17.04.2012 16:23, Bart Oldeman wrote:
> On 14 April 2012 07:17, Stas Sergeev <s...@list.ru> wrote:
>> I've just had a look into a possibility of using
>> LOWMEM() for the mappers, eliminating the
>> special meaning of MAPPING_LOWMEM and
>> MAPPING_HMA, which are adding lowmem_base
>> to the source. The mappers can use LOWMEM()
>> explicitly, instead of using the magic flags, because
>> the EMM can map from both the lowmem_base shm
>> and the malloc()'ed areas, so it needs to set the
>> source addr explicitly.
>> But it turned out that the mapping code is
>> over-complicated to do even that simple change.
> I wonder if the best way to solve the issue with using conventional
> memory to map EMS memory is to have the mapping code set up and
> maintain some kind of page table: (conv+hma)/4k=(1024+64)/4=272
> entries with pointers. That way, dosaddr_to_unixaddr just needs a
> quick table lookup like this:
>
> void* dosaddr_to_unixaddr(unsigned dosaddr)
> {
> /* use proper #define's in the real code ;) */
> if (dosaddr < (1024+64)*1024)
> return alias_table[dosaddr/4096] + (dosaddr & 4095);
> else
> /*DPMI memory, no alias */
> return &mem_base[dosaddr];
> }
Very probably; although we currently have kmem_map array
for the similar purpose with kmem: it would be good to not
have 2 lookup arrays, but just one.
I am currently busy with the mapping cleanup, have already
about a dozen of patches. :) dosaddr_to_unixaddr() is much
later (or you do).
------------------------------------------------------------------------------
Better than sec? Nothing is better than sec when it comes to
monitoring Big Data applications. Try Boundary one-second
resolution app monitoring today. Free.
http://p.sf.net/sfu/Boundary-dev2dev
_______________________________________________
Dosemu-devel mailing list
Dosemu-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dosemu-devel
