hwlat_detector.txt 3.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64

The module hwlat_detector is a special purpose kernel module that is used to
detect large system latencies induced by the behavior of certain underlying
hardware or firmware, independent of Linux itself. The code was developed
originally to detect SMIs (System Management Interrupts) on x86 systems,
however there is nothing x86 specific about this patchset. It was
originally written for use by the "RT" patch since the Real Time
kernel is highly latency sensitive.

SMIs are usually not serviced by the Linux kernel, which typically does not
even know that they are occuring. SMIs are instead are set up by BIOS code
and are serviced by BIOS code, usually for "critical" events such as
management of thermal sensors and fans. Sometimes though, SMIs are used for
other tasks and those tasks can spend an inordinate amount of time in the
handler (sometimes measured in milliseconds). Obviously this is a problem if
you are trying to keep event service latencies down in the microsecond range.

The hardware latency detector works by hogging all of the cpus for configurable
amounts of time (by calling stop_machine()), polling the CPU Time Stamp Counter
for some period, then looking for gaps in the TSC data. Any gap indicates a
time when the polling was interrupted and since the machine is stopped and
interrupts turned off the only thing that could do that would be an SMI.

Note that the SMI detector should *NEVER* be used in a production environment.
It is intended to be run manually to determine if the hardware platform has a
problem with long system firmware service routines.


Loading the module hwlat_detector passing the parameter "enabled=1" (or by
setting the "enable" entry in "hwlat_detector" debugfs toggled on) is the only
step required to start the hwlat_detector. It is possible to redefine the
threshold in microseconds (us) above which latency spikes will be taken
into account (parameter "threshold=").


	# modprobe hwlat_detector enabled=1 threshold=100

After the module is loaded, it creates a directory named "hwlat_detector" under
the debugfs mountpoint, "/debug/hwlat_detector" for this text. It is necessary
to have debugfs mounted, which might be on /sys/debug on your system.

The /debug/hwlat_detector interface contains the following files:

count			- number of latency spikes observed since last reset
enable			- a global enable/disable toggle (0/1), resets count
max			- maximum hardware latency actually observed (usecs)
sample			- a pipe from which to read current raw sample data
			  in the format <timestamp> <latency observed usecs>
			  (can be opened O_NONBLOCK for a single sample)
threshold		- minimum latency value to be considered (usecs)
width			- time period to sample with CPUs held (usecs)
			  must be less than the total window size (enforced)
window			- total period of sampling, width being inside (usecs)

By default we will set width to 500,000 and window to 1,000,000, meaning that
we will sample every 1,000,000 usecs (1s) for 500,000 usecs (0.5s). If we
observe any latencies that exceed the threshold (initially 100 usecs),
then we write to a global sample ring buffer of 8K samples, which is
consumed by reading from the "sample" (pipe) debugfs file interface.