DP Tool Club 8

home *** CD-ROM | disk | FTP | other *** search

/ DP Tool Club 8 / CDASC08.ISO / NEWS / 676 / XMSIF / XMSIF.DOC < prev next >

Wrap

Text File | 1993-10-07 | 44KB | 975 lines

XMSIF C Interface to XMS Functions XMSIF version 1.5 by James W. Birdsall 05/16/93 0. CONTENTS ----------- 0. CONTENTS I. INTRODUCTION I.1 WHAT IS SUPPORTED I.2 COPYRIGHT, LICENSE, AND WARRANTY DISCLAIMER II. COMPILING AND LINKING WITH THE LIBRARIES II.1 WITH C II.2 WITH C++ II.3 XMSTEST, THE EXAMPLE PROGRAM III. PROGRAMMING WITH XMSIF III.1 INITIALIZING THE LIBRARY III.2 ORDINARY USE III.3 TIPS ON USING EMBS III.4 TIPS ON USING UMBS III.5 OTHER TIPS III.6 FUNCTION GROUPINGS IV. LIBRARY REFERENCE IV.1 GLOBAL VARIABLES IV.2 EMB FUNCTIONS IV.3 UMB FUNCTIONS V. ERROR CODES V.1 INTERNAL ERRORS V.2 XMS DRIVER ERRORS VI. THE END VI.1 ACKNOWLEDGEMENTS I. INTRODUCTION --------------- XMSIF provides a high-level interface to XMS control functions for common operations such as allocating and freeing XMS extended memory blocks (EMBs) and upper memory blocks (UMBs), and copying data to and from EMBs. A raw interface to the driver has been included to allow more convenient access to driver functions not otherwise supported. XMSIF is written in assembly language for speed and assembled with Borland's Turbo Assembler (TASM) 2.5. The source code is not compatible with the Microsoft Assembler (MASM). I.1 WHAT IS SUPPORTED --------------------- XMSIF expressly supports the Microsoft eXtended Memory Specification (XMS) versions 2.0 and 3.0. Versions below 2.0 are not supported. Versions above 3.0 are supported as 3.0. XMSIF supports tiny, small, medium, compact, large, and huge memory models. The small model library supports both tiny and small models, so no library is provided specifically for tiny model. XMSIF supports any version of Turbo C, Turbo C++, or Borland C++, in both C and C++ modes, and Microsoft C 6.00/A and C/C++ 7.0. XMSIF has been tested with Borland C++ 2.0, Turbo C 2.0, Microsoft C 6.00A, and Microsoft C/C++ 7.0 in all of the supported memory models. XMSIF should work with earlier versions of Microsoft C and any other MS-DOS C compiler that 1) uses compatible parameter passing and return methods and 2) can use standard-format libraries. I.2 COPYRIGHT, LICENSE, AND WARRANTY DISCLAIMER ----------------------------------------------- XMSIF is not in the public domain. All the files are copyright 1991, 1992, 1993 by James W. Birdsall, all rights reserved. Permission is granted to do the following: You may freely redistribute this archive, so long as it contains all the files listed in the file MANIFEST, intact and unmodified. You may use the libraries in programs for your own use. You may not distribute programs linked with these libraries. Payment of the $5 shareware registration fee ($50 for commercial use) grants the following license, in addition to the permissions listed above: You may request the source to XMSIF. You may modify the source as necessary for use in your programs. However, you may not redistribute either the original or modified source. There is no additional charge for source. You may distribute programs linked with either the original libraries or libraries generated from source you have modified, without royalty, provided you (a) do not alter or remove copyright notices contained therein and (b) you indemnify, hold harmless, and defend the author from and against any claims or lawsuits, including attorney's fees, that arise or result from the use or distribution of your software product. For the purposes of this license, commercial use is defined as use by an incorporated entity in a software product that is regarded as the product of the corporation, no matter how the software product is distributed, but only if 100 or more copies of the product are expected to be made. Registered users will also receive update notices and bug reports, and are entitled to use future versions without further payment. The contents of the distribution archive, and all other related files, information, and services are provided "as is" and without warranty. To the extent permitted by applicable law, the author disclaims all warranties, express or implied, including but not limited to, any implied warranty of merchantability or fitness for a particular purpose. While effort has been made to ensure that the files, information, and services are accurate and correct, the author shall not be liable for damages arising out of the use of or inability to use this product, including but not limited to, loss of profit, data, or use of this software, or special, incidental, or consequential damages or other similar claims, even if the author has been specifically advised of the possibility of such damages. Some states do not allow the exclusion of incidental or consequential damages, so the foregoing limitation may not apply to you. Information on contacting the author is provided at the end of this file. II. COMPILING AND LINKING WITH THE LIBRARIES -------------------------------------------- This section describes