Kernel Function Trace
- 1 Introduction
- 2 Basic Use
- 3 Download
- 4 How To Use
- 5 Issues
- 6 Similar technologies
- 7 Filter Q&A
- 8 Sample results
Kernel Function Trace (KFT) is a kernel function tracing system, which uses the "-finstrument-functions" capability of the gcc compiler to add instrumentation callouts to every function entry and exit. The KFT system provides for capturing these callouts and generating a trace of events, with timing details. KFT is excellent at providing a good timing overview of kernel procedures, allowing you to see where time is spent in functions and sub-routines in the kernel.
The main mode of operation with KFT is to use the system with a dynamic trace configuration. That is, you can set a trace configuration after kernel
startup, using the
/proc/kft interface, and retrieve trace data immediately. However, another (special) mode of operation is available, called STATIC_RUN
mode, where the configuration for a KFT run is configured and compiled statically into the kernel. This mode is useful for getting a trace of kernel
operation during system bootup (before user space is running).
The KFT configuration lets you specify how to automatically start and stop a trace, whether to include interrupts as part of the trace, and whether to
filter the trace data by various criteria (for minimum function duration, only certain listed functions, etc.) KFT trace data is retrieved by reading from
/proc/kft_data after the trace is complete.
Tools are supplied to convert numeric trace data to kernel symbols, and to process and analyze the data in a KFT trace.
Documentation for KFT is available (as of 2.6.12) in Documentation/kft.txt, after applying the kft-all-in-one.patch.
Here's a presentation about KFT usage:
- Presentation: Learning the Kernel and Finding Performance Problems with KFI
- Sample trace used with presentation: omap-serial_init.trace.txt
KFT used to be called KFI (for Kernel Function Instrumentation). For prior releases of KFT, see KernelFunctionInstrumentation
- Download directory with recent versions: ftp://dslab.lzu.edu.cn/pub/kft/
- This is a fairly slow link - you can download the patch for 2.6.21 here: kft-all-in-one-2.6.21.patch
- Patches for Linux 220.127.116.11, 2.6.11 and 2.6.12: see the Patch Archive page (available as an all-in-one patch or a tar archive of broken-out patches)
- Patch for Linux 2.6.11: (can just download kfi-2.patch)
- Patch for Linux 2.6.7 (for x86 only): kfi-26-test1.patch
- Patch for CELF kernel (based on linux-2.4.20): kfi-24-test4.patch
KFT includes several helper scripts which are located in the kernel
- addr2sym - convert function addresses to symbols in the trace data
- kd - KFT dump - does filtering, sorting, analysis and trace formatting of KFT trace logs
- mkkftrun.pl - used during building the kernel to convert a configuration file into a C file to be compiled into the kernel
- sym2addr - convert function names to addresses in a KFT configuration file (for a dynamic trace)
See Documentation/kft.txt, in the kernel source tree after applying the patch, for instructions on using these programs.
How To Use
- download both the patch
- apply the patch in the kernel top-level directory:
- patch -p1 <kft.patch
- read the rest of the instructions in the Documentation/kft.txt file. (my apologies for being lazy!)
Adding platform support for the kft clock source
The current patch (from Sep 2005), uses sched_clock() as the clock source for kft_readclock(). sched_clock() is new in the 2.6 kernel, and returns a 64-bit value containing nanoseconds (not necessarily relative to any particular time base, but assumed to be monotonically increasing, and relatively frequency-stable.)
If your platform has good support for sched_clock(), then KFT should work for you unmodified. If not, you may wish to do one of two things:
- improve support for sched_clock() in your board port, or
- write a custom kft_readclock() routine.
A "good" sched_clock() routine will provide at least microsecond resolution on return values. Some architectures have sched_clock() returning values based on the
which on many embedded platforms only has resolution to 10 milliseconds.
There are some sample custom kft_readclock() routines in the current patch for different architectures.
Here is a list of things that need more work:
- may need to add noinstrument attributes for some time-critical code (need to check this)
- maybe check "Function Trace in KDB" patch for help with this
Mitsubishi measured the overhead of KFI (the predecessor to KFT). The period is from start_kernel() to smp_init().
