cranelift: Implement TLS on aarch64 Mach-O (Apple Silicon) (#5434)

* Implement TLS on Aarch64 Mach-O

* Add aarch64 macho TLS filetest

* Address review comments

- `Aarch64` instead of `AArch64` in comments
- Remove unnecessary guard in tls_value lowering
- Remove unnecessary regalloc metadata in emission

* Use x1 as temporary register in emission

- Instead of passing in a temporary register to use when emitting
the TLS code, just use `x1`, as it's already in the clobber set.
This also keeps the size of `aarch64::inst::Inst` at 32 bytes.
- Update filetest accordingly

* Update aarch64 mach-o TLS filetest

cranelift/codegen/src/binemit/mod.rs

@@ -56,12 +56,20 @@ pub enum Reloc {
     /// Mach-O x86_64 32 bit signed PC relative offset to a `__thread_vars` entry.
     MachOX86_64Tlv,
 
-    /// AArch64 TLS GD
+    /// Mach-O Aarch64 TLS
+    /// PC-relative distance to the page of the TLVP slot.
+    MachOAarch64TlsAdrPage21,
+
+    /// Mach-O Aarch64 TLS
+    /// Offset within page of TLVP slot.
+    MachOAarch64TlsAdrPageOff12,
+
+    /// Aarch64 TLS GD
     /// Set an ADRP immediate field to the top 21 bits of the final address. Checks for overflow.
     /// This is equivalent to `R_AARCH64_TLSGD_ADR_PAGE21` in the [aaelf64](https://github.com/ARM-software/abi-aa/blob/2bcab1e3b22d55170c563c3c7940134089176746/aaelf64/aaelf64.rst#relocations-for-thread-local-storage)
     Aarch64TlsGdAdrPage21,
 
-    /// AArch64 TLS GD
+    /// Aarch64 TLS GD
     /// Set the add immediate field to the low 12 bits of the final address. Does not check for overflow.
     /// This is equivalent to `R_AARCH64_TLSGD_ADD_LO12_NC` in the [aaelf64](https://github.com/ARM-software/abi-aa/blob/2bcab1e3b22d55170c563c3c7940134089176746/aaelf64/aaelf64.rst#relocations-for-thread-local-storage)
     Aarch64TlsGdAddLo12Nc,
@@ -109,6 +117,8 @@ impl fmt::Display for Reloc {
 
             Self::ElfX86_64TlsGd => write!(f, "ElfX86_64TlsGd"),
             Self::MachOX86_64Tlv => write!(f, "MachOX86_64Tlv"),
+            Self::MachOAarch64TlsAdrPage21 => write!(f, "MachOAarch64TlsAdrPage21"),
+            Self::MachOAarch64TlsAdrPageOff12 => write!(f, "MachOAarch64TlsAdrPageOff12"),
             Self::Aarch64TlsGdAdrPage21 => write!(f, "Aarch64TlsGdAdrPage21"),
             Self::Aarch64TlsGdAddLo12Nc => write!(f, "Aarch64TlsGdAddLo12Nc"),
             Self::Aarch64AdrGotPage21 => write!(f, "Aarch64AdrGotPage21"),

cranelift/codegen/src/isa/aarch64/inst.isle

@@ -927,6 +927,10 @@
         (symbol ExternalName)
         (rd WritableReg))
 
+       (MachOTlsGetAddr
+        (symbol ExternalName)
+        (rd WritableReg))
+
        ;; An unwind pseudo-instruction.
        (Unwind
         (inst UnwindInst))
@@ -3591,6 +3595,12 @@
             (_ Unit (emit (MInst.ElfTlsGetAddr name dst))))
         dst))
 
+(decl macho_tls_get_addr (ExternalName) Reg)
+(rule (macho_tls_get_addr name)
+      (let ((dst WritableReg (temp_writable_reg $I64))
+            (_ Unit (emit (MInst.MachOTlsGetAddr name dst))))
+        dst))
+
 ;; A tuple of `ProducesFlags` and `IntCC`.
 (type FlagsAndCC (enum (FlagsAndCC (flags ProducesFlags)
                                    (cc IntCC))))

cranelift/codegen/src/isa/aarch64/inst/emit.rs

@@ -3458,6 +3458,54 @@ impl MachInstEmit for Inst {
                 sink.put4(0xd503201f);
             }
 
