I build gentoo-sources with llvm and full lto. After switching the llvm tools from 18 to 19.1.6, I get the following warnings when building the kernel (both 6.12.7 and 6.12.8): LD vmlinux.o ld.lld: warning: <unknown>:0:0: stack frame size (2056) exceeds limit (2048) in function 'dml_core_mode_support' vmlinux.o: warning: objtool: start_unlink_intr+0xb7: can't find switch jump table I'm not absolutely sure and I don't have llvm 18 any longer to test, but I'm quite confident that both warnings were not present with llvm 18. Moreover, on every boot the syslog contains a warning which was definitely not present with llvm 18: kernel: WARNING: WARNING: Bad or missing .orc_unwind table. Disabling unwinder.
Created attachment 915755 [details] Kernel .config
As advised, please report this upstream to the LKML (lkml.org) and CC Maintainers AMD DISPLAY CORE - DML M: Chaitanya Dhere <chaitanya.dhere@amd.com> M: Jun Lei <jun.lei@amd.com> S: Supported F: drivers/gpu/drm/amd/display/dc/dml/ F: drivers/gpu/drm/amd/display/dc/dml2/
I've already reported both problems upstream. The stack frame size warning is the smaller problem, because it has no consequences. There exists an inofficial patch for it, I've no idea when it will be applied to mainline, because the amd people seem to ignore the patch and don't upstream it to amd's dc code base. The objtool warning is independent of dc and more serious, because it breaks the stack backtrace mechanism: It causes the unwind error on booting. It has been reproduced and analyzed by the llvm maintainers. A bug report has been / will be opened against llvm, but will take some time. Meanwhile, objtool will (hopefully) be patched to just ignore this kind of broken table entries produced by llvm. See https://lore.kernel.org/all/20250114171816.GA3416405@ax162/ https://lore.kernel.org/all/ee25c0b7e80113e950bd1d4c208b671d35774ff4.1736891751.git.jpoimboe@kernel.org/ https://github.com/ClangBuiltLinux/linux/issues/2048 https://github.com/ClangBuiltLinux/linux/issues/2064
Looks like a LLVM bisect resulted in a problematic commit. https://github.com/llvm/llvm-project/pull/76669 This commit was reverted, along with a few others.
(In reply to Klaus Kusche from comment #3) > I've already reported both problems upstream. > (In general, if you already have, please include those links upfront.) > The stack frame size warning is the smaller problem, > because it has no consequences. I think this isn't quite right -- it's very possible that with Clang, you get more inlining, and the function genuinely is too big.
