Add crash handler on Windows (#11570)

spec/std/exception/call_stack_spec.cr

@@ -55,7 +55,7 @@ describe "Backtrace" do
     error.to_s.should contain("IndexError")
   end
 
-  pending_win32 "prints crash backtrace to stderr" do
+  it "prints crash backtrace to stderr" do
     sample = datapath("crash_backtrace_sample")
 
     _, output, error = compile_and_run_file(sample)

spec/std/exception_spec.cr

@@ -35,16 +35,14 @@ describe "Exception" do
     ex.inspect_with_backtrace.should contain("inner")
   end
 
-  {% unless flag?(:win32) %}
-    it "collect memory within ensure block" do
-      sample = datapath("collect_within_ensure")
+  it "collect memory within ensure block" do
+    sample = datapath("collect_within_ensure")
 
-      _, output, error = compile_and_run_file(sample, ["--release"])
+    _, output, error = compile_and_run_file(sample, ["--release"])
 
-      output.to_s.empty?.should be_true
-      error.to_s.should contain("Unhandled exception: Oh no! (Exception)")
-      error.to_s.should_not contain("Invalid memory access")
-      error.to_s.should_not contain("Illegal instruction")
-    end
-  {% end %}
+    output.to_s.empty?.should be_true
+    error.to_s.should contain("Unhandled exception: Oh no! (Exception)")
+    error.to_s.should_not contain("Invalid memory access")
+    error.to_s.should_not contain("Illegal instruction")
+  end
 end

spec/std/kernel_spec.cr

@@ -239,8 +239,8 @@ describe "at_exit" do
   end
 end
 
-pending_win32 describe: "seg fault" do
-  it "reports SIGSEGV" do
+describe "hardware exception" do
+  it "reports invalid memory access" do
     status, _, error = compile_and_run_source <<-'CODE'
       puts Pointer(Int64).null.value
     CODE
@@ -261,7 +261,7 @@ pending_win32 describe: "seg fault" do
       # will address this.
       status, _, error = compile_and_run_source <<-'CODE'
       def foo
-        y = StaticArray(Int8,512).new(0)
+        y = StaticArray(Int8, 512).new(0)
         foo
       end
       foo
@@ -272,10 +272,10 @@ pending_win32 describe: "seg fault" do
     end
   {% end %}
 
-  it "detects stack overflow on a fiber stack" do
+  pending_win32 "detects stack overflow on a fiber stack" do
     status, _, error = compile_and_run_source <<-'CODE'
       def foo
-        y = StaticArray(Int8,512).new(0)
+        y = StaticArray(Int8, 512).new(0)
         foo
       end
 

src/exception/call_stack/libunwind.cr

@@ -32,6 +32,10 @@ struct Exception::CallStack
     end
   {% end %}
 
+  def self.setup_crash_handler
+    Signal.setup_segfault_handler
+  end
+
   {% if flag?(:interpreted) %} @[Primitive(:interpreter_call_stack_unwind)] {% end %}
   protected def self.unwind : Array(Void*)
     callstack = [] of Void*

src/exception/call_stack/stackwalk.cr

@@ -1,4 +1,5 @@
 require "c/dbghelp"
+require "c/malloc"
 
 # :nodoc:
 struct Exception::CallStack
@@ -31,7 +32,49 @@ struct Exception::CallStack
     LibC.SymSetOptions(LibC.SymGetOptions | LibC::SYMOPT_UNDNAME | LibC::SYMOPT_LOAD_LINES | LibC::SYMOPT_FAIL_CRITICAL_ERRORS | LibC::SYMOPT_NO_PROMPTS)
   end
 