how to use the XMSIF libraries with your programs. XMSIF is provided as Borland/Microsoft standard library files. Libraries are provided for small, medium, compact, large, and huge memory models (tiny model uses the small model library). The model for which a library is intended is indicated by the last letter of the filename proper, which is the same as the first letter for the model. For example, XMSIFL.LIB is the large model library. II.1 WITH C ----------- To use XMSIF in C programs, you must #include the file XMSIF.H in every source file that calls XMSIF functions, accesses XMSIF global variables, or uses #defined constants provided by XMSIF. The procedures for linking XMSIF with the rest of your program vary according to the compiler and method you are using. In general, you must include the appropriate library (the library corresponding to the memory model in which you have compiled the rest of your program) in the link. If you are compiling in the Integrated Development Environment of Turbo/Borland C[++], you should include the name of the appropriate library in the project file for your program. For example, if you are working in the compact memory model, you should include XMSIFC.LIB in your project file. If you are using a command-line compiler (bcc, tcc, or cl) to compile and link, simply place the full name of the appropriate XMSIF library on the command line. For example, "bcc -mc foo.c xmsifc.lib" will compile the file "foo.c" in the compact model and link it with xmsifc.lib. If you are linking manually (using TLINK or LINK), place the name of the appropriate library in with the other libraries. For example, tlink c0c.obj foo.obj, foo.exe, foo.map, cc.lib xmsifc.lib or link foo.obj, foo.exe, foo.map, xmsifc.lib ; will link the object "foo.obj" with the appropriate startup object and standard library (LINK automatically includes the startup object and standard library, so it is not necessary to explicitly include them) and the compact model XMSIF library. II.2 C++ -------- To use XMSIF in C++ programs, you must #include the file XMSIF.HPP in every file that calls XMSIF functions, accesses XMSIF global variables, or uses #defined constants provided by XMSIF. Be careful to include XMSIF.HPP instead of XMSIF.H. If you include the wrong one, you will probably see "undefined symbol" errors when linking. Otherwise, the procedures for using XMSIF with C++ programs are the same as for using it with C programs. II.3 XMSTEST, THE EXAMPLE PROGRAM --------------------------------- A large and complete example program and tester, XMSTEST, has been included in this distribution. It can be compiled in any of the supported memory models, with either Borland or Microsoft compilers. The test program has four source files: XMSTEST.C, XMSTEST2.C, XMSTEST3.C, and TESTUTIL.C, and two header files: XMSTEST.H and TESTUTIL.H. When compiling in tiny model with Borland compilers, an additional file, STACK.OBJ, is needed. STACK.ASM, which is the source for STACK.OBJ, has been included for completeness. Example makefiles have been included for Borland and Microsoft compilers. EXMAKEBC is the example makefile for Borland C++ 2.0, EXMAKETC is the example makefile for Turbo C[++], and EXMAKEMS is the example makefile for Microsoft C. Complete instructions for making the example program and using the example makefiles are included at the beginning of each makefile. More information about the program is also included at the beginning of XMSTEST.C. XMSTEST requires a single free EMB of at least 81,920 bytes in order to perform the copy-function tests. It requires at least one free UMB in order to perform all the UMB-function tests. If your system does not support UMBs, XMSTEST will note this fact and skip inappropriate UMB-function tests. If your system does support UMBs but none are free, XMSTEST will assume that your system does not support UMBs and will also skip inappropriate UMB-function tests. XMSTEST also requires at least 165,000 bytes of available conventional memory to run. XMSTEST may require the command-line parameters "-q" or "-w" to run on your system. You may also wish to use the "-3" parameter to perform more complete testing if your system is a 386 or better. A full explanation of these parameters and what they do may be found at the beginning of XMSTEST.C. III. PROGRAMMING WITH XMSIF --------------------------- This section describes how to make XMSIF calls in your program, and details various tricks and tips which you may find useful. III.1 INITIALIZATION -------------------- The library initialization function XMMlibinit() _must_ be called before any other XMSIF calls are made. All other XMSIF functions are guaranteed to fail if called before XMMlibinit(). XMMlibinit() determines whether an XMS driver is present and sets up various internal and global variables necessary to the functioning of XMSIF. III.2 ORDINARY USE ------------------ XMSIF provides several sets of functions. First, there is a set of functions which are intended to be orthogonal with the standard C functions malloc(), realloc(), and