fix(debug): handle mapping edges, stray traps and dying debuggees

Assisted-by: GLM 5.3

debug/repl.go

@@ -71,7 +71,12 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 		case "step", "s":
 			n := 1
 			if len(parts) > 1 {
-				n, _ = strconv.Atoi(parts[1])
+				v, err := strconv.Atoi(parts[1])
+				if err != nil || v < 0 {
+					fmt.Printf("invalid count: %s\n", parts[1])
+					continue
+				}
+				n = v
 			}
 			for range n {
 				if s.Exited() {
@@ -82,8 +87,13 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 					fmt.Println(err)
 					break
 				}
+				if sig := s.LastSignal(); sig != 0 {
+					regs, _ := s.GetRegs()
+					fmt.Printf("stopped on signal %v at %#x\n", sig, regs.GetPC())
+					break
+				}
 			}
-			if !s.Exited() {
+			if !s.Exited() && s.LastSignal() == 0 {
 				regs, _ := s.GetRegs()
 				pc := regs.GetPC()
 				text, _, _ := s.Disassemble(pc)
@@ -106,16 +116,16 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 				}
 				if err := s.Continue(); err != nil {
 					fmt.Println(err)
-					bm.Clear(afterAddr)
+					clearNextBp(bm, afterAddr)
 					continue
 				}
 				if s.Exited() {
-					bm.Clear(afterAddr)
+					clearNextBp(bm, afterAddr)
 					fmt.Println("debuggee exited")
 					continue
 				}
 				if sig := s.LastSignal(); sig != 0 {
-					bm.Clear(afterAddr)
+					clearNextBp(bm, afterAddr)
 					regs, _ := s.GetRegs()
 					fmt.Printf("stopped on signal %v at %#x\n", sig, regs.GetPC())
 					continue
@@ -125,12 +135,17 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 				// snapshot, and a stale SetRegs would clobber live state.
 				regs, _ = s.GetRegs()
 				bm.HandleTrap(&regs)
-				bm.Clear(afterAddr)
+				clearNextBp(bm, afterAddr)
 			} else {
 				if err := s.Step(); err != nil {
 					fmt.Println(err)
 					continue
 				}
+				if sig := s.LastSignal(); sig != 0 {
+					regs, _ := s.GetRegs()
+					fmt.Printf("stopped on signal %v at %#x\n", sig, regs.GetPC())
+					continue
+				}
 			}
 			if !s.Exited() {
 				regs, _ := s.GetRegs()
@@ -156,16 +171,16 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 			}
 			if err := s.Continue(); err != nil {
 				fmt.Println(err)
-				bm.Clear(retAddr)
+				clearNextBp(bm, retAddr)
 				continue
 			}
 			if s.Exited() {
-				bm.Clear(retAddr)
+				clearNextBp(bm, retAddr)
 				fmt.Println("debuggee exited")
 				continue
 			}
 			if sig := s.LastSignal(); sig != 0 {
-				bm.Clear(retAddr)
+				clearNextBp(bm, retAddr)
 				regs, _ := s.GetRegs()
 				fmt.Printf("stopped on signal %v at %#x\n", sig, regs.GetPC())
 				continue
@@ -174,7 +189,7 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 			// does: HandleTrap must see the PC the trap left behind.
 			regs, _ = s.GetRegs()
 			bm.HandleTrap(&regs)
-			bm.Clear(retAddr)
+			clearNextBp(bm, retAddr)
 			if s.Exited() {
 				fmt.Println("debuggee exited")
 			} else {
@@ -214,6 +229,7 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 					break
 				}
 				regs, _ := s.GetRegs()
+				trapPC := regs.GetPC()
 				if bp := bm.HandleTrap(&regs); bp != nil {
 					// Execute the instruction under the restored breakpoint
 					// so the next continue cannot re-trap on the same
@@ -229,6 +245,13 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 					fmt.Printf("breakpoint hit: %s (func+%#x)\n", name, bp.Addr-codeBase-uint64(funcOffset))
 					break
 				}
+				// The trap matched no breakpoint of ours.  When the PC still
+				// stands on the trapping instruction, resuming would re-execute
+				// it and trap forever, so surface the stop instead of spinning.
+				if TrapStray(s, reason, trapPC) {
+					fmt.Printf("SIGTRAP at %#x matches no breakpoint; the PC did not advance\n", trapPC-uint64(breakpointPCAdjust))
+					break
+				}
 			}
 
 		case "break", "b":
@@ -326,6 +349,10 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 			length := 64
 			if len(parts) > 1 {
 				addr, _ = resolveAddr(parts[1], codeBase, uint64(funcOffset), labels)
+				if addr == 0 {
+					fmt.Printf("unknown address: %s\n", parts[1])
+					continue
+				}
 			}
 			if len(parts) > 2 {
 				// A malformed or non-positive length would panic
@@ -399,9 +426,13 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 		case "disas", "u":
 			n := 5
 			if len(parts) > 1 {
-				n, _ = strconv.Atoi(parts[1])
-				if n <= 0 {
-					n = 5
+				v, err := strconv.Atoi(parts[1])
+				if err != nil {
+					fmt.Printf("invalid count: %s\n", parts[1])
+					continue
+				}
+				if v > 0 {
+					n = v
 				}
 			}
 			regs, _ := s.GetRegs()
@@ -496,10 +527,18 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 					typ = WatchRead
 				case "w":
 					typ = WatchWrite
+				default:
+					fmt.Printf("unknown watchpoint type: %s (want r or w)\n", parts[2])
+					continue
 				}
 			}
 			if len(parts) > 3 {
-				size, _ = strconv.Atoi(parts[3])
+				v, err := strconv.Atoi(parts[3])
+				if err != nil || v <= 0 {
+					fmt.Printf("invalid size: %s\n", parts[3])
+					continue
+				}
+				size = v
 			}
 			slot := s.FindFreeWatchpointSlot()
 			if slot < 0 {
@@ -543,6 +582,15 @@ func REPL(s *Session, bm *Breakpoints, codeBase uint64, funcOffset, funcSize, ar
 	s.Kill()
 }
 
+// clearNextBp removes one of the temporary breakpoints the next and finish
+// commands plant, reporting a failure instead of silently leaving the trap
+// instruction behind in the debuggee.
+func clearNextBp(bm *Breakpoints, addr uint64) {
+	if err := bm.Clear(addr); err != nil {
+		fmt.Printf("cannot remove temporary breakpoint at %#x: %v\n", addr, err)
+	}
+}
+
 func hexDump(addr uint64, data []byte) {
 	for i := 0; i < len(data); i += 16 {
 		end := min(i+16, len(data))