Platform was: SH7751R 240MHz (Memory Clock 80MHz)
With KFI : 922.419 msec
Without KFI : 666.982 msec
Overhead : 27.69%
There are other technologies for doing call traces or kernel profiling that are similar to KFT. Some of these are mentioned on the Kernel Instrumentation page. One that is very similar is a kernel trace mechanism for use with KDB. A patch was posted to LKML in January of 2002. See the message: http://www.uwsg.iu.edu/hypermail/linux/kernel/0201.3/0888.html
Tim asked the question:
Q. Is there a way to adjust the trigger or filters to reduce the memory usage?
A. The memory usage is determined by the size of the log, which is specified by
logentries in the KFT configuration. If
logentries is not specified, it defaults to a rather large number (20,000 in the current code). To use a smaller trace log, specify a smaller number of logentries in the KFT configuration.
The use of triggers and filters can help you fit more data (or more pertinent data) into the log, so you can more readily see the information you are interested in.
By setting start and stop triggers with a narrower "range" of operation, then the amount of data put into the log will be more limited. For example, the default configuration for a static trace uses
trigger start entry start_kernel
trigger stop entry to_userspace
This will trace EVERYTHING that the kernel does between those two routines. However, you can limit tracing to a much smaller time area of kernel initialization using better triggers. Here is an example showing a triggers for just watching mem_init():
trigger start entry mem_init
trigger stop exit mem_init
Filters are also vital to reduce the number of entries the trace log. With no time filters in place, KFT will log every single function executed by the kernel. This will quickly overrun the log (no matter what size you have reserved with
When using KFT to find long-duration functions in the kernel, we usually are not interested in routines that execute quickly, and instead use something like "filter mintime 500" to filter out routines taking less than 500 microseconds.
Here is an excerpt from a KFI log trace (processed with addr2sym). It shows all functions which lasted longer than 500 microseconds, from when the kernel entered start_kernel() to when it entered to_userspace().
kft log output (excerpt)
Kernel Instrumentation Run ID 0
Logging started at 6785045 usec by entry to function start_kernel Logging stopped at 8423650 usec by entry to function to_userspace
Filters: 500 usecs minimum execution time
Execution time filter count = 896348 Total entries filtered = 896348 Entries not found = 24
Number of entries after filters = 1757
Entry Delta PID Function Called At 1 0 0 start_kernel L6+0x0 14 8687 0 setup_arch start_kernel+0x35 39 891 0 setup_memory setup_arch+0x2a8 53 872 0 register_bootmem_low_pages setup_memory+0x8f 54 871 0 free_bootmem register_bootmem_low_pages+0x95 54 871 0 free_bootmem_core free_bootmem+0x34 930 7432 0 paging_init setup_arch+0x2af 935 7427 0 zone_sizes_init paging_init+0x4e 935 7427 0 free_area_init zone_sizes_init+0x83 935 7427 0 free_area_init_node free_area_init+0x4b 935 3759 0 __alloc_bootmem_node free_area_init_node+0xc5 935 3759 0 __alloc_bootmem_core __alloc_bootmem_node+0x43 4694 3668 0 free_area_init_core free_area_init_node+0x75 4817 3535 0 memmap_init_zone free_area_init_core+0x2bd 8807 266911 0 time_init start_kernel+0xb6 8807 261404 0 get_cmos_time time_init+0x1c 270211 5507 0 select_timer time_init+0x41 270211 5507 0 init_tsc select_timer+0x45 270211 5507 0 calibrate_tsc init_tsc+0x6c 275718 1638 0 console_init start_kernel+0xbb 275718 1638 0 con_init console_init+0x59 275954 733 0 vgacon_save_screen con_init+0x288 277376 6730 0 mem_init start_kernel+0xf8 277376 1691 0 free_all_bootmem mem_init+0x52 277376 1691 0 free_all_bootmem_core free_all_bootmem+0x24 284118 25027 0 calibrate_delay start_kernel+0x10f 293860 770 0 __delay calibrate_delay+0x62 293860 770 0 delay_tsc __delay+0x26 294951 1534 0 __delay calibrate_delay+0x62 294951 1534 0 delay_tsc __delay+0x26 297134 1149 0 __delay calibrate_delay+0xbe 297134 1149 0 delay_tsc __delay+0x26 . . . 1638605 0 145 filemap_nopage do_no_page+0xef 1638605 0 145 __lock_page filemap_nopage+0x286 1638605 0 145 io_schedule __lock_page+0x95 1638605 0 145 schedule io_schedule+0x24 1638605 0 5 schedule worker_thread+0x217 1638605 0 1 to_userspace init+0xa6
The log is attached here: kfiboot-9.lst A Delta value of 0 usually means the exit from the routine was not seen.
kft log analysis with 'kd'
Below is a
kd dump of the data from the above log.
