xref: /openbmc/qemu/docs/devel/tcg.rst (revision bad5cfcd)
1.. _tcg_internals:
2
3====================
4Translator Internals
5====================
6
7QEMU is a dynamic translator. When it first encounters a piece of code,
8it converts it to the host instruction set. Usually dynamic translators
9are very complicated and highly CPU dependent. QEMU uses some tricks
10which make it relatively easily portable and simple while achieving good
11performances.
12
13QEMU's dynamic translation backend is called TCG, for "Tiny Code
14Generator". For more information, please take a look at :ref:`tcg-ops-ref`.
15
16The following sections outline some notable features and implementation
17details of QEMU's dynamic translator.
18
19CPU state optimisations
20-----------------------
21
22The target CPUs have many internal states which change the way they
23evaluate instructions. In order to achieve a good speed, the
24translation phase considers that some state information of the virtual
25CPU cannot change in it. The state is recorded in the Translation
26Block (TB). If the state changes (e.g. privilege level), a new TB will
27be generated and the previous TB won't be used anymore until the state
28matches the state recorded in the previous TB. The same idea can be applied
29to other aspects of the CPU state.  For example, on x86, if the SS,
30DS and ES segments have a zero base, then the translator does not even
31generate an addition for the segment base.
32
33Direct block chaining
34---------------------
35
36After each translated basic block is executed, QEMU uses the simulated
37Program Counter (PC) and other CPU state information (such as the CS
38segment base value) to find the next basic block.
39
40In its simplest, less optimized form, this is done by exiting from the
41current TB, going through the TB epilogue, and then back to the
42main loop. That’s where QEMU looks for the next TB to execute,
43translating it from the guest architecture if it isn’t already available
44in memory. Then QEMU proceeds to execute this next TB, starting at the
45prologue and then moving on to the translated instructions.
46
47Exiting from the TB this way will cause the ``cpu_exec_interrupt()``
48callback to be re-evaluated before executing additional instructions.
49It is mandatory to exit this way after any CPU state changes that may
50unmask interrupts.
51
52In order to accelerate the cases where the TB for the new
53simulated PC is already available, QEMU has mechanisms that allow
54multiple TBs to be chained directly, without having to go back to the
55main loop as described above. These mechanisms are:
56
57``lookup_and_goto_ptr``
58^^^^^^^^^^^^^^^^^^^^^^^
59
60Calling ``tcg_gen_lookup_and_goto_ptr()`` will emit a call to
61``helper_lookup_tb_ptr``. This helper will look for an existing TB that
62matches the current CPU state. If the destination TB is available its
63code address is returned, otherwise the address of the JIT epilogue is
64returned. The call to the helper is always followed by the tcg ``goto_ptr``
65opcode, which branches to the returned address. In this way, we either
66branch to the next TB or return to the main loop.
67
68``goto_tb + exit_tb``
69^^^^^^^^^^^^^^^^^^^^^
70
71The translation code usually implements branching by performing the
72following steps:
73
741. Call ``tcg_gen_goto_tb()`` passing a jump slot index (either 0 or 1)
75   as a parameter.
76
772. Emit TCG instructions to update the CPU state with any information
78   that has been assumed constant and is required by the main loop to
79   correctly locate and execute the next TB. For most guests, this is
80   just the PC of the branch destination, but others may store additional
81   data. The information updated in this step must be inferable from both
82   ``cpu_get_tb_cpu_state()`` and ``cpu_restore_state()``.
83
843. Call ``tcg_gen_exit_tb()`` passing the address of the current TB and
85   the jump slot index again.
86
87Step 1, ``tcg_gen_goto_tb()``, will emit a ``goto_tb`` TCG
88instruction that later on gets translated to a jump to an address
89associated with the specified jump slot. Initially, this is the address
90of step 2's instructions, which update the CPU state information. Step 3,
91``tcg_gen_exit_tb()``, exits from the current TB returning a tagged
92pointer composed of the last executed TB’s address and the jump slot
93index.
94
95The first time this whole sequence is executed, step 1 simply jumps
96to step 2. Then the CPU state information gets updated and we exit from
97the current TB. As a result, the behavior is very similar to the less
98optimized form described earlier in this section.
99
100Next, the main loop looks for the next TB to execute using the
101current CPU state information (creating the TB if it wasn’t already
102available) and, before starting to execute the new TB’s instructions,
103patches the previously executed TB by associating one of its jump
104slots (the one specified in the call to ``tcg_gen_exit_tb()``) with the
105address of the new TB.
