#!/usr/bin/env python3 import argparse import binascii import random import struct import sys import tempfile import time import os import re import itertools import targets import testlib from testlib import assertEqual, assertNotEqual from testlib import assertIn, assertNotIn from testlib import assertGreater, assertRegex, assertLess from testlib import GdbTest, GdbSingleHartTest, TestFailed from testlib import TestNotApplicable, CompileError from testlib import UnknownThread from testlib import CouldNotReadRegisters, CommandException from testlib import ThreadTerminated MSTATUS_UIE = 0x00000001 MSTATUS_SIE = 0x00000002 MSTATUS_HIE = 0x00000004 MSTATUS_MIE = 0x00000008 MSTATUS_UPIE = 0x00000010 MSTATUS_SPIE = 0x00000020 MSTATUS_HPIE = 0x00000040 MSTATUS_MPIE = 0x00000080 MSTATUS_SPP = 0x00000100 MSTATUS_HPP = 0x00000600 MSTATUS_MPP = 0x00001800 MSTATUS_FS = 0x00006000 MSTATUS_XS = 0x00018000 MSTATUS_MPRV = 0x00020000 MSTATUS_PUM = 0x00040000 MSTATUS_MXR = 0x00080000 MSTATUS_VM = 0x1F000000 MSTATUS32_SD = 0x80000000 MSTATUS64_SD = 0x8000000000000000 # pylint: disable=abstract-method def ihex_line(address, record_type, data): assert len(data) < 128 line = f":{len(data):02X}{address:04X}{record_type:02X}" check = len(data) check += address % 256 check += address >> 8 check += record_type for char in data: value = ord(char) check += value line += f"{value:02X}" line += f"{(256 - check) % 256:02X}\n" return line def srec_parse(line): assert line.startswith(b'S') typ = line[:2] count = int(line[2:4], 16) data = "" if typ == b'S0': # header return 0, 0, 0 elif typ == b'S3': # data with 32-bit address # Any higher bits were chopped off. address = int(line[4:12], 16) for i in range(6, count+1): data += f"{int(line[2 * i:2 * i + 2], 16):c}" # Ignore the checksum. return 3, address, data elif typ == b'S7': # ignore execution start field return 7, 0, 0 else: raise TestFailed(f"Unsupported SREC type {typ!r}.") def readable_binary_string(s): return "".join(f"{ord(c):02x}" for c in s) class InfoTest(GdbTest): def test(self): output = self.gdb.command("monitor riscv info") info = {} for line in output.splitlines(): if re.search(r"Found \d+ triggers", line): continue if re.search(r"Disabling abstract command writes to CSRs.", line): continue if re.search( r"keep_alive.. was not invoked in the \d+ ms timelimit.", line): continue k, v = line.strip().split() info[k] = v assertEqual(int(info.get("hart.xlen")), self.hart.xlen) class SimpleRegisterTest(GdbTest): def check_reg(self, name, alias): a = random.randrange(1< last_pc and pc - last_pc <= 4: advances += 1 else: jumps += 1 last_pc = pc end = time.time() print(f"{(end - start) / count:.2f} seconds/step") # Some basic sanity that we're not running between breakpoints or # something. assertGreater(jumps, 1) assertGreater(advances, 5) class DebugExit(DebugTest): def test(self): self.exit() class DebugSymbols(DebugTest): def test(self): bp = self.gdb.b("main") output = self.gdb.c() assertIn(", main ", output) self.gdb.command(f"delete {bp}") bp = self.gdb.b("rot13") output = self.gdb.c() assertIn(", rot13 ", output) self.gdb.command(f"delete {bp}") class DebugBreakpoint(DebugTest): def test(self): self.gdb.b("rot13") # The breakpoint should be hit exactly 2 times. for _ in range(2): output = self.gdb.c() self.gdb.p("$pc") assertIn("Breakpoint ", output) assertIn("rot13 ", output) self.exit() class Hwbp1(DebugTest): def early_applicable(self): return self.hart.instruction_hardware_breakpoint_count > 0 def test(self): if not self.hart.honors_tdata1_hmode: # Run to main before setting the breakpoint, because startup code # will otherwise clear the trigger that we set. self.gdb.b("main") self.gdb.c() self.gdb.command("delete") self.gdb.hbreak("rot13") # The breakpoint should be hit exactly 2 times. for _ in range(2): output = self.gdb.c() self.gdb.p("$pc") assertRegex(output, r"[bB]reakpoint") assertIn("rot13 ", output) self.gdb.b("_exit") self.exit() def MCONTROL_TYPE(xlen): return 0xf<<((xlen)-4) def MCONTROL_DMODE(xlen): return 1<<((xlen)-5) def MCONTROL_MASKMAX(xlen): return 0x3<<((xlen)-11) MCONTROL_SELECT = 1<<19 MCONTROL_TIMING = 1<<18 MCONTROL_ACTION = 0x3f<<12 MCONTROL_CHAIN = 1<<11 MCONTROL_MATCH = 0xf<<7 MCONTROL_M = 1<<6 MCONTROL_H = 1<<5 MCONTROL_S = 1<<4 MCONTROL_U = 1<<3 MCONTROL_EXECUTE = 1<<2 MCONTROL_STORE = 1<<1 MCONTROL_LOAD = 1<<0 MCONTROL_TYPE_NONE = 0 MCONTROL_TYPE_MATCH = 2 MCONTROL_ACTION_DEBUG_EXCEPTION = 0 MCONTROL_ACTION_DEBUG_MODE = 1 MCONTROL_ACTION_TRACE_START = 2 MCONTROL_ACTION_TRACE_STOP = 3 MCONTROL_ACTION_TRACE_EMIT = 4 MCONTROL_MATCH_EQUAL = 0 MCONTROL_MATCH_NAPOT = 1 MCONTROL_MATCH_GE = 2 MCONTROL_MATCH_LT = 3 MCONTROL_MATCH_MASK_LOW = 4 MCONTROL_MATCH_MASK_HIGH = 5 def set_field(reg, mask, val): return ((reg) & ~(mask)) | (((val) * ((mask) & ~((mask) << 1))) & (mask)) class HwbpManual(DebugTest): """Make sure OpenOCD behaves "normal" when the user sets a trigger by writing the trigger registers themselves directly.""" def early_applicable(self): return self.target.support_manual_hwbp and \ self.hart.instruction_hardware_breakpoint_count >= 1 def test(self): if not self.hart.honors_tdata1_hmode: # Run to main before setting the breakpoint, because startup code # will otherwise clear the trigger that we set. self.gdb.b("main") self.gdb.c() self.gdb.command("delete") #self.gdb.hbreak("rot13") tdata1 = MCONTROL_DMODE(self.hart.xlen) tdata1 = set_field(tdata1, MCONTROL_ACTION, MCONTROL_ACTION_DEBUG_MODE) tdata1 = set_field(tdata1, MCONTROL_MATCH, MCONTROL_MATCH_EQUAL) tdata1 |= MCONTROL_M | MCONTROL_S | MCONTROL_U | MCONTROL_EXECUTE tselect = 0 while True: self.gdb.p(f"$tselect={tselect}") value = self.gdb.p("$tselect") if