the Chromium logo

The Chromium Projects
ChromiumOS > Guides >

Kernel Development

This document covers best practices for kernel development in ChromiumOS, including debugging tips, platform bringup info, committing changes, sending code upstream, and using upstream repos for testing & development.

*** promo Note for Googlers: There are additional Google-specific notes and work-in-progress notes at go/chromeos-kernel-tips-and-tricks.

Development workflow

Build and deploy

  1. Ensure the target machine is ready for kernel updates (not usually necessary)
  2. Build your kernel using either method:
  3. Deploy your kernel
  4. Recover from a bad installation (if necessary)

Preparing the target

First, make sure you're running a dev or test image. Then ensure that verity is disabled on the target before running the update_kernel.sh script, or it will complain and abort. Verity can be disabled using the command /usr/share/vboot/bin/make_dev_ssd.sh --remove_rootfs_verification --partition <partition number> on the target followed by a reboot.

Preparing to build

If you are making any changes to the kernel sources, and want to build the kernel with your changes, you must first tell the build system to use your local sources:

(chroot) $ cros workon --board ${BOARD} start chromeos-kernel-[x_y]  # first time

Building with emerge

The emerge-${BOARD} command is the standard way of building all packages, including the kernel. For instance:

(chroot) emerge-${BOARD} chromeos-kernel-[x_y]

Building with cros_workon_make

cros_workon_make runs an incremental build of the kernel. However, it doesn't resolve dependencies, therefore before using it, run emerge-${BOARD} once as described above. Then run:

(chroot) $ FEATURES="noclean" cros_workon_make --board=${BOARD} \
    --install chromeos-kernel-[x_y]

To enable debug options like lockdep and KASAN, add USE="debug" to the command line above. This is highly recommended because the default build is optimized for performance rather than debugging purpose. Note that the debug build bloats the size of kernel image, and the image may not be able to fit into its partition on some older devices. The debug build also takes much longer to boot.

You can also enable serial port at the same time by USE="debug pcserial".

*** note Note that using cros_workon_make leaves build artifacts in your source directory under the build directory. When you do a regular emerge of the kernel (and are cros-workon-ed) this will slow things down because the entire source directory gets copied. So delete the build directory when you're done.

Deploy your kernel

Update the kernel on the target (if you are sure that it will boot):

(chroot) $ ~/chromiumos/src/scripts/update_kernel.sh --remote=$DUT

*** note Note: cros deploy does not currently support deploying kernel packages.

Deploy your kernel with recovery in mind.

Alternatively flash a known good working image to the device first and then use update_kernel.sh to target the other kernel partition (typically KERN-B when KERN-A is live) instead of the live kernel partition. Boot the device and then run update_kernel.sh with --ab_update option to specify the other kernel partition.:

(chroot) ./update_kernel.sh --remote=$DUT --bootonce --ab_update

The bootloader will attempt to boot the kernel on the other partition (typically KERN-B) and kernel modules will be updated to the matching rootfs partition (matching ROOT-B partition). If the kernel crashes early on then a reboot will fallback to the A slot kernel and rootfs that is known to be good and working. If the boot is considered successful the partition is marked with a successful boot and will be used from the next time.

Be aware, however, that after you reboot again, if your boot was not yet marked as successful you're back to your old kernel (which can be very confusing... "hey where did my new feature go??").

See disk format for more info on partition layouts, as you may need to use a different partition number depending on how you installed your kernel or which one you want to replace. The cgpt utility can also be used to view a device's partition layout, as well as to modify the priority of its kernel partitions. To read a partition's priority, where -i specifies the partition in question:

(DUT) # cgpt show -i 2 /dev/mmcblk0
start size part contents
20480 32768 2 Label: "KERN-A"
Type: ChromeOS kernel
UUID: BCD6FC1E-528F-494B-9B06-FB723EA37672
Attr: priority=1 tries=0 successful=1

To switch between kernel primary/backup partition, execute (this can also be done from a USB stick):

(DUT) # cgpt prioritize -P2 -i <partition to be used> /dev/mmcblk0
Dealing with partition corruption due to bad kernel recovery

One time I really screwed up my system by recovering (after bad kernel installation) with 'dd if=/dev/sdb of=/dev/sda'. I forgot the '2' after each drive specification. This overwrote my internal partition table with an exact copy of the USB stick's partition table, including the GUIDs. When I subsequently tried to boot USB, the system always seemed to boot off the internal disk. 'rootdev -s' reported (internal partition) /dev/sda3. After an hour or so, consultation with Bill showed that I really was booting the kernel from /dev/sda2, but the kernel found the matching GUID on sda before even looking at sdb. This was recovered with:

(DUT) $ a=$(uuidgen)
(DUT) $ cgpt add -i 3 -u $a /dev/sda

which generates and installs a new GUID for sda3.

Recover from a bad kernel update

One issue is often to figure out how to recover if you flash a bad kernel. Booting from USB and running chromeos-install is one solution, but that's slow. There are a couple approaches that can be useful to recover quickly.

  1. Always flash a known good working image to the device so that you don't have to do this.

  2. Using chromeos-install-kernel to update the kernel:

    • Always have a good USB stick connected to the device.
    • (optional) Make sure you use a serial-enabled coreboot firmware.
    • If the kernel on internal storage does not boot anymore:

    1. Boot from USB (press Ctrl-U during FW bootup, you may have to do this repeatedly if on a serial console)

    2. Copy kernel and modules back to internal storage using the chromeos-install-kernel script:

      (DUT) # chromeos-install-kernel
(DUT) # reboot

    The system should boot from internal storage again.

