Merge pull request #169 from Decompollaborate/develop

1.28.0

CHANGELOG.md

@@ -7,6 +7,43 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.28.0] - 2024-08-09
+
+### Added
+
+- New `SectionText.gpRelHack` setting.
+  - Turning it on changes all instructions that use a `%gp_rel` reloc into macro
+    instructions that do not specify the relocation explicitly nor reference the
+    `$gp` register.
+  - This may even change some instruction mnemonics, like replacing `addiu` into
+    `la`.
+  - This is required by old assemblers that do not support explicit `%gp_rel`
+    relocations, but instead they infer the relocation to be used by checking
+    if the symbol was defined in the assembly file and its size fits on the
+    passed `-G` parameter.
+  - WARNING: It is the user's responsability to provide those symbol definitions
+    to the assembler, otherwise those instructions will be expanded into
+    multiple instructions and produce a shifted build.
+- elfObjDisasm's readelf:
+  - Add `MIPS_SCOMMON` and `MIPS_SUNDEFINED` support in symtab.
+  - Use the section name in the ndx column instead of a plain number for
+    `OBJECT`s and `FUNC`s.
+
+### Changed
+
+- Try to detect function pointers used on tail call optimizations and try to not
+  confuse them with detected jumptables.
+- rabbitizer 1.12.0 or above is required.
+
+### Fixed
+
+- Fix rodata addresses referenced _only_ by other rodata symbols on the same
+  file not being properly symbolized.
+- elfObjDisasm's readelf:
+  - Fix name column not displaying the section's name.
+  - Fix relocation sections not displaying anything on the name columns for
+    relocations relative to a section instead of a symbol.
+
 ## [1.27.0] - 2024-07-10
 
 ### Added
@@ -1556,7 +1593,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Version 1.0.0
 
 [unreleased]: https://github.com/Decompollaborate/spimdisasm/compare/master...develop
-[1.27.0]: https://github.com/Decompollaborate/spimdisasm/compare/1.27.0...1.27.0
+[1.28.0]: https://github.com/Decompollaborate/spimdisasm/compare/1.27.0...1.28.0
+[1.27.0]: https://github.com/Decompollaborate/spimdisasm/compare/1.26.1...1.27.0
 [1.26.1]: https://github.com/Decompollaborate/spimdisasm/compare/1.26.0...1.26.1
 [1.26.0]: https://github.com/Decompollaborate/spimdisasm/compare/1.25.1...1.26.0
 [1.25.1]: https://github.com/Decompollaborate/spimdisasm/compare/1.25.0...1.25.1

README.md

@@ -69,7 +69,7 @@ If you use a `requirements.txt` file in your repository, then you can add
 this library with the following line:
 
 ```txt
-spimdisasm>=1.27.0,<2.0.0
+spimdisasm>=1.28.0,<2.0.0
 ```
 
 ### Development version

pyproject.toml

@@ -4,7 +4,7 @@
 [project]
 name = "spimdisasm"
 # Version should be synced with spimdisasm/__init__.py
-version = "1.27.0"
+version = "1.28.0"
 description = "MIPS disassembler"
 readme = "README.md"
 license = {file = "LICENSE"}

requirements.txt

@@ -1 +1 @@
-rabbitizer>=1.10.0,<2.0.0
+rabbitizer>=1.12.0,<2.0.0

spimdisasm/__init__.py

@@ -5,8 +5,8 @@
 
 from __future__ import annotations
 
-__version_info__: tuple[int, int, int] = (1, 27, 0)
-__version__ = ".".join(map(str, __version_info__))# + ".dev0"
+__version_info__: tuple[int, int, int] = (1, 28, 0)
+__version__ = ".".join(map(str, __version_info__))# + "-dev0"
 __author__ = "Decompollaborate"
 
 from . import common as common

spimdisasm/common/ContextSymbols.py

@@ -426,6 +426,23 @@ def isLateRodata(self) -> bool:
         # if self.referenceCounter > 1: return False # ?
         return self.isJumpTable() or self.isFloat() or self.isDouble()
 
+
+    def notPointerByType(self) -> bool:
+        if self.isByte():
+            return True
+        if self.isShort():
+            return True
+        if self.isFloat():
+            return True
+        if self.isDouble():
+            return True
+        if self.isString():
+            return True
+        if self.isPascalString():
+            return True
+        return False
+
+
     def hasUserDeclaredSize(self) -> bool:
         return self.userDeclaredSize is not None
 

spimdisasm/common/Relocation.py

@@ -175,11 +175,13 @@ def getName(self, isSplittedSymbol: bool=False) -> str:
             return f"{name} - 0x{-self.addend:X}"
         return f"{name} + 0x{self.addend:X}"
 
