Last edited 2 years ago

Ftrace

1. Article purpose[edit source]

This article provides the basic information needed to start using the Linux® kernel tool: ftrace[1].

2. Introduction[edit source]

The following table provides a brief description of the tool, as well as its availability depending on the software packages:

Yes: this tool is either present (ready to use or to be activated), or can be integrated and activated on the software package.

No: this tool is not present and cannot be integrated, or it is present but cannot be activated on the software package.

Tool STM32MPU Embedded Software distribution STM32MPU Embedded Softwaredistribution for Android™
Warning white.png Warning
The STM32MPU distribution for Android™ is not yet available in the v3 ecosystem releases: refer to the Ftrace page for the v2 ecosystem releases (in archived wiki).
Name Category Purpose Starter Package Developer Package Distribution Package Starter Package Developer Package Distribution Package
ftrace Tracing tools ftrace[1] (Function Tracer) is a powerful kernel tracing utility that is able, for instance, to trace every kernel function calls and kernel events without adding any extra code in your kernel source code No No Yes No No Yes

Note: Before Linux kernel 4.1, all the ftrace tracing control files were within the debugfs file system, which is typically located at /sys/kernel/debug/tracing. Now, it is located in /sys/kernel/tracing, and independent from debugfs.

For backward compatibility, when mounting the debugfs file system, the tracefs file system is automatically mounted at: /sys/kernel/debug/tracing.

All files located in the tracefs file system are located in that debugfs file system directory as well.

Please note that all functions present in the symbol table are available for ftrace. To know if a function is available in the symbol list, you can use the command "nm vmlinux | grep <function_name>"

3. Installing the trace and debug tool on your target board[edit source]

ftrace is a kernel feature which is not activated by default in the OpenSTLinux distributions as there is an impact on the Linux kernel size (around 1.5% increase of vmlinux), and also an impact on the overall performance, because of an additional treatment done to the trace kernel events and function calls.

In order to use 'Kernel Function Tracer' required for ftrace, the Linux kernel configuration must activate CONFIG_FUNCTION_TRACER and CONFIG_FUNCTION_GRAPH_TRACER using the Linux kernel menuconfig tool:

Symbol: FUNCTION_TRACER
Location:
  Kernel Hacking --->
    Tracers -->
      [*] Kernel Function Tracer

Symbol: FUNCTION_GRAPH_TRACER
Location:
  Kernel Hacking --->
    Tracers -->
      [*] Kernel Function Tracer
      [*] Kernel Function Graph Tracer

3.1. Using the STM32MPU Embedded Software distribution[edit source]

3.1.1. Developer Package[edit source]

It is not recommended to enable the ftrace kernel configuration by using the Developer Package, as all external modules should be also recompiled (e.g. gcnano driver for GPU STM32MP1), and this is not possible with the Developer Package, which does not necessary provide all the sources.

That is the reason why this is set as not supported for Developer Package.

3.1.2. Distribution Package[edit source]

  • Enable the required Linux kernel configuration
To enable CONFIG_FUNCTION_TRACER and CONFIG_FUNCTION_GRAPH_TRACER in the Linux kernel configuration, please refer to Menuconfig or how to configure kernel article to get instructions for modifying the configuration and recompiling the Linux kernel image in the Distribution Package context.
  • You must also recompile the external Linux kernel module(s) (if existing) being not part of the Linux kernel source tree.
Example for gcnano driver of GPU STM32MP1:
 bitbake gcnano-driver-stm32mp
  • Re-build the full OpenSTLinux image, in order to recompile all dependencies and have correct rootfs including the external Linux kernel modules
 bitbake st-image-weston
Warning white.png Warning
As explained before, the size of the uncompressed Linux kernel image increases when enabling the ftrace configuration.

