[llvm] 2f54efd - [MachineOutliner] Don't outline ADRP pair to avoid incorrect ICF (#160232)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Nov 10 11:54:20 PST 2025
Author: Pranav Kant
Date: 2025-11-10T11:54:16-08:00
New Revision: 2f54efd1bf96f08e564dbd092f65c04b16e9d4dc
URL: https://github.com/llvm/llvm-project/commit/2f54efd1bf96f08e564dbd092f65c04b16e9d4dc
DIFF: https://github.com/llvm/llvm-project/commit/2f54efd1bf96f08e564dbd092f65c04b16e9d4dc.diff
LOG: [MachineOutliner] Don't outline ADRP pair to avoid incorrect ICF (#160232)
On AArch64, ADRP and its user instructions (LDR, ADD, etc.), that are
referencing a GOT symbol, when separated into different functions by
machine outliner exposes a correctness issue in the linker ICF. In such
cases, user instructions can end up pointing to a folded section (with
its canonical folded symbol), while ADRP instruction point to a GOT
entry corresponding to the original symbol. This leads to loading from
incorrect memory address after ICF. #129122 explains how this can happen
in detail.
This addresses #131660 which should fix two things:
1. Hide the correctness issue described above in the LLVM linker.
2. Allows optimizations that could relax GOT addressing to PC-relative
addressing.
Added:
llvm/test/CodeGen/AArch64/machine-outliner-adrp-got-split.mir
Modified:
llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
index 4b4073365483e..55a04cca4c394 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
@@ -9590,6 +9590,27 @@ AArch64InstrInfo::getOutliningCandidateInfo(
unsigned NumBytesToCreateFrame = 0;
+ // Avoid splitting ADRP ADD/LDR pair into outlined functions.
+ // These instructions are fused together by the scheduler.
+ // Any candidate where ADRP is the last instruction should be rejected
+ // as that will lead to splitting ADRP pair.
+ MachineInstr &LastMI = RepeatedSequenceLocs[0].back();
+ MachineInstr &FirstMI = RepeatedSequenceLocs[0].front();
+ if (LastMI.getOpcode() == AArch64::ADRP &&
+ (LastMI.getOperand(1).getTargetFlags() & AArch64II::MO_PAGE) != 0 &&
+ (LastMI.getOperand(1).getTargetFlags() & AArch64II::MO_GOT) != 0) {
+ return std::nullopt;
+ }
+
+ // Similarly any candidate where the first instruction is ADD/LDR with a
+ // page offset should be rejected to avoid ADRP splitting.
+ if ((FirstMI.getOpcode() == AArch64::ADDXri ||
+ FirstMI.getOpcode() == AArch64::LDRXui) &&
+ (FirstMI.getOperand(2).getTargetFlags() & AArch64II::MO_PAGEOFF) != 0 &&
+ (FirstMI.getOperand(2).getTargetFlags() & AArch64II::MO_GOT) != 0) {
+ return std::nullopt;
+ }
+
// We only allow outlining for functions having exactly matching return
// address signing attributes, i.e., all share the same value for the
// attribute "sign-return-address" and all share the same type of key they
diff --git a/llvm/test/CodeGen/AArch64/machine-outliner-adrp-got-split.mir b/llvm/test/CodeGen/AArch64/machine-outliner-adrp-got-split.mir
new file mode 100644
index 0000000000000..c397953b68f5e
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/machine-outliner-adrp-got-split.mir
@@ -0,0 +1,133 @@
+# RUN: llc -mtriple=aarch64--- -run-pass=machine-outliner -verify-machineinstrs %s -o - | FileCheck %s
+--- |
+
+ @x = common global i32 0, align 4
+
+ define i32 @adrp_add() #0 {
+ ret i32 0
+ }
+
+ define i32 @adrp_ldr() #0 {
+ ret i32 0
+ }
+
+ attributes #0 = { noinline noredzone }
+...
