xref: /openbmc/qemu/docs/system/replay.rst (revision afb81fe8)
1.. _replay:
2
3..
4    Copyright (c) 2010-2022 Institute for System Programming
5                        of the Russian Academy of Sciences.
6
7    This work is licensed under the terms of the GNU GPL, version 2 or later.
8    See the COPYING file in the top-level directory.
9
10Record/replay
11=============
12
13Record/replay functions are used for the deterministic replay of qemu execution.
14Execution recording writes a non-deterministic events log, which can be later
15used for replaying the execution anywhere and for unlimited number of times.
16It also supports checkpointing for faster rewind to the specific replay moment.
17Execution replaying reads the log and replays all non-deterministic events
18including external input, hardware clocks, and interrupts.
19
20Deterministic replay has the following features:
21
22 * Deterministically replays whole system execution and all contents of
23   the memory, state of the hardware devices, clocks, and screen of the VM.
24 * Writes execution log into the file for later replaying for multiple times
25   on different machines.
26 * Supports i386, x86_64, ARM, AArch64, Risc-V, MIPS, MIPS64, S390X, Alpha,
27   PowerPC, PowerPC64, M68000, Microblaze, OpenRISC, Nios II, SPARC,
28   and Xtensa hardware platforms.
29 * Performs deterministic replay of all operations with keyboard and mouse
30   input devices, serial ports, and network.
31
32Usage of the record/replay:
33
34 * First, record the execution with the following command line:
35
36    .. parsed-literal::
37        |qemu_system| \\
38        -icount shift=auto,rr=record,rrfile=replay.bin \\
39        -drive file=disk.qcow2,if=none,snapshot,id=img-direct \\
40        -drive driver=blkreplay,if=none,image=img-direct,id=img-blkreplay \\
41        -device ide-hd,drive=img-blkreplay \\
42        -netdev user,id=net1 -device rtl8139,netdev=net1 \\
43        -object filter-replay,id=replay,netdev=net1
44
45 * After recording, you can replay it by using another command line:
46
47    .. parsed-literal::
48        |qemu_system| \\
49        -icount shift=auto,rr=replay,rrfile=replay.bin \\
50        -drive file=disk.qcow2,if=none,snapshot,id=img-direct \\
51        -drive driver=blkreplay,if=none,image=img-direct,id=img-blkreplay \\
52        -device ide-hd,drive=img-blkreplay \\
53        -netdev user,id=net1 -device rtl8139,netdev=net1 \\
54        -object filter-replay,id=replay,netdev=net1
55
56   The only difference with recording is changing the rr option
57   from record to replay.
58 * Block device images are not actually changed in the recording mode,
59   because all of the changes are written to the temporary overlay file.
60   This behavior is enabled by using blkreplay driver. It should be used
61   for every enabled block device, as described in :ref:`block-label` section.
62 * ``-net none`` option should be specified when network is not used,
63   because QEMU adds network card by default. When network is needed,
64   it should be configured explicitly with replay filter, as described
65   in :ref:`network-label` section.
66 * Interaction with audio devices and serial ports are recorded and replayed
67   automatically when such devices are enabled.
68
69Core idea
70---------
71
72Record/replay system is based on saving and replaying non-deterministic
73events (e.g. keyboard input) and simulating deterministic ones (e.g. reading
74from HDD or memory of the VM). Saving only non-deterministic events makes
75log file smaller and simulation faster.
76
77The following non-deterministic data from peripheral devices is saved into
78the log: mouse and keyboard input, network packets, audio controller input,
79serial port input, and hardware clocks (they are non-deterministic
80too, because their values are taken from the host machine). Inputs from
81simulated hardware, memory of VM, software interrupts, and execution of
82instructions are not saved into the log, because they are deterministic and
83can be replayed by simulating the behavior of virtual machine starting from
84initial state.
85
86Instruction counting
87--------------------
88
89QEMU should work in icount mode to use record/replay feature. icount was
90designed to allow deterministic execution in absence of external inputs
91of the virtual machine. Record/replay feature is enabled through ``-icount``
92command-line option, making possible deterministic execution of the machine,
93interacting with user or network.
94
95.. _block-label:
96
97Block devices
98-------------
99
100Block devices record/replay module intercepts calls of
101bdrv coroutine functions at the top of block drivers stack.
102To record and replay block operations the drive must be configured
103as following:
104
105.. parsed-literal::
106    -drive file=disk.qcow2,if=none,snapshot,id=img-direct
107    -drive driver=blkreplay,if=none,image=img-direct,id=img-blkreplay
108    -device ide-hd,drive=img-blkreplay
109
110blkreplay driver should be inserted between disk image and virtual driver
111controller. Therefore all disk requests may be recorded and replayed.
