• Immutable Page
  • Info
  • Attachments

Diff for "OPWfirstpatch"

Differences between revisions 92 and 93

Deletions are marked like this. Additions are marked like this.
Line 32: Line 32:
= Tips = = Tips for Applying to be an OPW Kernel intern =

Hooray! Thanks for your interest in working on the Linux kernel. The next step is to apply to OPW, and use this tutorial to create your first patch to the Linux kernel.


If you run into any issues with this tutorial, please ask questions on the #kernel-opw IRC channel on irc.oftc.net, or on the opw-kernel mailing list.

Contribution Instructions


Instead, send your patch to the opw-kernel mailing list.


This tutorial will cover how to get your first patch submitted. You have three choices for how to complete this tutorial:

  1. Run Linux in VMPlayer from Windows.

  2. Run Linux natively on your own machine.

  3. Run Linux within VMPlayer on Linux.

We recommend running Linux natively. Most Linux kernel developers run Linux natively, so you may as well get used to it. :) If you want to run Linux in VMPlayer, follow these directions. Note, you will not be able to compile the Linux kernel on a Mac, because the filesystem defaults to case-insensitive.

This tutorial assumes you are running Ubuntu or Debian. If you are running Fedora, Suse, Arch, or Gentoo, the package installation commands or package names may be slightly different. Ask for help on #kernel-newbies on irc.oftc.net if you get stuck. If (and only if) you are an applicant for the FOSS Outreach Program for Women (OPW), then you should ask for help on #kernel-opw on irc.oftc.net.


This tutorial will show you how to:

  1. Intro
  2. Setup your tools
  3. Explore the kernel tree
  4. Play with some git basics
  5. Update your kernel
  6. Make a driver change
  7. Start creating your first patch
  8. Understand patch best practices
  9. Find a driver to clean up
  10. Committing your changes
  11. Submit a patch
  12. Send your patch to your mentors
  13. Responding to emails
  14. Revising your patches
  15. Submitting a patchset

Setup your tools

You'll need to install, configure, and download some software to get started. You should follow the setup directions here.

Setup vim

(Note, if you're running a native Linux install and you're used to another editor like emacs or nano, you can still use that editor, and you can skip this step. You may not be able to use gvim.)

  • Tip: Vim is a simple text editor that has a couple modes. It starts out in standard mode, and you can move the cursor down or up with the arrow keys (or the 'j' or 'k' keys), and move the cursor left or right with the arrow keys (or the 'h' and 'l' keys). You can go into "Insert mode" by typing 'i'. Now you can change text. To get back into standard mode, type <Escape>. To write a file, get into standard mode, and type :w<enter>. To quit vim, type :q<enter>. If you want to learn more about vim, the VIM adventures game is quite fun.

First, we need to make sure to enable the C indentation module in our default text editor (vim). Turning on this module will ensure that lines automatically get indented to the right level as you're editing. It saves you from hitting <tab> a lot. You can turn on automatic indentation based on the file type. First run:

vim ~/.vimrc

Then add this line:

filetype plugin indent on

You'll also want to add a couple more lines, to turn syntax highlighting on, and show the file name in the terminal title bar:

syn on se title

Setup vim as your default editor

Next, we'll need to set up mutt to use vim as the default editor, instead of nano. Run:

sudo update-alternatives --config editor

and chose /usr/bin/vim.basic as the default editor.

Set up email

To be able to send Linux kernel patches, you'll need to be able to send email from the Linux VM image (or your computer that is natively running Linux). The VM image comes installed with esmtp, and if you were following the native Linux install directions you should have that installed on your computer as well. Esmtp is a mail transport agent. It routes email to your mail server, such as gmail. To know what information to give esmtp, you will need to look up your mail server settings.

You will also need to install an approved email client, and use it to respond to message on the opw-kernel mailing list. These instructions assume you're using mutt, but you may find a GUI mail client like Evolution to be easier to use.

Gmail set up

In gmail, go click the gear icon, go to "Settings", go to the tab "Forwarding POP/IMAP", and click the "Configuration instructions" link at the very bottom of the page.

Then click "I want to set up IMAP". At the bottom of the page, under the paragraph about configuring your mail client, select "Other". Note the outgoing mail server information, and copy it into the .esmtprc file, as shown in the next section.

Configure esmtp

  • Note: If you already have another mail transfer agent (MTA) installed, you do not need to install esmtp. Instead, change the .muttrc file "sendmail" line to be the path to your MTA. Mutt uses ssmtp by default, so if your MTA is ssmtp, you can leave that line out entirely.

First, create a .esmtprc file with the right permissions:

touch ~/.esmtprc
chmod g-rwx ~/.esmtprc
chmod o-rwx ~/.esmtprc

Edit the .esmtprc in your home directory, and add lines like this:

identity "my.email@gmail.com"
hostname smtp.gmail.com:587
username "my.email@gmail.com"
password "ThisIsNotARealPassWord"
starttls required 

Next, set up the mail client, mutt, with some defaults, but creating a .muttrc file in your homedirectory:

set sendmail="/usr/bin/esmtp"
set envelope_from=yes
set from="Your Name <my.email@gmail.com>"
set use_from=yes
set edit_headers=yes

Test your email setup

Next, let's send a test email message with mutt. Run this command:


Say "no" to creating an inbox for now. Type 'm' to create a new message. Specify your own email address (or a secondary email) to send the test message to. Set the Subject however you want to. Type a message in the body, and then save and quit. Hit 'y' to send the message, hit 'e' to edit the message again, or hit 'q' to abort sending the message.

