Semi-working REL analysis & splitting

Cargo.toml

@@ -29,7 +29,6 @@ base64 = "0.21.2"
 byteorder = "1.4.3"
 cwdemangle = "0.1.5"
 dol = { git = "https://github.com/encounter/ppc750cl", rev = "5f6e991bf495388c4104f188d2e90c79da9f78de" }
-env_logger = "0.10.0"
 filetime = "0.2.21"
 fixedbitset = "0.4.2"
 flagset = { version = "0.4.3", features = ["serde"] }
@@ -45,15 +44,20 @@ multimap = "0.9.0"
 num_enum = "0.6.1"
 object = { version = "0.31.1", features = ["read_core", "std", "elf", "write_std"], default-features = false }
 once_cell = "1.18.0"
+path-slash = "0.2.1"
 petgraph = "0.6.3"
 ppc750cl = { git = "https://github.com/encounter/ppc750cl", rev = "5f6e991bf495388c4104f188d2e90c79da9f78de" }
+rayon = "1.7.0"
 regex = "1.9.0"
 serde = "1.0.166"
 serde_json = "1.0.104"
 serde_repr = "0.1.14"
 serde_yaml = "0.9.22"
 sha-1 = "0.10.1"
 smallvec = "1.11.0"
+tracing = "0.1.37"
+tracing-attributes = "0.1.26"
+tracing-subscriber = "0.3.17"
 
 [build-dependencies]
 anyhow = { version = "1.0.71", features = ["backtrace"] }

assets/ldscript_partial.lcf

@@ -0,0 +1,26 @@
+SECTIONS
+{
+    GROUP:
+    {
+        .init   :{}
+        .text   :{}
+        .ctors  :{}
+        .dtors  :{}
+        .rodata :{}
+        .data   :{ *(.data) *(extabindex) *(extab) }
+        .bss    :{}
+    }
+}
+
+FORCEACTIVE
+{
+	_unresolved
+	_prolog
+	_epilog
+	$FORCEACTIVE
+}
+
+FORCEFILES
+{
+    $FORCEFILES
+}

src/analysis/cfa.rs

@@ -24,13 +24,13 @@ pub struct SectionAddress {
 
 impl Debug for SectionAddress {
     fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{}:{:#010X}", self.section as isize, self.address)
+        write!(f, "{}:{:#X}", self.section as isize, self.address)
     }
 }
 
 impl Display for SectionAddress {
     fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
-        write!(f, "{}:{:#010X}", self.section as isize, self.address)
+        write!(f, "{}:{:#X}", self.section as isize, self.address)
     }
 }
 
@@ -131,9 +131,19 @@ impl AnalyzerState {
                 section.address,
                 section.address + section.size
             );
+            let address_str = if obj.module_id == 0 {
+                format!("{:08X}", addr.address)
+            } else {
+                format!(
+                    "{}_{}_{:X}",
+                    obj.module_id,
+                    section.name.trim_start_matches('.'),
+                    addr.address
+                )
+            };
             obj.add_symbol(
                 ObjSymbol {
-                    name: format!("jumptable_{:08X}", addr.address),
+                    name: format!("jumptable_{}", address_str),
                     demangled_name: None,
                     address: addr.address as u64,
                     section: Some(addr.section),

src/analysis/objects.rs

@@ -5,7 +5,7 @@ use crate::{
     util::split::is_linker_generated_label,
 };
 
-pub fn detect_object_boundaries(obj: &mut ObjInfo) -> Result<()> {
+pub fn detect_objects(obj: &mut ObjInfo) -> Result<()> {
     for (section_index, section) in
         obj.sections.iter_mut().filter(|(_, s)| s.kind != ObjSectionKind::Code)
     {

src/analysis/pass.rs

@@ -151,8 +151,9 @@ impl AnalysisPass for FindRelCtorsDtors {
         let possible_sections = obj
             .sections
             .iter()
-            .filter(|&(_, section)| {
+            .filter(|&(index, section)| {
                 if section.section_known
+                    || state.known_sections.contains_key(&index)
                     || !matches!(section.kind, ObjSectionKind::Data | ObjSectionKind::ReadOnlyData)
                     || section.size < 4
                 {
@@ -283,3 +284,40 @@ impl AnalysisPass for FindRelCtorsDtors {
         Ok(())
     }
 }
+
+pub struct FindRelRodataData {}
+
+impl AnalysisPass for FindRelRodataData {
+    fn execute(state: &mut AnalyzerState, obj: &ObjInfo) -> Result<()> {
+        ensure!(obj.kind == ObjKind::Relocatable);
+
+        match (obj.sections.by_name(".rodata")?, obj.sections.by_name(".data")?) {
+            (None, None) => {}
+            _ => return Ok(()),
+        }
+
+        let possible_sections = obj
+            .sections
+            .iter()
+            .filter(|&(index, section)| {
+                !section.section_known
+                    && !state.known_sections.contains_key(&index)
+                    && matches!(section.kind, ObjSectionKind::Data | ObjSectionKind::ReadOnlyData)
+            })
+            .collect_vec();
+
+        if possible_sections.len() != 2 {
+            log::warn!("Failed to find .rodata and .data");
+            return Ok(());
+        }
+
+        log::debug!("Found .rodata and .data: {:?}", possible_sections);
+        let rodata_section_index = possible_sections[0].0;
+        state.known_sections.insert(rodata_section_index, ".rodata".to_string());
+
+        let data_section_index = possible_sections[1].0;
+        state.known_sections.insert(data_section_index, ".data".to_string());
+
+        Ok(())
+    }
+}

