kernel/tools/power/pm-graph/sleepgraph.8
2024-07-22 17:22:30 +08:00

345 lines
12 KiB
Groff

.TH SLEEPGRAPH 8
.SH NAME
sleepgraph \- Suspend/Resume timing analysis
.SH SYNOPSIS
.ft B
.B sleepgraph
.RB [ OPTIONS ]
.RB [ COMMAND ]
.SH DESCRIPTION
\fBsleepgraph \fP is designed to assist kernel and OS developers
in optimizing their linux stack's suspend/resume time. Using a kernel
image built with a few extra options enabled, the tool will execute a
suspend and capture dmesg and ftrace data until resume is complete.
This data is transformed into a device timeline and an optional
callgraph to give a detailed view of which devices/subsystems are
taking the most time in suspend/resume.
.PP
If no specific command is given, the default behavior is to initiate
a suspend/resume.
.PP
Generates output files in subdirectory: suspend-yymmdd-HHMMSS
html timeline : <hostname>_<mode>.html
raw dmesg file : <hostname>_<mode>_dmesg.txt
raw ftrace file : <hostname>_<mode>_ftrace.txt
.SH OPTIONS
.TP
\fB-h\fR
Print the help text.
.TP
\fB-v\fR
Print the current tool version.
.TP
\fB-verbose\fR
Print extra information during execution and analysis.
.TP
\fB-config \fIfile\fR
Pull arguments and config options from a file.
.TP
\fB-m \fImode\fR
Mode to initiate for suspend e.g. standby, freeze, mem (default: mem).
.TP
\fB-o \fIname\fR
Overrides the output subdirectory name when running a new test.
Use {date}, {time}, {hostname} for current values.
.sp
e.g. suspend-{hostname}-{date}-{time}
.TP
\fB-rtcwake \fIt\fR | off
Use rtcwake to autoresume after \fIt\fR seconds (default: 15). Set t to "off" to
disable rtcwake and require a user keypress to resume.
.TP
\fB-addlogs\fR
Add the dmesg and ftrace logs to the html output. They will be viewable by
clicking buttons in the timeline.
.TP
\fB-noturbostat\fR
By default, if turbostat is found and the requested mode is freeze, sleepgraph
will execute the suspend via turbostat and collect data in the timeline log.
This option disables the use of turbostat.
.TP
\fB-result \fIfile\fR
Export a results table to a text file for parsing.
.TP
\fB-sync\fR
Sync the filesystems before starting the test. This reduces the size of
the sys_sync call which happens in the suspend_prepare phase.
.TP
\fB-rs \fIenable/disable\fR
During test, enable/disable runtime suspend for all devices. The test is delayed
by 5 seconds to allow runtime suspend changes to occur. The settings are restored
after the test is complete.
.TP
\fB-display \fIon/off/standby/suspend\fR
Switch the display to the requested mode for the test using the xset command.
This helps maintain the consistency of test data for better comparison.
.TP
\fB-wifi\fR
If a wifi connection is available, check that it reconnects after resume. Include
the reconnect time in the total resume time calculation and treat wifi timeouts
as resume failures.
.SS "advanced"
.TP
\fB-gzip\fR
Gzip the trace and dmesg logs to save space. The tool can also read in gzipped
logs for processing.
.TP
\fB-cmd \fIstr\fR
Run the timeline over a custom suspend command, e.g. pm-suspend. By default
the tool forces suspend via /sys/power/state so this allows testing over
an OS's official suspend method. The output file will change to
hostname_command.html and will autodetect which suspend mode was triggered.
.TP
\fB-filter \fI"d1,d2,..."\fR
Filter out all but these device callbacks. These strings can be device names
or module names. e.g. 0000:00:02.0, ata5, i915, usb, etc.
.TP
\fB-mindev \fIt\fR
Discard all device callbacks shorter than \fIt\fR milliseconds (default: 0.0).
This reduces the html file size as there can be many tiny callbacks which are barely
visible. The value is a float: e.g. 0.001 represents 1 us.
.TP
\fB-proc\fR
Add usermode process info into the timeline (default: disabled).
.TP
\fB-dev\fR
Add kernel source calls and threads to the timeline (default: disabled).
.TP
\fB-x2\fR
Run two suspend/resumes back to back (default: disabled).
.TP
\fB-x2delay \fIt\fR
Include \fIt\fR ms delay between multiple test runs (default: 0 ms).
.TP
\fB-predelay \fIt\fR
Include \fIt\fR ms delay before 1st suspend (default: 0 ms).
.TP
\fB-postdelay \fIt\fR
Include \fIt\fR ms delay after last resume (default: 0 ms).
