[BOLT,test] Link against a shared object to test PLT #125625

MaskRay · 2025-02-04T04:19:01Z

A few tests generate a statically-linked position-independent executable
with -nostdlib -Wl,--unresolved-symbols=ignore-all -pie (%clang) and
test PLT handling. (--unresolved-symbols=ignore-all suppresses undefined
symbol errors and serves as a convenience hack.)

This relies on an unguaranteed linker behavior: a statically-linked PIE
does not necessarily generate PLT entries.
While current lld generates a PLT entry, it will change to suppress the
PLT entry to simplify internal handling and improve consistency.

(The behavior has no consistency in GNU ld, some ports generated a
.dynsym entry while some don't. While most seem to generate a PLT entry
but some ports use a weird R_*_NONE relocation.)

Created using spr 1.3.5-bogner

llvmbot · 2025-02-04T04:20:52Z

@llvm/pr-subscribers-bolt

Author: Fangrui Song (MaskRay)

Changes

A few tests generate a statically-linked position-independent executable
with -nostdlib -Wl,--unresolved-symbols=ignore-all -pie (%clang) and
test PLT handling. (--unresolved-symbols=ignore-all suppresses undefined
symbol errors and serves as a convenience hack.)

This relies on an unguaranteed linker behavior: a statically-linked PIE
does not necessarily generate PLT entries.
While current lld generates a PLT entry, it will change to suppress the
PLT entry to simplify internal handling and improve consistency.

(The behavior has no consistency in GNU ld, some ports generated a
.dynsym entry while some don't. While most seem to generate a PLT entry
but some ports use a weird R_*_NONE relocation.)

Full diff: https://github.com/llvm/llvm-project/pull/125625.diff

7 Files Affected:

(modified) bolt/test/AArch64/exceptions-plt.cpp (+3-1)
(modified) bolt/test/AArch64/plt-call.test (+3-1)
(modified) bolt/test/X86/callcont-fallthru.s (+3-1)
(modified) bolt/test/X86/cfi-instrs-reordered.s (+3-1)
(modified) bolt/test/X86/plt-call.test (+3-1)
(modified) bolt/test/runtime/exceptions-plt.cpp (+3-1)
(modified) bolt/test/runtime/plt-lld.test (+4-3)

diff --git a/bolt/test/AArch64/exceptions-plt.cpp b/bolt/test/AArch64/exceptions-plt.cpp
index 576f0fc91a9d84..33c28406ca8d6b 100644
--- a/bolt/test/AArch64/exceptions-plt.cpp
+++ b/bolt/test/AArch64/exceptions-plt.cpp
@@ -2,7 +2,9 @@
 
 // REQUIRES: system-linux
 
-// RUN: %clangxx %cxxflags -O1 -Wl,-q,-znow %s -o %t.exe
+// RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
+// Link against a DSO to ensure PLT entries.
+// RUN: %clangxx %cxxflags -O1 -Wl,-q,-znow %s %t.so -o %t.exe
 // RUN: llvm-bolt %t.exe -o %t.bolt.exe --plt=all --print-only=.*main.* \
 // RUN:   --print-finalized 2>&1 | FileCheck %s
 
diff --git a/bolt/test/AArch64/plt-call.test b/bolt/test/AArch64/plt-call.test
index da307d4a6c01e6..1fa62c4a36aafa 100644
--- a/bolt/test/AArch64/plt-call.test
+++ b/bolt/test/AArch64/plt-call.test
@@ -1,6 +1,8 @@
 // Verify that PLTCall optimization works.
 
-RUN: %clang %cflags %p/../Inputs/plt-tailcall.c \
+RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
+// Link against a DSO to ensure PLT entries.
+RUN: %clang %cflags %p/../Inputs/plt-tailcall.c %t.so \
 RUN:    -o %t -Wl,-q
 RUN: llvm-bolt %t -o %t.bolt --plt=all --print-plt  --print-only=foo | FileCheck %s
 
diff --git a/bolt/test/X86/callcont-fallthru.s b/bolt/test/X86/callcont-fallthru.s
index 31a7910d7fa3f4..d76f869c971fd1 100644
--- a/bolt/test/X86/callcont-fallthru.s
+++ b/bolt/test/X86/callcont-fallthru.s
@@ -1,7 +1,9 @@
 ## Ensures that a call continuation fallthrough count is set when using
 ## pre-aggregated perf data.
 