value != tselect: raise TestNotApplicable self.gdb.p(f"$tdata1=0x{tdata1:x}") value = self.gdb.p("$tselect") if value == tdata1: break self.gdb.p("$tdata1=0") tselect += 1 self.gdb.p("$tdata2=&rot13") # The breakpoint should be hit exactly 2 times. for _ in range(2): output = self.gdb.c(ops=2) self.gdb.p("$pc") assertRegex(output, r"[bB]reakpoint") assertIn("rot13 ", output) self.gdb.p("$tdata2=&crc32a") self.gdb.c() before = self.gdb.p("$pc") assertEqual(before, self.gdb.p("&crc32a")) self.gdb.stepi() after = self.gdb.p("$pc") assertNotEqual(before, after) self.gdb.b("_exit") self.exit() class Hwbp2(DebugTest): def early_applicable(self): return self.hart.instruction_hardware_breakpoint_count >= 2 def test(self): self.gdb.command("delete") self.gdb.hbreak("main") self.gdb.hbreak("rot13") # We should hit 3 breakpoints. for expected in ("main", "rot13", "rot13"): output = self.gdb.c() self.gdb.p("$pc") assertRegex(output, r"[bB]reakpoint") assertIn(f"{expected} ", output) self.gdb.command("delete") self.gdb.b("_exit") self.exit() class TooManyHwbp(DebugTest): def test(self): for i in range(30): self.gdb.hbreak(f"*rot13 + {i * 4}") output = self.gdb.c(checkOutput=False) assertIn("Cannot insert hardware breakpoint", output) # There used to be a bug where this would fail if done twice in a row. output = self.gdb.c(checkOutput=False) assertIn("Cannot insert hardware breakpoint", output) # Clean up, otherwise the hardware breakpoints stay set and future # tests may fail. self.gdb.command("delete") self.gdb.b("_exit") self.exit() class Registers(DebugTest): def test(self): # Get to a point in the code where some registers have actually been # used. self.gdb.b("rot13") self.gdb.c() self.gdb.c() # Try both forms to test gdb. for cmd in ("info all-registers", "info registers all"): output = self.gdb.command(cmd, ops=20) for reg in ('zero', 'ra', 'sp', 'gp', 'tp'): assertIn(reg, output) for line in output.splitlines(): assertRegex(line, r"^\S") #TODO # mcpuid is one of the few registers that should have the high bit set # (for rv64). # Leave this commented out until gdb and spike agree on the encoding of # mcpuid (which is going to be renamed to misa in any case). #assertRegex(output, ".*mcpuid *0x80") #TODO: # The instret register should always be changing. #last_instret = None #for _ in range(5): # instret = self.gdb.p("$instret") # assertNotEqual(instret, last_instret) # last_instret = instret # self.gdb.stepi() self.exit() class UserInterrupt(DebugTest): def test(self): """Sending gdb ^C while the program is running should cause it to halt.""" self.gdb.b("main:start") self.gdb.c() self.gdb.p("i=123") self.gdb.c(wait=False) time.sleep(2) output = self.gdb.interrupt() assert "main" in output assertGreater(self.gdb.p("j"), 10) self.gdb.p("i=0") self.exit() class GdbServerError(Exception): pass class MemorySampleTest(DebugTest): def early_applicable(self): return self.target.support_memory_sampling def setup(self): DebugTest.setup(self) self.gdb.b("main:start") self.gdb.c() self.gdb.p("i=123") @staticmethod def check_incrementing_samples(raw_samples, check_addr, tolerance=0x200000): first_timestamp = None end = None total_samples = 0 previous_value = None for line in raw_samples.splitlines(): m = re.match(r"^timestamp \w+: (\d+)", line) if m: timestamp = int(m.group(1)) if not first_timestamp: first_timestamp = timestamp else: end = (timestamp, total_samples) else: assertRegex(line, r"^0x[0-f]+: 0x[0-f]+$") address, value = line.split(': ') address = int(address, 16) if address == check_addr: value = int(value, 16) if not previous_value is None: # TODO: what if the counter wraps? assertGreater(value, previous_value) assertLess(value, previous_value + tolerance) previous_value = value total_samples += 1 if end and total_samples > 0: samples_per_second = 1000 * end[1] / (end[0] - first_timestamp) print(f"{samples_per_second} samples/second") else: raise GdbServerError("No samples collected.") @staticmethod def check_samples_equal(raw_samples, check_addr, check_value): total_samples = 0 for line in raw_samples.splitlines(): if not line.startswith("timestamp "): address, value = line.split(': ') address = int(address, 16) if address == check_addr: value = int(value, 16) assertEqual(value, check_value) total_samples += 1 assertGreater(total_samples, 0) def collect_samples(self): self.gdb.c(wait=False) time.sleep(5) output = self.gdb.interrupt() assert "main" in output return self.gdb.command("monitor riscv dump_sample_buf", ops=5) class MemorySampleSingle(MemorySampleTest): def test(self): addr = self.gdb.p("&j") sizeof_j = self.gdb.p("sizeof(j)") self.gdb.command(f"monitor riscv memory_sample 0 0x{addr:x} {sizeof_j}") raw_samples = self.collect_samples() self.check_incrementing_samples(raw_samples, addr, tolerance=0x500000) # Buffer should have been emptied by dumping. raw_samples = self.gdb.command("monitor riscv dump_sample_buf", ops=5) assertEqual(len(raw_samples), 0) class MemorySampleMixed(MemorySampleTest): def test(self): addr = {} test_vars = ["j", "i32"] if self.hart.xlen >= 64: test_vars.append("i64") for i, name in enumerate(test_vars): addr[name] = self.gdb.p(f"&{name}") sizeof = self.gdb.p(f"sizeof({name})") self.gdb.command(f"monitor riscv memory_sample {i} " f"0x{addr[name]:x} {sizeof}") raw_samples = self.collect_samples() self.check_incrementing_samples(raw_samples, addr["j"], tolerance=0x500000) self.check_samples_equal(raw_samples, addr["i32"], 0xdeadbeef) if self.hart.xlen >= 64: self.check_samples_equal(raw_samples, addr["i64"], 0x1122334455667788) class RepeatReadTest(DebugTest): def early_applicable(self): return self.target.supports_clint_mtime warning_re = re.compile(r"\[(?P[^\]]+)\] Re-reading memory " r"from addresses 0x(?P[\da-f]+) and 0x(?P=addr)\.") def test(self): self.gdb.b("main:start") self.gdb.c() mtime_addr = self.target.clint_addr + 0xbff8 count = 1024 output = self.gdb.command( f"monitor riscv repeat_read {count} 0x{mtime_addr:x} 4") values = [] def is_valid_warning(line): match = self.warning_re.match(line) if match is None: return False assertEqual(int(match["addr"], 16), mtime_addr, "The repeat read is reading from the wrong address") return True for line in itertools.dropwhile(is_valid_warning, output.splitlines()): # This `if` is to be removed after # https://github.com/riscv/riscv-openocd/pull/871 is merged. if line.startswith("Batch memory"): continue for v in line.split(): values.append(int(v, 16)) assertEqual(len(values), count) # mtime should only ever increase, unless it wraps slop = 0x100000 for i in range(1, len(values)): if values[i] < values[i-1]: # wrapped assertLess(values[i], slop) else: assertGreater(values[i], values[i-1]) assertLess(values[i], values[i-1] + slop) output = self.gdb.command( f"monitor riscv repeat_read 0 0x{mtime_addr:x} 4") assertEqual(output, "") class Semihosting(GdbSingleHartTest): # Include malloc so that gdb can assign a string. compile_args = ("programs/semihosting.c", "programs/tiny-malloc.c", "-DDEFINE_MALLOC", "-DDEFINE_FREE") def early_applicable(self): return self.target.test_semihosting def setup(self): self.gdb.load() self.parkOtherHarts() self.gdb.b("_exit") def test(self): with tempfile.NamedTemporaryFile(suffix=".data") as temp: self.gdb.b("main:begin") self.gdb.c() self.gdb.p(f'filename="{temp.name}"', ops=3) self.exit() with open(temp.name, "r", encoding='utf-8') as fd: contents = fd.readlines() assertIn("Hello, world!\n", contents) # stdout should end up in the OpenOCD log with open(self.server.logname, "r", encoding='utf-8') as fd: log = fd.read() assertIn("Do re mi fa so la ti do!", log) class SemihostingFileio(Semihosting): def early_applicable(self): # Semihosting file i/o doesn't work right when there are multiple harts # in SMP mode, and the semihosting call comes from a hart other than the # first one. # The problem is that semihosting_common_fileio_info() is called only # for the first target in an SMP list. Either the caller needs to be # made aware of SMP targets, or that function needs to walk the list # itself. (Or maybe we need to make a separate function just for RISC-V # that does that.) return len(self.target.harts) == 1 def setup(self): self.gdb.command("monitor foreach t [target names] { " "targets $t; arm semihosting_fileio enable }") super().setup() def test(self): with tempfile.NamedTemporaryFile(suffix=".data") as temp: self.gdb.b("main:begin") self.gdb.c() self.gdb.p(f'filename="{temp.name}"', ops=3) output = self.exit() # stdout should end up in gdb's CLI assertIn("Do re mi fa so la ti do!", output) with open(temp.name, "r", encoding='utf-8') as fd: contents = fd.readlines() assertIn("Hello, world!\n", contents) class InterruptTest(GdbSingleHartTest): compile_args = ("programs/interrupt.c",) def early_applicable(self): return self.target.supports_clint_mtime def setup(self): self.gdb.load() def test(self): self.gdb.b("main") output = self.gdb.c() assertIn(" main ", output) self.gdb.b("trap_entry") output = self.gdb.c() assertIn(" trap_entry ", output) assertEqual(self.gdb.p("$mip") & 0x80, 0x80) assertEqual(self.gdb.p("interrupt_count"), 0) # You'd expect local to still be 0, but it looks like spike doesn't # jump to the interrupt handler immediately after the write to # mtimecmp. assertLess(self.gdb.p("local"), 1000) self.gdb.command("delete breakpoints") for _ in range(10): self.gdb.c(wait=False) time.sleep(2) self.gdb.interrupt() interrupt_count = self.gdb.p("interrupt_count") local = self.gdb.p("local") if interrupt_count > 1000 and \ local > 1000: self.disable_timer() return self.disable_timer() assertGreater(interrupt_count, 1000) assertGreater(local, 1000) def postMortem(self): GdbSingleHartTest.postMortem(self) self.gdb.p(f"*((long long*) 0x{self.target.clint_addr + 0xbff8:x})") self.gdb.p(f"*((long long*) 0x{self.target.clint_addr + 0x4000:x})") self.gdb.p("interrupt_count") self.gdb.p("local") class MulticoreRegTest(GdbTest): compile_args = ("programs/infinite_loop.S", "-DMULTICORE") def early_applicable(self): return len(self.target.harts) > 1 def setup(self): self.gdb.load() for hart in self.target.harts: self.gdb.select_hart(hart) self.gdb.p("$pc=_start") def test(self): # We use time instead of breakpoints, because otherwise we can't # guarantee that every hart runs all the way through the loop. (The # problem is that we can't guarantee resuming at the same time, so the # first hart that is resumed will hit a breakpoint at the end of the # loop before another hart has executed the whole loop.) # Run through the whole loop. self.gdb.c_all(wait=False) time.sleep(1) self.gdb.interrupt_all() hart_ids = set() for hart in self.target.harts: self.gdb.select_hart(hart) assertIn("main_end", self.gdb.where()) # Check register values. x1 = self.gdb.p("$x1") hart_id = self.gdb.p("$mhartid") assertEqual(x1, hart_id << 8) assertNotIn((hart.system, hart_id), hart_ids) hart_ids.add((hart.system, hart_id)) for n in range(2, 32): value = self.gdb.p(f"$x{n}") assertEqual(value, (hart_id << 8) + n - 1) # Confirmed that we read different register values for different harts. # Write a new value to x1, and run through the add sequence again. for hart in self.target.harts: self.gdb.select_hart(hart) self.gdb.p(f"$x1=0x{hart.index * 4096:x}") self.gdb.p("$pc=main_post_csrr") # Run through the whole loop. self.gdb.c_all(wait=False) time.sleep(1) self.gdb.interrupt_all() for