Look in your email to double check you received a message. If you send the email to yourself, for some mail services like gmail, the message will not show up in your inbox, and you will have to look in your Sent Mail folder.

Setup git

First, you need to tell git what your name and email address is, so that it can be used in the authorship information in the git commit. Create a file called .gitconfig and add lines like these to it:

   name = Your Name
   email = your.email@example.com 

Make sure that the email you specify here is the same email you used to set up sending mail. The Linux kernel developers will not accept a patch where the "From" email differs from the "Signed-off-by" line, which is what will happen if these two emails do not match.

Make sure you store your full, legal name in the 'name' line. By adding your Signed-off-by line to a patch, you are certifying that you have read and understood the Developer's Certificate of Origin. Please read though that document before you send patches to the kernel.

Explore the kernel tree

First, open a terminal, by clicking the black screen icon with the ">_" text in it.

  • Tip: You can exit out of a terminal tab or window by pressing <CTRL>d at any time. This is the recommended way of closing the terminal, since it won't kill any processes you have running in the background. Get used to exiting the terminal this way by opening and closing the terminal a couple times.

Change directories to your git checkout you set up earlier:

cd git/kernels/staging/

This is the Linux kernel tree. You can explore it by using the `ls` and `cd` commands. If you run `ls, you'll see several different folders:

intern@ubuntu:~/git/kernels/staging$ ls
arch                init             modules.order     System.map block               ipc              Module.symvers    tags COPYING             Kbuild           net               tools CREDITS             Kconfig          README            usr crypto              kernel           REPORTING-BUGS    virt Documentation       lib              samples           vmlinux drivers             MAINTAINERS      scripts           vmlinux.o extra_certificates  Makefile         security          x509.genkey firmware            mm               signing_key.priv fs                  Module.markers   signing_key.x509 include             modules.builtin  sound 

There's more to this directory than meets the eye! If you run ls -A, you'll see there's a hidden directory called .git. This contains all the meta information that git uses to track branches, remote repositories, and changes to files in the local directory.

You can view the commit history by running

git log

If you want a more compact form, you can run a command to see just the "short description" for each commit, with an abbrevated git commit ID:

git log --pretty=oneline --abbrev-commit

Play with some git basics

Git is a distributed revision control system, which means you can hack on your version of the code without having to coordinate with other developers. Think of your git checkout as a separate copy of the kernel respository.

Git includes support for branches. Each branch can contain a completely different set of patches. Kernel developers typically use one branch per patchset. For example, you might have one branch that includes bug fixes, and another branch that contains commits for a new feature you're working on.

You can run `git branch` to see which branch you're on, and what other branches are available:

intern@ubuntu:~/git/kernels/staging$ git branch
 * staging-next 

In this case, there is only one branch, called staging-next. The star indicates that the "staging-next" branch is the one you are currently on. In git speak, we say that you currently have the master branch "checked out".

Create a new branch called 'first-patch', and checkout that branch by running:

git checkout -b first-patch

Now if you run git branch, you'll see that there are two branches, and you are currently on the "first-patch" branch:

intern@ubuntu:~/git/kernels/staging$ git branch
* first-patch 

You can also use the git branch command to show branches on the staging remote repository. Run the command:

intern@ubuntu:~/git/kernels/staging$ git branch -a
* first-patch
  remotes/origin/HEAD -> origin/master

The first remote repository that is used to create the git checkout is called "origin". For now, just remember that "origin" means Greg Kroah-Hartman's staging git repository. Here, you can see the staging remote has three branches: master, staging-linus, and staging-next. The staging-linus branch contains bug fix patches for the current kernel release candidate, and the staging-next branch contains patches for the next kernel release. Your patches will all go into staging-next (since they will be code clean up, not bug fixes), so you want to base all your branches on the staging-next branch.

Update your kernel

When you create your first application clean-up patches, you want to create them on top of the latest commit from the staging-next tree. If your patch is out-of-date and doesn't apply to the latest tree, it may be rejected. You'll need to use git to fetch the latest changes:

git fetch origin

The third word in that command is the name of the remote repository you are fetching from. In this case, it's origin, which is the remote repository we initially cloned from (the staging repository).

That command will fetch the changes from the remote, but it won't actually change in files in the working copy (i.e. the files in this directory). If you run:

git log

You will see that your current working directory still points to the original commit. So where are the staging tree current changes?

The answer is that git stores the changes in a special hidden directory called .git. You can view the history of the staging repository by giving git log the "staging-next" branch of the "origin" remote repository:

git log origin/staging-next

Next, we need to update our branch to include the changes in the staging tree. The safest way to do this is to "rebase" your branch. This means that if you have any commits on your branch, they will be placed on top of the staging tree commits. Sometimes you may have to edit your commits if there are conflicts, but you should ask your mentor for help with this. For now, run:

git rebase origin/staging-next

If you run `git log` to show your staging branch history and then `git log origin/staging-next` to show the staging-next branch history, you should see that they have exactly the same commits.

Configure the kernel

The next step is to create a configuration file, compile the new kernel, and install it.

The first thing to know is that the Linux kernel is completely configurable. Each driver can be separately configured to be installed or not. There are three choices for driver installation:

  • disable the driver completely,

  • build the driver into the main kernel file (vmlinuz),

  • or build it as a module.

