jit/x86: add emit_indirect_branch_target #288
Conversation
riptl
force-pushed
the
anchor-endbr64
branch
3 times, most recently
from
e36a585
to
279e7d9
Compare
| @@ -866,6 +867,11 @@ fn emit_set_exception_kind<E: UserDefinedError>(jit: &mut JitCompiler, err: Ebpf | |||
| X86Instruction::store_immediate(OperandSize::S64, R10, X86IndirectAccess::Offset(8), err_kind as i64).emit(jit) | |||
| } | |||
|
|
|||
| #[inline] | |||
| fn emit_indirect_branch_target<E: UserDefinedError>(jit: &mut JitCompiler) -> Result<(), EbpfError<E>> { | |||
| X86Instruction::end_branch().emit(jit) | |||
There was a problem hiding this comment.
Should we put a feature flag here?
There was a problem hiding this comment.
Not necessary, in fact I would remove the entire emit_indirect_branch_target function and only let the X86Instruction::end_branch() in.
|
Why put the Edit: I would expect |
|
I think I'm fundamentally misunderstanding call and return sites in the JIT. 😅 @Lichtso What do you think of just addressing a small PR first that adds the |
riptl
changed the title
You're right. Updated PR to omit this change. |
| @@ -96,6 +98,11 @@ impl Default for X86Instruction { | |||
|
|
|||
| impl X86Instruction { | |||
| pub fn emit<E: UserDefinedError>(&self, jit: &mut JitCompiler) -> Result<(), EbpfError<E>> { | |||
| match self.repeat_string { | |||
There was a problem hiding this comment.
Maybe turn this into an enum:
- Repeat Non Zero
- Repeat Zero
- Don't Repeat
| @@ -866,6 +867,11 @@ fn emit_set_exception_kind<E: UserDefinedError>(jit: &mut JitCompiler, err: Ebpf | |||
| X86Instruction::store_immediate(OperandSize::S64, R10, X86IndirectAccess::Offset(8), err_kind as i64).emit(jit) | |||
| } | |||
|
|
|||
| #[inline] | |||
| fn emit_indirect_branch_target<E: UserDefinedError>(jit: &mut JitCompiler) -> Result<(), EbpfError<E>> { | |||
| X86Instruction::end_branch().emit(jit) | |||
There was a problem hiding this comment.
Not necessary, in fact I would remove the entire emit_indirect_branch_target function and only let the X86Instruction::end_branch() in.
| @@ -980,6 +986,22 @@ impl JitCompiler { | |||
| }) | |||
| } | |||
|
|
|||
| #[cfg(test)] | |||
| #[allow(dead_code)] | |||
| pub(crate) fn new_mock() -> Self { | |||
There was a problem hiding this comment.
Should be moved over in x86.rs mod test { ... } as this is only used by the instruction emitter.
|
|
||
| #[cfg(test)] | ||
| #[allow(dead_code)] | ||
| pub(crate) fn get_text_result(&self) -> &[u8] { |
There was a problem hiding this comment.
Probably not needed, just make the result accessible crate wide.
|
Closing for now as we currently have no need for it and it got unresolved conflicts. |
This PR adds the ability to emit the
endbr64instruction to begin the work on Indirect Branch Tracking (IBT).This does not introduce any functional changes to the JIT compiler.
The bulk of the work to enable IBT, which involves adding
endbr64s to code translated from SBF, is not implemented in this PR.Relates to #287
Changes
JitCompiler::new_mock()andJitCompiler::get_text_result()methods to allow test case assertions on x86-compiled code.emit_indirect_branch_targetfunction to emitendbr64per Intel CET.