Depending of the memory configuration of your target board (defined in the device tree), an increase of your kernel image can overlap some reserved regions placed after.
In that case you have a compilation error highlighted.
If this overlap has a minor impact (meaning that some features are no more functional but being not critical), this is possible to bypass the compilation error by activating the Linux kernel configuration CONFIG_SECTION_MISMATCH_WARN_ONLY using the Linux kernel Menuconfig tool (Menuconfig or how to configure kernel)

Symbol: SECTION_MISMATCH_WARN_ONLY
Location:
  Kernel Hacking --->
    Compile-time checks and compiler options -->
      [*] Make section mismatch errors non-fatal

3.2. Using the STM32MPU Embedded Software distribution for Android™[edit source]

Warning white.png Warning
The STM32MPU distribution for Android™ is not yet available in the v3 ecosystem releases: refer to the Ftrace page for the v2 ecosystem releases (in archived wiki).


4. Getting started[edit source]

4.1. Using ftrace at runtime[edit source]

First of all, you need to enable/activate the ftrace feature from target.

  • Once target booted and logged, mount tracefs:
 mount -t tracefs nodev /sys/kernel/tracing
  • At that step, all the 'ftrace' features take place in the file system directory path /sys/kernel/tracing.

To find out which tracers are available, simply cat the available_tracers file in the tracing directory:

 cat /sys/kernel/tracing/available_tracers
 function_graph function nop

More tracers can be added by kernel build configurations. Please refer to To go further section.

4.2. Filter option[edit source]

Warning white.png Warning
ftrace uses a function / graph_function filter, not a driver filter. As a consequence, tracing *myDriver* functions will not trace the myHelper function from myDriver

You can get the list of available filter function with the following command:

 cat /sys/kernel/tracing/available_filter_functions

4.2.1. Function tracer mode[edit source]

  • Start a tracing session
 echo 1 > /sys/kernel/tracing/tracing_on
  • To enable the function tracer, just write function to the current_tracer file. You can then verify the current value:
 echo function > /sys/kernel/tracing/current_tracer
 cat /sys/kernel/tracing/current_tracer
  function
 cat /sys/kernel/tracing/trace | head -20
 # tracer: function
 #
 # entries-in-buffer/entries-written: 144045/33695515   #P:2
 #
 #                              _-----=> irqs-off
 #                             / _----=> need-resched
 #                            | / _---=> hardirq/softirq
 #                            || / _--=> preempt-depth
 #                            ||| /     delay
 #           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
 #              | |       |   ||||       |         |
             date-3591  [001] ...3  3278.796042: memblock_is_map_memory <-pfn_valid
             date-3591  [001] ...3  3278.796046: unlock_page <-filemap_map_pages
             date-3591  [001] ...3  3278.796051: alloc_set_pte <-filemap_map_pages
             date-3591  [001] ...3  3278.796053: add_mm_counter_fast <-alloc_set_pte
             date-3591  [001] ...3  3278.796055: page_add_file_rmap <-alloc_set_pte
             date-3591  [001] ...3  3278.796057: __sync_icache_dcache <-alloc_set_pte
             date-3591  [001] ...3  3278.796059: pfn_valid <-__sync_icache_dcache
             date-3591  [001] ...3  3278.796061: memblock_is_map_memory <-pfn_valid
             date-3591  [001] ...3  3278.796064: unlock_page <-filemap_map_pages 
  • To apply function(s) filter you can set value(s) in /sys/kernel/tracing/set_ftrace_filter, then check for new content of the trace:
# Here we take the example with all uart functions
 echo "*uart*" > /sys/kernel/tracing/set_ftrace_filter

