On Tue, Oct 06, 2020 at 06:09:55PM +0300, Pavel Dovgalyuk wrote: > On 06.10.2020 16:36, Cleber Rosa wrote: > > On Sat, Oct 03, 2020 at 08:14:06PM +0300, Pavel Dovgalyuk wrote: > > > From: Pavel Dovgalyuk > > > > > > This is a test for GDB reverse debugging commands: reverse step and reverse continue. > > > Every test in this suite consists of two phases: record and replay. > > > Recording saves the execution of some instructions and makes an initial > > > VM snapshot to allow reverse execution. > > > Replay saves the order of the first instructions and then checks that they > > > are executed backwards in the correct order. > > > After that the execution is replayed to the end, and reverse continue > > > command is checked by setting several breakpoints, and asserting > > > that the execution is stopped at the last of them. > > > > > > Signed-off-by: Pavel Dovgalyuk > > > Reviewed-by: Philippe Mathieu-Daudé > > > Reviewed-by: Willian Rampazzo > > > > > > -- > > > > > > v5: > > > - disabled (as some other tests) when running on gitlab > > > due to the unidentified timeout problem > > > --- > > > MAINTAINERS | 1 > > > tests/acceptance/reverse_debugging.py | 208 +++++++++++++++++++++++++++++++++ > > > 2 files changed, 209 insertions(+) > > > create mode 100644 tests/acceptance/reverse_debugging.py > > > > > > diff --git a/MAINTAINERS b/MAINTAINERS > > > index ea4fa3e481..bd3a7efb75 100644 > > > --- a/MAINTAINERS > > > +++ b/MAINTAINERS > > > @@ -2693,6 +2693,7 @@ F: include/sysemu/replay.h > > > F: docs/replay.txt > > > F: stubs/replay.c > > > F: tests/acceptance/replay_kernel.py > > > +F: tests/acceptance/reverse_debugging.py > > > F: qapi/replay.json > > > IOVA Tree > > > diff --git a/tests/acceptance/reverse_debugging.py b/tests/acceptance/reverse_debugging.py > > > new file mode 100644 > > > index 0000000000..b72fdf6cdc > > > --- /dev/null > > > +++ b/tests/acceptance/reverse_debugging.py > > > @@ -0,0 +1,208 @@ > > > +# Reverse debugging test > > > +# > > > +# Copyright (c) 2020 ISP RAS > > > +# > > > +# Author: > > > +# Pavel Dovgalyuk > > > +# > > > +# This work is licensed under the terms of the GNU GPL, version 2 or > > > +# later. See the COPYING file in the top-level directory. > > > +import os > > > +import logging > > > + > > > +from avocado import skipIf > > > +from avocado_qemu import BUILD_DIR > > > +from avocado.utils import gdb > > > +from avocado.utils import process > > > +from avocado.utils.path import find_command > > > +from boot_linux_console import LinuxKernelTest > > > + > > > +class ReverseDebugging(LinuxKernelTest): > > > + """ > > > + Test GDB reverse debugging commands: reverse step and reverse continue. > > > + Recording saves the execution of some instructions and makes an initial > > > + VM snapshot to allow reverse execution. > > > + Replay saves the order of the first instructions and then checks that they > > > + are executed backwards in the correct order. > > > + After that the execution is replayed to the end, and reverse continue > > > + command is checked by setting several breakpoints, and asserting > > > + that the execution is stopped at the last of them. > > > + """ > > > + > > > + timeout = 10 > > > + STEPS = 10 > > > + endian_is_le = True > > > > Have you attmepted a "be" test too? I'm curious about why this is > > defined (and used later) but it's never used as `False`. > > It was intended to be used with PPC, but PPCs record-replay is not reliable > enough. > OK, thanks for the explanation. > > > > > + > > > + def run_vm(self, record, shift, args, replay_path, image_path): > > > + logger = logging.getLogger('replay') > > > + vm = self.get_vm() > > > + vm.set_console() > > > + if record: > > > + logger.info('recording the execution...') > > > + mode = 'record' > > > + else: > > > + logger.info('replaying the execution...') > > > + mode = 'replay' > > > + vm.add_args('-s', '-S') > > > + vm.add_args('-icount', 'shift=%s,rr=%s,rrfile=%s,rrsnapshot=init' % > > > + (shift, mode, replay_path), > > > + '-net', 'none') > > > + vm.add_args('-drive', 'file=%s,if=none' % image_path) > > > + if args: > > > + vm.add_args(*args) > > > + vm.launch() > > > + return