Kernel tries to send syscall to itself

[kennystrawnmusic]

Going to need to write my own Dma<T> and Mmio<T> structures instead of relying on the ones provided by the redox_syscall crate after running objdump on the instruction pointer returned by the #UD fault in the AHCI driver:

$ objdump -d target/x86_64-uefi-cryptos/release/cryptos | grep 56aae
   56aae:	0f 05                	syscall

Mmio<T> is easy to rewrite ― had one such structure embedded in my kernel myself temporarily before I began to run into compiler errors for whatever reason while trying to use it. Dma<T> not so much ― it's got some dependencies that need to be replaced with lower-level abstractions. Opening this issue as a reminder in the meantime.

[kennystrawnmusic]

Here's the complete call stack that needs to be worked around:

1. Dma::zeroed() -> Dma::from_physbox::zeroed()
2. Dma::from_physbox::zeroed() -> Dma::from_physbox::uninit()
3. Dma::from_physbox::uninit() -> redox_syscall::call::physmap()
4. redox_syscall::call::physmap() -> redox_syscall::syscall3()
5. redox_syscall::syscall3() -> asm!("syscall //...

Executing the syscall assembly instruction from inside the kernel is a #UD waiting to happen. So of course this needs to be worked around with re-implementation.