Hi Greg and Nick,
On 2021-02-11 10:46 -0800, Nick Desaulniers wrote:
On Thu, Feb 11, 2021 at 5:55 AM Greg Kroah-Hartman gregkh@linuxfoundation.org wrote:
On Thu, Feb 11, 2021 at 09:32:03PM +0800, Xi Ruoyao wrote:
Hi all,
The latest GNU assembler (binutils-2.36.1) is removing unused section symbols like Clang [1]. So linux-5.10.15 can't be built with binutils-2.36.1 now. It has been reported as https://bugzilla.kernel.org/show_bug.cgi?id=211693.
Xi, Happy Lunar New Year to you, too, and thanks for the report. Did you observe such segfaults for older branches of stable?
We found this issue building Linux From Scratch (http://www.linuxfromscratch.org/lfs/view/development/). We use the latest stable kernel in the project so we have no report of older branches.
I tried 5.4.97 with a configuration "looks like" the one triggers the segfault in 5.10.15. But 5.4.97 build does not segfault.
However, I guess for "some" config older branches may segfault too.
2.36 of binutils fails to build the 4.4.y tree right now as well, but as objtool isn't there, I don't know what to do about it :(
Greg, There may be multiple issues in the latest binutils release for the kernel; we should still avoid segfaults in host tools so I do recommend considering this patch for inclusion at least into 5.10.y.
I strongly agree, or it may affect "rolling-release" distros (which are likely to use binutils-2.36.1 and linux-5.10.y together) in an annoying way.
Arnd's report in https://github.com/ClangBuiltLinux/linux/issues/1207 mentions this was found via randconfig testing, so likely some set of configs is needed to reproduce reliably.
I attached a config in https://bugzilla.kernel.org/show_bug.cgi?id=211693 (also attached in this mail). It reproduces the issue for 5.10.15 very reliably. At least three of us tried this configuration on the system personally used (with binutils-2.36.1) and we got exactly same error: objtool segfaults on apic.o.
Do you have more info about the failure you're observing? Trolling lore, I only see: https://lore.kernel.org/stable/YCLeJcQFsDIsrAEc@kroah.com/ (Maybe it was reported on a different list; I only searched stable ML).