  3. Using update_kernel to update kernel:

    update_kernel.sh can be used to update the kernel in internal storage after booting from USB. It will try to copy modules over also but currently has issues if rootfs verification is enabled on USB. If your kernel change doesn't depend on the modules on the rootfs, you can go ahead and use --ignore_verity to update only the kernel. Example below assume NVME for KERN-A.

    TODO(b/202778198): fix update_kernel.sh to update root fs with modules.

    (chroot) $  ./update_kernel.sh --remote=$DUT --board=$BOARD \
  --device=/dev/nvme0n1p --ignore_verity --partition /dev/nvme0n1p2

Kernel configuration

Kernel configuration in ChromiumOS has an extra level of indirection from the normal .config file. So do the instructions - see this page for more information.

See also the cros-kernel eclass documentation.

Inspect kernel config

The built kernel config is available at /build/$BOARD/boot/config in the chroot.

On a running system, the kernel config is not loaded by default (to save memory), so you'll need to use modprobe to load it first:

(DUT)# modprobe configs; zcat /proc/config.gz

Kconfig changes

Kconfig changes (changes that affect chromeos/config) should be normalized by running chromeos/scripts/kernelconfig olddefconfig

Modifying the kernel command line

There are several ways to modify the kernel command line. These vary in ease of use, as well as type of target system (e.g., real hardware running the CrOS verified boot chain behaves differently than a QEMU VM booting via BIOS).

Modify kernel command line via update_kernel.sh

The built kernel command line is available at ~/chromiumos/src/build/images/$BOARD/latest/config.txt in the chroot. The update_kernel.sh script will use this file for the command line when updating the device if it exists. Otherwise, it will try to reuse the target device's args. The --remote-bootargs flag can also force reusing the target args.

*** note Note: while update_kernel.sh can update the kernel on a system using legacy BIOS with syslinux (such as BOARD=amd64-generic on QEMU), it does not currently know how to update the command line.

Modify kernel command line on device

It's possible to modify the command line of an installation on a target device as well.

For example, to enable the console on a recovery image on USB stick /dev/sdb:

(DUT) # /usr/share/vboot/bin/make_dev_ssd.sh -i /dev/sdb --partitions 2 --save_config ./foo
(DUT) # vi ./foo
  # add the updated command line, for example: earlycon=uart,mmio32,0xfedc6000,115200,48000000
  # save & exit vi
(DUT) # /usr/share/vboot/bin/make_dev_ssd.sh -i /dev/sdb --partitions 2 --set_config ./foo
(DUT) # /usr/share/vboot/bin/make_dev_ssd.sh -i /dev/sdb --recovery_key

This extracts the command line from the kernel partition using vbutil, allowing you to edit it and write it back.

Instead of --save_config and --set_config, you can also use --edit_config to edit the config in-place if you want.

*** note Note: this only works on systems using the CrOS vboot boot chain (i.e., all CrOS hardware). Notably, QEMU VMs do not use CrOS vboot.

Modify kernel command line in depthcharge

*** note This option is heavy-handed and difficult, as updating your firmware may brick your device. Proceed only if you know what you're doing.

If you're booting with depthcharge, you can modify the command line thusly:

(chroot) $ cros-workon-${BOARD} start depthcharge
(chroot) $ vi ~/chromiumos/src/platform/depthcharge/src/board/${board}/board.c

Call the commandline_append() function containing your command line addition:

#include "boot/commandline.h"

static int board_setup(void)
{
	commandline_append("earlycon=uart,mmio32,0xfedc6000,115200,48000000");
}

Rebuild depthcharge, build it into the firmware image, and flash it to your device.

Modify kernel command line on a legacy BIOS

*** note Note: no supported CrOS hardware boots via legacy BIOS. Legacy BIOS is typically used by QEMU VMs (e.g., cros_vm).

First, locate the EFI system partition (a.k.a., boot partition). This is often partition 12, such as /dev/sda12. Mount this partition to view and edit the syslinux boot configuration:

(DUT) # mkdir /tmp/mnt
(DUT) # mount /dev/sda12 /tmp/mnt
(DUT) # find /tmp/mnt/syslinux
/tmp/mnt/syslinux
/tmp/mnt/syslinux/syslinux.cfg
/tmp/mnt/syslinux/default.cfg
/tmp/mnt/syslinux/usb.A.cfg
/tmp/mnt/syslinux/root.A.cfg
/tmp/mnt/syslinux/root.B.cfg
/tmp/mnt/syslinux/README
/tmp/mnt/syslinux/vmlinuz.A
/tmp/mnt/syslinux/vmlinuz.B
/tmp/mnt/syslinux/ldlinux.sys
/tmp/mnt/syslinux/ldlinux.c32
(DUT) # cat /tmp/mnt/syslinux/README
Partition 12 contains the active bootloader configuration when
booting from a non-ChromeOS BIOS.  EFI BIOSes use /efi/*
and legacy BIOSes use this syslinux configuration.
(DUT) # cat /tmp/mnt/syslinux/default.cfg
DEFAULT chromeos-usb.A

Now, edit the relevant boot entry; this is almost always usb.A.cfg:

(DUT) # cat /tmp/mnt/syslinux/usb.A.cfg
label chromeos-usb.A
  menu label chromeos-usb.A
  kernel vmlinuz.A
  append init=/sbin/init rootwait ro noresume loglevel=7 noinitrd console=ttyS0  root=PARTUUID=80BF3EED-EB79-4D99-A837-06C95D8574B9 i915.modeset=1 cros_legacy cros_debug
[...]