-  def self.unwind
+  def self.setup_crash_handler
+    LibC.AddVectoredExceptionHandler(1, ->(exception_info) do
+      case status = exception_info.value.exceptionRecord.value.exceptionCode
+      when LibC::EXCEPTION_ACCESS_VIOLATION
+        addr = exception_info.value.exceptionRecord.value.exceptionInformation[1]
+        Crystal::System.print_error "Invalid memory access (C0000005) at address 0x%llx\n", addr
+        print_backtrace(exception_info)
+        LibC._exit(1)
+      when LibC::EXCEPTION_STACK_OVERFLOW
+        LibC._resetstkoflw
+        Crystal::System.print_error "Stack overflow (e.g., infinite or very deep recursion)\n"
+        print_backtrace(exception_info)
+        LibC._exit(1)
+      else
+        LibC::EXCEPTION_CONTINUE_SEARCH
+      end
+    end)
+
+    # ensure that even in the case of stack overflow there is enough reserved
+    # stack space for recovery
+    stack_size = LibC::DWORD.new!(0x10000)
+    LibC.SetThreadStackGuarantee(pointerof(stack_size))
+  end
+
+  {% if flag?(:interpreted) %} @[Primitive(:interpreter_call_stack_unwind)] {% end %}
+  protected def self.unwind : Array(Void*)
+    # TODO: use stack if possible (must be 16-byte aligned)
+    context = Pointer(LibC::CONTEXT).malloc(1)
+    context.value.contextFlags = LibC::CONTEXT_FULL
+    LibC.RtlCaptureContext(context)
+
+    stack = [] of Void*
+    each_frame(context) do |frame|
+      (frame.count + 1).times do
+        stack << frame.ip
+      end
+    end
+    stack
+  end
+
+  private def self.each_frame(context, &)
+    # unlike DWARF, this is required on Windows to even be able to produce
+    # correct stack traces, so we do it here but not in `libunwind.cr`
     load_debug_info
 
     machine_type = {% if flag?(:x86_64) %}
@@ -43,11 +86,6 @@ struct Exception::CallStack
                      {% raise "architecture not supported" %}
                    {% end %}
 
-    # TODO: use stack if possible (must be 16-byte aligned)
-    context = Pointer(LibC::CONTEXT).malloc(1)
-    context.value.contextFlags = LibC::CONTEXT_FULL
-    LibC.RtlCaptureContext(context)
-
     stack_frame = LibC::STACKFRAME64.new
     stack_frame.addrPC.mode = LibC::ADDRESS_MODE::AddrModeFlat
     stack_frame.addrFrame.mode = LibC::ADDRESS_MODE::AddrModeFlat
@@ -57,13 +95,15 @@ struct Exception::CallStack
     stack_frame.addrFrame.offset = context.value.rbp
     stack_frame.addrStack.offset = context.value.rsp
 
-    stack = [] of Void*
+    last_frame = nil
+    cur_proc = LibC.GetCurrentProcess
+    cur_thread = LibC.GetCurrentThread
 
     while true
       ret = LibC.StackWalk64(
         machine_type,
-        LibC.GetCurrentProcess,
-        LibC.GetCurrentThread,
+        cur_proc,
+        cur_thread,
         pointerof(stack_frame),
         context,
         nil,
@@ -72,10 +112,70 @@ struct Exception::CallStack
         nil
       )
       break if ret == 0
-      stack << Pointer(Void).new(stack_frame.addrPC.offset)
+
+      ip = Pointer(Void).new(stack_frame.addrPC.offset)
+      if last_frame
+        if ip != last_frame.ip
+          yield last_frame
+          last_frame = RepeatedFrame.new(ip)
+        else
+          last_frame.incr
+        end
+      else
+        last_frame = RepeatedFrame.new(ip)
+      end
     end
 
