[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: Legacy Terminology RE: DMA across PCI



[I'd like to offer my input because there still appears to be confusion.]

I often hear the term "DMA" used loosely when referring to a general burst 
across any type of parallel bus. As stated previously, this is not accurate.

To "burst data" across a bus, especially with PCI, has nothing to do with 
the term "Direct Memory Access" unless a controller (like the mentioned 
Galileo) is used which off-loads the actual burst control from the system's 
central processor.

It appears that the concept of a particular controller chip using DMA in a 
PCI application adds confusion for someone wondering if they can "do DMA" 
with the PCI bus.

To clarify:

The PCI bus itself is fundamentally based on burst transfers to optimize 
performance with minimal overhead associated with bus master arbitration.

DMA can be defined as a mechanism based on an independent controller that 
allows burst transfers to occur autonomously among a set of peripherals 
(memories, FIFOs, etc.) for the purpose of off-loading a central processor.

A PCI controller chip with DMA capability provides one solution for bursting 
data between a set of local peripherals to the PCI bus. Other PCI-based 
solutions will typically rely on a PCI bridge chip or a semi-custom design 
such as an FPGA incorporating PCI IP core design.

John Cappello
Optimal Design, Inc.
jcappello@optimal-design.com
856-582-4838


>From: Lame Brooks-G14738 <brooks.lame@motorola.com>
>To: "'ted.firlit@utmc.aeroflex.com'" <ted.firlit@utmc.aeroflex.com>
>CC: "PCISIG List (E-mail)" <pci-sig@znyx.com>
>Subject: RE: Legacy Terminology   RE: DMA across PCI
>Date: Thu, 14 Dec 2000 14:01:19 -0800
>
>I think you just said it yourself, Ted: a "DMA engine."   That's the old 
>ISA
>terminology refering to a state machine that manages data transfers.  I'm
>only pointing out that the term DMA in reference to PCI does not indicate
>any specific implementation nor performance capabilities.  And, because the
>term is so non-specific, it has little meaning to engineers in PCI.  With
>marketing liturature on the other hand, they throw in all the buzz words
>they  can.   :P     -- BrooksL
>
> > -----Original Message-----
> > From: Ted Firlit [mailto:firlit@utmc.aeroflex.com]
> > Sent: Thursday, 14 December, 2000 12:46
> > To: Lame Brooks-G14738
> > Cc: 'Daniel DeConinck'; PCISIG List (E-mail)
> > Subject: Re: Legacy Terminology RE: DMA across PCI
> >
> >
> > I'm using a Galileo PCI controller chip (GT64120A) which specifically
> > implements a 4-channel DMA engine for the host CPU to set up DMA
> > transfers.  I didn't think this was so unusual for a PCI
> > controller chip
> > to do.  If and when we change out this chip, I'll expect the
> > replacement
> > to implement DMA as well,  unless you know something about DMA and PCI
> > which I don't.
> >
> > TedF
> >
> > Lame Brooks-G14738 wrote:
> > >
> > > Yeah, years ago when I was a newbie to PCI, I got smacked
> > on the head for
> > > using DMA in reference to PCI.  DMA is an ISA bus related
> > acronym that
> > > stands for Direct Memory Access.  PCI is by definition all
> > about direct
> > > memory access - that's one of the features that made it
> > superior to ISA!
> > > Therefore, the old term DMA itself has little meaning in
> > reference to PCI.
> > > I think most of the people who try to use it now days are
> > meaning a concept
> > > analgous to the legacy ISA DMA Transfer Engine which is
> > simply a state
> > > machine that can manage a block data transfer given the
> > start and end
> > > addresses.  I'd stay away from legacy terminology when talking about
> > > specific bus behaviour like burst transfers.   -- BrooksL
> > >
> > > -----Original Message-----
> > > From: Daniel DeConinck [mailto:daniel.deconinck@sympatico.ca]
> > > Sent: Saturday, 14 October, 2000 09:48
> > > To: pci-sig@znyx.com
> > > Subject: DMA across PCI
> > >
> > > Hello,
> > >
> > > I looked in the index of the PCI spec 2.1 and the index of
> > "PCI System
> > > Architecture", by Mindshare, Inc. Niether had a reference to DMA.
> > >
> > > I am guessing that there is no such thing as DMA in the PCI
> > world. Is that
> > > the case ?
> > >
> > > I further guess that the closest thing to DMA is a Master
> > read or a Master
> > > write cylce.
> > >
> > > Please enlighten me.
> > >
> > > Sincerely
> > > Daniel DeConinck
> > > High Res Technologies, Inc.
> > >
> > >
> >
> > --
> > Ted Firlit		      		(719) 594-8138
> > Senior Design Engineer
> > UTMC Microelectronics Systems
> > 4350 Centennial Blvd., MS 1004
> > Colorado Springs, CO 80907
> >
>

_____________________________________________________________________________________
Get more from the Web.  FREE MSN Explorer download : http://explorer.msn.com