GitHub - kumarak/racehound: Data race detector for Linux kernel modules

kumarak / racehound Public

forked from kmrov/racehound

Notifications You must be signed in to change notification settings
Fork 0
Star 0

Data race detector for Linux kernel modules

GPL-2.0 license

0 stars 6 forks Branches Tags Activity

Star

Notifications

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 82 Commits
cmake		cmake
core		core
AUTHORS		AUTHORS
CMakeLists.txt		CMakeLists.txt
LICENSE		LICENSE
README		README
cmake_uninstall.cmake.in		cmake_uninstall.cmake.in
config.h.in		config.h.in

Repository files navigation

RaceHound
--------------
RaceHound project provides the tools for detect data races in Linux
kernel modules (device drivers, file system modules,
etc.) in runtime.

Currently, RaceHound is intended for systems with x86 architecture, 
both 32- and 64-bit.

Project site: http://github.com/winnukem/racehound

Basic Usage
--------------

Use case 1.
 
Suppose you would like to find out if there is a race in the kernel module 
named "foo_driver" between the instruction at the offset 0xf00d in the 
function "bar".

Here we assume RaceHound is installed to the default location (/usr/local/...).

1. Unload foo_driver if it is already loaded.

2. If debugfs is not mounted, mount it:

    # mount -t debugfs none /sys/kernel/debug

3. Load RaceHound and specify the name of the target:
  
    # insmod race_hound.ko target_name=foo_driver

4. You may instruct RaceHound to monitor the instruction at bar+0xf00d
already (or you may do so after you load the target module).

    # echo 'bar+0xf00d' > /sys/kernel/debug/rhound/breakpoints

You can use RaceHound to monitor more than one instruction at a time, if you
like. Just echo their locations to the 'breakpoints' file one at a time, as
shown above.

If you would like RaceHound to stop monitoring the instruction, prepend '-'
to the appropriate string:

    # echo '-bar+0xf00d' > /sys/kernel/debug/rhound/breakpoints

'cat /sys/kernel/debug/rhound/breakpoints' will show you the current list of
monitored instructions.

5. Now load foo_driver and do something with it.

6. If RaceHound detects a race, it will output a message similar to the 
following one to the system log:

  [rh] Detected a data race on the memory at ffff8800377214c0 
  between the instruction at ffffffffa0392b6d 
    (e1000_clean+0x18d/0x950 [e1000], CPU=0, 
     task_struct=ffff8800b061dc40, comm: "kworker/0:1") 
  and the instruction right before ffffffffa0390b18 
    (e1000_xmit_frame+0x928/0x1020 [e1000], CPU=1, 
     task_struct=ffff88011a535c40, comm: "Socket Thread") 

If foo_driver is built with debug info, you can use addr2line or a similar 
tool to find the source lines corresponding to the conflicting instruction.

7. When you are through with foo_driver, unload it.

8. Unload RaceHound.