On Wed, Nov 22, 2023 at 10:32 PM Dan Williams <dan.j.williams@xxxxxxxxx> wrote: > It concerns me that neither linux-next nor 0day robot exposure reported > this problem. > > Does ARM32 require manual compilation coverage these days or was this > just a series of unfortunate events that the build bots missed this? It's not just ARM32, I saw it on ARM64 as well and I'm pretty sure it appears on any bare metal "none" compiler. kernel.org host "nolibc" cross compilers (Arnd makes these): https://mirrors.edge.kernel.org/pub/tools/crosstool/ and those WORK, because they use the kernel minimal libc which defines __linux__. So a "nolibc" compiler works but not "none" compilers. I think the test robots all use Arnds nolibc compilers or the compilers from distributions so they don't see this. A typical example of breaking compilers: ARMs supported "none" compilers: https://developer.arm.com/downloads/-/arm-gnu-toolchain-downloads Yours, Linus Walleij