If you build the driver into the main kernel file, it will be loaded at boot time. The downside is that the kernel will have to load more code at boot for drivers that may not even corespond to hardware on the system. To avoid this, kernel developers often compile drivers as "modules". A module is a stand-alone .ko driver file that is loaded when the kernel detects hardware that matches the driver. For example, you could configure your wifi driver as a module, and the kernel will load it when it detects the wifi card.

The Linux kernel make system uses a special file called .config that stores what drivers are compiled in, or compiled as modules. Most Linux distributions store the .config file they used to compile your distro kernel in the /boot/ directory:

intern@ubuntu:~/git/kernels/staging$ ls /boot/
abi-3.5.0-23-generic         initrd.img-3.9.0-rc8+ abi-3.5.0-27-generic         memtest86+.bin config-3.5.0-23-generic      memtest86+_multiboot.bin config-3.5.0-27-generic      System.map-3.5.0-23-generic config-3.9.0                 System.map-3.5.0-27-generic grub                         System.map-3.9.0 initrd.img-3.5.0-23-generic  vmlinuz-3.5.0-23-generic initrd.img-3.5.0-27-generic  vmlinuz-3.5.0-27-generic initrd.img-3.9.0             vmlinuz-3.9.0 

You can duplicate the distro's configuration by copying one of the config-* files to a .config file in your git tree. This has already been done for you in the VM image.

Compile the kernel

Next, you'll need to run make to compile your new kernel. Optionally, make can take a flag that indicates how many threads to spawn to start separate compilations. Usually you want to pick a number that is equal to the number of CPUs you have in your machine. For example, if you had a dual core system, you would run:

make -j2

That may take a while. I would suggest reading some of the Linux Device Drivers book while you're waiting.

Make a driver change

These next couple of steps will allow you to make a change to a driver, and test that you've correctly compiled and installed the modified driver.

Modifying a driver under the VM

If you are running Linux in a VM, follow these directions. Otherwise, if you are running Linux natively on your machine, go to the next section.

One driver that's included in all VM images is the e1000 driver, the Intel ethernet driver. If you're running Linux natively, you will need to find a different driver. See the next driver section for how to find an appropriate driver.

The e1000 driver is found in the networking portion of the kernel:

intern@ubuntu:~/git/kernels/staging$ ls drivers/net/ethernet/intel/e1000/
e1000_ethtool.c  e1000.h  e1000_hw.c  e1000_hw.h  e1000_main.c  e1000_osdep.h  e1000_param.c  Makefile 

Let's make a small change to the probe function of the e1000 driver. A probe function is called when the driver is loaded. Let's edit e1000_main.c:

vim drivers/net/ethernet/intel/e1000/e1000_main.c

Next, find the probe function. You can search for text by typing '/' in standard mode. Once you've found the probe function, add a printk line to it:

static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent) {
         struct net_device *netdev;
         struct e1000_adapter *adapter;
         struct e1000_hw *hw;

         printk(KERN_DEBUG "I can modify the Linux kernel!\n");
         static int cards_found = 0;

Then type :wq<enter> to save the file and quit.

A printk function causes a message to be written to the kernel log buffer, which can then be viewed using the dmesg command.

Modifying a driver on native Linux

Your native Linux system will have many different drivers than the ones loaded on a Linux system running in VMPlayer. You will not necessarily have hardware that the e1000 driver can run on. Instead, you must find out which drivers are loaded on your system, and modify one of them.

First, use lsmod to see what drivers are loaded, and pick a name from that list to modify. If you have a varient of the e1000 driver, like e1000e, you may want to use that. Or you can find your wireless driver and modify that.

Once you have the name of a driver, it's time to find out where the .c and .h files for that driver are in the Linux kernel repository. You can do that by searching through the Makefiles to find out what C files are compiled into the driver binary. The best way to do that is with `git grep`. Unlike normal grep, git grep only searches through checked-in files in the repository. Normal grep will also search the binary files, and even the .git directory, which isn't useful and wastes your time.

For example, if you wanted to search for the files that are compiled into the xhci-hcd driver, you would run this command:

git grep xhci-hcd -- '*Makefile'

Once you've identified the files for your driver, you will need to make a change to the probe function as described in the previous section.

Compile your changes

Recompile your kernel, by running make (with an optional -jN flag):

make -j2

You may need to fixed some compilation errors. Also fix any new warnings that are due to your changes. In the Linux kernel, we strive to make sure that drivers do not produce warnings on anyone's system (this includes 32-bit and 64-bit systems, as well as different architectures, such as PPC, ARM, or x86). New features or bug fix patches that add additional warnings may not get merged.

Install your changes

After you've compiled the driver, you need to install your changes by running:

sudo make modules_install install

Test your changes

Since you've compiled a completely new kernel, you need to reboot into that new kernel in order to test your module changes. Reboot your VM (or computer), and then run:

dmesg | less

Search for your printk in the log file by typing "/I can modify". You should be able to find this message within the driver output during boot. If you don't see this message, ask for help on the #kernel-opw IRC channel on irc.oftc.net, or on the opw-kernel mailing list

Revert your changes

Since that was just a simple test, and you probably don't want to commit that change, you can revert your changes. Exit out of your editor by typing :q<enter> and running this command:

git reset --hard HEAD

That will revert ALL FILES in your current working directory to the last known commit (the HEAD commit), wiping out all your uncommited changes. Read the git reset manual for more information on ways to reset the state for specific files.

