Hands on PCI interface ...
in [Embedded]
|
Next: at89c52 bootloader
From: John Larkin on 2 Feb 2005 18:12 On Wed, 02 Feb 2005 21:48:45 +0100, Rene Tschaggelar <none(a)none.net> wrote: >Jonathan Kirwan wrote: >> On Wed, 02 Feb 2005 19:34:29 GMT, nico(a)puntnl.niks (Nico Coesel) wrote: >> >> >>>Creating a PCI implementation is difficult, but there are numerous >>>ways to get a PCI core or bridge (like the PLX chips). No need to >>>bother with tedious timing. >>>The PCI specification can be downloaded from several sites. It just >>>comes down to a proper PCB layout on the PCI side. >>> >>>The fastest way to get started with PCI is using a PLX chip. >> >> >> It's just not a hobbyist thing, frankly. The ISA is dirt easy and anyone can do >> it. > >Thanks for the hardware intro. >How about writing a driver ? There won't be single >stepping through code I'm afraid... > We wrote a little DOS app that makes PCI bios calls. It can find a PCI board and drag it down into a hole in the 640K-1M real address space where you can bang the registers all you want. Works under DOS or Win9x. I wonder if there are any equivalent true-Windows programs. John
From: Jonathan Kirwan on 2 Feb 2005 18:40 On Wed, 02 Feb 2005 23:43:25 +0100, Lasse Langwadt Christensen <langwadt(a)ieee.org> wrote: >AFAIK the PLX chips have a rather straight-forward backend bus, and I think some >of them can even do PCI to ISA if you really want to, so apart from the requirement that >you need to have a decent PCB if you want it to work reliably it shouldn't >be that difficult and PLX most likely have an app-note on how the PCB should look. There is still a high learning curve if something does NOT work as you expect it to. You need the tools and the knowledge. Of course, if everything works right out of the box, so to speak, then no problem. But then reality does impinge. I still do NOT consider PCI development to be hobbyist stuff. Jon
From: Andras Tantos on 2 Feb 2005 19:11 >> On Wed, 02 Feb 2005 19:34:29 GMT, nico(a)puntnl.niks (Nico Coesel) wrote: >> >> >>>Creating a PCI implementation is difficult, but there are numerous >>>ways to get a PCI core or bridge (like the PLX chips). No need to >>>bother with tedious timing. >>>The PCI specification can be downloaded from several sites. It just >>>comes down to a proper PCB layout on the PCI side. >>> >>>The fastest way to get started with PCI is using a PLX chip. >> >> >> It's just not a hobbyist thing, frankly. The ISA is dirt easy and anyone >> can do >> it. > > Thanks for the hardware intro. > How about writing a driver ? There won't be single > stepping through code I'm afraid... > You might want to take a look at this: http://h-storm.tantos.homedns.org/hc_pci.htm I've made this card exactly for this application: to allow myself to interface to a modern PC without fighting with the PCI bus every time. It is basically a simple ISA-like PCI bridge. You can attach a douther-card on top of it and add whaterver circuit you like. You don't even need a driver if you can go without interrupts. Regards, Andras Tantos
From: keith on 2 Feb 2005 21:45 On Wed, 02 Feb 2005 19:00:33 +0000, Jonathan Kirwan wrote: > On Thu, 3 Feb 2005 04:26:11 +1100, <Nico> wrote: > >>Interfacing via ISA bus is very easy but slow and it is getting difficult to >>find ISA based PCs. > > This is the reason I keep ISA bus based PCs around and working. > >>USB bus easy to use but its packet stucture and therefore the latency is an >>issue for some real-time applications. >> >>So, I want to learn how to build and work/play with PCI interfaces. Can you >>suggest what is the easiest and low-cost way of involving with the PCI bus >>to learn with hands on experiments? > > Well, ponying up to PCI will not be cheap. First, you need to understand the > difference between reflection wave and incident wave -- PCI is reflection wave. You really don't have to get into it in that detail unless you're pushing the envelope. > Second, there are very tight constraints implied by the technology. Your > PCI clock line must be 1.5" +/- 0.1" in length, 2.5" +- /1", though that isn't generally a problem. For a product, sure. > signal lines are to be less than 2.5" (if memory serves), Must be less than .75", IIRC. They really need to be as short as possible. 2.5" likely *won't* work. Stubs are badness! > and you will often find weird > serpentine clock lines to meet that 1.5" requirement. Yes, almost always (2.5"). > Third, because of > the loading requirements (at 33MHz, some 10pF total; at 66Mhz, 5pF), you > will be using an ASIC. No discrete logic with multiple loads on single > lines, for example. Yes, and the receivers aren't typical CMOS. Many FPGAs have PCI I/O. Either is a bad plan for one-off designs though. As mentioned before in this thread, PLX bridges are the way to go. They have PCI<->ISA brifges that work quite well. There are also PCI<->ISA cards. > Fourth, you will pay much more dearly for > instruments that can monitor and display PCI bus signals. Not really true. There are some relatively cheap bus monitors. If one sticks to a known bridge design a scope is all that's needed. I got a PLX-9054 based card running with no more than a scope. > Fifth, PCI mandates plug-and-play and certain minimum register > requirements and the ability to assign block addresses, if needed. Another reason to go with a known design. ;-) > You can get low cost ASICs. Not ASICs. ASSPs (Application Specific Standard Products). > But the rest makes this not low-cost and > there is a high threshold of knowledge required, as well. PCI was > almost designed from the ground up to exclude basement developers. Designed to exclude? Are you implying that they intentionally raised the entry bar? That's some charge! There is a reason for complication. It makes life simpler. ;-) -- Keith
From: keith on 2 Feb 2005 21:49
On Wed, 02 Feb 2005 23:40:56 +0000, Jonathan Kirwan wrote: > On Wed, 02 Feb 2005 23:43:25 +0100, Lasse Langwadt Christensen > <langwadt(a)ieee.org> wrote: > >>AFAIK the PLX chips have a rather straight-forward backend bus, and I think some >>of them can even do PCI to ISA if you really want to, so apart from the requirement that >>you need to have a decent PCB if you want it to work reliably it shouldn't >>be that difficult and PLX most likely have an app-note on how the PCB should look. > > There is still a high learning curve if something does NOT work as you expect it > to. You need the tools and the knowledge. It's not *that* difficult. The design kits have all the information needed. > Of course, if everything works right out of the box, so to speak, then no > problem. But then reality does impinge. > > I still do NOT consider PCI development to be hobbyist stuff. It's certainly within the upper-end hobbyist's realm. ...at least with the PLX bridge chips or prototype cards. -- Keith |