free(), and the Borland/Turbo C[++] function coreleft(), for EMBs. These functions are XMMalloc(), XMMrealloc(), XMMfree(), XMMcoreleft(), and XMMallcoreleft(). Second, there is a group of functions for copying data to and from EMBs. These functions are XMMcopyto(), XMMcopyfrom(), _XMMcopy(), XMMicopyto(), XMMicopyfrom(), and _XMMicopy(). Third, there is a group of miscellaneous functions: XMMlibinit(), XMMgetversion(), XMMrawcall(), and XMMraw3(). Finally, there is a set of functions which are intended to be orthogonal with the standard C functions malloc() and free(), and the Borland/Turbo C[++] function coreleft(), for UMBs. These functions are UMBalloc(), UMBfree(), UMBcoreleft(), and UMBallcoreleft(). For details on these functions and all the others listed above, see section IV.2. To use EMBs (Extended Memory Blocks), first it is necessary to allocate some. This is done with XMMalloc(), which takes a size in bytes and returns a handle which will be used to reference the allocated memory. EMBs are subject to fragmentation problems; it is not possible to allocate with XMMalloc() more extended memory than is in the largest free block. The size of the largest free block, in bytes, is returned by XMMcoreleft(). The total of all free EMBs is returned by XMMallcoreleft(). Other considerations about using EMBs are discussed in section III.3 below. To adjust the size of an existing EMB, use XMMrealloc(). The only way to get data to or from an EMB is via the XMS driver's copying functions. The XMSIF copying functions provide handy interfaces to the underlying driver functions and provide extra processing to remove some of the limitations on the driver functions (such as the inability to copy an odd number of bytes). The miscellaneous functions listed above initialize XMSIF (XMMlibinit(), see section III.1), return the XMS version implemented by the driver (XMMgetversion()), and allow programs to call the driver more directly, specifying values for all the relevant registers (XMMrawcall() and XMMraw3()). Finally, to use UMBs (Upper Memory Blocks, which may exist in unused spaces between 640K and 1M), first it is necessary to allocate one (or more). This is done with UMBalloc(), which takes a size in bytes and returns a far pointer to the UMB. UMBs are subject to fragmentation problems; it is not possible to allocate with UMBalloc() more memory than is in the largest free block. The size of the largest free block, in bytes, is returned by UMBcoreleft(). The total of all free UMBs is returned by UMBallcoreleft(). Other problems with using UMBs are discussed in section III.4 below. Since UMBs can be accessed directly, no copying functions are provided for them. Many functions return a status directly. For those functions, a return value of 0 indicates success and a return value of XMMOOPS indicates failure. The global variable _XMMerror contains an error code which gives further information on why the function failed, or has value 0 if the function succeeded. Many functions return some other value instead of a status code (0 or XMMOOPS). In these cases, the value of _XMMerror should be checked upon return. The recommended method of error-checking is indicated for each function in section IV.2 below. III.3 TIPS ON USING EMBS ------------------------ As mentioned above, EMBs are subject to fragmentation problems, and the limit on allocation is the size of the largest free block. This has some interesting side effects. For one, there is no single measure of free extended memory. In some cases, the size of the largest free block will be most useful, in other cases, the total of all free blocks. Thus, both XMMcoreleft() and XMMallcoreleft() are provided. But a consequence of having multiple measures is that allocations may change the values in unanticipated ways. For example, while any allocation will change the total of all free blocks, it may or may not change the size of the largest free block depending on the size of the allocation request and the allocation strategy followed by the XMS driver. For example, if the allocation request is small enough to fit into a smaller free block, the XMS driver may perform the allocation from the smaller block, leaving the size of the largest free block unchanged. Or it may always perform allocations from the largest free block, in which case the size of the largest free block will change. In summary, do not depend on the size of the largest free block changing when you allocate extended memory. Furthermore, not all XMS drivers behave in exactly the same way. While writing XMSTEST, I observed distinct differences between QEMM 5.12 and HIMEM.SYS 2.77. The most obvious difference is that when allocating under QEMM, the total of all free blocks drops by the size of the allocation rounded up to the nearest 16K, instead of the size rounded up to the nearest 1K (the minimum allocation unit for EMBs). This is probably related to QEMM's support of EMS, which has a minimum allocation unit of 16K. XMSTEST's -q parameter is a direct response to this difference in