# Clean the existing trace
 echo > /sys/kernel/tracing/trace

# Display new trace content (in that case, please do some actions in the console to get some traces)
 cat /sys/kernel/tracing/trace | head -20
# tracer: function
#
#                              _-----=> irqs-off
#                             / _----=> need-resched
#                            | / _---=> hardirq/softirq
#                            || / _--=> preempt-depth
#                            ||| /     delay
#           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
#              | |       |   ||||       |         |
              sh-343   [000] ....  9313.041827: uart_ioctl <-tty_ioctl
              sh-343   [000] ....  9313.041855: uart_ioctl <-tty_ioctl
              sh-343   [000] ....  9313.041866: uart_chars_in_buffer <-tty_wait_until_sent
              sh-343   [000] ....  9313.041870: uart_wait_until_sent <-tty_wait_until_sent
              sh-343   [000] ....  9313.041875: uart_set_termios <-tty_set_termios
              sh-343   [000] ....  9313.041968: uart_write_room <-tty_write_room
              sh-343   [000] ....  9313.041974: uart_write <-n_tty_write
              sh-343   [000] d..1  9313.041979: __uart_start <-uart_write
              sh-343   [000] d..1  9313.041987: uart_write_wakeup <-stm32_transmit_chars
              sh-343   [000] d.h2  9313.042007: uart_write_wakeup <-stm32_transmit_chars
              sh-343   [000] d.h2  9313.042022: uart_write_wakeup <-stm32_transmit_chars
More information about filtering option and configuration is available in the Linux documentation for ftrace[2].
  • To clear out the filter so that all functions are recorded again:
 echo > /sys/kernel/tracing/set_ftrace_filter

4.2.2. Graph function tracer mode[edit source]

Start a tracing session:

 echo 1 > /sys/kernel/tracing/tracing_on

To enable the function tracer, just write function_graph into the current_tracer file. You can then verify the current value:

 echo function_graph > /sys/kernel/tracing/current_tracer
 cat /sys/kernel/tracing/current_tracer
  function_graph
 cat /sys/kernel/tracing/trace | head -20
 # tracer: function_graph
 #
 # CPU  DURATION                  FUNCTION CALLS
 # |     |   |                     |   |   |   |
  1)   1.015 us    |        _spin_lock_irqsave();
  1)   0.476 us    |        internal_add_timer();
  1)   0.423 us    |        wake_up_idle_cpu();
  1)   0.461 us    |        _spin_unlock_irqrestore();
  1)   4.770 us    |      }
  1)   5.725 us    |    }
  1)   0.450 us    |    mutex_unlock();
  1) + 24.243 us   |  }
  1)   0.483 us    |  _spin_lock_irq();
  1)   0.517 us    |  _spin_unlock_irq();
  1)               |  prepare_to_wait() {
  1)   0.468 us    |    _spin_lock_irqsave();
  1)   0.502 us    |    _spin_unlock_irqrestore();
  1)   2.411 us    |  }
  1)   0.449 us    |  kthread_should_stop();
  1)               |  schedule() {

To apply a graph function(s) filter, you can set value(s) in /sys/kernel/tracing/set_graph_function, then check for the new content of the trace:

# Here we take the example with all uart functions
 echo "*uart*" > /sys/kernel/tracing/set_graph_function

# Clean the existing trace
 echo > /sys/kernel/tracing/trace
# Display the new trace content (in that case, please do some action in the console to get some traces)
 cat /sys/kernel/tracing/trace | head -20
# tracer: function_graph
#
# CPU  DURATION                  FUNCTION CALLS
# |     |   |                     |   |   |   |
 1)               |  uart_ioctl() {
 1)   0.875 us    |    mutex_lock();
 1)   0.792 us    |    mutex_unlock();
 1) + 15.542 us   |  }
 1)               |  uart_ioctl() {
 1)   0.583 us    |    mutex_lock();
 1)   0.584 us    |    mutex_unlock();
 1)   9.792 us    |  }
 1)               |  uart_chars_in_buffer() {
 1)               |    _raw_spin_lock_irqsave() {
 1)   0.667 us    |      preempt_count_add();
 1)   5.458 us    |    }
 1)               |    _raw_spin_unlock_irqrestore() {
 1)   0.583 us    |      preempt_count_sub();
 1)   5.000 us    |    }
 1) + 19.459 us   |  }
 1)   1.541 us    |  uart_wait_until_sent();
 1)               |  uart_set_termios() {
 1)   0.583 us    |    mutex_lock();
 1)   0.583 us    |    mutex_unlock();
 1) + 10.291 us   |  }
 1)               |  uart_write_room() {
 1)               |    _raw_spin_lock_irqsave() {
 1)   0.666 us    |      preempt_count_add();
 1)   5.333 us    |    }
 1)               |    _raw_spin_unlock_irqrestore() {
 1)   0.583 us    |      preempt_count_sub();
 1)   5.000 us    |    }
 1) + 19.625 us   |  }
 1)               |  uart_write() {
 1)               |    _raw_spin_lock_irqsave() {
 1)   0.625 us    |      preempt_count_add();
 1)   5.209 us    |    }
 1)               |    __uart_start() {
 1)               |      stm32_start_tx() {
 1)               |        stm32_transmit_chars() {
More information about filtering option and configuration on the Linux documentation for ftrace[2].
  • To clear out this special filter so that all functions will be recorded again:
 echo > /sys/kernel/tracing/set_graph_function

