2 Copyright (C) 2017, Emilio G. Cota <cota@braap.org>
3 Copyright (c) 2019, Linaro Limited
4 Written by Emilio Cota and Alex Bennée
9 QEMU TCG plugins provide a way for users to run experiments taking
10 advantage of the total system control emulation can have over a guest.
11 It provides a mechanism for plugins to subscribe to events during
12 translation and execution and optionally callback into the plugin
13 during these events. TCG plugins are unable to change the system state
14 only monitor it passively. However they can do this down to an
15 individual instruction granularity including potentially subscribing
16 to all load and store operations.
21 Any QEMU binary with TCG support has plugins enabled by default.
22 Earlier releases needed to be explicitly enabled with::
24 configure --enable-plugins
26 Once built a program can be run with multiple plugins loaded each with
29 $QEMU $OTHER_QEMU_ARGS \
30 -plugin contrib/plugin/libhowvec.so,inline=on,count=hint \
31 -plugin contrib/plugin/libhotblocks.so
33 Arguments are plugin specific and can be used to modify their
34 behaviour. In this case the howvec plugin is being asked to use inline
35 ops to count and break down the hint instructions by type.
37 Linux user-mode emulation also evaluates the environment variable
40 QEMU_PLUGIN="file=contrib/plugins/libhowvec.so,inline=on,count=hint" $QEMU
48 This is a new feature for QEMU and it does allow people to develop
49 out-of-tree plugins that can be dynamically linked into a running QEMU
50 process. However the project reserves the right to change or break the
51 API should it need to do so. The best way to avoid this is to submit
52 your plugin upstream so they can be updated if/when the API changes.
54 All plugins need to declare a symbol which exports the plugin API
55 version they were built against. This can be done simply by::
57 QEMU_PLUGIN_EXPORT int qemu_plugin_version = QEMU_PLUGIN_VERSION;
59 The core code will refuse to load a plugin that doesn't export a
60 ``qemu_plugin_version`` symbol or if plugin version is outside of QEMU's
61 supported range of API versions.
63 Additionally the ``qemu_info_t`` structure which is passed to the
64 ``qemu_plugin_install`` method of a plugin will detail the minimum and
65 current API versions supported by QEMU. The API version will be
66 incremented if new APIs are added. The minimum API version will be
67 incremented if existing APIs are changed or removed.
69 Lifetime of the query handle
70 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
72 Each callback provides an opaque anonymous information handle which
73 can usually be further queried to find out information about a
74 translation, instruction or operation. The handles themselves are only
75 valid during the lifetime of the callback so it is important that any
76 information that is needed is extracted during the callback and saved
82 First the plugin is loaded and the public qemu_plugin_install function
83 is called. The plugin will then register callbacks for various plugin
84 events. Generally plugins will register a handler for the *atexit*
85 if they want to dump a summary of collected information once the
86 program/system has finished running.
88 When a registered event occurs the plugin callback is invoked. The
89 callbacks may provide additional information. In the case of a
90 translation event the plugin has an option to enumerate the
91 instructions in a block of instructions and optionally register
92 callbacks to some or all instructions when they are executed.
94 There is also a facility to add an inline event where code to
95 increment a counter can be directly inlined with the translation.
96 Currently only a simple increment is supported. This is not atomic so
97 can miss counts. If you want absolute precision you should use a
98 callback which can then ensure atomicity itself.
100 Finally when QEMU exits all the registered *atexit* callbacks are
103 Exposure of QEMU internals
104 ~~~~~~~~~~~~~~~~~~~~~~~~~~
106 The plugin architecture actively avoids leaking implementation details
107 about how QEMU's translation works to the plugins. While there are
108 conceptions such as translation time and translation blocks the
109 details are opaque to plugins. The plugin is able to query select
110 details of instructions and system configuration only through the
111 exported *qemu_plugin* functions.
116 .. kernel-doc:: include/qemu/qemu-plugin.h
124 We have to ensure we cannot deadlock, particularly under MTTCG. For
125 this we acquire a lock when called from plugin code. We also keep the
126 list of callbacks under RCU so that we do not have to hold the lock
127 when calling the callbacks. This is also for performance, since some
128 callbacks (e.g. memory access callbacks) might be called very
131 * A consequence of this is that we keep our own list of CPUs, so that
132 we do not have to worry about locking order wrt cpu_list_lock.