+            &Inst::MachOTlsGetAddr { ref symbol, rd } => {
+                // Each thread local variable gets a descriptor, where the first xword of the descriptor is a pointer
+                // to a function that takes the descriptor address in x0, and after the function returns x0
+                // contains the address for the thread local variable
+                //
+                // what we want to emit is basically:
+                //
+                // adrp x0, <label>@TLVPPAGE  ; Load the address of the page of the thread local variable pointer (TLVP)
+                // ldr x0, [x0, <label>@TLVPPAGEOFF] ; Load the descriptor's address into x0
+                // ldr x1, [x0] ; Load the function pointer (the first part of the descriptor)
+                // blr x1 ; Call the function pointer with the descriptor address in x0
+                // ; x0 now contains the TLV address
+
+                let rd = allocs.next_writable(rd);
+                assert_eq!(xreg(0), rd.to_reg());
+                let rtmp = writable_xreg(1);
+
+                // adrp x0, <label>@TLVPPAGE
+                sink.add_reloc(Reloc::MachOAarch64TlsAdrPage21, symbol, 0);
+                sink.put4(0x90000000);
+
+                // ldr x0, [x0, <label>@TLVPPAGEOFF]
+                sink.add_reloc(Reloc::MachOAarch64TlsAdrPageOff12, symbol, 0);
+                sink.put4(0xf9400000);
+
+                // load [x0] into temp register
+                Inst::ULoad64 {
+                    rd: rtmp,
+                    mem: AMode::reg(rd.to_reg()),
+                    flags: MemFlags::trusted(),
+                }
+                .emit(&[], sink, emit_info, state);
+
+                // call function pointer in temp register
+                Inst::CallInd {
+                    info: crate::isa::Box::new(CallIndInfo {
+                        rn: rtmp.to_reg(),
+                        uses: smallvec![],
+                        defs: smallvec![],
+                        clobbers: PRegSet::empty(),
+                        opcode: Opcode::CallIndirect,
+                        caller_callconv: CallConv::AppleAarch64,
+                        callee_callconv: CallConv::AppleAarch64,
+                    }),
+                }
+                .emit(&[], sink, emit_info, state);
+            }
+
             &Inst::Unwind { ref inst } => {
                 sink.add_unwind(inst.clone());
             }

cranelift/codegen/src/isa/aarch64/inst/mod.rs

@@ -907,6 +907,13 @@ fn aarch64_get_operands<F: Fn(VReg) -> VReg>(inst: &Inst, collector: &mut Operan
             clobbers.remove(regs::xreg_preg(0));
             collector.reg_clobbers(clobbers);
         }
+        &Inst::MachOTlsGetAddr { rd, .. } => {
+            collector.reg_fixed_def(rd, regs::xreg(0));
+            let mut clobbers =
+                AArch64MachineDeps::get_regs_clobbered_by_call(CallConv::AppleAarch64);
+            clobbers.remove(regs::xreg_preg(0));
+            collector.reg_clobbers(clobbers);
+        }
         &Inst::Unwind { .. } => {}
         &Inst::EmitIsland { .. } => {}
         &Inst::DummyUse { reg } => {
@@ -2701,6 +2708,10 @@ impl Inst {
                 let rd = pretty_print_reg(rd.to_reg(), allocs);
                 format!("elf_tls_get_addr {}, {}", rd, symbol.display(None))
             }
+            &Inst::MachOTlsGetAddr { ref symbol, rd } => {
+                let rd = pretty_print_reg(rd.to_reg(), allocs);
+                format!("macho_tls_get_addr {}, {}", rd, symbol.display(None))
+            }
             &Inst::Unwind { ref inst } => {
                 format!("unwind {:?}", inst)
             }

cranelift/codegen/src/isa/aarch64/lower.isle

@@ -2574,10 +2574,12 @@
 
 ;;; Rules for `tls_value` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
-(rule (lower (tls_value (symbol_value_data name _ _)))
-      (if (tls_model_is_elf_gd))
+(rule (lower (has_type (tls_model (TlsModel.ElfGd)) (tls_value (symbol_value_data name _ _))))
       (elf_tls_get_addr name))
 
