Date: Sat, 28 Jan 2012 19:23:34 +1100 (EST) From: Bruce Evans <brde@optusnet.com.au> To: Marius Strobl <marius@freebsd.org> Cc: svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org Subject: Re: svn commit: r230628 - head/sys/sparc64/include Message-ID: <20120128184815.L1114@besplex.bde.org> In-Reply-To: <201201272204.q0RM4hRH062478@svn.freebsd.org> References: <201201272204.q0RM4hRH062478@svn.freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, 27 Jan 2012, Marius Strobl wrote:
> Log:
> Mark cpu_{halt,reset}() as __dead2 as appropriate.
>
> Modified:
> head/sys/sparc64/include/cpu.h
>
> Modified: head/sys/sparc64/include/cpu.h
> ==============================================================================
> --- head/sys/sparc64/include/cpu.h Fri Jan 27 21:52:59 2012 (r230627)
> +++ head/sys/sparc64/include/cpu.h Fri Jan 27 22:04:43 2012 (r230628)
> @@ -53,8 +53,8 @@ extern char btext[];
> extern char etext[];
>
> void cheetah_init(u_int cpu_impl);
> -void cpu_halt(void);
> -void cpu_reset(void);
> +void cpu_halt(void) __dead2;
> +void cpu_reset(void) __dead2;
> void fork_trampoline(void);
> void swi_vm(void *v);
> void zeus_init(u_int cpu_impl);
This reminds me that these functions and many others shouldn't be
in <machine/cpu.h> since they necessarily have a MI interface (so
that MI code can call them). A few interfaces in <machine/cpu.h>
need to be there so that the can be inlines or macros, but most
don't. Especially these 2. Since they are extremely non-time-
critical and not likely to be magical, they can be implemented as
a wrapper extern function even if the MD code prefers to use an
in inline or macro. Duplicating these 2 in ${N_ARCH} cpu.h files
mainly allows some of the files to forget to declare them as
__dead2.
I put these in <sys/systm.h> and didn't forget __dead2 for them.
systm.h already has prototypes for 11 functions named cpu_*. These
are unsorted of course. It also has prototypes for a few functions
that are misnamed (without a cpu_ as a prefix, or perhaps without
cpu_ or cpu at all) so that they are naturally unsorted and hard to
find.
No MD declaration of these 2 in -current except the above has the
__dead2's. IIRC, I noticed that __dead2 was missing for them from
the style bug that code after calls to them has /* NOTREACHED */
comments (shutdown_halt() still does, except in my version).
__dead2 should have made most of these comments go away 15-20 years
ago.
A more important "cpu_" function whose comments annoy me is
cpu_throw(). This has always been declared in an MI file
(<sys/proc.h>, which shows that we can't expect all these functions
to be declared in <sys/systm.h> if they are MI). Calls to
cpu_throw() have the unnecessary comments /* doesn't return */ in
2 places (lint won't understand these, so if you are going to
comment on the obvious then it should be /* NOTREACHED */), and a
bogus "teapot" panic and a /* NOTREACHED */ comment in another
place. The panic is so bogus that gcc removes it (as allowed by
cpu_throw() being declared __dead2). The first 2 places are
sched_throw() in 2 schedulers. This can't return, but isn't
declared as __dead2. The other place is thread_exit(). This is
one of very few extern functions that is declared as __dead2.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20120128184815.L1114>