Attached is a fix for this part of the sysrq code.

++ 21/09/01 17:08 -0400 - Crutcher Dunnavant:
> ++ 19/09/01 08:56 -0700 - Randy.Dunlap:
> > It always sets console_loglevel and then restores
> > console_loglevel from orig_log_level, so Alt+SysRq+#
> > handling is severely broken.
> > 
> > If someone (Crutcher ?) wants to patch it, that's fine.
> > If I patched it, I would just add a
> >   next_loglevel = -1;
> > at the beginning of __handle_sysrq_nolock() and then
> > let the loglevel handler(s) set next_loglevel.
> > If next_loglevel != -1 at the end of __handle_sysrq_nolock(),
> > set console_loglevel to next_loglevel.
> 
> I'm looking real close at this right now, and there are a couple of
> problems, and a simple, but ugly solution.
> 
> The entire reason that console_loglevel is touched _after_ the call to
> the second level handler is actually for the loglevel handler's
> printout. I was trying to minimize change in the display, but horked it.
> 
> Here is the problem.
> 
> SysRq events use action messages which get printed by the top level
> handler before calling the second level handler, the call line is:
> 
>         orig_log_level = console_loglevel;
>         console_loglevel = 7;
>         printk(KERN_INFO "SysRq : ");
> 
>         op_p = __sysrq_get_key_op(key);
> 	...
>         printk ("%s", op_p->action_msg);
>         op_p->handler(key, pt_regs, kbd, tty);
> 	...
>         console_loglevel = orig_log_level;
> 
> 
> The killer here is the fact that the action message format string does
> not carry a newline, allowing people to register strings which leave the
> printk state open. The loglevel handler then fills in the loglevel, and
> closes the printk state.
> 
> There was a time when I thought that was a good idea.
> 
> Go ahead, laugh.
> 
> Anyway, that sort of unresolved state is bad, and is the source of all
> of this song and dance. I think the right answer is to force handlers to
> open their own calls to printk, and to keep whats going on with the
> console_loglevel and printk buffer nice and clean.
> 
> The cost is that messages like this:
> 
> SysRq : Loglevel switched to X
> 
> will have to become more like this:
> 
> SysRq : Loglevel
> Loglevel switched to X
> 
> 
> Again, appologies, and a patch is forthcoming.

And here is the patch.

-- 
Crutcher        <crutcher@datastacks.com>
GCS d--- s+:>+:- a-- C++++$ UL++++$ L+++$>++++ !E PS+++ PE Y+ PGP+>++++
    R-(+++) !tv(+++) b+(++++) G+ e>++++ h+>++ r* y+>*$