Edit the chromeos-usb.A entry for append [...], save the file, and reboot:

(DUT) # vi /tmp/mnt/syslinux/usb.A.cfg
(DUT) # umount /tmp/mnt
(DUT) # reboot

Using modules

Loading Kernel modules from outside the root filesystem

If you need to load kernel modules from a location other than the root filesystem, module locking must be disabled. Either a kernel command line option can be used:

lsm.module_locking=0

Or, on images with dm-verity disabled (--noenable_rootfs_verification), the restriction can be disabled via the exposed sysctl:

(DUT) # echo 0 >/proc/sys/kernel/chromiumos/module_locking

Blocking kernel modules for individual overlays

If you need to block kernel modules for specific overlays, modify the overlay-/chromeos-base/chromeos-bsp-/chromeos-bsp--.ebuild file.

Add the following two lines to the end of the src_install() function:

insinto "/etc/modprobe.d"
doins "${FILESDIR}/<blocklist>"

The ${FILESDIR} variable points to the files/ directory within the chromeos-bsp-<name>/ directory. Within this directory, add your <blocklist> (ex cros-blocklist.conf).

For each kernel module you wish to block, add the following line to <blocklist>:

blacklist <module name>

You can also use # comments within these files to explain why the kernel module needs to be blocked.

Creating changelists (CLs)

When adding new files to the kernel, please add a regular Google copyright header to them. In particular this is true for any code that will eventually find its way upstream (which should include practically everything we do). The main reason for this is that there's no concept of "The ChromiumOS Authors" outside of our project, since it refers to the AUTHORS file that isn't bundled with the kernel.

Each file type has its own SPDX comment format, discussed here:

C header files:

/* SPDX-License-Identifier: GPL-2.0 */
/*
 * <short description>
 *
 * Copyright 2019 Google LLC.
 */

C source files:

// SPDX-License-Identifier: GPL-2.0
/*
 * <short description>
 *
 * Copyright 2019 Google LLC.
 */

For reference, old drivers already existing in upstream might still have the full text format, which would look like below.

/*
 * Copyright 2018 Google LLC.
 *
 * This software is licensed under the terms of the GNU General Public
 * License version 2, as published by the Free Software Foundation, and
 * may be copied, distributed, and modified under those terms.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
 * GNU General Public License for more details.
 */

(Compile) test

Make sure that your patch builds fine with allmodconfig:

mkdir -p ../build/x86-64 ../build/arm64
# Native build (x86-64)
make O=../build/x86-64 allmodconfig
make O=../build/x86-64 all -j50 2>&1|tee ../build/x86-64/build.log
# arm64 build
CROSS_COMPILE=aarch64-cros-linux-gnu- ARCH=arm64 O=../build/arm64 make allmodconfig
CROSS_COMPILE=aarch64-cros-linux-gnu- ARCH=arm64 O=../build/arm64 make -j64 >/dev/null

Test build with ChromeOS config:

cd src/third_party/kernel/v4.19
git checkout linux-next/master
# Checkout config options only
git checkout m/master -- chromeos
# Normal emerge
(chroot) emerge-${BOARD} -av chromeos-kernel-4_19

Commit messages & summary lines (CHROMIUM, UPSTREAM, FROMLIST, BACKPORT)

See the Kernel Design page for some more details.

For changes which cannot be submitted upstream to the official Linux Kernel repository, the commit message is important. We use the following conventions:

An example subject line is: CHROMIUM: ARM: tegra: Add initial support for aebl

If not sure, use git log to check commit messages of earlier commits for the same file or other files in the same directory.

Do not include configuration changes (i.e. changes to files within chromeos/config) with other code changes. See the next section for these.

Files may not be suitable for submission upstream because they have ChromiumOS-specific information, or may be based on other changes which are local to the ChromiumOS project. Such changes may not be upstreamed, but the ChromiumOS project team will continue to maintain the changes.

Configuration Changes

When a commit involves configuration changes, make sure that any code changes are separated out into a different commit. The configuration commit should contain only changes to files within the chromeos/config directory tree.

The commit message should start with CHROMIUM: config:

An example message is: CHROMIUM: config: enable aebl config

Committing

See the Contributing Guide for details on how to upload your changes, get them tested & reviewed, and ultimately get them into the tree.

Debugging

Finding issues

So the first step is to figure out what are the problems:

Enabling crash collection

Run the following commands on the target. This needs to be done just once after an install.

touch /var/lib/crash_sender_paused
touch /home/chronos/"Consent To Send Stats"
chown chronos:chronos /var/lib/crash_sender_paused
chown chronos:chronos /home/chronos/"Consent To Send Stats"
sync; sync; sync

The crashes will then appear in /var/spool/crash.

printk debugging

One advanced debugging technique is to use printk and other syslog output functions to tell you what your code is doing.

Sometimes adding too many printks changes behaviour (Heisenbug), or makes the system unusable, so be careful where you put them (probably a bad idea to put them on every timer tick for example, or on each incoming network packet in a wifi driver). If such granularity is needed however, there are some options:

Seeing early debug messages

If you need to see kernel log messages (e.g., over UART) before the full console driver is running, earlyprintk or earlycon may help you. Find more info in the kernel parameters guide.

Note that unlike with earlyprintk, you often don't need any hardware-specific arguments to use earlycon -- you only need to add earlycon to the kernel command line. The kernel can pick up the appropriate console parameters from either the Device Tree (via /chosen/stdout-path) or ACPI (via the SPCR table).

*** note Pitfall: ChromeOS kernel command lines typically include an empty console= parameter by default, which prevents directing kernel logs to the default console (earlycon or otherwise). Remove this if you want to direct kernel logs to your console.

