On Wed, Apr 08, 2015 at 06:06:08PM +0200, Pavel Machek wrote:
On Wed 2015-04-08 11:53:38, Tom Rini wrote:
On Wed, Apr 08, 2015 at 04:06:34PM +0200, Pavel Machek wrote:
On Wed 2015-04-08 09:49:53, Tom Rini wrote:
On Wed, Apr 08, 2015 at 02:09:57PM +0200, Pavel Machek wrote:
Hi!
>>maybe this helps? I just stepped into same problem on an am335x, >>but had not yet time to look deeper in it ... > >I did not try the cleanup_before_linux(), but "dcache off" should have >same effect, right?
Hmm.. not exactly... cleanup_before_linux() does also disable the mmu.
Seems like I found the reason. CONFIG_THUMB_BUILD was set, and that causes problems in put_dec_trunc() function (and elsewhere).
commit 1602a502985d42bafb58cc6862b469ac03a7d634 Author: Pavel pavel@ucw.cz Date: Wed Apr 8 14:08:18 2015 +0200
Disable thumb build; it breaks put_dec_trunc() function on toolchain from eldk-5.6 and eldk-5.4, and some other place, too. Signed-off-by: Pavel Machek <pavel@denx.de>diff --git a/include/configs/socfpga_common.h b/include/configs/socfpga_common.h index 6d93472..6608d7d 100644 --- a/include/configs/socfpga_common.h +++ b/include/configs/socfpga_common.h @@ -11,7 +11,7 @@ /* Virtual target or real hardware */ #undef CONFIG_SOCFPGA_VIRTUAL_TARGET
-#define CONFIG_SYS_THUMB_BUILD +#undef CONFIG_SYS_THUMB_BUILD
#define CONFIG_SOCFPGA
OK, this is kind of scary. What is wrong with those toolchains OR how we're telling them to build things as no, we should be able to use thumb mode just fine and need to for size reasons in various cases.
Well.. it is scary, as put_dec_trunc() is very straightforward function. It works in 2014.10, even with thumb on.
If this is a generic problem then for example, omap4_panda + ELDK 5.4/5.6 just shouldn't work either and we _must_ use thumb there for at least SPL.
I don't know enough about thumb in u-boot. And yes, I'd like people to confirm if u-boot works for them on socfpga...
I'll fire up ELDK 5.4 + Pandaboard later today (which uses thumb) but I think you need to bisect down to when exactly things break since my gut is telling me it's not toolchain / thumb but something else that broke things.
Thanks.
Well, it broke between 2014.10 and 2015.01, so I think it should be toolchain (because broken socfpga for >4 months... someone would notice, no?).
Marek told me I should mention... I'm using USE_PRIVATE_LIBGCC because compilation does not work without that.
I think the answer is that it's bisect time.