4.3. Buffer size[edit source]

A buffer is allocated for each CPU. For making a trace analysis you can change this buffer size (increase or decrease).

  • This is possible to read the given size value per CPU, or the total (value is given in kilobytes):
# Per CPU
 cat /sys/kernel/tracing/buffer_size_kb
1411
or
 cat /sys/kernel/tracing/per_cpu/cpuX/buffer_size_kb
1411
# Total for all CPUs: combined size of all the trace buffers
 cat /sys/kernel/tracing/buffer_total_size_kb
2822
  • To change the value (note that the trace buffers are allocated in pages (blocks of memory that the kernel uses for allocation, usually 4 Kbytes in size))
# Same value for each CPU (here 1000*4096/1024=4000)
 echo 4000 > /sys/kernel/tracing/buffer_size_kb
or
# Change buffer size value for a specific CPU X (here 1000*4096/1024=4000)
 echo 4000 > /sys/kernel/tracing/per_cpu/cpuX/buffer_size_kb

4.4. Using ftrace at boot time[edit source]

You can use ftrace from the kernel boot, which can be very useful to debug the boot issues.

For this, you have to use the kernel command-line parameters:
- ftrace and also ftrace_filter or ftrace_graph_filter if you want to add filter.

4.4.1. With STM32MPU Embedded Software package[edit source]

For instance, to modify the kernel bootargs you can do it in the following ways:

  • Mount a boot partition from the Linux kernel console, and then update the extlinux.conf file using the vi editor (see man page[3], or introduction page[4] ). In example:
 mount /dev/mmcblk0p4 /boot
# As example for SDCard boot on STM32MP15 Evaluation board, otherwise /boot/<bootdevice>_<platform>-<boardId>_extlinux/extlinux.conf
 vi /boot/mmc0_stm32mp157c-ev1_extlinux/extlinux.conf

Update kernel command-line by adding ftrace parameter:

- function tracer mode
root=/dev/mmcblk0p5 rootwait rw console=ttyS3,115200 ftrace=function ftrace_filter=*uart*

- function_graph tracer mode
root=/dev/mmcblk0p5 rootwait rw console=ttyS3,115200 ftrace=function_graph ftrace_graph_filter=*uart*

Save and quit file update, and then reboot the board

or

  • Edit the extlinux.conf file from the microSD™ card (if used as boot device)
Warning white.png Warning
Admin rights required
- Insert microSD card on host PC
- Check for boot partition mounted (i.e /media/$USER/bootfs)
- Edit the extlinux file corresponding to your setup (i.e /media/$USER/bootfs/mmc0_stm32mp157c-ev1_extlinux/extlinux.conf)
- Modify the command-line following your ftrace tracer configuration required (see above)
- Save modification, then insert the microSD card on your target
- Boot and check for kernel command-line

4.4.2. With STM32MPU Embedded Software package for Android[edit source]

Warning white.png Warning
The STM32MPU distribution for Android™ is not yet available in the v3 ecosystem releases: refer to the Ftrace page for the v2 ecosystem releases (in archived wiki).