-    stack
+    yield last_frame if last_frame
+  end
+
+  struct RepeatedFrame
+    getter ip : Void*, count : Int32
+
+    def initialize(@ip : Void*)
+      @count = 0
+    end
+
+    def incr
+      @count += 1
+    end
+  end
+
+  private record StackContext, context : LibC::CONTEXT*, thread : LibC::HANDLE
+
+  def self.print_backtrace(exception_info) : Nil
+    each_frame(exception_info.value.contextRecord) do |frame|
+      print_frame(frame)
+    end
+  end
+
+  private def self.print_frame(repeated_frame)
+    if name = decode_function_name(repeated_frame.ip.address)
+      file, line, _ = decode_line_number(repeated_frame.ip.address)
+      if file != "??" && line != 0
+        if repeated_frame.count == 0
+          Crystal::System.print_error "[0x%llx] %s at %s:%ld\n", repeated_frame.ip, name, file, line
+        else
+          Crystal::System.print_error "[0x%llx] %s at %s:%ld (%ld times)\n", repeated_frame.ip, name, file, line, repeated_frame.count + 1
+        end
+        return
+      end
+    end
+
+    if frame = decode_frame(repeated_frame.ip)
+      offset, sname, fname = frame
+      if repeated_frame.count == 0
+        Crystal::System.print_error "[0x%llx] %s +%lld in %s\n", repeated_frame.ip, sname, offset, fname
+      else
+        Crystal::System.print_error "[0x%llx] %s +%lld in %s (%ld times)\n", repeated_frame.ip, sname, offset, fname, repeated_frame.count + 1
+      end
+    else
+      if repeated_frame.count == 0
+        Crystal::System.print_error "[0x%llx] ???\n", repeated_frame.ip
+      else
+        Crystal::System.print_error "[0x%llx] ??? (%ld times)\n", repeated_frame.ip, repeated_frame.count + 1
+      end
+    end
   end
 
   protected def self.decode_line_number(pc)
@@ -86,19 +186,15 @@ struct Exception::CallStack
 
     if LibC.SymGetLineFromAddrW64(LibC.GetCurrentProcess, pc, out displacement, pointerof(line_info)) != 0
       file_name = String.from_utf16(line_info.fileName)[0]
-      line_number = line_info.lineNumber
+      line_number = line_info.lineNumber.to_i32
     else
       line_number = 0
     end
 
     unless file_name
-      module_info = Pointer(LibC::IMAGEHLP_MODULEW64).malloc(1)
-      module_info.value.sizeOfStruct = sizeof(LibC::IMAGEHLP_MODULEW64)
-
-      if LibC.SymGetModuleInfoW64(LibC.GetCurrentProcess, pc, module_info) != 0
-        mod_displacement = pc - LibC.SymGetModuleBase64(LibC.GetCurrentProcess, pc)
-        image_name = String.from_utf16(module_info.value.loadedImageName.to_unsafe)[0]
-        file_name = "#{image_name} +#{mod_displacement}"
+      if m_info = sym_get_module_info(pc)
+        offset, image_name = m_info
+        file_name = "#{image_name} +#{offset}"
       else
         file_name = "??"
       end
@@ -108,6 +204,37 @@ struct Exception::CallStack
   end
 
   protected def self.decode_function_name(pc)
+    if sym = sym_from_addr(pc)
+      _, sname = sym
+      sname
+    end
+  end
+
+  protected def self.decode_frame(ip)
+    pc = decode_address(ip)
+    if sym = sym_from_addr(pc)
+      if m_info = sym_get_module_info(pc)
+        offset, sname = sym
+        _, fname = m_info
+        {offset, sname, fname}
+      end
+    end
+  end
+
+  private def self.sym_get_module_info(pc)
+    load_debug_info
+
+    module_info = Pointer(LibC::IMAGEHLP_MODULEW64).malloc(1)
+    module_info.value.sizeOfStruct = sizeof(LibC::IMAGEHLP_MODULEW64)
+
+    if LibC.SymGetModuleInfoW64(LibC.GetCurrentProcess, pc, module_info) != 0
+      mod_displacement = pc - LibC.SymGetModuleBase64(LibC.GetCurrentProcess, pc)
+      image_name = String.from_utf16(module_info.value.loadedImageName.to_unsafe)[0]
+      {mod_displacement, image_name}
+    end
+  end
+
+  private def self.sym_from_addr(pc)
     load_debug_info
 
