diff --git a/cranelift/codegen/src/isa/aarch64/mod.rs b/cranelift/codegen/src/isa/aarch64/mod.rs
index 9e6918d011..0498875f2e 100644
--- a/cranelift/codegen/src/isa/aarch64/mod.rs
+++ b/cranelift/codegen/src/isa/aarch64/mod.rs
@@ -95,6 +95,7 @@ impl TargetIsa for AArch64Backend {
             dynamic_stackslot_offsets,
             bb_starts: emit_result.bb_offsets,
             bb_edges: emit_result.bb_edges,
+            alignment: emit_result.alignment,
         })
     }
 
@@ -179,6 +180,12 @@ impl TargetIsa for AArch64Backend {
     fn map_regalloc_reg_to_dwarf(&self, reg: Reg) -> Result<u16, systemv::RegisterMappingError> {
         inst::unwind::systemv::map_reg(reg).map(|reg| reg.0)
     }
+
+    fn function_alignment(&self) -> u32 {
+        // We use 32-byte alignment for performance reasons, but for correctness we would only need
+        // 4-byte alignment.
+        32
+    }
 }
 
 impl fmt::Display for AArch64Backend {
diff --git a/cranelift/codegen/src/isa/mod.rs b/cranelift/codegen/src/isa/mod.rs
index b9496a14a4..98364bb22f 100644
--- a/cranelift/codegen/src/isa/mod.rs
+++ b/cranelift/codegen/src/isa/mod.rs
@@ -277,6 +277,9 @@ pub trait TargetIsa: fmt::Display + Send + Sync {
     /// will be "labeled" or might have calls between them, typically the number
     /// of defined functions in the object file.
     fn text_section_builder(&self, num_labeled_funcs: u32) -> Box<dyn TextSectionBuilder>;
+
+    /// The function alignment required by this ISA.
+    fn function_alignment(&self) -> u32;
 }
 
 /// Methods implemented for free for target ISA!
diff --git a/cranelift/codegen/src/isa/s390x/mod.rs b/cranelift/codegen/src/isa/s390x/mod.rs
index 25886a8a0a..877a7374f0 100644
--- a/cranelift/codegen/src/isa/s390x/mod.rs
+++ b/cranelift/codegen/src/isa/s390x/mod.rs
@@ -93,6 +93,7 @@ impl TargetIsa for S390xBackend {
             dynamic_stackslot_offsets,
             bb_starts: emit_result.bb_offsets,
             bb_edges: emit_result.bb_edges,
+            alignment: emit_result.alignment,
         })
     }
 
@@ -161,6 +162,10 @@ impl TargetIsa for S390xBackend {
     fn text_section_builder(&self, num_funcs: u32) -> Box<dyn TextSectionBuilder> {
         Box::new(MachTextSectionBuilder::<inst::Inst>::new(num_funcs))
     }
+
+    fn function_alignment(&self) -> u32 {
+        4
+    }
 }
 
 impl fmt::Display for S390xBackend {
diff --git a/cranelift/codegen/src/isa/x64/mod.rs b/cranelift/codegen/src/isa/x64/mod.rs
index b57b6f7dac..ddb3523578 100644
--- a/cranelift/codegen/src/isa/x64/mod.rs
+++ b/cranelift/codegen/src/isa/x64/mod.rs
@@ -88,6 +88,7 @@ impl TargetIsa for X64Backend {
             dynamic_stackslot_offsets,
             bb_starts: emit_result.bb_offsets,
             bb_edges: emit_result.bb_edges,
+            alignment: emit_result.alignment,
         })
     }
 