133 * Use a recursive lock, since we can get registration calls from
136 As a result registering/unregistering callbacks is "slow", since it
137 takes a lock. But this is very infrequent; we want performance when
138 calling (or not calling) callbacks, not when registering them. Using
139 RCU is great for this.
141 We support the uninstallation of a plugin at any time (e.g. from
142 plugin callbacks). This allows plugins to remove themselves if they no
143 longer want to instrument the code. This operation is asynchronous
144 which means callbacks may still occur after the uninstall operation is
145 requested. The plugin isn't completely uninstalled until the safe work
146 has executed while all vCPUs are quiescent.
151 There are a number of plugins included with QEMU and you are
152 encouraged to contribute your own plugins plugins upstream. There is a
153 ``contrib/plugins`` directory where they can go.
157 These are some basic plugins that are used to test and exercise the
158 API during the ``make check-tcg`` target.
160 - contrib/plugins/hotblocks.c
162 The hotblocks plugin allows you to examine the where hot paths of
163 execution are in your program. Once the program has finished you will
164 get a sorted list of blocks reporting the starting PC, translation
165 count, number of instructions and execution count. This will work best
166 with linux-user execution as system emulation tends to generate
167 re-translations as blocks from different programs get swapped in and
168 out of system memory.
170 If your program is single-threaded you can use the ``inline`` option for
171 slightly faster (but not thread safe) counters.
175 ./aarch64-linux-user/qemu-aarch64 \
176 -plugin contrib/plugins/libhotblocks.so -d plugin \
177 ./tests/tcg/aarch64-linux-user/sha1
178 SHA1=15dd99a1991e0b3826fede3deffc1feba42278e6
179 collected 903 entries in the hash table
180 pc, tcount, icount, ecount
181 0x0000000041ed10, 1, 5, 66087
182 0x000000004002b0, 1, 4, 66087
185 - contrib/plugins/hotpages.c
187 Similar to hotblocks but this time tracks memory accesses::
189 ./aarch64-linux-user/qemu-aarch64 \
190 -plugin contrib/plugins/libhotpages.so -d plugin \
191 ./tests/tcg/aarch64-linux-user/sha1
192 SHA1=15dd99a1991e0b3826fede3deffc1feba42278e6
193 Addr, RCPUs, Reads, WCPUs, Writes
194 0x000055007fe000, 0x0001, 31747952, 0x0001, 8835161
195 0x000055007ff000, 0x0001, 29001054, 0x0001, 8780625
196 0x00005500800000, 0x0001, 687465, 0x0001, 335857
197 0x0000000048b000, 0x0001, 130594, 0x0001, 355
198 0x0000000048a000, 0x0001, 1826, 0x0001, 11
200 The hotpages plugin can be configured using the following arguments:
202 * sortby=reads|writes|address
204 Log the data sorted by either the number of reads, the number of writes, or
205 memory address. (Default: entries are sorted by the sum of reads and writes)
209 Track IO addresses. Only relevant to full system emulation. (Default: off)
213 The page size used. (Default: N = 4096)
215 - contrib/plugins/howvec.c
217 This is an instruction classifier so can be used to count different
218 types of instructions. It has a number of options to refine which get
219 counted. You can give a value to the ``count`` argument for a class of
220 instructions to break it down fully, so for example to see all the system
223 ./aarch64-softmmu/qemu-system-aarch64 $(QEMU_ARGS) \
224 -append "root=/dev/sda2 systemd.unit=benchmark.service" \
225 -smp 4 -plugin ./contrib/plugins/libhowvec.so,count=sreg -d plugin
227 which will lead to a sorted list after the class breakdown::
230 Class: UDEF not counted
232 Class: PCrel addr (47789483 hits)
233 Class: Add/Sub (imm) (192817388 hits)
234 Class: Logical (imm) (93852565 hits)
235 Class: Move Wide (imm) (76398116 hits)