.TP
\fB-multi \fIn d\fR
Used for endurance testing. If \fIn\fR is entirely numeric, it's treated as a count:
Execute \fIn\fR consecutive tests at \fId\fR second intervals.
If \fIn\fR is an integer followed by a "d", "h", or "m", it's treated as a duration:
Execute tests continuously over \fIn\fR days, hours, or minutes at \fId\fR second intervals.
The outputs will be created in a new subdirectory, for count: suspend-{date}-{time}-xN,
for duration: suspend-{date}-{time}-Nm. When the multitest run is done, the \fI-summary\fR
command is called automatically to create summary html files for all the data (unless you
use \fI-skiphtml\fR). \fI-skiphtml\fR will speed up the testing by not creating timelines
or summary html files. You can then run the tool again at a later time with \fI-summary\fR
and \fI-genhtml\fR to create the timelines.
.TP
\fB-maxfail \fIn\fR
Abort a -multi run after \fIn\fR consecutive fails. 0 means never abort (default = 0).
.TP
\fB-skiphtml\fR
Run the test and capture the trace logs, but skip the timeline generation.
You can generate the html timelines later with \fI-dmesg\fR & \fI-ftrace\fR, or
by running \fI-summary\fR and \fI-genhtml\fR.
.SS "ftrace debug"
.TP
\fB-f\fR
Use ftrace to create device callgraphs (default: disabled). This can produce
very large outputs, i.e. 10MB - 100MB.
.TP
\fB-ftop\fR
Use ftrace on the top level call: "suspend_devices_and_enter" only (default: disabled).
This option implies -f and creates a single callgraph covering all of suspend/resume.
.TP
\fB-maxdepth \fIlevel\fR
limit the callgraph trace depth to \fIlevel\fR (default: 0=all). This is
the best way to limit the output size when using callgraphs via -f.
.TP
\fB-expandcg\fR
pre-expand the callgraph data in the html output (default: disabled)
.TP
\fB-fadd \fIfile\fR
Add functions to be graphed in the timeline from a list in a text file
.TP
\fB-mincg \fIt\fR
Discard all callgraphs shorter than \fIt\fR milliseconds (default: 0.0).
This reduces the html file size as there can be many tiny callgraphs
which are barely visible in the timeline.
The value is a float: e.g. 0.001 represents 1 us.
.TP
\fB-cgfilter \fI"func1,func2,..."\fR
Reduce callgraph output in the timeline by limiting it certain devices. The
argument can be a single device name or a comma delimited list.
(default: none)
.TP
\fB-cgskip \fIfile\fR
Reduce callgraph timeline size by skipping over uninteresting functions
in the trace, e.g. printk or console_unlock. The functions listed
in this file will show up as empty leaves in the callgraph with only the start/end
times displayed. cgskip.txt is used automatically if found in the path, so
use "off" to disable completely (default: cgskip.txt)
.TP
\fB-cgphase \fIp\fR
Only show callgraph data for phase \fIp\fR (e.g. suspend_late).
.TP
\fB-cgtest \fIn\fR
In an x2 run, only show callgraph data for test \fIn\fR (e.g. 0 or 1).
.TP
\fB-timeprec \fIn\fR
Number of significant digits in timestamps (0:S, [3:ms], 6:us).
.TP
\fB-bufsize \fIN\fR
Set trace buffer size to N kilo-bytes (default: all of free memory up to 3GB)
.SH COMMANDS
.TP
\fB-summary \fIindir\fR
Create a set of summary pages for all tests in \fIindir\fR recursively.
Creates summary.html, summary-issues.html, and summary-devices.html in the current folder.
summary.html is a table of tests with relevant info sorted by kernel/host/mode,
and links to the test html files. It identifies the minimum, maximum, and median
suspend and resume times for you with highlights and links in the header.
summary-issues.html is a list of kernel issues found in dmesg from all the tests.
summary-devices.html is a list of devices and times from all the tests.
Use \fI-genhtml\fR to regenerate any tests with missing html.
.TP
\fB-genhtml\fR
Used with \fI-summary\fR to regenerate any missing html timelines from their
dmesg and ftrace logs. This will require a significant amount of time if there
are thousands of tests.
.TP
\fB-modes\fR
List available suspend modes.
.TP
\fB-status\fR
Test to see if the system is able to run this tool. Use this along
with any options you intend to use to see if they will work.
.TP
\fB-fpdt\fR
Print out the contents of the ACPI Firmware Performance Data Table.
.TP
\fB-wificheck\fR
Print out wifi status and connection details.
.TP
\fB-xon/-xoff/-xstandby/-xsuspend\fR
Test xset by attempting to switch the display to the given mode. This
is the same command which will be issued by \fB-display \fImode\fR.