-    def getNameWithReloc(self, isSplittedSymbol: bool=False) -> str:
+    def getNameWithReloc(self, *, isSplittedSymbol: bool=False, ignoredRelocs: set[RelocType]=set()) -> str:
         name = self.getName(isSplittedSymbol=isSplittedSymbol)
 
         percentRel = self.relocType.getPercentRel()
         if percentRel is not None:
+            if self.relocType in ignoredRelocs:
+                return f"({name})"
             return f"{percentRel}({name})"
 
         wordRel = self.relocType.getWordRel()

spimdisasm/elf32/Elf32Constants.py

@@ -346,9 +346,11 @@ class Elf32SectionHeaderNumber(enum.Enum):
     UNDEF           = 0
     ABS             = 0xFFF1
     COMMON          = 0xFFF2
-    MIPS_ACOMMON    = 0xFF00
-    MIPS_TEXT       = 0xFF01
-    MIPS_DATA       = 0xFF02
+    MIPS_ACOMMON    = 0xFF00 # Allocated common symbols.
+    MIPS_TEXT       = 0xFF01 # Allocated test symbols.
+    MIPS_DATA       = 0xFF02 # Allocated data symbols.
+    MIPS_SCOMMON    = 0xFF03 # Small common symbols.
+    MIPS_SUNDEFINED = 0xFF04 # Small undefined symbols.
 
     @staticmethod
     def fromValue(value: int) -> Elf32SectionHeaderNumber|None:

spimdisasm/elf32/Elf32File.py

@@ -543,7 +543,7 @@ def _readelf_symbol_table(self, symbolTable: Elf32Syms, stringTable: Elf32String
 
         print(f"Symbol table '{symbolTableName}' contains {len(symbolTable.symbols)} entries:")
 
-        print(f" {'Num':>5}: {'Value':>8} {'Size':>5} {'Type':7} {'Bind':6} {'Vis':9} {'Ndx':>12} {'Name'}")
+        print(f" {'Num':>5}: {'Value':>8} {'Size':>5} {'Type':7} {'Bind':6} {'Vis':9} {'Ndx':>15} {'Name'}")
 
         for i, sym in enumerate(symbolTable.symbols):
             entryType = Elf32SymbolTableType(sym.stType)
@@ -562,11 +562,23 @@ def _readelf_symbol_table(self, symbolTable: Elf32Syms, stringTable: Elf32String
             shndx = Elf32SectionHeaderNumber.fromValue(sym.shndx)
             if shndx is not None:
                 ndx = shndx.name
+            elif entryType in { Elf32SymbolTableType.OBJECT, Elf32SymbolTableType.FUNC }:
+                # spimdisasm-extension: instead of a number we use the section name if available
+                section = self.sectionHeaders[sym.shndx]
+                if section is not None:
+                    ndx = self.shstrtab[section.name]
 
             symName = ""
             if stringTable is not None:
                 symName = stringTable[sym.name]
-            print(f" {i:>5}: {sym.value:08X} {sym.size:>5X} {entryType.name:7} {bind:6} {visibility:9} {ndx:>12} {symName}")
+            if symName == "":
+                # GNU readelf uses the section's name as the name column for sections instead of the symbol's name.
+                if entryType == Elf32SymbolTableType.SECTION:
+                    section = self.sectionHeaders[sym.shndx]
+                    if section is not None:
+                        symName = self.shstrtab[section.name]
+
+            print(f" {i:>5}: {sym.value:08X} {sym.size:>5X} {entryType.name:7} {bind:6} {visibility:9} {ndx:>15} {symName}")
 
         print()
 
@@ -601,6 +613,13 @@ def readelf_relocs(self) -> None:
                     symValue = f"{sym.value:08X}"
                     if self.strtab is not None:
                         symName = self.strtab[sym.name]
+                if symName == "":
+                    # Some relocations are an offset to a section on the current object instead of to a symtab symbol.
+                    # TODO: what is the proper way to check this?
+                    section = self.sectionHeaders[sym.shndx]
+                    if section is not None:
+                        symName = self.shstrtab[section.name]
+
                 print(f" {rel.offset:08X} {rel.info:08X} {relType:<12} {symValue:>9} {symName}")
 
             print()

spimdisasm/elf32/Elf32SectionHeaders.py

@@ -64,6 +64,12 @@ def __getitem__(self, key: int) -> Elf32SectionHeaderEntry | None:
             return self.mipsText
         if key == Elf32SectionHeaderNumber.MIPS_DATA.value:
             return self.mipsData
+        if key == Elf32SectionHeaderNumber.MIPS_SCOMMON.value:
+            common.Utils.eprint("Warning: Elf32SectionHeaderNumber.MIPS_SCOMMON not implemented\n")
+            return None
+        if key == Elf32SectionHeaderNumber.MIPS_SUNDEFINED.value:
+            common.Utils.eprint("Warning: Elf32SectionHeaderNumber.MIPS_SUNDEFINED not implemented\n")
+            return None
         if key > len(self.sections):
             return None
         return self.sections[key]