4.4.3. Checking for trace[edit source]

When booted, to check for the trace, you have to mount the tracefs:

 mount -t tracefs nodev /sys/kernel/tracing

Then look at for trace content (i.e. for function trace):

 cat /sys/kernel/tracing/trace | head -20
# tracer: function
#
#                              _-----=> irqs-off
#                             / _----=> need-resched
#                            | / _---=> hardirq/softirq
#                            || / _--=> preempt-depth
#                            ||| /     delay
#           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION
#              | |       |   ||||       |         |
         systemd-1     [000] ....     1.087213: uart_register_driver <-usart_init
         systemd-1     [000] ....     1.087847: uart_get_rs485_mode <-stm32_serial_probe
         systemd-1     [000] ....     1.088436: uart_add_one_port <-stm32_serial_probe
         systemd-1     [000] ....     1.098000: uart_parse_options <-stm32_console_setup
         systemd-1     [000] ....     1.098007: uart_set_options <-stm32_console_setup
         systemd-1     [000] ....     1.098014: uart_get_baud_rate <-stm32_set_termios
         systemd-1     [000] d..1     1.098019: uart_update_timeout <-stm32_set_termios
         systemd-1     [000] d..1     1.098090: uart_console_write <-stm32_console_write
         systemd-1     [000] d..1     1.105231: uart_console_write <-stm32_console_write
         systemd-1     [000] d..1     1.114697: uart_console_write <-stm32_console_write
         systemd-1     [000] d..1     1.120300: uart_console_write <-stm32_console_write

4.4.4. Capturing an oops (from startup) to the serial console[edit source]

An interesting application to enable ftrace at boot, is to capture the function calls leading up to a panic by placing the following parameters on the kernel command line:

root=/dev/mmcblk0p5 rootwait rw console=ttyS3,115200 ftrace=function ftrace_dump_on_oops

When the oops occurs, the ftrace buffer will be automatically dumped on the console message.

4.5. Erasing trace[edit source]

This is possible to erase content of trace for ftrace with the following command:

 echo > /sys/kernel/tracing/trace

5. To go further[edit source]

5.1. Adding print information for ftrace[edit source]

In addition to the printing Linux kernel functions, it is possible to trace the specific debug information with ftrace by using trace_printk function.

It can be used just like printk(), and can also be used in any context (interrupt code, NMI code, and scheduler code).

trace_printk does not output to the console, but writes to the ftrace ring buffer and can be read via the trace file.

To use trace_printk function, you have to include linux/ftrace.h in your source code:

...
#include <linux/ftrace.h>
...

Then use trace_printk syntax as printk (see below example):

...
trace_printk("%s: %d uart_tx_stopped(port) %i\n", __FUNCTION__, __LINE__, uart_tx_stopped(port));
...

5.2. Stack Trace[edit source]

Extracted from Kernel documentation for ftrace[2].

Since the kernel has a fixed sized stack, it is important to not waste it in functions. A kernel developer must be aware of what the functions allocate on the stack. If they add too much size, the system can be in danger of a stack overflow, and a corruption will occur, usually leading to a system panic.

There are some tools that check this, usually with interrupts periodically checking the usage. But if you can perform a check at every function call that will become very useful. As ftrace provides a function tracer, it makes it convenient to check the stack size at every function call. This is enabled via the stack tracer.

The Linux kernel configuration option CONFIG_STACK_TRACER enables the ftrace stack tracing functionality.

