🚀 go-pugleaf

In article <mats.astrom.24.0@neurofys.uu.se>, mats.astrom@neurofys.uu.se (Mats Astrom) writes:
|> Path: rhbnc!doc.ic.ac.uk!uknet!mcsun!sunic!corax.udac.uu.se!neurofys.uu.se!mats.astrom
|> From: mats.astrom@neurofys.uu.se (Mats Astrom)
|> Newsgroups: comp.os.msdos.programmer
|> Subject: DMA : number of transfer
|> Message-ID: <mats.astrom.24.0@neurofys.uu.se>
|> Date: Thu, 1 Oct 92 08:46:13 GMT-0:08
|> Organization: Department of Clinical Neurophysiology, University of Uppsala
|> Lines: 52
|> NNTP-Posting-Host: mats.neurofys.uu.se
|> 
|> While doing DMA-transfer it's possible to read the number of transfers done
|> by reading a specific register of the DMA-circuit. I've manage to do so but
|> the result seem to be somewhat instable ! The instability is due to reads of
|> the register in the middle of a transfer, isn't it ????
|> Assuming I'm setting up everthing correctly, how do I read the number of
|> transferred bytes to get the correct value every time ?????
 

I expect I'm just being dumb, but why would you expect the number of transfers to
be stable? The whole point of DMA is that it goes on in the background while your
program is executing, and in principle on a PC every other bus cycle can be DMA
so of course its changing. If DMA masked off then in my experience the registers
are stable, so I guess you must have a programming problem.

Incidentally, when DMA is running flat out the mainline code in general gets 
much less than 50% of the CPUtime because wait states are automatically added 
for DMA cycles making them longer than typical program access cycles.

                       Adrian