Caveats apply: architecture and driver support varies. For example, ACPI/SPCR earlycon support is not fully integrated in ChromeOS as of this writing.

Dynamic Debugging (dev_dbg / pr_debug)

Dynamic debugging allows one to enable/disable debugging messages in kernel code at runtime (e.g., calls to dev_dbg or pr_debug).

Enabling

Using dynamic debugging requires the CONFIG_DYNAMIC_DEBUG config option to be enabled. By default dynamic debug is disabled on ChromeOS.

Flag USE=dyndebug will enable dynamic debug on the ChromeOS kernel ebuilds.

If using menuconfig, the following enables it:

Kernel hacking
     ---> printk and dmesg options
          ---> [*] Enable dynamic printk() support

Once the kernel is compiled with CONFIG_DYNAMIC_DEBUG, you can use the following commands to control the output.

Enable all dynamic debugging
echo "+p" > /sys/kernel/debug/dynamic_debug/control
Disable all dynamic debugging
echo "-p" > /sys/kernel/debug/dynamic_debug/control
Enable dynamic debugging for specific modules
echo "module cros_ec_spi +p" > /sys/kernel/debug/dynamic_debug/control
echo "module cros_ec_proto +p" > /sys/kernel/debug/dynamic_debug/control
View all of the individual statements that can be enabled
cat /sys/kernel/debug/dynamic_debug/control

See Dynamic Debug for complete details and syntax.

ftrace debugging

Another example:

cd /sys/kernel/debug/tracing
# Sample output: blk function_graph function nop. These are valid values you can echo into current_tracer
cat available_tracers

# By default this should output 'nop'
cat current_tracer

# function_graph is useful too
echo "function" > current_tracer

# This should output "all functions enabled" by default
cat set_ftrace_filter

# You can also append with "echo *nl80211* >> set_ftrace_filter"
echo *nl80211* > set_ftrace_filter

# Should be the number of functions enabled.
wc -l set_ftrace_filter

# Clear out the tracing pipe of the previous junk. You will need to Ctrl-C kill this after a while
cat trace_pipe > /dev/null

# You should see nothing, now start performing actions that will lead to your module/code being called.
cat trace_pipe

Another ftrace article: https://lwn.net/Articles/370423/

Other tricks:

Getting backtraces with BUG/WARN

BUG/WARN and friends provide nice backtraces. These can be very useful for figuring out what code path is triggering a hard to reproduce issue.

Decoding backtraces

~/chromiumos/src/platform/dev/contrib/kernel_decode_stack -b kukui

Sometimes gdb is more useful (aarch64, update as needed):

aarch64-cros-linux-gnu-gdb /build/kukui/usr/lib/debug/boot/vmlinux
disas /m function

Debugging kernel crashes

TODO: This is anecdotal, and may not be an optimal or fully correct solution. Please verify and remove the TODO.

You have a few options:

1. Googler-only: Check out go/xstability. Clicking on sample crashes here go/crash with the filter set for that particular crash. Click on a sample report. Below the "Report Time" and "Client ID" you should "Files" with a link to "upload_file_kcrash". This has the stack trace towards the end.

TODO: Add more details on this

2. If you are debugging a local crash on your device, look for the crash in /var/log/messages (unlikely that it would be saved there) or /sys/fs/pstore/console-ramoops. You may see some symbols preceded by question marks in the stack trace, something like the below.

<5>[ 25.801932] Call Trace:
<5>[ 25.801947] [<ffffffffc008c064>] ieee80211_amsdu_to_8023s+0xec/0x2df [cfg80211]
<5>[ 25.801968] [<ffffffffc02af0f2>] __iwl7000_ieee80211_sta_ps_transition+0x154a/0x21dc [iwl7000_mac80211]
<5>[ 25.801987] [<ffffffffc03154e4>] ? iwl_mvm_send_lq_cmd+0x8e/0x9c [iwlmvm]
<5>[ 25.802003] [<ffffffffc0324409>] ? iwl_mvm_rs_tx_status+0xf9c/0x1f5cd /4 [iwlmvm]
<5>[ 25.802023] [<ffffffffc02b06f2>] __iwl7000_ieee80211_mark_rx_ba_filtered_frames+0x96e/0xcb0 [iwl7000_mac80211]
<5>[ 25.802041] [<ffffffff9e4ee0f0>] ? kmem_cache_free+0x8a/0xc5
<5>[ 25.802059] [<ffffffffc02b08a1>] __iwl7000_ieee80211_mark_rx_ba_filtered_frames+0xb1d/0xcb0 [iwl7000_mac80211]
<5>[ 25.802080] [<ffffffffc02b0dc6>] __iwl7000_ieee80211_rx_napi+0x392/0x46a [iwl7000_mac80211]
<5>[ 25.802098] [<ffffffffc0316578>] iwl_mvm_rx_rx_mpdu+0x749/0x78b [iwlmvm]
<5>[ 25.802113] [<ffffffffc0310f16>] iwl_mvm_enter_d0i3+0x359/0xe7f [iwlmvm]
<5>[ 25.802128] [<ffffffffc023d504>] iwl_pci_unregister_driver+0xfdb/0x1439 [iwlwifi]
<5>[ 25.802143] [<ffffffffc023e883>] iwl_pcie_irq_handler+0x57d/0x7d1 [iwlwifi]
<5>[ 25.802157] [<ffffffff9e48c255>] ? free_irq+0x8a/0x8a
<5>[ 25.802168] [<ffffffff9e48c272>] irq_thread_fn+0x1d/0x3c
<5>[ 25.802179] [<ffffffff9e48be1a>] irq_thread+0x117/0x21a
<5>[ 25.802191] [<ffffffff9e921dda>] ? __schedule+0x589/0x5d3
<5>[ 25.802202] [<ffffffff9e48b863>] ? kzalloc.constprop.37+0x1c/0x1c
<5>[ 25.802214] [<ffffffff9e48bd03>] ? irq_thread_check_affinity+0x8f/0x8f
<5>[ 25.802227] [<ffffffff9e45183b>] kthread+0xc0/0xc8
<5>[ 25.802238] [<ffffffff9e45177b>] ? __kthread_parkme+0x6b/0x6b
<5>[ 25.802249] [<ffffffff9e92389c>] ret_from_fork+0x7c/0xb0
<5>[ 25.802259] [<ffffffff9e45177b>] ? __kthread_parkme+0x6b/0x6b