src/analysis/slices.rs

@@ -309,8 +309,8 @@ impl FunctionSlices {
                                 }
                             }
                         }
-                        BranchTarget::JumpTable { .. } => {
-                            bail!("Conditional jump table unsupported @ {:#010X}", ins_addr);
+                        BranchTarget::JumpTable { address, size } => {
+                            bail!("Conditional jump table unsupported @ {:#010X} -> {:#010X} size {:#X?}", ins_addr, address, size);
                         }
                     }
                 }

src/analysis/tracker.rs

@@ -5,6 +5,8 @@ use std::{
 
 use anyhow::{bail, Result};
 use ppc750cl::Opcode;
+use tracing::{debug_span, info_span};
+use tracing_attributes::instrument;
 
 use crate::{
     analysis::{
@@ -88,8 +90,8 @@ impl Tracker {
         }
     }
 
+    #[instrument(name = "tracker", skip(self, obj))]
     pub fn process(&mut self, obj: &ObjInfo) -> Result<()> {
-        log::debug!("Processing code sections");
         self.process_code(obj)?;
         for (section_index, section) in obj
             .sections
@@ -151,6 +153,7 @@ impl Tracker {
     ) -> Result<ExecCbResult<()>> {
         let ExecCbData { executor, vm, result, ins_addr, section: _, ins, block_start: _ } = data;
         let is_function_addr = |addr: SectionAddress| addr >= function_start && addr < function_end;
+        let _span = debug_span!("ins", addr = %ins_addr, op = ?ins.op).entered();
 
         match result {
             StepResult::Continue => {
@@ -310,8 +313,20 @@ impl Tracker {
                                 executor.push(addr, branch.vm, true);
                             }
                         }
-                        BranchTarget::JumpTable { .. } => {
-                            bail!("Conditional jump table unsupported @ {:#010X}", ins_addr)
+                        BranchTarget::JumpTable { address, size } => {
+                            let (entries, _) = uniq_jump_table_entries(
+                                obj,
+                                address,
+                                size,
+                                ins_addr,
+                                function_start,
+                                Some(function_end),
+                            )?;
+                            for target in entries {
+                                if is_function_addr(target) {
+                                    executor.push(target, branch.vm.clone_all(), true);
+                                }
+                            }
                         }
                     }
                 }
@@ -326,6 +341,9 @@ impl Tracker {
         };
         let function_start = SectionAddress::new(section_index, symbol.address as u32);
         let function_end = function_start + symbol.size as u32;
+        let _span =
+            info_span!("fn", name = %symbol.name, start = %function_start, end = %function_end)
+                .entered();
 
         // The compiler can sometimes create impossible-to-reach branches,
         // but we still want to track them.
@@ -461,6 +479,7 @@ impl Tracker {
             .or_else(|| check_symbol(self.sda_base, "_SDA_BASE_"))
     }
 
+    #[instrument(name = "apply", skip(self, obj))]
     pub fn apply(&self, obj: &mut ObjInfo, replace: bool) -> Result<()> {
         fn apply_section_name(section: &mut ObjSection, name: &str) {
             let module_id = if let Some((_, b)) = section.name.split_once(':') {

src/analysis/vm.rs

@@ -126,8 +126,11 @@ pub fn section_address_for(
         let (section_index, _) = obj.sections.at_address(target_addr).ok()?;
         return Some(SectionAddress::new(section_index, target_addr));
     }
-    // TODO: relative jumps within relocatable objects?
-    None
+    if obj.sections[ins_addr.section].contains(target_addr) {
+        Some(SectionAddress::new(ins_addr.section, target_addr))
+    } else {
+        None
+    }
 }
 
 impl VM {
@@ -180,11 +183,11 @@ impl VM {
     pub fn clone_all(&self) -> Box<Self> { Box::new(self.clone()) }
 
     pub fn step(&mut self, obj: &ObjInfo, ins_addr: SectionAddress, ins: &Ins) -> StepResult {
-        let relocation_target = relocation_target_for(obj, ins_addr, None).ok().flatten();
-        if let Some(_target) = relocation_target {
-            let _defs = ins.defs();
-            // TODO
-        }
+        // let relocation_target = relocation_target_for(obj, ins_addr, None).ok().flatten();
+        // if let Some(_target) = relocation_target {
+        //     let _defs = ins.defs();
+        //     // TODO
+        // }
 
         match ins.op {
             Opcode::Illegal => {