Hi Andy, On Mon, Jan 27, 2020 at 7:05 PM Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> wrote: > > On Fri, Jan 24, 2020 at 09:58:59PM +0200, Andy Shevchenko wrote: > > When build on, for example, x86 using `make O=... -j64` the version > > in the built kernel comes from include/generated/compile.h, which is: > > > > #define UTS_VERSION "#351 SMP Fri Jan 24 18:46:34 EET 2020" > > > > While at the end the x86 specific Makefile prints the contents of > > the .version file: > > > > Kernel: arch/x86/boot/bzImage is ready (#352) > > > > Obviously the latter is not true. This happens because we first > > check compile.h and update it and then generate new version, which is > > incorrect flow: > > > > CHK include/generated/compile.h > > UPD include/generated/compile.h > > ... > > GEN .version > > > > In order to fix this, move the version generation from link-vmlinux.sh > > to scripts/version.sh and re-use it in init/Makefile. > > > > Additionally provide a unified way to get the current version of the build > > and use this in few callers. This will respect the KBUILD_BUILD_VERSION > > in case it's provided. > > Hmm... It looks like a mess in my build tree. > I have to setup more experiments here. > Sorry for the late reply. I remember I was also hit by this one month ago or so. I did not dig into it at that time because the problem disappeared after doing something. Today, I took a look at this again. This is a regression caused by 56d589361572 This patch should fix it: https://patchwork.kernel.org/patch/11374047/ -- Best Regards Masahiro Yamada