@@ -158,6 +159,12 @@ impl TargetIsa for X64Backend {
     fn text_section_builder(&self, num_funcs: u32) -> Box<dyn TextSectionBuilder> {
         Box::new(MachTextSectionBuilder::<inst::Inst>::new(num_funcs))
     }
+
+    /// Align functions on x86 to 16 bytes, ensuring that rip-relative loads to SSE registers are
+    /// always from aligned memory.
+    fn function_alignment(&self) -> u32 {
+        16
+    }
 }
 
 impl fmt::Display for X64Backend {
diff --git a/cranelift/codegen/src/machinst/buffer.rs b/cranelift/codegen/src/machinst/buffer.rs
index 226feaf899..c4eb5dd2da 100644
--- a/cranelift/codegen/src/machinst/buffer.rs
+++ b/cranelift/codegen/src/machinst/buffer.rs
@@ -468,7 +468,11 @@ impl<I: VCodeInst> MachBuffer<I> {
     /// Align up to the given alignment.
     pub fn align_to(&mut self, align_to: CodeOffset) {
         trace!("MachBuffer: align to {}", align_to);
-        assert!(align_to.is_power_of_two());
+        assert!(
+            align_to.is_power_of_two(),
+            "{} is not a power of two",
+            align_to
+        );
         while self.cur_offset() & (align_to - 1) != 0 {
             self.put1(0);
         }
@@ -1620,7 +1624,7 @@ impl<I: VCodeInst> MachTextSectionBuilder<I> {
 }
 
 impl<I: VCodeInst> TextSectionBuilder for MachTextSectionBuilder<I> {
-    fn append(&mut self, named: bool, func: &[u8], align: Option<u32>) -> u64 {
+    fn append(&mut self, named: bool, func: &[u8], align: u32) -> u64 {
         // Conditionally emit an island if it's necessary to resolve jumps
         // between functions which are too far away.
         let size = func.len() as u32;
@@ -1628,7 +1632,7 @@ impl<I: VCodeInst> TextSectionBuilder for MachTextSectionBuilder<I> {
             self.buf.emit_island_maybe_forced(self.force_veneers, size);
         }
 
-        self.buf.align_to(align.unwrap_or(I::LabelUse::ALIGN));
+        self.buf.align_to(align);
         let pos = self.buf.cur_offset();
         if named {
             self.buf
diff --git a/cranelift/codegen/src/machinst/mod.rs b/cranelift/codegen/src/machinst/mod.rs
index 00a1741982..a2aa97064f 100644
--- a/cranelift/codegen/src/machinst/mod.rs
+++ b/cranelift/codegen/src/machinst/mod.rs
@@ -300,6 +300,9 @@ pub struct CompiledCodeBase<T: CompilePhase> {
     /// This info is generated only if the `machine_code_cfg_info`
     /// flag is set.
     pub bb_edges: Vec<(CodeOffset, CodeOffset)>,
+    /// Minimum alignment for the function, derived from the use of any
+    /// pc-relative loads.
+    pub alignment: u32,
 }
 
 impl CompiledCodeStencil {
@@ -314,6 +317,7 @@ impl CompiledCodeStencil {
             dynamic_stackslot_offsets: self.dynamic_stackslot_offsets,
             bb_starts: self.bb_starts,
             bb_edges: self.bb_edges,
+            alignment: self.alignment,
         }
     }
 }
@@ -355,7 +359,7 @@ pub trait TextSectionBuilder {
     ///
     /// This function returns the offset at which the data was placed in the
     /// text section.
-    fn append(&mut self, labeled: bool, data: &[u8], align: Option<u32>) -> u64;
+    fn append(&mut self, labeled: bool, data: &[u8], align: u32) -> u64;
 
     /// Attempts to resolve a relocation for this function.
     ///
diff --git a/cranelift/codegen/src/machinst/vcode.rs b/cranelift/codegen/src/machinst/vcode.rs
index 336284503b..bd458227b2 100644
--- a/cranelift/codegen/src/machinst/vcode.rs
+++ b/cranelift/codegen/src/machinst/vcode.rs
@@ -221,6 +221,9 @@ pub struct EmitResult<I: VCodeInst> {
 
     /// Stack frame size.
     pub frame_size: u32,
+
+    /// The alignment requirement for pc-relative loads.
+    pub alignment: u32,
 }
 
 /// A builder for a VCode function body.
@@ -1058,7 +1061,10 @@ impl<I: VCodeInst> VCode<I> {
         }
 
