This message is from the T13 list server.
On Sun, 14 Apr 2002 15:16:04 -0600, Pat LaVarre wrote: >This message is from the T13 list server. >An AtapiPio device can ask to copy In an odd count of bytes In, >an AtapiDma device cannot. No, the ATAPI/SCSI CDB tells the device and the host how much data to move. I would expect at the inferface between the application program and the OS drivers there would be some parameter that indicated the size of the application's I/O buffer. I would expect that parameter value to be equal to or greater than the size of the data transfer defined by the the ATAPI/SCSI CDB. Otherwise you have yet another problem: What happens to data the device wants to move that is beyond the end of the application program's I/O buffer? >It is T13 that has denied AtapiDma devices the privilege of >asking to copy In an odd count of bytes. ATA/ATAPI DMA, be it Single Word, Multiword or UltraDMA, never had such a option. I don't recall anyone saying this was a problem before. ATA/ATAPI PIO is not able to move an odd number of bytes. Yes, the last byte might be a pad byte but only the ATAPI/SCSI CDB will tell you that. >How is that not a T13 problem? Pio works, Dma doesn't. It only fails to work when the ATA/ATAPI host side is not properly implemented. All the things you say about Windows device drivers not working correctly only add to the widely known fact that the x86, PCI bus and Windows OSs are just toys that should not be used for high reliability computing. >Who but T13 cares about the distinction between Pio and Dma? Please make a proposal to T13 that defines what you are trying to fix and what the fix is. >Now that you feel you understand me, can you tell me again why >I'm wrong to want Atapi Dma to be as capable of counting bytes as >is Atapi Pio? I absolutely agree that a host should know the number of bytes that were actually transferred in ATA/ATAPI DMA. However, Intel doesn't agree with that. Counting the bytes is a host hardware issue and Intel has so far refused to add a byte (or word) counter to their PCI bus DMA engines. And again, I will note that the old ISA bus MW DMA engines did count words. And again, we have lost host side functionality in the current PCI bus implementations. But your X-to-ATA/ATAPI bridge should be able to implement that counter? Pat, I should add that I have not seen a single email (public or private) that says I am telling you the wrong things. I do make lots of mistakes and if I have done that during these discussions I am sure someone would tell me. Jim, Michael, anyone else? *** Hale Landis *** www.ata-atapi.com ***