vm > > > + > > > + @staticmethod > > > + def get_reg_le(g, reg): > > > + res = g.cmd(b'p%x' % reg) > > > + num = 0 > > > + for i in range(len(res))[-2::-2]: > > > + num = 0x100 * num + int(res[i:i + 2], 16) > > > + return num > > > + > > > + @staticmethod > > > + def get_reg_be(g, reg): > > > + res = g.cmd(b'p%x' % reg) > > > + return int(res, 16) > > > + > > > + def get_reg(self, g, reg): > > > + # value may be encoded in BE or LE order > > > + if self.endian_is_le: > > > + return self.get_reg_le(g, reg) > > > + else: > > > + return self.get_reg_be(g, reg) > > > + > > > + def get_pc(self, g): > > > + return self.get_reg(g, self.REG_PC) > > > + > > > + def check_pc(self, g, addr): > > > + pc = self.get_pc(g) > > > + if pc != addr: > > > + self.fail('Invalid PC (read %x instead of %x)' % (pc, addr)) > > > + > > > + @staticmethod > > > + def gdb_step(g): > > > + g.cmd(b's', b'T05thread:01;') > > > + > > > + @staticmethod > > > + def gdb_bstep(g): > > > + g.cmd(b'bs', b'T05thread:01;') > > > + > > > + @staticmethod > > > + def vm_get_icount(vm): > > > + return vm.qmp('query-replay')['return']['icount'] > > > + > > > + def reverse_debugging(self, shift=7, args=None): > > > + logger = logging.getLogger('replay') > > > + > > > + # create qcow2 for snapshots > > > + logger.info('creating qcow2 image for VM snapshots') > > > + image_path = os.path.join(self.workdir, 'disk.qcow2') > > > + qemu_img = os.path.join(BUILD_DIR, 'qemu-img') > > > + if not os.path.exists(qemu_img): > > > + qemu_img = find_command('qemu-img', False) > > > + if qemu_img is False: > > > + self.cancel('Could not find "qemu-img", which is required to ' > > > + 'create the temporary qcow2 image') > > > > This snippet is clearly modeled after the snippet in > > `boot_linux.BootLinuxBase.download_boot()`. I'm adding an action > > item to create a generic utility: > > > > https://gitlab.com/cleber.gnu/qemu/-/issues/1 > > > > > + cmd = '%s create -f qcow2 %s 128M' % (qemu_img, image_path) > > > + process.run(cmd) > > > + > > > + replay_path = os.path.join(self.workdir, 'replay.bin') > > > + > > > + # record the log > > > + vm = self.run_vm(True, shift, args, replay_path, image_path) > > > + while self.vm_get_icount(vm) <= self.STEPS: > > > + pass > > > + last_icount = self.vm_get_icount(vm) > > > + vm.shutdown() > > > + > > > + logger.info("recorded log with %s+ steps" % last_icount) > > > + > > > + # replay and run debug commands > > > + vm = self.run_vm(False, shift, args, replay_path, image_path) > > > + logger.info('connecting to gdbstub') > > > + g = gdb.GDBRemote('127.0.0.1', 1234, False, False) > > > + g.connect() > > > + r = g.cmd(b'qSupported') > > > + if b'qXfer:features:read+' in r: > > > + g.cmd(b'qXfer:features:read:target.xml:0,ffb') > > > + if b'ReverseStep+' not in r: > > > + self.fail('Reverse step is not supported by QEMU') > > > + if b'ReverseContinue+' not in r: > > > + self.fail('Reverse continue is not supported by QEMU') > > > + > > > + logger.info('stepping forward') > > > + steps = [] > > > + # record first instruction addresses > > > + for _ in range(self.STEPS): > > > + pc = self.get_pc(g) > > > + logger.info('saving position %x' % pc) > > > + steps.append(pc) > > > + self.gdb_step(g) > > > > Do you think it'd make sense to have more utility methods, such as > > `step()` and `bstep()` in `avocado.utils.gdb.GDBRemote` itself? > > I thought about it, but it was easier to not have the dependency on newer > avocado version. Agreed. > But now we can move these functions into avocado in two steps. > OK. I think the versions of these functions in `avocado.utils.gdb.GDBRemote` can benefit from parsing the reply packets. With that, in addition to using a strict expected reponse (like you've done with b'T05thread:01;') the caller may inspect only the aspects that it deems important. For instance, one may be interested in asserting that the signal was a SIGTRAP, but may not care about the thread ID. Anyway, I'm opening an issue on the Avocado project page: https://github.com/avocado-framework/avocado/issues/4258 If you have ideas about the interface, please let me know. Thanks, - Cleber.