Skip to content

A lightweight WebAssembly runtime that is fast, secure, and standards-compliant

Cranelift: reject too-large stackslot alignment. (#14650)

Chris Fallin4f5cd519f67b
diff --git a/cranelift/codegen/src/machinst/abi.rs b/cranelift/codegen/src/machinst/abi.rs
index ba3c0ec223..55ab974abc 100644
--- a/cranelift/codegen/src/machinst/abi.rs
+++ b/cranelift/codegen/src/machinst/abi.rs
@@ -1264,7 +1264,17 @@ impl<M: ABIMachineSpec> Callee<M> {
// We always at least machine-word-align slots, but also
// satisfy the user's requested alignment.
debug_assert!(data.align_shift < 32);
- let align = core::cmp::max(M::word_bytes(), 1u32 << data.align_shift);
+ // The frame is only as aligned as the ABI's stack alignment, so a
+ // slot can't be more aligned than that.
+ let requested_align = 1u32 << data.align_shift;
+ if requested_align > M::stack_align(call_conv) {
+ return Err(CodegenError::Unsupported(format!(
+ "stack slot alignment of {requested_align} bytes is larger than the \
+ stack alignment of {} bytes",
+ M::stack_align(call_conv)
+ )));
+ }
+ let align = core::cmp::max(M::word_bytes(), requested_align);
let mask = align - 1;
let start_offset = checked_round_up(unaligned_start_offset, mask)
.ok_or(CodegenError::ImplLimitExceeded)?;
diff --git a/cranelift/filetests/filetests/egraph/exponential-selects.clif b/cranelift/filetests/filetests/egraph/exponential-selects.clif
index d683dcafb8..f1b5bb0970 100644
--- a/cranelift/filetests/filetests/egraph/exponential-selects.clif
+++ b/cranelift/filetests/filetests/egraph/exponential-selects.clif
@@ -10,18 +10,18 @@ target s390x
function u1:0(i64x2, i128 uext, i16x8, f64, i64, i16, i32x4, f32, f32x4, i8x16, f64x2, i32, i8 sext) -> i8, i8 sext, i8 uext, i8x16, i8 sext, i32 sext, i8 sext, i8 sext, i16 sext, f32x4, i8 sext, i8, i64x2, f64 system_v {
ss0 = explicit_slot 81
ss1 = explicit_slot 80
- ss2 = explicit_slot 16, align = 16
+ ss2 = explicit_slot 16, align = 8
ss3 = explicit_slot 2, align = 2
ss4 = explicit_slot 4, align = 4
ss5 = explicit_slot 2, align = 2
- ss6 = explicit_slot 16, align = 16
- ss7 = explicit_slot 16, align = 16
+ ss6 = explicit_slot 16, align = 8
+ ss7 = explicit_slot 16, align = 8
ss8 = explicit_slot 8, align = 8
ss9 = explicit_slot 2, align = 2
- ss10 = explicit_slot 16, align = 16
+ ss10 = explicit_slot 16, align = 8
ss11 = explicit_slot 8, align = 8
ss12 = explicit_slot 4, align = 4
- ss13 = explicit_slot 16, align = 16
+ ss13 = explicit_slot 16, align = 8
ss14 = explicit_slot 2, align = 2
sig0 = (i8 sext, i64x2, f64, f64, f64x2, i64 uext, i16x8, f64x2, i16 sext) fast
sig1 = (f32) -> f32 system_v
diff --git a/cranelift/filetests/filetests/isa/riscv64/issue8847-1.clif b/cranelift/filetests/filetests/isa/riscv64/issue8847-1.clif
index f6a72e51ae..9607bb64bf 100644
--- a/cranelift/filetests/filetests/isa/riscv64/issue8847-1.clif
+++ b/cranelift/filetests/filetests/isa/riscv64/issue8847-1.clif
@@ -4,12 +4,12 @@ test compile
target riscv64
function u1:0() tail {
- ss0 = explicit_slot 50, align = 512
+ ss0 = explicit_slot 50, align = 16
ss1 = explicit_slot 47, align = 4
- ss2 = explicit_slot 34, align = 32
- ss3 = explicit_slot 103, align = 1024
- ss4 = explicit_slot 110, align = 512
- ss5 = explicit_slot 126, align = 512
+ ss2 = explicit_slot 34, align = 16
+ ss3 = explicit_slot 103, align = 16
+ ss4 = explicit_slot 110, align = 16
+ ss5 = explicit_slot 126, align = 16
