Re: mmotm 2009-01-28-02-17 uploaded (i386 build errors: cpusets &kexec)

[Neil Horman]

On Thu, Jan 29, 2009 at 11:48:59AM -0800, Andrew Morton wrote:
> On Thu, 29 Jan 2009 14:27:34 -0500
> Neil Horman <nhorman@xxxxxxxxxxxxx> wrote:
> 
> > On Thu, Jan 29, 2009 at 10:01:57AM -0800, Andrew Morton wrote:
> > > On Thu, 29 Jan 2009 07:09:35 -0500 Neil Horman <nhorman@xxxxxxxxxxxxx> wrote:
> > > 
> > > > > > So does that.
> > > > > > 
> > > > > > --- a/kernel/printk.c~kexec-add-dmesg-log-symbols-to-proc-vmcoreinfo-lists-fix-fix
> > > > > > +++ a/kernel/printk.c
> > > > > > @@ -137,11 +137,13 @@ static char *log_buf = __log_buf;
> > > > > >  static int log_buf_len = __LOG_BUF_LEN;
> > > > > >  static unsigned logged_chars; /* Number of chars produced since last read+clear operation */
> > > > > >  
> > > > > > +#ifdef CONFIG_KEXEC
> > > > > >  void log_buf_kexec_setup(void)
> > > > > >  {
> > > > > >  	VMCOREINFO_SYMBOL(log_buf);
> > > > > >  	VMCOREINFO_SYMBOL(log_end);
> > > > > >  }
> > > > > > +#endif
> > > > > >  
> > > > > >  static int __init log_buf_len_setup(char *str)
> > > > > >  {
> > > > > > _
> > > > > 
> > > > > Weird, I thought I checked that just now. In any case this email
> > > > > was already in my inbox. In any case, I can now produce the problem,
> > > > > and this patch resolves it.
> > > > > 
> > > > 
> > > > Sorry, i didn't think to check on conditions when CONFIG_KEXEC wasn't set.  I
> > > > assumed we had a stub for all the VMCOREINFO defines.  Why don't we fix it in
> > > > that manner?  I mean we have serveral arch specific additions to vmcoreinfo, all
> > > > of which apparently must be ifdefd for this to work.  Why not just fix it in
> > > > conditional definition?
> > > 
> > > I didn't do that because then we'd have an empty function log_buf_kexec_setup()
> > > floating about in every vmlinux.
> > > 
> > > Hardly a big issue, but a bit regrettable.
> > > 
> > Won't the compiler just eliminate that symbol during final link, given that its
> > not an exported symbol?
> 
> I don't think it's that smart.
> 
> akpm:/home/akpm> cat t.c
> 
> void foo(void)
> {
> }
> 	
> main()
> {
> }
> akpm:/home/akpm> gcc -O2 t.c
> akpm:/home/akpm> nm a.out|grep foo
> 08048300 T foo
> 
> With -ffunction-sections or other fancy options, maybe.  And it'll be
> version-dependent.
> 
huh, how about that.  I always thought we stripped out empty functions.  Perhaps
that a later stage in most of the build processes I've used (a strip or
something later on perhaps).  Anywho, I stand corrected.  In light of this, its
probably best that we keep what we have, and just ifdef out the callers.

Thanks Andrew!
Neil

> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/