     symbol_size = sizeof(LibC::SYMBOL_INFOW) + (LibC::MAX_SYM_NAME - 1) * sizeof(LibC::WCHAR)
@@ -117,13 +244,11 @@ struct Exception::CallStack
 
     sym_displacement = LibC::DWORD64.zero
     if LibC.SymFromAddrW(LibC.GetCurrentProcess, pc, pointerof(sym_displacement), symbol) != 0
-      String.from_utf16(symbol.value.name.to_unsafe.to_slice(symbol.value.nameLen))
+      symbol_str = String.from_utf16(symbol.value.name.to_unsafe.to_slice(symbol.value.nameLen))
+      {sym_displacement, symbol_str}
     end
   end
 
-  protected def self.decode_frame(pc)
-  end
-
   protected def self.decode_address(ip)
     ip.address
   end

src/kernel.cr

@@ -534,7 +534,6 @@ end
     end
 
     Signal.setup_default_handlers
-    Signal.setup_segfault_handler
   {% end %}
 
   # load debug info on start up of the program is executed with CRYSTAL_LOAD_DEBUG_INFO=1
@@ -544,6 +543,7 @@ end
   # - CRYSTAL_LOAD_DEBUG_INFO=1 will load debug info on startup
   # - Other values will load debug info on demand: when the backtrace of the first exception is generated
   Exception::CallStack.load_debug_info if ENV["CRYSTAL_LOAD_DEBUG_INFO"]? == "1"
+  Exception::CallStack.setup_crash_handler
 
   {% if flag?(:preview_mt) %}
     Crystal::Scheduler.init_workers

src/lib_c/x86_64-windows-msvc/c/errhandlingapi.cr

@@ -1,6 +1,14 @@
 require "c/int_safe"
 
 lib LibC
+  EXCEPTION_CONTINUE_SEARCH = LONG.new!(0)
+
+  EXCEPTION_ACCESS_VIOLATION = 0xC0000005_u32
+  EXCEPTION_STACK_OVERFLOW   = 0xC00000FD_u32
+
+  alias PVECTORED_EXCEPTION_HANDLER = EXCEPTION_POINTERS* -> LONG
+
   fun GetLastError : DWORD
   fun SetLastError(dwErrCode : DWORD)
+  fun AddVectoredExceptionHandler(first : DWORD, handler : PVECTORED_EXCEPTION_HANDLER) : Void*
 end

src/lib_c/x86_64-windows-msvc/c/malloc.cr

@@ -0,0 +1,3 @@
+lib LibC
+  fun _resetstkoflw : Int
+end

src/lib_c/x86_64-windows-msvc/c/processthreadsapi.cr

@@ -43,6 +43,7 @@ lib LibC
                      bInheritHandles : BOOL, dwCreationFlags : DWORD,
                      lpEnvironment : Void*, lpCurrentDirectory : LPWSTR,
                      lpStartupInfo : STARTUPINFOW*, lpProcessInformation : PROCESS_INFORMATION*) : BOOL
+  fun SetThreadStackGuarantee(stackSizeInBytes : DWORD*) : BOOL
   fun GetProcessTimes(hProcess : HANDLE, lpCreationTime : FILETIME*, lpExitTime : FILETIME*,
                       lpKernelTime : FILETIME*, lpUserTime : FILETIME*) : BOOL
 

src/lib_c/x86_64-windows-msvc/c/winnt.cr

@@ -153,4 +153,19 @@ lib LibC
   {% end %}
 
   fun RtlCaptureContext(contextRecord : CONTEXT*)
+
+  struct EXCEPTION_RECORD64
+    exceptionCode : DWORD
+    exceptionFlags : DWORD
+    exceptionRecord : DWORD64
+    exceptionAddress : DWORD64
+    numberParameters : DWORD
+    __unusedAlignment : DWORD
+    exceptionInformation : DWORD64[15]
+  end
+
+  struct EXCEPTION_POINTERS
+    exceptionRecord : EXCEPTION_RECORD64*
+    contextRecord : CONTEXT*
+  end
 end