behavior. When allocating from a DOS session under MS-Windows 3.1, the total of all free blocks drops by the size of the allocation rounded up to the nearest 4K. This is probably because the 386 supports memory mapping in units of 4K. XMSTEST's -w parameter is a direct response to this difference in behavior. In summary, do not depend on the total of all free blocks changing by the amount you think it will. Finally, there are some speed considerations when using EMBs. For maximum speed, copy the largest blocks you can. Each copy of an even number of bytes makes one call to the XMS driver. Each copy of an odd number of bytes makes two calls to the XMS driver. And each copy of a single byte requires three calls to the XMS driver in order to reliably copy the desired byte without corrupting surrounding data (remember that the XMS driver itself can only copy even numbers of bytes). To get a good idea of relative speeds of various sizes of copies, run XMSTEST -- it performs some speed tests, returning times in ticks (at 18.2 ticks per second). Large copies may be hundreds of times faster than multiple small copies to move the same number of bytes. If you have to make lots of small copies, try using the interval copy functions (_XMMicopy() et al.). They make the same number of calls to the XMS driver, but the loops are tightly coded in assembly and are marginally faster than performing the same loop yourself. And, as a last consideration, note that different XMS drivers may be faster or slower. Results from XMSTEST showed that QEMM was roughly 40% faster than HIMEM.SYS. III.4 TIPS ON USING UMBS ------------------------ It is probably better not to use UMBs if there isn't a powerful reason to do so. UMBs are not present on many systems, may be all used on other systems, and some XMS drivers have buggy UMB support functions. UMBs are blocks of memory between 640K (the top of conventional memory) and 1M (in general, the top of memory addressable in real mode). Systems based on the 80286 do not have UMBs unless there is special hardware support on the motherboard or elsewhere. Even if the hardware support is present, an XMS driver is also required to provide the UMB management functions. Systems based on the 80386 and up do not require the hardware support, but do require a driver (such as QEMM, EMM386.SYS, 386-to-the-Max, etc.) which uses the processor's on-board memory mapping capability to make extended memory appear between 640K and 1M, in unused holes between video memory, ROMs, and anything else inhabiting that space. Then they may require a separate XMS driver to actually provide the UMB management functions. QEMM and 386-to-the-Max include XMS drivers within themselves; EMM386.SYS requires HIMEM.SYS. Even if the system has the requisite support and drivers, UMBs may not be available to your application. DOS 5.0, for example, depending on the configuration, may allocate all UMBs to itself. Lastly, some XMS drivers have buggy UMB support functions. While testing XMSIF, I discovered that QEMM 5.12 may return a random error code when there are no more free UMBs. In fact, it may not return an error code at all -- it may leave the contents of that register untouched. Various measures to combat this problem have been taken within XMSIF with sufficient success that XMSTEST can perform the UMB-function tests under QEMM 5.12 correctly, but other XMS drivers may have other bugs. In summary, I would advise that you avoid the use of UMBs unless there is a powerful reason to do so, and even then your program should be ready to deal with error returns or unusual results. III.5 OTHER TIPS ---------------- Neither EMBs nor UMBs are deallocated automatically when a program exits. If the program does not deallocate EMBs or UMBs that it has allocated, those EMBs and/or UMBs are stuck, not available to any other program until the machine is rebooted. Under normal circumstances, it is easy enough to free EMBs and UMBs when they are no longer needed. However, an emergency exit due to the user hitting control-break or due to a hardware error (e.g. the famous "Abort, Retry, Fail?") can cause EMBs or UMBs to become stuck unless your program takes special measures to intercept these errors and perform cleanup before exiting. There are a number of ways to do this and they vary from compiler to compiler. Look for functions named things like ctrlbrk(), harderr(), and signal(). If you are copying array elements, XMMicopyto(), XMMicopyfrom(), or _XMMicopy() may be more efficient. These functions allow copying of elements of fixed size which are separated by gaps also of fixed size. For example, if you have an array of integers and wish to copy all the even-indexed elements (a[0], a[2], a[4], etc.), these functions are faster than calling a standard copying function (XMMcopyto(), XMMcopyfrom(), or _XMMcopy()) from within a loop. III.6 FUNCTION GROUPING ----------------------- The XMSIF functions have been arranged in the library in such a way as to reduce the number of unnecessary functions linked into your program. There are