         // Emit the constants used by the function.
+        let mut alignment = 1;
         for (constant, data) in self.constants.iter() {
+            alignment = data.alignment().max(alignment);
+
             let label = buffer.get_label_for_constant(constant);
             buffer.defer_constant(label, data.alignment(), data.as_slice(), u32::max_value());
         }
@@ -1101,6 +1107,7 @@ impl<I: VCodeInst> VCode<I> {
             dynamic_stackslot_offsets: self.abi.dynamic_stackslot_offsets().clone(),
             value_labels_ranges,
             frame_size,
+            alignment,
         }
     }
 
diff --git a/cranelift/jit/src/backend.rs b/cranelift/jit/src/backend.rs
index a24d723a40..796a4e00c2 100644
--- a/cranelift/jit/src/backend.rs
+++ b/cranelift/jit/src/backend.rs
@@ -21,7 +21,6 @@ use std::ptr::NonNull;
 use std::sync::atomic::{AtomicPtr, Ordering};
 use target_lexicon::PointerWidth;
 
-const EXECUTABLE_DATA_ALIGNMENT: u64 = 0x10;
 const WRITABLE_DATA_ALIGNMENT: u64 = 0x8;
 const READONLY_DATA_ALIGNMENT: u64 = 0x1;
 
@@ -234,7 +233,12 @@ impl JITModule {
         let plt_entry = self
             .memory
             .code
-            .allocate(std::mem::size_of::<[u8; 16]>(), EXECUTABLE_DATA_ALIGNMENT)
+            .allocate(
+                std::mem::size_of::<[u8; 16]>(),
+                self.isa
+                    .symbol_alignment()
+                    .max(self.isa.function_alignment() as u64),
+            )
             .unwrap()
             .cast::<[u8; 16]>();
         unsafe {
@@ -680,16 +684,20 @@ impl Module for JITModule {
         }
 
         // work around borrow-checker to allow reuse of ctx below
-        let _ = ctx.compile(self.isa())?;
+        let res = ctx.compile(self.isa())?;
+        let alignment = res.alignment as u64;
         let compiled_code = ctx.compiled_code().unwrap();
 
         let code_size = compiled_code.code_info().total_size;
 
         let size = code_size as usize;
+        let align = alignment
+            .max(self.isa.function_alignment() as u64)
+            .max(self.isa.symbol_alignment());
         let ptr = self
             .memory
             .code
-            .allocate(size, EXECUTABLE_DATA_ALIGNMENT)
+            .allocate(size, align)
             .expect("TODO: handle OOM etc.");
 
         {
@@ -745,6 +753,7 @@ impl Module for JITModule {
         &mut self,
         id: FuncId,
         func: &ir::Function,
+        alignment: u64,
         bytes: &[u8],
         relocs: &[MachReloc],
     ) -> ModuleResult<ModuleCompiledFunction> {
@@ -764,10 +773,13 @@ impl Module for JITModule {
         }
 
         let size = bytes.len();
+        let align = alignment
+            .max(self.isa.function_alignment() as u64)
+            .max(self.isa.symbol_alignment());
         let ptr = self
             .memory
             .code
-            .allocate(size, EXECUTABLE_DATA_ALIGNMENT)
+            .allocate(size, align)
             .expect("TODO: handle OOM etc.");
 
         unsafe {
diff --git a/cranelift/module/src/module.rs b/cranelift/module/src/module.rs
index 187fed3184..d38f96ec99 100644
--- a/cranelift/module/src/module.rs
+++ b/cranelift/module/src/module.rs
@@ -640,6 +640,7 @@ pub trait Module {
         &mut self,
         func_id: FuncId,
         func: &ir::Function,
+        alignment: u64,
         bytes: &[u8],
         relocs: &[MachReloc],
     ) -> ModuleResult<ModuleCompiledFunction>;
@@ -736,10 +737,11 @@ impl<M: Module> Module for &mut M {
         &mut self,
         func_id: FuncId,
         func: &ir::Function,
+        alignment: u64,
         bytes: &[u8],
         relocs: &[MachReloc],
     ) -> ModuleResult<ModuleCompiledFunction> {
-        (**self).define_function_bytes(func_id, func, bytes, relocs)
+        (**self).define_function_bytes(func_id, func, alignment, bytes, relocs)
     }
 
     fn define_data(&mut self, data: DataId, data_ctx: &DataContext) -> ModuleResult<()> {
diff --git a/cranelift/object/src/backend.rs b/cranelift/object/src/backend.rs
index 845b279986..8f44b2cc28 100644
--- a/cranelift/object/src/backend.rs
+++ b/cranelift/object/src/backend.rs
@@ -314,11 +314,13 @@ impl Module for ObjectModule {
         info!("defining function {}: {}", func_id, ctx.func.display());
         let mut code: Vec<u8> = Vec::new();
 