-# RUN: %clangxx %cxxflags %s -o %t -Wl,-q -nostdlib
+# RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
+## Link against a DSO to ensure PLT entries.
+# RUN: %clangxx %cxxflags %s %t.so -o %t -Wl,-q -nostdlib
 # RUN: link_fdata %s %t %t.pa1 PREAGG
 # RUN: link_fdata %s %t %t.pa2 PREAGG2
 # RUN: link_fdata %s %t %t.pa3 PREAGG3
diff --git a/bolt/test/X86/cfi-instrs-reordered.s b/bolt/test/X86/cfi-instrs-reordered.s
index c325aaf1ad8b18..5173fa6c3c7d0a 100644
--- a/bolt/test/X86/cfi-instrs-reordered.s
+++ b/bolt/test/X86/cfi-instrs-reordered.s
@@ -3,7 +3,9 @@
 
 # RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown %s -o %t.o
 # RUN: llvm-strip --strip-unneeded %t.o
-# RUN: %clangxx %cflags %t.o -o %t.exe
+# RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
+## Link against a DSO to ensure PLT entries.
+# RUN: %clangxx %cflags %t.o %t.so -o %t.exe
 # RUN: llvm-bolt %t.exe -o %t --reorder-blocks=cache --print-after-lowering \
 # RUN:   --print-only=_Z10SolveCubicddddPiPd 2>&1 | FileCheck %s
 #
diff --git a/bolt/test/X86/plt-call.test b/bolt/test/X86/plt-call.test
index e6ae86c179d279..aeee3024ac1708 100644
--- a/bolt/test/X86/plt-call.test
+++ b/bolt/test/X86/plt-call.test
@@ -1,6 +1,8 @@
 // Verify that PLTCall optimization works.
 
-RUN: %clang %cflags %p/../Inputs/plt-tailcall.c \
+RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
+// Link against a DSO to ensure PLT entries.
+RUN: %clang %cflags %p/../Inputs/plt-tailcall.c %t.so \
 RUN:    -o %t -Wl,-q
 RUN: llvm-bolt %t -o %t.bolt --plt=all --print-plt  --print-only=foo | FileCheck %s
 
diff --git a/bolt/test/runtime/exceptions-plt.cpp b/bolt/test/runtime/exceptions-plt.cpp
index 8a75a3cb384b90..3d8e7a5133e2c1 100644
--- a/bolt/test/runtime/exceptions-plt.cpp
+++ b/bolt/test/runtime/exceptions-plt.cpp
@@ -2,7 +2,9 @@
 
 // REQUIRES: system-linux
 
-// RUN: %clangxx %cxxflags -O1 -Wl,-q,-znow %s -o %t.exe
+// RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
+// Link against a DSO to ensure PLT entries.
+// RUN: %clangxx %cxxflags -O1 -Wl,-q,-znow %s %t.so -o %t.exe
 // RUN: llvm-bolt %t.exe -o %t.bolt.exe --plt=all
 // RUN: %t.bolt.exe
 
diff --git a/bolt/test/runtime/plt-lld.test b/bolt/test/runtime/plt-lld.test
index b505a191f90abf..3432e18bf4daf4 100644
--- a/bolt/test/runtime/plt-lld.test
+++ b/bolt/test/runtime/plt-lld.test
@@ -1,14 +1,15 @@
 // This test checks that the pointers to PLT are properly updated.
-// The test is using lld linker.
+// The test uses lld and links against a DSO to ensure PLT entries.
+RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
 
