diff --git a/cranelift/codegen/src/isa/aarch64/abi.rs b/cranelift/codegen/src/isa/aarch64/abi.rs
index 235d9031a8..9439ef55df 100644
--- a/cranelift/codegen/src/isa/aarch64/abi.rs
+++ b/cranelift/codegen/src/isa/aarch64/abi.rs
@@ -73,9 +73,9 @@
 //! https://searchfox.org/mozilla-central/rev/bc3600def806859c31b2c7ac06e3d69271052a89/js/src/wasm/WasmStubs.h#134
 //!
 //! In brief:
-//! - Returns are processed in *reverse* order.
-//! - The first return in this order (so the last return) goes into the ordinary
-//!   return register, X0.
+//! - Return values are processed in *reverse* order.
+//! - The first return value in this order (so the last return) goes into the
+//!   ordinary return register, X0.
 //! - Any further returns go in a struct-return area, allocated upwards (in
 //!   address order) during the reverse traversal.
 //! - This struct-return area is provided by the caller, and a pointer to its
@@ -98,6 +98,7 @@ use crate::isa;
 use crate::isa::aarch64::{inst::*, lower::ty_bits};
 use crate::machinst::*;
 use crate::settings;
+use crate::{CodegenError, CodegenResult};
 
 use alloc::boxed::Box;
 use alloc::vec::Vec;
@@ -134,6 +135,11 @@ struct ABISig {
     call_conv: isa::CallConv,
 }
 
+/// This is the limit for the size of argument and return-value areas on the
+/// stack. We place a reasonable limit here to avoid integer overflow issues
+/// with 32-bit arithmetic: for now, 128 MB.
+static STACK_ARG_RET_SIZE_LIMIT: u64 = 128 * 1024 * 1024;
+
 // Spidermonkey specific ABI convention.
 
 /// This is SpiderMonkey's `WasmTableCallSigReg`.
@@ -208,14 +214,15 @@ enum ArgsOrRets {
 /// Process a list of parameters or return values and allocate them to X-regs,
 /// V-regs, and stack slots.
 ///
-/// Returns the list of argument locations, and the stack-space used (rounded up
-/// to a 16-byte-aligned boundary).
+/// Returns the list of argument locations, the stack-space used (rounded up
+/// to a 16-byte-aligned boundary), and if `add_ret_area_ptr` was passed, the
+/// index of the extra synthetic arg that was added.
 fn compute_arg_locs(
     call_conv: isa::CallConv,
     params: &[ir::AbiParam],
     args_or_rets: ArgsOrRets,
     add_ret_area_ptr: bool,
-) -> (Vec<ABIArg>, i64) {
+) -> CodegenResult<(Vec<ABIArg>, i64, Option<usize>)> {
     let is_baldrdash = call_conv.extends_baldrdash();
 
     // See AArch64 ABI (https://c9x.me/compile/bib/abi-arm64.pdf), sections 5.4.
@@ -290,7 +297,7 @@ fn compute_arg_locs(
         ret.reverse();
     }
 
-    if add_ret_area_ptr {
+    let extra_arg = if add_ret_area_ptr {
         debug_assert!(args_or_rets == ArgsOrRets::Args);
         if next_xreg < max_reg_vals {
             ret.push(ABIArg::Reg(xreg(next_xreg).to_real_reg(), I64));
@@ -298,35 +305,39 @@ fn compute_arg_locs(
             ret.push(ABIArg::Stack(next_stack as i64, I64));
             next_stack += 8;
         }
-    }
+        Some(ret.len() - 1)
+    } else {
+        None
+    };
 
     next_stack = (next_stack + 15) & !15;
 
-    (ret, next_stack as i64)
+    // To avoid overflow issues, limit the arg/return size to something
+    // reasonable -- here, 128 MB.
+    if next_stack > STACK_ARG_RET_SIZE_LIMIT {
+        return Err(CodegenError::ImplLimitExceeded);
+    }
+
+    Ok((ret, next_stack as i64, extra_arg))
 }
 
 impl ABISig {
-    fn from_func_sig(sig: &ir::Signature) -> ABISig {
+    fn from_func_sig(sig: &ir::Signature) -> CodegenResult<ABISig> {
         // Compute args and retvals from signature. Handle retvals first,
         // because we may need to add a return-area arg to the args.
-        let (rets, stack_ret_space) = compute_arg_locs(
+        let (rets, stack_ret_space, _) = compute_arg_locs(
             sig.call_conv,
             &sig.returns,
             ArgsOrRets::Rets,
             /* extra ret-area ptr = */ false,
-        );
+        )?;
         let need_stack_return_area = stack_ret_space > 0;
-        let (args, stack_arg_space) = compute_arg_locs(
+        let (args, stack_arg_space, stack_ret_arg) = compute_arg_locs(
             sig.call_conv,
             &sig.params,
             ArgsOrRets::Args,
             need_stack_return_area,
-        );
-        let stack_ret_arg = if need_stack_return_area {
-            Some(args.len() - 1)
-        } else {
-            None
-        };
+        )?;
 
         trace!(
             "ABISig: sig {:?} => args = {:?} rets = {:?} arg stack = {} ret stack = {} stack_ret_arg = {:?}",
@@ -338,14 +349,14 @@ impl ABISig {
             stack_ret_arg
         );
 
-        ABISig {
+        Ok(ABISig {
             args,
             rets,
             stack_arg_space,
             stack_ret_space,
             stack_ret_arg,
             call_conv: sig.call_conv,
-        }
+        })
     }
 }
 
