Merge branch 'lsda_sym_minus_sym' into 'main'

Check on both symbols in symbol_minus_symbol for END symbols

Closes #599 and debloating/dykondo#128

See merge request rewriting/ddisasm!1229

CHANGELOG.md

@@ -22,6 +22,8 @@
 * Fix 16-Thumb STM instructions considered to be invalid if the same register
   is used in reglist and register operands with writeback enabled.
 * Fixed bug that could result in missed symbolic expressions with TLS variables for `local-executable` TLS model
+* Fix bug that caused assembling error due to wrong `symbol_minus_symbol`
+  for lsda entries with references to the end of `.gcc_except_table`
 
 # 1.9.0
 

examples/asm_examples/ex_exceptions5/Makefile

@@ -0,0 +1,18 @@
+CXX="g++"
+EXEC=
+
+all: ex_original.s
+	# NOTE: `-T linker-script.ld`
+	# This is to place .gcc_except_table *before* .note.gnu.build-id
+	# so that the reference label for the end of the section is aligned
+	# with the adjacent section. Otherwise, the next section can be some
+	# discarded one, which would prevent from demonstrating the issue.
+	#
+	$(CXX) -T linker-script.ld -o ex ex_original.s
+	@ $(EXEC) ./ex > out.txt
+clean:
+	rm -f ex out.txt
+	rm -fr ex.unstripped *.old*  dl_files *.gtirb
+check:
+	@ $(EXEC) ./ex >/tmp/res.txt
+	@ diff out.txt /tmp/res.txt && echo TEST OK

examples/asm_examples/ex_exceptions5/ex_original.s

@@ -0,0 +1,67 @@
+.section .text
+
+#-----------------------------------
+.type foo, @function
+foo:
+
+.cfi_startproc
+.cfi_lsda 27, .L_call_site_start
+    endbr64
+    pushq %rbp
+    movq %rsp,%rbp
+    nop
+    popq %rbp
+    retq
+.cfi_endproc
+
+# Entry point
+.globl main
+.type main, @function
+main:
+    call callThrower    # Call a function that may "throw" an exception
+    mov $60, %rax       # syscall: exit
+    xor %rdi, %rdi      # status: 0
+    syscall
+
+# Dummy function simulating an exception thrower
+.type callThrower, @function
+callThrower:
+    ret                 # Simply return (replace with an actual throw in C++)
+
+# Exception Handling Table
+.section .gcc_except_table, "a", @progbits
+.align 4
+
+.L_entry_start:
+    .byte 0x1           # Entry indicating an exception
+    .byte 0x0
+    .byte 0x0
+    .byte 0x7d
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+    .byte 0x0
+
+.L_call_site_start:
+    .byte 0xff          # Start of a call site
+    .byte 0xff          # Additional indicator byte
+    .byte 0x1           # Additional indicator byte
+    #
+    # With this example, if a boundary_sym_expr is not correctly created
+    # for symbol_minus_symbol (either the first or the second symbol, or
+    # both) or an END symbol is not chosen for such a symbol,
+    # the assembler will fail with an error, such as
+    #
+    # "Error: invalid operands (.note.gnu.property and .gcc_except_table
+    # sections) for '-'"
+    #
+    .uleb128 .L_end - .L_end  # Zero-length entry at the very end of the table
+.L_end:

examples/asm_examples/ex_exceptions5/linker-script.ld

@@ -0,0 +1,4 @@
+SECTIONS {
+  .gcc_except_table : { *(.gcc_except_table) }
+}
+INSERT BEFORE .note.gnu.build-id;

src/datalog/binary/elf/exceptions.dl

@@ -176,20 +176,20 @@ labeled_ea(Personality):-
 Same as `symbol_minus_symbol`: this is to avoid cyclic aggregation by
 `boundary_sym_expr` using `symbol_minus_symbol`.
 */
-.decl lsda_symbol_minus_symbol(EA:address,Size:unsigned,Symbol1:address,Symbol2:address,Scale:unsigned,Offset:number)
+.decl lsda_symbol_minus_symbol(EA:address,Size:unsigned,Symbol1:address,Symbol2:address,Scale:unsigned,Offset:number,Reason:symbol)
 
 symbol_minus_symbol(EA,Size,Symbol1,Symbol2,Scale,Offset):-
-    lsda_symbol_minus_symbol(EA,Size,Symbol1,Symbol2,Scale,Offset).
+    lsda_symbol_minus_symbol(EA,Size,Symbol1,Symbol2,Scale,Offset,_).
 
 // If the second address is the end of the section, create `boundary_sym_expr`
 // for the address.
 boundary_sym_expr(EA, Dest):-
-    lsda_symbol_minus_symbol(EA,_,_,Dest,_,_),
+    lsda_symbol_minus_symbol(EA,_,_,Dest,_,_,_),
     loaded_section(_,Dest,_).
 