currently two groups: FUNCTIONS VARIABLES --------- --------- GROUP 1: XMMlibinit(), XMMgetversion(), _XMMerror, _XMMversion, XMMcoreleft(), XMMallcoreleft(), _xmsif_vers_vers, XMMalloc(), XMMrealloc(), XMMfree(), _xmsif_vers_date, _XMMcopy(), _XMMicopy() _xmsif_vers_time GROUP 2: UMBcoreleft(), UMBallcoreleft(), none UMBalloc(), UMBfree() GROUP 3: XMMrawcall(), XMMraw3() none If your program references any of the functions or variables in a group, all the functions and variables in that group will be linked in. Note that group one will always be linked, since it contains XMMlibinit(), which all XMSIF-using programs must call. IV. LIBRARY REFERENCE --------------------- IV.1 GLOBAL VARIABLES --------------------- _XMMerror --------- unsigned char const _XMMerror; _XMMerror contains the error code from the last XMSIF call. If the call succeeded, _XMMerror will be 0. If the call failed because the XMS driver returned an error, _XMMerror contains the error code returned by the XMS driver. If the call failed because XMSIF detected an error internally, _XMMerror contains an internal error code. A list of error code values, their meanings, and #defined constants for them is in section V. _XMMversion ----------- unsigned int const _XMMversion; _XMMversion contains the XMS version implemented by the XMS driver in packed BCD format. For example, if the XMS version is 2.0, the value of _XMMversion will be 0x0200. The high byte of _XMMversion represents two digits to the left of the decimal point (one digit per nibble) and the low byte represents two digits to the right of the decimal point (again, one digit per nibble). This value is the same as that returned by XMMgetversion(). xmsif_vers_vers, xmsif_vers_date, xmsif_vers_time ------------------------------------------------- char const xmsif_vers_vers[]; char const xmsif_vers_date[]; char const xmsif_vers_time[]; These are null-terminated strings containing information about the name and version of the library, the date of assembly, and the time of assembly, respectively. IV.2 EMB FUNCTIONS ------------------ _XMMcopy() - copy data ---------- int _XMMcopy(unsigned long clen, int shan, unsigned long soff, int dhan, unsigned long doff); int XMMcopyfrom(unsigned long clen, int handle, unsigned long soff, unsigned char far *dest); int XMMcopyto(unsigned long clen, unsigned char far *src, int handle, unsigned long doff); _XMMcopy() allows painless copying from conventional memory to an EMB, from an EMB to conventional memory, from one EMB to another, or even from one point in conventional memory to another. Areas larger than 64K can be copied without special treatment. Clen bytes of data are copied. If shan is nonzero, it is assumed to be the handle of an EMB, and soff is treated as a byte offset into the EMB owned by shan. If shan is zero, soff is treated as a segment:offset far pointer to conventional memory. The pair of shan and soff indicate where the data is to be copied from (the source). Dhan and doff work similarly, but indicate where the data is to be copied to (the destination). This function returns 0 on success or XMMOOPS on error. The specific error may be determined from _XMMerror. If clen is 0, the function returns immediately without error. XMMcopyfrom() is a macro which provides a convenient interface to _XMMcopy() for copying data from an EMB to conventional memory. Clen bytes of data are copied. Handle is the handle of an EMB, and soff is a byte offset into the EMB owned by handle. The pair of handle and soff indicate where the data is to be copied from. Dest is a far pointer to conventional memory and indicates where the data is to be copied to. The returns are the same as for _XMMcopy(). XMMcopyto() is a macro which provides a convenient interface to _XMMcopy() for copying data to an EMB from conventional memory. Clen bytes of data are copied. Src is a far pointer to conventional memory and indicates where the data is to be copied from. Handle is the handle of an EMB, and doff is a byte offset into the EMB owned by handle. The pair of handle and doff indicate where the data is to be copied to. The returns are the same as for _XMMcopy(). _XMMicopy() - copy data by intervals ----------- int _XMMicopy(unsigned long nelem, int elsize, unsigned int sskip, int shan, unsigned long soff, int dhan, unsigned long doff, unsigned int dskip); int XMMicopyfrom(unsigned long nelem, int elsize, unsigned int byteskip, int handle, unsigned long soff, unsigned char far *dest); int XMMicopyto(unsigned long nelem, int elsize, unsigned int byteskip, unsigned char far *src, int handle, unsigned long doff); _XMMicopy() allows painless copying of array elements from conventional memory to an EMB, from an EMB to conventional memory, from one EMB to another, or even from one point in conventional memory to another. Calculation of addresses, skipping of unwanted elements, etc., are all handled by the function. Elements spread across more than 64K or totalling more than 64K in length can be copied without special treatment. Nelem elements are copied, each of