sigma0 memory space

[Adam Lackorzynski]

On Wed May 29, 2013 at 14:52:26 +0300, joel at nider.org wrote:
> I have been reading about the structure of L4Ka::Pistachio, and
> discovered that they use idempotent (1:1) page table mapping in
> their sigma0 process.  I am wondering if Fiasco also takes this
> approach?

Sigma0 uses 1:1 virt-phys mappings.

> If so, does the kernel also use idempotent mapping?

The kernel is linked to the upper 1G of the address space, so no.

> I would imagine that this would have some implications on demand
> paging in the kernel.  Does the kernel even use demand paging?  (I

No, it does not.

> mean at some point during boot the kernel will start to use memory
> to store kernel objects - either the kernel 'guesses' how much it
> will ever need and statically allocates that memory, or it's going
> to have to ask for pages later on, which will be owned by some other
> process).
> 
> I tried to read the code, but found it very hard to traverse - the
> C++ classes don't make it easy to follow a flow.  Can you please
> point out the file(s) where the kernel sets up the memory space for
> sigma0, and where it passes the initial ownership (from kernel space
> to sigma0 space)?

Setup happens in kernel_thread-std.cpp, the actual mappings are
generated in function handle_sigma0_page_fault() which creates identity
mappings.



Adam
-- 
Adam                 adam at os.inf.tu-dresden.de
  Lackorzynski         http://os.inf.tu-dresden.de/~adam/