Start creating your first patch

Next, you'll get to do some useful modifications to the kernel, create your first git commit, and send out your first patch. Before you make your first commit using git, you'll need to do some setup.

Git pre-commit hooks

Git includes some "hooks" for scripts that can be run before specific git commands are executed. The "pre-commit" hook is run before you make a git commit with the `git commit`` command.

Linux kernel developer have very stringent guidelines for coding style. They're so picky, they created a script called checkpatch.pl that you can run over your patches to make sure the patch compiles to the kernel coding style.

To ensure that you create good commits that comply with the coding style, you can run checkpatch.pl over your commit with the "pre-commit" hook. That means git will refuse to commit your changes if it doesn't compily with kernel coding style.

If you already have a .git/hooks/pre-commit file, move it to .git/hooks/pre-commit.sample. git will not execute files with the .sample extension.

Then edit the .git/hooks/pre-commit file and add the following lines:

exec git diff --cached | scripts/checkpatch.pl --no-signoff - || true

Next, make sure the file is executable:

chmod a+x .git/hooks/pre-commit

After you commit, this hook will output any checkpatch errors or warnings that your patch creates. If you see warnings or errors that you know you added, you can amend the commit by changing the file, using git add to add the changes, and then using git commit --amend to commit the changes.

Checking that the pre-commit hook works

You can check that the pre-commit hook is working by adding a deliberate change that will trigger checkpatch (such as adding a really long line or adding trailing whitespace to a line), and then attempting to run git commit. After you commit, you should see the additional checkpatch.pl warning or error.

If your pre-commit hook is not working, please ask for help on IRC.

Understand patch best practices

Before you create your patch, you need to understand how to create good patches. Otherwise your patches will be less likely to be accepted by maintainers, and you will have to go through more revisions than necessary.

First, read about PatchPhilosophy. That document will help you create patches that are easy to read, and have a better chance of being applied by maintainers. Please also read CodingStyle (which is also available in the kernel code repository under Documentation). This will help you understand how to write code that the Linux kernel community will accept, and the rules here are the basis for the script checkpatch.pl.

Note that checkpatch.pl is failable! Use your best judgement when deciding whether it makes sense to make the change checkpatch.pl suggests. The end goal is for the code to be more readable. If checkpatch.pl suggests a change and you think the end result is not more readable, don't make the change. For example, if a line is 81 characters long, but breaking it makes the resulting code look ugly, don't break that line. Please read the CheckpatchTips page for how to avoid common mistakes when cleaning up drivers.

Find a driver to clean up

The staging tree, in drivers/staging/ is full of drivers that are not quite up to kernel coding style, or that use deprecated API. Drivers get placed here in order to get cleaned up. Some drivers have a TODO file in their parent directory, that lists things that need to be done to it:

find drivers/staging -name TODO

You can either tackle one of those TODO items, or you can do a simple coding style cleanup.

Running checkpatch.pl

If you pick a driver in staging, you can run the script that checks whether a file conforms to kernel coding style:

perl scripts/checkpatch.pl -f drivers/staging/csr/* | less

Pick a warning, and try to fix it. For your first patch, only pick one warning. In the future you can group multiple changes into one patch, but only if you follow the PatchPhilosophy of breaking each patch into logical changes.

Recompiling the driver

You'll need to make sure the driver you're changing is configured as a module. Run:

make menuconfig

This opens up a text-based GUI that allows you to explore the configuration options.

Use the arrow keys to go to Device Drivers ->  and hit <enter>. Then go down to Staging drivers. At any time, you can hit '?', which will show you the help text for that kernel configuration option. You can search for the driver you're modifying by '/', in order to get the driver's longer name. Make sure the driver you're working on is compiled as a module ('M'), instead of being built-in ('*'). You can change a driver to being compiled as a module by typing 'm' when the driver is selected in the menu. Hitting <enter> will change the driver to being built-in.

Once you've enabled the driver you're modifying, use <tab> or the right arrow key to move the cursor from 'Select' to 'Exit' and hit <enter>. Continue to do this until you get to the main menu. When it asks you to save your configuration, chose 'Yes'.

Then recompile the kernel with:

make -j2

You should reboot your kernel, load the driver with modprobe. You'll be able to see that the driver is loaded by running lsmod. Loading the driver at least makes sure that the driver probe function works.

  • Note: Do not work on drivers that show that they depend on CONFIG_BROKEN. If you search for a driver after running `make menuconfig`, and you notice the "Depends on:" line includes BROKEN, do not work on this driver.

Driver dependencies

Sometimes it's hard to find your driver. Maybe you searched for the CONFIG option, but you can't find it in the menus where it should be. This may be because a driver or subsystem it depends on is not enabled, and so this driver is hidden from the menu. You need to enable all the dependencies in order to make the menu option visible, and then you can enable the driver you're modifying.

For example, say I was modifying drivers/usb/host/xhci-ring.c. If I look in the Kconfig file in the parent directory (drivers/usb/host/Kconfig), I can see an option for the xHCI driver:

         tristate "xHCI HCD (USB 3.0) support"
         depends on USB && USB_ARCH_HAS_XHCI
           The eXtensible Host Controller Interface (xHCI) is standard for USB 3.0
           "SuperSpeed" host controller hardware.

           To compile this driver as a module, choose M here: the
           module will be called xhci-hcd.