hart in self.target.harts: self.gdb.select_hart(hart) assertIn("main_end", self.gdb.where()) # Check register values. for n in range(1, 32): value = self.gdb.p(f"$x{n}") assertEqual(value, hart.index * 0x1000 + n - 1) #class MulticoreRunHaltStepiTest(GdbTest): # compile_args = ("programs/multicore.c", "-DMULTICORE") # # def early_applicable(self): # return len(self.target.harts) > 1 # # def setup(self): # self.gdb.load() # for hart in self.target.harts: # self.gdb.select_hart(hart) # self.gdb.p("$mhartid") # self.gdb.p("$pc=_start") # # def test(self): # previous_hart_count = [0 for h in self.target.harts] # previous_interrupt_count = [0 for h in self.target.harts] # # Check 10 times # for i in range(10): # # 3 attempts for each time we want the check to pass # for attempt in range(3): # self.gdb.global_command("echo round %d attempt %d\\n" % (i, # attempt)) # self.gdb.c_all(wait=False) # time.sleep(2) # self.gdb.interrupt_all() # hart_count = self.gdb.p("hart_count") # interrupt_count = self.gdb.p("interrupt_count") # ok = True # for i, h in enumerate(self.target.harts): # if hart_count[i] <= previous_hart_count[i]: # ok = False # break # if interrupt_count[i] <= previous_interrupt_count[i]: # ok = False # break # self.gdb.p("$mie") # self.gdb.p("$mip") # self.gdb.p("$mstatus") # self.gdb.p("$priv") # self.gdb.p("buf", fmt="") # self.gdb.select_hart(h) # pc = self.gdb.p("$pc") # self.gdb.stepi() # stepped_pc = self.gdb.p("$pc") # assertNotEqual(pc, stepped_pc) # previous_hart_count = hart_count # previous_interrupt_count = interrupt_count # if ok: # break # else: # assert False, \ # "hart count or interrupt didn't increment as expected" class MulticoreRunAllHaltOne(GdbTest): compile_args = ("programs/multicore.c", "-DMULTICORE") def early_applicable(self): return len(self.target.harts) > 1 def setup(self): self.gdb.select_hart(self.target.harts[0]) self.gdb.load() for hart in self.target.harts: self.gdb.select_hart(hart) self.gdb.p("$pc=_start") def test(self): if not self.gdb.one_hart_per_gdb(): raise TestNotApplicable # Run harts in reverse order for h in reversed(self.target.harts): self.gdb.select_hart(h) self.gdb.c(wait=False) self.gdb.interrupt() # Give OpenOCD time to call poll() on both harts, which is what causes # the bug. time.sleep(1) self.gdb.p("buf", fmt="") self.disable_timer(interrupt=True) class MulticoreRtosSwitchActiveHartTest(GdbTest): compile_args = ("programs/multicore.c", "-DMULTICORE") def early_applicable(self): return len(self.target.harts) > 1 def setup(self): self.gdb.select_hart(self.target.harts[0]) self.gdb.load() for hart in self.target.harts: self.gdb.select_hart(hart) self.gdb.p("$pc=_start") def test(self): if self.gdb.one_hart_per_gdb(): raise TestNotApplicable # Set breakpoint near '_start' label to increase the chances of a # situation when all harts hit breakpoint immediately and # simultaneously. self.gdb.b("set_trap_handler") # Check that all harts hit breakpoint one by one. for _ in range(len(self.target.harts)): output = self.gdb.c() assertIn("hit Breakpoint", output) assertIn("set_trap_handler", output) assertNotIn("received signal SIGTRAP", output) self.disable_timer() class SmpSimultaneousRunHalt(GdbTest): compile_args = ("programs/run_halt_timing.S", "-DMULTICORE") def early_applicable(self): return len(self.target.harts) > 1 and self.target.support_hasel def setup(self): self.gdb.select_hart(self.target.harts[0]) self.gdb.load() for hart in self.target.harts: self.gdb.select_hart(hart) self.gdb.p("$pc=_start") def test(self): if self.gdb.one_hart_per_gdb() or not self.server.smp(): raise TestNotApplicable old_mtime = set() for _ in range(5): self.gdb.c_all(wait=False) time.sleep(2) self.gdb.interrupt_all() mtime_value = [] counter = [] for hart in self.target.harts: self.gdb.select_hart(hart) mv = self.gdb.p("$s2") assertNotIn(mv, old_mtime, "mtime doesn't appear to be changing at all") mtime_value.append(mv) c = self.gdb.p("$s0") assertNotEqual(c, 0, "counter didn't increment; code didn't run?") counter.append(c) # Reset the counter for the next round. self.gdb.p("$s0=0") old_mtime.update(mtime_value) mtime_spread = max(mtime_value) - min(mtime_value) print("mtime_spread:", mtime_spread) counter_spread = max(counter) - min(counter) print("counter_spread:", counter_spread) assertLess(mtime_spread, 101 * (len(self.target.harts) - 1), "Harts don't halt around the same time.") # spike executes normal code 5000 instructions at a time, so we # expect 5k instructions to be executed on one hart before the # other gets to go. Our loop (unoptimized) is quite a few # instructions, but allow for 5k anyway. assertLess(counter_spread, 5001 * (len(self.target.harts) - 1), "Harts don't resume around the same time.") class StepTest(GdbSingleHartTest): compile_args = ("programs/step.S", ) def setup(self): self.gdb.load() self.gdb.b("main") self.gdb.c() def test(self): main_address = self.gdb.p("$pc") if self.hart.extensionSupported("c"): sequence = (4, 8, 0xc, 0xe, 0x14, 0x18, 0x22, 0x1c, 0x24, 0x24) else: sequence = (4, 8, 0xc, 0x10, 0x18, 0x1c, 0x28, 0x20, 0x2c, 0x2c) for expected in sequence: self.gdb.stepi() pc = self.gdb.p("$pc") assertEqual(f"{pc - main_address:x}", f"{expected:x}") class JumpHbreak(GdbSingleHartTest): """'jump' resumes execution at location. Execution stops again immediately if there is a breakpoint there. That second line can be trouble.""" compile_args = ("programs/trigger.S", ) def early_applicable(self): return self.hart.instruction_hardware_breakpoint_count >= 1 def setup(self): self.gdb.load() self.gdb.hbreak("main") self.gdb.c() self.gdb.command("delete 1") def test(self): self.gdb.b("read_loop") self.gdb.command("hbreak