@@ -446,15 +457,7 @@ fn gen_stack_limit(f: &ir::Function, abi: &ABISig, gv: ir::GlobalValue) -> (Reg,
             } => {
                 let base = generate_gv(f, abi, base, insts);
                 let into_reg = writable_spilltmp_reg();
-                let mem = if let Some(offset) =
-                    UImm12Scaled::maybe_from_i64(offset.into(), ir::types::I8)
-                {
-                    MemArg::UnsignedOffset(base, offset)
-                } else {
-                    let offset: i64 = offset.into();
-                    insts.extend(Inst::load_constant(into_reg, offset as u64));
-                    MemArg::RegReg(base, into_reg.to_reg())
-                };
+                let mem = MemArg::RegOffset(base, offset.into(), I64);
                 insts.push(Inst::ULoad64 {
                     rd: into_reg,
                     mem,
@@ -481,10 +484,10 @@ fn get_special_purpose_param_register(
 
 impl AArch64ABIBody {
     /// Create a new body ABI instance.
-    pub fn new(f: &ir::Function, flags: settings::Flags) -> Self {
+    pub fn new(f: &ir::Function, flags: settings::Flags) -> CodegenResult<Self> {
         debug!("AArch64 ABI: func signature {:?}", f.signature);
 
-        let sig = ABISig::from_func_sig(&f.signature);
+        let sig = ABISig::from_func_sig(&f.signature)?;
 
         let call_conv = f.signature.call_conv;
         // Only these calling conventions are supported.
@@ -517,7 +520,7 @@ impl AArch64ABIBody {
                 .map(|reg| (reg, Vec::new()))
                 .or_else(|| f.stack_limit.map(|gv| gen_stack_limit(f, &sig, gv)));
 
-        Self {
+        Ok(Self {
             sig,
             stackslots,
             stackslots_size: stack_offset,
@@ -529,7 +532,7 @@ impl AArch64ABIBody {
             flags,
             is_leaf: f.is_leaf(),
             stack_limit,
-        }
+        })
     }
 
     /// Returns the offset from FP to the argument area, i.e., jumping over the saved FP, return
@@ -635,15 +638,22 @@ impl AArch64ABIBody {
 
 fn load_stack(mem: MemArg, into_reg: Writable<Reg>, ty: Type) -> Inst {
     match ty {
-        types::B1
-        | types::B8
-        | types::I8
-        | types::B16
-        | types::I16
-        | types::B32
-        | types::I32
-        | types::B64
-        | types::I64 => Inst::ULoad64 {
+        types::B1 | types::B8 | types::I8 => Inst::ULoad8 {
+            rd: into_reg,
+            mem,
+            srcloc: None,
+        },
+        types::B16 | types::I16 => Inst::ULoad16 {
+            rd: into_reg,
+            mem,
+            srcloc: None,
+        },
+        types::B32 | types::I32 => Inst::ULoad32 {
+            rd: into_reg,
+            mem,
+            srcloc: None,
+        },
+        types::B64 | types::I64 => Inst::ULoad64 {
             rd: into_reg,
             mem,
             srcloc: None,
@@ -664,15 +674,22 @@ fn load_stack(mem: MemArg, into_reg: Writable<Reg>, ty: Type) -> Inst {
 
 fn store_stack(mem: MemArg, from_reg: Reg, ty: Type) -> Inst {
     match ty {
-        types::B1
-        | types::B8
-        | types::I8
-        | types::B16
-        | types::I16
-        | types::B32
-        | types::I32
-        | types::B64
-        | types::I64 => Inst::Store64 {
+        types::B1 | types::B8 | types::I8 => Inst::Store8 {
+            rd: from_reg,
+            mem,
+            srcloc: None,
+        },
+        types::B16 | types::I16 => Inst::Store16 {
+            rd: from_reg,
+            mem,
+            srcloc: None,
+        },
+        types::B32 | types::I32 => Inst::Store32 {
+            rd: from_reg,
+            mem,
+            srcloc: None,
+        },
+        types::B64 | types::I64 => Inst::Store64 {
             rd: from_reg,
             mem,
             srcloc: None,
@@ -791,17 +808,14 @@ fn get_caller_saves(call_conv: isa::CallConv) -> Vec<Writable<Reg>> {
 impl ABIBody for AArch64ABIBody {
     type I = Inst;
 
-    fn needed_tmps(&self) -> usize {
-        if self.sig.stack_ret_arg.is_some() {
-            1
-        } else {
-            0
-        }
+    fn temp_needed(&self) -> bool {
+        self.sig.stack_ret_arg.is_some()
     }
 
-    fn init_with_tmps(&mut self, tmps: &[Writable<Reg>]) {
+    fn init(&mut self, maybe_tmp: Option<Writable<Reg>>) {
         if self.sig.stack_ret_arg.is_some() {
-            self.ret_area_ptr = Some(tmps[0]);
+            assert!(maybe_tmp.is_some());
+            self.ret_area_ptr = maybe_tmp;
         }
     }
 
@@ -845,14 +859,14 @@ impl ABIBody for AArch64ABIBody {
         match &self.sig.args[idx] {
             &ABIArg::Reg(r, ty) => Inst::gen_move(into_reg, r.to_reg(), ty),
             &ABIArg::Stack(off, ty) => load_stack(
-                MemArg::FPOffset(self.fp_to_arg_offset() + off),
+                MemArg::FPOffset(self.fp_to_arg_offset() + off, ty),
                 into_reg,
                 ty,
             ),
         }
     }
 
