel 2 will be 0x9F, Sel 3 will be 0xA7, and Sel 4 will be 0xAF.

Only Sel 1 is used to create a huge pointer to the block. As you increment through the block, the compiler generates the correct code to switch from Sel 1 to Sel 2 to Sel 3 to Sel 4 automatically. The code listed in the "Code Sample Two" section of this article demonstrates this.

Code Sample One

This example shows how to allocate exactly one data selector. Code segments in 16-bit code are always less than 64K, so you can ensure that you don't allocate multiple selectors by allocating a copy of a code selector and converting it into a data selector:
     UINT codeSelector, dataSelector;

   _asm {
        mov ax, cs
        mov codeSelector, ax

   dataSelector = AllocSelector(codeSelector);
   if (!dataSelector)
      return NULL;

   // Change dataSelector from a code selector into a data selector
   if (PrestoChangoSelector(codeSelector, dataSelector))
      // Set the desired base address and limit
      SetSelectorBase(dataSelector, dwLinearBase);
      DPMISetSelectorLimit(dataSelector, dwLimit);
      // If you get here, you couldn't change dataSelector so you need
      // to free it because you can't use it as a code selector
      return NULL;

   // You now have a single data selector. Use it, and then free it by
   // calling FreeSelector.

Code Sample Two

This example shows how to automatically allocate an array of tiled selectors that points to a specified region of memory, and obtain a huge pointer from the selector array:
    // The application creates, uses, and frees a huge pointer here
   char __huge * hpMem;

   hpMem = CreateHugePointer (dwBaseAddress, dwLength);

   // hpMem now points to a region of pre-allocated memory, such
   // as a memory-mapped hardware device's buffer. Use it as you
   // would any pointer.

   FreeHugePointer (hpMem);

// The following three functions show how to allocate an array
// of tiled selectors that point to a specified region of memory.

  This function creates a huge pointer with the proper number
  of selectors to access physical memory. The huge pointer may be
  used by C or C++ code. dwLinearBase is a 32-bit linear address,
  and dwLength is the number of bytes that the huge pointer will
  be able to access. This function returns the huge pointer if it
  succeeds, or it returns NULL if it fails.

void __huge * CreateHugePointer (DWORD dwLinearBase,
                                 DWORD dwLength)
   WORD tempSelector = NULL;
   WORD codeSelector = NULL;
   WORD dataSelector = NULL;
   DWORD dwLimit;

     A segment's limit is defined as the last accessible offset in
     the segment. Because the limit is the last accessible offset, it
     is the desired length of the segment minus 1. For example, if
     you want a 64K segment, then you need a limit of 0xFFFF, not
     0x10000, because the segment contains byte offsets 0 to 0xFFFF.
     Note that a segment with a limit of 0 is actually a single byte
     in length. Thus, this function considers a length of zero
   if (dwLength == 0)
      return NULL;

   dwLimit = dwLength - 1;
     Allocate a single temporary selector by making a copy of the
     code segment selector and converting the copy to a data
     selector. Code segments are always less than or equal to
     64K in length, so you are guaranteed to get a single temporary
     selector and be sure to free a single selector.

     Once you have the temporary selector, set its base address and
     limit to the desired values, which may be larger than 64K.
     Because the memory must be accessed by 16-bit code, you must
     allocate an array of tiled selectors. The temporary selector is
     used to force AllocSelector to allocate an array of the proper
     number of tiled selectors, each with the proper base and limit.
     Then you can free the single temporary selector.

     If you fail anywhere along the way, clean up whatever has been
     done, and return NULL.

   _asm {
        mov ax, cs
        mov codeSelector, ax

   tempSelector = AllocSelector (codeSelector);
   if (!tempSelector)
      return NULL;

     If you can successfully change the tempSelector into a
     data selector, set its base address and limit to the
     desired base and limit, and then allocate the real selector
     array. Otherwise, prepare to return NULL.

