Add EL2 examples

[thejpster]

Adds another batch of examples which all run at EL2.

• HVC calls at EL2 go into the SVC handler
• Information about Data Aborts, Prefetch Aborts, Undefined Exceptions, etc, appear in the Hyp Syndrome Register, and not where EL1 expects to find them (DFSR, IFAR, etc)
• You return from an IRQ, HVC, or whatever, with an eret, so a while new batch of assembly trampolines were required
• A feature was added to aarch32-rt to enable these new EL2 trampolines. This feature also skips putting you into EL1 automatically.

[jonathanpallant]

Squashed and rebased.

[wt]

I was curious. Could we use naked functions to make this a little less ugly?

[thejpster]

I was curious. Could we use naked functions to make this a little less ugly?

Naked functions didn’t play nice with the interworking attribute. There was a bug and it only got fixed recently. But I guess it’s okay for the examples because they only build with a recent version of nightly.

[thejpster]

I tried making some examples use naked functions, but the compiler added prologue threw off my careful counting of bytes when it comes to verifying where the exception was throw. So I don't think it makes it better.

[wt]

Is this a bug in naked function support in the compiler?

[wt]

BTW, otherwise LGTM.

[thejpster]

Is this a bug in naked function support in the compiler?

Ah, I was wrong. It's not the prologue (there isn't one). It's that naked functions with the instruction_set(arm::t32) attribute get marked as Arm functions in the symbol table, so the processor jumps to the function in the wrong mode. This was rust-lang/rust#151946, which is fixed if I move up to a new enough nightly version.

In this case I was testing the UDF trap as a naked function, and it did trap, but not because it hit a UDF instruction but instead because it hit a Thumb instruction in Arm mode. The trap handler didn't like the error that was reported.

TL;DR: Naked functions are OK if I use a newer nightly Rust. This is changed over in #111.

[thejpster]

TFW PRs have been open so long you've forgotten about bugs, and the fact you already fixed them months ago, and debug them all over again.