which is elsize bytes long. If shan is nonzero, it is assumed to be the handle of an EMB, and soff is treated as a byte offset into the EMB owned by shan. If shan is zero, soff is treated as a segment:offset far pointer to conventional memory. The pair of shan and soff indicate where the data is to be copied from (the source). Dhan and doff work similarly, but indicate where the data is to be copied to (the destination). Sskip and dskip indicate the number of bytes between elements in the source and destination, respectively. If nelem or elsize is zero, the function returns immediately without error. If nelem is one, _XMMcopy() is called internally. Elsize must be in the range 0 through 16384, or the function will return with error XMM_ELTOOBIG (in _XMMerror). Sskip and dskip must both be in the range 0 through 32768, or the function will return with error XMM_SKTOOBIG (in _XMMerror). This function returns 0 on success or XMMOOPS on error. The specific error may be determined from _XMMerror. XMMicopyfrom() is a macro which provides a more convenient interface to _XMMicopy() for copying data from an EMB to conventional memory with sskip equal to dskip. Nelem elements are copied, each of which is elsize bytes long. Handle is the handle of an EMB, and soff is a byte offset into the EMB owned by handle. The pair of handle and soff indicate where the data is to be copied from. Dest is a far pointer to conventional memory and indicates where the data is to be copied to. Byteskip is the number of bytes between elements, the same for both source and destination. The returns are the same as for _XMMicopy(). XMMicopyto() is a macro which provides a more convenient interface to _XMMicopy() for copying data to an EMB from conventional memory with sskip equal to dskip. Nelem elements are copied, each of which is elsize bytes long. Src is a far pointer to conventional memory and indicates where the data is to be copied from. Handle is the handle of an EMB, and doff is a byte offset into the EMB owned by handle. The pair of handle and doff indicate where the data is to be copied to. Byteskip is the number of bytes between elements, the same for both source and destination. The returns are the same as for _XMMicopy(). XMMallcoreleft() - get total of all free EMBs ---------------- unsigned long XMMallcoreleft(void); unsigned long XMMcoreleft(void); XMMallcoreleft() returns the total size of all free EMBs, in bytes. Note that when all EMBs have been allocated, the driver itself returns an error (0xA0, XMM_NOFREEX). This behavior is masked by XMMallcoreleft() -- if there are no free EMBs, XMMallcoreleft() returns 0L, not an error. _XMMerror should be checked after calling this function (it is reset to 0 if there are no free EMBs). XMMcoreleft() returns the size of the largest free EMB, in bytes. Note that when all EMBs have been allocated, the driver itself returns an error (0xA0, XMM_NOFREEX). This behavior is masked by XMMcoreleft() -- if there are no free EMBs, XMMcoreleft() returns 0L, not an error. _XMMerror should be checked after calling this function (it is reset to 0 if there are no free EMBs). XMMalloc() - allocate EMB ---------- int XMMalloc(unsigned long bytes); int XMMrealloc(int handle, unsigned long bytes) XMMalloc() allocates an EMB. It takes the given number of bytes and rounds it up to the nearest kilobyte, then allocates an EMB that many kilobytes in size. It returns the EMB handle assigned by the XMS driver. _XMMerror should be checked after calling this function. Note that EMBs cannot be allocated in units smaller than a kilobyte (1024 bytes). Requesting one byte will allocate 1K; 1025 bytes will allocate 2K. Requesting zero bytes allocates a handle but no memory. Note that it is not possible to allocate in one XMMalloc() call more extended memory than is in the largest free EMB. This is a limitation of the XMS driver. See section III.3 above for more information on EMB allocation limitations. XMMrealloc() adjusts the size of an EMB previous allocated with XMMalloc(). Handle is the handle of the EMB to be adjusted. The given number of bytes is rounded up to the nearest kilobyte, then the EMB is adjusted to that size, if possible. XMMrealloc() returns the handle of the new block, which is always the same as the handle of the old block. _XMMerror should be checked after calling this function. XMMcopyfrom() - copy data from EMB ------------- int XMMcopyfrom(unsigned long clen, int handle, unsigned long soff, unsigned char far *dest); See _XMMcopy(). XMMcopyto() - copy data to EMB ----------- int XMMcopyto(unsigned long clen, unsigned char far *src, int handle, unsigned long doff); See _XMMcopy(); XMMcoreleft() - get size of largest free EMB ------------- unsigned long XMMcoreleft(void); See XMMallcoreleft(). XMMfree() - deallocate an EMB --------- int XMMfree(int handle); This function accepts an EMB handle (as returned by XMMalloc() or otherwise obtained from the XMS driver) and releases it and the EMB associated with it (if any). This function returns 0 on success or XMMOOPS on error. The