236 Class: Bitfield (44706084 hits)
237 Class: Extract (5499257 hits)
238 Class: Cond Branch (imm) (147202932 hits)
239 Class: Exception Gen (193581 hits)
240 Class: NOP not counted
241 Class: Hints (6652291 hits)
242 Class: Barriers (8001661 hits)
243 Class: PSTATE (1801695 hits)
244 Class: System Insn (6385349 hits)
245 Class: System Reg counted individually
246 Class: Branch (reg) (69497127 hits)
247 Class: Branch (imm) (84393665 hits)
248 Class: Cmp & Branch (110929659 hits)
249 Class: Tst & Branch (44681442 hits)
250 Class: AdvSimd ldstmult (736 hits)
251 Class: ldst excl (9098783 hits)
252 Class: Load Reg (lit) (87189424 hits)
253 Class: ldst noalloc pair (3264433 hits)
254 Class: ldst pair (412526434 hits)
255 Class: ldst reg (imm) (314734576 hits)
256 Class: Loads & Stores (2117774 hits)
257 Class: Data Proc Reg (223519077 hits)
258 Class: Scalar FP (31657954 hits)
259 Individual Instructions:
260 Instr: mrs x0, sp_el0 (2682661 hits) (op=0xd5384100/ System Reg)
261 Instr: mrs x1, tpidr_el2 (1789339 hits) (op=0xd53cd041/ System Reg)
262 Instr: mrs x2, tpidr_el2 (1513494 hits) (op=0xd53cd042/ System Reg)
263 Instr: mrs x0, tpidr_el2 (1490823 hits) (op=0xd53cd040/ System Reg)
264 Instr: mrs x1, sp_el0 (933793 hits) (op=0xd5384101/ System Reg)
265 Instr: mrs x2, sp_el0 (699516 hits) (op=0xd5384102/ System Reg)
266 Instr: mrs x4, tpidr_el2 (528437 hits) (op=0xd53cd044/ System Reg)
267 Instr: mrs x30, ttbr1_el1 (480776 hits) (op=0xd538203e/ System Reg)
268 Instr: msr ttbr1_el1, x30 (480713 hits) (op=0xd518203e/ System Reg)
269 Instr: msr vbar_el1, x30 (480671 hits) (op=0xd518c01e/ System Reg)
272 To find the argument shorthand for the class you need to examine the
273 source code of the plugin at the moment, specifically the ``*opt``
274 argument in the InsnClassExecCount tables.
276 - contrib/plugins/lockstep.c
278 This is a debugging tool for developers who want to find out when and
279 where execution diverges after a subtle change to TCG code generation.
280 It is not an exact science and results are likely to be mixed once
281 asynchronous events are introduced. While the use of -icount can
282 introduce determinism to the execution flow it doesn't always follow
283 the translation sequence will be exactly the same. Typically this is
284 caused by a timer firing to service the GUI causing a block to end
285 early. However in some cases it has proved to be useful in pointing
286 people at roughly where execution diverges. The only argument you need
287 for the plugin is a path for the socket the two instances will
291 ./sparc-softmmu/qemu-system-sparc -monitor none -parallel none \
292 -net none -M SS-20 -m 256 -kernel day11/zImage.elf \
293 -plugin ./contrib/plugins/liblockstep.so,sockpath=lockstep-sparc.sock \
296 which will eventually report::
298 qemu-system-sparc: warning: nic lance.0 has no peer
299 @ 0x000000ffd06678 vs 0x000000ffd001e0 (2/1 since last)
300 @ 0x000000ffd07d9c vs 0x000000ffd06678 (3/1 since last)
301 Δ insn_count @ 0x000000ffd07d9c (809900609) vs 0x000000ffd06678 (809900612)
302 previously @ 0x000000ffd06678/10 (809900609 insns)
303 previously @ 0x000000ffd001e0/4 (809900599 insns)
304 previously @ 0x000000ffd080ac/2 (809900595 insns)
305 previously @ 0x000000ffd08098/5 (809900593 insns)
306 previously @ 0x000000ffd080c0/1 (809900588 insns)
308 - contrib/plugins/hwprofile.c
310 The hwprofile tool can only be used with system emulation and allows
311 the user to see what hardware is accessed how often. It has a number of options:
313 * track=read or track=write
315 By default the plugin tracks both reads and writes. You can use one
316 of these options to limit the tracking to just one class of accesses.
