18.13 Using DMA with DJGPP

Q: How do I use DMA with DJGPP programs?

Q: I want to use DMA, but I don't know how to get the physical address of the buffer I allocate for that purpose.

A: The main problem in using DMA with DJGPP is how to get the physical address of a buffer, which is required to program the DMA controller. In protected-mode environment, memory addresses that your program manipulates are actually offsets from the base address of the data segment. You can obtain the base address of the data segment by calling the __dpmi_get_segment_base_address library function and add it to the address of your buffer, but the resulting address is a logical address, translated into a physical address by the memory-mapping unit which is part of the CPU.

You have several alternatives to get the physical address of your buffer:

• Allocate the buffer in conventional memory, below the 1MB mark. This memory is mapped 1:1 by all DPMI servers, so the linear address is equal to the physical one. You can allocate a buffer in conventional memory using the library function __dpmi_allocate_dos_memory.

  This method has a disadvantage of using conventional memory which is at a premium, and is therefore generally ill-suited for large DMA buffers.

• Use VDS, the Virtual DMA Services API. This is implemented by a bunch of functions of interrupt 4Bh; see Ralf Brown's Interrupt List for the details. The VDS method has a drawback that it needs a real memory manager, such as EMM386 or QEMM, to run, since only memory managers and Windows support the VDS API (except that Windows/NT doesn't, or so I'm told). In other words, if you use VDS, your program won't work on a system where CWSDPMI³⁸ is used as the DPMI server, allocating memory by raw XMS calls or via HIMEM.

  The following snippet tests bit 5 of the BIOS data area at 0040:007B, to see whether the VDS API is supported:

  #include <go32.h>
  #include <sys/farptr.h>
  
   int vds_available = (_farpeekb (_dos_ds, 0x0047b) & 0x20) != 0;
  

  (However, an explicit call to the VDS API, after the test above shows that the bit is set, is a more reliable way to detect the VDS support.)

  To use VDS, you allocate a buffer, then use the Lock DMA Buffer Region function (AX=8103h) to get its physical address. Then you call the Disable DMA Translation function (AX=810Bh), program the DMA controller with the physical address returned by the Lock DMA Region function, and start the transfer. After the transfer, you need to call the Enable DMA Translation (AX=810Ch) and Unlock DMA Buffer Region (AX=8104h) functions.

  Alternatively, you could use the Request DMA Buffer and Release DMA Buffer services, but then you will need to copy the data to and from the DMA buffer (e.g. using the movedata function).

  The VDS method is convenient when your program needs to work in several different environments, such as both DOS and Windows, and if you don't want to waste the conventional memory. Experience shows, however, that VDS is inconvenient for buffers larger than 128KB, because many implementations of VDS fail for large buffers, in particular in plain DOS. If you need large DMA buffers, use the XMS method, described next.

• Use XMS, the extended memory allocation API. The XMS services are invoked by calling the XMS driver entry point, which is returned in ES:BX by function 4310h of the software interrupt 2Fh. To call the XMS driver from a DJGPP program, use the library function __dpmi_simulate_real_mode_procedure_retf (don't forget to zero out the SS and SP registers!) passing it the address you got from the 2Fh/4310h call. You need to allocate a buffer with XMS function 09h, then lock that buffer with function 0Ch; the last call returns the 32-bit physical base address of the allocated block, with which you program the DMA controller. You can then map the physical address to a linear address and build a descriptor that spans the entire buffer, in the same manner as you would map a memory-mapped device, see mapping physical address to linear address. This gives you a selector which you can use to copy data between your program and the DMA buffer with movedata and "farptr" functions.

  The XMS method is especially suited to very large DMA buffers, like 2MB (these obviously cannot be allocated in conventional memory). It is also supported by more system configurations used to run DJGPP, since even HIMEM supports XMS.