[Hexagon] Add support for decoding PLT symbols #123425

quic-areg · 2025-01-17T23:54:17Z

Describes PLT entries for hexagon.

llvmbot · 2025-01-17T23:54:52Z

@llvm/pr-subscribers-lld-elf
@llvm/pr-subscribers-lld

@llvm/pr-subscribers-llvm-binary-utilities

Author: None (quic-areg)

Changes

Describes PLT entries for hexagon.

Full diff: https://github.com/llvm/llvm-project/pull/123425.diff

3 Files Affected:

(modified) llvm/lib/Object/ELFObjectFile.cpp (+4)
(modified) llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp (+37-1)
(added) llvm/test/tools/llvm-objdump/ELF/Hexagon/plt.test (+42)

diff --git a/llvm/lib/Object/ELFObjectFile.cpp b/llvm/lib/Object/ELFObjectFile.cpp
index 1ddfadaf1e2716..2d3d70db50c393 100644
--- a/llvm/lib/Object/ELFObjectFile.cpp
+++ b/llvm/lib/Object/ELFObjectFile.cpp
@@ -802,6 +802,10 @@ std::vector<ELFPltEntry> ELFObjectFileBase::getPltEntries() const {
     case Triple::aarch64_be:
       JumpSlotReloc = ELF::R_AARCH64_JUMP_SLOT;
       break;
+    case Triple::hexagon:
+      JumpSlotReloc = ELF::R_HEX_JMP_SLOT;
+      GlobDatReloc = ELF::R_HEX_GLOB_DAT;
+      break;
     default:
       return {};
   }
diff --git a/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp b/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
index a98f6048b051cc..d9fe05d4e1ac3f 100644
--- a/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
+++ b/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
@@ -734,8 +734,44 @@ class HexagonMCInstrAnalysis : public MCInstrAnalysis {
     Target = Value;
     return true;
   }
+
+  uint32_t getValueFromMask(uint32_t Instruction, uint32_t Mask) const {
+    uint32_t Result = 0;
+    size_t Off = 0;
+    for (uint32_t Bit = 0; Bit != sizeof(uint32_t) * CHAR_BIT; ++Bit) {
+      const uint8_t ValBit = (Instruction >> Bit) & 1;
+      const bool MaskBit = (Mask >> Bit) & 1;
+      if (MaskBit) {
+        Result |= (ValBit << Off);
+        ++Off;
+      }
+    }
+    return Result;
+  }
+
+  std::vector<std::pair<uint64_t, uint64_t>>
+  findPltEntries(uint64_t PltSectionVA, ArrayRef<uint8_t> PltContents,
+                 const Triple &TargetTriple) const override {
+    // Do a lightweight parsing of PLT entries.
+    std::vector<std::pair<uint64_t, uint64_t>> Result;
+    for (uint64_t Byte = 0x0, End = PltContents.size(); Byte < End; Byte += 4) {
+      // Recognize immext(##gotpltn)
+      uint32_t ImmExt = support::endian::read32le(PltContents.data() + Byte);
+      if ((ImmExt & 0x00004000) != 0x00004000)
+        continue;
+      uint32_t LoadGotPlt =
+          support::endian::read32le(PltContents.data() + Byte + 4);
+      if ((LoadGotPlt & 0x6a49c00c) != 0x6a49c00c)
+        continue;
+      uint32_t Address = (getValueFromMask(ImmExt, 0xfff3fff) << 6) +
+                         getValueFromMask(LoadGotPlt, 0x1f80) + PltSectionVA +
+                         Byte;
+      Result.push_back(std::make_pair(PltSectionVA + Byte, Address));
+    }
+    return Result;
+  }
 };
-}
+} // namespace
 
