Re: Linux 5.16-rc1

[Guenter Roeck]

On 11/15/21 10:10 AM, Linus Torvalds wrote:
> On Mon, Nov 15, 2021 at 9:07 AM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> > parisc:allmodconfig: Lots of build failures in arch/parisc/include/asm/jump_label.h.
> > Not fixed in -next. The problem seens to be related to the thread_info changes,
> > or at least bisect points to commit 01463374c50e ("Merge tag 'cpu-to-thread_info-v5.16-rc1'
> > of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux").
>
> I don't think they are related to the thread_info changes except very
> incidentally.
>
> > From the errors ("expected ':' before '__stringify'") it looks like
> __stringify() isn't defined there, and I do think that header file
> should include <linux/compiler.h> to get it.
>
> I don't actually see how that merge commit you bisected to could
> possibly matter (the diff doesn't show any header file include
> changes, or any parisc ones), but I suspect you might have bisected to
> it because we did end up having actual thread_info errors on parsic
> that weren't fixed until commit 2a2e8202c7a1 ("parisc: move CPU field
> back into thread_info").
>
> So I suspect the parisc build problems started with thread_info issues
> at that merge commit you pinpoint, and then by commit 2a2e8202c7a1
> some other header file simplification had exposed that "no
> __stringify()" thing).
>
> Helge - apparently from allmodconfig and others, we have:
>
>     arch/parisc/include/asm/jump_label.h: In function 'arch_static_branch':
>     arch/parisc/include/asm/jump_label.h:18:18: error: expected ':'
> before '__stringify'
>
> and others, which does look like just a missing header file (that was
> presumably previously implicitly included earlier, and now the
> implicit include is gone).

Geert pinpointed this problem to

    due to static_branch_likely() in crypto/api.c

Commit adad556efcdd4 ("crypto: api - Fix built-in testing dependency failures")
added "#include <linux/jump_label.h>". That extra include isn't needed because
crypto/internal.h already includes it. Removing it from crypto/api.c
fixes the problem for parisc, but I don't know if that is the correct fix.
Alternatively, linux/completion.h would have to be included from crypto/api.c
before including linux/jump_label.h.

Guenter