The WSL2-Linux-Kernel repo contains the kernel source code and configuration files for the WSL2 kernel.
This fork integrates the RamShared Hardware-Accelerated VRAM Block Driver (CONFIG_BLK_DEV_RAMSHARED=m) alongside upstream userspace block drivers (CONFIG_BLK_DEV_UBLK=m) and ZRAM writeback (CONFIG_ZRAM_WRITEBACK=y) to enable seamless 4-tier memory cascading for Linux and WSL2.
- 4-Tier Progressive Memory Hierarchy: Physical RAM (16 GB) ➔ ZRAM LZ4 (1,024 MB) ➔ GPU VRAM via PCIe DMA (4,096 MB) ➔ Host SSD.
- Ultra-Low Latency & High Bandwidth: 8.74 GiB/s Host-to-Device DMA, 231 µs median 4KB Direct I/O latency, and 16.4 TB/s instant atomic memory reclaim.
- Anti-Hang Immunity: Fully qualified under 100% full-capacity saturation holding 9,160 MB active swap across 40 continuous cycles without system lockups or kernel panics.
If you discover an issue relating to WSL or the WSL2 kernel, please report it on the WSL GitHub project. It is not possible to report issues on the WSL2-Linux-Kernel project.
If you're able to determine that the bug is present in the upstream Linux kernel, you may want to work directly with the upstream developers. Please note that there are separate processes for reporting a normal bug and a security bug.
Is there a missing feature that you'd like to see? Please request it on the WSL GitHub project.
If you're able and interested in contributing kernel code for your feature request, we encourage you to submit the change upstream.
Instructions for building an x86_64 WSL2 kernel with an Ubuntu distribution using bash are as follows:
-
Install the build dependencies:
sudo apt install build-essential flex bison dwarves libssl-dev libelf-dev cpio qemu-utils rsync -
Modify WSL2 kernel configs (optional):
make menuconfig KCONFIG_CONFIG=Microsoft/config-wsl -
Build the kernel using the WSL2 kernel configuration and put the modules in a
modulesfolder under the current working directory:
make KCONFIG_CONFIG=Microsoft/config-wsl && make INSTALL_MOD_PATH="$PWD/modules" modules_installYou may wish to include
-j$(nproc)on the firstmakecommand to build in parallel. -
Install the kernel's UAPI headers into a
headersfolder:
make headers_install INSTALL_HDR_PATH="$PWD/headers" -
Build the
perftooling into aperffolder:
make -C tools/perf NO_JEVENTS=1 NO_JVMTI=1 NO_LIBTRACEEVENT=1 install DESTDIR="$PWD/perf" prefix=/
Then, you can use a provided script to create a VHDX containing the modules, headers, and perf
tooling:
./Microsoft/scripts/gen_artifacts_vhdx.sh "$PWD/modules" "$PWD/headers" "$PWD/perf" $(make -s kernelrelease) modules.vhdx
To save space, you can now delete the compilation artifacts:
make clean && rm -r "$PWD/modules" "$PWD/headers" "$PWD/perf"
If you prefer, you can also build the VHDX manually as follows. WSL expects the artifacts laid
out under <kernelrelease>/{modules,linux-headers,perf}, so first assemble a staging directory:
-
Stage the modules, headers, and perf tooling under the kernel release directory:
release=$(make -s kernelrelease) mkdir -p "$PWD/staging/$release/linux-headers" cp -r "$PWD/modules/lib/modules/$release" "$PWD/staging/$release/modules" rm -f "$PWD/staging/$release/modules/build" "$PWD/staging/$release/modules/source" cp -r "$PWD/headers/." "$PWD/staging/$release/linux-headers" cp -r "$PWD/perf" "$PWD/staging/$release/perf" -
Calculate the image size (plus 256MiB for slack):
image_size=$(du -bs "$PWD/staging" | awk '{print $1;}'); image_size=$((image_size + (256 * (1<<20)))); -
Build a populated ext4 image:
mke2fs -L '' -d "$PWD/staging" -N $(( $(find "$PWD/staging" | wc -l) + 4096 )) -b 1024 -t ext4 "$PWD/modules.img" $((image_size / 1024)) -
Convert the img to VHDX:
qemu-img convert -O vhdx "$PWD/modules.img" "$PWD/modules.vhdx" -
Clean up:
rm "$PWD/modules.img" # optionally the $PWD/staging, $PWD/modules, $PWD/headers, and $PWD/perf dirs too
Please see the documentation on the .wslconfig configuration file for information on using a custom built kernel.