+(rule (lower (has_type (tls_model (TlsModel.Macho)) (tls_value (symbol_value_data name _ _))))
+      (macho_tls_get_addr name))
+
 ;;; Rules for `fcvt_low_from_sint` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
 (rule (lower (has_type $F64X2 (fcvt_low_from_sint val)))

cranelift/filetests/filetests/isa/aarch64/tls-macho.clif

@@ -0,0 +1,58 @@
+test compile precise-output
+set tls_model=macho
+target aarch64
+
+function u0:0(i32) -> i32, i64 {
+gv0 = symbol colocated tls u1:0
+
+block0(v0: i32):
+    v1 = global_value.i64 gv0
+    return v0, v1
+}
+
+; VCode:
+;   stp fp, lr, [sp, #-16]!
+;   mov fp, sp
+;   str x24, [sp, #-16]!
+;   stp d14, d15, [sp, #-16]!
+;   stp d12, d13, [sp, #-16]!
+;   stp d10, d11, [sp, #-16]!
+;   stp d8, d9, [sp, #-16]!
+; block0:
+;   mov x24, x0
+;   macho_tls_get_addr x0, userextname0
+;   mov x1, x0
+;   mov x0, x24
+;   ldp d8, d9, [sp], #16
+;   ldp d10, d11, [sp], #16
+;   ldp d12, d13, [sp], #16
+;   ldp d14, d15, [sp], #16
+;   ldr x24, [sp], #16
+;   ldp fp, lr, [sp], #16
+;   ret
+; 
+; Disassembled:
+; block0: ; offset 0x0
+;   stp x29, x30, [sp, #-0x10]!
+;   mov x29, sp
+;   str x24, [sp, #-0x10]!
+;   stp d14, d15, [sp, #-0x10]!
+;   stp d12, d13, [sp, #-0x10]!
+;   stp d10, d11, [sp, #-0x10]!
+;   stp d8, d9, [sp, #-0x10]!
+; block1: ; offset 0x1c
+;   mov x24, x0
+;   adrp x0, #0 ; reloc_external MachOAarch64TlsAdrPage21 u1:0 0
+;   ldr x0, [x0] ; reloc_external MachOAarch64TlsAdrPageOff12 u1:0 0
+;   ldr x1, [x0]
+;   blr x1
+;   mov x1, x0
+;   mov x0, x24
+;   ldp d8, d9, [sp], #0x10
+;   ldp d10, d11, [sp], #0x10
+;   ldp d12, d13, [sp], #0x10
+;   ldp d14, d15, [sp], #0x10
+;   ldr x24, [sp], #0x10
+;   ldp x29, x30, [sp], #0x10
+;   ret
+

cranelift/object/src/backend.rs

@@ -654,6 +654,36 @@ impl ObjectModule {
                     32,
                 )
             }
+            Reloc::MachOAarch64TlsAdrPage21 => {
+                assert_eq!(
+                    self.object.format(),
+                    object::BinaryFormat::MachO,
+                    "MachOAarch64TlsAdrPage21 is not supported for this file format"
+                );
+                (
+                    RelocationKind::MachO {
+                        value: object::macho::ARM64_RELOC_TLVP_LOAD_PAGE21,
+                        relative: true,
+                    },
+                    RelocationEncoding::Generic,
+                    21,
+                )
+            }
+            Reloc::MachOAarch64TlsAdrPageOff12 => {
+                assert_eq!(
+                    self.object.format(),
+                    object::BinaryFormat::MachO,
+                    "MachOAarch64TlsAdrPageOff12 is not supported for this file format"
+                );
+                (
+                    RelocationKind::MachO {
+                        value: object::macho::ARM64_RELOC_TLVP_LOAD_PAGEOFF12,
+                        relative: false,
+                    },
+                    RelocationEncoding::Generic,
+                    12,
+                )
+            }
             Reloc::Aarch64TlsGdAdrPage21 => {
                 assert_eq!(
                     self.object.format(),