There are a few ways you can resolve the "? some_symbol + 0xoffset" format into a line of source code. For example, you can enter the cros_sdk chroot and load up the vmlinux file in gdb.

Be careful to use the gdb binary from the cross-toolchain of the $BOARD you are debugging on. TODO(crbug.com/995661): ChromiumOS runs 32-bit ARM userspace on ARM64 boards and there is no good programmatic way of getting the right gdb tuple in such case, so just use aarch64-cros-linux-gnu-gdb with them for the time being.

(cr) user@machine /build/samus $ gdb="$(portageq-$BOARD envvar CHOST)-gdb"
(cr) user@machine /build/samus $ file /build/$BOARD/usr/lib/debug/boot/vmlinux | grep -q aarch64 && gdb="aarch64-cros-linux-gnu-gdb"
(cr) user@machine /build/samus $ ${gdb} /build/$BOARD/usr/lib/debug/boot/vmlinux

Next, use the list command to print the code at given address

Reading symbols from /build/samus/usr/lib/debug/boot/vmlinux...done.
(gdb) list *( iwl_mvm_send_lq_cmd+0x8e)
0x12b5 is in iwl_mvm_send_lq_cmd (/mnt/host/source/src/third_party/kernel/v4.14/drivers/net/wireless/iwl7000/iwlwifi/mvm/utils.c:752).
747  };
748
749  if (WARN_ON(lq->sta_id == IWL_MVM_STATION_COUNT))
750  return -EINVAL;
751
752  return iwl_mvm_send_cmd(mvm, &cmd);
753  }
754
755  /**
756  * iwl_mvm_update_smps - Get a request to change the SMPS mode
(gdb)

3. A slightly more tedious way of getting symbols is to symbolize the whole kernel using objdump -

cd /build/samus/var/cache/portage/sys-kernel/chromeos-kernel-4_14
# Pick a proper output location - the resulting file is > 2GB in size!
objdump -e vmlinux > /tmp/objdump-output.txt
grep your_kernel_symbol /tmp/objdump-output.txt

More information here and here.

Debugging with KGDB/KDB

KGDB is an in-kernel debugger implementation, which allows developers to attach a local GDB instance on their development machine to debug the kernel on a remote test machine, using a serial connection. You can find some information here:

To use KGDB with ChromiumOS requires two steps for the test machine:

  1. Enable KGDB in the kernel configuration
  2. Set kernel parameters to enable the appropriate debug console

Step 1 can be done by building with USE="kgdb":

USE="kgdb vtconsole pcserial" emerge-${BOARD} chromeos-kernel-${VER}

(pcserial is only needed on x86 builds)

Step 2 can be done by adding kgdboc=$TTY to the kernel config.txt, where $TTY depends on the board -- for many systems, this should be ttyS0, but some ARM SoCs use ttyS2.

Once you configure the target device, you can break into debug mode with the Alt-SysRq-G shortcut (see Linux SysRq docs); e.g.:

then attach to the target console with your cross-targeted GDB:

# If running servod, TTY_PORT can be derived from dut-control:
TTY_PORT="$(dut-control cpu_uart_pty | cut -d: -f2)"
# ${CROSS_ARCH} is something like x86_64-cros-linux-gnu-,
# aarch64-cros-linux-gnu-, etc., depending on your target kernel's
# architecture.
(chroot) $ ${CROSS_ARCH}-gdb \
         /build/${BOARD}/usr/lib/debug/boot/vmlinux \
         -ex "target remote ${TTY_PORT}" \
         -ex "set remotebaud 115200"

Once attached, you can use standard GDB commands, though report has it that not everything works well (e.g., stepping and breakpoints) -- YMMV.

Besides basic GDB commands, you can make use of Linux-specific KDB commands via the monitor command. For more info, run this while attached:

(gdb) monitor help
Command Usage Description
----------------------------------------------------------
[...]

Debugging modules

You can get a list of modules and addresses in kgdb with monitor lsmod. Then you can add symbol files using the base addresses found there:

add-symbol-file /build/${BOARD}/usr/lib/debug/lib/modules/5.4.109/kernel/drivers/net/wireless/marvell/mwifiex/mwifiex.ko.debug 0xbf077000
add-symbol-file /build/${BOARD}/usr/lib/debug/lib/modules/5.4.109/kernel/drivers/net/wireless/marvell/mwifiex/mwifiex_sdio.ko.debug 0xbf0a0000

If you're in kgdb and want to get back to kdb:

maintenance packet 3
Ctrl-Z
kill -9 %

QEMU notes

Debugging a QEMU system (such as one launched via cros_vm) requires a few tweaks.

