Merge #1183

1183: Make full preemption an optional feature. r=syrusakbary a=losfair

Full preemption requires two additional memory loads on loop backedges and function calls. This PR allows disabling full preemption at code generation time, and disables it by default.

Co-authored-by: losfair <zhy20000919@hotmail.com>
Co-authored-by: Heyang Zhou <zhy20000919@hotmail.com>

lib/runtime-core/src/backend.rs

@@ -109,9 +109,28 @@ impl BackendCompilerConfig {
 pub struct CompilerConfig {
     /// Symbol information generated from emscripten; used for more detailed debug messages
     pub symbol_map: Option<HashMap<u32, String>>,
+
+    /// How to make the decision whether to emit bounds checks for memory accesses.
     pub memory_bound_check_mode: MemoryBoundCheckMode,
+
+    /// Whether to generate explicit native stack checks against `stack_lower_bound` in `InternalCtx`.
+    ///
+    /// Usually it's adequate to use hardware memory protection mechanisms such as `mprotect` on Unix to
+    /// prevent stack overflow. But for low-level environments, e.g. the kernel, faults are generally
+    /// not expected and relying on hardware memory protection would add too much complexity.
     pub enforce_stack_check: bool,
+
+    /// Whether to enable state tracking. Necessary for managed mode.
     pub track_state: bool,
+
+    /// Whether to enable full preemption checkpoint generation.
+    ///
+    /// This inserts checkpoints at critical locations such as loop backedges and function calls,
+    /// allowing preemptive unwinding/task switching.
+    ///
+    /// When enabled there can be a small amount of runtime performance overhead.
+    pub full_preemption: bool,
+
     pub features: Features,
 
     // Target info. Presently only supported by LLVM.