Re: [PATCH] um: Fix filtering '-mno-global-merge'

[Masahiro Yamada]

On Sun, Mar 27, 2022 at 5:00 AM Nathan Chancellor <nathan@xxxxxxxxxx> wrote:
>
> On Sat, Mar 26, 2022 at 12:29:55PM +0900, Masahiro Yamada wrote:
> > On Wed, Mar 23, 2022 at 2:39 AM Nathan Chancellor <nathan@xxxxxxxxxx> wrote:
> > >
> > > When booting a clang compiled UML kernel, the kernel panics when trying
> > > to run init:
> > >
> > >   wait_stub_done : failed to wait for SIGTRAP, pid = 651294, n = 651294, errno = 0, status = 0xb7f
> > >   Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
> > >
> > > After the commit in Fixes, many flags from KBUILD_CFLAGS do not appear
> > > in USER_CFLAGS, likely due to USER_CFLAGS initially being a recursive
> > > variable ("VAR =") then being switched to a simple ("VAR :=") variable.
> > > For example, diffing arch/x86/um/.ptrace_user.o.cmd shows flags such as
> > > '-Os' and '-fno-delete-null-pointer-checks' getting dropped, which both
> > > impact code generation.
> > >
> > > Rework the filtering to use filter-out instead of patsubst, which allows
> > > all the patterns that USER_CFLAGS cares about to be excluded in one
> > > command and ensures all flags from KBUILD_CFLAGS are transferred over to
> > > USER_CFLAGS properly, which resolves the boot issue noted above.
> > >
> > > Fixes: 6580c5c18fb3 ("um: clang: Strip out -mno-global-merge from USER_CFLAGS")
> > > Signed-off-by: Nathan Chancellor <nathan@xxxxxxxxxx>
> >
> >
> >
> > Can we remove -mno-global-merge entirely?
> >
> >
> > 61163efae02040f66a95 was a very old commit,
> > without enough explanation.
> >
> > Shall we remove  -mno-global-merge, and do compile-tests.
> > If we are hit by problems for arm/arm64, we can re-add it.
>
> Yes, I think that was the conversation that we had on the commit that
> this fixes:
>
> https://lore.kernel.org/r/YiD7R2wRxoWxtVq7@dev-arch.thelio-3990X/
>
> I can test that on my Raspberry Pi 3 and 4 on Monday, although I would
> like for this patch to be picked up in the meantime so that it is
> possible to test UML on -next with clang. We can remove
> -mno-global-merge in a follow up change, if you do not have any
> objections?
>
> Cheers,
> Nathano


This patch is unneeded now
because I picked up the alternative:

https://lore.kernel.org/all/CAK7LNAS6C6Uj9cCQ0o=bYF1F-EVD=VgdR8YYx-1PJc9toX_HZA@xxxxxxxxxxxxxx/



-- 
Best Regards
Masahiro Yamada