Fix multiple issues related to generic-sandbox when running on MacOS aarch64

[aman4150]

generic-sandbox: Fix signal handling on macOS

ucontext_t is different on macOS and Linux. On macOS, SIGSEGV is not
sent for page faults, instead SIGBUS is sent. This commit updates the
signal handler to check for SIGBUS on macOS.

Finally, we also fix the page fault address extraction on macOS.

Signed-off-by: Aman <[email protected]>

---

generic-sandbox: Fix invalid jump detection on macOS / Rosetta 2

On Rosetta 2, indirect jump to a non-canonical address doesn't trigger
a page fault (SIGBUS) right away. Instead, it jumps to the address, and then
triggers a page fault (SIGBUS) when trying to fetch the instruction.

This means that previous program counter is now lost, and now we have no
way of detecting if the guest program triggered an invalid jump or not.

We fix this program by updating the vmctx before making the jump and in
the signal handler we check if the current program counter is non-canonical.

Signed-off-by: Aman <[email protected]>

---

ci: Run generic-sandbox tests in CI on aarch64 MacOS using Rosetta 2

Signed-off-by: Aman <[email protected]>

[koute]

Please remove this.

We deliberately keep the number of dependencies to a minimum, especially dependencies which don't really bring any extra functionality and are just quality-of-life improvements like paste here it. Currently the only external dependency we have is the log crate, and I'd like to keep it that way. (The schnellru crate is my own crate, so it doesn't count. :P)

You should be able to do something like this instead of using the paste crate:

#[cfg(target_os = "macos")]
macro_rules! reg_field {
    (rax) => {__rax};
    (rbx) => {__rbx};
}

#[cfg(not(target_os = "macos"))]
macro_rules! reg_field {
    ($reg:ident) => {$reg};
}

And then use reg_field! in the fetch_reg.

[aman4150]

Sure.