-    fn gen_retval_area_setup(&self) -> Vec<Inst> {
+    fn gen_retval_area_setup(&self) -> Option<Inst> {
         if let Some(i) = self.sig.stack_ret_arg {
             let inst = self.gen_copy_arg_to_reg(i, self.ret_area_ptr.unwrap());
             trace!(
@@ -860,10 +874,10 @@ impl ABIBody for AArch64ABIBody {
                 inst,
                 self.ret_area_ptr.unwrap().to_reg()
             );
-            vec![inst]
+            Some(inst)
         } else {
             trace!("gen_retval_area_setup: not needed");
-            vec![]
+            None
         }
     }
 
@@ -924,8 +938,7 @@ impl ABIBody for AArch64ABIBody {
                     }
                     _ => {}
                 };
-                let mem = MemArg::reg_maybe_offset(self.ret_area_ptr.unwrap().to_reg(), off, ty)
-                    .expect("Return-value area is too large");
+                let mem = MemArg::RegOffset(self.ret_area_ptr.unwrap().to_reg(), off, ty);
                 ret.push(store_stack(mem, from_reg.to_reg(), ty))
             }
         }
@@ -961,7 +974,7 @@ impl ABIBody for AArch64ABIBody {
         let stack_off = self.stackslots[slot.as_u32() as usize] as i64;
         let sp_off: i64 = stack_off + (offset as i64);
         trace!("load_stackslot: slot {} -> sp_off {}", slot, sp_off);
-        load_stack(MemArg::NominalSPOffset(sp_off), into_reg, ty)
+        load_stack(MemArg::NominalSPOffset(sp_off, ty), into_reg, ty)
     }
 
     /// Store to a stackslot.
@@ -971,7 +984,7 @@ impl ABIBody for AArch64ABIBody {
         let stack_off = self.stackslots[slot.as_u32() as usize] as i64;
         let sp_off: i64 = stack_off + (offset as i64);
         trace!("store_stackslot: slot {} -> sp_off {}", slot, sp_off);
-        store_stack(MemArg::NominalSPOffset(sp_off), from_reg, ty)
+        store_stack(MemArg::NominalSPOffset(sp_off, ty), from_reg, ty)
     }
 
     /// Produce an instruction that computes a stackslot address.
@@ -982,7 +995,7 @@ impl ABIBody for AArch64ABIBody {
         let sp_off: i64 = stack_off + (offset as i64);
         Inst::LoadAddr {
             rd: into_reg,
-            mem: MemArg::NominalSPOffset(sp_off),
+            mem: MemArg::NominalSPOffset(sp_off, I8),
         }
     }
 
@@ -993,7 +1006,7 @@ impl ABIBody for AArch64ABIBody {
         let spill_off = islot * 8;
         let sp_off = self.stackslots_size as i64 + spill_off;
         trace!("load_spillslot: slot {:?} -> sp_off {}", slot, sp_off);
-        load_stack(MemArg::NominalSPOffset(sp_off), into_reg, ty)
+        load_stack(MemArg::NominalSPOffset(sp_off, ty), into_reg, ty)
     }
 
     /// Store to a spillslot.
@@ -1003,7 +1016,7 @@ impl ABIBody for AArch64ABIBody {
         let spill_off = islot * 8;
         let sp_off = self.stackslots_size as i64 + spill_off;
         trace!("store_spillslot: slot {:?} -> sp_off {}", slot, sp_off);
-        store_stack(MemArg::NominalSPOffset(sp_off), from_reg, ty)
+        store_stack(MemArg::NominalSPOffset(sp_off, ty), from_reg, ty)
     }
 
     fn gen_prologue(&mut self) -> Vec<Inst> {
@@ -1290,17 +1303,17 @@ impl AArch64ABICall {
         extname: &ir::ExternalName,
         dist: RelocDistance,
         loc: ir::SourceLoc,
-    ) -> AArch64ABICall {
-        let sig = ABISig::from_func_sig(sig);
+    ) -> CodegenResult<AArch64ABICall> {
+        let sig = ABISig::from_func_sig(sig)?;
         let (uses, defs) = abisig_to_uses_and_defs(&sig);
-        AArch64ABICall {
+        Ok(AArch64ABICall {
             sig,
             uses,
             defs,
             dest: CallDest::ExtName(extname.clone(), dist),
             loc,
             opcode: ir::Opcode::Call,
-        }
+        })
     }
 
     /// Create a callsite ABI object for a call to a function pointer with the
@@ -1310,17 +1323,17 @@ impl AArch64ABICall {
         ptr: Reg,
         loc: ir::SourceLoc,
         opcode: ir::Opcode,
-    ) -> AArch64ABICall {
-        let sig = ABISig::from_func_sig(sig);
+    ) -> CodegenResult<AArch64ABICall> {
+        let sig = ABISig::from_func_sig(sig)?;
         let (uses, defs) = abisig_to_uses_and_defs(&sig);
-        AArch64ABICall {
+        Ok(AArch64ABICall {
             sig,
             uses,
             defs,
             dest: CallDest::Reg(ptr),
             loc,
             opcode,
-        }
+        })
     }
 }
 
@@ -1394,7 +1407,9 @@ impl ABICall for AArch64ABICall {
                 from_reg,
                 ty,
             )),
-            &ABIArg::Stack(off, ty) => ctx.emit(store_stack(MemArg::SPOffset(off), from_reg, ty)),
+            &ABIArg::Stack(off, ty) => {
+                ctx.emit(store_stack(MemArg::SPOffset(off, ty), from_reg, ty))
+            }
         }
     }
 