 // type table pointer
 symbol_special_encoding(TypeTablePointerLocation,"uleb128"),
-lsda_symbol_minus_symbol(TypeTablePointerLocation,Size,CallsiteTablePointerLoc-1,LsdaTypeTableAddress,1,0):-
+lsda_symbol_minus_symbol(TypeTablePointerLocation,Size,CallsiteTablePointerLoc-1,LsdaTypeTableAddress,1,0,"type-table-pointer"):-
     lsda_pointer_locations(Lsda,TypeTablePointerLocation,CallsiteTablePointerLoc),
     lsda(Lsda,_,_,_,LsdaTypeTableAddress, _, _),
     LsdaTypeTableAddress != 0,
@@ -198,7 +198,7 @@ lsda_symbol_minus_symbol(TypeTablePointerLocation,Size,CallsiteTablePointerLoc-1
 
 //callsite table size
 symbol_special_encoding(CallsiteTablePointerLoc,"uleb128"),
-lsda_symbol_minus_symbol(CallsiteTablePointerLoc,Size,CallSiteTable_address,CallSiteTable_address+CallSiteTableLength,1,0):-
+lsda_symbol_minus_symbol(CallsiteTablePointerLoc,Size,CallSiteTable_address,CallSiteTable_address+CallSiteTableLength,1,0,"callsite-table-size"):-
     lsda_pointer_locations(Lsda,_,CallsiteTablePointerLoc),
     lsda(Lsda,CallSiteTable_address,_,CallSiteTableLength,_, _, _),
     Size = CallSiteTable_address - CallsiteTablePointerLoc,
@@ -217,23 +217,23 @@ symbol_special_encoding(EA_landingPad,EncodingName):-
 
 
 //region beginning
-lsda_symbol_minus_symbol(EA_start,Size,LandingPadBaseAddress,StartRegion,1,0):-
+lsda_symbol_minus_symbol(EA_start,Size,LandingPadBaseAddress,StartRegion,1,0,"region-beginning"):-
     lsda_callsite(CallSiteTable_address, EA_start, StartRegion,EA_end,_,_,_,_),
     lsda(_,CallSiteTable_address,_,_,_,_,LandingPadBaseAddress),
     Size = EA_end-EA_start,
     Size > 0.
 
 //region end
 boundary_sym_expr(EA_end,EndRegion),
-lsda_symbol_minus_symbol(EA_end,Size,StartRegion,EndRegion,1,0):-
+lsda_symbol_minus_symbol(EA_end,Size,StartRegion,EndRegion,1,0,"region-end"):-
     lsda_callsite(_, _, StartRegion,EA_end,EndRegion,EA_landingPad,_,_),
     Size = EA_landingPad-EA_end,
     Size > 0,
     // Boundary is at the end of the block preceding the EndRegion label.
     block_boundaries(_,_,EndRegion).
 
 //landing pad
-lsda_symbol_minus_symbol(EA_landingPad,Size,LandingPadBaseAddress,LandingPad,1,0):-
+lsda_symbol_minus_symbol(EA_landingPad,Size,LandingPadBaseAddress,LandingPad,1,0,"landing-pad"):-
     lsda_callsite(CallSiteTable_address, _, _,_,_,EA_landingPad,LandingPad,EA_endLandingPad),
     lsda(_,CallSiteTable_address,_,_,_,_,LandingPadBaseAddress),
     LandingPad != 0,
@@ -251,7 +251,7 @@ symbolic_data(EAType,EncodingSize,Type):-
     EAType=LsdaTypeTableAddress-(EncodingSize*(Index+1)),
     Type != 0.
 
-lsda_symbol_minus_symbol(EAType,EncodingSize,EAType,Type,1,0):-
+lsda_symbol_minus_symbol(EAType,EncodingSize,EAType,Type,1,0,"type-table"):-
     lsda_type_entry(LsdaTypeTableAddress, Index,Type),
     lsda(_,_,_,_,LsdaTypeTableAddress, TypeTableEncoding, _),
     TypeTableEncoding band 0xF = Encoding,

src/datalog/symbolization.dl

@@ -954,7 +954,11 @@ symbolic_expr_symbol_minus_symbol(EA,Size,SymbolName1,SymbolName2,Scale,FinalOff
         NewSymbol2 = Symbol2,
         FinalOffset = Offset1
     ),
-    best_symexpr_symbol(NewSymbol1,SymbolName1,"Beg"),
+    (
+        NewSymbol1 = NewSymbol2, Symbol1Loc = Symbol2Loc;
+        NewSymbol1 != NewSymbol2, Symbol1Loc = "Beg"
+    ),
+    best_symexpr_symbol(NewSymbol1,SymbolName1,Symbol1Loc),
     (
         boundary_sym_expr(EA,NewSymbol2), Symbol2Loc = "End"
         ;

tests/linux-elf-x64.yaml

@@ -512,9 +512,11 @@ tests:
     <<: *assembly
     binary: fun.so
 
+  - name: ex_exceptions5
+    <<: *assembly
+
   - name: ex_tls_local_exec
     <<: *assembly
-    binary: ex
 
   # ----------------------------------------------------------------------------
   # Assembly examples. (stripped)
@@ -601,10 +603,13 @@ tests:
     <<: *test-strip-default
     binary: fun.so
 
+  - name: ex_exceptions5
+    <<: *assembly
+    <<: *test-strip-default
+
   - name: ex_tls_local_exec
     <<: *assembly
     <<: *test-strip-default
-    binary: ex
 
   # ----------------------------------------------------------------------------
   # Relocatable ELF objects (.o).