[compiler-rt] [SystemZ][XRay] XRay runtime support for SystemZ (PR #113252)

Kai Nacke via llvm-commits llvm-commits at lists.llvm.org
Tue Oct 22 10:19:31 PDT 2024


https://github.com/redstar updated https://github.com/llvm/llvm-project/pull/113252

>From 2ce5bddcc1a1050f86927efd825b59615fb930c1 Mon Sep 17 00:00:00 2001
From: Kai Nacke <kai.peter.nacke at ibm.com>
Date: Mon, 21 Oct 2024 22:24:46 -0400
Subject: [PATCH 1/2] [SystemZ][XRay] XRay runtime support for SystemZ

Adds the runtime support routines for XRay on SystemZ.
Only function entry/exit is implemented.
---
 .../cmake/Modules/AllSupportedArchDefs.cmake  |   2 +-
 compiler-rt/lib/xray/CMakeLists.txt           |   7 ++
 compiler-rt/lib/xray/xray_interface.cpp       |   2 +
 compiler-rt/lib/xray/xray_s390x.cpp           |  90 ++++++++++++++
 compiler-rt/lib/xray/xray_trampoline_s390x.S  | 111 ++++++++++++++++++
 compiler-rt/lib/xray/xray_tsc.h               |  22 ++++
 6 files changed, 233 insertions(+), 1 deletion(-)
 create mode 100644 compiler-rt/lib/xray/xray_s390x.cpp
 create mode 100644 compiler-rt/lib/xray/xray_trampoline_s390x.S

diff --git a/compiler-rt/cmake/Modules/AllSupportedArchDefs.cmake b/compiler-rt/cmake/Modules/AllSupportedArchDefs.cmake
index 809e9277156912..bd51db766faab9 100644
--- a/compiler-rt/cmake/Modules/AllSupportedArchDefs.cmake
+++ b/compiler-rt/cmake/Modules/AllSupportedArchDefs.cmake
@@ -102,7 +102,7 @@ if(APPLE)
 set(ALL_XRAY_SUPPORTED_ARCH ${X86_64} ${ARM64})
 else()
 set(ALL_XRAY_SUPPORTED_ARCH ${X86_64} ${ARM32} ${ARM64} ${MIPS32} ${MIPS64}
-		powerpc64le ${HEXAGON} ${LOONGARCH64})
+		powerpc64le ${HEXAGON} ${LOONGARCH64} ${S390X})
 endif()
 set(ALL_SHADOWCALLSTACK_SUPPORTED_ARCH ${ARM64})
 
diff --git a/compiler-rt/lib/xray/CMakeLists.txt b/compiler-rt/lib/xray/CMakeLists.txt
index cf7b5062aae32d..8e1ef2db2873a8 100644
--- a/compiler-rt/lib/xray/CMakeLists.txt
+++ b/compiler-rt/lib/xray/CMakeLists.txt
@@ -83,6 +83,11 @@ set(hexagon_SOURCES
   xray_trampoline_hexagon.S
   )
 
+set(s390x_SOURCES
+  xray_s390x.cpp
+  xray_trampoline_s390x.S
+  )
+
 set(XRAY_SOURCE_ARCHS
   arm
   armhf
@@ -93,6 +98,7 @@ set(XRAY_SOURCE_ARCHS
   mips64
   mips64el
   powerpc64le
+  s390x
   x86_64
   )
 