sig0 = (i64 sext, i64 sext, i64 sext, i64 sext, i64 sext, i64 sext, i64 sext, i8 uext, i16 uext, i16, i64 sext, i64 sext, i128 uext, i8 sext, f32) tail
block0:
diff --git a/cranelift/filetests/filetests/isa/riscv64/issue8847.clif b/cranelift/filetests/filetests/isa/riscv64/issue8847.clif
index 91ab3bdea4..1ed7326975 100644
--- a/cranelift/filetests/filetests/isa/riscv64/issue8847.clif
+++ b/cranelift/filetests/filetests/isa/riscv64/issue8847.clif
@@ -10,12 +10,12 @@ set enable_heap_access_spectre_mitigation=false
target riscv64 has_zcd has_zbkb has_zbc has_zbs has_zicond has_zvl32b has_zvl64b has_zvl128b has_zvl1024b has_zvl2048b has_zvl4096b has_zvl8192b has_zvl16384b has_zvl32768b
function u1:0() tail {
- ss0 = explicit_slot 50, align = 512
+ ss0 = explicit_slot 50, align = 16
ss1 = explicit_slot 47, align = 4
- ss2 = explicit_slot 34, align = 32
- ss3 = explicit_slot 103, align = 1024
- ss4 = explicit_slot 110, align = 512
- ss5 = explicit_slot 126, align = 512
+ ss2 = explicit_slot 34, align = 16
+ ss3 = explicit_slot 103, align = 16
+ ss4 = explicit_slot 110, align = 16
+ ss5 = explicit_slot 126, align = 16
sig0 = (f32, f64, f64, f32, i8 uext, i128, i8 uext, i32, i16 sext, i64, i64 sext, i128, i8 sext, i8, i64, i64 sext) -> i16 sext, i64 sext, f64, i32 sext, f64, i8 sext, i64 sext, f32
sig1 = () system_v
sig2 = (i128, i16 sext, i128 sext, i32 sext, i16, i64 uext, f32, i8 sext, f32, i8, i64, i64, i64, i64 uext, f64) -> f64, i8 uext, f32, i128, i64 uext, i8, i16 sext, i64 sext tail
diff --git a/cranelift/filetests/filetests/isa/riscv64/issue8866.clif b/cranelift/filetests/filetests/isa/riscv64/issue8866.clif
index d462f9bf41..50e3e637b4 100644
--- a/cranelift/filetests/filetests/isa/riscv64/issue8866.clif
+++ b/cranelift/filetests/filetests/isa/riscv64/issue8866.clif
@@ -10,12 +10,12 @@ set enable_heap_access_spectre_mitigation=false
target riscv64 has_zcd has_zbkb has_zbc has_zbs has_zicond has_zvl32b has_zvl64b has_zvl128b has_zvl1024b has_zvl2048b has_zvl4096b has_zvl8192b has_zvl16384b has_zvl32768b
function u1:0() tail {
- ss0 = explicit_slot 50, align = 512
+ ss0 = explicit_slot 50, align = 16
ss1 = explicit_slot 47, align = 4
- ss2 = explicit_slot 34, align = 32
- ss3 = explicit_slot 103, align = 1024
- ss4 = explicit_slot 110, align = 512
- ss5 = explicit_slot 126, align = 512
+ ss2 = explicit_slot 34, align = 16
+ ss3 = explicit_slot 103, align = 16
+ ss4 = explicit_slot 110, align = 16
+ ss5 = explicit_slot 126, align = 16
sig0 = (f32, f64, f64, f32, i8 uext, i128, i8 uext, i32, i16 sext, i64, i64 sext, i128, i8 sext, i8, i64, i64 sext) -> i16 sext, i64 sext, f64, i32 sext, f64, i8 sext, i64 sext, f32 system_v
sig1 = () system_v
sig2 = (i128, i16 sext, i128 sext, i32 sext, i16, i64 uext, f32, i8 sext, f32, i8, i64, i64, i64, i64 uext, f64) -> f64, i8 uext, f32, i128, i64 uext, i8, i16 sext, i64 sext tail
diff --git a/cranelift/filetests/filetests/isa/s390x/stackslot-align-too-large.clif b/cranelift/filetests/filetests/isa/s390x/stackslot-align-too-large.clif
new file mode 100644
index 0000000000..5106640871
--- /dev/null
+++ b/cranelift/filetests/filetests/isa/s390x/stackslot-align-too-large.clif
@@ -0,0 +1,12 @@
+; Stack slots can't be more aligned than the ABI's stack alignment (8 bytes on
+; s390x), since the frame itself is only that aligned. Requesting more is an
+; unsupported-feature error rather than a silently misaligned slot.
