xref: /openbmc/qemu/docs/devel/tcg.rst (revision f76b348e)
1====================
2Translator Internals
3====================
4
5QEMU is a dynamic translator. When it first encounters a piece of code,
6it converts it to the host instruction set. Usually dynamic translators
7are very complicated and highly CPU dependent. QEMU uses some tricks
8which make it relatively easily portable and simple while achieving good
9performances.
10
11QEMU's dynamic translation backend is called TCG, for "Tiny Code
12Generator". For more information, please take a look at ``tcg/README``.
13
14Some notable features of QEMU's dynamic translator are:
15
16CPU state optimisations
17-----------------------
18
19The target CPUs have many internal states which change the way it
20evaluates instructions. In order to achieve a good speed, the
21translation phase considers that some state information of the virtual
22CPU cannot change in it. The state is recorded in the Translation
23Block (TB). If the state changes (e.g. privilege level), a new TB will
24be generated and the previous TB won't be used anymore until the state
25matches the state recorded in the previous TB. The same idea can be applied
26to other aspects of the CPU state.  For example, on x86, if the SS,
27DS and ES segments have a zero base, then the translator does not even
28generate an addition for the segment base.
29
30Direct block chaining
31---------------------
32
33After each translated basic block is executed, QEMU uses the simulated
34Program Counter (PC) and other cpu state information (such as the CS
35segment base value) to find the next basic block.
36
37In order to accelerate the most common cases where the new simulated PC
38is known, QEMU can patch a basic block so that it jumps directly to the
39next one.
40
41The most portable code uses an indirect jump. An indirect jump makes
42it easier to make the jump target modification atomic. On some host
43architectures (such as x86 or PowerPC), the ``JUMP`` opcode is
44directly patched so that the block chaining has no overhead.
45
46Self-modifying code and translated code invalidation
47----------------------------------------------------
48
49Self-modifying code is a special challenge in x86 emulation because no
50instruction cache invalidation is signaled by the application when code
51is modified.
52
53User-mode emulation marks a host page as write-protected (if it is
54not already read-only) every time translated code is generated for a
55basic block.  Then, if a write access is done to the page, Linux raises
56a SEGV signal. QEMU then invalidates all the translated code in the page
57and enables write accesses to the page.  For system emulation, write
58protection is achieved through the software MMU.
59
60Correct translated code invalidation is done efficiently by maintaining
61a linked list of every translated block contained in a given page. Other
62linked lists are also maintained to undo direct block chaining.
63
64On RISC targets, correctly written software uses memory barriers and
65cache flushes, so some of the protection above would not be
66necessary. However, QEMU still requires that the generated code always
67matches the target instructions in memory in order to handle
68exceptions correctly.
69
70Exception support
71-----------------
72
73longjmp() is used when an exception such as division by zero is
74encountered.
75
76The host SIGSEGV and SIGBUS signal handlers are used to get invalid
77memory accesses.  QEMU keeps a map from host program counter to
78target program counter, and looks up where the exception happened
79based on the host program counter at the exception point.
80
81On some targets, some bits of the virtual CPU's state are not flushed to the
82memory until the end of the translation block.  This is done for internal
83emulation state that is rarely accessed directly by the program and/or changes
84very often throughout the execution of a translation block---this includes
85condition codes on x86, delay slots on SPARC, conditional execution on
86Arm, and so on.  This state is stored for each target instruction, and
87looked up on exceptions.
88
89MMU emulation
90-------------
91
92For system emulation QEMU uses a software MMU. In that mode, the MMU
93virtual to physical address translation is done at every memory
94access.
95
96QEMU uses an address translation cache (TLB) to speed up the translation.
97In order to avoid flushing the translated code each time the MMU
98mappings change, all caches in QEMU are physically indexed.  This
99means that each basic block is indexed with its physical address.
100
101In order to avoid invalidating the basic block chain when MMU mappings
102change, chaining is only performed when the destination of the jump
103shares a page with the basic block that is performing the jump.
104
105The MMU can also distinguish RAM and ROM memory areas from MMIO memory
106areas.  Access is faster for RAM and ROM because the translation cache also
107hosts the offset between guest address and host memory.  Accessing MMIO
108memory areas instead calls out to C code for device emulation.
109Finally, the MMU helps tracking dirty pages and pages pointed to by
110translation blocks.
111
112