specific error may be determined from _XMMerror. XMMgetversion() - obtain XMS version implemented by XMS driver --------------- int XMMgetversion(void); This function returns the XMS version implemented by the XMS driver. The version number is in packed BCD format as in the global variable _XMMversion, and is in fact the same value. _XMMerror should be checked after calling this function. XMMicopyfrom() - copy data by intervals from EMB -------------- int XMMicopyfrom(unsigned long nelem, int elsize, unsigned int byteskip, int handle, unsigned long soff, unsigned char far *dest); See _XMMicopy(). XMMicopyto() - copy data by intervals to EMB ------------ int XMMicopyto(unsigned long nelem, int elsize, unsigned int byteskip, unsigned char far *src, int handle, unsigned long doff); See _XMMicopy(). XMMlibinit() - initialize XMSIF ------------ int XMMlibinit(void); XMMlibinit() initializes XMSIF. Any other XMSIF function will return with error XMM_NOINIT (in _XMMerror) if called before XMMlibinit() has been called. This function returns 0 on success, XMMOOPS on error, or NOXMM if no XMS driver is detected. The specific error may be determined from _XMMerror. XMMrawcall() - call XMS driver directly ------------ int XMMrawcall(struct XMMregs *regs); void XMMraw3(struct XMMbigregs *regs); XMMrawcall() is a raw interface directly to the XMS driver, supporting calls which use only 16-bit registers. The XMMregs structure has fields for the AX, BX, DX, SI, and DS registers, which include all the 16-bit registers used as parameters for calls to the XMS driver. The values in the structure are placed in the appropriate registers before the XMS driver is called, and the registers are copied back into the structure after the XMS driver call returns. Check XMSIF.H or XMSIF.HPP for the exact format of the XMMregs structure. Since the XMS specification provides a uniform error indication for all XMS calls, XMMrawcall() is able to determine if the call performed caused an error. It returns 0 on success or XMMOOPS on error. The specific error may be determined from _XMMerror; the error code can also be found in the low byte of the regBX field of the XMMregs structure. XMMraw3() is a raw interface directly to the XMS driver, supporting the calls added in version 3.0 which use 32-bit registers. The XMMbigregs structure has fields for the EAX, EBX, ECX, and EDX regusters, which include all the 32-bit registers used as parameters for calls to the XMS driver. The values in the structure are placed in the appropriate registers before the XMS driver is called, and the registers are copied back into the structure after the XMS driver call returns. Check XMSIF.H or XMSIF.HPP for the exact format of the XMMbigregs sturcture. These calls do not provide a uniform error indication, so XMMraw3() does not return anything. The low 8 bits of EBX (register BL) are copied to _XMMerror, but they may not be meaningful. In general, it is necessary to look at the register contents to determine success or failure. Not all calls use all of each register, so be careful when examining values. If the XMS driver is version 2.0, _XMMerror is set to XMM_BADVERS. Note that this function must not be used on 286 (or earlier) machines. It uses 386 instructions to access the registers, which will probably crash older processors. Furthermore, HIMEM versions 3.03 through 3.07 are known to crash when these calls are made on 286 or earlier machines. It is up to the application to ensure that this call is only used on 386 or better machines. XMMraw3() - call XMS driver directly with extended registers --------- See XMMrawcall(). XMMrealloc() - adjust the size of an EMB ------------ See XMMalloc(). IV.3 UMB FUNCTIONS ------------------ UMBallcoreleft() - get total of all free UMBs ---------------- unsigned long UMBallcoreleft(void); unsigned long UMBcoreleft(void); UMBallcoreleft() returns the total size of all free UMBs, in bytes. Note that when all UMBs have been allocated, the driver itself returns an error (0xB1, UMB_NOFREEUMB). This behavior is masked by UMBallcoreleft() -- if there are no free UMBs, UMBallcoreleft() returns 0L, not an error. If the XMS driver does not support UMBs at all, it returns an error (0x80, XMM_UNIMP). This is also masked by UMBallcoreleft(). Again, 0L will be returned, not an error. _XMMerror should be checked after calling this function (it is reset to 0 when a masked error occurs). UMBcoreleft() returns the size of the largest free UMB, in bytes. Note that when all UMBs have been allocated, the driver itself returns an error (0xB1, UMB_NOFREEUMB). This behavior is masked by UMBcoreleft() -- if there are no free UMBs, UMBcoreleft() returns 0L, not an error. If the XMS driver does not support UMBs at all, it returns an error (0x80, XMM_UNIMP). This is also masked by UMBallcoreleft(). Again, 0L will be returned, not an error. _XMMerror should be checked after calling this function (it is reset to 0 when a masked error occurs). UMBalloc() - allocate UMB ---------- void far *UMBalloc(unsigned long bytes, unsigned long *finalsize); This