+test compile expect-fail
+target s390x
+
+function %f() -> i64 {
+ ss0 = explicit_slot 16, align = 16
+block0:
+ v0 = stack_addr.i64 ss0
+ return v0
+}
diff --git a/cranelift/filetests/filetests/isa/s390x/user_stack_maps.clif b/cranelift/filetests/filetests/isa/s390x/user_stack_maps.clif
index a482622dbd..128cedd2fb 100644
--- a/cranelift/filetests/filetests/isa/s390x/user_stack_maps.clif
+++ b/cranelift/filetests/filetests/isa/s390x/user_stack_maps.clif
@@ -106,7 +106,7 @@ function %different_types(i8, i16, i32, i64, f32, f64) -> i8, i16, i32, i64, f32
ss1 = explicit_slot 2, align = 2
ss2 = explicit_slot 8, align = 4
ss3 = explicit_slot 16, align = 8
- ss4 = explicit_slot 48, align = 16
+ ss4 = explicit_slot 48, align = 8
sig0 = () system_v
fn0 = colocated u0:0 sig0
@@ -129,15 +129,15 @@ block0(v0: i8, v1: i16, v2: i32, v3: i64, v4: f32, v5: f64):
; VCode:
; stmg %r6, %r15, 48(%r15)
-; aghi %r15, -352
-; std %f8, 288(%r15)
-; std %f9, 296(%r15)
-; std %f10, 304(%r15)
-; std %f11, 312(%r15)
-; std %f12, 320(%r15)
-; std %f13, 328(%r15)
-; std %f14, 336(%r15)
-; std %f15, 344(%r15)
+; aghi %r15, -344
+; std %f8, 280(%r15)
+; std %f9, 288(%r15)
+; std %f10, 296(%r15)
+; std %f11, 304(%r15)
+; std %f12, 312(%r15)
+; std %f13, 320(%r15)
+; std %f14, 328(%r15)
+; std %f15, 336(%r15)
; block0:
; la %r12, 160(%r15)
; stc %r2, 0(%r12)
@@ -150,44 +150,44 @@ block0(v0: i8, v1: i16, v2: i32, v3: i64, v4: f32, v5: f64):
; lgr %r7, %r4
; la %r2, 180(%r15)
; ste %f0, 0(%r2)
-; vst %v0, 256(%r15)
+; vst %v0, 248(%r15)
; la %r2, 184(%r15)
; stg %r5, 0(%r2)
; lgr %r6, %r5
; la %r2, 192(%r15)
; std %f2, 0(%r2)
-; vst %v2, 272(%r15)
+; vst %v2, 264(%r15)
; brasl %r14, userextname0
; ; UserStackMap { by_type: [(types::I8, CompoundBitSet {0}), (types::I16, CompoundBitSet {8}), (types::I32, CompoundBitSet {16}), (types::F32, CompoundBitSet {20}), (types::I64, CompoundBitSet {24}), (types::F64, CompoundBitSet {32})], sp_to_sized_stack_slots: None }
; lgr %r2, %r9
; lgr %r3, %r8
; lgr %r4, %r7
; lgr %r5, %r6
-; vl %v0, 256(%r15)
-; vl %v2, 272(%r15)
-; ld %f8, 288(%r15)
-; ld %f9, 296(%r15)
-; ld %f10, 304(%r15)
-; ld %f11, 312(%r15)
-; ld %f12, 320(%r15)
-; ld %f13, 328(%r15)
-; ld %f14, 336(%r15)
-; ld %f15, 344(%r15)
-; lmg %r6, %r15, 400(%r15)
+; vl %v0, 248(%r15)
+; vl %v2, 264(%r15)