For one, you need to establish a virtual serial console by adding -serial pty to your QEMU command. For example:

(chroot) $ cros_vm [...] --start --qemu-args="-serial pty"

This will eventually print out a line like:

char device redirected to /dev/pts/10 (label serial1)

Thus, we use /dev/pts/10 for our ${TTY_PORT} when invoking gdb's target remote ${TTY_PORT}.

Then, this new console typically becomes ttyS1 within the VM, because cros_vm establishes the first serial device to monitor kernel logs. So your modified kernel command line should include kgdboc=ttyS1.

*** note QEMU also has its own GDB support, which can be easier to set up than KGDB. See the QEMU GDB docs.

i.e., cros_vm [...] --start --qemu-args="-s -S", and ${CROSS_ARCH}-gdb [...] -ex "target remote localhost:1234".

Multiplexing the console

Easy method

Use dut-console script with -k parameter:

dut-console -p 9999 -c cpu -k

dut-console will also print instructions on how to attach gdb from inside the chroot.

Detailed method

If you want to use both KGDB and a standard serial console over the same serial port, you need to run a program like kdmx or agent-proxy to multiplex your connection. Both can be found at:

https://kernel.googlesource.com/pub/scm/utils/kernel/kgdb/agent-proxy/

kdmx is probably easier to deal with. If your serial port is at /dev/pts/80, you can start it with:

agent-proxy/kdmx/kdmx -n -b 115200 -p /dev/pts/80 -s /tmp/kdmx_tty_

You can find the ttys to use for console in /tmp/kdmx_tty_trm and for gdb in /tmp/kdmx_tty_gdb. Thus connect to the terminal with:

cu --nostop -l $(cat /tmp/kdmx_tty_trm)

and attach gdb with:

${CROSS_ARCH}-gdb \
	/build/${BOARD}/usr/lib/debug/boot/vmlinux \
	-ex "target remote $(cat /tmp/kdmx_tty_gdb)"

If telnet is more your style, use agent-proxy with:

agent-proxy 127.0.0.1:5510^127.0.0.1:5511 0 /dev/pts/80,115200

Then connect to the terminal with:

telnet localhost 5510

and attach gdb with:

${CROSS_ARCH}-gdb \
         /build/${BOARD}/usr/lib/debug/boot/vmlinux \
         -ex "target remote localhost:5511"

Errata

Bisecting a stable branch merge

To bisect along the upstream stable branch, first identify and test the merge commit and the chromeos branch.

Merge:

commit f5edda0c2aefe22f338c3a00c0aa52161976d4b1
Merge: ce070a331d16 399849e4654e
Author: Guenter Roeck <groeck@chromium.org>
Date:   Wed Jul 1 08:17:19 2020 -0700

    CHROMIUM: Merge 'v4.19.131' into chromeos-4.19

    Merge of v4.19.131 into chromeos-4.19

    Changelog:
    ----------------------------------------------------------------
    Aaron Plattner (1):
          ALSA: hda: Add NVIDIA codec IDs 9a & 9d through a0 to patch table

    Aditya Pakki (1):
          rocker: fix incorrect error handling in dma_rings_init

ChromeOS:

commit ce070a331d1697048ebfdb9011be299bc77940dc
Author: Benjamin Gordon <bmgordon@chromium.org>
Date:   Thu Mar 26 13:23:28 2020 -0600

    CHROMIUM: LSM: Convert symlink checks to MNT_NOSYMFOLLOW

Start a regular git bisect, identifying the merge as bad and chromeos branch as good:

git bisect
git checkout f5edda0c2aef
# Build and test
git bisect bad
git checkout ce070a331d16
# Build and test
git bisect good

Git is smart enough to bisect along the upstream branch, rooted at the common branch point (the previous upstream merge).

At each bisection point, you need to merge in the chromeos branch. The device may not boot and function correctly if you do not do this:

git merge --no-commit ce070a331d16
# Resolve merge conflicts

# Build/deploy/test kernel

# Reset git state and continue bisection
git reset --hard
git bisect [good|bad]

Performance analysis

Use perf! One easy perf method is to use perf top on the target device to look for hotspots.

Another quick way to profile is to use cycle profiling. On the DUT, use

perf record -a -g -e cycles

with an optional time period and/or command if you want to profile a specific binary. Then on your workstation, generate the report with something like:

perf report -g --symfs /path/to/chroot/build/$BOARD

This should resolve symbols and give you a good idea of where cycles were spent during the perf record above.

On specific platforms, other tools may be available. For example on Intel, both socwatch and VTune can be used provided the proper kernel drivers are loaded.

More information on profiling Chrome and (for Googlers) uploading to pprof can be found in CPU Profiling Chrome.

FIXME: add some stuff on importing perf output into flamegraph tool, etc.

Upstream development

How do I backport an upstream patch?

Let's suppose you've spotted a juicy new commit in Linus's upstream linux kernel tree that you just must have. Instead of creating a new branch and manually applying the changes, use git cherry-pick to do it for you. In addition, the repository maintainers appreciate it if the cherry-picked commit still contains the original author and git hash of the original upstream commit.

For "simple" UPSTREAM cherry-picks, one should first try using fromupstream.py script to prepare CLs "automagically". Look below for examples.

Otherwise, the follow steps use git cherry-pick -x to do most of the work:

NAME
        git-cherry-pick - Apply the changes introduced by some existing commits

SYNOPSIS
        git cherry-pick [--edit] [-n] [-m parent-number] [-s] [-x] [--ff] <commit>...

DESCRIPTION
	Given one or more existing commits, apply the change each one
	introduces, recording a new commit for each. This requires your working
	tree to be clean (no modifications from the HEAD commit).