Symbol: STACK_TRACER
Location:
  Kernel Hacking --->
    Tracers -->
      [*] Trace max stack

To enable it, write a '1' into /proc/sys/kernel/stack_tracer_enabled.

 echo 1 > /proc/sys/kernel/stack_tracer_enabled

You can also enable it from the kernel command line to trace the stack size of the kernel during boot up, by adding "stacktrace" to the kernel command line parameter.

root=/dev/mmcblk0p5 rootwait rw console=ttyS3,115200 stacktrace

When booted, to check for the trace, you have to mount first the tracefs, then display the trace content:

 mount -t tracefs nodev /sys/kernel/tracing

 cat /sys/kernel/tracing/stack_max_size
2928

 cat /sys/kernel/tracing/stack_trace
        Depth    Size   Location    (82 entries)
       -----    ----   --------
 0)     4328       4   __rcu_read_unlock+0x14/0x68
 1)     4324     180   select_task_rq_fair+0x8ac/0xb7c
 2)     4144      64   try_to_wake_up+0x100/0x3fc
 3)     4080      16   wake_up_process+0x20/0x24
 4)     4064      24   swake_up_locked.part.0+0x20/0x38
 5)     4040      24   swake_up+0x38/0x48
 6)     4016      16   rcu_gp_kthread_wake+0x4c/0x50
 7)     4000      24   rcu_report_qs_rsp+0x50/0x84
 8)     3976     120   rcu_report_qs_rnp+0x258/0x2ec
 9)     3856      80   rcu_process_callbacks+0x290/0x43c
10)     3776      96   __do_softirq+0x12c/0x3ec
11)     3680      16   irq_exit+0xd0/0x118
12)     3664      48   __handle_domain_irq+0x90/0xfc
13)     3616      40   gic_handle_irq+0x5c/0xa0
14)     3576      68   __irq_svc+0x6c/0xa8
15)     3508      28   unwind_get_byte+0x20/0x74
16)     3480     160   unwind_frame+0x1a8/0x6b0
17)     3320      32   walk_stackframe+0x34/0x40
18)     3288      56   __save_stack_trace+0xa4/0xa8
19)     3232      16   save_stack_trace+0x30/0x34
20)     3216      72   create_object+0x120/0x278
21)     3144      40   kmemleak_alloc+0x8c/0xd4
22)     3104      64   kmem_cache_alloc+0x184/0x2f0
23)     3040      64   __kernfs_new_node+0x58/0x15c
24)     2976      24   kernfs_new_node+0x2c/0x48
25)     2952      24   __kernfs_create_file+0x28/0xb8
26)     2928      56   sysfs_add_file_mode_ns+0xc4/0x1a0
27)     2872      24   sysfs_create_file_ns+0x4c/0x58
28)     2848      56   kobject_add_internal+0x174/0x358
29)     2792      40   kobject_add+0x50/0x98
30)     2752      32   irq_sysfs_add+0x44/0x60
31)     2720      72   __irq_alloc_descs+0x174/0x234
32)     2648      48   irq_domain_alloc_descs+0x64/0xe4
33)     2600      56   irq_create_mapping+0x108/0x1fc
34)     2544      56   irq_create_fwspec_mapping+0x140/0x318
35)     2488      88   irq_create_of_mapping+0x5c/0x64
36)     2400     168   of_irq_get+0x68/0x78
37)     2232      24   stpmu1_regulator_parse_dt+0x68/0x80
38)     2208      96   regulator_register+0x218/0x970
39)     2112      32   devm_regulator_register+0x54/0x84
40)     2080     136   stpmu1_regulator_probe+0x350/0x5f4
41)     1944      32   platform_drv_probe+0x60/0xbc
42)     1912      64   driver_probe_device+0x2f4/0x488
43)     1848      32   __device_attach_driver+0xac/0x14c
44)     1816      40   bus_for_each_drv+0x54/0xa4
45)     1776      40   __device_attach+0xc0/0x150
46)     1736      16   device_initial_probe+0x1c/0x20
47)     1720      32   bus_probe_device+0x94/0x9c
48)     1688      64   device_add+0x3c0/0x5d0
49)     1624      16   of_device_add+0x44/0x4c
50)     1608      40   of_platform_device_create_pdata+0x84/0xb4
51)     1568     104   of_platform_bus_create+0x160/0x2f8
52)     1464      56   of_platform_populate+0x9c/0x134
53)     1408      32   stpmu1_probe+0x6c/0xac
54)     1376      40   i2c_device_probe+0x290/0x2dc
55)     1336      64   driver_probe_device+0x2f4/0x488
56)     1272      32   __device_attach_driver+0xac/0x14c
57)     1240      40   bus_for_each_drv+0x54/0xa4
58)     1200      40   __device_attach+0xc0/0x150
59)     1160      16   device_initial_probe+0x1c/0x20
60)     1144      32   bus_probe_device+0x94/0x9c
61)     1112      64   device_add+0x3c0/0x5d0
62)     1048      24   device_register+0x24/0x28
63)     1024      48   i2c_new_device+0x14c/0x2f4
64)      976      96   of_i2c_register_device+0x134/0x1dc
65)      880      40   of_i2c_register_devices+0x8c/0x100
66)      840      48   i2c_register_adapter+0x184/0x404
67)      792      48   i2c_add_adapter+0xa4/0x138
68)      744     160   stm32f7_i2c_probe+0x954/0xd08
69)      584      32   platform_drv_probe+0x60/0xbc
70)      552      64   driver_probe_device+0x2f4/0x488
71)      488      32   __driver_attach+0x110/0x12c
72)      456      40   bus_for_each_dev+0x5c/0xac
73)      416      16   driver_attach+0x2c/0x30
74)      400      48   bus_add_driver+0x1d0/0x274
75)      352      24   driver_register+0x88/0x104
76)      328      16   __platform_driver_register+0x50/0x58
77)      312      16   stm32f7_i2c_driver_init+0x24/0x28
78)      296     112   do_one_initcall+0x54/0x178
79)      184      72   kernel_init_freeable+0x1dc/0x274
80)      112      24   kernel_init+0x18/0x124
81)       88      88   ret_from_fork+0x14/0x24