Now, I run `make menuconfig` and type '/' to search for USB_XHCI_HCD. The search entry shows:

Symbol: USB_XHCI_HCD [=m]
 Type  : tristate
 Prompt: xHCI HCD (USB 3.0) support
    -> Device Drivers
 (1)   -> USB support (USB_SUPPORT [=y]
   Defined at drivers/usb/host/Kconfig:20
   Depends on: USB_SUPPORT [=y] && USB [=n] && USB_ARCH_HAS_XHCI [=y])

Look at the "Depends on" line. This is a boolean equation that represents the driver dependencies that need to be enabled in order to even show the driver option in the menus. A 'y' means the driver or subsystem is built into the kernel and an 'm' means the driver is built as a module. Both these options are equivalent to '1' in boolean logic. A 'n' means the driver or subsystem is disabled. An 'n' is equivalent to a '0' in boolean logic.

  • Tip: If you don't know boolean logic, you can take a look at these tutorials. If you just need a brush up on boolean logic, here's a crib sheet. If you're lazy, here's a boolean algebra calculator, or you can use the "Programming Mode" for the calculator application installed in Ubuntu by default.

In this case, the xHCI driver config option is not being shown in the menus because USB is set to 'n'. If I search for USB, I see lots of results, and finally find this relevant one:

Symbol: USB [=n]  
 Type  : tristate
 Prompt: Support for Host-side USB
     -> Device Drivers
 (1)   -> USB support (USB_SUPPORT [=n]) 
   Defined at drivers/usb/Kconfig:51
   Depends on: USB_SUPPORT [=n] && USB_ARCH_HAS_HCD [=n]

If I look under the Device Drivers menu, I can find "USB Support" and set it to 'm'. Once that's done, I can find CONFIG_XHCI_HCD under the "USB Support" menu.

It may take recursively enabling several different configuration options before you can even see your driver in the menu. Ask for help if you're stuck.

  • Note: Do not work on drivers that show that they depend on CONFIG_BROKEN. If you search for a driver after running `make menuconfig`, and you notice the "Depends on:" line includes BROKEN, do not work on this driver.

Reloading modules

If you're running a kernel that has the same release version (`uname -r`) as the new code you're compiling, you can test your changes without rebooting. Simply install the module in /lib/modules, and unload and reload the driver:

make -j2 && sudo make modules_install
sudo modprobe -r <module_name>
sudo modprobe <module_name> 

How do you know what the module name is? If you've compiled the driver as a module, there should be a .ko file in the parent directory. For example, after we configure the CSR driver to be compiled as a module, we can run this command:

intern@ubuntu:~/git/kernel/staging$ ls drivers/staging/csr/*.ko
 drivers/staging/csr/csr_helper.ko  drivers/staging/csr/csr_wifi.ko 

So, there are two drivers that we need to load with modprobe. You can load those drivers one at a time by passing modprobe the base filename:

sudo modprobe csr_wifi 

To ensure the driver got loaded, you can run:

lsmod | less

In this case, you'll notice that both the csr_wifi and csr_helper modules got loaded. This is because the csr_wifi driver depends on the csr_helper driver, and modprobe automatically figured out it needed to load the csr_helper driver before it loaded the csr_wifi driver.

Committing your changes

In this example, assume we've addressed a warning in the CSR driver, modified the file, recompiled the driver, and tested our changes.

Viewing your changes

Git keeps track of changes in the working directory. Git can be told to ignore binary files (like .o or .ko files), so it won't track changes to those files. You can see which files have been modified by running:

git status

git can also show you a diff stat of what changed:

git diff

Commit your changes

Assuming we want to include all of our changes in one git commit, you can use git to add the changed file to the list of changes to be committed (the "staging area"):

git add <file>

If you run `git diff` again, you'll notice it doesn't list any changed files. That's because, by default, git diff only shows you the unstaged changes. If you run this command instead, you'll see the staged changes:

git diff --cached

That command will show you the changes to be committed.

Reverting your staged changes

If you don't want to commit those changes, you can remove those changes from the staging area by running:

git reset <file>

Committing parts of files

You can also add parts of files to the staging area by using the {{-p}} flag:

git add -p

That will allow you to add hunks of the file to the staging area, or even edit hunks that you want to commit. This is useful, for instance, if you've made whitespace changes, and also made a camel-case variable name fix, but those changes are on the same line. You can edit the line to revert the camel-case name change, and just add the whitespace change to the staging area. Then when you commit, you will just be committing the whitespace change.

Committing changes

Finally, you can commit your staged changes:

git commit -s -v

That will add the Signed-off-by line that is needed at the end of your patch description. The -v flag will show you the diff that you're committing. This is very useful to decide whether you are committing the correct code.

Make sure that when you create your patch, you follow the PatchPhilosophy guidelines. Make sure to include a blank line between your short description (what will become the Subject line of your patch) and the body of your patch. Make sure there is a blank line between the body of your patch and your Signed-off-by line. Again, make sure you read PatchPhilosophy before you make your first commit.

Editing your commits

If you should need to edit your commits, please see the "Editing commits" and "Editing patchsets" sections.

Viewing your commit

Make sure your commit looks fine by running these commands:

git show HEAD

This will show the latest commit. If you want git to show a different commit, you can pass the commit ID (the long number that's shown in `git log`, or the short number that's shown in `git log --pretty=oneline --abbrev-commit`). Read the "Specifying Revisions section" of the `git rev-parse` manual page for more details on what you can in place of a commit ID.

