Amiga Plus 2004 #9/Amiga Plus CD - 2004 - No. 09.iso/amigaplus/tools/emulator/winuae/installwinuae0990.exe/Docs/README.pci

Text File | 2000-12-15 | 17.8 KB | 368 lines

This describes how to get Virge and Virge/DX cards working under
UAE-JIT if you are using Linux.

************************************************************************
* EVERYTHING DESCRIBED IN HERE IS A HACK. IF YOU MANAGE TO DAMAGE YOUR *
* HARDWARE, SOFTWARE OR DATA --- BAD LUCK! I AM NOT LIABLE IN ANY WAY! *
* IF THE IDEA OF USING A COMMAND-LINE FRIGHTENS YOU, DO NOT READ ON!   *
************************************************************************

OK, scared yet? No need to be --- just follow the instructions, and if
things don't work, experiment a bit before you send a cry for help my
way, OK?


A) The Easy Part --- a single Virge
===================================

To use a Virge PCI card under UAE-JIT, you need several things:

   * A Virge card (well, duh!)
   * Another gfx card that you will use for your X display (This is
     important! You must not use the Virge for this! If all else fails,
     try a remote X server)
   * The CGX3 Virge drivers from http://www.vgr.com/mediator/
     Get the October 26th release.
   * The pci.library that comes with UAE-JIT
   * XFree86 4.0x
   * A halfway recent linux kernel
   * A pci enabled version of UAE-JIT (i.e. something "bleeding edge",
     and newer than Everton. You probably already have it if you read
     this!)

Now put the Virge card into a PCI slot on your machine, make sure it
is *not* the card the machine uses to boot (look for "initialize VGA first"
in the BIOS, sometimes hidden in some rather unintuitive place, and set
it to "AGP"). Boot into linux, and at a command line type

    # pcitweak -l

(the '#' is the shell prompt I use. Don't type it!)
You should get output that looks something like this:

    PCI: Probing config type using method 1
    PCI: Config type is 1
    PCI: PCI scan (all values are in hex)
    PCI: 00:00:0: chip 8086,7190 card 0000,0000 rev 03 class 06,00,00 hdr 00
    PCI: 00:01:0: chip 8086,7191 card 0000,0000 rev 03 class 06,04,00 hdr 01
    PCI: 00:07:0: chip 8086,7110 card 0000,0000 rev 02 class 06,01,00 hdr 80
    PCI: 00:07:1: chip 8086,7111 card 0000,0000 rev 01 class 01,01,80 hdr 00
    PCI: 00:07:2: chip 8086,7112 card 0000,0000 rev 01 class 0c,03,00 hdr 00
    PCI: 00:07:3: chip 8086,7113 card 0000,0000 rev 02 class 06,80,00 hdr 00
    PCI: 00:09:0: chip 10ec,8029 card 0000,0000 rev 00 class 02,00,00 hdr 00
    PCI: 00:0d:0: chip 5333,8a01 card 5333,8a01 rev 01 class 03,00,00 hdr 00
    PCI: 00:0f:0: chip 5333,8a01 card 5333,8a01 rev 01 class 03,00,00 hdr 00
    PCI: 01:01:0: chip 1023,9750 card 1023,9750 rev f3 class 03,00,00 hdr 00
    PCI: End of PCI scan

What you are looking for are lines that have "5333,8a01" or "5333,5631" in
them (those are Virge/DX and Virge cards, respectively). Take down the
field that looks like "00:0d:0" for those lines. That's the identifier for
those cards. Yes, I have two in the machine I took the example from.
You might also want to write down the identifier for your "main" graphics
card. That should be the only other one that has "class 03", and often
lives on its own bus (that's the first field in the identifier).

Next, you need to tell UAE-JIT about the PCI cards it is supposed to see.
The pci library emulation will not allow the emulated Amiga access to anything
else, which should reduce the risk of crashing your machine a tad....
To tell UAE about the devices, add a line like this to your UAE config
file:

   pci_devices=!0.f.0:!0.d.0:!*1.1.0

OK, this obviously needs explanation... "pci_devices" is assigned a string.
That string consists of a number of device descriptions, separated by ':'.
So the above has three device descriptions: "!0.f.0", "!0.d.0" and "!*1.1.0".
Except for the '!' and '*', those correspond to the identifiers you took down
in the previous step --- for me, "0.f.0" is one of my Virge/DX cards,
"0.d.0" is the other, and "1.1.0" is the Trident card that I run X on
(I assume that you are flexible enough to replace colons with periods ;-).
Each device can optionally be prefixed by '!' and '*' (in that order!).