@@ -1409,7 +1424,7 @@ impl ABICall for AArch64ABICall {
             &ABIArg::Stack(off, ty) => {
                 let ret_area_base = self.sig.stack_arg_space;
                 ctx.emit(load_stack(
-                    MemArg::SPOffset(off + ret_area_base),
+                    MemArg::SPOffset(off + ret_area_base, ty),
                     into_reg,
                     ty,
                 ));
@@ -1427,7 +1442,7 @@ impl ABICall for AArch64ABICall {
             let ret_area_base = self.sig.stack_arg_space;
             ctx.emit(Inst::LoadAddr {
                 rd,
-                mem: MemArg::SPOffset(ret_area_base),
+                mem: MemArg::SPOffset(ret_area_base, I8),
             });
             self.emit_copy_reg_to_arg(ctx, i, rd.to_reg());
         }
diff --git a/cranelift/codegen/src/isa/aarch64/inst/args.rs b/cranelift/codegen/src/isa/aarch64/inst/args.rs
index 0ea61a0404..dd41912479 100644
--- a/cranelift/codegen/src/isa/aarch64/inst/args.rs
+++ b/cranelift/codegen/src/isa/aarch64/inst/args.rs
@@ -145,11 +145,15 @@ pub enum MemArg {
     /// Reference to a "label": e.g., a symbol.
     Label(MemLabel),
 
+    /// Arbitrary offset from a register. Converted to generation of large
+    /// offsets with multiple instructions as necessary during code emission.
+    RegOffset(Reg, i64, Type),
+
     /// Offset from the stack pointer.
-    SPOffset(i64),
+    SPOffset(i64, Type),
 
     /// Offset from the frame pointer.
-    FPOffset(i64),
+    FPOffset(i64, Type),
 
     /// Offset from the "nominal stack pointer", which is where the real SP is
     /// just after stack and spill slots are allocated in the function prologue.
@@ -163,7 +167,7 @@ pub enum MemArg {
     /// SP" is where the actual SP is after the function prologue and before
     /// clobber pushes. See the diagram in the documentation for
     /// [crate::isa::aarch64::abi](the ABI module) for more details.
-    NominalSPOffset(i64),
+    NominalSPOffset(i64, Type),
 }
 
 impl MemArg {
@@ -174,17 +178,6 @@ impl MemArg {
         MemArg::UnsignedOffset(reg, UImm12Scaled::zero(I64))
     }
 
-    /// Memory reference using an address in a register and an offset, if possible.
-    pub fn reg_maybe_offset(reg: Reg, offset: i64, value_type: Type) -> Option<MemArg> {
-        if let Some(simm9) = SImm9::maybe_from_i64(offset) {
-            Some(MemArg::Unscaled(reg, simm9))
-        } else if let Some(uimm12s) = UImm12Scaled::maybe_from_i64(offset, value_type) {
-            Some(MemArg::UnsignedOffset(reg, uimm12s))
-        } else {
-            None
-        }
-    }
-
     /// Memory reference using the sum of two registers as an address.
     pub fn reg_plus_reg(reg1: Reg, reg2: Reg) -> MemArg {
         MemArg::RegReg(reg1, reg2)
@@ -431,8 +424,11 @@ impl ShowWithRRU for MemArg {
                 simm9.show_rru(mb_rru)
             ),
             // Eliminated by `mem_finalize()`.
-            &MemArg::SPOffset(..) | &MemArg::FPOffset(..) | &MemArg::NominalSPOffset(..) => {
-                panic!("Unexpected stack-offset mem-arg mode!")
+            &MemArg::SPOffset(..)
+            | &MemArg::FPOffset(..)
+            | &MemArg::NominalSPOffset(..)
+            | &MemArg::RegOffset(..) => {
+                panic!("Unexpected pseudo mem-arg mode (stack-offset or generic reg-offset)!")
             }
         }
     }
diff --git a/cranelift/codegen/src/isa/aarch64/inst/emit.rs b/cranelift/codegen/src/isa/aarch64/inst/emit.rs
index 99aade9b30..81b238adc4 100644
--- a/cranelift/codegen/src/isa/aarch64/inst/emit.rs
+++ b/cranelift/codegen/src/isa/aarch64/inst/emit.rs
@@ -5,6 +5,7 @@ use crate::ir::constant::ConstantData;
 use crate::ir::types::*;
 use crate::ir::TrapCode;
 use crate::isa::aarch64::inst::*;
+use crate::isa::aarch64::lower::ty_bits;
 
 use regalloc::{Reg, RegClass, Writable};
 