You'll also want to make sure your commit looks fine when you run these two commands:

git log
git log --pretty=oneline --abbrev-commit

Submit a patch

The first step to sending a patch is to figure out who to send it to. (If you are an applicant for OPW, please DO NOT SEND YOUR FIRST PATCH TO THE MAIN LINUX MAILING LISTS! See the explanation in the "Send your patch to your mentors" section below.)

If you are not an OPW applicant, you need to find the maintainer of the code you're patching, and Cc the correct mailing list. If you simply send it off to LKML, it will get ignored. The perl script in the kernel source directory scripts/get_maintainer.pl will take either take a git commit or a file, and tell you who to send your patch to:

$ git show --pretty=oneline --abbrev-commit HEAD
0680eb1f485b timekeeping: Another fix to the VSYSCALL_OLD update_vsyscall
diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index f36b02838a47..fb4a9c2cf8d9 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -338,10 +338,11 @@ EXPORT_SYMBOL_GPL(ktime_get_mono_fast_ns);
 static inline void update_vsyscall(struct timekeeper *tk)
-       struct timespec xt;
+       struct timespec xt, wm;
        xt = timespec64_to_timespec(tk_xtime(tk));
-       update_vsyscall_old(&xt, &tk->wall_to_monotonic, tk->tkr.clock, tk->tkr.mult,
+       wm = timespec64_to_timespec(tk->wall_to_monotonic);
+       update_vsyscall_old(&xt, &wm, tk->tkr.clock, tk->tkr.mult,

$ git show HEAD | perl scripts/get_maintainer.pl 
John Stultz <john.stultz@linaro.org> (supporter:TIMEKEEPING, CLOC...)
Thomas Gleixner <tglx@linutronix.de> (supporter:TIMEKEEPING, CLOC...)
linux-kernel@vger.kernel.org (open list:TIMEKEEPING, CLOC...)
$ perl scripts/get_maintainer.pl -f kernel/time/timekeeping.c
John Stultz <john.stultz@linaro.org> (supporter:TIMEKEEPING, CLOC...)
Thomas Gleixner <tglx@linutronix.de> (supporter:TIMEKEEPING, CLOC...)
linux-kernel@vger.kernel.org (open list:TIMEKEEPING, CLOC...)

The second step to submitting a patch is to create and send a patch as an email. You cannot send patches as attachments to the mailing list. Instead, you will have to craft a special email, and send the patch inline.

There are two ways to send a patch. Either way is correct, and we suggest you get comfortable with sending patches both ways. We suggest sending your first patch to yourself, in order to make sure it works.

Creating a patch to send with mutt

First, create a patch that describes the, using `git format-patch`. That command takes a starting commit ID (and optionally) an ending commit ID, in order to create patches for the commit after the starting commit ID. We want to create a patch for the first commit in our history (the HEAD commit). To specify the commit before the HEAD commit, you can use either "HEAD^" or "HEAD~".

git format-patch -o /tmp/ HEAD^

The -o flag specifies where to put the patch. If you've run the command correctly, you should see the command output a filename in /tmp/. If it outputs nothing, you've forgotten the '^'.

Sending your patch with mutt

Next, tell mutt to use that patch as a draft email, with the -H flag:

mutt -H /tmp/0001-<whatever your filename is>

Send this first patch to yourself. Leave the Subject line intact. Now you can see how your patch translates to an inline email. The git short description becomes the email subject line, and the patch body and diff become the body of the email.

Write the mail, quit out of vim, and send the mail with 'y'.

Sending patches with git send-email

You can also send a patch with the command-line tool `git send-email`. You can either pass it the file that `git format-patch` generated, or you can give it the same commit ID range you gave `git format-patch`:

git send-email --annotate HEAD^

git send email will prompt you with who to send the message to, and other odd questions:

Who should the emails be sent to (if any)?
Message-ID to be used as In-Reply-To for the first email (if any)?

Put in your mentor's address in the first line, and leave the second blank unless you want it to appear as a thread in an existing conversation.

At this point `git send-email` will look for Cc: lines in your commit message, and include them in the email headers. It will then show you the resulting email header, and ask you to confirm:

Send this email? ([y]es|[n]o|[q]uit|[a]ll):

Send with 'y' (or 'a': git send-email can be used to send multiple commits at once).

Tips and Tricks

Please read the patch tips and tricks page for an explanation of patch tags (e.g. Reviewed-by and Signed-off-by), and when to use [PATCH] vs. [RFC] in your patch subject prefix.

Send your patch to your mentors

Once you've send your patch to your own email, and ensured that it looks fine, it's time to send your patch off.

If you are not an OPW applicant, send your patch to the maintainer and lists that the get_maintainer.pl script tells you to.

If you are an OPW applicant, please send your patches to opw-kernel mailing list. Make sure you have your mentor's ACK on the patch through the opw-kernel mailing list before sending it off to the main mailing lists. There may be protocol around who to send the email to, or which driver files should not be cleaned up with checkpatch.pl. Again, make sure you have your mentor's ACK before sending email off to the main mailing lists.

Responding to emails

When responding to emails on the mailing list, it's important to use a communication style consistent with what the open source community expects. This means you will need to:

  • Use one of the standard email clients listed in Documentation/email-clients.txt. Do NOT use the gmail web interface to respond to patch feedback, as it line-wraps the mail (even when in plain text mode). Do NOT use outlook, as it mangles patches (turns tabs into spaces).