+; ld %f8, 280(%r15)
+; ld %f9, 288(%r15)
+; ld %f10, 296(%r15)
+; ld %f11, 304(%r15)
+; ld %f12, 312(%r15)
+; ld %f13, 320(%r15)
+; ld %f14, 328(%r15)
+; ld %f15, 336(%r15)
+; lmg %r6, %r15, 392(%r15)
; br %r14
;
; Disassembled:
; block0: ; offset 0x0
; stmg %r6, %r15, 0x30(%r15)
-; aghi %r15, -0x160
-; std %f8, 0x120(%r15)
-; std %f9, 0x128(%r15)
-; std %f10, 0x130(%r15)
-; std %f11, 0x138(%r15)
-; std %f12, 0x140(%r15)
-; std %f13, 0x148(%r15)
-; std %f14, 0x150(%r15)
-; std %f15, 0x158(%r15)
+; aghi %r15, -0x158
+; std %f8, 0x118(%r15)
+; std %f9, 0x120(%r15)
+; std %f10, 0x128(%r15)
+; std %f11, 0x130(%r15)
+; std %f12, 0x138(%r15)
+; std %f13, 0x140(%r15)
+; std %f14, 0x148(%r15)
+; std %f15, 0x150(%r15)
; block1: ; offset 0x2a
; la %r12, 0xa0(%r15)
; stc %r2, 0(%r12)
@@ -200,28 +200,28 @@ block0(v0: i8, v1: i16, v2: i32, v3: i64, v4: f32, v5: f64):
; lgr %r7, %r4
; la %r2, 0xb4(%r15)
; ste %f0, 0(%r2)
-; vst %v0, 0x100(%r15)
+; vst %v0, 0xf8(%r15)
; la %r2, 0xb8(%r15)
; stg %r5, 0(%r2)
; lgr %r6, %r5
; la %r2, 0xc0(%r15)
; std %f2, 0(%r2)
-; vst %v2, 0x110(%r15)
+; vst %v2, 0x108(%r15)
; brasl %r14, 0x78 ; reloc_external PLTRel32Dbl u0:0 2
; lgr %r2, %r9
; lgr %r3, %r8
; lgr %r4, %r7
; lgr %r5, %r6
-; vl %v0, 0x100(%r15)
-; vl %v2, 0x110(%r15)
-; ld %f8, 0x120(%r15)
-; ld %f9, 0x128(%r15)
-; ld %f10, 0x130(%r15)
-; ld %f11, 0x138(%r15)
-; ld %f12, 0x140(%r15)
-; ld %f13, 0x148(%r15)
-; ld %f14, 0x150(%r15)
-; ld %f15, 0x158(%r15)
-; lmg %r6, %r15, 0x190(%r15)
+; vl %v0, 0xf8(%r15)
+; vl %v2, 0x108(%r15)
+; ld %f8, 0x118(%r15)
+; ld %f9, 0x120(%r15)
+; ld %f10, 0x128(%r15)
+; ld %f11, 0x130(%r15)
+; ld %f12, 0x138(%r15)
+; ld %f13, 0x140(%r15)
+; ld %f14, 0x148(%r15)
+; ld %f15, 0x150(%r15)
+; lmg %r6, %r15, 0x188(%r15)
; br %r14
diff --git a/cranelift/filetests/filetests/isa/x64/stackslot-align-too-large.clif b/cranelift/filetests/filetests/isa/x64/stackslot-align-too-large.clif
new file mode 100644
index 0000000000..1d70993657
--- /dev/null
+++ b/cranelift/filetests/filetests/isa/x64/stackslot-align-too-large.clif
@@ -0,0 +1,13 @@
+; Stack slots can't be more aligned than the ABI's stack alignment (16 bytes
+; on x86-64), since the frame itself is only that aligned. Requesting more is
+; an unsupported-feature error rather than a silently misaligned slot.
