home *** CD-ROM | disk | FTP | other *** search
-
- > Hi Doug,
-
- > I believe it's supposed to be just 2 banks, with 32MB space for each one, but
-
- > That's another interesting thing: two banks with 32 MB space is 64 MB.
- > But Overscan says that AB can have up to 128 MB RAM.
-
- I know. I can only find 2 banks, and they are separated by 32MB. There are only
- 2 slots, so 128MB seems impossible.
-
- Maybe it reconfigures itself with larger simms, but I really doubt it. Even if
- it did, we couldn't support that without details.
-
- > OK. Can I test it somehow simply on my own AB?
-
- I could send you a version of the 68040 driver which sets up a flat, unprotected
- tree over the first 64MB of ram. This will make things much easier to test.
-
- > I think the program is OK, isn't it?
-
- No, because if you test an 8MB block, and it consists of 2 mirrored blocks of
- 4MB, then the test will fail right away.
-
- When the first location of the second 4MB block is written, it will overwrite
- the same first location of the first 4MB block. When the testing pass takes place
- the first 4MB block will be considered invalid and fail, even if it is ok.
-
- The only simple way round this is to find blocks of ram with addresses that DO still
- match up, but these will be the last (highest up) copies of any mirrors which might
- be available. This means you would need to test all 64MB of space and find all
- the little fragments which check out, and join them together!
-
- I did say it wouldn't be easy... :I
-
- > I know about this problem (you already told me), but now I try to
- > find my other 4 MB of my 8 MB SIMM. Later, when I buy the 16 MB one, I'll
- > probably get into these problems with repeated 8 MB blocks, but for now I
- > would like to find whole 8 MB of my current borrowed SIMM.
-
- If you only have one simm fitted, then it should be 8 contiguous megs of
- ram starting at $1000000. If my program only detects 4MB, then there may
- only be 4MB starting at $1000000 with mirrors just above it. I don't know
- where the rest might be.
-
- I'll look further into this problem to see if I can develop a fail-safe method
- for counting RAM.
-
- I've just had an idea - if I send you a small program which fills memory at 8k
- intervals with the appropriate addresses, then saves these filled locations
- to disk (64MB/8k= 8000 locations or 32k of data), you could send me the resulting
- file and I could study it, looking for mirrors. When I find all the unique 'bits',
- I might then have a better idea of what's going on.
-
- > BTW, an idea: could I read the setup from MMU and send it to you? We
- > could find out where the AB-RAM is with the original AB's SW!
-
- I don't trust that software. My friend fits 16MB to one slot and that
- code produces 32MB on the desktop. Any further information is bound to
- be unreliable. The original drivers don't test the ram properly.
-
- > 1) with 8 MB SIMM it recognises only 4 MB of it, but I have checked with
- > the original AB's SW and a little test program (see above) that I
- > actually have all 8 MB of FastRAM.
-
- Ok, I'll keep looking into this. It is made harder by my own machine using
- different simms and the board appears to behave differently too (no mirrors).
-
- The mirrors indicate a problem related to incomplete memory decoding, but
- I can't tell if it is deliberate, harmless or even a hardware fault.
-
- > 2) some programs are confused with the beginning of FastRAM at $2E6,
- > namely Profile, SysInfo and Linux boot loader. As for Linux, it can be
- > solved with special cookie, but the system info programs will always
- > report bad numbers with your current patch.
-
- If Linux sets up it's own MMU, then the changing 68040 patch will not help
- anyway, but for other programs it is worth considering.
-
- > 3) some programs don't run with your patch (but do run with orig AB's SW):
- > - TTRAMTST says Not enough memory (maybe it expects 4 MB of FastRAM at
- > least but I have only 3 MB with your patch - I don't know)
- > - Music Analyzer ("M_ANALYZ.PRG") gives two bombs
-
- Hmmm... Sounds like something is changing the cache or MMU tree. There will
- be some things I can't program round without ripping features out of the
- patch. I'll see what I can do about it.
-
- > That's all I remember. I hope that you don't take it as a criticism of
- > your work - I just wanted to tell you what you could improve a bit if you
- > would want to.
-
- Yep, well I might not be able to solve everything, but if these are the only
- real problems you are having, then the outlook seems good.
-
- Doug.
-
-