320 Will include a detailed break down of what the guest PC that made the
321 access was. Not compatible with the pattern option. Example output::
323 cirrus-low-memory @ 0xfffffd00000a0000
324 pc:fffffc0000005cdc, 1, 256
325 pc:fffffc0000005ce8, 1, 256
326 pc:fffffc0000005cec, 1, 256
330 Instead break down the accesses based on the offset into the HW
331 region. This can be useful for seeing the most used registers of a
332 device. Example output::
334 pci0-conf @ 0xfffffd01fe000000
343 - contrib/plugins/execlog.c
345 The execlog tool traces executed instructions with memory access. It can be used
346 for debugging and security analysis purposes.
347 Please be aware that this will generate a lot of output.
349 The plugin takes no argument::
351 qemu-system-arm $(QEMU_ARGS) \
352 -plugin ./contrib/plugins/libexeclog.so -d plugin
354 which will output an execution trace following this structure::
356 # vCPU, vAddr, opcode, disassembly[, load/store, memory addr, device]...
357 0, 0xa12, 0xf8012400, "movs r4, #0"
358 0, 0xa14, 0xf87f42b4, "cmp r4, r6"
359 0, 0xa16, 0xd206, "bhs #0xa26"
360 0, 0xa18, 0xfff94803, "ldr r0, [pc, #0xc]", load, 0x00010a28, RAM
361 0, 0xa1a, 0xf989f000, "bl #0xd30"
362 0, 0xd30, 0xfff9b510, "push {r4, lr}", store, 0x20003ee0, RAM, store, 0x20003ee4, RAM
363 0, 0xd32, 0xf9893014, "adds r0, #0x14"
364 0, 0xd34, 0xf9c8f000, "bl #0x10c8"
365 0, 0x10c8, 0xfff96c43, "ldr r3, [r0, #0x44]", load, 0x200000e4, RAM
367 - contrib/plugins/cache.c
369 Cache modelling plugin that measures the performance of a given L1 cache
370 configuration, and optionally a unified L2 per-core cache when a given working
373 qemu-x86_64 -plugin ./contrib/plugins/libcache.so \
374 -d plugin -D cache.log ./tests/tcg/x86_64-linux-user/float_convs
376 will report the following::
378 core #, data accesses, data misses, dmiss rate, insn accesses, insn misses, imiss rate
379 0 996695 508 0.0510% 2642799 18617 0.7044%
381 address, data misses, instruction
382 0x424f1e (_int_malloc), 109, movq %rax, 8(%rcx)
383 0x41f395 (_IO_default_xsputn), 49, movb %dl, (%rdi, %rax)
384 0x42584d (ptmalloc_init.part.0), 33, movaps %xmm0, (%rax)
385 0x454d48 (__tunables_init), 20, cmpb $0, (%r8)
388 address, fetch misses, instruction
389 0x4160a0 (__vfprintf_internal), 744, movl $1, %ebx
390 0x41f0a0 (_IO_setb), 744, endbr64
391 0x415882 (__vfprintf_internal), 744, movq %r12, %rdi
392 0x4268a0 (__malloc), 696, andq $0xfffffffffffffff0, %rax
395 The plugin has a number of arguments, all of them are optional:
399 Print top N icache and dcache thrashing instructions along with their
400 address, number of misses, and its disassembly. (default: 32)
406 Instruction cache configuration arguments. They specify the cache size, block
407 size, and associativity of the instruction cache, respectively.
408 (default: N = 16384, B = 64, A = 8)
414 Data cache configuration arguments. They specify the cache size, block size,
415 and associativity of the data cache, respectively.
416 (default: N = 16384, B = 64, A = 8)
420 Sets the eviction policy to POLICY. Available policies are: :code:`lru`,
421 :code:`fifo`, and :code:`rand`. The plugin will use the specified policy for
422 both instruction and data caches. (default: POLICY = :code:`lru`)
426 Sets the number of cores for which we maintain separate icache and dcache.
427 (default: for linux-user, N = 1, for full system emulation: N = cores
432 Simulates a unified L2 cache (stores blocks for both instructions and data)
433 using the default L2 configuration (cache size = 2MB, associativity = 16-way,
440 L2 cache configuration arguments. They specify the cache size, block size, and
441 associativity of the L2 cache, respectively. Setting any of the L2
442 configuration arguments implies ``l2=on``.
443 (default: N = 2097152 (2MB), B = 64, A = 16)