+test compile expect-fail
+target x86_64
+
+function %f() -> i64 {
+ ss0 = explicit_slot 8
+ ss1 = explicit_slot 64, align = 32
+block0:
+ v0 = stack_addr.i64 ss1
+ return v0
+}
diff --git a/cranelift/filetests/filetests/runtests/simd-fcmp-unaligned.clif b/cranelift/filetests/filetests/runtests/simd-fcmp-unaligned.clif
index 8784c72dc3..c2340067a6 100644
--- a/cranelift/filetests/filetests/runtests/simd-fcmp-unaligned.clif
+++ b/cranelift/filetests/filetests/runtests/simd-fcmp-unaligned.clif
@@ -12,7 +12,7 @@ target pulley64
target pulley64be
function %fcmp_eq_unaligned_f32x4(f32x4) -> i32x4 {
- ss0 = explicit_slot 48, align = 16
+ ss0 = explicit_slot 48, align = 8
block0(v0: f32x4):
v1 = stack_addr.i64 ss0
store.f32x4 v0, v1+1
@@ -23,7 +23,7 @@ block0(v0: f32x4):
; run: %fcmp_eq_unaligned_f32x4([0x1.0 0x2.0 0x3.0 0x4.0]) == [-1 -1 -1 -1]
function %fcmp_lt_unaligned_f64x2(f64x2) -> i64x2 {
- ss0 = explicit_slot 48, align = 16
+ ss0 = explicit_slot 48, align = 8
block0(v0: f64x2):
v1 = stack_addr.i64 ss0
store.f64x2 v0, v1+1
diff --git a/crates/cranelift/src/compiler.rs b/crates/cranelift/src/compiler.rs
index 29fee69dc7..fc33677345 100644
--- a/crates/cranelift/src/compiler.rs
+++ b/crates/cranelift/src/compiler.rs
@@ -1180,10 +1180,11 @@ impl Compiler {
let values_vec_byte_size = u32::try_from(value_size * values_vec_len).unwrap();
let values_vec_len = u32::try_from(values_vec_len).unwrap();
+ // `ValRaw` is 8-byte aligned: its `v128` field is a byte array.
let slot = builder.func.create_sized_stack_slot(ir::StackSlotData::new(
ir::StackSlotKind::ExplicitSlot,
values_vec_byte_size,
- 4,
+ 3,
));
let values_vec_ptr = builder.ins().stack_addr(pointer_type, slot, 0);
diff --git a/crates/cranelift/src/func_environ/stack_switching/instructions.rs b/crates/cranelift/src/func_environ/stack_switching/instructions.rs
index 27ec7185e0..4a18503237 100644
--- a/crates/cranelift/src/func_environ/stack_switching/instructions.rs
+++ b/crates/cranelift/src/func_environ/stack_switching/instructions.rs
@@ -96,19 +96,23 @@ pub(crate) mod stack_switching_helpers {
/// Provides information about the layout of a type when it is used as an
/// element in a host array. This is used for `VMHostArrayRef`.
pub(crate) trait VMHostArrayEntry {
- /// Returns `(align, size)` in bytes.