  • When responding to email, make sure you respond inline, rather than top-posting. This is a good example of responding inline.

  • Make sure your email client appends '>' characters to inline mail when you respond to it.

  • Don't include quotes in your signature.

Revising your patches

Editing your commits

If you should need to edit your commits, please see the "Editing commits" and "Editing patchsets" sections.

Versioning one patch revision

If you receive feedback on a patch, and were asked to update the patch, you need to version the patches that you re-send. A new version number lets reviewers know you made changes to the patch, and they should review it again.

An example of what this would look like is:

[PATCH] Foo: Fix these things

And the updated versioning for a second revision:

[PATCH v2] Foo: Fix these things better

It's fairly simple to accomplish this, and there's certainly a few ways to do this. If you generate your patches using `git format-patch`, then it's simple to do this. Just add the --subject-prefix option like this:

git format-patch --subject-prefix="PATCH v2"

or whatever version you are currently on (3, 4, etc.).

Make sure to include a summary of what changed from one version of the patch (or patchset) to the next. Do not include a version change log in the patch description, because it won't make sense when the final version of the patch makes it into the kernel. Instead, edit your patches before you send them to include a change log below the "---" line. Git will ignore this change log when the patch is applied. Here's a good example of a patch with a change log:

Subject: [PATCH v2 1/2] USB: at91: fix the number of endpoint parameter

In sama5d3 SoC, there are 16 endpoints, which is different with
earlier SoCs (only have 7 endpoints). The USBA_NR_ENDPOINTS micro
is not suitable for sama5d3. So, get the endpoints number through
the udc->num_ep, which get from platform data for non-dt kernel,
or parse from dt node.

Signed-off-by: Bo Shen <voice.shen@atmel.com>
Changes in v2:
  - Make the commit message more clearer.

 drivers/usb/gadget/atmel_usba_udc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/usb/gadget/atmel_usba_udc.c b/drivers/usb/gadget/atmel_usba_udc.c

Versioning patchsets

When you send out a new version of a patchset, instead of indicating on each patch what changed, you should include a coverletter that explains all the changes. Example cover letter. You can include a coverletter in your patcheset by using the --cover-letter option to git format-patch, e.g.

git format-patch -n --subject-prefix="PATCH vY" --cover-letter

where Y is the version of the patch you are currently sending. The cover-letter option will create a "PATCH 0/Y" that you can add a change log to. Make sure to change the SUBJECT HERE on the coverletter subject line. Note that cover letters are discarded when applying patches, so any information that you want to preserve in the git log should be in the patch descriptions, not the cover letter. The cover letter is for introducing what problem you're trying to solve, and including version change logs.

Make sure to include all of the patches you sent in the patchset before (i.e. if you sent three patches and you had to revise the second patch, send all three again).

An example of what this would look like is:

[PATCH 0/3] comedi: Fix these things
[PATCH 1/3] comedi: Fix the first thing
[PATCH 2/3] comedi: Fix the second thing
[PATCH 3/3] comedi: Fix the third thing

And the updated versioning for a second revision:

[PATCH v2 0/3] comedi: Fix these things
[PATCH v2 1/3] comedi: Fix the first thing
[PATCH v2 2/3] comedi: Fix the second thing
[PATCH v2 3/3] comedi: Fix the third thing

Submitting a patchset

Sometimes you need to send multiple related patches. This is useful for grouping, say, to group driver clean up patches for one particular driver into a set, or grouping patches that are part of a new feature into one set.

For example, take a look at this patch set:

Typically, the subject prefix for patches in the patchset are [PATCH X/Y] or [RFC X/Y], where Y is the total number of patches, and X is the current patch number. Patchsets often have a "cover letter" that is [PATCH 0/Y] or [RFC 0/Y]. A cover letter is used to explain why the patchset is necessary, and provide an overview of the end result of the patches. Cover letters are also great places to ask for help in reviewing specific patches in the patchset.

In order to create patchsets like this, you will need to use either `git send-email``}} or {{{`git format-patch`. These tools will generate the right "In-Reply-To" Headers, so that in a text mail client, the patches will appear next to each other, rather than having unrelated email in between. Otherwise, patches may get jumbled, depending on when they were received.

Using git format-patch to send patchsets

First, use `git log --pretty=oneline --abbrev-commit` to figure out the first commit ID you want to send. For example, say that my tree had this git log history:

b7ca36a Docs: Move ref to Frohwalt Egerer to end of REPORTING-BUGS
bf6adaf Docs: Add a tips section to REPORTING-BUGS.
bc6bed4 Docs: Expectations for bug reporters and maintainers
2c97a63 Docs: Add info on supported kernels to REPORTING-BUGS.
7883a25 Docs: Add "Gather info" section to REPORTING-BUGS.
d60418b Docs: Step-by-step directions for reporting bugs.
3b12c21 Trivial: docs: Remove six-space indentation in REPORTING-BUGS.
bb33db7 Merge branches 'timers-urgent-for-linus', 'irq-urgent-for-linus' and 'core-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
41ef2d5 Linux 3.9-rc7

The first commit I want to send as part of the patchset has commit ID 3b12c21. The last patch I want to send has commit ID b7ca36a. So, I want to pass the commit range 3b12c21^..b7ca36a to git format-patch. (Remember, the git format-patch range starting commit must be the commit before the first commit you want to send, so we use the '^' to specify the patch before commit 3b12c21.) The command will look something like this:

git format-patch -o /tmp/ --cover-letter -n --thread=shallow --cc="linux-usb@vger.kernel.org" 3b12c21^..b7ca36a

Again, the -o flag specifies where to put the email files. The -n flag says to add numbering to each patch (e.g. [PATCH 2/5]). The --thread=shallow flag specifies that all patches will be In-Reply-To your cover letter.

