Right now, if layout.ld or target.json is changed, the firmware will
not automatically relink when running "cargo xbuild". This can make
working on layout.ld quite tedious, as you have to manually change the
Rust code to get the firmware to rebuild.

We solve this by adding a tiny build script which simply rebuilds the
program if either layout.ld or target.json changes.

We also stop using "-s" (aka "--strip-all") in the target options.
Instead, we add a second /DISCARD/ section in the linker script. This
puts all of our logic about sections in a single place. It also makes
it easier for a developer to get the symbols (if they want them).

Signed-off-by: Joe Richey <[email protected]>

This ensures that the stack region is mapped by the ELF loader and
doesn't conflict with any hypervisor memory regions. To do this without
increasing binary size, we place the stack right after the .bss section.

Note that we still need an assert to make sure that our minimal page
table setup in ram32.s covers our program + stack.

Also, being able to see the stack as a section makes debugging easier.

Signed-off-by: Joe Richey <[email protected]>

The structure of the ELF note can be done with pure Rust code. We can
definie the Name and Desc types and use a static struct to hold the
note.

Due to Rust's limitations on "pointer-to-integer cast", we have to have
Desc have a function pointer type, which means that field is now 8 bytes
long instead of 4. However, this doesn't seem to be an issue. The binary
still works w/ PVH Boot on QEMU and CH.

Signed-off-by: Joe Richey <[email protected]>

The GDT is just some data in static memory, so there's not a good
reason to have this code in assembly. Ideally, we would use types
from the x86_64 crate for this. However,

  - We can't use x86_64::structures::gdt::DescriptorFlags for the
    contents of the descriptors, as that definition is missing flags.
  - We can't use x86_64::structures::DescriptorTablePointer for the
    GDT pointers because its base is a u64, and Rust doesn't allow
    "pointer-to-integer cast" in statics.

So we have to roll our own. The definitions aren't that bad. The only
downside is that we have to add the bitflags dependency, but it was
already a dependancy of x86_64, so that's also not bad.

Signed-off-by: Joe Richey <[email protected]>

Signed-off-by: Joe Richey <[email protected]>