+---
+# Check that main function body doesn't split ADRP pair
+#
+# CHECK-LABEL: name: adrp_add
+# CHECK-DAG: bb.0:
+# CHECK: BL @OUTLINED_FUNCTION_[[F0:[0-9]+]]
+# CHECK-NEXT: BL @OUTLINED_FUNCTION_[[F2:[0-9]+]]
+# CHECK-NEXT: $lr = ORRXri $xzr, 1
+name: adrp_add
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $lr
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $x9 = ADRP target-flags(aarch64-page, aarch64-got) @x
+ $x12 = ADDXri $x9, target-flags(aarch64-pageoff, aarch64-got) @x, 0
+ $lr = ORRXri $xzr, 1
+ bb.1:
+ liveins: $lr
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $x9 = ADRP target-flags(aarch64-page, aarch64-got) @x
+ $x12 = ADDXri $x9, target-flags(aarch64-pageoff, aarch64-got) @x, 0
+ $lr = ORRXri $xzr, 1
+ bb.2:
+ liveins: $lr
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $x9 = ADRP target-flags(aarch64-page, aarch64-got) @x
+ $x12 = ADDXri $x9, target-flags(aarch64-pageoff, aarch64-got) @x, 0
+ $lr = ORRXri $xzr, 1
+ bb.3:
+ liveins: $lr
+ RET undef $lr
+...
+---
+# Check that main function body doesn't split ADRP pair
+#
+# CHECK-LABEL: name: adrp_ldr
+# CHECK-DAG: bb.0:
+# CHECK: BL @OUTLINED_FUNCTION_[[F0]]
+# CHECK-NEXT: BL @OUTLINED_FUNCTION_[[F1:[0-9]+]]
+# CHECK-NEXT: $lr = ORRXri $xzr, 1
+name: adrp_ldr
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $lr
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $x9 = ADRP target-flags(aarch64-page, aarch64-got) @x
+ $x12 = LDRXui $x9, target-flags(aarch64-pageoff, aarch64-got) @x
+ $lr = ORRXri $xzr, 1
+ bb.1:
+ liveins: $lr
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $x9 = ADRP target-flags(aarch64-page, aarch64-got) @x
+ $x12 = LDRXui $x9, target-flags(aarch64-pageoff, aarch64-got) @x
+ $lr = ORRXri $xzr, 1
+ bb.2:
+ liveins: $lr
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $w12 = ORRWri $wzr, 1
+ $x9 = ADRP target-flags(aarch64-page, aarch64-got) @x
+ $x12 = LDRXui $x9, target-flags(aarch64-pageoff, aarch64-got) @x
+ $lr = ORRXri $xzr, 1
+ bb.3:
+ liveins: $lr
+ RET undef $lr
+
+# Check that no outlined function split the ADRP pair apart
+#
+# CHECK: OUTLINED_FUNCTION_[[F0]]
+# CHECK-DAG: bb.0
+# CHECK: $w12 = ORRWri $wzr, 1
+# CHECK-NEXT: $w12 = ORRWri $wzr, 1
+# CHECK-NEXT: $w12 = ORRWri $wzr, 1
+# CHECK-NEXT: $w12 = ORRWri $wzr, 1
+# CHECK-NEXT: $w12 = ORRWri $wzr, 1
+# CHECK-NEXT: RET $lr
+
+# CHECK: OUTLINED_FUNCTION_[[F1]]
+# CHECK-DAG: bb.0
+# CHECK: $w12 = ORRWri $wzr, 1
+# CHECK-NEXT: $x9 = ADRP target-flags(aarch64-page, aarch64-got) @x
+# CHECK-NEXT: $x12 = LDRXui $x9, target-flags(aarch64-pageoff, aarch64-got) @x
+
+# CHECK: name: OUTLINED_FUNCTION_[[F2]]
+# CHECK-DAG: bb.0
+# CHECK: $w12 = ORRWri $wzr, 1
+# CHECK-NEXT: $x9 = ADRP target-flags(aarch64-page, aarch64-got) @x
+# CHECK-NEXT: $x12 = ADDXri $x9, target-flags(aarch64-pageoff, aarch64-got) @x, 0
More information about the llvm-commits
mailing list