spimdisasm/mips/sections/MipsSectionData.py

@@ -54,7 +54,6 @@ def analyze(self) -> None:
             localOffset = 0
             for w in self.words:
                 currentVram = self.getVramOffset(localOffset)
-                currentVrom = self.getVromOffset(localOffset)
 
                 if self.popPointerInDataReference(currentVram) is not None and localOffset not in localOffsetsWithSymbols:
                     contextSym = self._addOwnedSymbol(localOffset)

spimdisasm/mips/sections/MipsSectionRodata.py

@@ -72,8 +72,11 @@ def _analyze_processJumptable(self, localOffset: int, w: int, contextSym: common
     def analyze(self) -> None:
         lastVramSymbol: common.ContextSymbol = self._checkAndCreateFirstSymbol()
 
-        symbolList: list[tuple[int, int]] = []
+        symbolList: list[tuple[int, common.ContextSymbol]] = []
         localOffset = 0
+        localOffsetsWithSymbols: set[int] = set()
+
+        needsFurtherAnalyzis = False
 
         jumpTableSym: common.ContextSymbol|None = None
         firstJumptableWord = -1
@@ -97,17 +100,43 @@ def analyze(self) -> None:
                 self.checkWordIsASymbolReference(w)
 
             if contextSym is not None:
-                self.symbolsVRams.add(currentVram)
-                symbolList.append((localOffset, currentVram))
+                symbolList.append((localOffset, contextSym))
+                localOffsetsWithSymbols.add(localOffset)
 
                 self._createAutoPadFromSymbol(localOffset, contextSym)
 
+            elif jumpTableSym is None and self.popPointerInDataReference(currentVram) is not None:
+                contextSym = self._addOwnedSymbol(localOffset)
+                symbolList.append((localOffset, contextSym))
+                localOffsetsWithSymbols.add(localOffset)
+
+            if not lastVramSymbol.notPointerByType():
+                if self.checkWordIsASymbolReference(w):
+                    if w < currentVram and self.containsVram(w):
+                        # References a data symbol from this section and it is behind this current symbol
+                        needsFurtherAnalyzis = True
+
             localOffset += 4
 
+        if needsFurtherAnalyzis:
+            localOffset = 0
+            for w in self.words:
+                currentVram = self.getVramOffset(localOffset)
+
+                if self.popPointerInDataReference(currentVram) is not None and localOffset not in localOffsetsWithSymbols:
+                    contextSym = self._addOwnedSymbol(localOffset)
+                    symbolList.append((localOffset, contextSym))
+                    localOffsetsWithSymbols.add(localOffset)
+
+                localOffset += 4
+
+            # Since we appended new symbols, this list is not sorted anymore
+            symbolList.sort()
+
         previousSymbolWasLateRodata = False
         previousSymbolExtraPadding = 0
 
-        for i, (offset, vram) in enumerate(symbolList):
+        for i, (offset, contextSym) in enumerate(symbolList):
             if i + 1 == len(symbolList):
                 words = self.words[offset//4:]
             else:
@@ -116,12 +145,13 @@ def analyze(self) -> None:
 
             vrom = self.getVromOffset(offset)
             vromEnd = vrom + len(words)*4
-            sym = symbols.SymbolRodata(self.context, vrom, vromEnd, offset + self.inFileOffset, vram, words, self.segmentVromStart, self.overlayCategory)
+            sym = symbols.SymbolRodata(self.context, vrom, vromEnd, offset + self.inFileOffset, contextSym.vram, words, self.segmentVromStart, self.overlayCategory)
             sym.parent = self
             sym.setCommentOffset(self.commentOffset)
             sym.stringEncoding = self.stringEncoding
             sym.analyze()
             self.symbolList.append(sym)
+            self.symbolsVRams.add(contextSym.vram)
 
             # File boundaries detection
             if sym.inFileOffset % 16 == 0:

spimdisasm/mips/sections/MipsSectionText.py

@@ -21,6 +21,8 @@ def __init__(self, context: common.Context, vromStart: int, vromEnd: int, vram:
 
         self.instrCat: rabbitizer.Enum = rabbitizer.InstrCategory.CPU
         self.detectRedundantFunctionEnd: bool|None = None
+        self.gpRelHack: bool = False
+        """Get rid of `%gp_rel` and `$gp` since old assemblers don't support `%gp_rel`."""
 
         self.enableStringGuessing = False
 
@@ -319,6 +321,7 @@ def analyze(self) -> None:
             func.hasUnimplementedIntrs = hasUnimplementedIntrs
             func.parent = self
             func.isRsp = self.instrCat == rabbitizer.InstrCategory.RSP
+            func.gpRelHack = self.gpRelHack
             func.analyze()
             self.symbolList.append(func)