 // Non-PIE:
-RUN: %clang %cflags -no-pie %p/../Inputs/plt.c -fuse-ld=lld \
+RUN: %clang %cflags -no-pie %p/../Inputs/plt.c %t.so -fuse-ld=lld \
 RUN:    -o %t.lld.exe -Wl,-q
 RUN: llvm-bolt %t.lld.exe -o %t.lld.bolt.exe --use-old-text=0 --lite=0
 RUN: %t.lld.bolt.exe | FileCheck %s
 
 // PIE:
-RUN: %clang %cflags -fPIC -pie %p/../Inputs/plt.c -fuse-ld=lld \
+RUN: %clang %cflags -fPIC -pie %p/../Inputs/plt.c %t.so -fuse-ld=lld \
 RUN:    -o %t.lld.pie.exe -Wl,-q
 RUN: llvm-bolt %t.lld.pie.exe -o %t.lld.bolt.pie.exe --use-old-text=0 --lite=0
 RUN: %t.lld.bolt.pie.exe | FileCheck %s

maksfb

Thanks!

MaskRay · 2025-02-08T05:44:14Z

/cherry-pick a907008

llvmbot · 2025-02-08T05:49:56Z

/pull-request #126351

A few tests generate a statically-linked position-independent executable with `-nostdlib -Wl,--unresolved-symbols=ignore-all -pie` (`%clang`) and test PLT handling. (--unresolved-symbols=ignore-all suppresses undefined symbol errors and serves as a convenience hack.) This relies on an unguaranteed linker behavior: a statically-linked PIE does not necessarily generate PLT entries. While current lld generates a PLT entry, it will change to suppress the PLT entry to simplify internal handling and improve consistency. (The behavior has no consistency in GNU ld, some ports generated a .dynsym entry while some don't. While most seem to generate a PLT entry but some ports use a weird `R_*_NONE` relocation.) (cherry picked from commit a907008)

A few tests generate a statically-linked position-independent executable with `-nostdlib -Wl,--unresolved-symbols=ignore-all -pie` (`%clang`) and test PLT handling. (--unresolved-symbols=ignore-all suppresses undefined symbol errors and serves as a convenience hack.) This relies on an unguaranteed linker behavior: a statically-linked PIE does not necessarily generate PLT entries. While current lld generates a PLT entry, it will change to suppress the PLT entry to simplify internal handling and improve consistency. (The behavior has no consistency in GNU ld, some ports generated a .dynsym entry while some don't. While most seem to generate a PLT entry but some ports use a weird `R_*_NONE` relocation.)

[𝘀𝗽𝗿] initial version

c364e78

Created using spr 1.3.5-bogner

MaskRay requested review from aaupov, maksfb, rafaelauler, ayermolo, dcci and yota9 as code owners February 4, 2025 04:19

llvmbot added the BOLT label Feb 4, 2025

MaskRay mentioned this pull request Feb 4, 2025

[BOLT][AArch64] Implement PLTCall optimization #93584

Merged

maksfb approved these changes Feb 4, 2025

View reviewed changes

MaskRay merged commit a907008 into main Feb 5, 2025
9 checks passed

MaskRay deleted the users/MaskRay/spr/bolttest-link-against-a-shared-object-to-test-plt branch February 5, 2025 17:32

MaskRay added this to the LLVM 20.X Release milestone Feb 8, 2025

MaskRay added the release:backport label Feb 8, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[BOLT,test] Link against a shared object to test PLT #125625

[BOLT,test] Link against a shared object to test PLT #125625

Uh oh!

MaskRay commented Feb 4, 2025

Uh oh!

llvmbot commented Feb 4, 2025

Uh oh!

maksfb left a comment

Uh oh!

Uh oh!

MaskRay commented Feb 8, 2025

Uh oh!

llvmbot commented Feb 8, 2025

Uh oh!

Uh oh!

[BOLT,test] Link against a shared object to test PLT #125625

[BOLT,test] Link against a shared object to test PLT #125625

Uh oh!

Conversation

MaskRay commented Feb 4, 2025

Uh oh!

llvmbot commented Feb 4, 2025

Uh oh!

maksfb left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

MaskRay commented Feb 8, 2025

Uh oh!

llvmbot commented Feb 8, 2025

Uh oh!

Uh oh!