@@ -141,6 +147,7 @@ set(XRAY_ALL_SOURCE_FILES
   ${mips64_SOURCES}
   ${mips64el_SOURCES}
   ${powerpc64le_SOURCES}
+  ${s390x_SOURCES}
   ${XRAY_IMPL_HEADERS}
   )
 list(REMOVE_DUPLICATES XRAY_ALL_SOURCE_FILES)
diff --git a/compiler-rt/lib/xray/xray_interface.cpp b/compiler-rt/lib/xray/xray_interface.cpp
index 5839043fcb93a8..fee7dccf695ca0 100644
--- a/compiler-rt/lib/xray/xray_interface.cpp
+++ b/compiler-rt/lib/xray/xray_interface.cpp
@@ -56,6 +56,8 @@ static const int16_t cSledLength = 64;
 static const int16_t cSledLength = 8;
 #elif defined(__hexagon__)
 static const int16_t cSledLength = 20;
+#elif defined(__s390x__)
+static const int16_t cSledLength = 18;
 #else
 #error "Unsupported CPU Architecture"
 #endif /* CPU architecture */
diff --git a/compiler-rt/lib/xray/xray_s390x.cpp b/compiler-rt/lib/xray/xray_s390x.cpp
new file mode 100644
index 00000000000000..135994e54a2d7a
--- /dev/null
+++ b/compiler-rt/lib/xray/xray_s390x.cpp
@@ -0,0 +1,90 @@
+//===-- xray_s390x.cpp ------------------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of XRay, a dynamic runtime instrumentation system.
+//
+// Implementation of s390x routines.
+//
+//===----------------------------------------------------------------------===//
+#include "sanitizer_common/sanitizer_common.h"
+#include "xray_defs.h"
+#include "xray_interface_internal.h"
+#include <cassert>
+#include <cstring>
+
+namespace __xray {
+
+bool patchFunctionEntry(const bool Enable, uint32_t FuncId,
+                        const XRaySledEntry &Sled,
+                        void (*Trampoline)()) XRAY_NEVER_INSTRUMENT {
+  const uint64_t Address = Sled.address();
+  if (Enable) {
+    // The resulting code is:
+    //   stmg    %r2, %r15, 16(%r15)
+    //   llilf   %2, FuncID
+    //   brasl   %r14, __xray_FunctionEntry at GOT
+    // The FuncId and the stmg instruction must be written.
+
+    // Write FuncId into llilf.
+    reinterpret_cast<uint32_t *>(Address)[2] = FuncId;
+    // Write last part of stmg.
+    reinterpret_cast<uint16_t *>(Address)[2] = 0x24;
+    // Write first part of stmg.
+    reinterpret_cast<uint32_t *>(Address)[0] = 0xeb2ff010;
+  } else {
+    // j +16 instructions.
+    *reinterpret_cast<uint32_t *>(Address) = 0xa7f4000b;
+  }
+  return true;
+}
+
+bool patchFunctionExit(const bool Enable, uint32_t FuncId,
+                       const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
+  const uint64_t Address = Sled.address();
+  if (Enable) {
+    // The resulting code is:
+    //   stmg    %r2, %r15, 24(%r15)
+    //   llilf   %2,FuncID
+    //   j       __xray_FunctionEntry at GOT
+    // The FuncId and the stmg instruction must be written.
+
+    // Write FuncId into llilf.
+    reinterpret_cast<uint32_t *>(Address)[2] = FuncId;
+    // Write last part of of stmg.
+    reinterpret_cast<uint16_t *>(Address)[2] = 0x24;
+    // Write first part of stmg.
+    reinterpret_cast<uint32_t *>(Address)[0] = 0xeb2ff010;
+  } else {
+    // br %14 instruction.
+    *reinterpret_cast<uint16_t *>(Address) = 0x07fe;
+  }
+  return true;
+}
+
+bool patchFunctionTailExit(const bool Enable, const uint32_t FuncId,
+                           const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
+  return patchFunctionExit(Enable, FuncId, Sled);
+}
+
+bool patchCustomEvent(const bool Enable, const uint32_t FuncId,
+                      const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
+  // FIXME: Implement.
+  return false;
+}
+
+bool patchTypedEvent(const bool Enable, const uint32_t FuncId,
+                     const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
+  // FIXME: Implement.
+  return false;
+}
+
+} // namespace __xray
+
+extern "C" void __xray_ArgLoggerEntry() XRAY_NEVER_INSTRUMENT {
+  // FIXME: this will have to be implemented in the trampoline assembly file.
+}