@@ -29,8 +30,12 @@ pub fn mem_finalize(
     state: &EmitState,
 ) -> (SmallVec<[Inst; 4]>, MemArg) {
     match mem {
-        &MemArg::SPOffset(off) | &MemArg::FPOffset(off) | &MemArg::NominalSPOffset(off) => {
+        &MemArg::RegOffset(_, off, ty)
+        | &MemArg::SPOffset(off, ty)
+        | &MemArg::FPOffset(off, ty)
+        | &MemArg::NominalSPOffset(off, ty) => {
             let basereg = match mem {
+                &MemArg::RegOffset(reg, _, _) => reg,
                 &MemArg::SPOffset(..) | &MemArg::NominalSPOffset(..) => stack_reg(),
                 &MemArg::FPOffset(..) => fp_reg(),
                 _ => unreachable!(),
@@ -52,6 +57,9 @@ pub fn mem_finalize(
             if let Some(simm9) = SImm9::maybe_from_i64(off) {
                 let mem = MemArg::Unscaled(basereg, simm9);
                 (smallvec![], mem)
+            } else if let Some(uimm12s) = UImm12Scaled::maybe_from_i64(off, ty) {
+                let mem = MemArg::UnsignedOffset(basereg, uimm12s);
+                (smallvec![], mem)
             } else {
                 let tmp = writable_spilltmp_reg();
                 let mut const_insts = Inst::load_constant(tmp, off as u64);
@@ -654,17 +662,17 @@ impl MachInstEmit for Inst {
                 // This is the base opcode (top 10 bits) for the "unscaled
                 // immediate" form (Unscaled). Other addressing modes will OR in
                 // other values for bits 24/25 (bits 1/2 of this constant).
-                let op = match self {
-                    &Inst::ULoad8 { .. } => 0b0011100001,
-                    &Inst::SLoad8 { .. } => 0b0011100010,
-                    &Inst::ULoad16 { .. } => 0b0111100001,
-                    &Inst::SLoad16 { .. } => 0b0111100010,
-                    &Inst::ULoad32 { .. } => 0b1011100001,
-                    &Inst::SLoad32 { .. } => 0b1011100010,
-                    &Inst::ULoad64 { .. } => 0b1111100001,
-                    &Inst::FpuLoad32 { .. } => 0b1011110001,
-                    &Inst::FpuLoad64 { .. } => 0b1111110001,
-                    &Inst::FpuLoad128 { .. } => 0b0011110011,
+                let (op, bits) = match self {
+                    &Inst::ULoad8 { .. } => (0b0011100001, 8),
+                    &Inst::SLoad8 { .. } => (0b0011100010, 8),
+                    &Inst::ULoad16 { .. } => (0b0111100001, 16),
+                    &Inst::SLoad16 { .. } => (0b0111100010, 16),
+                    &Inst::ULoad32 { .. } => (0b1011100001, 32),
+                    &Inst::SLoad32 { .. } => (0b1011100010, 32),
+                    &Inst::ULoad64 { .. } => (0b1111100001, 64),
+                    &Inst::FpuLoad32 { .. } => (0b1011110001, 32),
+                    &Inst::FpuLoad64 { .. } => (0b1111110001, 64),
+                    &Inst::FpuLoad128 { .. } => (0b0011110011, 128),
                     _ => unreachable!(),
                 };
 
@@ -678,6 +686,9 @@ impl MachInstEmit for Inst {
                         sink.put4(enc_ldst_simm9(op, simm9, 0b00, reg, rd));
                     }
                     &MemArg::UnsignedOffset(reg, uimm12scaled) => {
+                        if uimm12scaled.value() != 0 {
+                            assert_eq!(bits, ty_bits(uimm12scaled.scale_ty()));
+                        }
                         sink.put4(enc_ldst_uimm12(op, uimm12scaled, reg, rd));
                     }
                     &MemArg::RegReg(r1, r2) => {
@@ -686,19 +697,7 @@ impl MachInstEmit for Inst {
                         ));
                     }
                     &MemArg::RegScaled(r1, r2, ty) | &MemArg::RegScaledExtended(r1, r2, ty, _) => {
-                        match (ty, self) {
-                            (I8, &Inst::ULoad8 { .. }) => {}
-                            (I8, &Inst::SLoad8 { .. }) => {}
-                            (I16, &Inst::ULoad16 { .. }) => {}
-                            (I16, &Inst::SLoad16 { .. }) => {}
-                            (I32, &Inst::ULoad32 { .. }) => {}
-                            (I32, &Inst::SLoad32 { .. }) => {}
-                            (I64, &Inst::ULoad64 { .. }) => {}
-                            (F32, &Inst::FpuLoad32 { .. }) => {}
-                            (F64, &Inst::FpuLoad64 { .. }) => {}
-                            (I128, &Inst::FpuLoad128 { .. }) => {}
-                            _ => panic!("Mismatching reg-scaling type in MemArg"),
-                        }
+                        assert_eq!(bits, ty_bits(ty));
                         let extendop = match &mem {
                             &MemArg::RegScaled(..) => None,
                             &MemArg::RegScaledExtended(_, _, _, op) => Some(op),
@@ -746,6 +745,7 @@ impl MachInstEmit for Inst {
                     &MemArg::SPOffset(..)
                     | &MemArg::FPOffset(..)
                     | &MemArg::NominalSPOffset(..) => panic!("Should not see stack-offset here!"),
+                    &MemArg::RegOffset(..) => panic!("SHould not see generic reg-offset here!"),
                 }
             }
 
@@ -791,14 +791,14 @@ impl MachInstEmit for Inst {
                     inst.emit(sink, flags, state);
                 }
 
-                let op = match self {
-                    &Inst::Store8 { .. } => 0b0011100000,
-                    &Inst::Store16 { .. } => 0b0111100000,
-                    &Inst::Store32 { .. } => 0b1011100000,
-                    &Inst::Store64 { .. } => 0b1111100000,
-                    &Inst::FpuStore32 { .. } => 0b1011110000,
-                    &Inst::FpuStore64 { .. } => 0b1111110000,
-                    &Inst::FpuStore128 { .. } => 0b0011110010,
+                let (op, bits) = match self {
+                    &Inst::Store8 { .. } => (0b0011100000, 8),
+                    &Inst::Store16 { .. } => (0b0111100000, 16),
+                    &Inst::Store32 { .. } => (0b1011100000, 32),
+                    &Inst::Store64 { .. } => (0b1111100000, 64),
+                    &Inst::FpuStore32 { .. } => (0b1011110000, 32),
+                    &Inst::FpuStore64 { .. } => (0b1111110000, 64),
+                    &Inst::FpuStore128 { .. } => (0b0011110010, 128),
                     _ => unreachable!(),
                 };
 