just_before_read_loop") output = self.gdb.command("jump just_before_read_loop") assertRegex(output, r"Breakpoint \d, just_before_read_loop ") output = self.gdb.c() assertRegex(output, r"Breakpoint \d, read_loop ") class TriggerTest(GdbSingleHartTest): compile_args = ("programs/trigger.S", ) def setup(self): self.gdb.load() self.gdb.b("main") self.gdb.c() self.gdb.command("delete") class TriggerExecuteInstant(TriggerTest): """Test an execute breakpoint on the first instruction executed out of debug mode.""" def test(self): main_address = self.gdb.p("$pc") self.gdb.command(f"hbreak *0x{main_address + 4:x}") self.gdb.c() assertEqual(self.gdb.p("$pc"), main_address+4) # FIXME: Triggers aren't quite working yet #class TriggerLoadAddress(TriggerTest): # def test(self): # self.gdb.command("rwatch *((&data)+1)") # output = self.gdb.c() # assertIn("read_loop", output) # assertEqual(self.gdb.p("$a0"), # self.gdb.p("(&data)+1")) # self.exit() class TriggerLoadAddressInstant(TriggerTest): """Test a load address breakpoint on the first instruction executed out of debug mode.""" def test(self): self.gdb.command("b just_before_read_loop") self.gdb.c() read_loop = self.gdb.p("&read_loop") read_again = self.gdb.p("&read_again") data = self.gdb.p("&data") self.gdb.command(f"rwatch *0x{data:x}") self.gdb.c() # Accept hitting the breakpoint before or after the load instruction. assertIn(self.gdb.p("$pc"), [read_loop, read_loop + 4]) assertEqual(self.gdb.p("$a0"), self.gdb.p("&data")) self.gdb.c() assertIn(self.gdb.p("$pc"), [read_again, read_again + 4]) assertEqual(self.gdb.p("$a0"), self.gdb.p("&data")) # FIXME: Triggers aren't quite working yet #class TriggerStoreAddress(TriggerTest): # def test(self): # self.gdb.command("watch *((&data)+3)") # output = self.gdb.c() # assertIn("write_loop", output) # assertEqual(self.gdb.p("$a0"), # self.gdb.p("(&data)+3")) # self.exit() class TriggerStoreAddressInstant(TriggerTest): def test(self): """Test a store address breakpoint on the first instruction executed out of debug mode.""" self.gdb.command("b just_before_write_loop") self.gdb.c() write_loop = self.gdb.p("&write_loop") data = self.gdb.p("&data") self.gdb.command(f"watch *0x{data:x}") self.gdb.c() # Accept hitting the breakpoint before or after the store instruction. assertIn(self.gdb.p("$pc"), [write_loop, write_loop + 4]) assertEqual(self.gdb.p("$a0"), self.gdb.p("&data")) class TriggerDmode(TriggerTest): def early_applicable(self): return self.hart.honors_tdata1_hmode and \ self.hart.instruction_hardware_breakpoint_count > 0 def check_triggers(self, tdata1_lsbs, tdata2): dmode = 1 << (self.hart.xlen-5) triggers = [] if self.hart.xlen == 32: xlen_type = 'int' elif self.hart.xlen == 64: xlen_type = 'long long' else: raise NotImplementedError dmode_count = 0 i = 0 for i in range(16): tdata1 = self.gdb.p(f"(({xlen_type} *)&data)[{2*i}]") if tdata1 == 0: break tdata2 = self.gdb.p(f"(({xlen_type} *)&data)[{2*i+1}]") if tdata1 & dmode: dmode_count += 1 else: assertEqual(tdata1 & 0xffff, tdata1_lsbs) assertEqual(tdata2, tdata2) assertGreater(i, 1) assertEqual(dmode_count, 1) return triggers def test(self): # If we want this test to run from flash, we can't have any software # breakpoints set. self.gdb.command("hbreak write_load_trigger") self.gdb.p("$pc=write_store_trigger") output = self.gdb.c() assertIn("write_load_trigger", output) self.check_triggers((1<<6) | (1<<1), 0xdeadbee0) self.gdb.command("delete") self.gdb.command("hbreak clear_triggers") output = self.gdb.c() assertIn("clear_triggers", output) self.check_triggers((1<<6) | (1<<0), 0xfeedac00) self.gdb.command("delete") self.exit() class RegsTest(GdbSingleHartTest): compile_args = ("programs/regs.S", ) def setup(self): self.gdb.load() main_bp = self.gdb.b("main") output = self.gdb.c() assertIn("Breakpoint ", output) assertIn("main", output) self.gdb.command(f"delete {main_bp}") self.gdb.b("handle_trap") class WriteGprs(RegsTest): def test(self): if self.hart.extensionSupported('E'): regs = [(f"x{n}") for n in range(2, 16)] else: regs = [(f"x{n}") for n in range(2, 32)] self.gdb.p("$pc=write_regs") for i, r in enumerate(regs): self.gdb.p(f"${r}={(0xdeadbeef<\n") self.download_c.write( b"unsigned int crc32a(uint8_t *message, unsigned int size);\n") self.download_c.write(b"const uint32_t length = %d;\n" % length) self.download_c.write(b"const uint8_t d[%d] = {\n" % length) self.crc = 0 assert length % 16 == 0 for i in range(length // 16): self.download_c.write(f" /* 0x{i * 16:04x} */ ".encode()) for _ in range(16): value = random.randrange(1<<8) self.download_c.write(f"0x{value:02x}, ".encode()) self.crc = binascii.crc32(struct.pack("B", value), self.crc) self.download_c.write(b"\n") self.download_c.write(b"};\n") self.download_c.write(b"uint8_t *data = &d[0];\n") self.download_c.write( b"uint32_t main() { return crc32a(data, length); }\n") self.download_c.flush() if self.crc < 0: self.crc += 2**32 compiled = {} for hart in self.target.harts: key = hart.system if key not in compiled: compiled[key] = self.target.compile(hart, self.download_c.name, "programs/checksum.c") self.gdb.select_hart(hart) self.gdb.command(f"file {compiled.get(key)}") self.gdb.select_hart(self.hart) def test(self): self.gdb.load() self.parkOtherHarts() self.gdb.command("b _exit") #self.gdb.c(ops=100) self.gdb.c() assertEqual(self.gdb.p("status"), self.crc) os.unlink(self.download_c.name) #class MprvTest(GdbSingleHartTest): # compile_args = ("programs/mprv.S", ) # def setup(self): # self.gdb.load() # # def test(self): # """Test that the debugger can access memory when MPRV is set.""" # self.gdb.c(wait=False) # time.sleep(0.5) # self.gdb.interrupt() # output = self.gdb.command("p/x *(int*)(((char*)&data)-0x80000000)") # assertIn("0xbead", output) class PrivTest(GdbSingleHartTest): """Base class for a few tests that change privilege levels.""" compile_args = ("programs/priv.S", ) def setup(self): # pylint: disable=attribute-defined-outside-init self.gdb.load() misa = self.hart.misa self.supported = set() if misa & (1<<20): self.supported.add(0) if misa & (1<<18): self.supported.add(1) if misa & (1<<7): self.supported.add(2) self.supported.add(3) self.disable_pmp() # Ensure Virtual Memory is disabled if applicable (SATP register is not # reset) try: self.gdb.p("$satp=0") except testlib.CouldNotFetch: # SATP only exists if you have S mode. pass class PrivRw(PrivTest): """Test reading/writing priv.""" def test(self): self.write_nop_program(4) for privilege in range(4): self.gdb.p(f"$priv={privilege}") self.gdb.stepi() actual = self.gdb.p("$priv") assertIn(actual, self.supported) if privilege in self.supported: assertEqual(actual, privilege) class PrivChange(PrivTest): """Test that the core's privilege level actually changes when the debugger writes it.""" def test(self): if 0 not in self.supported: raise TestNotApplicable self.gdb.b("main") self.gdb.c() # Machine mode self.gdb.p("$priv=3") main_address = self.gdb.p("$pc") self.gdb.stepi() assertEqual(f"{self.gdb.p('$pc'):x}", f"{main_address + 4:x}") # User mode self.gdb.p("$priv=0") self.gdb.stepi() # Should have taken an exception, so be at trap_entry pc = self.gdb.p("$pc") trap_entry = self.gdb.p("&trap_entry") assertEqual(pc, trap_entry) class CheckMisa(GdbTest): """Make sure the misa we're using is actually what the target exposes.""" def test(self): for hart in self.target.harts: self.gdb.select_hart(hart) misa = self.gdb.p("$misa") assertEqual(misa, hart.misa) class TranslateTest(GdbSingleHartTest): compile_args = ("programs/translate.c", ) def early_applicable(self): return self.hart.ram_size >= 32 * 1024 def setup(self): self.disable_pmp() self.gdb.load() self.gdb.b("main") output = self.gdb.c() assertRegex(output, r"\bmain\b") def check_satp(self, mode): if self.hart.xlen == 32: satp = mode << 31 else: satp = mode << 60 try: self.gdb.p(f"$satp=0x{satp:x}") except testlib.CouldNotFetch as cnf: raise TestNotApplicable from cnf readback = self.gdb.p("$satp") self.gdb.p("$satp=0") if readback != satp: raise TestNotApplicable def test_translation(self): self.gdb.b("error") self.gdb.b("handle_trap") self.gdb.b("main:active") output = self.gdb.c() assertRegex(output, r"\bmain\b") assertEqual(0xdeadbeef, self.gdb.p("physical[0]")) assertEqual(0x55667788, self.gdb.p("physical[1]")) assertEqual(0xdeadbeef, self.gdb.p("virtual[0]")) assertEqual(0x55667788, self.gdb.p("virtual[1]")) SATP_MODE_OFF = 0 SATP_MODE_SV32 = 1 SATP_MODE_SV39 = 8 SATP_MODE_SV48 = 9 SATP_MODE_SV57 = 10 SATP_MODE_SV64 = 11 class Sv32Test(TranslateTest): def early_applicable(self): return TranslateTest.early_applicable(self) and self.hart.xlen == 32 def test(self): self.check_satp(SATP_MODE_SV32) self.gdb.p("vms=&sv32") self.test_translation() class Sv39Test(TranslateTest): def early_applicable(self): return TranslateTest.early_applicable(self) and self.hart.xlen > 32 def test(self): self.check_satp(SATP_MODE_SV39) self.gdb.p("vms=&sv39") self.test_translation() class Sv48Test(TranslateTest): def early_applicable(self): return TranslateTest.early_applicable(self) and self.hart.xlen > 32 def test(self): self.check_satp(SATP_MODE_SV48) self.gdb.p("vms=&sv48") self.test_translation() class VectorTest(GdbSingleHartTest): compile_args = ("programs/vectors.S", ) def early_applicable(self): if not self.hart.extensionSupported('V'): return False # If the compiler can't build this test, say it's not applicable. At # some time all compilers will support the V extension, but we're not # there yet. try: self.compile() except CompileError as e: if b"Error: unknown CSR `vlenb'" in e.stderr: return False return True def setup(self): self.gdb.load() self.gdb.b("main") self.gdb.c() def test(self): vlenb = self.gdb.p("$vlenb") self.gdb.command("delete") self.gdb.b("_exit") self.gdb.b("trap_entry") self.gdb.b("test0") output = self.gdb.c() assertIn("Breakpoint", output) assertIn("test0", output) # I'm not convinced that writing 0 is supported on every vector # implementation. If this test fails, that might be why. for regname in ('$vl', '$vtype'): value = self.gdb.p(regname) assertNotEqual(value, 0) self.gdb.p(f"{regname}=0") self.gdb.command("maintenance flush register-cache") assertEqual(self.gdb.p(regname), 0) self.gdb.p(f"{regname}=0x{value:x}") self.gdb.command("maintenance flush register-cache") assertEqual(self.gdb.p(regname), value) assertEqual(self.gdb.p("$a0"), 0) a = self.gdb.x("&a", 'b', vlenb) b = self.gdb.x("&b", 'b', vlenb) v4 = self.gdb.p("$v4") assertEqual(a, b) assertEqual(b, v4["b"]) assertEqual(0, self.gdb.p("$a0")) self.gdb.b("test1") output = self.gdb.c() assertIn("Breakpoint", output) assertIn("test1", output) assertEqual(self.gdb.p("$a0"), 0) b = self.gdb.x("&b", 'b', vlenb) c = self.gdb.x("&c", 'b', vlenb) v4 = self.gdb.p("$v4") assertEqual(b, c) assertEqual(c, v4["b"]) assertEqual(0, self.gdb.p("$a0")) output = self.gdb.c() assertIn("Breakpoint", output) assertIn("_exit", output) assertEqual(self.gdb.p("status"), 0) class EbreakTest(GdbSingleHartTest): """Test that we work correctly when somebody puts an ebreak directly into their code.""" compile_args = ("programs/ebreak.c", ) def setup(self): self.gdb.load() self.gdb.b("_exit") def test(self): # Should hit ebreak in the code. output = self.gdb.c() assertIn("ebreak", output) ebreak_pc = self.gdb.p("$pc") # Simple resume, we should hit the same ebreak again. output = self.gdb.c() assertIn("ebreak", output) assertEqual(ebreak_pc, self.gdb.p("$pc")) # Test getting past the ebreak by changing the PC. for _ in range(2): self.gdb.p("$pc=$pc+4") output = self.gdb.c() assertIn("ebreak", output) assertEqual(ebreak_pc, self.gdb.p("$pc")) self.gdb.p("$pc=$pc+4") output = self.gdb.c() assertIn("_exit", output) class UnavailableMultiTest(GdbTest): """Test that we work correctly when a hart becomes unavailable (e.g. because it's powered down).""" compile_args = ("programs/counting_loop.c", "-DDEFINE_MALLOC", "-DDEFINE_FREE") def early_applicable(self): return (self.hart.support_cease or self.target.support_unavailable_control) \ and len(self.target.harts) > 1 def setup(self): ProgramTest.setup(self) self.parkOtherHarts() def test(self): # Run all the way to the infinite loop in exit self.gdb.c_all(wait=False) # Other hart should have become unavailable. if self.target.support_unavailable_control: self.server.wait_until_running(self.target.harts) self.server.set_available([self.hart]) self.gdb.expect(r"\S+ became unavailable.") self.gdb.interrupt() for hart in self.target.harts: # Try to read misa on the ceased harts if hart != self.hart: try: self.gdb.select_hart(hart) self.gdb.p("$misa") assert False, \ "Shouldn't be able to access unavailable hart." except (UnknownThread, CommandException): pass # Check that the main hart can still be debugged. self.gdb.select_hart(self.hart) main_addr = self.gdb.p("$pc=main") self.gdb.stepi() # Assume the first instruction of main is not a jump. pc = self.gdb.p("$pc") assertGreater(pc, main_addr) assertLess(pc, main_addr + 8) self.gdb.p("$pc=_start") self.exit() class CeaseStepiTest(ProgramTest): """Test that we work correctly when the hart we're debugging ceases to respond.""" def early_applicable(self): return self.hart.support_cease def test(self): self.gdb.b("main") output = self.gdb.c() assertIn("Breakpoint", output) assertIn("main", output) self.gdb.p("$pc=cease") self.gdb.stepi(wait=False) self.gdb.expect(r"\S+ became unavailable.") self.gdb.interrupt() try: self.gdb.p("$pc") assert False, ("Registers shouldn't be accessible when the hart is " "unavailable.") except CouldNotReadRegisters: pass class UnavailableRunTest(ProgramTest): """Test that we work correctly when the hart we're debugging ceases to respond.""" def early_applicable(self): return self.hart.support_cease or \ self.target.support_unavailable_control def test(self): self.gdb.b("main") output = self.gdb.c() assertIn("Breakpoint", output) assertIn("main", output) if self.target.support_unavailable_control: self.gdb.p("$pc=loop_forever") else: self.gdb.p("$pc=cease") self.gdb.c(wait=False) if self.target.support_unavailable_control: self.server.wait_until_running([self.hart]) self.server.set_available( [h for h in self.target.harts if h != self.hart]) self.gdb.expect(r"\S+ became unavailable.") self.gdb.interrupt() # gdb might automatically switch to the available hart. try: self.gdb.select_hart(self.hart) except ThreadTerminated: # GDB sees that the thread is gone. Count this as success. return try: self.gdb.p("$pc") assert False, ("Registers shouldn't be accessible when the hart is " "unavailable.") except CouldNotReadRegisters: pass class UnavailableCycleTest(ProgramTest): """Test that harts can be debugged after becoming temporarily unavailable.""" def early_applicable(self): return self.target.support_unavailable_control def test(self): self.gdb.b("main") output = self.gdb.c() assertIn("Breakpoint", output) assertIn("main", output) self.gdb.p("$pc=loop_forever") self.gdb.c(wait=False) self.server.wait_until_running([self.hart]) self.server.set_available( [h for h in self.target.harts if h != self.hart]) self.gdb.expect(r"\S+ became unavailable.") # Now send a DMI command through OpenOCD to make the hart available # again. self.server.set_available(self.target.harts) self.gdb.expect(r"\S+ became available") self.gdb.interrupt() self.gdb.p("$pc") class UnavailableHaltedTest(ProgramTest): """Test behavior when the current hart becomes unavailable while halted.""" def early_applicable(self): return self.target.support_unavailable_control def test_resume(self, c_expect=None): # Confirm things don't completely fall apart on `c` self.gdb.c(wait=False) if c_expect: self.gdb.expect(c_expect) else: time.sleep(1) # Now send a DMI command through OpenOCD to make the hart available # again. self.server.set_available(self.target.harts) # The hart will show up as halted. That's just how spike behaves when we # make a hart unavailable while it's halted. self.gdb.expect("became available") self.gdb.p("$minstret") def test(self): self.gdb.b("main") output = self.gdb.c() assertIn("Breakpoint", output) assertIn("main", output) self.server.set_available( [h for h in self.target.harts if h != self.hart]) self.gdb.command(f"# disabled hart {self.hart.id}") # gdb won't show that the hart became unavailable, because it thinks # nothing can changed on a halted Linux thread. try: # We can't try this with something reasonable like $pc, because gdb # has cached it, and it assumes the target can't change while it's # halted. self.gdb.p("$minstret") assert False, ("Registers shouldn't be accessible when the hart is " "unavailable.") except testlib.CouldNotFetch: pass # There's a breakpoint set, so gdb will single step. You can't single # step an unavailable target, so gdb should get a message to that # effect. self.test_resume(c_expect="unavailable") # Delete breakpoints self.gdb.command("delete") self.server.set_available( [h for h in self.target.harts if h != self.hart]) # Resume again. With breakpoints cleared, gdb will send vCont;c instead # of step. There should be no error this time, since