diff --git a/compiler-rt/lib/xray/xray_trampoline_s390x.S b/compiler-rt/lib/xray/xray_trampoline_s390x.S
new file mode 100644
index 00000000000000..49a290ceb11f0a
--- /dev/null
+++ b/compiler-rt/lib/xray/xray_trampoline_s390x.S
@@ -0,0 +1,111 @@
+//===-- xray_trampoline_s390x.s ---------------------------------*- ASM -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file is a part of XRay, a dynamic runtime instrumentation system.
+//
+// This implements the s390x-specific assembler for the trampolines.
+//
+//===----------------------------------------------------------------------===//
+
+    .text
+
+#if __VX__
+// Minimal stack frame size (160) plus space for 8 vector registers a 16 bytes.
+#define STACKSZ  288
+#else
+// Minimal stack frame size
+#define STACKSZ  160
+#endif
+
+//===----------------------------------------------------------------------===//
+
+    .globl  __xray_FunctionEntry
+    .p2align    4
+    .type   __xray_FunctionEntry, at function
+__xray_FunctionEntry:
+    # The registers r2-15 of the instrumented function are already saved in the
+    # stack frame. On entry, r2 contains the function id, and %r14 the address
+    # of the first instruction of the instrumented function.
+    # Register r14 will be stored in the slot reserved for compiler use.
+    stg     %r14, 8(%r15)
+    std     %f0, 128(%r15)
+    std     %f2, 136(%r15)
+    std     %f4, 144(%r15)
+    std     %f6, 152(%r15)
+    aghi    %r15, -STACKSZ
+#if __VX__
+    vstm    %v24, %v31, 160(%r15)
+#endif
+
+    lgrl    %r1, _ZN6__xray19XRayPatchedFunctionE at GOT
+    ltg     %r1, 0(%r1)
+    je      .Lrestore1
+
+    # Set r3 to XRayEntryType::ENTRY = 0.
+    # The FuncId is still stored in r2.
+    lghi    %r3, 0
+    basr    %r14, %r1
+
+.Lrestore1:
+#if __VX__
+    vlm     %v24, %v31, 160(%r15)
+#endif
+    ld      %f6, STACKSZ+152(%r15)
+    ld      %f4, STACKSZ+144(%r15)
+    ld      %f2, STACKSZ+136(%r15)
+    ld      %f0, STACKSZ+128(%r15)
+    lmg     %r1, %r15, STACKSZ+8(%r15)
+    br      %r1
+.Lfunc_end0:
+    .size    __xray_FunctionEntry, .Lfunc_end0-__xray_FunctionEntry
+
+//===----------------------------------------------------------------------===//
+
+    .globl  __xray_FunctionExit
+    .p2align    4
+    .type   __xray_FunctionExit, at function
+__xray_FunctionExit:
+    # The registers r2-15 of the instrumented function are already saved in the
+    # stack frame. On entry, the register r2 contains the function id.
+    # At the end, the function jumps to the address saved in the slot for r14,
+    # which contains the return address into the caller of the instrumented
+    # function.
+    std     %f0, 128(%r15)
+    std     %f2, 136(%r15)
+    std     %f4, 144(%r15)
+    std     %f6, 152(%r15)
+    aghi    %r15, -STACKSZ
+#if __VX__
+    vstm    %v24, %v31, 160(%r15)
+#endif
+
+    lgrl    %r1, _ZN6__xray19XRayPatchedFunctionE at GOT
+    ltg     %r1, 0(%r1)
+    je      .Lrestore2
+
+    # Set r3 to XRayEntryType::EXIT = 1.
+    # The FuncId is still stored in r2.
+    lghi    %r3, 1
+    basr    %r14, %r1
+
+.Lrestore2:
+#if __VX__
+    vlm     %v24, %v31, 160(%r15)
+#endif
+    ld      %f6, STACKSZ+152(%r15)
+    ld      %f4, STACKSZ+144(%r15)
+    ld      %f2, STACKSZ+136(%r15)
+    ld      %f0, STACKSZ+128(%r15)
+    lmg     %r2, %r15, STACKSZ+16(%r15)
+    br      %r14
+.Lfunc_end1:
+    .size    __xray_FunctionExit, .Lfunc_end1-__xray_FunctionExit
+
+//===----------------------------------------------------------------------===//
+
+    .section    ".note.GNU-stack","", at progbits
diff --git a/compiler-rt/lib/xray/xray_tsc.h b/compiler-rt/lib/xray/xray_tsc.h
index e1cafe1bf11d2d..5b2ba49b85f5d6 100644
--- a/compiler-rt/lib/xray/xray_tsc.h
+++ b/compiler-rt/lib/xray/xray_tsc.h
@@ -83,6 +83,28 @@ inline uint64_t getTSCFrequency() XRAY_NEVER_INSTRUMENT {
 