OPTIONS
...
    -x
	When recording the commit, append to the original commit message a note
	that indicates which commit this change was cherry-picked from. Append
	the note only for cherry picks without conflicts. Do not use this
	option if you are cherry-picking from your private branch because the
	information is useless to the recipient. If on the other hand you are
	cherry-picking between two publicly visible branches (e.g. backporting
	a fix to a maintenance branch for an older release from a development
        branch), adding this information can be useful.

First, add Linus's tree as a remote to the chromium-os kernel tree (assuming the chromium-os root is ~/chromiumos):

cd ~/chromiumos/src/third_party/kernel/<version>
git remote add upstream git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
git remote update

This will take a little while as git fetches all upstream commits. Luckily, git is smart and won't refetch commits already in the chromium-os tree.

Once the tree is updated, take a brief look at whats been happening upstream recently to a particular path (--oneline shows short-form upstream hashes and the brief commit message):

git log --oneline upstream/master /path/of/interest

We can view that juicy commit using its upstream hash:

git show <upstream_commit_hash>

To backport the commit to the chromium-os tree, first start a new branch from the current Tip of Tree (ToT). Then cherry-pick with -x to preserve the original author and hash, and -s to sign-off-by the commit:

repo sync .
repo start my_upstream_commit .
git cherry-pick -x -s <upstream_commit_hash>

Add TEST= and BUG= lines at the bottom of the patch description. Also, remember to keep the patch subject intact with only an addition of UPSTREAM: or BACKPORT: as a new prefix. Use UPSTREAM: if you are applying an upstream patch as-is, or BACKPORT: if you had to change the code to make it run with an older kernel version.

NOTE: Do not make functional changes to backported patches! Downstream changes in backports should be strictly limited to resolving conflicts. If you need to make a functional change to a backport (ie: changing a delay, tweaking a default value, etc), backport the change from upstream as-is and follow up with a separate patch with CHROMIUM prefix.

Now, the upstream commit is on its own branch, let's upload it to gerrit, like usual:

repo upload .

This will generate a gerrit change for review.

After review, submit the patch in gerrit like usual.

UPSTREAM, BACKPORT, FROMLIST, and you

When backporting patches from Linus's kernel tree, you should tag your patch with UPSTREAM (or BACKPORT, if modifications were needed). But what about patches that are "on their way" upstream, but haven't been merged for an official release yet?

Previous discussions defining this practice:

Picking patches from mailing lists / upstream

UPSTREAM

../../../platform/dev/contrib/fromupstream.py -b b:195958998 \
  -t "Deploy reven kernel on RFE type 2 Realtek wireless and test that wireless works" \
  linux://5d6651fe85837b11564a2e2c3c6279c057d078d6

FROMGIT

../../../platform/dev/contrib/fromupstream.py -b b:123489157 \
  -t "Deploy kukui kernel with USE=kmemleak, no kmemleak warning in __arm_v7s_alloc_table" \
  'git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu.git#next/032ebd8548c9d05e8d2bdc7a7ec2fe29454b0ad0'

FROMLIST

Add project url in ~/.pwclientrc

[options]
default=kernel

[kernel]
url=https://patchwork.kernel.org/

[lore]
url=https://lore.kernel.org/patchwork/xmlrpc/

Then run:

../../../platform/dev/contrib/fromupstream.py -b b:132314838 -t "no crash with CONFIG_FAILSLAB" 'pw://10957015'
# or
../../../platform/dev/contrib/fromupstream.py -b b:132314838 -t "no crash with CONFIG_FAILSLAB" 'pw://kernel/10957015'

For patches that go only to LKML (for which there is no patchwork), replace the "pw://" argument with "msgid://". That'll still try to look for a patchwork but if there is none then it'll just give the lore link.

Submitting patch series by gerrit cmd tool

In CrOS chroot (gerrit deps prints dependencies from top to bottom, so its better to use tac so that the bottom-most CL is set to ready first):

# If the CL of interest is HEAD, else substitute the gerrit CL number.
cl=$(git log -1 --format='%(trailers:key=Change-Id,valueonly)')
deps=( $(gerrit --raw deps "${cl}" | tac) )
gerrit label-v "${deps[@]}" 1
gerrit label-cq "${deps[@]}" 2

Downloading patches from upstream

If you want to review or apply a patchset from a mailing list that you are not subscribed, you can download it from lore. Lore is an email archive maintained by kernel.org.

You can reach lore via their web interface: https://lore.kernel.org . Simply navigate to the mail thread that you are interested and click on the mbox.gz link to obtain the mailbox file.

You can also use the b4 console tool. Simply run b4 mbox $Message-id and you will obtain the mailbox file. You can also apply the all patches with b4 using b4 am $Message-id. b4 will run some attestations of the patches.

Review patches from a mailbox file

If you use mutt, you can simply run mutt -f patch.mbox.

If you prefer to forward the emails to an email account with IMAP access, you can use imap-upload:

  1. Clone the imap-upload repo.
  2. python2.7 ./imap_upload.py patch.mbox --gmail
  3. Use @chromium.org account.
  4. Find the email in your mailbox, and reply!

How do I build an upstream kernel?

There are various ways of building mainline Linux, but it can be useful to use existing ChromeOS tooling to build a non-ChromeOS-flavored kernel. See the cros-kernel eclass documentation for tips on how to use the "fallback" configuration system to build any (e.g., mainline) kernel tree within the existing Portage-based flow.

*** note Note: ChromeOS kernels often support hardware that is not yet supported in an upstream kernel release. Ensuring hardware support for your system is not covered here.

