xref: /openbmc/qemu/docs/devel/tcg-icount.rst (revision c16de0d9)
1..
2   Copyright (c) 2020, Linaro Limited
3   Written by Alex Bennée
4
5
6========================
7TCG Instruction Counting
8========================
9
10TCG has long supported a feature known as icount which allows for
11instruction counting during execution. This should not be confused
12with cycle accurate emulation - QEMU does not attempt to emulate how
13long an instruction would take on real hardware. That is a job for
14other more detailed (and slower) tools that simulate the rest of a
15micro-architecture.
16
17This feature is only available for system emulation and is
18incompatible with multi-threaded TCG. It can be used to better align
19execution time with wall-clock time so a "slow" device doesn't run too
20fast on modern hardware. It can also provides for a degree of
21deterministic execution and is an essential part of the record/replay
22support in QEMU.
23
24Core Concepts
25=============
26
27At its heart icount is simply a count of executed instructions which
28is stored in the TimersState of QEMU's timer sub-system. The number of
29executed instructions can then be used to calculate QEMU_CLOCK_VIRTUAL
30which represents the amount of elapsed time in the system since
31execution started. Depending on the icount mode this may either be a
32fixed number of ns per instruction or adjusted as execution continues
33to keep wall clock time and virtual time in sync.
34
35To be able to calculate the number of executed instructions the
36translator starts by allocating a budget of instructions to be
37executed. The budget of instructions is limited by how long it will be
38until the next timer will expire. We store this budget as part of a
39vCPU icount_decr field which shared with the machinery for handling
40cpu_exit(). The whole field is checked at the start of every
41translated block and will cause a return to the outer loop to deal
42with whatever caused the exit.
43
44In the case of icount, before the flag is checked we subtract the
45number of instructions the translation block would execute. If this
46would cause the instruction budget to go negative we exit the main
47loop and regenerate a new translation block with exactly the right
48number of instructions to take the budget to 0 meaning whatever timer
49was due to expire will expire exactly when we exit the main run loop.
50
51Dealing with MMIO
52-----------------
53
54While we can adjust the instruction budget for known events like timer
55expiry we cannot do the same for MMIO. Every load/store we execute
56might potentially trigger an I/O event, at which point we will need an
57up to date and accurate reading of the icount number.
58
59To deal with this case, when an I/O access is made we:
60
61  - restore un-executed instructions to the icount budget
62  - re-compile a single [1]_ instruction block for the current PC
63  - exit the cpu loop and execute the re-compiled block
64
65.. [1] sometimes two instructions if dealing with delay slots
66
67Other I/O operations
68--------------------
69
70MMIO isn't the only type of operation for which we might need a
71correct and accurate clock. IO port instructions and accesses to
72system registers are the common examples here. These instructions have
73to be handled by the individual translators which have the knowledge
74of which operations are I/O operations.
75
76When the translator is handling an instruction of this kind:
77
78* it must call gen_io_start() if icount is enabled, at some
79   point before the generation of the code which actually does
80   the I/O, using a code fragment similar to:
81
82.. code:: c
83
84    if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) {
85        gen_io_start();
86    }
87
88* it must end the TB immediately after this instruction
89