RE: Invoking int 0x10 from a service

Hello Alex,

> I'm currently planning to write a VESA driver for our Pistachio-based
> OS. Video mode setting is done by calling the BIOS interrupt 0x10,
> which uses real-mode addressing, and therefore cannot be run from
> protected mode. What I thought about was to use VM86 mode for that
> call, but it looks like there is no way to do that with Pistachio.

You're right, Pistachio doesn't offer support for vm86 mode. I don't know
too much about VESA drivers, but I am not sure if vm86 would solve your
problem actually. vm86 mode only allows you to run 16-bit user-mode code; on
interrupts or exceptions, the processor automatically enters the kernel and
switches back to protected mode again. Thus, executing int $0x10 in a vm86
compartment should bring you back into the kernel rather than to the
corresponding BIOS function. I guess what you really need is to switch to
real mode and then call int $0x10??? 

> Have I missed something? Do you know a way to workaround this without
> modifying Pistachio?

I can't imagine a solution that doesn't involve modifying the kernel. We do
have support for compatibility 32-bit mode on x64 processors, though, and it
is pretty similar to vm86 on x32 in that it also uses different segment
descriptors. For x64 we used a bit in space_control to denote that this
space should be running in compatibility 32-bit mode. We then changed the
segments for threads within that particular address space, and offer a
32-bit KIP and code to translate system calls (e.g., IPC) into their 64-bit
counterparts. For vm86 support you could start from there. For real-mode
support, the first hack that comes into my mind is to allow privileged tasks
to somehow switch to real mode (e.g, via processor_control) and provide an
entry point to switch back to protected mode. 

-Jan


--
Jan Stoess
System Architecture Group
University of Karlsruhe