@@ -812,6 +812,9 @@ impl MachInstEmit for Inst {
                         sink.put4(enc_ldst_simm9(op, simm9, 0b00, reg, rd));
                     }
                     &MemArg::UnsignedOffset(reg, uimm12scaled) => {
+                        if uimm12scaled.value() != 0 {
+                            assert_eq!(bits, ty_bits(uimm12scaled.scale_ty()));
+                        }
                         sink.put4(enc_ldst_uimm12(op, uimm12scaled, reg, rd));
                     }
                     &MemArg::RegReg(r1, r2) => {
@@ -843,6 +846,7 @@ impl MachInstEmit for Inst {
                     &MemArg::SPOffset(..)
                     | &MemArg::FPOffset(..)
                     | &MemArg::NominalSPOffset(..) => panic!("Should not see stack-offset here!"),
+                    &MemArg::RegOffset(..) => panic!("SHould not see generic reg-offset here!"),
                 }
             }
 
diff --git a/cranelift/codegen/src/isa/aarch64/inst/emit_tests.rs b/cranelift/codegen/src/isa/aarch64/inst/emit_tests.rs
index f9cd9237f9..6f302501d2 100644
--- a/cranelift/codegen/src/isa/aarch64/inst/emit_tests.rs
+++ b/cranelift/codegen/src/isa/aarch64/inst/emit_tests.rs
@@ -1311,7 +1311,7 @@ fn test_aarch64_binemit() {
     insns.push((
         Inst::ULoad64 {
             rd: writable_xreg(1),
-            mem: MemArg::FPOffset(32768),
+            mem: MemArg::FPOffset(32768, I8),
             srcloc: None,
         },
         "100090D2B063308B010240F9",
@@ -1320,7 +1320,7 @@ fn test_aarch64_binemit() {
     insns.push((
         Inst::ULoad64 {
             rd: writable_xreg(1),
-            mem: MemArg::FPOffset(-32768),
+            mem: MemArg::FPOffset(-32768, I8),
             srcloc: None,
         },
         "F0FF8F92B063308B010240F9",
@@ -1329,7 +1329,7 @@ fn test_aarch64_binemit() {
     insns.push((
         Inst::ULoad64 {
             rd: writable_xreg(1),
-            mem: MemArg::FPOffset(1048576), // 2^20
+            mem: MemArg::FPOffset(1048576, I8), // 2^20
             srcloc: None,
         },
         "1002A0D2B063308B010240F9",
@@ -1338,13 +1338,43 @@ fn test_aarch64_binemit() {
     insns.push((
         Inst::ULoad64 {
             rd: writable_xreg(1),
-            mem: MemArg::FPOffset(1048576 + 1), // 2^20 + 1
+            mem: MemArg::FPOffset(1048576 + 1, I8), // 2^20 + 1
             srcloc: None,
         },
         "300080D21002A0F2B063308B010240F9",
         "movz x16, #1 ; movk x16, #16, LSL #16 ; add x16, fp, x16, UXTX ; ldr x1, [x16]",
     ));
 
+    insns.push((
+        Inst::ULoad64 {
+            rd: writable_xreg(1),
+            mem: MemArg::RegOffset(xreg(7), 8, I64),
+            srcloc: None,
+        },
+        "E18040F8",
+        "ldur x1, [x7, #8]",
+    ));
+
+    insns.push((
+        Inst::ULoad64 {
+            rd: writable_xreg(1),
+            mem: MemArg::RegOffset(xreg(7), 1024, I64),
+            srcloc: None,
+        },
+        "E10042F9",
+        "ldr x1, [x7, #1024]",
+    ));
+
+    insns.push((
+        Inst::ULoad64 {
+            rd: writable_xreg(1),
+            mem: MemArg::RegOffset(xreg(7), 1048576, I64),
+            srcloc: None,
+        },
+        "1002A0D2F060308B010240F9",
+        "movz x16, #16, LSL #16 ; add x16, x7, x16, UXTX ; ldr x1, [x16]",
+    ));
+
     insns.push((
         Inst::Store8 {
             rd: xreg(1),
diff --git a/cranelift/codegen/src/isa/aarch64/inst/imms.rs b/cranelift/codegen/src/isa/aarch64/inst/imms.rs
index 7c473a83d2..20a8225e7a 100644
--- a/cranelift/codegen/src/isa/aarch64/inst/imms.rs
+++ b/cranelift/codegen/src/isa/aarch64/inst/imms.rs
@@ -259,7 +259,12 @@ impl UImm12Scaled {
 
     /// Value after scaling.
     pub fn value(&self) -> u32 {
-        self.value as u32 * self.scale_ty.bytes()
+        self.value as u32
+    }
+
+    /// The value type which is the scaling base.
+    pub fn scale_ty(&self) -> Type {
+        self.scale_ty
     }
 }
 
diff --git a/cranelift/codegen/src/isa/aarch64/inst/mod.rs b/cranelift/codegen/src/isa/aarch64/inst/mod.rs
index 1d21613e04..6fb559dbb9 100644
--- a/cranelift/codegen/src/isa/aarch64/inst/mod.rs
+++ b/cranelift/codegen/src/isa/aarch64/inst/mod.rs
@@ -1004,6 +1004,9 @@ fn memarg_regs(memarg: &MemArg, collector: &mut RegUsageCollector) {
         &MemArg::SPOffset(..) | &MemArg::NominalSPOffset(..) => {
             collector.add_use(stack_reg());
         }
+        &MemArg::RegOffset(r, ..) => {
+            collector.add_use(r);
+        }
     }
 }
 
