On Wed, Mar 04, 2015 at 08:42:58AM +0100, Heiko Schocher wrote:
Hello Tom,
Am 02.03.2015 14:59, schrieb Tom Rini:
On Mon, Mar 02, 2015 at 07:56:41AM +0100, Heiko Schocher wrote:
Hello Simon,
Am 24.02.2015 14:31, schrieb Simon Glass:
Hi Heiko,
On 23 February 2015 at 23:18, Heiko Schocher hs@denx.de wrote:
a6b541b090: TI ARMv7: Don't use GD before crt0.S has set it
moves the init of the debug uart at the very end of SPL code. Enable it for the siemens board earlier, as they print ddr settings ... all debug output before board_init_r() is here currently useless. Maybe we must rework this globally?
Assuming we are talking about U-Boot proper, the DDR init should happen in board_init_f(), specifically dram_init(). so I think this code should be updated.
If it is SPL, then DDR init should happen in SPL's board_init_f().
It is in SPL...
sdram_init() is called from:
./arch/arm/cpu/armv7/am33xx/board.c from s_init() ...
I sent a series a few weeks ago (available at u-boot-dm branch spl-working) related to this topic:
Ah ... Hmm... so "./arch/arm/cpu/armv7/am33xx/board.c" needs a rework, right?
Is a simple rename s_init() -> board_init_f() correct?
Right so, no, we can't just rename s_init to board_init_f. This is what I was talking about in the thread about the function Hans wants to add to enable some bits in CP15 on sunxi, iirc.
In short, armv7 has a different set of abstraction hooks than the previous ARM cores (armv8 followed what we have for v7) and I'm not convinced in the end that it really won us anything. See http://lists.denx.de/pipermail/u-boot/2015-January/202350.html
For today you need to rework the Siemens code to print out the DDR values (when desired) in spl_board_init() as we do not, or will not shortly, have gd prior to board_init_f running.
Hmm... first I thought, ok, no problem, move the output from the RAM parameters to spl_board_init() ... but thats only the half of the story ... They read the RAM parameters from an i2c eeprom, and if there are errors, they print this errors ... currently this does not work, and thats I think the more important case ... and I could not move this error printfs to somewhere, because if RAM is not working ... there is no later ...
So I have to enable the console early ... maybe I missed something, but this worked fine in the past (and I think we should not break this, as this is an essential feature).
OK, I missed something too. I think this gets better now once I merge Simon's SPL series as we do all of this from board_init_f() and the siemens code should just work again.
If I look into arch/arm/cpu/armv7/am33xx/board.c s_init() there gets the console also in the case:
#if defined(CONFIG_NOR_BOOT) || defined(CONFIG_QSPI_BOOT)
very early enabled ... is this buggy?
Digging out one of the QSPI bootable boards (which I only _think_ I have) or setting up and confirming the beaglebone white + NOR cape I have still functions at the HW level and then re-testing on mainline are on my list as yes, I'm a bit worried these paths might be broken again.