-        ctx.compile_and_emit(self.isa(), &mut code)?;
+        let res = ctx.compile_and_emit(self.isa(), &mut code)?;
+        let alignment = res.alignment as u64;
 
         self.define_function_bytes(
             func_id,
             &ctx.func,
+            alignment,
             &code,
             ctx.compiled_code().unwrap().buffer.relocs(),
         )
@@ -328,6 +330,7 @@ impl Module for ObjectModule {
         &mut self,
         func_id: FuncId,
         func: &ir::Function,
+        alignment: u64,
         bytes: &[u8],
         relocs: &[MachReloc],
     ) -> ModuleResult<ModuleCompiledFunction> {
@@ -348,7 +351,10 @@ impl Module for ObjectModule {
         }
         *defined = true;
 
-        let align = std::cmp::max(self.function_alignment, self.isa.symbol_alignment());
+        let align = self
+            .function_alignment
+            .max(self.isa.symbol_alignment())
+            .max(alignment);
         let (section, offset) = if self.per_function_section {
             let symbol_name = self.object.symbol(symbol).name.clone();
             let (section, offset) =
diff --git a/crates/cranelift/src/compiler.rs b/crates/cranelift/src/compiler.rs
index 3d6829f1cc..198b0e6303 100644
--- a/crates/cranelift/src/compiler.rs
+++ b/crates/cranelift/src/compiler.rs
@@ -271,7 +271,8 @@ impl wasmtime_environ::Compiler for Compiler {
             &mut func_env,
         )?;
 
-        let (_, code_buf) = compile_maybe_cached(&mut context, isa, cache_ctx.as_mut())?;
+        let (code, code_buf) = compile_maybe_cached(&mut context, isa, cache_ctx.as_mut())?;
+        let alignment = code.alignment;
 
         let compiled_code = context.compiled_code().unwrap();
         let func_relocs = compiled_code
@@ -333,6 +334,7 @@ impl wasmtime_environ::Compiler for Compiler {
                 stack_maps,
                 start: 0,
                 length,
+                alignment,
             },
             address_map: address_transform,
         }))
diff --git a/crates/cranelift/src/obj.rs b/crates/cranelift/src/obj.rs
index 58bb4c6eca..28d7094980 100644
--- a/crates/cranelift/src/obj.rs
+++ b/crates/cranelift/src/obj.rs
@@ -95,7 +95,11 @@ impl<'a> ModuleTextBuilder<'a> {
         func: &'a CompiledFunction,
     ) -> (SymbolId, Range<u64>) {
         let body_len = func.body.len() as u64;
-        let off = self.text.append(labeled, &func.body, None);
+        let off = self.text.append(
+            labeled,
+            &func.body,
+            self.isa.function_alignment().max(func.info.alignment),
+        );
 
         let symbol_id = self.obj.add_symbol(Symbol {
             name,
@@ -198,7 +202,7 @@ impl<'a> ModuleTextBuilder<'a> {
         if padding == 0 {
             return;
         }
-        self.text.append(false, &vec![0; padding], Some(1));
+        self.text.append(false, &vec![0; padding], 1);
     }
 
     /// Indicates that the text section has been written completely and this
diff --git a/crates/environ/src/compilation.rs b/crates/environ/src/compilation.rs
index 3322ce2003..9fc56e75fa 100644
--- a/crates/environ/src/compilation.rs
+++ b/crates/environ/src/compilation.rs
@@ -28,6 +28,9 @@ pub struct FunctionInfo {
     pub start: u64,
     /// The size of the compiled function, in bytes.
     pub length: u32,
+
+    /// The alignment requirements of this function, in bytes.
+    pub alignment: u32,
 }
 
 /// Information about a compiled trampoline which the host can call to enter