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