there is no # observable difference between an unavailable thread and a running # thread. self.test_resume() class FreeRtosTest(GdbTest): def early_applicable(self): return self.target.freertos_binary def freertos(self): return True def test(self): self.gdb.command(f"file {self.target.freertos_binary}") self.gdb.load() output = self.gdb.command("monitor riscv_freertos_stacking mainline") # Turn off htif, which doesn't work when the file is loaded into spike # through gdb. It only works when spike loads the ELF file itself. bp = self.gdb.b("main") self.gdb.c() self.gdb.command(f"delete {bp}") self.gdb.p("*((int*) &use_htif) = 0") # Need this, otherwise gdb complains that there is no current active # thread. self.gdb.threads() # fake 'Current Execution' thread is created. threads = self.gdb.threads() assertEqual(len(threads), 1) assertIn("Current Execution", threads[0]) bp = self.gdb.b("vTaskStartScheduler") self.gdb.c() self.gdb.command(f"delete {bp}") # 'Current Execution' is still there before the scheduler has been # started. Now there are 3 threads: Current Execution, Rx, Tx. threads = self.gdb.threads() assertEqual(len(threads), 3) bp = self.gdb.b("prvQueueReceiveTask") self.gdb.c() self.gdb.command(f"delete {bp}") bp = self.gdb.b("prvQueueSendTask") self.gdb.c() self.gdb.command(f"delete {bp}") # no more fake 'Current Execution' thread. # Now there are 4 threads: Rx, Tx, IDLE, Tmr Svc. threads = self.gdb.threads() assertEqual(len(threads), 4) values = {} for thread in threads: assertNotIn("No Name", thread[1]) self.gdb.thread(thread) assertEqual(self.gdb.p("$zero"), 0) output = self.gdb.command("info reg sp") assertIn("ucHeap", output) self.gdb.command("info reg mstatus") values[thread.id] = self.gdb.p("$s11") self.gdb.p(f"$s11=0x{values[thread.id] ^ int(thread.id):x}") # Test that writing worked self.gdb.stepi() for thread in self.gdb.threads(): self.gdb.thread(thread) assertEqual(self.gdb.p("$s11"), values[thread.id] ^ int(thread.id)) class StepThread2Test(GdbTest): # Check that we can do stepi on thread 2 without GDB switching to thread 1. # There was a bug where this could happen, because OpenOCD was mistakenly # omitting a thread ID in its stop reply. This was addressed in OpenOCD, # but if there is a regression in the future, this test should catch it) def early_applicable(self): return len(self.target.harts) > 1 def test(self): output = self.gdb.command("thread 2") if "Unknown thread" in output: raise TestNotApplicable before = self.gdb.command("thread") self.gdb.stepi() after = self.gdb.command("thread") # make sure that single-step doesn't alter # GDB's conception of the current thread assertEqual(before, after) class EtriggerTest(DebugTest): def setup(self): DebugTest.setup(self) self.gdb.b("main:start") self.gdb.c() self.gdb.b("handle_trap") def test(self): # Set trigger on Load access fault self.gdb.command("monitor riscv etrigger set m 0x20") # Set fox to a null pointer so we'll get a load access exception later. self.gdb.p("fox=(char*)0") output = self.gdb.c() # We should not be at handle_trap assertNotIn("handle_trap", output) # Instead, we should have hit a breakpoint at trap_entry, which is the # actual exception handler. assertIn("breakpoint", output) assertIn("trap_entry", self.gdb.where()) class IcountTest(DebugTest): compile_args = ("programs/infinite_loop.S", ) def setup(self): DebugTest.setup(self) self.gdb.b("main") self.gdb.c() def test(self): # Execute 2 instructions. output = self.gdb.command("monitor riscv icount set m 2") if self.target.icount_limit > 1: assertNotIn("Failed", output) else: assertIn("Failed", output) self.gdb.b("main_post_csrr") output = self.gdb.c() assertIn("main_post_csrr", output) main_post_csrr = self.gdb.p("&main_post_csrr") assertEqual(self.gdb.p("$pc"), main_post_csrr) self.gdb.command("monitor riscv icount clear") # Execute 1 instruction. output = self.gdb.command("monitor riscv icount set m 1") assertNotIn("Failed", output) output = self.gdb.c() assertIn("breakpoint", output) assertEqual(self.gdb.p("$pc"), main_post_csrr + 4) class ItriggerTest(GdbSingleHartTest): compile_args = ("programs/interrupt.c",) def early_applicable(self): return self.target.supports_clint_mtime def setup(self): self.gdb.load() def test(self): output = self.gdb.command("monitor riscv itrigger set 0x80") assertIn("Doesn't make sense", output) output = self.gdb.command("monitor riscv itrigger set m 0") assertIn("Doesn't make sense", output) output = self.gdb.command("monitor riscv itrigger clear") assertIn("No itrigger is set", output) self.gdb.command("monitor riscv itrigger set m 0x80") self.gdb.c() assertIn("trap_entry", self.gdb.where()) self.gdb.command("monitor riscv itrigger clear") self.gdb.p("keep_running=0") self.exit() parsed = None def main(): parser = argparse.ArgumentParser( description="Test that gdb can talk to a RISC-V target.", epilog=""" Example command line from the real world: Run all RegsTest cases against a physical FPGA, with custom openocd command: ./gdbserver.py --freedom-e300 --server_cmd "$HOME/SiFive/openocd/src/openocd -s $HOME/SiFive/openocd/tcl -d" Simple """) targets.add_target_options(parser) testlib.add_test_run_options(parser) # TODO: remove global global parsed # pylint: disable=global-statement parsed = parser.parse_args() target = targets.target(parsed) testlib.print_log_names = parsed.print_log_names module = sys.modules[__name__] return testlib.run_all_tests(module, target, parsed) # TROUBLESHOOTING TIPS # If a particular test fails, run just that one test, eg.: # ./gdbserver.py MprvTest.test_mprv # Then inspect gdb.log and spike.log to see what happened in more detail. if __name__ == '__main__': sys.exit(main())