 static MCInstrAnalysis *createHexagonMCInstrAnalysis(const MCInstrInfo *Info) {
   return new HexagonMCInstrAnalysis(Info);
diff --git a/llvm/test/tools/llvm-objdump/ELF/Hexagon/plt.test b/llvm/test/tools/llvm-objdump/ELF/Hexagon/plt.test
new file mode 100644
index 00000000000000..5d2b8576b9b848
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/ELF/Hexagon/plt.test
@@ -0,0 +1,42 @@
+# RUN: yaml2obj %s | llvm-objdump -d - | FileCheck %s
+
+#      CHECK: 00000310 <printf@plt>:
+# CHECK-NEXT:      310:       36 40 00 00     00004036 {      immext(#0xd80)
+# CHECK-NEXT:      314:       0e de 49 6a     6a49de0e        r14 = add(pc,##0xdbc) }
+# CHECK-NEXT:      318:       1c c0 8e 91     918ec01c {      r28 = memw(r14+#0x0) }
+# CHECK-NEXT:      31c:       00 c0 9c 52     529cc000 {      jumpr r28 }
+
+--- !ELF
+FileHeader:
+  Class:           ELFCLASS32
+  Data:            ELFDATA2LSB
+  Type:            ET_DYN
+  Machine:         EM_HEXAGON
+Sections:
+  - Name:            .rela.plt
+    Type:            SHT_RELA
+    Flags:           [ SHF_ALLOC ]
+    Info:            .got.plt
+    Relocations:
+      - Offset:          0x10CC
+        Symbol:          printf
+        Type:            R_HEX_JMP_SLOT
+  - Name:            .plt
+    Type:            SHT_PROGBITS
+    Flags:           [ SHF_ALLOC, SHF_EXECINSTR ]
+    Address:         0x2B0
+    Content:         384000001CC0496A0E429CE24F409C913CC09C910E420E8C00C09C520000000000000000000000000000000000000000374000000ED0496A1CC08E9100C09C52374000000ECA496A1CC08E9100C09C52374000000EC4496A1CC08E9100C09C52364000000EDE496A1CC08E9100C09C52
+  - Name:            .text
+    Type:            SHT_PROGBITS
+    Flags:           [ SHF_ALLOC, SHF_EXECINSTR ]
+    Address:         0x320
+    Content:         0240096AC97FFF0F01C6007802C221F3FF7FFF0F80C7007800C002F300C0809100C0007500C05F5300C0096ADAFFFF5901C09DA082FFFEBF00C0423C0140000000D4496AD6FFFF5B00C000781EC01E96
+  - Name:            .got.plt
+    Type:            SHT_PROGBITS
+    Flags:           [ SHF_WRITE, SHF_ALLOC ]
+    Address:         0x10B0
+    Content:         00000000000000000000000000000000B0020000B0020000B0020000B0020000
+Symbols:
+  - Name:            printf
+    Binding:         STB_GLOBAL
+...

llvmbot · 2025-01-17T23:54:54Z

@llvm/pr-subscribers-backend-hexagon

Author: None (quic-areg)

Changes

Describes PLT entries for hexagon.

Full diff: https://github.com/llvm/llvm-project/pull/123425.diff

3 Files Affected:

(modified) llvm/lib/Object/ELFObjectFile.cpp (+4)
(modified) llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp (+37-1)
(added) llvm/test/tools/llvm-objdump/ELF/Hexagon/plt.test (+42)

diff --git a/llvm/lib/Object/ELFObjectFile.cpp b/llvm/lib/Object/ELFObjectFile.cpp
index 1ddfadaf1e2716..2d3d70db50c393 100644
--- a/llvm/lib/Object/ELFObjectFile.cpp
+++ b/llvm/lib/Object/ELFObjectFile.cpp
@@ -802,6 +802,10 @@ std::vector<ELFPltEntry> ELFObjectFileBase::getPltEntries() const {
     case Triple::aarch64_be:
       JumpSlotReloc = ELF::R_AARCH64_JUMP_SLOT;
       break;
+    case Triple::hexagon:
+      JumpSlotReloc = ELF::R_HEX_JMP_SLOT;
+      GlobDatReloc = ELF::R_HEX_GLOB_DAT;
+      break;
     default:
       return {};
   }
diff --git a/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp b/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
index a98f6048b051cc..d9fe05d4e1ac3f 100644
--- a/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
+++ b/llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp
@@ -734,8 +734,44 @@ class HexagonMCInstrAnalysis : public MCInstrAnalysis {
     Target = Value;
     return true;
   }
+
+  uint32_t getValueFromMask(uint32_t Instruction, uint32_t Mask) const {
+    uint32_t Result = 0;
+    size_t Off = 0;
+    for (uint32_t Bit = 0; Bit != sizeof(uint32_t) * CHAR_BIT; ++Bit) {
+      const uint8_t ValBit = (Instruction >> Bit) & 1;
+      const bool MaskBit = (Mask >> Bit) & 1;
+      if (MaskBit) {
+        Result |= (ValBit << Off);
+        ++Off;
+      }
+    }
+    return Result;
+  }
+
+  std::vector<std::pair<uint64_t, uint64_t>>
+  findPltEntries(uint64_t PltSectionVA, ArrayRef<uint8_t> PltContents,
+                 const Triple &TargetTriple) const override {
+    // Do a lightweight parsing of PLT entries.
+    std::vector<std::pair<uint64_t, uint64_t>> Result;
+    for (uint64_t Byte = 0x0, End = PltContents.size(); Byte < End; Byte += 4) {
+      // Recognize immext(##gotpltn)
+      uint32_t ImmExt = support::endian::read32le(PltContents.data() + Byte);
+      if ((ImmExt & 0x00004000) != 0x00004000)
+        continue;
+      uint32_t LoadGotPlt =
+          support::endian::read32le(PltContents.data() + Byte + 4);
+      if ((LoadGotPlt & 0x6a49c00c) != 0x6a49c00c)
+        continue;
+      uint32_t Address = (getValueFromMask(ImmExt, 0xfff3fff) << 6) +
+                         getValueFromMask(LoadGotPlt, 0x1f80) + PltSectionVA +
+                         Byte;
+      Result.push_back(std::make_pair(PltSectionVA + Byte, Address));
+    }
+    return Result;
+  }
 };
-}
+} // namespace
 
