Way back (again) -- farmalloc.

[Kennedy]

I'm stuck in a time warp. . .

I'm using 16-bit Dr. DOS and need to be able to move large amounts of memory. What I'm attempting to do is to grab up a huge block of memory (400K), copy a file into that block of memory, then copy that block of memory from the low memory to the upper memory (I'm writing my own version of loadlin but attempting to not use Assembly except in my high-move code).

From reading the data available to me in TC 3.5, using an char far *buf and farmalloc will do the trick for the first part. I'm guessing I need a near pointer to do my fread(), then use movedata() to copy the < 0x10000 chunks to the upper bound memory. Next, I'll move the far chunks by going PM.

I have, however, several questions to which I'm unable to find answers.

1) Will I be able to move between PM and RM without totally hosing up DOS, if I pusha and popa before and after I go into PM? I'm trying to not have to do assembly reads to the raw disk, but just use DOS to feed me data.

2) I've got to move some data (the Linux real mode code) into 0x90000. Does DOS use this as part of its far heap? It isn't a problem if it does, I'll just copy that data at the end, then make sure that I copy my code segment to a location out of the way (to a location that I know), then copy that segment to a lower location (thereby getting me out of the 0x90000 segment), then copy back up.

3) Will the far pointers map to memory in the ASM code correctly? (according to an old post of Bubba, there is a way to do this, but again, I just want to make sure that I'm not chasing my tail.)

Thanks in advance for you help/advice.
Andy

[Kennedy]

To add insult to injury: TC 3.0 does not recognize the e-extensions of the registers. Thus, I'm now having to go back and nasm the 32-bit code and load it into the data section of the code, then form a jmp to it in the code. . . Man oh man I really hate TC 3.0.

[iMalc]

What is it you're wanting to do all this for?

[Kennedy]

:::Sighs heavily:::

Man oh man, this is really a pain in the rear. . .

TC 3.0 allows inline asm, as we all know/expect. It *claims* to allow labels, however, it only allows labels in the past and apparently no forward labels. How can I, in the following code snip, declare the label before I call the label? Is this even possible?

        asm {
                pusha
                push    es
                push    ds
                push    cs
                mov     ax,hmnext
                push    ax
                mov     ax,newcs
                mov     bx,newip
                mov     di,bx
                mov     es,ax
                jmp     es:di
        hmnext: pop     ds
                pop     es
                popa
        }

I really really really hate TC 3.0.

[Kennedy]

My further woes with TC 3.0 have continued, to the point where I have preformed the following hack:

        asm {
                pusha
                push    es
                push    ds
                mov     ax,newcs
                mov     bx,newip
                dd      55aa55aah
                dd      90909090h
                dd      90909090h <=====  This repeats to 127 iterations.
                pop     ds
                pop     es
                popa
        }

After I compile the 16-bit code with TC 3.0, I then use a hex editor to delete the 512 bytes from 55aa55aa -> 90909090, then insert a 512 block of machine code from a file that saves the current code segment, then jumps to another location (defined by ax:bx).

For those who may wonder, 55aa55aa is just an identifier so that I can locate the block in the file. 90h, however, is the no-op statement (or xchg ax,ax) and is a safe instruction to use for padding.

Now, we'll just see if this works.

[Adak]

You know your relatives can use this stuff against you in your sanity hearing, don't you?

Good luck!

[Kennedy]

Just one more thing to add to the docket, I guess.