.TP
\fB-xstat\fR
Get the current DPMS display mode.
.TP
\fB-sysinfo\fR
Print out system info extracted from BIOS. Reads /dev/mem directly instead of going through dmidecode.
.TP
\fB-devinfo\fR
Print out the pm settings of all devices which support runtime suspend.
.TP
\fB-cmdinfo\fR
Print out all the platform data collected from the system that makes it into the logs.
.TP
\fB-flist\fR
Print the list of ftrace functions currently being captured. Functions
that are not available as symbols in the current kernel are shown in red.
By default, the tool traces a list of important suspend/resume functions
in order to better fill out the timeline. If the user has added their own
with -fadd they will also be checked.
.TP
\fB-flistall\fR
Print all ftrace functions capable of being captured. These are all the
possible values you can add to trace via the -fadd argument.
.SS "rebuild"
.TP
\fB-ftrace \fIfile\fR
Create HTML output from an existing ftrace file.
.TP
\fB-dmesg \fIfile\fR
Create HTML output from an existing dmesg file.
.SH EXAMPLES
.SS "simple commands"
Check which suspend modes are currently supported.
.IP
\f(CW$ sleepgraph -modes\fR
.PP
Read the Firmware Performance Data Table (FPDT)
.IP
\f(CW$ sudo sleepgraph -fpdt\fR
.PP
Print out the current USB power topology
.IP
\f(CW$ sleepgraph -usbtopo
.PP
Verify that you can run a command with a set of arguments
.IP
\f(CW$ sudo sleepgraph -f -rtcwake 30 -status
.PP
Generate a summary of all timelines in a particular folder.
.IP
\f(CW$ sleepgraph -summary ~/workspace/myresults/\fR
.PP
.SS "capturing basic timelines"
Execute a mem suspend with a 15 second wakeup. Include the logs in the html.
.IP
\f(CW$ sudo sleepgraph -rtcwake 15 -addlogs\fR
.PP
Execute a standby with a 15 second wakeup. Change the output folder name.
.IP
\f(CW$ sudo sleepgraph -m standby -rtcwake 15 -o "standby-{host}-{date}-{time}"\fR
.PP
Execute a freeze with no wakeup (require keypress). Change output folder name.
.IP
\f(CW$ sudo sleepgraph -m freeze -rtcwake off -o "freeze-{hostname}-{date}-{time}"\fR
.PP
.SS "capturing advanced timelines"
Execute a suspend & include dev mode source calls, limit callbacks to 5ms or larger.
.IP
\f(CW$ sudo sleepgraph -m mem -rtcwake 15 -dev -mindev 5\fR
.PP
Run two suspends back to back, include a 500ms delay before, after, and in between runs.
.IP
\f(CW$ sudo sleepgraph -m mem -rtcwake 15 -x2 -predelay 500 -x2delay 500 -postdelay 500\fR
.PP
Execute a suspend using a custom command.
.IP
\f(CW$ sudo sleepgraph -cmd "echo mem > /sys/power/state" -rtcwake 15\fR
.PP
.SS "endurance testing using -multi"
.PP
Do a batch run of 10 freezes with 30 seconds delay between runs.
.IP
\f(CW$ sudo sleepgraph -m freeze -rtcwake 15 -multi 10 30\fR
.PP
Do a batch run of freezes for 24 hours.
.IP
\f(CW$ sudo sleepgraph -m freeze -rtcwake 15 -multi 24h 0\fR
.SS "adding callgraph data"
Add device callgraphs. Limit the trace depth and only show callgraphs 10ms or larger.
.IP
\f(CW$ sudo sleepgraph -m mem -rtcwake 15 -f -maxdepth 5 -mincg 10\fR
.PP
Capture a full callgraph across all suspend, then filter the html by a single phase.
.IP
\f(CW$ sudo sleepgraph -m mem -rtcwake 15 -f\fR
.IP
\f(CW$ sleepgraph -dmesg host_mem_dmesg.txt -ftrace host_mem_ftrace.txt -f -cgphase resume
.PP
.SS "rebuild timeline from logs"
.PP
Rebuild the html from a previous run's logs, using the same options.
.IP
\f(CW$ sleepgraph -dmesg dmesg.txt -ftrace ftrace.txt -callgraph\fR
.PP
Rebuild the html with different options.
.IP
\f(CW$ sleepgraph -dmesg dmesg.txt -ftrace ftrace.txt -addlogs -srgap\fR
.SH "SEE ALSO"
dmesg(1)
.PP
.SH AUTHOR
.nf
Written by Todd Brandt <todd.e.brandt@linux.intel.com>