 static MCInstrAnalysis *createHexagonMCInstrAnalysis(const MCInstrInfo *Info) {
   return new HexagonMCInstrAnalysis(Info);
diff --git a/llvm/test/tools/llvm-objdump/ELF/Hexagon/plt.test b/llvm/test/tools/llvm-objdump/ELF/Hexagon/plt.test
new file mode 100644
index 00000000000000..5d2b8576b9b848
--- /dev/null
+++ b/llvm/test/tools/llvm-objdump/ELF/Hexagon/plt.test
@@ -0,0 +1,42 @@
+# RUN: yaml2obj %s | llvm-objdump -d - | FileCheck %s
+
+#      CHECK: 00000310 <printf@plt>:
+# CHECK-NEXT:      310:       36 40 00 00     00004036 {      immext(#0xd80)
+# CHECK-NEXT:      314:       0e de 49 6a     6a49de0e        r14 = add(pc,##0xdbc) }
+# CHECK-NEXT:      318:       1c c0 8e 91     918ec01c {      r28 = memw(r14+#0x0) }
+# CHECK-NEXT:      31c:       00 c0 9c 52     529cc000 {      jumpr r28 }
+
+--- !ELF
+FileHeader:
+  Class:           ELFCLASS32
+  Data:            ELFDATA2LSB
+  Type:            ET_DYN
+  Machine:         EM_HEXAGON
+Sections:
+  - Name:            .rela.plt
+    Type:            SHT_RELA
+    Flags:           [ SHF_ALLOC ]
+    Info:            .got.plt
+    Relocations:
+      - Offset:          0x10CC
+        Symbol:          printf
+        Type:            R_HEX_JMP_SLOT
+  - Name:            .plt
+    Type:            SHT_PROGBITS
+    Flags:           [ SHF_ALLOC, SHF_EXECINSTR ]
+    Address:         0x2B0
+    Content:         384000001CC0496A0E429CE24F409C913CC09C910E420E8C00C09C520000000000000000000000000000000000000000374000000ED0496A1CC08E9100C09C52374000000ECA496A1CC08E9100C09C52374000000EC4496A1CC08E9100C09C52364000000EDE496A1CC08E9100C09C52
+  - Name:            .text
+    Type:            SHT_PROGBITS
+    Flags:           [ SHF_ALLOC, SHF_EXECINSTR ]
+    Address:         0x320
+    Content:         0240096AC97FFF0F01C6007802C221F3FF7FFF0F80C7007800C002F300C0809100C0007500C05F5300C0096ADAFFFF5901C09DA082FFFEBF00C0423C0140000000D4496AD6FFFF5B00C000781EC01E96
+  - Name:            .got.plt
+    Type:            SHT_PROGBITS
+    Flags:           [ SHF_WRITE, SHF_ALLOC ]
+    Address:         0x10B0
+    Content:         00000000000000000000000000000000B0020000B0020000B0020000B0020000
+Symbols:
+  - Name:            printf
+    Binding:         STB_GLOBAL
+...

quic-areg · 2025-01-17T23:58:02Z

@quic-akaryaki @androm3da @MaskRay

androm3da

LGTM but someone else should approve.

MaskRay · 2025-01-18T06:33:17Z

llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp

+    size_t Off = 0;
+    for (uint32_t Bit = 0; Bit != sizeof(uint32_t) * CHAR_BIT; ++Bit) {
+      const uint8_t ValBit = (Instruction >> Bit) & 1;
+      const bool MaskBit = (Mask >> Bit) & 1;


inline the only-used-once variable

MaskRay · 2025-01-18T06:34:46Z

llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp

+  uint32_t getValueFromMask(uint32_t Instruction, uint32_t Mask) const {
+    uint32_t Result = 0;
+    size_t Off = 0;
+    for (uint32_t Bit = 0; Bit != sizeof(uint32_t) * CHAR_BIT; ++Bit) {


This function seems inefficient. You can compute Result based on popcount(Instruction & Mask)

tried to make it more efficient. the loop should run popcount(Mask) times.

tried to make it more efficient. the loop should run popcount(Mask) times.