How do I send a patch upstream?

Changes to parts of the kernel which are not purely ChromeOS-specific should be upstreamed where possible. This includes just about any part of the kernel: ARM- and x86-specific changes, driver patches and changes within the main kernel and mm source. You can start with a code review if you like. Take a look on the kernel mailing list to get a feel for how people submit and review patches.

1) Prepare your local repository state

To upstream, create a remote to track upstream.

For example the main kernel:

git remote add upstream git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
git fetch upstream
git checkout -b send-upstream upstream/master

You can then create a commit within this branch. This can be done either by cherry-picking the commit from another branch and perhaps changing the commit message:

git cherry-pick my-change
git commit --amend
# edit the message and save

or using git am to turn a patch into a commit:

git am my-change.patch

or manually applying a patch, and then committing:

patch -p1 < my-change.patch
git add ...
git commit
# create a suitable message

2) Check if your patches are correctly formatted

There are two aspects of having correct patches to send upstream: not having ChromiumOS-specific details, and meeting all the Linux kernel requirements.

Remove ChromiumOS-specific Details

Verifying these details is as simple as loading the patch file in your favorite editor. Edit the file manually to become compliant; this will, of course, have no affect on the source or commit message stored by git.

Once all of the above is true, you can move on to checking for compliance with the Linux Kernel guidelines.

Check for Compliance with Linux Kernel Requirements

First off, make sure the Kernel builds with patch applied.

For style, the patman tool (see below) will automatically run checkpatch.pl on your change. If you'd like to run the checkpatch.pl tool manually, here'a an example workflow:

git format-patch HEAD~
scripts/checkpatch.pl 0001-my-change.patch
# make improvements
git add ...
git commit --amend
# rinse and repeat

3a) Option 1: Send out the patch using b4

Currently, it is recommended to use b4 to send your patch upstream because it is maintained by the people behind the kernel infra.

You may use the original documentation. Alternatively, this tutorial is also helpful.

While following this workflow, keep the following in mind:

3b) Option 2: Send out the patch using patman

It is possible to send out patches using git send-email manually, but for most usecases using the patman CLI is sufficient and can save a lot of time.

(See the next section for first-time credential setup for using patman and git send-email.)

Patman automates patch creation, checking, change list creation, cover letter, sending to the mailing list, etc. You can find patman in the U-Boot tree (src/third_party/u-boot/files/tools/patman). It usually should be run outside of the chroot, so you could create an alias, or a symlink to somewhere in your path:

alias patman='~/chromiumos/src/third_party/u-boot/files/tools/patman/patman'
# or
ln -s ~/chromiumos/src/third_party/u-boot/files/tools/patman/patman ~/bin

To use patman, amend your top commit to have the line:

Series-to: LKML <linux-kernel@vger.kernel.org>
Series-cc: (anyone you want to Cc all patches in the series to)

Then type:

patman -n

to generate patches, check that they will go to the right place, and send them. Or:

patman

to generate patches and send them.

Various options are available. Particularly useful ones are:

Full documentation is available in the README (patman -h) or here. Take a look at the automated change list creation and the alias support also.

Troubleshooting

You may get some python ModuleNotFoundError errors when running patman. This may be resolved by pip install pygit2 requests

First-Time Email Setup

If you have never sent email from the command-line, or from git send-email, then there is some setup required.

1. Install git send-email
2. Decide on the email address, password, and mail server to use

You must configure git's send-email command with the details of how to send email from your identity. The rest of this section will explain how to set up a google-mail based account (e.g. an @gmail.com address, @chromium.org, etc). If you have a different mail server, please contact the system administrator (or check some help docs related to your email service) for the correct settings.

NOTE: For Googlers, note that DMARC restrictions prevent usage of your @google.com email address. Use http://go/chromium-account to obtain an @chromium.org address.

3. Set up email credentials

For google-mail-based addresses, it's recommended to use an "App Password" for convenience when storing your real password on disk is undesirable (which should be most cases). Follow these instructions to obtain an App Password, and use it as the smtppass value in the next section.

Edit your ~/.gitconfig

Open up your ~/.gitconfig file to include the following stanza:

[sendemail]
  smtpserver = smtp.gmail.com
  smtpserverport = 587
  smtpencryption = tls
  smtpuser = YOUR_EMAIL_ADDRESS
  smtppass = PASSWORD
  confirm = always

Remember to swap in the YOUR_EMAIL_ADDRESS with your full email address, and PASSWORD with your password (or App Password).

Optional: Automating the Compliance Checks

To use the following script, you will need to have created a patch file using git format-patch. Also note that you will have to recreate the patch file, and re-check your patch file each time you check in code to your source tree.

This script might be useful also, as it checks a series of patches, checks for ChromeOS-specific commit tags and prints a summary at the end. Put it in your path and run it from anywhere.

#! /bin/sh
# Pass a list of patchfiles to check for compliance

KERNEL=./scripts/
OUT=$(tempfile)
while (( "$#" )); do
	ERRCP=
	ERR=
	"${KERNEL}/checkpatch.pl" $1 || ERRCP=1
	grep BUG= $1 && ERR="$ERR BUG"
	grep TEST= $1 && ERR="$ERR TEST"
	grep "Change-Id" $1 && ERR="$ERR Change-Id"
	grep "Review URL" $1 && ERR="$ERR Review URL"
	if [ -n "${ERR}" ]; then
		echo "Bad $1 ($ERR)" >>$OUT
	else
		echo "OK $1" >>$OUT
	fi
	shift
done
cat $OUT