 } // namespace __xray
 
+#elif defined(__s390x__)
+#include "sanitizer_common/sanitizer_common.h"
+#include "sanitizer_common/sanitizer_internal_defs.h"
+#include "xray_defs.h"
+#include <cerrno>
+#include <cstdint>
+#include <time.h>
+
+namespace __xray {
+
+inline bool probeRequiredCPUFeatures() XRAY_NEVER_INSTRUMENT { return true; }
+
+ALWAYS_INLINE uint64_t readTSC(uint8_t &CPU) XRAY_NEVER_INSTRUMENT {
+  return __builtin_readcyclecounter();
+}
+
+inline uint64_t getTSCFrequency() XRAY_NEVER_INSTRUMENT {
+  return NanosecondsPerSecond;
+}
+
+} // namespace __xray
+
 #else
 #error Target architecture is not supported.
 #endif // CPU architecture

>From a6accc5dc26b76efec59d437c8b1fb2386488dc4 Mon Sep 17 00:00:00 2001
From: Kai Nacke <kai.peter.nacke at ibm.com>
Date: Tue, 22 Oct 2024 13:19:16 -0400
Subject: [PATCH 2/2] Changes based on reviewer comments

---
 compiler-rt/lib/xray/xray_s390x.cpp | 34 ++++++++++++++---------------
 1 file changed, 16 insertions(+), 18 deletions(-)

diff --git a/compiler-rt/lib/xray/xray_s390x.cpp b/compiler-rt/lib/xray/xray_s390x.cpp
index 135994e54a2d7a..ded7aa1f23d817 100644
--- a/compiler-rt/lib/xray/xray_s390x.cpp
+++ b/compiler-rt/lib/xray/xray_s390x.cpp
@@ -17,11 +17,9 @@
 #include <cassert>
 #include <cstring>
 
-namespace __xray {
-
-bool patchFunctionEntry(const bool Enable, uint32_t FuncId,
-                        const XRaySledEntry &Sled,
-                        void (*Trampoline)()) XRAY_NEVER_INSTRUMENT {
+bool __xray::patchFunctionEntry(const bool Enable, uint32_t FuncId,
+                                const XRaySledEntry &Sled,
+                                void (*Trampoline)()) XRAY_NEVER_INSTRUMENT {
   const uint64_t Address = Sled.address();
   if (Enable) {
     // The resulting code is:
@@ -43,8 +41,9 @@ bool patchFunctionEntry(const bool Enable, uint32_t FuncId,
   return true;
 }
 
-bool patchFunctionExit(const bool Enable, uint32_t FuncId,
-                       const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
+bool __xray::patchFunctionExit(const bool Enable, uint32_t FuncId,
+                               const XRaySledEntry &Sled)
+    XRAY_NEVER_INSTRUMENT {
   const uint64_t Address = Sled.address();
   if (Enable) {
     // The resulting code is:
@@ -66,25 +65,24 @@ bool patchFunctionExit(const bool Enable, uint32_t FuncId,
   return true;
 }
 
-bool patchFunctionTailExit(const bool Enable, const uint32_t FuncId,
-                           const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
+bool __xray::patchFunctionTailExit(const bool Enable, const uint32_t FuncId,
+                                   const XRaySledEntry &Sled)
+    XRAY_NEVER_INSTRUMENT {
   return patchFunctionExit(Enable, FuncId, Sled);
 }
 
-bool patchCustomEvent(const bool Enable, const uint32_t FuncId,
-                      const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
-  // FIXME: Implement.
+bool __xray::patchCustomEvent(const bool Enable, const uint32_t FuncId,
+                              const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
+  // TODO Implement.
   return false;
 }
 
-bool patchTypedEvent(const bool Enable, const uint32_t FuncId,
-                     const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
-  // FIXME: Implement.
+bool __xray::patchTypedEvent(const bool Enable, const uint32_t FuncId,
+                             const XRaySledEntry &Sled) XRAY_NEVER_INSTRUMENT {
+  // TODO Implement.
   return false;
 }
 
-} // namespace __xray
-
 extern "C" void __xray_ArgLoggerEntry() XRAY_NEVER_INSTRUMENT {
-  // FIXME: this will have to be implemented in the trampoline assembly file.
+  // TODO this will have to be implemented in the trampoline assembly file.
 }



More information about the llvm-commits mailing list