The '!' means that UAE should not be concerned about the device already
being active. This is "active" in a PCI sense. Unfortunately, UAE needs
to activate a device to use it, and doesn't currently deactivate it. So
you pretty much need the '!' for now, because otherwise you will only
ever be able to use the PCI emulation once....
The '*' indicates your main display. UAE will never make this visible to
the Amiga. In fact, at the moment it doesn't do *anything* with it.
Eventually, however, it will need to know about that VGA device to do
proper initialization of the Virge, so it's probably a good idea to
put it in there already.

Unfortunately, there still is a problem with getting the Virge cards
properly initialized by the CGX3 drivers --- it seems as if I/O to the
cards never makes it through. That's not really a showstopper, as the
Virge has memory-mapped I/O which can be used instead, and the pci emulation
takes care of redirecting the I/O. BUT(!) you need to get the Virge cards
initialized.
That's what you need XFree86 4.0x for. You will need to set up a config
file in some place to run it on your Virge card. Make sure you point it
at the right card, and tell it to use the Virge driver. I have a device
section like this

    Section "Device"
        Identifier  "virge"
        BusID      "PCI:0:13:0"
        Driver      "s3virge"
    EndSection

(the "13" is the "0d" from the identifier --- a=10, b=11, c=12, d=13, e=14,
f=15). If I wanted to use the Virge on 00:0f:0, I'd use 15 instead of 13.

Setting up X can be tricky, but XF86 4.0 has a great config file, some good
documentation (and some not so good :( ), and quite frankly, I can't help
you. If you can't get this going, ask in the appropriate newsgroup or
read some more documentation, ok?

Once your config file is done, you can start a bare-bones X server on
the Virge card by typing

    # XFree86 -xf86config your_virge_config_file :1 &

This will start a _second_ X server, in addition to the one you are already
running on your "real" gfx card (you are, right?). Note that there won't
be anything to see, just the stippled background. That's all that's needed
to set up the card.

Once you have managed to do this, its easy the rest of the way. So don't
give up just yet....

Now switch back to your "real" X server (usually CTRL-ALT-F7 will do the trick)
and start UAE. Copy "pci.library" to your libs: directory.

Next, uncompress the Elbox archive, and choose "Install".  WARNING!
WARNING! WARNING! This will trash your P96 setup. So make sure you
have a backup to go back to if things don't work out, or if you just
want P96 back. You have been warned!

When the install is complete, quit UAE and restart it. If all went well,
you will now see the Cybergraphics logo appear on the monitor connected
to the Virge, and then have Workbench appear on it. Hooray! You are now
running Cybergfx, and the card is fully under the control of the Elbox
driver. You can do everything with it that you could do with a "real"
Mediator and Virge --- define your own modes with cgxmode, run software
on it, drag screens.....
Note, however, that UAE itself doesn't know anything about that gfx card.
It still talks to you through the window on the "real" X server, and if
you move the mouse out of that window, UAE loses it. So you might want
to go fullscreen (F12-s if you have DGA modes enabled).

Worse yet, UAE doesn't know that you probably aren't interested in the
ECS output, and thus keeps on emulating the native Amiga video system.
Sometimes, that might be what you want, but usually it isn't. So you
might want to disable that part of the emulation, by pressing F12-i.
Just remember to press F12-i again when you actually want to see the
ECS output, for example because you started a game ;-). UAE-JIT will
put "NAT" at the end of the lines it prints every second if the native
ECS graphics are still being emulated, so if you can see that output,
you'll get a reminder.

Got it working? Great! Enjoy it, and maybe have a look at the next section.

Didn't get it working? Try again. And again. And if it still doesn't work,
and you think you should mail me ([e-mail]) for help, be
sure to include as much detail as possible about your system, what happens,
what doesn't happen, and so on.


