1 =========================
2 Hardware Latency Detector
3 =========================
4
5 Introduction
6 -------------
7
8 The tracer hwlat_detector is a special purpose
9 detect large system latencies induced by the b
10 hardware or firmware, independent of Linux its
11 originally to detect SMIs (System Management I
12 however there is nothing x86 specific about th
13 originally written for use by the "RT" patch s
14 kernel is highly latency sensitive.
15
16 SMIs are not serviced by the Linux kernel, whi
17 even know that they are occurring. SMIs are in
18 and are serviced by BIOS code, usually for "cr
19 management of thermal sensors and fans. Someti
20 other tasks and those tasks can spend an inord
21 handler (sometimes measured in milliseconds).
22 you are trying to keep event service latencies
23
24 The hardware latency detector works by hogging
25 amounts of time (with interrupts disabled), po
26 for some period, then looking for gaps in the
27 time when the polling was interrupted and sinc
28 the only thing that could do that would be an
29 (or an NMI, but those can be tracked).
30
31 Note that the hwlat detector should *NEVER* be
32 It is intended to be run manually to determine
33 problem with long system firmware service rout
34
35 Usage
36 ------
37
38 Write the ASCII text "hwlat" into the current_
39 (mounted at /sys/kernel/tracing or /sys/kernel
40 redefine the threshold in microseconds (us) ab
41 be taken into account.
42
43 Example::
44
45 # echo hwlat > /sys/kernel/tracing/cur
46 # echo 100 > /sys/kernel/tracing/traci
47
48 The /sys/kernel/tracing/hwlat_detector interfa
49
50 - width - time period to sample with CPUs he
51 must be less than the total window
52 - window - total period of sampling, width b
53
54 By default the width is set to 500,000 and win
55 for every 1,000,000 usecs (1s) the hwlat detec
56 (0.5s). If tracing_thresh contains zero when h
57 change to a default of 10 usecs. If any latenc
58 observed then the data will be written to the
59
60 The minimum sleep time between periods is 1 mi
61 is less than 1 millisecond apart from window,
62 be totally starved.
63
64 If tracing_thresh was zero when hwlat detector
65 back to zero if another tracer is loaded. Note
66 tracing_thresh that hwlat detector had will be
67 be restored in tracing_thresh if it is still z
68 started again.
69
70 The following tracing directory files are used
71
72 in /sys/kernel/tracing:
73
74 - tracing_threshold - minimum latency valu
75 - tracing_max_latency - maximum hardware lat
76 - tracing_cpumask - the CPUs to move the
77 - hwlat_detector/width - specified amount of
78 - hwlat_detector/window - amount of ti
79 - hwlat_detector/mode - the thread mode
80
81 By default, one hwlat detector's kernel thread
82 specified in cpumask at the beginning of a new
83 fashion. This behavior can be changed by chang
84 the available options are:
85
86 - none: do not force migration
87 - round-robin: migrate across each CPU specif
88 - per-cpu: create one thread for each cpu
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.