[U-Boot] [PATCH v4 0/4] Factorize ARM relocate_code instances

[Albert ARIBAUD]

Hi Simon,

On Mon, 27 May 2013 07:56:48 -0700, Simon Glass <sjg at chromium.org>
wrote:

> Hi Albert,
> 
> On Mon, May 20, 2013 at 2:26 AM, Albert ARIBAUD
> <albert.u.boot at aribaud.net>wrote:
> 
> > Hi Benoît,
> >
> > On Sun, 19 May 2013 17:57:59 +0200 (CEST), Benoît Thébaudeau
> > <benoit.thebaudeau at advansee.com> wrote:
> >
> > > Hi Albert,
> > >
> > > On Sunday, May 19, 2013 1:48:11 PM, Albert ARIBAUD wrote:
> > > > This series replaces all instances of relocate_code in
> > > > various start.S files from the ARM architecture with a
> > > > single instance in arch/arm/lib/relocate.S.
> > > >
> > > > This is done in steps, each step keeping the whole of ARM
> > > > U-Boot buildable and runnable and touching as little code
> > > > as necessary.
> > > >
> > > > 1. Drop mx31pdk SPL call to relocate_code
> > > >
> > > > In SPL of mx31pdk, board_init_f calls relocate_code to
> > > > move SPL from the NAND read buffer to SRAM. This move
> > > > is not a relocation and is thus better implemented as
> > > > a simple ad hoc copy loop.
> > > >
> > > > 2. Drop tx25 SPL call to relocate_code
> > > >
> > > > Same as mx31pdk.
> > > >
> > > > 3. Prevent all SPLs from assembling relocate_code
> > > >
> > > > Now that no SPL uses relocate_code any more, remove
> > > > it completely from SPL builds by putting it entirely
> > > > between #ifndef CONFIG_SPL_BUILD and #endif preprocessor
> > > > conditional.
> > > >
> > > > 4. Actual factorization of relocate_code
> > > >
> > > > Remove all instances of relocate_code from start.S and
> > > > create a new file, arch/arm/lib/relocate.S, to provide
> > > > the new single instance of relocate_code.
> > > >
> > > > The only non-trivial change is that for PXA25X, a call
> > > > to unlock (actually disable) dcache is moved from before
> > > > relocating to after relocating. As this cache is either
> > > > locked-as-RAM or disabled, but never used as actual DDR
> > > > cache, this move has no performance effect on relocation
> > > > and no functional impact either.
> > > >
> > > > The relocation routine was step-tested on versatileqemu
> > > > under gdb. Especially, both relocation offset (from
> > > > _start to addr_moni and local offsets (from relocate_code
> > > > to several linker script symbols) have been verified.
> > > >
> > > > Changes in v4:
> > > > - added blank line after PXA25X conditional code
> > > > - fixed typos in relocate_code comments
> > > > - brought back 'mov pc,lr' instead of 'bx lr' for ARMv4
> > >
> > > [...]
> > >
> > > Perfect!
> > >
> > > For this v4 series:
> > > Reviewed-by: Benoît Thébaudeau <benoit.thebaudeau at advansee.com>
> >
> > Thanks Benoît!
> >
> > Now I need some people to regression-test it on their
> > HW, both ARM and non-ARM just in case.
> >
> 
> Tested on snow (ARMv7) and link (x86), seems fine.
> 
> The series:
> 
> Tested-by: Simon Glass <sjg at chromium.org>

thanks a lot!

> Re your second series containing 'arm: optimize relocate_code
> routine<http://patchwork.ozlabs.org/patch/243817/>',
> I ended up with conflicts here - perhaps because the first series was
> updated. I would like to test that also, as it should make it possible to
> remove CONFIG_SYS_TEXT_BASE which is currently of interest.

Thanks Simon. I have to rebase and repost the second series indeed;
will do this tonight. Note that it was first posted after merge window
closure, so if you need it for things planned in 2013.07, then I'll
have to ask for Tom's agreement to put the second series in 2013.07.

> Regards,
> Simon

Amicalement,
-- 
Albert.