unwinding: perform more bounds checking

[japaric]

as seen in PR #184, the probe-rs API does not perform bounds checking so reading invalid (RAM) memory returns 0 (observed behavior so far; it may return junk with different probes). Thus we should do some bound checking on the probe-run side. That PR adds bound checking to the operation of "read stacked registers" but we should be more thorough and validate all memory operations (be it RAM or Flash). (The "read core register" operation does not need to be validated I think)

Instances that should be checked:

• read_word_32 in

  probe-run/src/registers.rs

  Line 70 in e8be32a

  self.cache.insert(reg.0, self.core.read_word_32(addr)?);

[Urhengulas]

Using VS Code global search, I found the following read operations (excluding read_core_reg):

• canary.rs

  • probe-run/src/canary.rs

    Line 359 in 6f3a3a4

    core.read_8(low_addr as u64, &mut buf)?;

  • probe-run/src/canary.rs

    Line 378 in 6f3a3a4

    .read_word_32(word_addr as u64)?

• main.rs

  • probe-run/src/main.rs

    Line 227 in 6f3a3a4

    core.read_32(rtt_buffer_address.into(), channel_flags)?;

• registers.rs

  • probe-run/src/registers.rs

    Line 71 in 6f3a3a4

    let value = self.core.read_word_32(addr.into())?;

• stacked.rs

  • probe-run/src/stacked.rs

    Line 67 in 6f3a3a4

    core.read_32(start.into(), registers)?;

---

canary.rs: Should be okay since the addresses are derived from StackInfo.

main.rs: Should be okay as well, since we get the address from the __SEGGER_RTT symbol.

registers.rs: It should be sufficient to store the StackInfo (or just the ram_range) in the Registers struct and use that for the bounds check.

stacked.rs: Already performs a bounds check.

---

Should write operations perform a bounds check as well?