B) For Bravehearts Only --- using more than one Virge
=====================================================

The free Elbox drivers don't actually support more than one card. This
is due to some hardware limitations of the Mediator, which it is aimed
at. So to get more than one to work under UAE requires some trickery.
What essentially happens is that each card gets "its own" PCI interface,
and is the only one using it (and thus works).

Unfortunately, the drivers weren't meant to do that, either, and have
some names and values hardcoded in. You'll need to put the included
"patchVirge" program in the same directory as the "MediatorViRGE" driver
file. Then run it (from a linux command line --- it's a linux/x86 program!)
by typing

   # ./patchVirge 1

which should output

    replacing pattern 1, pos 577
    replacing pattern 2, pos 617
    replacing pattern 2, pos 745
    replacing pattern 3, pos 1806
    replacing pattern 2, pos 1818
    replacing pattern 2, pos 1832
    replacing pattern 2, pos 1862
    replacing pattern 2, pos 1910
    replacing pattern 2, pos 2002
    replacing pattern 4, pos 2486
    replacing pattern 1, pos 35118
    replacing pattern 1, pos 35184
    replacing pattern 1, pos 35210
    replacing pattern 1, pos 35228
    replacing pattern 2, pos 52118
    replacing pattern 1, pos 53076
    replacing pattern 1, pos 53100
    replacing pattern 1, pos 53116
    Replaced 18 patterns

and produce a file "MediatorViRGE.1". Repeat with

   # ./patchVirge 2

and

   # ./patchVirge 3

For the sake of symmetry, you can also run

   # ./patchVirge 0

and get a "MediatorViRGE.0", but apart from slightly different mode names,
that one is identical to the original MediatorViRGE.

Each of these files is a completely functional driver. Each of them will
produce screen modes with slightly different modenames, and (more importantly)
with different mode ids (and don't ask how long it took to find what bytes
to patch for *that* ;-). Back in UAE, copy as many of them as you have
Virge cards into the Devs/Monitors of your boot disk. WARNING! The files
might not be marked executable. After copying, make sure you select each,
use the "icons->information.." menu item, activate "executable" and
"save" the changed info.
If you use MediatorViRGE.0 (I do, so it might be a good idea), you might
want to remove the original MediatorViRGE driver from Devs/Monitors, because
otherwise you might get mode-id conflicts (which are a real pain to track
down, as CGX is very quiet about them).
One more thing --- each of those drivers uses its own, individual copy of
the "pci.library". MediatorViRGE.0 uses "pc0.library", MediatorViRGE.1 uses
"pc1.library" and so on. So you need to copy the right libraries to libs:,
or things won't work. Except for the name, the libraries are exactly
identical, but you need to have separate ones. You cannot just copy one
library over and over again, or even use links --- the names of the libraries
are contained inside as well.
Oh, and one more "one more thing". You need to copy a few files in
"prefs/env-archive/cybergraphx" on your boot drive. There should be a file
"MediatorViRGEMonitor" there --- copy it to "Mediato1ViRGEMonitor",
"Mediato2VirgeMonitor" and so on, and of course "Mediato0ViRGEMonitor" as
well if you use that. Don't worry about the original file; You can keep it
without causing problems.

Now exit and restart UAE, and the cards should be assigned to the drivers
in the order you mention them in the config file. You can't currently
use more than 4 cards, and you should not have more than 4 drivers
installed at any one time --- but apart from that, it doesn't matter if
you have more drivers installed than you have cards, because each card
will only be found once. For example, I have 4 drivers installed, and
currently only have 2 Virge cards.
[Note --- I consider it unlikely, but it is possible that you actually need
to install the full CGX3 package, rather than just the Elbox version. I did
so while bug hunting, so I cannot promise that it works without. I can't see
why it wouldn't, though.]

C) Known Bugs
=============
You certainly don't want to do a soft reset --- the PCI emulation isn't
up to that yet, and will not release the cards. Things will crash one
way or another. Just don't do it!

You *CANNOT* use anything but the Virge driver. No other drivers are
publicly available, anyway, but just in case you somehow get your hands
on one --- IT WON'T WORK. If you can legally send it my way, I might be
able to make it work, but right now, there is quite a bit of Virge
specific stuff in there.

For some reason, the bottom line of the screen is displayed above the top
line. I think this is a driver problem, and has nothing to do with the
pci emulation.

Some 2-monitor configurations in Shapeshifter may give corrupt screen
display on one monitor. Once again, I suspect the problem is on the
Amiga side.

D) How did this come about?
===========================

A lot of trial and error, and a lot of "where the $@%#$^@ did that value
come from", which UAE is just great for. And of course the good old
"Let's return some bogus value, and see where it shows up", which UAE
is also great for.

I have never seen anything but a picture of a Mediator, and I have
never seen any Mediator related code other than the Virge driver. This
means that anything the Virge driver doesn't use, the emulation
doesn't support. There are many pci.library functions for which I have
absolutely no idea what they are supposed to do --- the Virge drivers
only use two.

Personally, I think that a proper pci.library emulation, developed from
a full and detailed spec, would be a GREAT tool for driver development.
Developing in an emulation allows for great logging and debugging support
that working on the real iron just can't provide (Logging every byte ever
written to your device is just so much more useful than simply seeing
your Amiga crash and burn). But it appears that Elbox likes to play
their hand close to their chest, and thus are unwilling to open up
their specs. I think that's a mistake, but I am not in a position to
make any decisions....


E) If you have a look at the source...
======================================

...you will notice a couple of #define's at in the file pcilib.c, switching
on or off features.

The first one is "USE_REAL_IO". It is related to the already mentioned
problems with initializing the Virge cards --- if you set it to 1,
the pci emulation will use the PCI I/O space rather than the memory
mapped I/O of the Virge cards. In that case, you *must* provide an
entry in pci_devices for your main VGA card (it needs to be deactivated
temporarily when I/O to the Virge occurs). This has never worked for me.
It might work in a machine that has just one PCI bus, and no AGP bus,
or if your main VGA card is on the same PCI bus as the Virge. If you get
this to work, tell me!
If the PCI emulation ever is to support other drivers, such as the network
one that seems to float around, this will need to be set to 1. But before
that happens, I need to find out why XF86 can wake up the Virge in my
machine, but the emulated Amiga's Virge driver can't.

The second one is called "EVIL_HACK". It's evil. It's a hack. It also
saves loads and loads of CPU time ;-).
The Mediator is limited to an 8M address window, which means that it
cannot at the same time have access to the VGA frame buffer of the
Virge *and* the memory-mapped I/O area (which is at
frame_buffer_address+16MB). So each time it wants to access I/O, it
switches the window 16MB forward, does its access, and switches
the window back. And the driver isn't very clever about it; If it
wants to do two accesses to I/O, it still switches the window
back in between.
Now, on a Mediator, switching the window is cheap --- store the value in
some hardware register, and whatever is in that register gets added to
the address of each access to the PCI space. Under UAE, it's not cheap.
It requires un-mmapping 8M of memory, and mmapping another 8M. Each time.
Quite often only to reverse the mappings immediately afterwards. Not good!
Fortunately, UAE is not limited to a 24 bit address space. So if you enable
"EVIL_HACK", the pci emulation will search for a 32M address window instead,
and map the whole 32M space of the Virge at the same time.
So far, so good. However, the Virge driver doesn't expect this, and
rather assumes a window switch to provide new stuff in the first 8M of
Mediator space. And unfortunately, the "switch window" routine doesn't
return the base address of the Mediator space to use (Bad Thinking there,
Elbox! You are locking yourself into your current scheme!), so changing
that assumption is not possible in a non-evil way. However, it *is* possible
to change in a completely evil way --- after each time the window switch
function is called, the driver loads the Mediator window base address from
its own, driver-local storage space. It always loads it from the same spot.
So the evil thing to do is to "mess with its mind", i.e. to have the
window switch emulation leave the mappings alone, and instead change
that base address variable.
Now, mind you --- it's not the pci.library's to change, and where it
can be found could quite well be completely different for the next
driver version.  Messing with the internals of another program is
certainly not "the done thing". But if it saves CPU cycles, who am I
to reject it?