5.3. More tracers[edit source]

More tracers are available for ftrace. Please refer to the Linux kernel documentation for ftrace[2].

Tracer name Descriptipn
blk Block tracer. The tracer used by the blktrace user application
hwlat Hardware Latency tracer. It is used to detect if the hardware produces any latency
irqsoff Traces the areas that disable interrupts and saves the trace with the longest max latency
preemptoff Similar to irqsoff but traces and records the amount of time for which preemption is disabled
preemptirqsoff Similar to irqsoff and preemptoff, but traces and records the largest time for which irqs and/or preemption is disabled
wakeup Traces and records the max latency that it takes for the highest priority task to get scheduled after it has been woken up
wakeup_rt Traces and records the max latency that it takes for just RT tasks (as the current "wakeup" does)
wakeup_dl Traces and records the max latency that it takes for a SCHED_DEADLINE task to be woken (as the "wakeup" and "wakeup_rt" does)
mmiotrace A special tracer that is used to trace binary module. It traces all the calls that a module makes to the hardware
branch This tracer can be configured when tracing likely/unlikely calls within the kernel
nop This is the "trace nothing" tracer

5.4. Complementary tools[edit source]

While the debugfs interface is rather simple, it can also be awkward to work with. Some tools are proposed to simplify the user experience with ftrace.

trace-cmd is a command-line tool that works and interacts with ftrace instead of echoing commands into specific files and reading the result from another file. It proposes a high level user command interface to ease the ftrace usage.
Associated to trace-cmd, kernelshark tool proposes a trace viewer useful for analysing the traces.
LTTng is an open source tracing framework for Linux partially based on ftrace, which also proposes a high level user interface through command line. We associate Trace Compass on Host PC side to a log viewer.

6. References[edit source]


  • Useful external links
Document link Document Type Description
ftrace (kernel.org docs) Standard Documentation from Linux kernel sources
Debugging the kernel using Ftrace - part1 User Guide http://lwn.net
Debugging the kernel using Ftrace - part2 User Guide http://lwn.net
Using the TRACE_EVENT() macro (with CREATE_TRACE_POINTS) Training http://lwn.net
Load perf analysis using ftrace User Guide Linaro