Hello Hannes, Tom,
On di, 2014-09-16 at 09:56 -0400, Tom Rini wrote:
On Tue, Sep 16, 2014 at 02:17:01PM +0200, Hannes Petermaier wrote:
Hi,
i've just pulled newest u-boot and tried it on my custom AM335x board without success. It hangs after (successful) running init_s function.
.... Enable clock module - 44e0001c Enable clock module - 44e000d0 Enable clock module - 44e00028
is the last debug output. After some time the watchdog is doing his job and restarts again.
Okay ... thinking maybe something had become incompatible with my customization ... so let's try the beaglebone board. Same behaviour :-(
I debugged around and found finally this commit, which breaks the support: / //commit id: f0c3a6c4ad09210d5d4aeafe87685ee75e5683d6// // // ARM: SPL: do not set gd again// //// // Just before calling board_init_f, crt0.S has already// // reserved space for the initial gd on the stack. There// // should be no need to allocate it again.// //// // cc: Albert ARIBAUD albert.u.boot@aribaud.net// // Signed-off-by: Jeroen Hofstee jeroen@myspectrum.nl/
I have reverted this commit in my branch and everything works fine. Finally i have to admit, that i do not fully understand whats going on there in detail.
What i can see, is that there live more than one instance of the "gd". One is setup in arch/arm/lib/cpu/armv7/lowlevel_init.S, another one is setup in arch/arm/lib/crt0.S. Within the "gd-ptr" is loaded to &gdata (gdata __attribute__ ((section(".data")))
I am a bit confused all about this :-)
I just finished bisecting down to this being a problem commit as well. As far as I can see, the commit in question doesn't make sense. Or perhaps rather, the comments in crt0.S aren't quite right in the SPL case where we don't place the initial gd on the stack, it's in the .data section.
As far as I can tell there are two gdata's is the SPL case. One on the stack and one in gdata. Can't the later not be removed, so we end up with a single gd init in crt0.S? Before this patch the code switches back and forth between the two, which is at least a bit weird.
I will have a closer look at the call path, especially why it used to work...
Regards, Jeroen