6.2 KiB
Bringing up the Kernel
Add notes on what you've tried w.r.t bringing up the kernel
wkt Sun Mar 13 12:26:06 AEST 2016
We can make at the top level to assemble the kernel, some utilities and build the filesystem with the utilities, dd and system directories.
In the build directory, make run starts Simh running, loads the kernel into memory and sets the PC to 0100.
I'm sure that Phil has got further than this, but by setting a breakpoint at 01000 (set br 1000) and doiing a c(ontinue), we see:
simh> show cpu history
...
003351 1 000072 000000 JMP 3361
003361 1 000072 000000 LAC 4257
003362 1 000454 000000 JMP I 3272
004544 1 000454 000000 JMP 10000
which is the end of coldboot in s8.s. The first few instructions are definitely from init.s. By breaking at 001115 (fork from a.lst), we can see two forks, which correspond to the two in init.s.
We do get to the 010052: 020005 sys write; m1; m1s code in init to write "\nlogin:" on the terminal, but this isn't being displayed.
We make it to 01363, the first instruction in wttyo, so the write syscall is hitting the right device. We do get into dsprestart at 03516, and we do make it to the dpcf instruction at 03527, but we don't get past here. We don't make it down to the sma instruction at 03531 or the ksf instruction at 03537,
wkt Sun Mar 13 14:41:30 AEST 2016
I've commented out some of the Graphics-2 code around dsprestart and pibreak in s7.s. With this code removed, I now see the login: prompt being written by the init child. It's not responding to keyboard characters, but I haven't traced the code yet.
wkt Sun Mar 13 15:06:53 AEST 2016
Using the modified kernel and the normal init.s, I've set breakpoints at 0352, the dac after krb in dsprestart, and at 10253 the lac char straight after the sys read in init.s.
I'm seeing breaks at 0352 for the four characters "ken\r", but I'm not seeing any breaks after the sys read in init.s. The read only reads one word, so I was expecting to see it react after one or perhaps two characters typed at the keyboard, or even the \r character.
Sun Mar 13 21:11:26 AEST 2016
Phil made essentially the same change. He noted in an e-mail: entering dmr<CTRL/J> at the login: prompt got me a password: prompt. Entering dmr<CTRL/J> there gave me a question mark. Which may be either: I was wrong in my guess about "password" file format, or where home directories should be located...
So the next step is to dissect init.s in more detail. I've changed the as7 assembler so that we can #ifdef the code around dsprestart in s7.s but keep the Graphics-2 code still in the file.
wkt Mon Mar 14 19:51:08 AEST 2016
Phil realised that it was the link syscall which was failing. It seems to need a directory at i-num 4. I fiddled a bit and added some more kernel comments. It seems that "dd" has to be a i-num 4 and "system" at i-num 3. I've modified mkfs7 and the proto file to allow this to occur. I've also make link counts negative. This now gets init past the link syscall, but it then dies on the open("sh") immediately after that.
wkt Mon Mar 14 20:20:44 AEST 2016
A transcription error in s2.s link caused the new link's d.inum to be zero and not the correct value. Fixed. I've eyeballed the fs differences and they look OK to me. init is still erroring. Not sure why yet.
It's the open call directly after the link that's failing. Why, if the open is for the link we just created.
When I dike out the chdir syscalls and put a sh in system, we can open this one, bounce up to high mem, read the shell into 10000 and then exec this code. So the rest of init is OK, but there's still a bug opening the link we just created.
wkt Mon Mar 14 21:23:30 AEST 2016
I had wrong permission bit values in mkfs7 which was stopping the shell open once we had changed user-id to the real user. Now fixed, and we can now get to a shell prompt.
wkt Wed Mar 16 06:20:24 AEST 2016
The shell is working, but we have to move the binaries into the user's directory to get them to work. ls wasn't working because there was no way to open the current directory. I've taken the decision to add a "." entry to each directory in the filesystem, so now we can run things like ls, cat, date. We have a minimally working kernel!
wkt Wed Mar 16 09:50:13 AEST 2016
I uncommented the code in Phil's shell to link binaries from the system directory so that we can run them in the current directory. I fixed a few home-grown utilities, and I added world-read permissions to dd and system. cp works, so does stat. A few things not working yet like chrm, mv, ln.
wkt Sat Mar 19 06:31:52 AEST 2016
The shell now has code to link binaries in from the system directory, and mkfs7 has been modified to optionally create . and .., so things are mostly working now including chdir.
wkt Wed Mar 23 14:16:15 AEST 2016
To keep things sensible, we have set the main build rules to build from the untouched historical sources. Phil has managed to get SimH to simulate enough of the Graphics-2 device to use it as a second console. We have removed all the #ifdefs from the code.
Warren has been working on an alternative version which sees the system advance by about a year to when it has . and .. directory entries, no dd entry, plus things like mkdir. To build this, do make alt, make altrun. The src/alt area is used to hold the "alternative" code.
Essentially, it all works now except a few minor glitches. We now need to get as up and running fully, get ed rescanned and working. Eventually it would be good to have a working roff; but someone is going to have to write it from scratch.
wkt Thu Mar 24 17:53:55 AEST 2016
The B compiler works and can compile itself, thanks to the hard work done by Robert Swierczek, so it's now on the filesystem image; there's an example B program in dmr's home directory.
Phil has written a proper RIM bootstrap for the system with the kernel loaded onto physical track 180 (logical 80). The mkfs7 tools has been altered to place the kernel there, so now we can boot from a paper tape that is only 237 bytes (23.7 inches) long instead of 14133 bytes (1413 inches) long :-) That will help when we get to the LCM machine.
We have a new scan from Norman Wilson, and some transcription errors in ed have been fixed. ed now assembles and runs, but it isn't working perfectly yet.