xref: /openbmc/qemu/docs/devel/tcg-plugins.rst (revision 3ef11c99)
1..
2   Copyright (C) 2017, Emilio G. Cota <cota@braap.org>
3   Copyright (c) 2019, Linaro Limited
4   Written by Emilio Cota and Alex Bennée
5
6.. _TCG Plugins:
7
8QEMU TCG Plugins
9================
10
11
12Writing plugins
13---------------
14
15API versioning
16~~~~~~~~~~~~~~
17
18This is a new feature for QEMU and it does allow people to develop
19out-of-tree plugins that can be dynamically linked into a running QEMU
20process. However the project reserves the right to change or break the
21API should it need to do so. The best way to avoid this is to submit
22your plugin upstream so they can be updated if/when the API changes.
23
24All plugins need to declare a symbol which exports the plugin API
25version they were built against. This can be done simply by::
26
27  QEMU_PLUGIN_EXPORT int qemu_plugin_version = QEMU_PLUGIN_VERSION;
28
29The core code will refuse to load a plugin that doesn't export a
30``qemu_plugin_version`` symbol or if plugin version is outside of QEMU's
31supported range of API versions.
32
33Additionally the ``qemu_info_t`` structure which is passed to the
34``qemu_plugin_install`` method of a plugin will detail the minimum and
35current API versions supported by QEMU. The API version will be
36incremented if new APIs are added. The minimum API version will be
37incremented if existing APIs are changed or removed.
38
39Lifetime of the query handle
40~~~~~~~~~~~~~~~~~~~~~~~~~~~~
41
42Each callback provides an opaque anonymous information handle which
43can usually be further queried to find out information about a
44translation, instruction or operation. The handles themselves are only
45valid during the lifetime of the callback so it is important that any
46information that is needed is extracted during the callback and saved
47by the plugin.
48
49Plugin life cycle
50~~~~~~~~~~~~~~~~~
51
52First the plugin is loaded and the public qemu_plugin_install function
53is called. The plugin will then register callbacks for various plugin
54events. Generally plugins will register a handler for the *atexit*
55if they want to dump a summary of collected information once the
56program/system has finished running.
57
58When a registered event occurs the plugin callback is invoked. The
59callbacks may provide additional information. In the case of a
60translation event the plugin has an option to enumerate the
61instructions in a block of instructions and optionally register
62callbacks to some or all instructions when they are executed.
63
64There is also a facility to add an inline event where code to
65increment a counter can be directly inlined with the translation.
66Currently only a simple increment is supported. This is not atomic so
67can miss counts. If you want absolute precision you should use a
68callback which can then ensure atomicity itself.
69
70Finally when QEMU exits all the registered *atexit* callbacks are
71invoked.
72
73Exposure of QEMU internals
74~~~~~~~~~~~~~~~~~~~~~~~~~~
75
76The plugin architecture actively avoids leaking implementation details
77about how QEMU's translation works to the plugins. While there are
78conceptions such as translation time and translation blocks the
79details are opaque to plugins. The plugin is able to query select
80details of instructions and system configuration only through the
81exported *qemu_plugin* functions.
82
83However the following assumptions can be made:
84
85Translation Blocks
86++++++++++++++++++
87
88All code will go through a translation phase although not all
89translations will be necessarily be executed. You need to instrument
90actual executions to track what is happening.
91
92It is quite normal to see the same address translated multiple times.
93If you want to track the code in system emulation you should examine
94the underlying physical address (``qemu_plugin_insn_haddr``) to take
95into account the effects of virtual memory although if the system does
96paging this will change too.
97
98Not all instructions in a block will always execute so if its
99important to track individual instruction execution you need to
100instrument them directly. However asynchronous interrupts will not
101change control flow mid-block.
102
103Instructions
104++++++++++++
105
106Instruction instrumentation runs before the instruction executes. You
107can be can be sure the instruction will be dispatched, but you can't
108be sure it will complete. Generally this will be because of a
109synchronous exception (e.g. SIGILL) triggered by the instruction
110attempting to execute. If you want to be sure you will need to
111instrument the next instruction as well. See the ``execlog.c`` plugin
112for examples of how to track this and finalise details after execution.
113
114Memory Accesses
115+++++++++++++++
116
117Memory callbacks are called after a successful load or store.
118Unsuccessful operations (i.e. faults) will not be visible to memory
119instrumentation although the execution side effects can be observed
120(e.g. entering a exception handler).
121
122System Idle and Resume States
123+++++++++++++++++++++++++++++
124
125The ``qemu_plugin_register_vcpu_idle_cb`` and
126``qemu_plugin_register_vcpu_resume_cb`` functions can be used to track
127when CPUs go into and return from sleep states when waiting for
128external I/O. Be aware though that these may occur less frequently
129than in real HW due to the inefficiencies of emulation giving less
130chance for the CPU to idle.
131
132Internals
133---------
134
135Locking
136~~~~~~~
137
138We have to ensure we cannot deadlock, particularly under MTTCG. For
139this we acquire a lock when called from plugin code. We also keep the
140list of callbacks under RCU so that we do not have to hold the lock
141when calling the callbacks. This is also for performance, since some
142callbacks (e.g. memory access callbacks) might be called very
143frequently.
144
145  * A consequence of this is that we keep our own list of CPUs, so that
146    we do not have to worry about locking order wrt cpu_list_lock.
147  * Use a recursive lock, since we can get registration calls from
148    callbacks.
149
150As a result registering/unregistering callbacks is "slow", since it
151takes a lock. But this is very infrequent; we want performance when
152calling (or not calling) callbacks, not when registering them. Using
153RCU is great for this.
154
155We support the uninstallation of a plugin at any time (e.g. from
156plugin callbacks). This allows plugins to remove themselves if they no
157longer want to instrument the code. This operation is asynchronous
158which means callbacks may still occur after the uninstall operation is
159requested. The plugin isn't completely uninstalled until the safe work
160has executed while all vCPUs are quiescent.
161
162Plugin API
163==========
164
165The following API is generated from the inline documentation in
166``include/qemu/qemu-plugin.h``. Please ensure any updates to the API
167include the full kernel-doc annotations.
168
169.. kernel-doc:: include/qemu/qemu-plugin.h
170