+ /// Returns `(align_shift, size)`: the log2 of the alignment, as used by
+ /// `StackSlotData`, and the size in bytes.
fn vmhostarray_entry_layout<P: wasmtime_environ::PtrSize>(p: &P) -> (u8, u32);
}
impl VMHostArrayEntry for u128 {
fn vmhostarray_entry_layout<P: wasmtime_environ::PtrSize>(_p: &P) -> (u8, u32) {
- (16, 16)
+ (4, 16)
}
}
impl<T> VMHostArrayEntry for *mut T {
fn vmhostarray_entry_layout<P: wasmtime_environ::PtrSize>(p: &P) -> (u8, u32) {
- (p.size(), p.size().into())
+ (
+ u8::try_from(p.size().trailing_zeros()).unwrap(),
+ p.size().into(),
+ )
}
}
@@ -2470,7 +2474,7 @@ pub(crate) fn translate_switch<'a>(
let slot_size = ir::StackSlotData::new(
ir::StackSlotKind::ExplicitSlot,
u32::from(cctx_size),
- u8::try_from(env.pointer_type().bytes()).unwrap(),
+ u8::try_from(env.pointer_type().bytes().trailing_zeros()).unwrap(),
);
let slot = builder.create_sized_stack_slot(slot_size);
let tmp_control_context = builder.ins().stack_addr(env.pointer_type(), slot, 0);
diff --git a/tests/disas/stack-switching/resume-suspend-data-passing.wat b/tests/disas/stack-switching/resume-suspend-data-passing.wat
index b7d82b9cef..6a35fbafb4 100644
--- a/tests/disas/stack-switching/resume-suspend-data-passing.wat
+++ b/tests/disas/stack-switching/resume-suspend-data-passing.wat
@@ -38,7 +38,7 @@
)
;; function u0:0(i64 vmctx, i64) tail {
-;; ss0 = explicit_slot 16, align = 65536
+;; ss0 = explicit_slot 16, align = 16
;; region0 = 123 ""
;; region1 = 160 ""
;; region2 = 106 ""
@@ -179,7 +179,7 @@
;; }
;;
;; function u0:1(i64 vmctx, i64) tail {
-;; ss0 = explicit_slot 8, align = 256
+;; ss0 = explicit_slot 8, align = 8
;; region0 = 123 ""
;; region1 = 160 ""
;; region2 = 28 ""
diff --git a/tests/disas/stack-switching/resume-suspend.wat b/tests/disas/stack-switching/resume-suspend.wat
index 7713591d7f..de6f00a38f 100644
--- a/tests/disas/stack-switching/resume-suspend.wat
+++ b/tests/disas/stack-switching/resume-suspend.wat
@@ -23,7 +23,7 @@
)
;; function u0:0(i64 vmctx, i64) tail {
-;; ss0 = explicit_slot 16, align = 65536
+;; ss0 = explicit_slot 16, align = 16
;; region0 = 123 ""
;; region1 = 160 ""
;; region2 = 106 ""
@@ -145,7 +145,7 @@
;; }
;;
;; function u0:1(i64 vmctx, i64) tail {
-;; ss0 = explicit_slot 8, align = 256
+;; ss0 = explicit_slot 8, align = 8
;; region0 = 123 ""
;; region1 = 160 ""
;; region2 = 28 ""
diff --git a/tests/disas/stack-switching/symmetric-switch.wat b/tests/disas/stack-switching/symmetric-switch.wat
index 7085d0649a..b57ef0b94a 100644
--- a/tests/disas/stack-switching/symmetric-switch.wat
+++ b/tests/disas/stack-switching/symmetric-switch.wat
@@ -26,8 +26,8 @@
)
;; function u0:0(i64 vmctx, i64) tail {
-;; ss0 = explicit_slot 16, align = 65536
-;; ss1 = explicit_slot 24, align = 256
+;; ss0 = explicit_slot 16, align = 16
+;; ss1 = explicit_slot 24, align = 8
;; region0 = 123 ""
;; region1 = 160 ""
;; region2 = 28 ""
@@ -307,7 +307,7 @@
;; }
;;
;; function u0:2(i64 vmctx, i64) tail {
-;; ss0 = explicit_slot 8, align = 256
+;; ss0 = explicit_slot 8, align = 8
;; region0 = 123 ""
;; region1 = 160 ""
;; region2 = 28 ""