RE: Compile problems w/gcc 9.4.0 in linux-next

[Michael Kelley <mhklinux@outlook.com>]

From: Ard Biesheuvel <ardb@kernel.org> Sent: Thursday, March 20, 2025 12:20 PM
> 
> Hi Michael,
> 
> On Thu, 20 Mar 2025 at 19:29, Michael Kelley <mhklinux@outlook.com> wrote:
> >
> > Brian/Ingo/Ard -
> >
> > I'm seeing a compile problem in linux-next-20250319 with gcc 9.4.0. I've bisected
> > to a group of 6 commits that you authored:
> >
> > a9a76b38aaf5 x86/boot: Disable stack protector for early boot code
> > f58b63857ae3 x86/pvh: Use fixed_percpu_data for early boot GSBASE
> > cb7927fda002 x86/relocs: Handle R_X86_64_REX_GOTPCRELX relocations
> > 78c4374ef8b8 x86/module: Deal with GOT based stack cookie load on Clang < 17
> > 80d47defddc0 x86/stackprotector/64: Convert to normal per-CPU variable
> > 9d7de2aa8b41 x86/percpu/64: Use relative percpu offsets
> >
> > As of a9a76b38aaf5, everything compiles correctly.
> >
> > For f58b63857ae3 thru 80d47defddc0, this error occurs:
> >
> >     arch/x86/platform/pvh/head.S: Assembler messages:
> >     arch/x86/platform/pvh/head.S:183: Error: operand size mismatch for `movq'
> >
> 
> This is simply a bug in patch f58b63857ae3, which uses movq on
> EDX/EAX. This breaks bisection (as you've noticed), annoyingly, and
> I'll leave it to the tip maintainers to decide whether this warrants a
> rebase (given that the fix is already queued up as well)
> 
> > As of 9d7de2aa8b41 and later, the head.S error goes away, and this
> > error occurs:
> >
> >     lib/zstd/compress/huf_compress.c:1033:1: internal compiler error: Segmentation fault
> >      1033 | {
> >           | ^
> >     Please submit a full bug report,
> >     with preprocessed source if appropriate.
> >     See <file:///usr/share/doc/gcc-9/README.Bugs> for instructions.
> >
> > Interestingly, if DYNAMIC_BMI2 in lib/zstd/common/portability_macros.h
> > is changed on line 61 to always be defined as 0, the compile problem in
> > huf_compress.c goes away.
> >
> 
> Without having looked at the crashing code, I suspect that the only
> meaningful course of action here is to tweak the conditional logic
> that defines DYNAMIC_BMI2 to only enable it for GCC >= 10.1 (assuming
> those are unaffected) In any case, GCC 9.4 is not going to get fixed
> in the field, so even if this gets fixed on the GCC side, we will have
> to deal with it here.

Nick and David --

What are your thoughts as maintainers of lib/zstd?

Michael

> 
> > I'm using gcc 9.4.0 in an Ubuntu 20.04 install as it's the standard compiler
> > for that version of Ubuntu. I'm guessing that getting a fix for that
> > relatively old version of gcc isn't likely. I have not tested with other
> > gcc versions.