That will output files into /tmp, and you can edit them in mutt in multiple terminal tabs:


Resending patches

If you find a mistake in your initial patch submission, and you need to correct it, make sure to version your new patch as described in the previous section.

If your patch has not been responded to within three business days (not including weekend days), the mentors may have missed it, and you may want to resend it. Make sure to use RESEND in the subject line if the patch itself has not been changed (meaning no changes in the patch description or body). You can use the --subject-prefix argument to git format-patch to add RESEND, by running

git format-patch --subject-prefix="RESEND PATCH"

That command will create an email you can send with mutt (or your approved mail client) that will have a subject that looks like:

[RESEND PATCH] Foo: Fix these things

If your patch hasn't been responded to within a week, ping a mentor (or sarahsharp) on the IRC channel. Do not ping mentors immediately after sending patches. We know you're excited/worried about submitting your first patch, but mentors need to take time to properly respond and review patches.

Tips for Applying to be an OPW Kernel intern

After you complete the first patch tutorial and get a patch accepted into the Linux kernel, you may wonder what to do next. We recommend getting 3-5 single cleanup patches accepted, and then moving on to create a patchset with several different cleanups for the same staging driver. Please limit your patchsets to no more than 10 patches, unless you've been told to break the patchset down further. Please try to limit your patches to no more than 100 to 200 lines total, to avoid mentors getting patch review fatigue. If you're bored and losing focus when you're tediously creating the patch, it's highly likely the mentors will get tired while reviewing it. :) Once you've gotten 3-5 patchsets accepted, move onto exploring more complex tools, like sparse and coccinelle. Try to tackle some code refactoring, or take on a small task that a mentor has recommended for their project. Mentors are more likely to pick you for their project if they see you tackling tasks from that project or asking questions about that project.

Tackling increasingly complex patches mean the OPW mentors will get a sense of how good a C programmer you are. It's very easy for beginner C programmers to do some of the more simple checkpatch cleanups, but it takes more skill to complete our mentor's small tasks, or sparse and coccinelle patches. You may find yourself trying to understand advanced C topics, such as when to declare a function static. If you are attempting silence a checkpatch.pl warning, and you don't understand what the warning message means, please ask on IRC or the mailing list before blindly trying to silence the warning. We value applicants that know when they need help, and we are happy to share knowledge and expand your skillset.

Please remember that we value consistently clean patches, so make sure to compile your code and ensure that no checkpatch errors or build warnings were added. Do this every single time for every patch revision, immediately before you send your patches out. We value applicants that learn fast enough to avoid making the same mistakes twice. It's a good practice to keep a log of improvements people have suggested for your previous patches, and double check that before submitting a new patch or patchset.

We look for applicants that can revise patches based on feedback from mentors. If you get feedback on a patch, change the code and send a revised patch. Don't just go onto the next patch. We value applicants that consistently send or revise patches every day, or every other day, rather than being sporatic about when patches are sent. Our best interns made the application process a part of their daily routine: prepare a patchset early in the day, go about their lives, and revise the patchset based on feedback received later in the day. Rinse and repeat.

We also value clear, concise communication. If you have a setup issue, tell us the exact failure mode, how to reproduce the failure, and what you've done to try and trouble shoot the failure. Please ensure that when you ask a question on IRC, you remain connected to the channel until someone answers your question. If no one answers in a couple hours, send the question to the mailing list.

We look for applicants that have a knack for seeking documentation and resources their own, while still knowing when to ask questions. If you have been stuck on a problem for more than a day, seek help! We love to see applicants ask questions on IRC and the mailing list, and we especially appreciate applicants that help each other out. Helping out someone who is struggling means you get to share your knowledge, and the act of writing down what you know will solidify that knowledge in your mind. It also shows the OPW mentors that you're learning new skills, and are willing to share that knowledge. Bonus points if you make a kernelnewbies account (email Sarah Sharp for edit privileges) and correct or add knowledge to this wiki.

During the intern selection process, mentors will often already have a particular applicant in mind they want to pick. This applicant has often worked directly with the mentor, asked them questions, or sent them a detailed timeline for their project. Often times one particular project is very popular, because the mentor is highly responsive on the mailing list, and applicants like working with them. In that case, that project may be the first or second choice for many of the OPW applicants. Therefore, it is important to list at least four projects you are interested in, and make sure to work on small tasks from at least 2-3 OPW mentors.

There are also a few factors outside your control that may impact whether you are selected for an OPW internship. We try to ensure that the timezone difference between mentors and interns is acceptable. If the project requires specialized hardware, and it is difficult to get packages through customs in your country, you may not be selected for that project. An OPW internship is a full-time commitment, so we cannot accept you if you're working a full-time job, or taking more than one or two classes during the internship period. If there are many very good candidates, you may not be selected to be an OPW intern this round. We encourage you to apply next round, since you'll have a head start over the other candidates.

Good luck, future Linux kernel developers!

Tell others about this page:

last edited 2014-09-02 00:09:43 by SarahSharp