[U-Boot] CONFIG_SYS_FDT_PAD: Why do we need this?

[U-Boot] CONFIG_SYS_FDT_PAD: Why do we need this?

[Dirk Behme]

In boot_relocate_fdt() [1] we increase the size of the loaded fdt by a 
predefined, hard coded CONFIG_SYS_FDT_PAD. The basic idea seem to be 
to create some space for adding additional information later, like the 
bootargs.

I wonder why we have to do this by a hard coded, predefined value 
(which is most probably always to large)? Why can't we increase 
'totalsize' in the fdt_header just if we add anything? Individually by 
the size of the element added, at runtime? E.g. by anything like [2]?

I debugged a system were

#define CONFIG_SYS_FDT_PAD 0x3000

was used. These 0x3000 are added to totalsize before something is 
added at all. Then, if an element is added, always the whole 0x3000 
bytes have to be memmoved additionally. And in the end only ~0x100 
additional space was really needed.

Best regards

Dirk

[1] 
http://git.denx.de/cgi-bin/gitweb.cgi?p=u-boot.git;a=blob;f=common/image.c;h=8c644b7dacdc9b7a196a5238f4dc5d94bd615d07;hb=HEAD#l1285

[2]

static int _fdt_splice_struct(void *fdt, void *p,
			      int oldlen, int newlen)
{
	int delta = newlen - oldlen;
	int err;

	if ((err = _fdt_splice(fdt, p, oldlen, newlen)))
		return err;

	fdt_set_size_dt_struct(fdt, fdt_size_dt_struct(fdt) + delta);
	fdt_set_off_dt_strings(fdt, fdt_off_dt_strings(fdt) + delta);
+	fdt_set_totalsize(fdt, fdt_totalsize(fdt) + delta);
	return 0;
}

in

http://git.denx.de/cgi-bin/gitweb.cgi?p=u-boot.git;a=blob;f=lib/libfdt/fdt_rw.c;h=5c27a677e3536ed7320ee399113666975e5d545b;hb=HEAD#l127

[U-Boot] CONFIG_SYS_FDT_PAD: Why do we need this?

[Simon Glass]

Hi Dirk,

On Sat, Mar 24, 2012 at 11:59 PM, Dirk Behme <dirk.behme@googlemail.com> wrote:
>
> In boot_relocate_fdt() [1] we increase the size of the loaded fdt by a
> predefined, hard coded CONFIG_SYS_FDT_PAD. The basic idea seem to be to
> create some space for adding additional information later, like the
> bootargs.
>
> I wonder why we have to do this by a hard coded, predefined value (which is
> most probably always to large)? Why can't we increase 'totalsize' in the
> fdt_header just if we add anything? Individually by the size of the element
> added, at runtime? E.g. by anything like [2]?
>
> I debugged a system were
>
> #define CONFIG_SYS_FDT_PAD 0x3000
>
> was used. These 0x3000 are added to totalsize before something is added at
> all. Then, if an element is added, always the whole 0x3000 bytes have to be
> memmoved additionally. And in the end only ~0x100 additional space was
> really needed.

I don't really know the answer, but I guess that just increasing the
FDT is dangerous, since we don't know how big it can grow before
running into something else. This way we set a maximum size that it
can grow to and will produce an error if it is exceeded.

There is also the practical problem that we would need to check for
FDT_ERR_NOSPACE, expand the space, and then retry. A bit of a
hassle...

I suppose you could use fdt_pack() afterwards, although I don't see a
command for this at present.

Regards,
Simon

>
> Best regards
>
> Dirk
>
> [1]
> http://git.denx.de/cgi-bin/gitweb.cgi?p=u-boot.git;a=blob;f=common/image.c;h=8c644b7dacdc9b7a196a5238f4dc5d94bd615d07;hb=HEAD#l1285
>
> [2]
>
> static int _fdt_splice_struct(void *fdt, void *p,
> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?int oldlen, int newlen)
> {
> ? ? ? ?int delta = newlen - oldlen;
> ? ? ? ?int err;
>
> ? ? ? ?if ((err = _fdt_splice(fdt, p, oldlen, newlen)))
> ? ? ? ? ? ? ? ?return err;
>
> ? ? ? ?fdt_set_size_dt_struct(fdt, fdt_size_dt_struct(fdt) + delta);
> ? ? ? ?fdt_set_off_dt_strings(fdt, fdt_off_dt_strings(fdt) + delta);
> + ? ? ? fdt_set_totalsize(fdt, fdt_totalsize(fdt) + delta);
> ? ? ? ?return 0;
> }
>
> in
>
> http://git.denx.de/cgi-bin/gitweb.cgi?p=u-boot.git;a=blob;f=lib/libfdt/fdt_rw.c;h=5c27a677e3536ed7320ee399113666975e5d545b;hb=HEAD#l127
> _______________________________________________
> U-Boot mailing list
> U-Boot at lists.denx.de
> http://lists.denx.de/mailman/listinfo/u-boot