Re: [PATCH 1/2] Enable a way to provide the reason for "being here"

["Uwe Kleine-König" <u.kleine-koenig@pengutronix.de>]

On Wed, Jun 20, 2012 at 09:59:14PM +0200, Juergen Beisert wrote:
> Uwe Kleine-König wrote:
> > On Wed, Jun 20, 2012 at 09:09:53PM +0200, Juergen Beisert wrote:
> > > Hi Uwe,
> > >
> > > Uwe Kleine-König wrote:
> > > > On Wed, Jun 20, 2012 at 05:08:52PM +0200, Juergen Beisert wrote:
> > > > > Marc Kleine-Budde wrote:
> > > > > > [...]
> > > > > >
> > > > > > > +
> > > > > > > +#include <common.h>
> > > > > > > +#include <init.h>
> > > > > > > +#include <environment.h>
> > > > > > > +#include <globalvar.h>
> > > > > > > +#include <reset_source.h>
> > > > > > > +
> > > > > > > +static const char name[] = "global.system.reset";
> > > > > > > +static const char unknown_reset[] = "unknown";
> > > > > > > +static const char power_on_reset[] = "POR";
> > > > > > > +static const char manual_reset[] = "RST";
> > > > > > > +static const char watchdog[] = "WDG";
> > > > > > > +static const char wake[] = "WKE";
> > > > > > > +static const char jtag[] = "JTAG";
> > > > > >
> > > > > > what about using an array
> > > > > >
> > > > > > static cost char *reset_reason_array[] = {
> > > > > > 	[RESET_UKWN] = "unknown",
> > > > > > 	...
> > > > > > };
> > > > >
> > > > > The result is not the same. The strings are no longer "const". But it
> > > > > should
> > > >
> > > > <kidding>Of course they are not "const" if you write "cost".</kidding>
> > > > Using
> > > >
> > > > 	static const char * const reset_reason_array[] = {
> > > >
> > > > should do the trick.
> > >
> > > No, it doesn't. Only this would:
> > >
> > > static const bla[] = "this is a really constant string";
> > > static const char * const reset_reason_array[] = {
> > > 	[0] = bla,
> > > [...]
> > > };
> >
> > What makes you think it's not const. I think it is, look:
> >
> > 	ukleinek@perseus:~/tmp$ cat test.c
> > 	static const char * const const_reset_reason_array[] = {
> > 		"jtag",
> > 		"por",
> > 	};
> >
> > 	static const char *reset_reason_array[] = {
> > 		"jtag",
> > 		"por",
> > 	};
> >
> > 	void somefunc(void)
> > 	{
> > 		char *reason = const_reset_reason_array[0];
> > 	}
> > 	ukleinek@perseus:~/tmp$ gcc -Wall -c test.c
> > 	test.c: In function ‘somefunc’:
> > 	test.c:13:17: warning: initialization discards ‘const’ qualifier from
> > pointer target type [enabled by default] test.c:13:8: warning: unused
> > variable ‘reason’ [-Wunused-variable] test.c: At top level:
> > 	test.c:6:20: warning: ‘reset_reason_array’ defined but not used
> > [-Wunused-variable]
> >
> > 	ukleinek@perseus:~/tmp$ objdump -D -j .rodata -j .data test.o
> >
> > 	test.o:     file format elf64-x86-64
> >
> >
> > 	Disassembly of section .data:
> >
> > 	0000000000000000 <reset_reason_array>:
> > 		...
> >
> > 	Disassembly of section .rodata:
> >
> > 	0000000000000000 <const_reset_reason_array-0x10>:
> > 	   0:	6a 74                	pushq  $0x74
> > 	   2:	61                   	(bad)
> > 	   3:	67 00 70 6f          	add    %dh,0x6f(%eax)
> > 	   7:	72 00                	jb     9 <const_reset_reason_array-0x7>
> > 	   9:	00 00                	add    %al,(%rax)
> > 	   b:	00 00                	add    %al,(%rax)
> > 	   d:	00 00                	add    %al,(%rax)
> > 		...
> >
> > 	0000000000000010 <const_reset_reason_array>:
> > 		...
> >
> > so const_reset_reason_array lives in .rodata and assinging
> > const_reset_reason_array[0] to a char * results in a warning.
> >
> > What am I missing?
> 
> const char friesel[] = "this is really a constant string";
> const char *frasel = "this is not really a constant string";
> 
> The compiler will place the "this is really a constant string" into 
> the .rodata, but "this is not really a constant string" will be still 
> in .data (writeable). This difference is very important to know, if you want 
> the compiler to place the strings into the flash of a microcontroller instead 
> of the RAM (which is most of the time very small).
Right, and if you say:

	const char *const frosel = "this is really a constant string, too";

it will end in .rodata, too. And that is the construct I suggested to
use.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox