Restoring a Sun 386i: What a mess, anyone have a running one?

[ajacocks]

What kind of module is it? Dallas?

- Alex

 

[czunit]

Yep, the normal Dallas semiconductor timekeeper chip. Forgot the model number but it's in all the FAQs.

Meantime I took the CPU board out of the 386i case to take a look at the CPU. With no boards (including no memory) the lights on the back should at least flash around a bit, but it's doing all lights on which means no CPU activity at all. Maybe the CPU chip is shot, I think I have spares. Pics in a moment.

 

[czunit]

Ok, here are pictures of the 386i.

 

[czunit]

In the meantime I'm going to start working on the second and third systems. The second one will at least start doing diagnostics, failing with a CPU error 3 (led 0,1 lit). The third does nothing and all LEDs are always lit. Note all tests are done on a bare motherboard, no memory or other cards.

Test 1: Try things on unit 2 to see what causes a complete failure (no LEDs)
Nothing: LEDs throw error 3. Good.
Pull CPU chip. LEDs do nothing but all on.
Put CPU chip in. LEDs throw error 3.

Fact: Removing the CPU will cause LEDs to just light and do nothing. CPU controls LED register.

Test 2: Try chips from unit 3 to see if they work.
386 CPU swap. Works, chip good.
386 CPU spare chip. Works, chip good.
387 FPU swap. Works, chip good.
387 removed. Works, 387 chip is not needed for initial start of diags.

Fact: CPU and FPU chips are good.
Fact: FPU is not needed for initial diags to start running.

Test 3: Try swapping timekeeper chip from unit 3.
Timekeeper swap. Works good.
Timekeeper chip removal: DIFFERENCE: System will flash the high 4 bit LEDs, but the low LEDs will stay at zero.
Timekeeper chip install: Works good, LEDs at error 3

Fact: Without the timekeeper chip something will happen, but diags will not execute.
Conjecture: With a dead battery the timekeeper may prevent system from getting past error 3.

Test 4: Try swapping 82380 chip from unit 3.
82380 removal: All LEDs on.
82380 swap: Works, LEDs at error 0
(Note this is because I forgot to put the timekeeper chip in.)
82380 reinstall: Works, LEDs at error 0
82380+Timekeeper: Works, LEDs at 3

Fact: 82380 chips appear to be good.
Fact: Removal of 82380 chip will brick the system.

Other tests to do later:
Test 5: Look into the socketed 20 pin chips. They might be small ROMs, or custom ASICs.
Test 6: Find and check clocks to see if they are running with scope.

Advice on other tests appreciated. Will keep the results here.

 

[czunit]

Well, I tried the timekeeper chip from the running 386i (the one I hacked 30 years ago) and plugging it into board 2 gives me the following:

Stops on test 3.

Putting in a valid memory card (only 4mb on a 16mb board) to test:

Passes diagnostics and goes to the cylon pattern indicating the ROM is up.

This is good. It means that firefly (my second 386i) is also up and running.

Now to see if I can get into it via serial port, and try updating the eeprom data. I also took the dead timekeeper out to the shed, ground down the back of it to expose the two pins, and I'll order another battery holder from Amazon to give it a try as well. On the first one I split the contacts completely, I wonder if a battery will work if the old battery pack is still connected.....

Then I have to think more about the third one. Maybe the Bios PROM is bad.

 

[hornbetw]

The Li cell in the timekeeper module is not made to be recharged. If you connect another battery in parallel with it, you'll recharge it and cause it to split open and leak. Disconnect the internal battery before hooking up the new one to the chip.

 

[czunit]

Noted, and good point. I need to find my dremel to make a nice clean cut of the interconnect. Now where did it go....

Meantime, I swapped the ROM from board 3 (not working) into board 2. Board 2 fired up properly, so the ROM is not bad (although it's 4.4 on board 3, and 4.5 on board 2).

Noted.

 

[czunit]

Well let's see: I dropped to the ROM monitor and did a q 100 to start entering sane values for the eeprom. Here is what I put in from my dump.

0x100 1
0x101 8
0x102 1 hard disk
0x103 1 diskette
0x104 80
0x105 1
0x106 0
0x107 8d
0x10F aa
0x110 10
0x111 3
0x112 0

Note: This makes no sense according to the FAQ but it does work. Ethernet is now up. Now how do I change the IP address on the system with yellow pages up? I forget, simply editing /etc/hosts and /etc/networks does not do it.

Next step is I am kermitting over the diag.img image to put on a floppy and will dd off the 3.0 diags. Note the 3.0 diag disk CPU tests would still fail hard, so there is something wrong with them, or they need the CPU board rev and such set in the EEPROM before they work. Chicken/egg thing. I'll try the 3.0 diags and see if CPU still fails before putting in the older 4.01 diags and running them.

Update: Diags disk is the same 3.0 diagnostics. Not sure if these were the same ones that were on the 4.01 disk set. However big problem is my 4.01 diag disk is totally unreadable. If there is an earlier version of the diag disk I'd like to find it...

Update: CPU test still fails. Did change the memory from 8 to 16, probably recomputed the checksums, but the important thing is the motherboard revision and the like is set properly. So ethernet works. Oh and look, I hardened this system by removing telnet and ftp daemons. Good for me, but......

God Kermit takes FOREVER! Note: You can speed it up some by going to 19200 on /dev/ttya and setting the send and receive packet size to 1000. Speeds it up a *LOT*. Oddly enough I was able to download a file from the 386i at 38400, not sure if sending would work at that speed.

 

[czunit]

More progress: The third memory card that was throwing errors on test #3 is now working again. What I did was pull the 82385 cache controller chip off the socket, then found that the pins were gunked with dust and junk. Cleaned the pins off with compressed air, put the chip back on, and now it's working.

So I have 3 memory boards now. The last thing to do is rebuild the power supplies and figure out why the last motherboard isn't doing anything. It's a complete dud, doesn't even turn off any of the LEDs on the back to start a test. I've pulled, tested, and re-seated all the big chips, maybe it's time to work on the smaller ones....

Also got the IP addresses changed, and updated the resolvers to point to my current domain master DNS server instead of using 30 year old cached BIND files. Turns out you just need to run the appropriate make (make hosts to update yp with the local copy of /etc/hosts for example) in /var/yp. So with that done Suntools is up, my old accounts are there, and the NCSA 1.0 web server from 1994 or so that I compiled is still there and working.

But man the system is slooooo compared to modern systems. That's ok, it does work. Wonder if I could patch sendmail, bind 4, sendmail 8, and build a ssh host strong enough to put it on the internet again......

Edit: Fired up Suntools and found the DOS image still has checkit and Lemmings! Man I haven't seen checkit in 30 years, it was a diagnostic/demo program that I used all the time and lost all the copies of. Fun to see it back.

Edit: Need to figure out how to set a modern date. Date command is useless, and although it will roll over to 2000 if set to 991231, setting a modern date throws it back to 1923...

 

[mdog69]

What kind of module is it? Dallas?

- Alex

MK48T02

Note that "dremelling" these devices is the way to go, as ISTR that there were different versions of silicon, and replacing an old one with a later version may not work.

For the record my IPX has a butchered MK48T02 with the top removed and a CR2032 in a holder and a new 32.768kHz crystal.

The crystal is soldered directly to the respective pins and folded back onto the top of the chip. The holder is hot glued on top.

 

[czunit]

 

[komradebob]

Side note: I worked in the 386i development group way back when for a brief period. Before we had actual 386i hardware, we used Compaq 386 PCs with a graphics card on them for development.

 

[komradebob]

View attachment 1248252

Congrats!!!

 

[czunit]

Really handy thing: The 386i with the Roadracer video card *does* work with a modern LCD monitor using a simple VGA-BNC cable adapter. Much easier to deal with than the NextStation 13Y3 color monitor. Back into the closet that thing goes....

 

[czunit]

Congrats!!!

Yes, I think I will FINALLY be getting the PC board standoffs tomorrow so I can start working on putting this franken-power supply into a SUN sized box. That will allow me to button up the box and have a funny running unit.

Then I can figure out the other two supplies and figure out why the third CPU board does nothing. It's weird, doesn't even start any of the diagnostics but the chips are good.

 

[czunit]

So let's see: Power supply is jammed into the Sun PS box. Could be worse, but it works.



And the system all buttoned up and running X windows and of course xsnow....

 

[czunit]

Side note: I worked in the 386i development group way back when for a brief period. Before we had actual 386i hardware, we used Compaq 386 PCs with a graphics card on them for development.

Really? Interesting, anything you can tell me about why the CPU will only use Intel 386's and not the Cyrix chips? I have a Cyrix 386/50 and it just does nothing in the box which doesn't make a whole lot of sense.

Currently the SunVGA card is odd, it seems to blank out the DOS window and make it very slow. Might have bumped some of the switch settings....

 

[DDS]

AS -> Arne Saknussemm

The markings on a cave wall are mentioned in "Journey to the Center of the Earth" by Jules Verne.

Journey to the Center of the Earth - Wikipedia

en.wikipedia.org

 

[czunit]

In the category of "minor stuff but still useful" I used my calipers to measure and create a 3d model of the drive sled bracket. The original is plastic, so measuring it exactly, figuring out where the holes and countersinks for the screws are, then building the 3d model gets me a really nice plate that fits and works....

I'll post a picture later but I have to say 3d printers are *GREAT* for fabricating old parts. I'll also put it on Thingverse.

 

[czunit]

Pictures of the hard drive caddy/plate. Came out pretty well, screws fit perfectly, seems even more solid than the original.