     SetSelectorLimit does not handle the granularity bit of
     selectors properly, which limits its usefulness only to ranges
     of addresses less than 1MB in length. This function calls
     DPMISetSelectorLimit, a function defined below, to overcome
     this limitation and allow you to create arrays of tiled
     selectors that can access more than 1MB.
   if (PrestoChangoSelector (codeSelector, tempSelector))
      SetSelectorBase(tempSelector, dwLinearBase);
      DPMISetSelectorLimit(tempSelector, dwLimit);
      dataSelector = AllocSelector(tempSelector);
      dataSelector = NULL;

   // Clean up temp selector
   DPMISetSelectorLimit(tempSelector, 0L);

   // dataSelector will be NULL if it could not be allocated
   // successfully, making this function return NULL.
   return (void __huge *)MAKELONG(0, dataSelector);

  This function frees pointers allocated by CreateHugePointer.
  It correctly frees all tiled selectors created to access the
  block of physical memory. It is very important that you call
  this on all pointers created by CreateHugePointer and that you
  do not call this function on pointers allocated by any way
  other than using CreateHugePointer.

void FreeHugePointer (void __huge * hPtr)
   if (hPtr)
      FreeSelector (HIWORD(hPtr));

  This function sets the limit of a selector using DPMI Function
  0008h (Set Segment Limit). This function is necessary if the
  segment size is greater than 1 MB because the Windows
  SetSelectorLimit() API function does not correctly set selector
  limits greater than 1 MB.

  Segments that are larger than 1MB are actually page granular,
  meaning that in the descriptor, the limit field is actually
  stored as the number of 4K pages rather than bytes. When you
  specify a limit greater than 1MB, this function rounds it up
  to the nearest page boundary.

  No matter the size of the segment, this function always accepts
  selector limits in number of bytes, never pages. The conversion
  between bytes and pages is handled internally.

  Note that this function takes a segment limit, which is one less
  than the number of bytes in the segment.

   BOOL DPMISetSelectorLimit (UINT selector, DWORD dwLimit)
      BOOL bRetVal=TRUE;

   // If the limit is >= 1MB, we need to make the limit a mulitple
   // of the page size or DPMISetSelectorLimit will fail.

      if( dwLimit >= 0x100000 )
         dwLimit |= 0x0FFF;

         mov  ax, 0008h
         mov  bx, selector
         mov  cx, word ptr [dwLimit+2]
         mov  dx, word ptr [dwLimit]
         int  31h
         jnc  success
         mov  bRetVal, FALSE
      return bRetVal;

Four Caveats to Keep in Mind

Keep the following in mind when using the suggestions in this article:

  • Allocating selectors does not actually allocate any memory. It merely creates a pointer that can be used to access existing memory (memory previously allocated or provided by a memory-mapped hardware device). Do not confuse allocating selectors with allocating memory.
  • Selectors that alias (point to) a memory block allocated by Windows are not updated if the memory block is moved. To ensure that the memory block is not moved, call GlobalFix on it before creating a selector that aliases it. However, if the allocated selector points to memory provided by a hardware device, there is no need to call GlobalFix because the device's memory was not allocated by Windows.
  • The Windows memory manager does not keep track of which task allocated selectors with these functions, so you must ensure that the task frees them correctly. In particular, make sure it does not free more or fewer selectors than it allocated. The sample code in this article demonstrates the proper way to allocate and free selectors with these functions.
  • Allocating large numbers of selectors is discouraged because selectors are a limited resource.


Article ID: 132005 - Last Review: November 6, 1999 - Revision: 1.0
  • Microsoft Windows Software Development Kit 3.1
kb16bitonly kbdocerr KB132005
Retired KB Content Disclaimer
This article was written about products for which Microsoft no longer offers support. Therefore, this article is offered "as is" and will no longer be updated.

Give Feedback


Contact us for more help

Contact us for more help
Connect with Answer Desk for expert help.
Get more support from