Re: [PATCH v2 5/5] arm64: defconfig: Switch Ethernet drivers to modules
Linus Walleij <[email protected]>
| Newsgroups | dev.linux.lists.soc,org.infradead.lists.linux-arm-kernel,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <CAD++jLkjOzfP5MLB0wuOUouM3keE1osBEJ+PWC84dwtd6h4SOg@mail.gmail.com> |
On Wed, Apr 29, 2026 at 8:50 PM Krzysztof Kozlowski <[email protected]> wrote: > Development of Linux kernel progressed over last 10 years and it is easy > to generate now initramfs for building own kernel, e.g. with Yocto or > mkosi. Therefore for a few years of reviews on mailing lists, all new > options enabled in arm64 defconfig were with assumption of having > initramfs which can load bare minimum of modules to mount filesystem > from network or disk. Basically network driver as built-in is not > anymore essential to boot the system, so switch almost all Ethernet > drivers to modules to save on kernel image size. > > Similarly 9P network filesystem for QEMU, especially that testing kernel > unuder QEMU does not have any size or build process constraints and can > use initramfs with -initrd argument. Note that having network drivers > does not break NFS root, because whatever loading method, e.g. TFTP, > which brought the kernel image can bring also the initramfs with network > adapters (and I have been using such method for years for my Samsung > boards). > > Notable exceptions / diff explanations: > > 1. Mark I2C as built-in, used by CONFIG_IGB as a module. > > 2. CONFIG_BCM4908_ENET and CONFIG_BCMASP appear in the diff, because > they were default=y (via ARCH_BCMBCA or ARCH_BCM_IPROC). > > 3. CONFIG_HNS3_HCLGE and CONFIG_HNS3_ENET are removed, because they are > default=m. > > Moving code to modules has positive impact on kernel image size, thus > boot time of all users not using above drivers and ability to flash > fixed-size boot partitions. > > Old Image size: 41.2 MiB (Image.gz: 14.8 MiB) > New Image size: 39.1 MiB (Image.gz: 13.8 MiB) > > bloat-o-meter of vmlinux: > add/remove: 3/6972 grow/shrink: 6/45 up/down: 66659/-2333659 (-2267000) > > Signed-off-by: Krzysztof Kozlowski <[email protected]> OK it's probably the right thing to do, no reaction from the netdev people so let's go. Reviewed-by: Linus Walleij <[email protected]> Yours, Linus Walleij