PPC: Guess reloc data type based on the instruction. (#108)

SquareMan · web-flow · commit a43320af1f92 · 2024-09-25T23:45:37.000-06:00
* Guess reloc data type based on the instruction.

Adds an entry to the reloc tooltip to show the inferred data type
and value.

* Fix clippy warning

* Match on Opcode rather than mnemonic string
diff --git a/objdiff-core/src/arch/mod.rs b/objdiff-core/src/arch/mod.rs
@@ -1,11 +1,13 @@
-use std::{borrow::Cow, collections::BTreeMap};
+use std::{borrow::Cow, collections::BTreeMap, ffi::CStr};
 
 use anyhow::{bail, Result};
+use byteorder::ByteOrder;
 use object::{Architecture, File, Object, ObjectSymbol, Relocation, RelocationFlags, Symbol};
 
 use crate::{
     diff::DiffObjConfig,
     obj::{ObjIns, ObjReloc, ObjSection},
+    util::ReallySigned,
 };
 
 #[cfg(feature = "arm")]
@@ -17,6 +19,97 @@ pub mod ppc;
 #[cfg(feature = "x86")]
 pub mod x86;
 
+/// Represents the type of data associated with an instruction
+pub enum DataType {
+    Int8,
+    Int16,
+    Int32,
+    Int64,
+    Int128,
+    Float,
+    Double,
+    Bytes,
+    String,
+}
+
+impl DataType {
+    pub fn display_bytes<Endian: ByteOrder>(&self, bytes: &[u8]) -> Option<String> {
+        if self.required_len().is_some_and(|l| bytes.len() < l) {
+            return None;
+        }
+
+        match self {
+            DataType::Int8 => {
+                let i = i8::from_ne_bytes(bytes.try_into().unwrap());
+                if i < 0 {
+                    format!("Int8: {:#x} ({:#x})", i, ReallySigned(i))
+                } else {
+                    format!("Int8: {:#x}", i)
+                }
+            }
+            DataType::Int16 => {
+                let i = Endian::read_i16(bytes);
+                if i < 0 {
+                    format!("Int16: {:#x} ({:#x})", i, ReallySigned(i))
+                } else {
+                    format!("Int16: {:#x}", i)
+                }
+            }
+            DataType::Int32 => {
+                let i = Endian::read_i32(bytes);
+                if i < 0 {
+                    format!("Int32: {:#x} ({:#x})", i, ReallySigned(i))
+                } else {
+                    format!("Int32: {:#x}", i)
+                }
+            }
+            DataType::Int64 => {
+                let i = Endian::read_i64(bytes);
+                if i < 0 {
+                    format!("Int64: {:#x} ({:#x})", i, ReallySigned(i))
+                } else {
+                    format!("Int64: {:#x}", i)
+                }
+            }
+            DataType::Int128 => {
+                let i = Endian::read_i128(bytes);
+                if i < 0 {
+                    format!("Int128: {:#x} ({:#x})", i, ReallySigned(i))
+                } else {
+                    format!("Int128: {:#x}", i)
+                }
+            }
+            DataType::Float => {
+                format!("Float: {}", Endian::read_f32(bytes))
+            }
+            DataType::Double => {
+                format!("Double: {}", Endian::read_f64(bytes))
+            }
+            DataType::Bytes => {
+                format!("Bytes: {:#?}", bytes)
+            }
+            DataType::String => {
+                format!("String: {:?}", CStr::from_bytes_until_nul(bytes).ok()?)
+            }
+        }
+        .into()
+    }
+
+    fn required_len(&self) -> Option<usize> {
+        match self {
+            DataType::Int8 => Some(1),
+            DataType::Int16 => Some(2),
+            DataType::Int32 => Some(4),
+            DataType::Int64 => Some(8),
+            DataType::Int128 => Some(16),
+            DataType::Float => Some(4),
+            DataType::Double => Some(8),
+            DataType::Bytes => None,
+            DataType::String => None,
+        }
+    }
+}
+
 pub trait ObjArch: Send + Sync {
     fn process_code(
         &self,
@@ -42,6 +135,12 @@ pub trait ObjArch: Send + Sync {
 
     fn symbol_address(&self, symbol: &Symbol) -> u64 { symbol.address() }
 
+    fn guess_data_type(&self, _instruction: &ObjIns) -> Option<DataType> { None }
+
+    fn display_data_type(&self, _ty: DataType, bytes: &[u8]) -> Option<String> {
+        Some(format!("Bytes: {:#x?}", bytes))
+    }
+
     // Downcast methods
     #[cfg(feature = "ppc")]
     fn ppc(&self) -> Option<&ppc::ObjArchPpc> { None }
diff --git a/objdiff-core/src/arch/ppc.rs b/objdiff-core/src/arch/ppc.rs
@@ -1,15 +1,16 @@
 use std::{borrow::Cow, collections::BTreeMap};
 
 use anyhow::{bail, ensure, Result};
+use byteorder::BigEndian;
 use cwextab::{decode_extab, ExceptionTableData};
 use object::{
     elf, File, Object, ObjectSection, ObjectSymbol, Relocation, RelocationFlags, RelocationTarget,
     Symbol, SymbolKind,
 };
-use ppc750cl::{Argument, InsIter, GPR};
+use ppc750cl::{Argument, InsIter, Opcode, GPR};
 
 use crate::{
-    arch::{ObjArch, ProcessCodeResult},
+    arch::{DataType, ObjArch, ProcessCodeResult},
     diff::DiffObjConfig,
     obj::{ObjIns, ObjInsArg, ObjInsArgValue, ObjReloc, ObjSection, ObjSymbol},
 };
@@ -186,6 +187,36 @@ impl ObjArch for ObjArchPpc {
         }
     }
 