@@ -1318,6 +1321,7 @@ fn aarch64_map_regs<RUM: RegUsageMapper>(inst: &mut Inst, mapper: &RUM) {
             &mut MemArg::FPOffset(..)
             | &mut MemArg::SPOffset(..)
             | &mut MemArg::NominalSPOffset(..) => {}
+            &mut MemArg::RegOffset(ref mut r, ..) => map_use(m, r),
         };
     }
 
diff --git a/cranelift/codegen/src/isa/aarch64/lower.rs b/cranelift/codegen/src/isa/aarch64/lower.rs
index 10db3b1f07..129b332295 100644
--- a/cranelift/codegen/src/isa/aarch64/lower.rs
+++ b/cranelift/codegen/src/isa/aarch64/lower.rs
@@ -539,12 +539,10 @@ pub(crate) fn lower_address<C: LowerCtx<I = Inst>>(
     // TODO: support base_reg + scale * index_reg. For this, we would need to pattern-match shl or
     // mul instructions (Load/StoreComplex don't include scale factors).
 
-    // Handle one reg and offset that fits in immediate, if possible.
+    // Handle one reg and offset.
     if addends.len() == 1 {
         let reg = input_to_reg(ctx, addends[0], NarrowValueMode::ZeroExtend64);
-        if let Some(memarg) = MemArg::reg_maybe_offset(reg, offset as i64, elem_ty) {
-            return memarg;
-        }
+        return MemArg::RegOffset(reg, offset as i64, elem_ty);
     }
 
     // Handle two regs and a zero offset, if possible.
diff --git a/cranelift/codegen/src/isa/aarch64/lower_inst.rs b/cranelift/codegen/src/isa/aarch64/lower_inst.rs
index f59c0a3f19..56fd628932 100644
--- a/cranelift/codegen/src/isa/aarch64/lower_inst.rs
+++ b/cranelift/codegen/src/isa/aarch64/lower_inst.rs
@@ -1335,7 +1335,7 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
                     assert!(inputs.len() == sig.params.len());
                     assert!(outputs.len() == sig.returns.len());
                     (
-                        AArch64ABICall::from_func(sig, &extname, dist, loc),
+                        AArch64ABICall::from_func(sig, &extname, dist, loc)?,
                         &inputs[..],
                     )
                 }
@@ -1344,7 +1344,7 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
                     let sig = ctx.call_sig(insn).unwrap();
                     assert!(inputs.len() - 1 == sig.params.len());
                     assert!(outputs.len() == sig.returns.len());
-                    (AArch64ABICall::from_ptr(sig, ptr, loc, op), &inputs[1..])
+                    (AArch64ABICall::from_ptr(sig, ptr, loc, op)?, &inputs[1..])
                 }
                 _ => unreachable!(),
             };
diff --git a/cranelift/codegen/src/isa/aarch64/mod.rs b/cranelift/codegen/src/isa/aarch64/mod.rs
index 3aa8c779aa..1b061045c0 100644
--- a/cranelift/codegen/src/isa/aarch64/mod.rs
+++ b/cranelift/codegen/src/isa/aarch64/mod.rs
@@ -46,7 +46,7 @@ impl AArch64Backend {
         func: &Function,
         flags: settings::Flags,
     ) -> CodegenResult<VCode<inst::Inst>> {
-        let abi = Box::new(abi::AArch64ABIBody::new(func, flags));
+        let abi = Box::new(abi::AArch64ABIBody::new(func, flags)?);
         compile::compile::<AArch64Backend>(func, self, abi)
     }
 }
diff --git a/cranelift/codegen/src/isa/x64/abi.rs b/cranelift/codegen/src/isa/x64/abi.rs
index 294bdc215a..4ba75e394c 100644
--- a/cranelift/codegen/src/isa/x64/abi.rs
+++ b/cranelift/codegen/src/isa/x64/abi.rs
@@ -184,11 +184,11 @@ impl X64ABIBody {
 impl ABIBody for X64ABIBody {
     type I = Inst;
 
-    fn needed_tmps(&self) -> usize {
-        0
+    fn temp_needed(&self) -> bool {
+        false
     }
 
-    fn init_with_tmps(&mut self, _: &[Writable<Reg>]) {}
+    fn init(&mut self, _: Option<Writable<Reg>>) {}
 
     fn flags(&self) -> &settings::Flags {
         &self.flags
@@ -239,8 +239,8 @@ impl ABIBody for X64ABIBody {
         }
     }
 
-    fn gen_retval_area_setup(&self) -> Vec<Inst> {
-        vec![]
+    fn gen_retval_area_setup(&self) -> Option<Inst> {
+        None
     }
 
     fn gen_copy_reg_to_retval(
diff --git a/cranelift/codegen/src/machinst/abi.rs b/cranelift/codegen/src/machinst/abi.rs
index 7075b10689..6e5170c3f6 100644
--- a/cranelift/codegen/src/machinst/abi.rs
+++ b/cranelift/codegen/src/machinst/abi.rs
@@ -12,11 +12,14 @@ pub trait ABIBody {
     /// The instruction type for the ISA associated with this ABI.
     type I: VCodeInst;
 
-    /// How many temps are needed?
-    fn needed_tmps(&self) -> usize;
+    /// Does the ABI-body code need a temp reg? One will be provided to `init()`
+    /// as the `maybe_tmp` arg if so.
+    fn temp_needed(&self) -> bool;
 
