Aarch64 EXT4FS inode checksum failures - seems to be weak memory ordering issues

[Nick Desaulniers]

On Fri, Jan 8, 2021 at 12:34 PM Arnd Bergmann <arnd at kernel.org> wrote:
>
> On Fri, Jan 8, 2021 at 9:02 PM Linus Torvalds
> <torvalds at linux-foundation.org> wrote:
> > On Fri, Jan 8, 2021 at 1:27 AM Will Deacon <will at kernel.org> wrote:
> > >
> > > On Fri, Jan 08, 2021 at 10:21:54AM +0100, Peter Zijlstra wrote:
> > > > On Thu, Jan 07, 2021 at 10:20:38PM +0100, Arnd Bergmann wrote:
> > > > > On Thu, Jan 7, 2021 at 2:37 PM Russell King - ARM Linux admin
> >
> > I appreciate Arnd pointing out "--std=gnu11", though. What are the
> > actual relevant language improvements?

It's hard to say, since a lot of new language features were already
GNU C extensions.

The only semantic difference I'm aware of is the semantics of `extern
inline` changed 100% from c89 to c99 (so jumping from gnu89 to gnu11
would change that).  We already #define inline to
__attribute__((__gnu_inline)) (there's also a -fgnu-inline flag), but
I worry for places that don't include that header or drop
KBUILD_CFLAGS (like every vdso), though `extern inline` is awful (and
I should be put in jail for introducing it to the kernel; now we have
__attribute__((no_stack_protector)) in both toolchains, and should be
using that instead, but we don't have it yet for all supported
compiler versions).

A quick grep through clang's sources shows mostly parser changes for
_Noreturn, _Alignof and friends etc..  New to me are unicode literal
strings (u or U suffix or prefix?) and something about loops expected
to make forward progress???

Another thing I've been worried about is Makefiles that reset
KBUILD_CFLAGS, since that's a constant source of pain/breakage for
cross compiling from Clang.  That tends to drop -std=gnu89.  For
instance:

$ make LLVM=1 -j71 defconfig
$ make LLVM=1 -j71 V=1 &>log.txt
$ grep -v std=gnu89 log.txt | grep clang | rev | cut -d ' ' -f 1 | rev
| grep -v \\.S
arch/x86/realmode/rm/wakemain.c
arch/x86/realmode/rm/video-mode.c
arch/x86/realmode/rm/regs.c
arch/x86/realmode/rm/video-vga.c
arch/x86/realmode/rm/video-vesa.c
arch/x86/realmode/rm/video-bios.c
drivers/firmware/efi/libstub/efi-stub-helper.c
drivers/firmware/efi/libstub/gop.c
drivers/firmware/efi/libstub/secureboot.c
drivers/firmware/efi/libstub/tpm.c
drivers/firmware/efi/libstub/file.c
drivers/firmware/efi/libstub/mem.c
drivers/firmware/efi/libstub/random.c
drivers/firmware/efi/libstub/randomalloc.c
drivers/firmware/efi/libstub/pci.c
drivers/firmware/efi/libstub/skip_spaces.c
lib/cmdline.c
lib/ctype.c
drivers/firmware/efi/libstub/alignedmem.c
drivers/firmware/efi/libstub/relocate.c
drivers/firmware/efi/libstub/vsprintf.c
drivers/firmware/efi/libstub/x86-stub.c
arch/x86/boot/a20.c
arch/x86/boot/cmdline.c
arch/x86/boot/cpuflags.c
arch/x86/boot/cpucheck.c
arch/x86/boot/early_serial_console.c
arch/x86/boot/edd.c
arch/x86/boot/main.c
arch/x86/boot/memory.c
arch/x86/boot/pm.c
arch/x86/boot/printf.c
arch/x86/boot/regs.c
arch/x86/boot/string.c
arch/x86/boot/tty.c
arch/x86/boot/video.c
arch/x86/boot/video-mode.c
arch/x86/boot/version.c
arch/x86/boot/video-vga.c
arch/x86/boot/video-vesa.c
arch/x86/boot/video-bios.c
arch/x86/boot/cpu.c
arch/x86/boot/compressed/string.c
arch/x86/boot/compressed/cmdline.c
arch/x86/boot/compressed/error.c
arch/x86/boot/compressed/cpuflags.c
arch/x86/boot/compressed/early_serial_console.c
arch/x86/boot/compressed/kaslr.c
arch/x86/boot/compressed/ident_map_64.c
arch/x86/boot/compressed/idt_64.c
arch/x86/boot/compressed/pgtable_64.c
arch/x86/boot/compressed/acpi.c
arch/x86/boot/compressed/misc.c

So it looks like parts of the tree are already built with -std=gnu11
or -std=gnu17, as they rely on the implicit default C language mode
when unspecified.  Oops?
-- 
Thanks,
~Nick Desaulniers