Message-Id: <s1ebb8a4.084@MAIL.TAIT.CO.NZ>
Date: Tue, 16 Jul 1996 15:43:21 +1200
From: Bill Currie <bill_currie AT MAIL DOT TAIT DOT CO DOT NZ>
To: djgpp AT delorie DOT com
Subject:  Re: Locking RAM for hardware interrupts

On 16/7/96  3:43 am, Orbital <vazndain AT zveris> did thus say >>>
> 
> 
> On Fri, 12 Jul 1996, Bruce Foley wrote:
> 
[snip]
> 
>   Now, I would like to continue with the FAT-DS thread. If the problem was
>   memory protection and not the NT and Linux DOS-box compatibility, then
>   you could just enable near pointers, do your stuff and then disable them.
>   However, this is slow, because __djgpp_nearptr.. functions issue DPMI calls
>   etc. I found a cool workaround for that:
>   
>   [1] Allocate an alias to your data descriptor and set its limit to 4GB.
>   [2] When you need near pointers, load ds and es with the alias.
>   [3] When done reload ds and es with the original selector.
> 
>   Voila. Steps [2] and [3] are much faster than __djgpp_nearptr.. calls. To
>   be more precise, there is no need to allocate the alias, it's already 
>   there, stored in __djgpp_ds_alias variable. Just fatten it and that's it.
>   There is also no need to store the original one, as it is stored in a
>   variable __djgpp_app_DS. Both of them are unsigned short ints. Now, a
>   nearptr_enavble() would look like this:
>   
>     asm volatile ("mov __djgpp_dos_alias,%es; mov __djgpp_dos_alias,%ds");
> 
>   and nearptr_disable() would be:
> 
>     asm volatile ("mov __djgpp_app_DS,%es; mov __djgpp_app_DS,%ds");
> 
>   Plus it would take a single call to
>  
>     __dpmi_set_selector_limit(__djgpp_ds_alias, -1);
> 
>   somewhere at startup. Hope someone finds this useful. Tell me if I'm
>   wrong somewhere as I have not tested it myself.
[snip]
> 
>   Enough for now. You must be pretty cool if you managed to read up to here.
>   Man, this *is* a long message. Sorry if I didn't write anything interesting. 
>   

You have GOT to be kidding!! Pretty cool idea, I'll have to check it out.  The
only possible problem that I can see is if the ds alias is ever different to the
app ds, but I believe that only happens during sbrk calls (unix mode), so the app
will never see them being different.

>   
>   Orbital
> 
>