I have a multitasking kernel ring3 protection vm86 using software taskswitching. now I´m trying to implement each task to have it own page directory. util now ok!
but to load a flat binary image to a memory that is linked into 0x400000 I had to map that 0x400000 into physical addres that binary image is loaded, ok? good.
but now is my big question...
do I have to map the kernel memory into page directory for all tasks ?
if I do that works! ?
I just need to change CR3 on irq00 when before popa/pop ds... .. .. right ?
or I do need to change cr3 into all irq00-15 isr0-31 hander ?
I´m trying to read intel manual 3 but is big..
question about paging
Hi,
Some of what you ask is a matter of design preference, but yes, I would generally map the kernel region in to all memory spaces. If you do this, you will not need to reload CR3 on each interrupt / system call. And yes - simply change CR3 before you pop the registers of the incoming task.
This can also assist you with a method of not needing to change ESP0 in your TSS - the stack can be mapped in to the same location in each page directory (pointing to different physical RAM).
Just one issue - I personally advise against flat binaries. I know they initially make relocation slightly easier, but you will find a format such as ELF/COFF much better in the long run (accurate BSS size etc...).
Cheers,
Adam
Some of what you ask is a matter of design preference, but yes, I would generally map the kernel region in to all memory spaces. If you do this, you will not need to reload CR3 on each interrupt / system call. And yes - simply change CR3 before you pop the registers of the incoming task.
This can also assist you with a method of not needing to change ESP0 in your TSS - the stack can be mapped in to the same location in each page directory (pointing to different physical RAM).
Just one issue - I personally advise against flat binaries. I know they initially make relocation slightly easier, but you will find a format such as ELF/COFF much better in the long run (accurate BSS size etc...).
Cheers,
Adam
I personally would still suggest flat binaries multi-tasked first (and preferably even allocating virtual memory dynamically) before messing with stuff like ELF which has like 10 thousand new things that can go bogus. 
Yes, in the long run ELF is a good idea, but it's impossible to write software such as a kernel without doing one thing at a time.
Yes, in the long run ELF is a good idea, but it's impossible to write software such as a kernel without doing one thing at a time.
The real problem with goto is not with the control transfer, but with environments. Properly tail-recursive closures get both right.