aboutsummaryrefslogtreecommitdiffstats
path: root/src
Commit message (Collapse)AuthorAge
* Modified ft to support fz overhead tracingHEADmasterztong2021-03-10
|
* Update ft2csv.cricardobtxr2019-06-03
| | | Retrieve UNLOCK overheads when UNLOCK_START and UNLOCK_END records are interleaved by SCHED and SCHED2 records.
* st-job-stats: output number of preemptions and migrationsBjoern Brandenburg2016-04-24
|
* Port st-dump and st-job-statsBjoern Brandenburg2016-03-23
| | | | | | Include st-dump (formerly 'st_show') and st-job-stats (formerly 'st_job_stats') from https://github.com/brandenburg/sched-trace-tools in this repository.
* ftcat: add -p (ping) optionBjoern Brandenburg2016-03-08
| | | | Ask ftcat to store its PID when it is done with initialization.
* ft2csv: add -l (list IDs in trace) optionBjoern Brandenburg2016-03-08
| | | | Use ft2csv -l <MY-TRACE-FILE> to get a list of all present events.
* Support new event IDsBjoern Brandenburg2016-03-08
| | | | ...XCALL, SEND_XCALL, SCHED_TIMER, and TIMER_LATENCY.
* Add support for QUANTUM_BOUNDARY_{START,END}Bjoern Brandenburg2014-06-07
|
* Implement -c (implausbile latency filter) in ftsortBjoern Brandenburg2014-01-14
| | | | | | | | | | | | | | | | | | | | | | | | | | | Given the option -c <CYCLES PER NANOSECOND> ftsort now looks for latencies that are implausibly large and filters them. Whether a latency sample is "implausibly large" is determined based on the following observation: the timestamps of earlier and later events indicate when a processor was in preemptable mode (i.e., points in time at which interrupts were not masked). The delta between two consecutive points imply an upper bound on the maximum "interrupts off" segment during the interval between the two points. No interrupt latency should be larger than this upper bound. Note that this option should *NOT* be required for publication-quality benchmarks in most circumstances. In particular, this should never be necessary when benchmarking LITMUS^RT kernels. If it becomes necessary, then this should be considered a bug and reported. For context, this filter was originally developed for benchmarking SCHED_DEADLINE, in which "jobs are released" (i.e., nanosleep() calls are completed) such that implausibly large latencies may arise if the system becomes overloaded.
* Implement -v (verbose messages) option in ftcatBjoern Brandenburg2014-01-14
|
* Add TSC calibration support to ftcatBjoern Brandenburg2014-01-14
| | | | | | Provide the -c option to calibrate the TSC offsets. Provide the -v option to get output about it.
* Remove two sources of outliersBjoern Brandenburg2012-10-19
| | | | | - Don't enforce sequentiality constraint on SEND_RESCHED samples. - Refuse samples with negative time deltas (possible due to clock skew).
* Be more verbose about aborted movesBjoern Brandenburg2012-10-19
|
* Add monotonicity filtering to ftsortBjoern Brandenburg2012-10-19
| | | | | | Timestamps and sequence numbers should be monotonic with regard to a single CPU and/or or task. Non-compliant samples were disturbed and should be rejected.
* Add -s (simulate) and -v (verbose) flags to ftsortBjoern Brandenburg2012-10-19
| | | | | | The -s flag prevents ftsort from writing back results to disk. The -v (verbose) flag enables more detailed output. These flags are useful when debugging traces and filtering.
* Add filtering of IRQ-disturbed samples (on by default)Bjoern Brandenburg2012-10-19
| | | | | | Interrupts can cause massive outliers. Instead of trying to filter them afterwards, let's not even include such samples in the first place.
* Support matching based on PID in ft2csvBjoern Brandenburg2012-10-19
|
* Update timestamp structure to reflect PID fieldBjoern Brandenburg2012-10-19
|
* Update locking timestamp IDsBjoern Brandenburg2012-10-19
| | | | ...to mirror those used in the kernel.
* Support terminating ftcat after a given number of samples has been readBjoern Brandenburg2012-10-19
|
* Remove hard-coded thresholdBjoern Brandenburg2012-08-01
| | | | | | The hard-coded threshold is a leftover from the beginnings. It can be surprising and is no longer needed: outlier filtering should happen during post-processing, if at all.
* Add support for IRQ flag and counter in binary trace format.Bjoern B. Brandenburg2012-01-30
| | | | | | | | This introduces the new bit allocation and adds support to ftdump for showing the values. Not yet included is any support for filtering based on IRQ interference.
* create nicer output in ftdumpBjoern B. Brandenburg2011-11-24
|
* update timestamp typesBjoern B. Brandenburg2011-07-25
|
* report ioctl errorsBjoern B. Brandenburg2011-07-25
|
* Switch to ioctl() interface for Feather-Trace devicesBjoern B. Brandenburg2011-07-23
|
* Fix compiler warning.Bjoern B. Brandenburg2011-07-23
| | | | picky gcc...
* Support filtering by CPU identity.Bjoern B. Brandenburg2011-02-21
|
* write back dirty pages and keep track of speedBjoern B. Brandenburg2011-02-18
|
* pre-page mapped filesBjoern B. Brandenburg2011-02-18
|
* allow binary output in ft2csvBjoern B. Brandenburg2011-02-18
| | | | This can help with large data sets.
* Remove endian swapping from ft2csv; don't skip over holesBjoern B. Brandenburg2011-02-17
|
* Add tool for sorting binary Feather-Trace streamsBjoern B. Brandenburg2011-02-17
|
* show non-consecutive sequence numbersBjoern B. Brandenburg2011-02-17
|
* add support for release latency; make ftdump a bit prettierBjoern B. Brandenburg2011-02-17
|
* Add TS_RELEASE_LATENCYBjoern B. Brandenburg2011-02-05
|
* Parametrize CYCLES per USAndrea Bastoni2010-04-21
|
* ft2csv: accept short names (without the trailing _START)Bjoern B. Brandenburg2009-05-05
| | | | Also, adjust line breaks to fit within 80 chars.
* silence a bunch of 64bit-specific warningsBjoern B. Brandenburg2009-05-05
|
* remove unused code to silence gcc warningBjoern B. Brandenburg2009-05-05
|
* add SEND_RESCHED eventBjoern B. Brandenburg2009-04-28
|
* fix the documentation fix...Bjoern B. Brandenburg2009-04-28
|
* document -b optionBjoern B. Brandenburg2009-04-28
|
* optionally look at non-rt overheadsBjoern B. Brandenburg2008-12-09
|
* timestamp.h: rip out old unneeded definitionsBjoern B. Brandenburg2008-12-09
|
* add license stuffBjoern B. Brandenburg2008-12-09
|
* properly disable eventsBjoern B. Brandenburg2008-12-09
|
* provide usage informationBjoern B. Brandenburg2008-12-09
|
* display bytes transferredBjoern B. Brandenburg2008-11-08
|
* bugfix: advance to next record in byte order conversionBjoern Brandenburg2008-09-25
|
generated by cgit v1.2.2 (git 2.25.0) at 2026-01-12 17:39:08 -0500