How to Program DMA Transfers in the Protected Mode Windows EnvLast reviewed: February 27, 1995Article ID: Q124729 |
The information in this article applies to:
SUMMARYThe process for performing DMA transfers in the protected mode Windows environment should be similiar to the process for performing DMA transfers in the real mode MS-DOS environment. In both environments, you must initialize the DMA controller with the starting base address, page address, transfer count, and mode. In both environments, you would usually install a hardware interrupt handler to handle terminal count interrupts when a DMA transfer is complete.
MORE INFORMATIONThe main difference, relevant to DMA transfers, between the real mode MS-DOS environment and the protected mode Windows environment is that with the use of 80386 virtual memory features, a linear address may be mapped to an arbitrary physical address. In addition, a contiguous linear address region that spans page boundaries may be mapped to physical pages that are not physically contiguous. Because the DMA controller bypasses the 80386 page tables and works directly with physical addresses, the DMA controller must be programmed with physical address and the DMA transfer buffer must be physically contiguous. In enhanced mode Windows, the Virtual DMA Device (VDMAD) attempts to hide the virtual memory issues from device drivers that are not fully aware of the virtual memory environment. The main purpose of VDMAD is to convert linear addresses virtually programmed into the DMA controller into physical addresses. This is necessary because Windows applications and DLLs that are not fully aware of the virtual memory environmentonly deal with linear addresses, but the DMA controller only deals with physical addresses. VDMAD virtualizes the DMA controller by trapping all port I/O access to the controller. Windows-based applications and DLLs do not have direct access to the DMA controller through port I/O. They only directly affect the VMDAD maintained virtual state of the DMA controller. The virtual state for each channel includes the mode, base address, and transfer count. When a DMA channel is virtually unmasked, VDMAD physically programs the DMA controller with appropriate values based on the channels virtual state. If the DMA region specified by the virtual base address and transfer count maps to physical memory that is physically contiguous and does not cross a 64K boundary on byte channels or a 128K boundary on word channels, the VDMAD will physically program the DMA controller with the physical base. If the DMA buffer is not physically contiguous, or crosses a 64K or 128K boundary, VDMAD double buffers the DMA tranfer using a physically contiguous DMA buffer that it allocated during its initialization. The default size of this buffer is 16K. It can be increased by using SYSTEM.INI [386Enh] DMABUFFERSIZE=. If the DMA buffer is not physically contiguous and it is larger than the size of the VDMAD buffer, VDMAD will terminate the VM (Virtual Machine) and display this message:
"The DMA buffer is too small. Set DMABUFFERSIZE=xxx in SYSTEM.INI in the [386Enh] section and restart your computer."The Windows 3.1 version of VDMAD does support AutoInit DMA transfers. The major restriction VDMAD places on AutoInit DMA transfers is that the DMA buffer must be physically contiguous. Unless a DMA buffer is allocated by a VxD, a Windows-based application or DLL should not attempt AutoInit transfers using a buffer larger than 4Kb. If you need a DMA buffer larger than 4K that is guaranteed to be physically contiguous, you can use _PageAllocate during the initialization of a VxD. GlobalDosAlloc() does not guarantee anything about the physical memory properties of the memory it allocates. In particular, the memory may not be physically contiguous, and it may be physically located above one megabyte. However, it does guarantee that the linear address of memory is below one megabyte. What this means in relation to not needing to use VDS is that the linear address can be programmed into a virtual 24-bit DMA controller, and then VDMAD can transparently take care of converting the linear address to a physical address. Memory returned by GlobalDosAlloc() is already fixed and page locked so neither GlobalFix() nor GlobalPageLock() are necessary. In enhanced mode Windows, GlobalAlloc() can allocate memory with a linear address above 16 megabytes. In particular, the address should be above two gigabytes if it is not below one megabyte. A linear address above 16 megabytes cannot be programmed into a virtual 24-bit DMA controller. You can use the VDS function Lock DMA Region to convert the linear address to a physical address. Then use the VDS function Disable Address Translation to disable the normal linear->physical address translation. Then program the DMA controller with the physical address of the buffer instead of the linear address. The VDS function Lock DMA Region will page lock a linear address region, but it has no knowledge of Kernel's global heap. If the memory moves in Kernel's global heap, the linear address of the memory will change and it will no longer map to the same physical pages. A DMA buffer allocated using GlobalAlloc() should be allocated with the flags GMEM_FIXED by a DLL, or if allocated by an application GlobalFix() or GlobalPageLock() should be used to keep the memory from moving in the linear address space. In summary, if AutoInit transfers are not used, a DMA buffer can be allocated using GlobalDosAlloc(), in which case the use of VDS should not be necessary, or a DMA buffer can be allocated using GlobalAlloc() (GMEM_FIXED in a DLL, or also use GlobalPageLock() in an application), in which case the use of VDS is necessary. If AutoInit transfers are used, and the buffer size must be larger than 4K, then the DMA buffer must be allocated using _PageAllocate in the initialization of a VxD, and VDS must be used to disable address translation.
|
Additional reference words: 3.10 3.11 VDMAD VDS
© 1998 Microsoft Corporation. All rights reserved. Terms of Use. |