-    /// Initialize, providing the requersted temps.
-    fn init_with_tmps(&mut self, tmps: &[Writable<Reg>]);
+    /// Initialize. This is called after the ABIBody is constructed because it
+    /// may be provided with a temp vreg, which can only be allocated once the
+    /// lowering context exists.
+    fn init(&mut self, maybe_tmp: Option<Writable<Reg>>);
 
     /// Get the settings controlling this function's compilation.
     fn flags(&self) -> &settings::Flags;
@@ -40,12 +43,12 @@ pub trait ABIBody {
     /// register.
     fn gen_copy_arg_to_reg(&self, idx: usize, into_reg: Writable<Reg>) -> Self::I;
 
-    /// Generate any setup instructions needed to save values to the
+    /// Generate any setup instruction needed to save values to the
     /// return-value area. This is usually used when were are multiple return
     /// values or an otherwise large return value that must be passed on the
     /// stack; typically the ABI specifies an extra hidden argument that is a
     /// pointer to that memory.
-    fn gen_retval_area_setup(&self) -> Vec<Self::I>;
+    fn gen_retval_area_setup(&self) -> Option<Self::I>;
 
     /// Generate an instruction which copies a source register to a return value slot.
     fn gen_copy_reg_to_retval(
diff --git a/cranelift/codegen/src/machinst/lower.rs b/cranelift/codegen/src/machinst/lower.rs
index b83d400b74..a4bd09cfa0 100644
--- a/cranelift/codegen/src/machinst/lower.rs
+++ b/cranelift/codegen/src/machinst/lower.rs
@@ -383,7 +383,7 @@ impl<'func, I: VCodeInst> Lower<'func, I> {
                 let insn = self.vcode.abi().gen_copy_arg_to_reg(i, reg);
                 self.emit(insn);
             }
-            for insn in self.vcode.abi().gen_retval_area_setup().into_iter() {
+            if let Some(insn) = self.vcode.abi().gen_retval_area_setup() {
                 self.emit(insn);
             }
         }
@@ -652,11 +652,13 @@ impl<'func, I: VCodeInst> Lower<'func, I> {
     pub fn lower<B: LowerBackend<MInst = I>>(mut self, backend: &B) -> CodegenResult<VCode<I>> {
         debug!("about to lower function: {:?}", self.f);
 
-        // Initialize the ABI object with any temps it needs.
-        let tmps: SmallVec<[Writable<Reg>; 4]> = (0..self.vcode.abi().needed_tmps())
-            .map(|_| self.alloc_tmp(RegClass::I64, I64))
-            .collect();
-        self.vcode.abi().init_with_tmps(&tmps[..]);
+        // Initialize the ABI object, giving it a temp if requested.
+        let maybe_tmp = if self.vcode.abi().temp_needed() {
+            Some(self.alloc_tmp(RegClass::I64, I64))
+        } else {
+            None
+        };
+        self.vcode.abi().init(maybe_tmp);
 
         // Get the pinned reg here (we only parameterize this function on `B`,
         // not the whole `Lower` impl).
diff --git a/cranelift/filetests/filetests/vcode/aarch64/stack-limit.clif b/cranelift/filetests/filetests/vcode/aarch64/stack-limit.clif
index b7dc3bf342..ffb8b9b599 100644
--- a/cranelift/filetests/filetests/vcode/aarch64/stack-limit.clif
+++ b/cranelift/filetests/filetests/vcode/aarch64/stack-limit.clif
@@ -64,8 +64,8 @@ block0(v0: i64):
 
 ; check:      stp fp, lr, [sp, #-16]!
 ; nextln:     mov fp, sp
-; nextln:     ldr x16, [x0]
-; nextln:     ldr x16, [x16, #4]
+; nextln:     ldur x16, [x0]
+; nextln:     ldur x16, [x16, #4]
 ; nextln:     subs xzr, sp, x16
 ; nextln:     b.hs 8
 ; nextln:     udf
@@ -128,8 +128,8 @@ block0(v0: i64):
 
 ; check:      stp fp, lr, [sp, #-16]!
 ; nextln:     mov fp, sp
-; nextln:     ldr x16, [x0]
-; nextln:     ldr x16, [x16, #4]
+; nextln:     ldur x16, [x0]
+; nextln:     ldur x16, [x16, #4]
 ; nextln:     add x16, x16, #32
 ; nextln:     subs xzr, sp, x16
 ; nextln:     b.hs 8
@@ -151,8 +151,8 @@ block0(v0: i64):
 
 ; check:      stp fp, lr, [sp, #-16]!
 ; nextln:     mov fp, sp
-; nextln:     ldr x16, [x0]
-; nextln:     ldr x16, [x16, #4]
+; nextln:     ldur x16, [x0]
+; nextln:     ldur x16, [x16, #4]
 ; nextln:     subs xzr, sp, x16
 ; nextln:     b.hs 8
 ; nextln:     udf
@@ -179,9 +179,7 @@ block0(v0: i64):
 
 ; check:      stp fp, lr, [sp, #-16]!
 ; nextln:     mov fp, sp
-; nextln:     movz x16, #6784
-; nextln:     movk x16, #6, LSL #16
-; nextln:     ldr x16, [x0, x16]
+; nextln:     movz x16, #6784 ; movk x16, #6, LSL #16 ; add x16, x0, x16, UXTX ; ldr x16, [x16]
 ; nextln:     add x16, x16, #32
 ; nextln:     subs xzr, sp, x16
 ; nextln:     b.hs 8