function allocates an UMB. It takes the given number of bytes and rounds it up to the nearest multiple of 16, then allocates a UMB that many bytes in size. The actual size in bytes of the UMB allocated is returned in the unsigned long pointed to by finalsize. UMBalloc() returns a far pointer to the UMB; this pointer should be cast to a type other than void before being used to access the UMB. _XMMerror should be checked after calling this function. Note that UMBs cannot be allocated in units smaller than a paragraph (16 bytes). Requesting one byte will allocate 16; 17 bytes will allocate 32. Note that it is not possible to allocate in one UMBalloc() call more extended memory than is in the largest free UMB. This is a limitation of the XMS driver. See section III.4 above for more information on UMB allocation limitations. UMBcoreleft() - get size of largest free UMB ------------- unsigned long UMBcoreleft(void); See UMBallcoreleft(). UMBfree() - deallocate an UMB --------- int UMBfree(void far *handle); This function accepts a pointer to a UMB (as returned by UMBalloc()) and releases that UMB. If the pointer has been modified, UMBfree() attempts to return it to canonical form before calling the XMS driver. However, for maximum reliability, UMBfree() should be called with an unmodified copy of the pointer returned by UMBalloc(). This function returns 0 on success or UMBOOPS on error. The specific error may be determined from _XMMerror. V. ERROR CODES -------------- This section is a list of the error codes to which the global variable _XMMerror may be set, and the values and meanings thereof. V.1 INTERNAL ERRORS ------------------- Internal codes indicate an error detected by XMSIF itself. NAME VALUE MEANING ---- ----- ------- XMM_NOINIT 0x40 XMMlibinit() must be called before any other XMSIF function can be called. XMM_UMBHUGE 0x41 Size of UMB requested is larger than XMS driver can handle (maximum UMB request size is 1M). XMM_BADPTR 0x42 The UMB pointer passed to UMBfree() is corrupt. XMM_ELTOOBIG 0x43 The element size passed to XMMicopyto(), XMMicopyfrom(), or _XMMicopy() is too big. XMM_SKTOOBIG 0x44 The skip size passed to XMMicopyto(), XMMicopyfrom(), or _XMMicopy() is too big. XMM_BADVERS 0x45 Certain functions are not supported under XMS version 2.0. V.2 XMS DRIVER ERRORS --------------------- These codes are defined in the XMS specification and are returned by the XMS driver. They are saved in _XMMerror by XMSIF without alteration. NAME VALUE MEANING ---- ----- ------- XMM_UNIMP 0x80 Function is not implemented XMM_VDISK 0x81 VDISK device driver was detected XMM_A20ERROR 0x82 A20 error occurred XMM_GENERROR 0x83 General driver error occurred XMM_UNRECERROR 0x84 Unrecoverable driver error occurred XMM_NOHMA 0x90 HMA (High Memory Area) does not exist XMM_HMAUSED 0x91 HMA already allocated XMM_HMATOOBIG 0x92 Request to allocate HMA denied because amount of HMA requested is less than minimum parameter given to XMS driver on its command line XMM_HMANOALLOC 0x93 HMA is not allocated XMM_A20STILLEN 0x94 A20 line is still enabled XMM_NOFREEX 0xA0 All EMBs (Extended Memory Blocks) are allocated XMM_NOFREEXHAN 0xA1 No free EMB handles XMM_BADXHAN 0xA2 EMB handle is invalid XMM_BADSRCHAN 0xA3 Source EMB handle is invalid XMM_BADSRCOFF 0xA4 Source offset in EMB is beyond end of EMB XMM_BADDESTHAN 0xA5 Destination EMB handle is invalid XMM_BADDESTOFF 0xA6 Destination offset in EMB is beyond end of EMB XMM_BADLENGTH 0xA7 Length is invalid XMM_COPYOVERLAP 0xA8 Overlap in copy request is invalid XMM_PARITY 0xA9 Parity error was detected XMM_NOLOCK 0xAA EMB is not locked XMM_LOCKED 0xAB EMB is locked XMM_TOOMANYLOCKS 0xAC EMB lock count overflowed XMM_LOCKFAIL 0xAD EMB lock failed XMM_UMBSMALLER 0xB0 UMB (Upper Memory Block) of size requested is not available; however, a smaller UMB is available XMM_NOFREEUMB 0xB1 All UMBs are allocated XMM_BADUMBHAN 0xB2 UMB handle (same as segment address of start of UMB) is invalid VI. THE END ----------- Technical support via email is available from the following addresses: INTERNET: The following are alternate addresses for the same place: support@picarefy.com picarefy!support@amc.com picarefy!support@netcom.com uunet!uw-coco!amc-gw!picarefy!support COMPUSERVE: 71261,1731 GENIE: J.BIRDSALL2 Registrations should be sent to: James W. Birdsall 11112 NE 124 LN #D204 Kirkland, WA 98034 If you have an email address on any of the networks listed above, please include it when registering, especially if you are requesting source code. It is much easier to send the source code by email. Also, please specify what sort of archive (ZIP, ZOO, ARC, LZH, ARJ, UNIX shar) you can handle most easily. NOTE: IF YOU DO NOT PROVIDE AN EMAIL ADDRESS, YOU WILL ONLY RECEIVE MAJOR VERSION UPDATES. YOU WILL NOT RECEIVE MINOR VERSIONS. PLEASE PROVIDE AN EMAIL ADDRESS IF YOU HAVE ANY WAY OF DOING SO. VI.1 ACKNOWLEDGEMENTS --------------------- Thanks to Bob Parsons of Parsons Technology Inc. for some good suggestions on the documentation and for providing the original C++ header file.