106
107The next time this previous TB is executed and we get to that same
108``goto_tb`` step, it will already be patched (assuming the destination TB
109is still in memory) and will jump directly to the first instruction of
110the destination TB, without going back to the main loop.
111
112For the ``goto_tb + exit_tb`` mechanism to be used, the following
113conditions need to be satisfied:
114
115* The change in CPU state must be constant, e.g., a direct branch and
116  not an indirect branch.
117
118* The direct branch cannot cross a page boundary. Memory mappings
119  may change, causing the code at the destination address to change.
120
121Note that, on step 3 (``tcg_gen_exit_tb()``), in addition to the
122jump slot index, the address of the TB just executed is also returned.
123This address corresponds to the TB that will be patched; it may be
124different than the one that was directly executed from the main loop
125if the latter had already been chained to other TBs.
126
127Self-modifying code and translated code invalidation
128----------------------------------------------------
129
130Self-modifying code is a special challenge in x86 emulation because no
131instruction cache invalidation is signaled by the application when code
132is modified.
133
134User-mode emulation marks a host page as write-protected (if it is
135not already read-only) every time translated code is generated for a
136basic block.  Then, if a write access is done to the page, Linux raises
137a SEGV signal. QEMU then invalidates all the translated code in the page
138and enables write accesses to the page.  For system emulation, write
139protection is achieved through the software MMU.
140
141Correct translated code invalidation is done efficiently by maintaining
142a linked list of every translated block contained in a given page. Other
143linked lists are also maintained to undo direct block chaining.
144
145On RISC targets, correctly written software uses memory barriers and
146cache flushes, so some of the protection above would not be
147necessary. However, QEMU still requires that the generated code always
148matches the target instructions in memory in order to handle
149exceptions correctly.
150
151Exception support
152-----------------
153
154longjmp() is used when an exception such as division by zero is
155encountered.
156
157The host SIGSEGV and SIGBUS signal handlers are used to get invalid
158memory accesses.  QEMU keeps a map from host program counter to
159target program counter, and looks up where the exception happened
160based on the host program counter at the exception point.
161
162On some targets, some bits of the virtual CPU's state are not flushed to the
163memory until the end of the translation block.  This is done for internal
164emulation state that is rarely accessed directly by the program and/or changes
165very often throughout the execution of a translation block---this includes
166condition codes on x86, delay slots on SPARC, conditional execution on
167Arm, and so on.  This state is stored for each target instruction, and
168looked up on exceptions.
169
170MMU emulation
171-------------
172
173For system emulation QEMU uses a software MMU. In that mode, the MMU
174virtual to physical address translation is done at every memory
175access.
176
177QEMU uses an address translation cache (TLB) to speed up the translation.
178In order to avoid flushing the translated code each time the MMU
179mappings change, all caches in QEMU are physically indexed.  This
180means that each basic block is indexed with its physical address.
181
182In order to avoid invalidating the basic block chain when MMU mappings
183change, chaining is only performed when the destination of the jump
184shares a page with the basic block that is performing the jump.
185
186The MMU can also distinguish RAM and ROM memory areas from MMIO memory
187areas.  Access is faster for RAM and ROM because the translation cache also
188hosts the offset between guest address and host memory.  Accessing MMIO
189memory areas instead calls out to C code for device emulation.
190Finally, the MMU helps tracking dirty pages and pages pointed to by
191translation blocks.
192
193Profiling JITted code
194---------------------
195
196The Linux ``perf`` tool will treat all JITted code as a single block as
197unlike the main code it can't use debug information to link individual
198program counter samples with larger functions. To overcome this
199limitation you can use the ``-perfmap`` or the ``-jitdump`` option to generate
200map files. ``-perfmap`` is lightweight and produces only guest-host mappings.
201``-jitdump`` additionally saves JITed code and guest debug information (if
202available); its output needs to be integrated with the ``perf.data`` file
203before the final report can be viewed.
204
205.. code::
206
207  perf record $QEMU -perfmap $REMAINING_ARGS
208  perf report
209
210  perf record -k 1 $QEMU -jitdump $REMAINING_ARGS
211  DEBUGINFOD_URLS= perf inject -j -i perf.data -o perf.data.jitted
212  perf report -i perf.data.jitted
213
214Note that qemu-system generates mappings only for ``-kernel`` files in ELF
215format.
216