What if you invoke llvm::popcount instead? I think that was the suggestion here.

uint32_t Result = llvm::popcount(Instruction & Mask);

This will give the number of set bits in Result, but was not sure where to go from there. I think I still need to extract and shift the bits accordingly. @MaskRay thoughts?

Is the value (1u << llvm::popcount(Instruction & Mask)) - 1 ?

If (Instruction & Mask) could be UINT32_MAX, 1u<<32 would trigger UB, but I suspect it would not happen here.

Is the value (1u << llvm::popcount(Instruction & Mask)) - 1 ?

It is not. The value will have popcount(Instruction & Mask) set bits but the number of significant bits is larger. This function tries to extract immediate value from an instruction encoding. It could be done by manually extracting and shifting the bits, but I thought this function was cleaner since the immediate bits are non-contiguous. The expression would look like this:

uint32_t Address = (((ImmExt & 0x0fff) | (ImmExt & 0xfff3 >> 16) >> 2) << 6) + ((LoadGotPlt & 0x1f80) >> 7) + PltSectionVA + Byte;

MaskRay · 2025-01-29T04:07:05Z

llvm/lib/Target/Hexagon/MCTargetDesc/HexagonMCTargetDesc.cpp

+      uint32_t Address = (getValueFromMask(ImmExt, 0xfff3fff) << 6) +
+                         getValueFromMask(LoadGotPlt, 0x1f80) + PltSectionVA +
+                         Byte;
+      Result.push_back(std::make_pair(PltSectionVA + Byte, Address));


emplace_back

Describes PLT entries for hexagon.

quic-areg · 2025-01-29T20:05:25Z

@androm3da can you merge this?

llvmbot added backend:Hexagon llvm:binary-utilities labels Jan 17, 2025

androm3da reviewed Jan 18, 2025

View reviewed changes

MaskRay reviewed Jan 18, 2025

View reviewed changes

quic-areg force-pushed the objdump-plt branch from 3352f35 to 79b7c01 Compare January 22, 2025 18:59

llvmbot added lld lld:ELF labels Jan 22, 2025

MaskRay reviewed Jan 29, 2025

View reviewed changes

MaskRay approved these changes Jan 29, 2025

View reviewed changes

quic-areg added 4 commits January 29, 2025 08:53

[Hexagon] Add support for decoding PLT symbols

a34ff65

Describes PLT entries for hexagon.

fix lld tests

63c169e

make getValueFromMask more efficient

e3c1804

emplace_back

998cdae

quic-areg force-pushed the objdump-plt branch from 79b7c01 to 998cdae Compare January 29, 2025 16:56

androm3da approved these changes Jan 29, 2025

View reviewed changes

androm3da merged commit 61ea63b into llvm:main Jan 29, 2025
8 checks passed

quic-areg mentioned this pull request Jan 30, 2025

Request Commit Access For quic-areg #125010

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Hexagon] Add support for decoding PLT symbols #123425

[Hexagon] Add support for decoding PLT symbols #123425

quic-areg commented Jan 17, 2025

llvmbot commented Jan 17, 2025 •

edited

Loading

llvmbot commented Jan 17, 2025

quic-areg commented Jan 17, 2025

androm3da left a comment

MaskRay Jan 18, 2025

MaskRay Jan 18, 2025

quic-areg Jan 22, 2025

androm3da Jan 22, 2025 •

edited

Loading

quic-areg Jan 24, 2025

MaskRay Jan 26, 2025

quic-areg Jan 27, 2025

MaskRay Jan 29, 2025

quic-areg Jan 29, 2025

quic-areg commented Jan 29, 2025

[Hexagon] Add support for decoding PLT symbols #123425

[Hexagon] Add support for decoding PLT symbols #123425

Conversation

quic-areg commented Jan 17, 2025

llvmbot commented Jan 17, 2025 • edited Loading

llvmbot commented Jan 17, 2025

quic-areg commented Jan 17, 2025

androm3da left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

androm3da Jan 22, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

quic-areg commented Jan 29, 2025

llvmbot commented Jan 17, 2025 •

edited

Loading

androm3da Jan 22, 2025 •

edited

Loading