For the purpose of finding areas of big time in the kernel, the functions with high "Local" time are important. For example,
delay_tsc() is called 156 times, resulting in 619 milliseconds of duration. Other time-consuming routines were:
The top line showing schedule() called 192 times and lasting over 5 seconds, is accounted wrong due to the switch in execution control inside the schedule routine. (The count of 192 calls is correct, but the duration is wrong.)
$ ~/work/kft/kft/kd -n 30 kftboot-9.lst
kft nested call trace with 'kd -c'
Below is a
kd -c trace of the data from a log taken from a PPC440g platform, from a (dynamic) trace of the function do_fork().
Here is the configuration file that was used:
trigger start entry do_fork
trigger stop exit do_fork
Here is the first part of the trace in nested call format:
Times (Entry, Duration and Local) are in micro-seconds. Note the timer interrupt during the routine.
Entry Duration Local Pid Trace 4 20428 209 33 do_fork 7 6 6 33 | alloc_pidmap 18 2643 84 33 | copy_process 21 114 19 33 | | dup_task_struct 24 8 6 33 | | | prepare_to_copy 27 2 2 33 | | | | sub_preempt_count 35 22 9 33 | | | kmem_cache_alloc 38 2 2 33 | | | | __might_sleep 43 11 9 33 | | | | cache_alloc_refill 49 2 2 33 | | | | | sub_preempt_count 60 65 6 33 | | | __get_free_pages 63 59 14 33 | | | | __alloc_pages 65 3 3 33 | | | | | __might_sleep 71 3 3 33 | | | | | zone_watermark_ok 77 37 17 33 | | | | | buffered_rmqueue 80 4 4 33 | | | | | | __rmqueue 86 3 3 33 | | | | | | sub_preempt_count 92 3 3 33 | | | | | | bad_range 98 2 2 33 | | | | | | __mod_page_state 103 8 5 33 | | | | | | prep_new_page 106 3 3 33 | | | | | | | set_page_refs 117 2 2 33 | | | | | zone_statistics 141 25 4 33 | | do_posix_clock_monotonic_gettime 143 21 6 33 | | | do_posix_clock_monotonic_get 146 15 6 33 | | | | do_posix_clock_monotonic_gettime_parts 149 9 6 33 | | | | | getnstimeofday 152 3 3 33 | | | | | | do_gettimeofday 169 3 3 33 | | copy_semundo 174 41 17 33 | | copy_files 177 19 9 33 | | | kmem_cache_alloc 180 2 2 33 | | | | __might_sleep 185 8 5 33 | | | | cache_alloc_refill 188 3 3 33 | | | | | sub_preempt_count 200 3 3 33 | | | count_open_files 209 2 2 33 | | | sub_preempt_count 218 19 8 33 | | kmem_cache_alloc 220 2 2 33 | | | __might_sleep 225 9 6 33 | | | cache_alloc_refill 229 3 3 33 | | | | sub_preempt_count 241 2 2 33 | | sub_preempt_count 246 216 9 33 | | kmem_cache_alloc 249 199 199 33 | | | __might_sleep !!!! start 253 151 63 33 timer_interrupt 256 8 6 -1 ! profile_tick 259 2 2 -1 ! ! profile_hit 267 61 15 -1 ! update_process_times 270 8 5 -1 ! ! account_system_time 273 3 3 -1 ! ! ! update_mem_hiwater 281 8 5 -1 ! ! run_local_timers 284 3 3 -1 ! ! ! raise_softirq 293 27 16 -1 ! ! scheduler_tick
. . .
To see the full trace, go to the KftDoForkTrace page.