+    fn guess_data_type(&self, instruction: &ObjIns) -> Option<super::DataType> {
+        // Always shows the first string of the table. Not ideal, but it's really hard to find
+        // the actual string being referenced.
+        if instruction.reloc.as_ref().is_some_and(|r| r.target.name.starts_with("@stringBase")) {
+            return Some(DataType::String);
+        }
+
+        // SAFETY: ppc750cl::Opcode is repr(u8) and op is originally obtained on PPC from casting
+        // an Opcode to a u8 so we know it's a valid value for Opcode.
+        match unsafe { std::mem::transmute::<u8, Opcode>(instruction.op as u8) } {
+            Opcode::Lbz | Opcode::Lbzu | Opcode::Lbzux | Opcode::Lbzx => Some(DataType::Int8),
+            Opcode::Lhz | Opcode::Lhzu | Opcode::Lhzux | Opcode::Lhzx => Some(DataType::Int16),
+            Opcode::Lha | Opcode::Lhau | Opcode::Lhaux | Opcode::Lhax => Some(DataType::Int16),
+            Opcode::Lwz | Opcode::Lwzu | Opcode::Lwzux | Opcode::Lwzx => Some(DataType::Int32),
+            Opcode::Lfs | Opcode::Lfsu | Opcode::Lfsux | Opcode::Lfsx => Some(DataType::Float),
+            Opcode::Lfd | Opcode::Lfdu | Opcode::Lfdux | Opcode::Lfdx => Some(DataType::Double),
+
+            Opcode::Stb | Opcode::Stbu | Opcode::Stbux | Opcode::Stbx => Some(DataType::Int8),
+            Opcode::Sth | Opcode::Sthu | Opcode::Sthux | Opcode::Sthx => Some(DataType::Int16),
+            Opcode::Stw | Opcode::Stwu | Opcode::Stwux | Opcode::Stwx => Some(DataType::Int32),
+            Opcode::Stfs | Opcode::Stfsu | Opcode::Stfsux | Opcode::Stfsx => Some(DataType::Float),
+            Opcode::Stfd | Opcode::Stfdu | Opcode::Stfdux | Opcode::Stfdx => Some(DataType::Double),
+            _ => None,
+        }
+    }
+
+    fn display_data_type(&self, ty: DataType, bytes: &[u8]) -> Option<String> {
+        ty.display_bytes::<BigEndian>(bytes)
+    }
+
     fn ppc(&self) -> Option<&ObjArchPpc> { Some(self) }
 }
 
diff --git a/objdiff-core/src/obj/mod.rs b/objdiff-core/src/obj/mod.rs
@@ -126,6 +126,7 @@ pub struct ObjSymbol {
     pub virtual_address: Option<u64>,
     /// Original index in object symbol table
     pub original_index: Option<usize>,
+    pub bytes: Vec<u8>,
 }
 
 pub struct ObjInfo {
diff --git a/objdiff-core/src/obj/read.rs b/objdiff-core/src/obj/read.rs
@@ -78,6 +78,16 @@ fn to_obj_symbol(
     let virtual_address = split_meta
         .and_then(|m| m.virtual_addresses.as_ref())
         .and_then(|v| v.get(symbol.index().0).cloned());
+
+    let bytes = symbol
+        .section_index()
+        .and_then(|idx| obj_file.section_by_index(idx).ok())
+        .and_then(|section| section.data().ok())
+        .and_then(|data| {
+            data.get(section_address as usize..(section_address + symbol.size()) as usize)
+        })
+        .unwrap_or(&[]);
+
     Ok(ObjSymbol {
         name: name.to_string(),
         demangled_name,
@@ -89,6 +99,7 @@ fn to_obj_symbol(
         addend,
         virtual_address,
         original_index: Some(symbol.index().0),
+        bytes: bytes.to_vec(),
     })
 }
 
@@ -179,6 +190,7 @@ fn symbols_by_section(
             addend: 0,
             virtual_address: None,
             original_index: None,
+            bytes: Vec::new(),
         });
     }
     Ok(result)
@@ -239,6 +251,7 @@ fn find_section_symbol(
         addend: offset_addr as i64,
         virtual_address: None,
         original_index: None,
+        bytes: Vec::new(),
     })
 }
 
@@ -521,6 +534,7 @@ fn update_combined_symbol(symbol: ObjSymbol, address_change: i64) -> Result<ObjS
             None
         },
         original_index: symbol.original_index,
+        bytes: symbol.bytes,
     })
 }
 
diff --git a/objdiff-gui/src/views/function_diff.rs b/objdiff-gui/src/views/function_diff.rs
@@ -79,6 +79,12 @@ fn ins_hover_ui(
                     appearance.highlight_color,
                     format!("Size: {:x}", reloc.target.size),
                 );
+                if let Some(s) = arch
+                    .guess_data_type(ins)
+                    .and_then(|ty| arch.display_data_type(ty, &reloc.target.bytes))
+                {
+                    ui.colored_label(appearance.highlight_color, s);
+                }
             } else {
                 ui.colored_label(appearance.highlight_color, "Extern".to_string());
             }

Original file line number	Diff line number	Diff line change
`@@ -126,6 +126,7 @@ pub struct ObjSymbol {`
`126`	`126`	`pub virtual_address: Option<u64>,`
`127`	`127`	`/// Original index in object symbol table`
`128`	`128`	`pub original_index: Option<usize>,`
	`129`	`+ pub bytes: Vec<u8>,`
`129`	`130`	`}`
`130`	`131`
`131`	`132`	`pub struct ObjInfo {`