home *** CD-ROM | disk | FTP | other *** search
-
- > Hahaaahaaa! Haaahaahahaha!
-
- > :-) not very funny for me, though.
-
- I'm not laughing at you - I'm laughing at Profile. :I
-
- > I've just worked it out after you pointed this out to me. (should have
- > guessed earlier!).
-
- > OK, I should check the mem variables earlier...
-
- No, I meant _I_ should have guessed earlier... :/
-
- > You are not going to believe ME! :)))))
-
- > I see, I should learn to use the correct English tense at last (not will
- > not believe, but are not going to :-)
-
- No, I just meant that you were not going to believe me, after you suggesting
- I might not believe you! I wasn't criticising your English (which was quite
- correct as far as I noticed).
-
- It appears everything we say to one another has several possible meanings, with
- the wrong one at the top of the list. :)))
-
- > 3) You have 1MB of this ram allocated for FASTROM (automatic RomSpeed). This
- > drops alt_ramtop to $1300000.
-
- > OK, but think about the automatic allocation - someone with little memory
- > probably doesn't want to spend 1/4 of his memory for ROM ;-)
-
- I didn't expect there to be many 4MB fastram users. I was going to make it
- configurable anyway - it's not complete yet. It already disables fastrom
- on machines with less than 4MB.
-
- > OK, but that's your not bug, but fault or something. Really.
-
- > First, AFAIK there's no documented way for finding out the start of FastRAM.
-
- No clear way anyway...
-
- > Second, it is documented, that FastRAM begins at $1000000.
-
- No, it's actually documented that TT ram always starts at %1000000 on real TT
- machines, but alternate ram may begin and end absolutely anywhere there is
- a valid CPU addressing space. Motorola guidelines suggest 32MB boundaries.
-
- > Third, programs do rely on the beginning at $1000000. At least all
- > programs for checking free RAM and, which is more important, Linux relies
- > on it, too. And Linux is the most important for me ATM.
-
- I wouldn't be concerned about ram-checkers if they make such assumptions, but
- Linux should know much better. It is designed for any platform with any memory
- map configuration - not the TT.
-
- I am prepared to change the driver to solve problems with Linux if they do exist,
- but I would be most disappointed if it doesn't already depend on an external driver
- to specify alternate ram pages. Any other way is unsafe.
-
- > I'll try to check it in a little program written in C, that will be the
- > fastest solution for now.
-
- It won't help you - the MMU is instructed to map any non-valid ram pages as 'missing'
- so no memory errors result from crashes (wraparound can cause good ram to be destroyed
- by unchecked writes).
-
- You will need to try it on a clean machine (if the MMU is not set up the same way)
- or with one of the AB drivers, also assuming they leave all invalid ram unprotected.
-
- > Anyway, Profile should not be assuming alt_ram begins at $1000000 - there are no
- > rules specifying this address as being a requirement. That's why the maddalt call
- > offers you a choice of destinations! :)
-
- > Well, that's true. But maybe it is because you can have e.g. two external
- > memory blocks, so the second begins at $1400000, but the first at
- > $1000000.
-
- Whatever the reasons, there are no rules for allocation of alternate ram. TT ram
- maybe, but not alternate ram. There is an important difference - altram encapsulates
- TT ram, but not the other way round. It's quite possible to have both TT ram and
- other altram at the same time. This allows external ram devices to be mapped at
- any location at all.
-
- > Anyway, in Atari Compendium and in other sources there's written Atari
- > memory map, where TT RAM always begins at $1000000...
-
- Agreed, but I did not expect anyone to assume Falcon altram MUST begin here. It
- could be anywhere - quite legally.
-
- > Could you please modify your patch to reflect that?
-
- I will, for the sake of Linux and other more important programs which may have
- problems, but this really shouldn't happen. The Falcon is not a TT. :)
-
- I would have done this originally, but for a number of reasons decided not to.
-
- One good reason is the difficulty of dynamically building an MMU tree half-way
- through a 32MB memory block, with unspecified length. I'll do my best to implement
- this when I have time.
-
- > Profile is just subtracting $1000000 from the end of detected fastram and coming
- > up with a figure 16 megs bigger than what is actually there. Not a good plan.
-
- > The only one, IMHO. Or how could you find out how much Fast RAM is there?
- > (maybe you don't need to know that, usually).
-
- They should use the memory pool to calculate used/free & total ram. They should
- not be using ramtop pointers, which cannot be relied upon.
-
- > So you can stop worrying about 19mb, and start worrying about 4 instead of 8!
-
- > So, _please_, try to modify your 68040.PRG to let the FastRAM begin at
- > $1000000, it helps me a lot and maybe people won't be confused by Profile
- > and other programs.
-
- To be honest, I knew about the TT address, but never believed any software would
- make such daring assumptions about the start of ALL kinds of fastram. In principle
- I should leave it alone out of protest, but you are right - people will be happier
- with it 'emulating' a TT as closely as possible. :)
-
- Doug.
-
-