112
113.. _snapshotting-label:
114
115Snapshotting
116------------
117
118New VM snapshots may be created in replay mode. They can be used later
119to recover the desired VM state. All VM states created in replay mode
120are associated with the moment of time in the replay scenario.
121After recovering the VM state replay will start from that position.
122
123Default starting snapshot name may be specified with icount field
124rrsnapshot as follows:
125
126.. parsed-literal::
127    -icount shift=auto,rr=record,rrfile=replay.bin,rrsnapshot=snapshot_name
128
129This snapshot is created at start of recording and restored at start
130of replaying. It also can be loaded while replaying to roll back
131the execution.
132
133``snapshot`` flag of the disk image must be removed to save the snapshots
134in the overlay (or original image) instead of using the temporary overlay.
135
136.. parsed-literal::
137    -drive file=disk.ovl,if=none,id=img-direct
138    -drive driver=blkreplay,if=none,image=img-direct,id=img-blkreplay
139    -device ide-hd,drive=img-blkreplay
140
141Use QEMU monitor to create additional snapshots. ``savevm <name>`` command
142created the snapshot and ``loadvm <name>`` restores it. To prevent corruption
143of the original disk image, use overlay files linked to the original images.
144Therefore all new snapshots (including the starting one) will be saved in
145overlays and the original image remains unchanged.
146
147When you need to use snapshots with diskless virtual machine,
148it must be started with "orphan" qcow2 image. This image will be used
149for storing VM snapshots. Here is the example of the command line for this:
150
151.. parsed-literal::
152    |qemu_system| \\
153      -icount shift=auto,rr=replay,rrfile=record.bin,rrsnapshot=init \\
154      -net none -drive file=empty.qcow2,if=none,id=rr
155
156``empty.qcow2`` drive does not connected to any virtual block device and used
157for VM snapshots only.
158
159.. _network-label:
160
161Network devices
162---------------
163
164Record and replay for network interactions is performed with the network filter.
165Each backend must have its own instance of the replay filter as follows:
166
167.. parsed-literal::
168    -netdev user,id=net1 -device rtl8139,netdev=net1
169    -object filter-replay,id=replay,netdev=net1
170
171Replay network filter is used to record and replay network packets. While
172recording the virtual machine this filter puts all packets coming from
173the outer world into the log. In replay mode packets from the log are
174injected into the network device. All interactions with network backend
175in replay mode are disabled.
176
177Audio devices
178-------------
179
180Audio data is recorded and replay automatically. The command line for recording
181and replaying must contain identical specifications of audio hardware, e.g.:
182
183.. parsed-literal::
184    -audio pa,model=ac97
185
186Serial ports
187------------
188
189Serial ports input is recorded and replay automatically. The command lines
190for recording and replaying must contain identical number of ports in record
191and replay modes, but their backends may differ.
192E.g., ``-serial stdio`` in record mode, and ``-serial null`` in replay mode.
193
194Reverse debugging
195-----------------
196
197Reverse debugging allows "executing" the program in reverse direction.
198GDB remote protocol supports "reverse step" and "reverse continue"
199commands. The first one steps single instruction backwards in time,
200and the second one finds the last breakpoint in the past.
201
202Recorded executions may be used to enable reverse debugging. QEMU can't
203execute the code in backwards direction, but can load a snapshot and
204replay forward to find the desired position or breakpoint.
205
206The following GDB commands are supported:
207
208 - ``reverse-stepi`` (or ``rsi``) - step one instruction backwards
209 - ``reverse-continue`` (or ``rc``) - find last breakpoint in the past
210
211Reverse step loads the nearest snapshot and replays the execution until
212the required instruction is met.
213
214Reverse continue may include several passes of examining the execution
215between the snapshots. Each of the passes include the following steps:
216
217 #. loading the snapshot
218 #. replaying to examine the breakpoints
219 #. if breakpoint or watchpoint was met
220
221    * loading the snapshot again
222    * replaying to the required breakpoint
223
224 #. else
225
226    * proceeding to the p.1 with the earlier snapshot
227
228Therefore usage of the reverse debugging requires at least one snapshot
229created. This can be done by omitting ``snapshot`` option
230for the block drives and adding ``rrsnapshot`` for both record and replay
231command lines.
232See the :ref:`snapshotting-label` section to learn more about running record/replay
233and creating the snapshot in these modes.
234
235When ``rrsnapshot`` is not used, then snapshot named ``start_debugging``
236created in temporary overlay. This allows using reverse debugging, but with
237temporary snapshots (existing within the session).
238